Initial commit of BoringSSL for Android.

10 files changed, 21454 insertions, 0 deletions

diff --git a/src/crypto/aes/asm/aes-586.pl b/src/crypto/aes/asm/aes-586.pl
new file mode 100755
index 0000000..07fb94c
--- /dev/null
+++ b/src/crypto/aes/asm/aes-586.pl
@@ -0,0 +1,2987 @@
+#!/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/.
+# ====================================================================
+#
+# Version 4.3.
+#
+# 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
+#
+# Version 3.7 reimplements outer rounds as "compact." Meaning that
+# first and last rounds reference compact 256 bytes S-box. This means
+# that first round consumes a lot more CPU cycles and that encrypt
+# and decrypt performance becomes asymmetric. Encrypt performance
+# drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is
+# aggressively pre-fetched.
+#
+# Version 4.0 effectively rolls back to 3.6 and instead implements
+# additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact,
+# which use exclusively 256 byte S-box. These functions are to be
+# called in modes not concealing plain text, such as ECB, or when
+# we're asked to process smaller amount of data [or unconditionally
+# on hyper-threading CPU]. Currently it's called unconditionally from
+# AES_[en|de]crypt, which affects all modes, but CBC. CBC routine
+# still needs to be modified to switch between slower and faster
+# mode when appropriate... But in either case benchmark landscape
+# changes dramatically and below numbers are CPU cycles per processed
+# byte for 128-bit key.
+#
+#		ECB encrypt	ECB decrypt	CBC large chunk
+# P4		52[54]		83[95]		23
+# AMD K8	46[41]		66[70]		18
+# PIII		41[50]		60[77]		24
+# Core 2	31[36]		45[64]		18.5
+# Atom		76[100]		96[138]		60
+# Pentium	115		150		77
+#
+# Version 4.1 switches to compact S-box even in key schedule setup.
+#
+# Version 4.2 prefetches compact S-box in every SSE round or in other
+# words every cache-line is *guaranteed* to be accessed within ~50
+# cycles window. Why just SSE? Because it's needed on hyper-threading
+# CPU! Which is also why it's prefetched with 64 byte stride. Best
+# part is that it has no negative effect on performance:-)  
+#
+# Version 4.3 implements switch between compact and non-compact block
+# functions in AES_cbc_encrypt depending on how much data was asked
+# to be processed in one stroke.
+#
+######################################################################
+# Timing attacks are classified in two classes: synchronous when
+# attacker consciously initiates cryptographic operation and collects
+# timing data of various character afterwards, and asynchronous when
+# malicious code is executed on same CPU simultaneously with AES,
+# instruments itself and performs statistical analysis of this data.
+#
+# As far as synchronous attacks go the root to the AES timing
+# vulnerability is twofold. Firstly, of 256 S-box elements at most 160
+# are referred to in single 128-bit block operation. Well, in C
+# implementation with 4 distinct tables it's actually as little as 40
+# references per 256 elements table, but anyway... Secondly, even
+# though S-box elements are clustered into smaller amount of cache-
+# lines, smaller than 160 and even 40, it turned out that for certain
+# plain-text pattern[s] or simply put chosen plain-text and given key
+# few cache-lines remain unaccessed during block operation. Now, if
+# attacker can figure out this access pattern, he can deduct the key
+# [or at least part of it]. The natural way to mitigate this kind of
+# attacks is to minimize the amount of cache-lines in S-box and/or
+# prefetch them to ensure that every one is accessed for more uniform
+# timing. But note that *if* plain-text was concealed in such way that
+# input to block function is distributed *uniformly*, then attack
+# wouldn't apply. Now note that some encryption modes, most notably
+# CBC, do mask the plain-text in this exact way [secure cipher output
+# is distributed uniformly]. Yes, one still might find input that
+# would reveal the information about given key, but if amount of
+# candidate inputs to be tried is larger than amount of possible key
+# combinations then attack becomes infeasible. This is why revised
+# AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
+# of data is to be processed in one stroke. The current size limit of
+# 512 bytes is chosen to provide same [diminishigly low] probability
+# for cache-line to remain untouched in large chunk operation with
+# large S-box as for single block operation with compact S-box and
+# surely needs more careful consideration...
+#
+# As for asynchronous attacks. There are two flavours: attacker code
+# being interleaved with AES on hyper-threading CPU at *instruction*
+# level, and two processes time sharing single core. As for latter.
+# Two vectors. 1. Given that attacker process has higher priority,
+# yield execution to process performing AES just before timer fires
+# off the scheduler, immediately regain control of CPU and analyze the
+# cache state. For this attack to be efficient attacker would have to
+# effectively slow down the operation by several *orders* of magnitute,
+# by ratio of time slice to duration of handful of AES rounds, which
+# unlikely to remain unnoticed. Not to mention that this also means
+# that he would spend correspondigly more time to collect enough
+# statistical data to mount the attack. It's probably appropriate to
+# say that if adeversary reckons that this attack is beneficial and
+# risks to be noticed, you probably have larger problems having him
+# mere opportunity. In other words suggested code design expects you
+# to preclude/mitigate this attack by overall system security design.
+# 2. Attacker manages to make his code interrupt driven. In order for
+# this kind of attack to be feasible, interrupt rate has to be high
+# enough, again comparable to duration of handful of AES rounds. But
+# is there interrupt source of such rate? Hardly, not even 1Gbps NIC
+# generates interrupts at such raging rate...
+#
+# And now back to the former, hyper-threading CPU or more specifically
+# Intel P4. Recall that asynchronous attack implies that malicious
+# code instruments itself. And naturally instrumentation granularity
+# has be noticeably lower than duration of codepath accessing S-box.
+# Given that all cache-lines are accessed during that time that is.
+# Current implementation accesses *all* cache-lines within ~50 cycles
+# window, which is actually *less* than RDTSC latency on Intel P4!
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],"aes-586.pl",$x86only = $ARGV[$#ARGV] eq "386");
+&static_label("AES_Te");
+&static_label("AES_Td");
+
+$s0="eax";
+$s1="ebx";
+$s2="ecx";
+$s3="edx";
+$key="edi";
+$acc="esi";
+$tbl="ebp";
+
+# stack frame layout in _[x86|sse]_AES_* routines, frame is allocated
+# by caller
+$__ra=&DWP(0,"esp");	# return address
+$__s0=&DWP(4,"esp");	# s0 backing store
+$__s1=&DWP(8,"esp");	# s1 backing store
+$__s2=&DWP(12,"esp");	# s2 backing store
+$__s3=&DWP(16,"esp");	# s3 backing store
+$__key=&DWP(20,"esp");	# pointer to key schedule
+$__end=&DWP(24,"esp");	# pointer to end of key schedule
+$__tbl=&DWP(28,"esp");	# %ebp backing store
+
+# stack frame layout in AES_[en|crypt] routines, which differs from
+# above by 4 and overlaps by %ebp backing store
+$_tbl=&DWP(24,"esp");
+$_esp=&DWP(28,"esp");
+
+sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
+
+$speed_limit=512;	# chunks smaller than $speed_limit are
+			# processed with compact routine in CBC mode
+$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 ($v0,$v1) = ($acc,$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,$__key);				# reincarnate v1 as key
+	&xor	($s[2],&DWP(1,$te,$v0,8));		# s1>>24
+}
+
+# Another experimental routine, which features "horizontal spin," but
+# eliminates one reference to stack. Strangely enough runs slower...
+sub enchoriz()
+{ my ($v0,$v1) = ($key,$acc);
+
+	&movz	($v0,&LB($s0));			#  3, 2, 1, 0*
+	&rotr	($s2,8);			#  8,11,10, 9
+	&mov	($v1,&DWP(0,$te,$v0,8));	#  0
+	&movz	($v0,&HB($s1));			#  7, 6, 5*, 4
+	&rotr	($s3,16);			# 13,12,15,14
+	&xor	($v1,&DWP(3,$te,$v0,8));	#  5
+	&movz	($v0,&HB($s2));			#  8,11,10*, 9
+	&rotr	($s0,16);			#  1, 0, 3, 2
+	&xor	($v1,&DWP(2,$te,$v0,8));	# 10
+	&movz	($v0,&HB($s3));			# 13,12,15*,14
+	&xor	($v1,&DWP(1,$te,$v0,8));	# 15, t[0] collected
+	&mov	($__s0,$v1);			# t[0] saved
+
+	&movz	($v0,&LB($s1));			#  7, 6, 5, 4*
+	&shr	($s1,16);			#  -, -, 7, 6
+	&mov	($v1,&DWP(0,$te,$v0,8));	#  4
+	&movz	($v0,&LB($s3));			# 13,12,15,14*
+	&xor	($v1,&DWP(2,$te,$v0,8));	# 14
+	&movz	($v0,&HB($s0));			#  1, 0, 3*, 2
+	&and	($s3,0xffff0000);		# 13,12, -, -
+	&xor	($v1,&DWP(1,$te,$v0,8));	#  3
+	&movz	($v0,&LB($s2));			#  8,11,10, 9*
+	&or	($s3,$s1);			# 13,12, 7, 6
+	&xor	($v1,&DWP(3,$te,$v0,8));	#  9, t[1] collected
+	&mov	($s1,$v1);			#  s[1]=t[1]
+
+	&movz	($v0,&LB($s0));			#  1, 0, 3, 2*
+	&shr	($s2,16);			#  -, -, 8,11
+	&mov	($v1,&DWP(2,$te,$v0,8));	#  2
+	&movz	($v0,&HB($s3));			# 13,12, 7*, 6
+	&xor	($v1,&DWP(1,$te,$v0,8));	#  7
+	&movz	($v0,&HB($s2));			#  -, -, 8*,11
+	&xor	($v1,&DWP(0,$te,$v0,8));	#  8
+	&mov	($v0,$s3);
+	&shr	($v0,24);			# 13
+	&xor	($v1,&DWP(3,$te,$v0,8));	# 13, t[2] collected
+
+	&movz	($v0,&LB($s2));			#  -, -, 8,11*
+	&shr	($s0,24);			#  1*
+	&mov	($s2,&DWP(1,$te,$v0,8));	# 11
+	&xor	($s2,&DWP(3,$te,$s0,8));	#  1
+	&mov	($s0,$__s0);			# s[0]=t[0]
+	&movz	($v0,&LB($s3));			# 13,12, 7, 6*
+	&shr	($s3,16);			#   ,  ,13,12
+	&xor	($s2,&DWP(2,$te,$v0,8));	#  6
+	&mov	($key,$__key);			# reincarnate v0 as key
+	&and	($s3,0xff);			#   ,  ,13,12*
+	&mov	($s3,&DWP(0,$te,$s3,8));	# 12
+	&xor	($s3,$s2);			# s[2]=t[3] collected
+	&mov	($s2,$v1);			# s[2]=t[2]
+}
+
+# More experimental code... SSE one... Even though this one eliminates
+# *all* references to stack, it's not faster...
+sub sse_encbody()
+{
+	&movz	($acc,&LB("eax"));		#  0
+	&mov	("ecx",&DWP(0,$tbl,$acc,8));	#  0
+	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
+	&movz	("edx",&HB("eax"));		#  1
+	&mov	("edx",&DWP(3,$tbl,"edx",8));	#  1
+	&shr	("eax",16);			#  5, 4
+
+	&movz	($acc,&LB("ebx"));		# 10
+	&xor	("ecx",&DWP(2,$tbl,$acc,8));	# 10
+	&pshufw	("mm6","mm4",0x08);		# 13,12, 9, 8
+	&movz	($acc,&HB("ebx"));		# 11
+	&xor	("edx",&DWP(1,$tbl,$acc,8));	# 11
+	&shr	("ebx",16);			# 15,14
+
+	&movz	($acc,&HB("eax"));		#  5
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	#  5
+	&movq	("mm3",QWP(16,$key));
+	&movz	($acc,&HB("ebx"));		# 15
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	# 15
+	&movd	("mm0","ecx");			# t[0] collected
+
+	&movz	($acc,&LB("eax"));		#  4
+	&mov	("ecx",&DWP(0,$tbl,$acc,8));	#  4
+	&movd	("eax","mm2");			#  7, 6, 3, 2
+	&movz	($acc,&LB("ebx"));		# 14
+	&xor	("ecx",&DWP(2,$tbl,$acc,8));	# 14
+	&movd	("ebx","mm6");			# 13,12, 9, 8
+
+	&movz	($acc,&HB("eax"));		#  3
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	#  3
+	&movz	($acc,&HB("ebx"));		#  9
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	#  9
+	&movd	("mm1","ecx");			# t[1] collected
+
+	&movz	($acc,&LB("eax"));		#  2
+	&mov	("ecx",&DWP(2,$tbl,$acc,8));	#  2
+	&shr	("eax",16);			#  7, 6
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+	&movz	($acc,&LB("ebx"));		#  8
+	&xor	("ecx",&DWP(0,$tbl,$acc,8));	#  8
+	&shr	("ebx",16);			# 13,12
+
+	&movz	($acc,&HB("eax"));		#  7
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	#  7
+	&pxor	("mm0","mm3");
+	&movz	("eax",&LB("eax"));		#  6
+	&xor	("edx",&DWP(2,$tbl,"eax",8));	#  6
+	&pshufw	("mm1","mm0",0x08);		#  5, 4, 1, 0
+	&movz	($acc,&HB("ebx"));		# 13
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	# 13
+	&xor	("ecx",&DWP(24,$key));		# t[2]
+	&movd	("mm4","ecx");			# t[2] collected
+	&movz	("ebx",&LB("ebx"));		# 12
+	&xor	("edx",&DWP(0,$tbl,"ebx",8));	# 12
+	&shr	("ecx",16);
+	&movd	("eax","mm1");			#  5, 4, 1, 0
+	&mov	("ebx",&DWP(28,$key));		# t[3]
+	&xor	("ebx","edx");
+	&movd	("mm5","ebx");			# t[3] collected
+	&and	("ebx",0xffff0000);
+	&or	("ebx","ecx");
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+######################################################################
+# "Compact" block function
+######################################################################
+
+sub enccompact()
+{ my $Fn = \&mov;
+  while ($#_>5) { pop(@_); $Fn=sub{}; }
+  my ($i,$te,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# $Fn is used in first compact round and its purpose is to
+	# void restoration of some values from stack, so that after
+	# 4xenccompact with extra argument $key value is left there...
+	if ($i==3)  {	&$Fn	($key,$__key);			}##%edx
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
+	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
+			&movz	($out,&BP(-128,$te,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}##%eax
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,8);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$__s0);		}##%ebx
+	else        {	&mov	($tmp,$s[2]);
+			&shr	($tmp,16);			}
+	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
+			&and	($tmp,0xFF);
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}##%ecx
+	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
+	else        {	&mov	($tmp,$s[3]);
+			&shr	($tmp,24);			}
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$acc);			}
+	&comment();
+}
+
+sub enctransform()
+{ my @s = ($s0,$s1,$s2,$s3);
+  my $i = shift;
+  my $tmp = $tbl;
+  my $r2  = $key ;
+
+	&and	($tmp,$s[$i]);
+	&lea	($r2,&DWP(0,$s[$i],$s[$i]));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&and	($r2,0xfefefefe);
+	&sub	($acc,$tmp);
+	&mov	($tmp,$s[$i]);
+	&and	($acc,0x1b1b1b1b);
+	&rotr	($tmp,16);
+	&xor	($acc,$r2);	# r2
+	&mov	($r2,$s[$i]);
+
+	&xor	($s[$i],$acc);	# r0 ^ r2
+	&rotr	($r2,16+8);
+	&xor	($acc,$tmp);
+	&rotl	($s[$i],24);
+	&xor	($acc,$r2);
+	&mov	($tmp,0x80808080)	if ($i!=1);
+	&xor	($s[$i],$acc);	# ROTATE(r2^r0,24) ^ r2
+}
+
+&function_begin_B("_x86_AES_encrypt_compact");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$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
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	# prefetch Te4
+	&mov	($key,&DWP(0-128,$tbl));
+	&mov	($acc,&DWP(32-128,$tbl));
+	&mov	($key,&DWP(64-128,$tbl));
+	&mov	($acc,&DWP(96-128,$tbl));
+	&mov	($key,&DWP(128-128,$tbl));
+	&mov	($acc,&DWP(160-128,$tbl));
+	&mov	($key,&DWP(192-128,$tbl));
+	&mov	($acc,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+
+		&enccompact(0,$tbl,$s0,$s1,$s2,$s3,1);
+		&enccompact(1,$tbl,$s1,$s2,$s3,$s0,1);
+		&enccompact(2,$tbl,$s2,$s3,$s0,$s1,1);
+		&enccompact(3,$tbl,$s3,$s0,$s1,$s2,1);
+		&mov	($tbl,0x80808080);
+		&enctransform(2);
+		&enctransform(3);
+		&enctransform(0);
+		&enctransform(1);
+		&mov 	($key,$__key);
+		&mov	($tbl,$__tbl);
+		&add	($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	($key,$__end);
+	&mov	($__key,$key);
+	&jb	(&label("loop"));
+
+	&enccompact(0,$tbl,$s0,$s1,$s2,$s3);
+	&enccompact(1,$tbl,$s1,$s2,$s3,$s0);
+	&enccompact(2,$tbl,$s2,$s3,$s0,$s1);
+	&enccompact(3,$tbl,$s3,$s0,$s1,$s2);
+
+	&xor	($s0,&DWP(16,$key));
+	&xor	($s1,&DWP(20,$key));
+	&xor	($s2,&DWP(24,$key));
+	&xor	($s3,&DWP(28,$key));
+
+	&ret	();
+&function_end_B("_x86_AES_encrypt_compact");
+
+######################################################################
+# "Compact" SSE block function.
+######################################################################
+#
+# Performance is not actually extraordinary in comparison to pure
+# x86 code. In particular encrypt performance is virtually the same.
+# Decrypt performance on the other hand is 15-20% better on newer
+# µ-archs [but we're thankful for *any* improvement here], and ~50%
+# better on PIII:-) And additionally on the pros side this code
+# eliminates redundant references to stack and thus relieves/
+# minimizes the pressure on the memory bus.
+#
+# MMX register layout                           lsb
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |          mm4          |          mm0          |
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |     s3    |     s2    |     s1    |     s0    |    
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0|
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+#
+# Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8.
+# In this terms encryption and decryption "compact" permutation
+# matrices can be depicted as following:
+#
+# encryption              lsb	# decryption              lsb
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t0 || 15 | 10 |  5 |  0 |	# | t0 ||  7 | 10 | 13 |  0 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t1 ||  3 | 14 |  9 |  4 |	# | t1 || 11 | 14 |  1 |  4 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t2 ||  7 |  2 | 13 |  8 |	# | t2 || 15 |  2 |  5 |  8 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t3 || 11 |  6 |  1 | 12 |	# | t3 ||  3 |  6 |  9 | 12 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+#
+######################################################################
+# Why not xmm registers? Short answer. It was actually tested and
+# was not any faster, but *contrary*, most notably on Intel CPUs.
+# Longer answer. Main advantage of using mm registers is that movd
+# latency is lower, especially on Intel P4. While arithmetic
+# instructions are twice as many, they can be scheduled every cycle
+# and not every second one when they are operating on xmm register,
+# so that "arithmetic throughput" remains virtually the same. And
+# finally the code can be executed even on elder SSE-only CPUs:-)
+
+sub sse_enccompact()
+{
+	&pshufw	("mm1","mm0",0x08);		#  5, 4, 1, 0
+	&pshufw	("mm5","mm4",0x0d);		# 15,14,11,10
+	&movd	("eax","mm1");			#  5, 4, 1, 0
+	&movd	("ebx","mm5");			# 15,14,11,10
+	&mov	($__key,$key);
+
+	&movz	($acc,&LB("eax"));		#  0
+	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
+	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
+	&movz	($key,&LB("ebx"));		# 10
+	&movz	("edx",&BP(-128,$tbl,"edx",1));	#  1
+	&shr	("eax",16);			#  5, 4
+	&shl	("edx",8);			#  1
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 10
+	&movz	($key,&HB("ebx"));		# 11
+	&shl	($acc,16);			# 10
+	&pshufw	("mm6","mm4",0x08);		# 13,12, 9, 8
+	&or	("ecx",$acc);			# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  5
+	&shl	($acc,24);			# 11
+	&shr	("ebx",16);			# 15,14
+	&or	("edx",$acc);			# 11
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  5
+	&movz	($key,&HB("ebx"));		# 15
+	&shl	($acc,8);			#  5
+	&or	("ecx",$acc);			#  5
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 15
+	&movz	($key,&LB("eax"));		#  4
+	&shl	($acc,24);			# 15
+	&or	("ecx",$acc);			# 15
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
+	&movd	("eax","mm2");			#  7, 6, 3, 2
+	&movd	("mm0","ecx");			# t[0] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	# 14
+	&movz	($key,&HB("eax"));		#  3
+	&shl	("ecx",16);			# 14
+	&movd	("ebx","mm6");			# 13,12, 9, 8
+	&or	("ecx",$acc);			# 14
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  3
+	&movz	($key,&HB("ebx"));		#  9
+	&shl	($acc,24);			#  3
+	&or	("ecx",$acc);			#  3
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  9
+	&movz	($key,&LB("ebx"));		#  8
+	&shl	($acc,8);			#  9
+	&shr	("ebx",16);			# 13,12
+	&or	("ecx",$acc);			#  9
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  8
+	&movz	($key,&LB("eax"));		#  2
+	&shr	("eax",16);			#  7, 6
+	&movd	("mm1","ecx");			# t[1] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	#  2
+	&movz	($key,&HB("eax"));		#  7
+	&shl	("ecx",16);			#  2
+	&and	("eax",0xff);			#  6
+	&or	("ecx",$acc);			#  2
+
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
+	&shl	($acc,24);			#  7
+	&and	("ebx",0xff);			# 12
+	&movz	("eax",&BP(-128,$tbl,"eax",1));	#  6
+	&or	("ecx",$acc);			#  7
+	&shl	("eax",16);			#  6
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 13
+	&or	("edx","eax");			#  6
+	&shl	($acc,8);			# 13
+	&movz	("ebx",&BP(-128,$tbl,"ebx",1));	# 12
+	&or	("ecx",$acc);			# 13
+	&or	("edx","ebx");			# 12
+	&mov	($key,$__key);
+	&movd	("mm4","ecx");			# t[2] collected
+	&movd	("mm5","edx");			# t[3] collected
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+					if (!$x86only) {
+&function_begin_B("_sse_AES_encrypt_compact");
+	&pxor	("mm0",&QWP(0,$key));	#  7, 6, 5, 4, 3, 2, 1, 0
+	&pxor	("mm4",&QWP(8,$key));	# 15,14,13,12,11,10, 9, 8
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	&mov	($s0,0x1b1b1b1b);		# magic constant
+	&mov	(&DWP(8,"esp"),$s0);
+	&mov	(&DWP(12,"esp"),$s0);
+
+	# prefetch Te4
+	&mov	($s0,&DWP(0-128,$tbl));
+	&mov	($s1,&DWP(32-128,$tbl));
+	&mov	($s2,&DWP(64-128,$tbl));
+	&mov	($s3,&DWP(96-128,$tbl));
+	&mov	($s0,&DWP(128-128,$tbl));
+	&mov	($s1,&DWP(160-128,$tbl));
+	&mov	($s2,&DWP(192-128,$tbl));
+	&mov	($s3,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+		&sse_enccompact();
+		&add	($key,16);
+		&cmp	($key,$__end);
+		&ja	(&label("out"));
+
+		&movq	("mm2",&QWP(8,"esp"));
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&movq	("mm1","mm0");		&movq	("mm5","mm4");	# r0
+		&pcmpgtb("mm3","mm0");		&pcmpgtb("mm7","mm4");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&pshufw	("mm2","mm0",0xb1);	&pshufw	("mm6","mm4",0xb1);# ROTATE(r0,16)
+		&paddb	("mm0","mm0");		&paddb	("mm4","mm4");
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# = r2
+		&pshufw	("mm3","mm2",0xb1);	&pshufw	("mm7","mm6",0xb1);# r0
+		&pxor	("mm1","mm0");		&pxor	("mm5","mm4");	# r0^r2
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= ROTATE(r0,16)
+
+		&movq	("mm2","mm3");		&movq	("mm6","mm7");
+		&pslld	("mm3",8);		&pslld	("mm7",8);
+		&psrld	("mm2",24);		&psrld	("mm6",24);
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= r0<<8
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= r0>>24
+
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&movq	("mm2",&QWP(0,$key));	&movq	("mm6",&QWP(8,$key));
+		&psrld	("mm1",8);		&psrld	("mm5",8);
+		&mov	($s0,&DWP(0-128,$tbl));
+		&pslld	("mm3",24);		&pslld	("mm7",24);
+		&mov	($s1,&DWP(64-128,$tbl));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= (r2^r0)<<8
+		&mov	($s2,&DWP(128-128,$tbl));
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= (r2^r0)>>24
+		&mov	($s3,&DWP(192-128,$tbl));
+
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");
+	&jmp	(&label("loop"));
+
+	&set_label("out",16);
+	&pxor	("mm0",&QWP(0,$key));
+	&pxor	("mm4",&QWP(8,$key));
+
+	&ret	();
+&function_end_B("_sse_AES_encrypt_compact");
+					}
+
+######################################################################
+# Vanilla block function.
+######################################################################
+
+sub encstep()
+{ my ($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	($key,$__key);			}##%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 ($s[1],$__s0);		}##%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 ($s[2],$__s1);		}##%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	($key,$__key);			}##%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 ($s[1],$__s0);		}##%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 ($s[2],$__s1);		}##%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);			}
+}
+
+&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",$acc="ecx");
+	}
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($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	($__end,$acc);		# end of key schedule
+
+	    &set_label("loop",16);
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$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,$__end);
+	    &mov	($__key,$key);
+	    &jb		(&label("loop"));
+	}
+	else {
+	    &cmp	($acc,10);
+	    &jle	(&label("10rounds"));
+	    &cmp	($acc,12);
+	    &jle	(&label("12rounds"));
+
+	&set_label("14rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$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	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("12rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$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	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("10rounds",4);
+	    for ($i=1;$i<10;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$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",$acc="esi");
+	}
+	&enclast(0,$tbl,$s0,$s1,$s2,$s3);
+	&enclast(1,$tbl,$s1,$s2,$s3,$s0);
+	&enclast(2,$tbl,$s2,$s3,$s0,$s1);
+	&enclast(3,$tbl,$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);
+
+#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+#rcon:
+	&data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
+	&data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
+	&data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000);
+	&data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000);
+&function_end_B("_x86_AES_encrypt");
+
+# void asm_AES_encrypt (const void *inp,void *out,const AES_KEY *key);
+&function_begin("asm_AES_encrypt");
+	&mov	($acc,&wparam(0));		# load inp
+	&mov	($key,&wparam(2));		# load key
+
+	&mov	($s0,"esp");
+	&sub	("esp",36);
+	&and	("esp",-64);			# align to cache-line
+
+	# place stack frame just "above" the key schedule
+	&lea	($s1,&DWP(-64-63,$key));
+	&sub	($s1,"esp");
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	("esp",$s1);
+	&add	("esp",4);	# 4 is reserved for caller's return address
+	&mov	($_esp,$s0);			# save stack pointer
+
+	&call   (&label("pic_point"));          # make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only);
+	&lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+
+	# pick Te4 copy which can't "overlap" with stack frame or key schedule
+	&lea	($s1,&DWP(768-4,"esp"));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),25);	# check for SSE bit
+	&jnc	(&label("x86"));
+
+	&movq	("mm0",&QWP(0,$acc));
+	&movq	("mm4",&QWP(8,$acc));
+	&call	("_sse_AES_encrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&movq	(&QWP(0,$acc),"mm0");		# write output data
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&function_end_A();
+					}
+	&set_label("x86",16);
+	&mov	($_tbl,$tbl);
+	&mov	($s0,&DWP(0,$acc));		# load input data
+	&mov	($s1,&DWP(4,$acc));
+	&mov	($s2,&DWP(8,$acc));
+	&mov	($s3,&DWP(12,$acc));
+	&call	("_x86_AES_encrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&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("asm_AES_encrypt");
+
+#--------------------------------------------------------------------#
+
+######################################################################
+# "Compact" block function
+######################################################################
+
+sub deccompact()
+{ my $Fn = \&mov;
+  while ($#_>5) { pop(@_); $Fn=sub{}; }
+  my ($i,$td,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# $Fn is used in first compact round and its purpose is to
+	# void restoration of some values from stack, so that after
+	# 4xdeccompact with extra argument $key, $s0 and $s1 values
+	# are left there...
+	if($i==3)   {	&$Fn	($key,$__key);			}
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+			&movz	($out,&BP(-128,$td,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(-128,$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	($tmp,&BP(-128,$td,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &$Fn ($s[2],$__s1);		}
+	else        {	&mov	($tmp,$s[3]);			}
+			&shr	($tmp,24);
+			&movz	($tmp,&BP(-128,$td,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&$Fn	($s[3],$__s0);			}
+}
+
+# must be called with 2,3,0,1 as argument sequence!!!
+sub dectransform()
+{ my @s = ($s0,$s1,$s2,$s3);
+  my $i = shift;
+  my $tmp = $key;
+  my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1);
+  my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
+  my $tp8 = $tbl;
+
+	&mov	($tmp,0x80808080);
+	&and	($tmp,$s[$i]);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp2,&DWP(0,$s[$i],$s[$i]));
+	&sub	($acc,$tmp);
+	&and	($tp2,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp2);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp4,&DWP(0,$tp2,$tp2));
+	&sub	($acc,$tmp);
+	&and	($tp4,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp2,$s[$i]);	# tp2^tp1
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp4);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp8,&DWP(0,$tp4,$tp4));
+	&sub	($acc,$tmp);
+	&and	($tp8,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp4,$s[$i]);	# tp4^tp1
+	 &rotl	($s[$i],8);	# = ROTATE(tp1,8)
+	&xor	($tp8,$acc);
+
+	&xor	($s[$i],$tp2);
+	&xor	($tp2,$tp8);
+	&xor	($s[$i],$tp4);
+	&xor	($tp4,$tp8);
+	&rotl	($tp2,24);
+	&xor	($s[$i],$tp8);	# ^= tp8^(tp4^tp1)^(tp2^tp1)
+	&rotl	($tp4,16);
+	&xor	($s[$i],$tp2);	# ^= ROTATE(tp8^tp2^tp1,24)
+	&rotl	($tp8,8);
+	&xor	($s[$i],$tp4);	# ^= ROTATE(tp8^tp4^tp1,16)
+	 &mov	($s[0],$__s0)			if($i==2); #prefetch $s0
+	 &mov	($s[1],$__s1)			if($i==3); #prefetch $s1
+	 &mov	($s[2],$__s2)			if($i==1);
+	&xor	($s[$i],$tp8);	# ^= ROTATE(tp8,8)
+
+	&mov	($s[3],$__s3)			if($i==1);
+	&mov	(&DWP(4+4*$i,"esp"),$s[$i])	if($i>=2);
+}
+
+&function_begin_B("_x86_AES_decrypt_compact");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$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
+
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	# prefetch Td4
+	&mov	($key,&DWP(0-128,$tbl));
+	&mov	($acc,&DWP(32-128,$tbl));
+	&mov	($key,&DWP(64-128,$tbl));
+	&mov	($acc,&DWP(96-128,$tbl));
+	&mov	($key,&DWP(128-128,$tbl));
+	&mov	($acc,&DWP(160-128,$tbl));
+	&mov	($key,&DWP(192-128,$tbl));
+	&mov	($acc,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+
+		&deccompact(0,$tbl,$s0,$s3,$s2,$s1,1);
+		&deccompact(1,$tbl,$s1,$s0,$s3,$s2,1);
+		&deccompact(2,$tbl,$s2,$s1,$s0,$s3,1);
+		&deccompact(3,$tbl,$s3,$s2,$s1,$s0,1);
+		&dectransform(2);
+		&dectransform(3);
+		&dectransform(0);
+		&dectransform(1);
+		&mov 	($key,$__key);
+		&mov	($tbl,$__tbl);
+		&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,$__end);
+	&mov	($__key,$key);
+	&jb	(&label("loop"));
+
+	&deccompact(0,$tbl,$s0,$s3,$s2,$s1);
+	&deccompact(1,$tbl,$s1,$s0,$s3,$s2);
+	&deccompact(2,$tbl,$s2,$s1,$s0,$s3);
+	&deccompact(3,$tbl,$s3,$s2,$s1,$s0);
+
+	&xor	($s0,&DWP(16,$key));
+	&xor	($s1,&DWP(20,$key));
+	&xor	($s2,&DWP(24,$key));
+	&xor	($s3,&DWP(28,$key));
+
+	&ret	();
+&function_end_B("_x86_AES_decrypt_compact");
+
+######################################################################
+# "Compact" SSE block function.
+######################################################################
+
+sub sse_deccompact()
+{
+	&pshufw	("mm1","mm0",0x0c);		#  7, 6, 1, 0
+	&pshufw	("mm5","mm4",0x09);		# 13,12,11,10
+	&movd	("eax","mm1");			#  7, 6, 1, 0
+	&movd	("ebx","mm5");			# 13,12,11,10
+	&mov	($__key,$key);
+
+	&movz	($acc,&LB("eax"));		#  0
+	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x06);		#  3, 2, 5, 4
+	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
+	&movz	($key,&LB("ebx"));		# 10
+	&movz	("edx",&BP(-128,$tbl,"edx",1));	#  1
+	&shr	("eax",16);			#  7, 6
+	&shl	("edx",8);			#  1
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 10
+	&movz	($key,&HB("ebx"));		# 11
+	&shl	($acc,16);			# 10
+	&pshufw	("mm6","mm4",0x03);		# 9, 8,15,14
+	&or	("ecx",$acc);			# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  7
+	&shl	($acc,24);			# 11
+	&shr	("ebx",16);			# 13,12
+	&or	("edx",$acc);			# 11
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
+	&shl	($acc,24);			#  7
+	&or	("ecx",$acc);			#  7
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 13
+	&movz	($key,&LB("eax"));		#  6
+	&shl	($acc,8);			# 13
+	&movd	("eax","mm2");			#  3, 2, 5, 4
+	&or	("ecx",$acc);			# 13
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  6
+	&movz	($key,&LB("ebx"));		# 12
+	&shl	($acc,16);			#  6
+	&movd	("ebx","mm6");			#  9, 8,15,14
+	&movd	("mm0","ecx");			# t[0] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	# 12
+	&movz	($key,&LB("eax"));		#  4
+	&or	("ecx",$acc);			# 12
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
+	&or	("edx",$acc);			#  4
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 14
+	&movz	($key,&HB("eax"));		#  5
+	&shl	($acc,16);			# 14
+	&shr	("eax",16);			#  3, 2
+	&or	("edx",$acc);			# 14
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  5
+	&movz	($key,&HB("ebx"));		# 15
+	&shr	("ebx",16);			#  9, 8
+	&shl	($acc,8);			#  5
+	&movd	("mm1","edx");			# t[1] collected
+	&movz	("edx",&BP(-128,$tbl,$key,1));	# 15
+	&movz	($key,&HB("ebx"));		#  9
+	&shl	("edx",24);			# 15
+	&and	("ebx",0xff);			#  8
+	&or	("edx",$acc);			# 15
+
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  9
+	&movz	($key,&LB("eax"));		#  2
+	&shl	($acc,8);			#  9
+	&movz	("eax",&HB("eax"));		#  3
+	&movz	("ebx",&BP(-128,$tbl,"ebx",1));	#  8
+	&or	("ecx",$acc);			#  9
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  2
+	&or	("edx","ebx");			#  8
+	&shl	($acc,16);			#  2
+	&movz	("eax",&BP(-128,$tbl,"eax",1));	#  3
+	&or	("edx",$acc);			#  2
+	&shl	("eax",24);			#  3
+	&or	("ecx","eax");			#  3
+	&mov	($key,$__key);
+	&movd	("mm4","edx");			# t[2] collected
+	&movd	("mm5","ecx");			# t[3] collected
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+					if (!$x86only) {
+&function_begin_B("_sse_AES_decrypt_compact");
+	&pxor	("mm0",&QWP(0,$key));	#  7, 6, 5, 4, 3, 2, 1, 0
+	&pxor	("mm4",&QWP(8,$key));	# 15,14,13,12,11,10, 9, 8
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	&mov	($s0,0x1b1b1b1b);		# magic constant
+	&mov	(&DWP(8,"esp"),$s0);
+	&mov	(&DWP(12,"esp"),$s0);
+
+	# prefetch Td4
+	&mov	($s0,&DWP(0-128,$tbl));
+	&mov	($s1,&DWP(32-128,$tbl));
+	&mov	($s2,&DWP(64-128,$tbl));
+	&mov	($s3,&DWP(96-128,$tbl));
+	&mov	($s0,&DWP(128-128,$tbl));
+	&mov	($s1,&DWP(160-128,$tbl));
+	&mov	($s2,&DWP(192-128,$tbl));
+	&mov	($s3,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+		&sse_deccompact();
+		&add	($key,16);
+		&cmp	($key,$__end);
+		&ja	(&label("out"));
+
+		# ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N)
+		&movq	("mm3","mm0");		&movq	("mm7","mm4");
+		&movq	("mm2","mm0",1);	&movq	("mm6","mm4",1);
+		&movq	("mm1","mm0");		&movq	("mm5","mm4");
+		&pshufw	("mm0","mm0",0xb1);	&pshufw	("mm4","mm4",0xb1);# = ROTATE(tp0,16)
+		&pslld	("mm2",8);		&pslld	("mm6",8);
+		&psrld	("mm3",8);		&psrld	("mm7",8);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp0<<8
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp0>>8
+		&pslld	("mm2",16);		&pslld	("mm6",16);
+		&psrld	("mm3",16);		&psrld	("mm7",16);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp0<<24
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp0>>24
+
+		&movq	("mm3",&QWP(8,"esp"));
+		&pxor	("mm2","mm2");		&pxor	("mm6","mm6");
+		&pcmpgtb("mm2","mm1");		&pcmpgtb("mm6","mm5");
+		&pand	("mm2","mm3");		&pand	("mm6","mm3");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm2");		&pxor	("mm5","mm6");	# tp2
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&movq	("mm2","mm1");		&movq	("mm6","mm5");
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp2
+		&pslld	("mm3",24);		&pslld	("mm7",24);
+		&psrld	("mm2",8);		&psrld	("mm6",8);
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp2<<24
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp2>>8
+
+		&movq	("mm2",&QWP(8,"esp"));
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&pcmpgtb("mm3","mm1");		&pcmpgtb("mm7","mm5");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm3");		&pxor	("mm5","mm7");	# tp4
+		&pshufw	("mm3","mm1",0xb1);	&pshufw	("mm7","mm5",0xb1);
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp4
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= ROTATE(tp4,16)	
+
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&pcmpgtb("mm3","mm1");		&pcmpgtb("mm7","mm5");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm3");		&pxor	("mm5","mm7");	# tp8
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&pshufw	("mm2","mm1",0xb1);	&pshufw	("mm6","mm5",0xb1);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= ROTATE(tp8,16)
+		&pslld	("mm1",8);		&pslld	("mm5",8);
+		&psrld	("mm3",8);		&psrld	("mm7",8);
+		&movq	("mm2",&QWP(0,$key));	&movq	("mm6",&QWP(8,$key));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8<<8
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp8>>8
+		&mov	($s0,&DWP(0-128,$tbl));
+		&pslld	("mm1",16);		&pslld	("mm5",16);
+		&mov	($s1,&DWP(64-128,$tbl));
+		&psrld	("mm3",16);		&psrld	("mm7",16);
+		&mov	($s2,&DWP(128-128,$tbl));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8<<24
+		&mov	($s3,&DWP(192-128,$tbl));
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp8>>24
+
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");
+	&jmp	(&label("loop"));
+
+	&set_label("out",16);
+	&pxor	("mm0",&QWP(0,$key));
+	&pxor	("mm4",&QWP(8,$key));
+
+	&ret	();
+&function_end_B("_sse_AES_decrypt_compact");
+					}
+
+######################################################################
+# Vanilla block function.
+######################################################################
+
+sub decstep()
+{ my ($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	($key,$__key);			}
+	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 ($s[2],$__s1);		}
+	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	($s[3],$__s0);			}
+			&comment();
+}
+
+sub declast()
+{ my ($i,$td,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	if($i==0)   {	&lea	($td,&DWP(2048+128,$td));
+			&mov	($tmp,&DWP(0-128,$td));
+			&mov	($acc,&DWP(32-128,$td));
+			&mov	($tmp,&DWP(64-128,$td));
+			&mov	($acc,&DWP(96-128,$td));
+			&mov	($tmp,&DWP(128-128,$td));
+			&mov	($acc,&DWP(160-128,$td));
+			&mov	($tmp,&DWP(192-128,$td));
+			&mov	($acc,&DWP(224-128,$td));
+			&lea	($td,&DWP(-128,$td));		}
+	if($i==3)   {	&mov	($key,$__key);			}
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+			&movz	($out,&BP(0,$td,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(0,$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	($tmp,&BP(0,$td,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}
+	else        {	&mov	($tmp,$s[3]);			}
+			&shr	($tmp,24);
+			&movz	($tmp,&BP(0,$td,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$__s0);
+			&lea	($td,&DWP(-2048,$td));		}
+}
+
+&function_begin_B("_x86_AES_decrypt");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($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	($__end,$acc);		# end of key schedule
+	    &set_label("loop",16);
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$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,$__end);
+	    &mov	($__key,$key);
+	    &jb		(&label("loop"));
+	}
+	else {
+	    &cmp	($acc,10);
+	    &jle	(&label("10rounds"));
+	    &cmp	($acc,12);
+	    &jle	(&label("12rounds"));
+
+	&set_label("14rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$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	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("12rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$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	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("10rounds",4);
+	    for ($i=1;$i<10;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$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,$tbl,$s0,$s3,$s2,$s1);
+	&declast(1,$tbl,$s1,$s0,$s3,$s2);
+	&declast(2,$tbl,$s2,$s1,$s0,$s3);
+	&declast(3,$tbl,$s3,$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:	# four copies of Td4 to choose from to avoid L1 aliasing
+	&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);
+
+	&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);
+
+	&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);
+
+	&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 asm_AES_decrypt (const void *inp,void *out,const AES_KEY *key);
+&function_begin("asm_AES_decrypt");
+	&mov	($acc,&wparam(0));		# load inp
+	&mov	($key,&wparam(2));		# load key
+
+	&mov	($s0,"esp");
+	&sub	("esp",36);
+	&and	("esp",-64);			# align to cache-line
+
+	# place stack frame just "above" the key schedule
+	&lea	($s1,&DWP(-64-63,$key));
+	&sub	($s1,"esp");
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	("esp",$s1);
+	&add	("esp",4);	# 4 is reserved for caller's return address
+	&mov	($_esp,$s0);	# save stack pointer
+
+	&call   (&label("pic_point"));          # make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
+	&lea    ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl));
+
+	# pick Td4 copy which can't "overlap" with stack frame or key schedule
+	&lea	($s1,&DWP(768-4,"esp"));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),25);	# check for SSE bit
+	&jnc	(&label("x86"));
+
+	&movq	("mm0",&QWP(0,$acc));
+	&movq	("mm4",&QWP(8,$acc));
+	&call	("_sse_AES_decrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&movq	(&QWP(0,$acc),"mm0");		# write output data
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&function_end_A();
+					}
+	&set_label("x86",16);
+	&mov	($_tbl,$tbl);
+	&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_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($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("asm_AES_decrypt");
+
+# void asm_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)		# return address	 0(%esp)
+#              0(%esp)		# s0 backing store	 4(%esp)	
+#              4(%esp)		# s1 backing store	 8(%esp)
+#              8(%esp)		# s2 backing store	12(%esp)
+#             12(%esp)		# s3 backing store	16(%esp)
+#             16(%esp)		# key backup		20(%esp)
+#             20(%esp)		# end of key schedule	24(%esp)
+#             24(%esp)		# %ebp backup		28(%esp)
+#             28(%esp)		# %esp backup
+my $_inp=&DWP(32,"esp");	# copy of wparam(0)
+my $_out=&DWP(36,"esp");	# copy of wparam(1)
+my $_len=&DWP(40,"esp");	# copy of wparam(2)
+my $_key=&DWP(44,"esp");	# copy of wparam(3)
+my $_ivp=&DWP(48,"esp");	# copy of wparam(4)
+my $_tmp=&DWP(52,"esp");	# volatile variable
+#
+my $ivec=&DWP(60,"esp");	# ivec[16]
+my $aes_key=&DWP(76,"esp");	# copy of aes_key
+my $mark=&DWP(76+240,"esp");	# copy of aes_key->rounds
+
+&function_begin("asm_AES_cbc_encrypt");
+	&mov	($s2 eq "ecx"? $s2 : "",&wparam(2));	# load len
+	&cmp	($s2,0);
+	&je	(&label("drop_out"));
+
+	&call   (&label("pic_point"));		# make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
+
+	&cmp	(&wparam(5),0);
+	&lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+	&jne	(&label("picked_te"));
+	&lea	($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl));
+	&set_label("picked_te");
+
+	# one can argue if this is required
+	&pushf	();
+	&cld	();
+
+	&cmp	($s2,$speed_limit);
+	&jb	(&label("slow_way"));
+	&test	($s2,15);
+	&jnz	(&label("slow_way"));
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),28);	# check for hyper-threading bit
+	&jc	(&label("slow_way"));
+					}
+	# pre-allocate aligned stack frame...
+	&lea	($acc,&DWP(-80-244,"esp"));
+	&and	($acc,-64);
+
+	# ... and make sure it doesn't alias with $tbl modulo 4096
+	&mov	($s0,$tbl);
+	&lea	($s1,&DWP(2048+256,$tbl));
+	&mov	($s3,$acc);
+	&and	($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("tbl_break_out"));
+	&sub	($s3,$s1);
+	&sub	($acc,$s3);
+	&jmp	(&label("tbl_ok"));
+	&set_label("tbl_break_out",4);	# else %esp -= (p-s)&0xfff + framesz;
+	&sub	($s3,$s0);
+	&and	($s3,0xfff);
+	&add	($s3,384);
+	&sub	($acc,$s3);
+	&set_label("tbl_ok",4);
+
+	&lea	($s3,&wparam(0));	# obtain pointer to parameter block
+	&exch	("esp",$acc);		# allocate stack frame
+	&add	("esp",4);		# reserve for return address!
+	&mov	($_tbl,$tbl);		# save %ebp
+	&mov	($_esp,$acc);		# save %esp
+
+	&mov	($s0,&DWP(0,$s3));	# load inp
+	&mov	($s1,&DWP(4,$s3));	# load out
+	#&mov	($s2,&DWP(8,$s3));	# load len
+	&mov	($key,&DWP(12,$s3));	# load key
+	&mov	($acc,&DWP(16,$s3));	# load ivp
+	&mov	($s3,&DWP(20,$s3));	# load enc flag
+
+	&mov	($_inp,$s0);		# save copy of inp
+	&mov	($_out,$s1);		# save copy of out
+	&mov	($_len,$s2);		# save copy of len
+	&mov	($_key,$key);		# save copy of key
+	&mov	($_ivp,$acc);		# save copy of ivp
+
+	&mov	($mark,0);		# copy of aes_key->rounds = 0;
+	# do we copy key schedule to stack?
+	&mov	($s1 eq "ebx" ? $s1 : "",$key);
+	&mov	($s2 eq "ecx" ? $s2 : "",244/4);
+	&sub	($s1,$tbl);
+	&mov	("esi",$key);
+	&and	($s1,0xfff);
+	&lea	("edi",$aes_key);
+	&cmp	($s1,2048+256);
+	&jb	(&label("do_copy"));
+	&cmp	($s1,4096-244);
+	&jb	(&label("skip_copy"));
+	&set_label("do_copy",4);
+		&mov	($_key,"edi");
+		&data_word(0xA5F3F689);	# rep movsd
+	&set_label("skip_copy");
+
+	&mov	($key,16);
+	&set_label("prefetch_tbl",4);
+		&mov	($s0,&DWP(0,$tbl));
+		&mov	($s1,&DWP(32,$tbl));
+		&mov	($s2,&DWP(64,$tbl));
+		&mov	($acc,&DWP(96,$tbl));
+		&lea	($tbl,&DWP(128,$tbl));
+		&sub	($key,1);
+	&jnz	(&label("prefetch_tbl"));
+	&sub	($tbl,2048);
+
+	&mov	($acc,$_inp);
+	&mov	($key,$_ivp);
+
+	&cmp	($s3,0);
+	&je	(&label("fast_decrypt"));
+
+#----------------------------- ENCRYPT -----------------------------#
+	&mov	($s0,&DWP(0,$key));		# load iv
+	&mov	($s1,&DWP(4,$key));
+
+	&set_label("fast_enc_loop",16);
+		&mov	($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),$s3);
+
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($s2,$_len);		# load len
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_enc_loop"));
+	&mov	($acc,$_ivp);		# load ivp
+	&mov	($s2,&DWP(8,$key));	# restore last 2 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);
+	&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");
+	&mov	("esp",$_esp);
+	&popf	();
+    &set_label("drop_out");
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+#----------------------------- DECRYPT -----------------------------#
+&set_label("fast_decrypt",16);
+
+	&cmp	($acc,$_out);
+	&je	(&label("fast_dec_in_place"));	# in-place processing...
+
+	&mov	($_tmp,$key);
+
+	&align	(4);
+	&set_label("fast_dec_loop",16);
+		&mov	($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,$key));
+
+		&mov	($key,$_out);		# load out
+		&mov	($acc,$_inp);		# load inp
+
+		&mov	(&DWP(0,$key),$s0);	# write output
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($s2,$_len);		# load len
+		&mov	($_tmp,$acc);		# save ivp
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($key,&DWP(16,$key));	# advance out
+		&mov	($_out,$key);		# save out
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_dec_loop"));
+	&mov	($key,$_tmp);		# load temp ivp
+	&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("fast_dec_out"));
+
+    &set_label("fast_dec_in_place",16);
+	&set_label("fast_dec_in_place_loop");
+		&mov	($s0,&DWP(0,$acc));	# read input
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&lea	($key,$ivec);
+		&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));	# advance out
+		&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
+		&mov	($s2,$_len);		# load len
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_dec_in_place_loop"));
+
+    &set_label("fast_dec_out",4);
+	&cmp	($mark,0);		# was the key schedule copied?
+	&mov	("edi",$_key);
+	&je	(&label("skip_dzero"));
+	# zero copy of key schedule
+	&mov	("ecx",240/4);
+	&xor	("eax","eax");
+	&align	(4);
+	&data_word(0xABF3F689);		# rep stosd
+	&set_label("skip_dzero");
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+#--------------------------- SLOW ROUTINE ---------------------------#
+&set_label("slow_way",16);
+
+	&mov	($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap
+	&mov	($key,&wparam(3));	# load key
+
+	# pre-allocate aligned stack frame...
+	&lea	($acc,&DWP(-80,"esp"));
+	&and	($acc,-64);
+
+	# ... and make sure it doesn't alias with $key modulo 1024
+	&lea	($s1,&DWP(-80-63,$key));
+	&sub	($s1,$acc);
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	($acc,$s1);
+
+	# pick S-box copy which can't overlap with stack frame or $key
+	&lea	($s1,&DWP(768,$acc));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+	&lea	($s3,&wparam(0));	# pointer to parameter block
+
+	&exch	("esp",$acc);
+	&add	("esp",4);		# reserve for return address!
+	&mov	($_tbl,$tbl);		# save %ebp
+	&mov	($_esp,$acc);		# save %esp
+	&mov	($_tmp,$s0);		# save OPENSSL_ia32cap
+
+	&mov	($s0,&DWP(0,$s3));	# load inp
+	&mov	($s1,&DWP(4,$s3));	# load out
+	#&mov	($s2,&DWP(8,$s3));	# load len
+	#&mov	($key,&DWP(12,$s3));	# load key
+	&mov	($acc,&DWP(16,$s3));	# load ivp
+	&mov	($s3,&DWP(20,$s3));	# load enc flag
+
+	&mov	($_inp,$s0);		# save copy of inp
+	&mov	($_out,$s1);		# save copy of out
+	&mov	($_len,$s2);		# save copy of len
+	&mov	($_key,$key);		# save copy of key
+	&mov	($_ivp,$acc);		# save copy of ivp
+
+	&mov	($key,$acc);
+	&mov	($acc,$s0);
+
+	&cmp	($s3,0);
+	&je	(&label("slow_decrypt"));
+
+#--------------------------- SLOW ENCRYPT ---------------------------#
+	&cmp	($s2,16);
+	&mov	($s3,$s1);
+	&jb	(&label("slow_enc_tail"));
+
+					if (!$x86only) {
+	&bt	($_tmp,25);		# check for SSE bit
+	&jnc	(&label("slow_enc_x86"));
+
+	&movq	("mm0",&QWP(0,$key));	# load iv
+	&movq	("mm4",&QWP(8,$key));
+
+	&set_label("slow_enc_loop_sse",16);
+		&pxor	("mm0",&QWP(0,$acc));	# xor input data
+		&pxor	("mm4",&QWP(8,$acc));
+
+		&mov	($key,$_key);
+		&call	("_sse_AES_encrypt_compact");
+
+		&mov	($acc,$_inp);		# load inp
+		&mov	($key,$_out);		# load out
+		&mov	($s2,$_len);		# load len
+
+		&movq	(&QWP(0,$key),"mm0");	# save output data
+		&movq	(&QWP(8,$key),"mm4");
+
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&cmp	($s2,16);
+		&mov	($_len,$s2);		# save len
+	&jae	(&label("slow_enc_loop_sse"));
+	&test	($s2,15);
+	&jnz	(&label("slow_enc_tail"));
+	&mov	($acc,$_ivp);		# load ivp
+	&movq	(&QWP(0,$acc),"mm0");	# save ivec
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+					}
+    &set_label("slow_enc_x86",16);
+	&mov	($s0,&DWP(0,$key));	# load iv
+	&mov	($s1,&DWP(4,$key));
+
+	&set_label("slow_enc_loop_x86",4);
+		&mov	($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,$_key);		# load key
+		&call	("_x86_AES_encrypt_compact");
+
+		&mov	($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,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&cmp	($s2,16);
+		&mov	($_len,$s2);		# save len
+	&jae	(&label("slow_enc_loop_x86"));
+	&test	($s2,15);
+	&jnz	(&label("slow_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);
+
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_enc_tail",16);
+	&emms	()	if (!$x86only);
+	&mov	($key eq "edi"? $key:"",$s3);	# load out to edi
+	&mov	($s1,16);
+	&sub	($s1,$s2);
+	&cmp	($key,$acc eq "esi"? $acc:"");	# compare with inp
+	&je	(&label("enc_in_place"));
+	&align	(4);
+	&data_word(0xA4F3F689);	# rep movsb	# copy input
+	&jmp	(&label("enc_skip_in_place"));
+    &set_label("enc_in_place");
+	&lea	($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
+
+	&mov	($key,$_ivp);			# restore ivp
+	&mov	($acc,$s3);			# output as input
+	&mov	($s0,&DWP(0,$key));
+	&mov	($s1,&DWP(4,$key));
+	&mov	($_len,16);			# len=16
+	&jmp	(&label("slow_enc_loop_x86"));	# one more spin...
+
+#--------------------------- SLOW DECRYPT ---------------------------#
+&set_label("slow_decrypt",16);
+					if (!$x86only) {
+	&bt	($_tmp,25);		# check for SSE bit
+	&jnc	(&label("slow_dec_loop_x86"));
+
+	&set_label("slow_dec_loop_sse",4);
+		&movq	("mm0",&QWP(0,$acc));	# read input
+		&movq	("mm4",&QWP(8,$acc));
+
+		&mov	($key,$_key);
+		&call	("_sse_AES_decrypt_compact");
+
+		&mov	($acc,$_inp);		# load inp
+		&lea	($s0,$ivec);
+		&mov	($s1,$_out);		# load out
+		&mov	($s2,$_len);		# load len
+		&mov	($key,$_ivp);		# load ivp
+
+		&movq	("mm1",&QWP(0,$acc));	# re-read input
+		&movq	("mm5",&QWP(8,$acc));
+
+		&pxor	("mm0",&QWP(0,$key));	# xor iv
+		&pxor	("mm4",&QWP(8,$key));
+
+		&movq	(&QWP(0,$key),"mm1");	# copy input to iv
+		&movq	(&QWP(8,$key),"mm5");
+
+		&sub	($s2,16);		# decrease len
+		&jc	(&label("slow_dec_partial_sse"));
+
+		&movq	(&QWP(0,$s1),"mm0");	# write output
+		&movq	(&QWP(8,$s1),"mm4");
+
+		&lea	($s1,&DWP(16,$s1));	# advance out
+		&mov	($_out,$s1);		# save out
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("slow_dec_loop_sse"));
+	&emms	();
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_dec_partial_sse",16);
+	&movq	(&QWP(0,$s0),"mm0");	# save output to temp
+	&movq	(&QWP(8,$s0),"mm4");
+	&emms	();
+
+	&add	($s2 eq "ecx" ? "ecx":"",16);
+	&mov	("edi",$s1);		# out
+	&mov	("esi",$s0);		# temp
+	&align	(4);
+	&data_word(0xA4F3F689);		# rep movsb # copy partial output
+
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+					}
+	&set_label("slow_dec_loop_x86",16);
+		&mov	($s0,&DWP(0,$acc));	# read input
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&lea	($key,$ivec);
+		&mov	(&DWP(0,$key),$s0);	# copy to temp
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_decrypt_compact");
+
+		&mov	($key,$_ivp);		# load ivp
+		&mov	($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("slow_dec_partial_x86"));
+
+		&mov	($_len,$acc);		# save len
+		&mov	($acc,$_out);		# load out
+
+		&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));	# advance out
+		&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 it to 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));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+	&jnz	(&label("slow_dec_loop_x86"));
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_dec_partial_x86",16);
+	&lea	($acc,$ivec);
+	&mov	(&DWP(0,$acc),$s0);	# save output to temp
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+
+	&mov	($acc,$_inp);
+	&mov	($s0,&DWP(0,$acc));	# re-read input
+	&mov	($s1,&DWP(4,$acc));
+	&mov	($s2,&DWP(8,$acc));
+	&mov	($s3,&DWP(12,$acc));
+
+	&mov	(&DWP(0,$key),$s0);	# copy it to iv
+	&mov	(&DWP(4,$key),$s1);
+	&mov	(&DWP(8,$key),$s2);
+	&mov	(&DWP(12,$key),$s3);
+
+	&mov	("ecx",$_len);
+	&mov	("edi",$_out);
+	&lea	("esi",$ivec);
+	&align	(4);
+	&data_word(0xA4F3F689);		# rep movsb # copy partial output
+
+	&mov	("esp",$_esp);
+	&popf	();
+&function_end("asm_AES_cbc_encrypt");
+}
+
+#------------------------------------------------------------------#
+
+sub enckey()
+{
+	&movz	("esi",&LB("edx"));		# rk[i]>>0
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&movz	("esi",&HB("edx"));		# rk[i]>>8
+	&shl	("ebx",24);
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&shr	("edx",16);
+	&movz	("esi",&LB("edx"));		# rk[i]>>16
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&movz	("esi",&HB("edx"));		# rk[i]>>24
+	&shl	("ebx",8);
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&shl	("ebx",16);
+	&xor	("eax","ebx");
+
+	&xor	("eax",&DWP(1024-128,$tbl,"ecx",4));	# rcon
+}
+
+&function_begin("_x86_AES_set_encrypt_key");
+	&mov	("esi",&wparam(1));		# user supplied key
+	&mov	("edi",&wparam(3));		# private key schedule
+
+	&test	("esi",-1);
+	&jz	(&label("badpointer"));
+	&test	("edi",-1);
+	&jz	(&label("badpointer"));
+
+	&call	(&label("pic_point"));
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&lea	($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+	&lea	($tbl,&DWP(2048+128,$tbl));
+
+	# prefetch Te4
+	&mov	("eax",&DWP(0-128,$tbl));
+	&mov	("ebx",&DWP(32-128,$tbl));
+	&mov	("ecx",&DWP(64-128,$tbl));
+	&mov	("edx",&DWP(96-128,$tbl));
+	&mov	("eax",&DWP(128-128,$tbl));
+	&mov	("ebx",&DWP(160-128,$tbl));
+	&mov	("ecx",&DWP(192-128,$tbl));
+	&mov	("edx",&DWP(224-128,$tbl));
+
+	&mov	("ecx",&wparam(2));		# 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
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&movz	("esi",&HB("edx"));		# rk[11]>>8
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&shr	("edx",16);
+		&shl	("ebx",8);
+		&movz	("esi",&LB("edx"));		# rk[11]>>16
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&movz	("esi",&HB("edx"));		# rk[11]>>24
+		&shl	("ebx",16);
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&shl	("ebx",24);
+		&xor	("eax","ebx");
+
+		&mov	(&DWP(48,"edi"),"eax");		# rk[12]
+		&xor	("eax",&DWP(20,"edi"));
+		&mov	(&DWP(52,"edi"),"eax");		# rk[13]
+		&xor	("eax",&DWP(24,"edi"));
+		&mov	(&DWP(56,"edi"),"eax");		# rk[14]
+		&xor	("eax",&DWP(28,"edi"));
+		&mov	(&DWP(60,"edi"),"eax");		# rk[15]
+
+		&add	("edi",32);
+	&jmp	(&label("14loop"));
+
+	&set_label("14break");
+	&mov	(&DWP(48,"edi"),14);		# setup number of rounds
+	&xor	("eax","eax");
+	&jmp	(&label("exit"));
+
+    &set_label("badpointer");
+	&mov	("eax",-1);
+    &set_label("exit");
+&function_end("_x86_AES_set_encrypt_key");
+
+# int asm_AES_set_encrypt_key(const unsigned char *userKey, const int bits,
+#                             AES_KEY *key)
+&function_begin_B("asm_AES_set_encrypt_key");
+	&call	("_x86_AES_set_encrypt_key");
+	&ret	();
+&function_end_B("asm_AES_set_encrypt_key");
+
+sub deckey()
+{ my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
+  my $tmp = $tbl;
+
+	&mov	($tmp,0x80808080);
+	&and	($tmp,$tp1);
+	&lea	($tp2,&DWP(0,$tp1,$tp1));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&sub	($acc,$tmp);
+	&and	($tp2,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp2);
+	&lea	($tp4,&DWP(0,$tp2,$tp2));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&sub	($acc,$tmp);
+	&and	($tp4,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp2,$tp1);	# tp2^tp1
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp4);
+	&lea	($tp8,&DWP(0,$tp4,$tp4));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	 &xor	($tp4,$tp1);	# tp4^tp1
+	&sub	($acc,$tmp);
+	&and	($tp8,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &rotl	($tp1,8);	# = ROTATE(tp1,8)
+	&xor	($tp8,$acc);
+
+	&mov	($tmp,&DWP(4*($i+1),$key));	# modulo-scheduled load
+
+	&xor	($tp1,$tp2);
+	&xor	($tp2,$tp8);
+	&xor	($tp1,$tp4);
+	&rotl	($tp2,24);
+	&xor	($tp4,$tp8);
+	&xor	($tp1,$tp8);	# ^= tp8^(tp4^tp1)^(tp2^tp1)
+	&rotl	($tp4,16);
+	&xor	($tp1,$tp2);	# ^= ROTATE(tp8^tp2^tp1,24)
+	&rotl	($tp8,8);
+	&xor	($tp1,$tp4);	# ^= ROTATE(tp8^tp4^tp1,16)
+	&mov	($tp2,$tmp);
+	&xor	($tp1,$tp8);	# ^= ROTATE(tp8,8)
+
+	&mov	(&DWP(4*$i,$key),$tp1);
+}
+
+# int asm_AES_set_decrypt_key(const unsigned char *userKey, const int bits,
+#                             AES_KEY *key)
+&function_begin_B("asm_AES_set_decrypt_key");
+	&call	("_x86_AES_set_encrypt_key");
+	&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
+
+	&set_label("invert",4);			# 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"));
+
+	&mov	($key,&wparam(2));
+	&mov	($acc,&DWP(240,$key));		# pull number of rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	(&wparam(2),$acc);
+
+	&mov	($s0,&DWP(16,$key));		# modulo-scheduled load
+	&set_label("permute",4);		# permute the key schedule
+		&add	($key,16);
+		&deckey	(0,$key,$s0,$s1,$s2,$s3);
+		&deckey	(1,$key,$s1,$s2,$s3,$s0);
+		&deckey	(2,$key,$s2,$s3,$s0,$s1);
+		&deckey	(3,$key,$s3,$s0,$s1,$s2);
+		&cmp	($key,&wparam(2));
+	&jb	(&label("permute"));
+
+	&xor	("eax","eax");			# return success
+&function_end("asm_AES_set_decrypt_key");
+&asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
+
+&asm_finish();
diff --git a/src/crypto/aes/asm/aes-armv4.pl b/src/crypto/aes/asm/aes-armv4.pl
new file mode 100644
index 0000000..3bd9a6d
--- /dev/null
+++ b/src/crypto/aes/asm/aes-armv4.pl
@@ -0,0 +1,1224 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> 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/.
+# ====================================================================
+
+# AES for ARMv4
+
+# January 2007.
+#
+# Code uses single 1K S-box and is >2 times faster than code generated
+# by gcc-3.4.1. This is thanks to unique feature of ARMv4 ISA, which
+# allows to merge logical or arithmetic operation with shift or rotate
+# in one instruction and emit combined result every cycle. The module
+# is endian-neutral. The performance is ~42 cycles/byte for 128-bit
+# key [on single-issue Xscale PXA250 core].
+
+# May 2007.
+#
+# AES_set_[en|de]crypt_key is added.
+
+# July 2010.
+#
+# Rescheduling for dual-issue pipeline resulted in 12% improvement on
+# Cortex A8 core and ~25 cycles per byte processed with 128-bit key.
+
+# February 2011.
+#
+# Profiler-assisted and platform-specific optimization resulted in 16%
+# improvement on Cortex A8 core and ~21.5 cycles per byte.
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
+$s0="r0";
+$s1="r1";
+$s2="r2";
+$s3="r3";
+$t1="r4";
+$t2="r5";
+$t3="r6";
+$i1="r7";
+$i2="r8";
+$i3="r9";
+
+$tbl="r10";
+$key="r11";
+$rounds="r12";
+
+$code=<<___;
+#if defined(__arm__)
+#ifndef __KERNEL__
+# include "arm_arch.h"
+#else
+# define __ARM_ARCH__ __LINUX_ARM_ARCH__
+#endif
+
+.text
+#if __ARM_ARCH__<7
+.code	32
+#else
+.syntax	unified
+# ifdef __thumb2__
+.thumb
+# else
+.code	32
+# endif
+#endif
+
+.type	AES_Te,%object
+.align	5
+AES_Te:
+.word	0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d
+.word	0xfff2f20d, 0xd66b6bbd, 0xde6f6fb1, 0x91c5c554
+.word	0x60303050, 0x02010103, 0xce6767a9, 0x562b2b7d
+.word	0xe7fefe19, 0xb5d7d762, 0x4dababe6, 0xec76769a
+.word	0x8fcaca45, 0x1f82829d, 0x89c9c940, 0xfa7d7d87
+.word	0xeffafa15, 0xb25959eb, 0x8e4747c9, 0xfbf0f00b
+.word	0x41adadec, 0xb3d4d467, 0x5fa2a2fd, 0x45afafea
+.word	0x239c9cbf, 0x53a4a4f7, 0xe4727296, 0x9bc0c05b
+.word	0x75b7b7c2, 0xe1fdfd1c, 0x3d9393ae, 0x4c26266a
+.word	0x6c36365a, 0x7e3f3f41, 0xf5f7f702, 0x83cccc4f
+.word	0x6834345c, 0x51a5a5f4, 0xd1e5e534, 0xf9f1f108
+.word	0xe2717193, 0xabd8d873, 0x62313153, 0x2a15153f
+.word	0x0804040c, 0x95c7c752, 0x46232365, 0x9dc3c35e
+.word	0x30181828, 0x379696a1, 0x0a05050f, 0x2f9a9ab5
+.word	0x0e070709, 0x24121236, 0x1b80809b, 0xdfe2e23d
+.word	0xcdebeb26, 0x4e272769, 0x7fb2b2cd, 0xea75759f
+.word	0x1209091b, 0x1d83839e, 0x582c2c74, 0x341a1a2e
+.word	0x361b1b2d, 0xdc6e6eb2, 0xb45a5aee, 0x5ba0a0fb
+.word	0xa45252f6, 0x763b3b4d, 0xb7d6d661, 0x7db3b3ce
+.word	0x5229297b, 0xdde3e33e, 0x5e2f2f71, 0x13848497
+.word	0xa65353f5, 0xb9d1d168, 0x00000000, 0xc1eded2c
+.word	0x40202060, 0xe3fcfc1f, 0x79b1b1c8, 0xb65b5bed
+.word	0xd46a6abe, 0x8dcbcb46, 0x67bebed9, 0x7239394b
+.word	0x944a4ade, 0x984c4cd4, 0xb05858e8, 0x85cfcf4a
+.word	0xbbd0d06b, 0xc5efef2a, 0x4faaaae5, 0xedfbfb16
+.word	0x864343c5, 0x9a4d4dd7, 0x66333355, 0x11858594
+.word	0x8a4545cf, 0xe9f9f910, 0x04020206, 0xfe7f7f81
+.word	0xa05050f0, 0x783c3c44, 0x259f9fba, 0x4ba8a8e3
+.word	0xa25151f3, 0x5da3a3fe, 0x804040c0, 0x058f8f8a
+.word	0x3f9292ad, 0x219d9dbc, 0x70383848, 0xf1f5f504
+.word	0x63bcbcdf, 0x77b6b6c1, 0xafdada75, 0x42212163
+.word	0x20101030, 0xe5ffff1a, 0xfdf3f30e, 0xbfd2d26d
+.word	0x81cdcd4c, 0x180c0c14, 0x26131335, 0xc3ecec2f
+.word	0xbe5f5fe1, 0x359797a2, 0x884444cc, 0x2e171739
+.word	0x93c4c457, 0x55a7a7f2, 0xfc7e7e82, 0x7a3d3d47
+.word	0xc86464ac, 0xba5d5de7, 0x3219192b, 0xe6737395
+.word	0xc06060a0, 0x19818198, 0x9e4f4fd1, 0xa3dcdc7f
+.word	0x44222266, 0x542a2a7e, 0x3b9090ab, 0x0b888883
+.word	0x8c4646ca, 0xc7eeee29, 0x6bb8b8d3, 0x2814143c
+.word	0xa7dede79, 0xbc5e5ee2, 0x160b0b1d, 0xaddbdb76
+.word	0xdbe0e03b, 0x64323256, 0x743a3a4e, 0x140a0a1e
+.word	0x924949db, 0x0c06060a, 0x4824246c, 0xb85c5ce4
+.word	0x9fc2c25d, 0xbdd3d36e, 0x43acacef, 0xc46262a6
+.word	0x399191a8, 0x319595a4, 0xd3e4e437, 0xf279798b
+.word	0xd5e7e732, 0x8bc8c843, 0x6e373759, 0xda6d6db7
+.word	0x018d8d8c, 0xb1d5d564, 0x9c4e4ed2, 0x49a9a9e0
+.word	0xd86c6cb4, 0xac5656fa, 0xf3f4f407, 0xcfeaea25
+.word	0xca6565af, 0xf47a7a8e, 0x47aeaee9, 0x10080818
+.word	0x6fbabad5, 0xf0787888, 0x4a25256f, 0x5c2e2e72
+.word	0x381c1c24, 0x57a6a6f1, 0x73b4b4c7, 0x97c6c651
+.word	0xcbe8e823, 0xa1dddd7c, 0xe874749c, 0x3e1f1f21
+.word	0x964b4bdd, 0x61bdbddc, 0x0d8b8b86, 0x0f8a8a85
+.word	0xe0707090, 0x7c3e3e42, 0x71b5b5c4, 0xcc6666aa
+.word	0x904848d8, 0x06030305, 0xf7f6f601, 0x1c0e0e12
+.word	0xc26161a3, 0x6a35355f, 0xae5757f9, 0x69b9b9d0
+.word	0x17868691, 0x99c1c158, 0x3a1d1d27, 0x279e9eb9
+.word	0xd9e1e138, 0xebf8f813, 0x2b9898b3, 0x22111133
+.word	0xd26969bb, 0xa9d9d970, 0x078e8e89, 0x339494a7
+.word	0x2d9b9bb6, 0x3c1e1e22, 0x15878792, 0xc9e9e920
+.word	0x87cece49, 0xaa5555ff, 0x50282878, 0xa5dfdf7a
+.word	0x038c8c8f, 0x59a1a1f8, 0x09898980, 0x1a0d0d17
+.word	0x65bfbfda, 0xd7e6e631, 0x844242c6, 0xd06868b8
+.word	0x824141c3, 0x299999b0, 0x5a2d2d77, 0x1e0f0f11
+.word	0x7bb0b0cb, 0xa85454fc, 0x6dbbbbd6, 0x2c16163a
+@ Te4[256]
+.byte	0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5
+.byte	0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76
+.byte	0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0
+.byte	0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0
+.byte	0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc
+.byte	0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15
+.byte	0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a
+.byte	0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75
+.byte	0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0
+.byte	0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84
+.byte	0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b
+.byte	0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf
+.byte	0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85
+.byte	0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8
+.byte	0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5
+.byte	0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2
+.byte	0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17
+.byte	0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73
+.byte	0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88
+.byte	0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb
+.byte	0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c
+.byte	0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79
+.byte	0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9
+.byte	0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08
+.byte	0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6
+.byte	0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a
+.byte	0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e
+.byte	0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e
+.byte	0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94
+.byte	0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf
+.byte	0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68
+.byte	0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16
+@ rcon[]
+.word	0x01000000, 0x02000000, 0x04000000, 0x08000000
+.word	0x10000000, 0x20000000, 0x40000000, 0x80000000
+.word	0x1B000000, 0x36000000, 0, 0, 0, 0, 0, 0
+.size	AES_Te,.-AES_Te
+
+@ void asm_AES_encrypt(const unsigned char *in, unsigned char *out,
+@ 		       const AES_KEY *key) {
+.global asm_AES_encrypt
+.hidden asm_AES_encrypt
+.type   asm_AES_encrypt,%function
+.align	5
+asm_AES_encrypt:
+#if __ARM_ARCH__<7
+	sub	r3,pc,#8		@ asm_AES_encrypt
+#else
+	adr	r3,asm_AES_encrypt
+#endif
+	stmdb   sp!,{r1,r4-r12,lr}
+	mov	$rounds,r0		@ inp
+	mov	$key,r2
+	sub	$tbl,r3,#asm_AES_encrypt-AES_Te	@ Te
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	orr	$s3,$s3,$t2,lsl#16
+	orr	$s3,$s3,$t3,lsl#24
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
+	bl	_armv4_AES_encrypt
+
+	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
+	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
+	mov	$t2,$s0,lsr#16		@ manner...
+	mov	$t3,$s0,lsr#8
+	strb	$t1,[$rounds,#0]
+	strb	$t2,[$rounds,#1]
+	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
+	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
+	mov	$t3,$s1,lsr#8
+	strb	$t1,[$rounds,#4]
+	strb	$t2,[$rounds,#5]
+	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
+	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
+	mov	$t3,$s2,lsr#8
+	strb	$t1,[$rounds,#8]
+	strb	$t2,[$rounds,#9]
+	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
+	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
+	mov	$t3,$s3,lsr#8
+	strb	$t1,[$rounds,#12]
+	strb	$t2,[$rounds,#13]
+	strb	$t3,[$rounds,#14]
+	strb	$s3,[$rounds,#15]
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	asm_AES_encrypt,.-asm_AES_encrypt
+
+.type   _armv4_AES_encrypt,%function
+.align	2
+_armv4_AES_encrypt:
+	str	lr,[sp,#-4]!		@ push lr
+	ldmia	$key!,{$t1-$i1}
+	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
+	eor	$s1,$s1,$t2
+	eor	$s2,$s2,$t3
+	eor	$s3,$s3,$i1
+	sub	$rounds,$rounds,#1
+	mov	lr,#255
+
+	and	$i1,lr,$s0
+	and	$i2,lr,$s0,lsr#8
+	and	$i3,lr,$s0,lsr#16
+	mov	$s0,$s0,lsr#24
+.Lenc_loop:
+	ldr	$t1,[$tbl,$i1,lsl#2]	@ Te3[s0>>0]
+	and	$i1,lr,$s1,lsr#16	@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Te2[s0>>8]
+	and	$i2,lr,$s1
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Te1[s0>>16]
+	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Te0[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te1[s1>>16]
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te3[s1>>0]
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te2[s1>>8]
+	eor	$s0,$s0,$i1,ror#8
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Te0[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,ror#8
+	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$t3,$i3,ror#8
+	and	$i3,lr,$s2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te2[s2>>8]
+	eor	$s1,$s1,$t1,ror#24
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te1[s2>>16]
+	mov	$s2,$s2,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te3[s2>>0]
+	eor	$s0,$s0,$i1,ror#16
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Te0[s2>>24]
+	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,ror#8
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$t3,$i3,ror#16
+	and	$i3,lr,$s3,lsr#16	@ i2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te3[s3>>0]
+	eor	$s2,$s2,$t2,ror#16
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te1[s3>>16]
+	eor	$s0,$s0,$i1,ror#24
+	ldr	$i1,[$key],#16
+	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Te0[s3>>24]
+	eor	$s2,$s2,$i3,ror#8
+	ldr	$t1,[$key,#-12]
+	eor	$s3,$s3,$t3,ror#8
+
+	ldr	$t2,[$key,#-8]
+	eor	$s0,$s0,$i1
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0,lsr#16
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
+
+	subs	$rounds,$rounds,#1
+	bne	.Lenc_loop
+
+	add	$tbl,$tbl,#2
+
+	ldrb	$t1,[$tbl,$i1,lsl#2]	@ Te4[s0>>0]
+	and	$i1,lr,$s1,lsr#16	@ i0
+	ldrb	$t2,[$tbl,$i2,lsl#2]	@ Te4[s0>>8]
+	and	$i2,lr,$s1
+	ldrb	$t3,[$tbl,$i3,lsl#2]	@ Te4[s0>>16]
+	and	$i3,lr,$s1,lsr#8
+	ldrb	$s0,[$tbl,$s0,lsl#2]	@ Te4[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s1>>16]
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s1>>0]
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s1>>8]
+	eor	$s0,$i1,$s0,lsl#8
+	ldrb	$s1,[$tbl,$s1,lsl#2]	@ Te4[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,lsl#8
+	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$i3,$t3,lsl#8
+	and	$i3,lr,$s2
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s2>>8]
+	eor	$s1,$t1,$s1,lsl#24
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s2>>16]
+	mov	$s2,$s2,lsr#24
+
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s2>>0]
+	eor	$s0,$i1,$s0,lsl#8
+	ldrb	$s2,[$tbl,$s2,lsl#2]	@ Te4[s2>>24]
+	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,lsl#16
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,lsl#8
+	and	$i3,lr,$s3,lsr#16	@ i2
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s3>>0]
+	eor	$s2,$t2,$s2,lsl#24
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s3>>16]
+	eor	$s0,$i1,$s0,lsl#8
+	ldr	$i1,[$key,#0]
+	ldrb	$s3,[$tbl,$s3,lsl#2]	@ Te4[s3>>24]
+	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
+	eor	$s2,$s2,$i3,lsl#16
+	ldr	$t2,[$key,#8]
+	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
+
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
+
+	sub	$tbl,$tbl,#2
+	ldr	pc,[sp],#4		@ pop and return
+.size	_armv4_AES_encrypt,.-_armv4_AES_encrypt
+
+.global asm_AES_set_encrypt_key
+.hidden asm_AES_set_encrypt_key
+.type   asm_AES_set_encrypt_key,%function
+.align	5
+asm_AES_set_encrypt_key:
+_armv4_AES_set_encrypt_key:
+#if __ARM_ARCH__<7
+	sub	r3,pc,#8		@ asm_AES_set_encrypt_key
+#else
+	adr	r3,asm_AES_set_encrypt_key
+#endif
+	teq	r0,#0
+#if __ARM_ARCH__>=7
+	itt	eq			@ Thumb2 thing, sanity check in ARM
+#endif
+	moveq	r0,#-1
+	beq	.Labrt
+	teq	r2,#0
+#if __ARM_ARCH__>=7
+	itt	eq			@ Thumb2 thing, sanity check in ARM
+#endif
+	moveq	r0,#-1
+	beq	.Labrt
+
+	teq	r1,#128
+	beq	.Lok
+	teq	r1,#192
+	beq	.Lok
+	teq	r1,#256
+#if __ARM_ARCH__>=7
+	itt	ne			@ Thumb2 thing, sanity check in ARM
+#endif
+	movne	r0,#-1
+	bne	.Labrt
+
+.Lok:	stmdb   sp!,{r4-r12,lr}
+	sub	$tbl,r3,#_armv4_AES_set_encrypt_key-AES_Te-1024	@ Te4
+
+	mov	$rounds,r0		@ inp
+	mov	lr,r1			@ bits
+	mov	$key,r2			@ key
+
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	str	$s0,[$key],#16
+	orr	$s3,$s3,$t2,lsl#16
+	str	$s1,[$key,#-12]
+	orr	$s3,$s3,$t3,lsl#24
+	str	$s2,[$key,#-8]
+	str	$s3,[$key,#-4]
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$key],#16
+	str	$s1,[$key,#-12]
+	str	$s2,[$key,#-8]
+	str	$s3,[$key,#-4]
+#endif
+
+	teq	lr,#128
+	bne	.Lnot128
+	mov	$rounds,#10
+	str	$rounds,[$key,#240-16]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+
+.L128_loop:
+	and	$t2,lr,$s3,lsr#24
+	and	$i1,lr,$s3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$t2,$t2,$t1
+	eor	$s0,$s0,$t2			@ rk[4]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[5]=rk[1]^rk[4]
+	str	$s0,[$key],#16
+	eor	$s2,$s2,$s1			@ rk[6]=rk[2]^rk[5]
+	str	$s1,[$key,#-12]
+	eor	$s3,$s3,$s2			@ rk[7]=rk[3]^rk[6]
+	str	$s2,[$key,#-8]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-4]
+	bne	.L128_loop
+	sub	r2,$key,#176
+	b	.Ldone
+
+.Lnot128:
+#if __ARM_ARCH__<7
+	ldrb	$i2,[$rounds,#19]
+	ldrb	$t1,[$rounds,#18]
+	ldrb	$t2,[$rounds,#17]
+	ldrb	$t3,[$rounds,#16]
+	orr	$i2,$i2,$t1,lsl#8
+	ldrb	$i3,[$rounds,#23]
+	orr	$i2,$i2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#22]
+	orr	$i2,$i2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#21]
+	ldrb	$t3,[$rounds,#20]
+	orr	$i3,$i3,$t1,lsl#8
+	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
+	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#16]
+	ldr	$i3,[$rounds,#20]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
+	str	$i2,[$key],#8
+	str	$i3,[$key,#-4]
+#endif
+
+	teq	lr,#192
+	bne	.Lnot192
+	mov	$rounds,#12
+	str	$rounds,[$key,#240-24]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+	mov	$rounds,#8
+
+.L192_loop:
+	and	$t2,lr,$i3,lsr#24
+	and	$i1,lr,$i3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$i3,$t2,$t1
+	eor	$s0,$s0,$i3			@ rk[6]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[7]=rk[1]^rk[6]
+	str	$s0,[$key],#24
+	eor	$s2,$s2,$s1			@ rk[8]=rk[2]^rk[7]
+	str	$s1,[$key,#-20]
+	eor	$s3,$s3,$s2			@ rk[9]=rk[3]^rk[8]
+	str	$s2,[$key,#-16]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-12]
+#if __ARM_ARCH__>=7
+	itt	eq				@ Thumb2 thing, sanity check in ARM
+#endif
+	subeq	r2,$key,#216
+	beq	.Ldone
+
+	ldr	$i1,[$key,#-32]
+	ldr	$i2,[$key,#-28]
+	eor	$i1,$i1,$s3			@ rk[10]=rk[4]^rk[9]
+	eor	$i3,$i2,$i1			@ rk[11]=rk[5]^rk[10]
+	str	$i1,[$key,#-8]
+	str	$i3,[$key,#-4]
+	b	.L192_loop
+
+.Lnot192:
+#if __ARM_ARCH__<7
+	ldrb	$i2,[$rounds,#27]
+	ldrb	$t1,[$rounds,#26]
+	ldrb	$t2,[$rounds,#25]
+	ldrb	$t3,[$rounds,#24]
+	orr	$i2,$i2,$t1,lsl#8
+	ldrb	$i3,[$rounds,#31]
+	orr	$i2,$i2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#30]
+	orr	$i2,$i2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#29]
+	ldrb	$t3,[$rounds,#28]
+	orr	$i3,$i3,$t1,lsl#8
+	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
+	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#24]
+	ldr	$i3,[$rounds,#28]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
+	str	$i2,[$key],#8
+	str	$i3,[$key,#-4]
+#endif
+
+	mov	$rounds,#14
+	str	$rounds,[$key,#240-32]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+	mov	$rounds,#7
+
+.L256_loop:
+	and	$t2,lr,$i3,lsr#24
+	and	$i1,lr,$i3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$i3,$t2,$t1
+	eor	$s0,$s0,$i3			@ rk[8]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[9]=rk[1]^rk[8]
+	str	$s0,[$key],#32
+	eor	$s2,$s2,$s1			@ rk[10]=rk[2]^rk[9]
+	str	$s1,[$key,#-28]
+	eor	$s3,$s3,$s2			@ rk[11]=rk[3]^rk[10]
+	str	$s2,[$key,#-24]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-20]
+#if __ARM_ARCH__>=7
+	itt	eq				@ Thumb2 thing, sanity check in ARM
+#endif
+	subeq	r2,$key,#256
+	beq	.Ldone
+
+	and	$t2,lr,$s3
+	and	$i1,lr,$s3,lsr#8
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#16
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3,lsr#24
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$key,#-48]
+	orr	$t2,$t2,$i3,lsl#24
+
+	ldr	$i1,[$key,#-44]
+	ldr	$i2,[$key,#-40]
+	eor	$t1,$t1,$t2			@ rk[12]=rk[4]^...
+	ldr	$i3,[$key,#-36]
+	eor	$i1,$i1,$t1			@ rk[13]=rk[5]^rk[12]
+	str	$t1,[$key,#-16]
+	eor	$i2,$i2,$i1			@ rk[14]=rk[6]^rk[13]
+	str	$i1,[$key,#-12]
+	eor	$i3,$i3,$i2			@ rk[15]=rk[7]^rk[14]
+	str	$i2,[$key,#-8]
+	str	$i3,[$key,#-4]
+	b	.L256_loop
+
+.align	2
+.Ldone:	mov	r0,#0
+	ldmia   sp!,{r4-r12,lr}
+.Labrt:
+#if defined(__thumb2__) && __ARM_ARCH__>=7
+	.short	0x4770			@ bx lr in Thumb2 encoding
+#else
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	asm_AES_set_encrypt_key,.-asm_AES_set_encrypt_key
+
+.global asm_AES_set_decrypt_key
+.hidden asm_AES_set_decrypt_key
+.type   asm_AES_set_decrypt_key,%function
+.align	5
+asm_AES_set_decrypt_key:
+	str	lr,[sp,#-4]!            @ push lr
+	bl	_armv4_AES_set_encrypt_key
+	teq	r0,#0
+	ldr	lr,[sp],#4              @ pop lr
+	bne	.Labrt
+
+	mov	r0,r2			@ asm_AES_set_encrypt_key preserves r2,
+	mov	r1,r2			@ which is AES_KEY *key
+	b	_armv4_AES_set_enc2dec_key
+.size	asm_AES_set_decrypt_key,.-asm_AES_set_decrypt_key
+
+@ void AES_set_enc2dec_key(const AES_KEY *inp,AES_KEY *out)
+.global	AES_set_enc2dec_key
+.hidden	AES_set_enc2dec_key
+.type	AES_set_enc2dec_key,%function
+.align	5
+AES_set_enc2dec_key:
+_armv4_AES_set_enc2dec_key:
+	stmdb   sp!,{r4-r12,lr}
+
+	ldr	$rounds,[r0,#240]
+	mov	$i1,r0			@ input
+	add	$i2,r0,$rounds,lsl#4
+	mov	$key,r1			@ ouput
+	add	$tbl,r1,$rounds,lsl#4
+	str	$rounds,[r1,#240]
+
+.Linv:	ldr	$s0,[$i1],#16
+	ldr	$s1,[$i1,#-12]
+	ldr	$s2,[$i1,#-8]
+	ldr	$s3,[$i1,#-4]
+	ldr	$t1,[$i2],#-16
+	ldr	$t2,[$i2,#16+4]
+	ldr	$t3,[$i2,#16+8]
+	ldr	$i3,[$i2,#16+12]
+	str	$s0,[$tbl],#-16
+	str	$s1,[$tbl,#16+4]
+	str	$s2,[$tbl,#16+8]
+	str	$s3,[$tbl,#16+12]
+	str	$t1,[$key],#16
+	str	$t2,[$key,#-12]
+	str	$t3,[$key,#-8]
+	str	$i3,[$key,#-4]
+	teq	$i1,$i2
+	bne	.Linv
+
+	ldr	$s0,[$i1]
+	ldr	$s1,[$i1,#4]
+	ldr	$s2,[$i1,#8]
+	ldr	$s3,[$i1,#12]
+	str	$s0,[$key]
+	str	$s1,[$key,#4]
+	str	$s2,[$key,#8]
+	str	$s3,[$key,#12]
+	sub	$key,$key,$rounds,lsl#3
+___
+$mask80=$i1;
+$mask1b=$i2;
+$mask7f=$i3;
+$code.=<<___;
+	ldr	$s0,[$key,#16]!		@ prefetch tp1
+	mov	$mask80,#0x80
+	mov	$mask1b,#0x1b
+	orr	$mask80,$mask80,#0x8000
+	orr	$mask1b,$mask1b,#0x1b00
+	orr	$mask80,$mask80,$mask80,lsl#16
+	orr	$mask1b,$mask1b,$mask1b,lsl#16
+	sub	$rounds,$rounds,#1
+	mvn	$mask7f,$mask80
+	mov	$rounds,$rounds,lsl#2	@ (rounds-1)*4
+
+.Lmix:	and	$t1,$s0,$mask80
+	and	$s1,$s0,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s1,$t1,$s1,lsl#1	@ tp2
+
+	and	$t1,$s1,$mask80
+	and	$s2,$s1,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s2,$t1,$s2,lsl#1	@ tp4
+
+	and	$t1,$s2,$mask80
+	and	$s3,$s2,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s3,$t1,$s3,lsl#1	@ tp8
+
+	eor	$t1,$s1,$s2
+	eor	$t2,$s0,$s3		@ tp9
+	eor	$t1,$t1,$s3		@ tpe
+	eor	$t1,$t1,$s1,ror#24
+	eor	$t1,$t1,$t2,ror#24	@ ^= ROTATE(tpb=tp9^tp2,8)
+	eor	$t1,$t1,$s2,ror#16
+	eor	$t1,$t1,$t2,ror#16	@ ^= ROTATE(tpd=tp9^tp4,16)
+	eor	$t1,$t1,$t2,ror#8	@ ^= ROTATE(tp9,24)
+
+	ldr	$s0,[$key,#4]		@ prefetch tp1
+	str	$t1,[$key],#4
+	subs	$rounds,$rounds,#1
+	bne	.Lmix
+
+	mov	r0,#0
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	AES_set_enc2dec_key,.-AES_set_enc2dec_key
+
+.type	AES_Td,%object
+.align	5
+AES_Td:
+.word	0x51f4a750, 0x7e416553, 0x1a17a4c3, 0x3a275e96
+.word	0x3bab6bcb, 0x1f9d45f1, 0xacfa58ab, 0x4be30393
+.word	0x2030fa55, 0xad766df6, 0x88cc7691, 0xf5024c25
+.word	0x4fe5d7fc, 0xc52acbd7, 0x26354480, 0xb562a38f
+.word	0xdeb15a49, 0x25ba1b67, 0x45ea0e98, 0x5dfec0e1
+.word	0xc32f7502, 0x814cf012, 0x8d4697a3, 0x6bd3f9c6
+.word	0x038f5fe7, 0x15929c95, 0xbf6d7aeb, 0x955259da
+.word	0xd4be832d, 0x587421d3, 0x49e06929, 0x8ec9c844
+.word	0x75c2896a, 0xf48e7978, 0x99583e6b, 0x27b971dd
+.word	0xbee14fb6, 0xf088ad17, 0xc920ac66, 0x7dce3ab4
+.word	0x63df4a18, 0xe51a3182, 0x97513360, 0x62537f45
+.word	0xb16477e0, 0xbb6bae84, 0xfe81a01c, 0xf9082b94
+.word	0x70486858, 0x8f45fd19, 0x94de6c87, 0x527bf8b7
+.word	0xab73d323, 0x724b02e2, 0xe31f8f57, 0x6655ab2a
+.word	0xb2eb2807, 0x2fb5c203, 0x86c57b9a, 0xd33708a5
+.word	0x302887f2, 0x23bfa5b2, 0x02036aba, 0xed16825c
+.word	0x8acf1c2b, 0xa779b492, 0xf307f2f0, 0x4e69e2a1
+.word	0x65daf4cd, 0x0605bed5, 0xd134621f, 0xc4a6fe8a
+.word	0x342e539d, 0xa2f355a0, 0x058ae132, 0xa4f6eb75
+.word	0x0b83ec39, 0x4060efaa, 0x5e719f06, 0xbd6e1051
+.word	0x3e218af9, 0x96dd063d, 0xdd3e05ae, 0x4de6bd46
+.word	0x91548db5, 0x71c45d05, 0x0406d46f, 0x605015ff
+.word	0x1998fb24, 0xd6bde997, 0x894043cc, 0x67d99e77
+.word	0xb0e842bd, 0x07898b88, 0xe7195b38, 0x79c8eedb
+.word	0xa17c0a47, 0x7c420fe9, 0xf8841ec9, 0x00000000
+.word	0x09808683, 0x322bed48, 0x1e1170ac, 0x6c5a724e
+.word	0xfd0efffb, 0x0f853856, 0x3daed51e, 0x362d3927
+.word	0x0a0fd964, 0x685ca621, 0x9b5b54d1, 0x24362e3a
+.word	0x0c0a67b1, 0x9357e70f, 0xb4ee96d2, 0x1b9b919e
+.word	0x80c0c54f, 0x61dc20a2, 0x5a774b69, 0x1c121a16
+.word	0xe293ba0a, 0xc0a02ae5, 0x3c22e043, 0x121b171d
+.word	0x0e090d0b, 0xf28bc7ad, 0x2db6a8b9, 0x141ea9c8
+.word	0x57f11985, 0xaf75074c, 0xee99ddbb, 0xa37f60fd
+.word	0xf701269f, 0x5c72f5bc, 0x44663bc5, 0x5bfb7e34
+.word	0x8b432976, 0xcb23c6dc, 0xb6edfc68, 0xb8e4f163
+.word	0xd731dcca, 0x42638510, 0x13972240, 0x84c61120
+.word	0x854a247d, 0xd2bb3df8, 0xaef93211, 0xc729a16d
+.word	0x1d9e2f4b, 0xdcb230f3, 0x0d8652ec, 0x77c1e3d0
+.word	0x2bb3166c, 0xa970b999, 0x119448fa, 0x47e96422
+.word	0xa8fc8cc4, 0xa0f03f1a, 0x567d2cd8, 0x223390ef
+.word	0x87494ec7, 0xd938d1c1, 0x8ccaa2fe, 0x98d40b36
+.word	0xa6f581cf, 0xa57ade28, 0xdab78e26, 0x3fadbfa4
+.word	0x2c3a9de4, 0x5078920d, 0x6a5fcc9b, 0x547e4662
+.word	0xf68d13c2, 0x90d8b8e8, 0x2e39f75e, 0x82c3aff5
+.word	0x9f5d80be, 0x69d0937c, 0x6fd52da9, 0xcf2512b3
+.word	0xc8ac993b, 0x10187da7, 0xe89c636e, 0xdb3bbb7b
+.word	0xcd267809, 0x6e5918f4, 0xec9ab701, 0x834f9aa8
+.word	0xe6956e65, 0xaaffe67e, 0x21bccf08, 0xef15e8e6
+.word	0xbae79bd9, 0x4a6f36ce, 0xea9f09d4, 0x29b07cd6
+.word	0x31a4b2af, 0x2a3f2331, 0xc6a59430, 0x35a266c0
+.word	0x744ebc37, 0xfc82caa6, 0xe090d0b0, 0x33a7d815
+.word	0xf104984a, 0x41ecdaf7, 0x7fcd500e, 0x1791f62f
+.word	0x764dd68d, 0x43efb04d, 0xccaa4d54, 0xe49604df
+.word	0x9ed1b5e3, 0x4c6a881b, 0xc12c1fb8, 0x4665517f
+.word	0x9d5eea04, 0x018c355d, 0xfa877473, 0xfb0b412e
+.word	0xb3671d5a, 0x92dbd252, 0xe9105633, 0x6dd64713
+.word	0x9ad7618c, 0x37a10c7a, 0x59f8148e, 0xeb133c89
+.word	0xcea927ee, 0xb761c935, 0xe11ce5ed, 0x7a47b13c
+.word	0x9cd2df59, 0x55f2733f, 0x1814ce79, 0x73c737bf
+.word	0x53f7cdea, 0x5ffdaa5b, 0xdf3d6f14, 0x7844db86
+.word	0xcaaff381, 0xb968c43e, 0x3824342c, 0xc2a3405f
+.word	0x161dc372, 0xbce2250c, 0x283c498b, 0xff0d9541
+.word	0x39a80171, 0x080cb3de, 0xd8b4e49c, 0x6456c190
+.word	0x7bcb8461, 0xd532b670, 0x486c5c74, 0xd0b85742
+@ Td4[256]
+.byte	0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38
+.byte	0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb
+.byte	0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87
+.byte	0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb
+.byte	0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d
+.byte	0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e
+.byte	0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2
+.byte	0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25
+.byte	0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16
+.byte	0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92
+.byte	0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda
+.byte	0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84
+.byte	0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a
+.byte	0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06
+.byte	0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02
+.byte	0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b
+.byte	0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea
+.byte	0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73
+.byte	0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85
+.byte	0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e
+.byte	0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89
+.byte	0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b
+.byte	0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20
+.byte	0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4
+.byte	0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31
+.byte	0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f
+.byte	0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d
+.byte	0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef
+.byte	0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0
+.byte	0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61
+.byte	0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26
+.byte	0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d
+.size	AES_Td,.-AES_Td
+
+@ void asm_AES_decrypt(const unsigned char *in, unsigned char *out,
+@ 		       const AES_KEY *key) {
+.global asm_AES_decrypt
+.hidden asm_AES_decrypt
+.type   asm_AES_decrypt,%function
+.align	5
+asm_AES_decrypt:
+#if __ARM_ARCH__<7
+	sub	r3,pc,#8		@ asm_AES_decrypt
+#else
+	adr	r3,asm_AES_decrypt
+#endif
+	stmdb   sp!,{r1,r4-r12,lr}
+	mov	$rounds,r0		@ inp
+	mov	$key,r2
+	sub	$tbl,r3,#asm_AES_decrypt-AES_Td		@ Td
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	orr	$s3,$s3,$t2,lsl#16
+	orr	$s3,$s3,$t3,lsl#24
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
+	bl	_armv4_AES_decrypt
+
+	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
+	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
+	mov	$t2,$s0,lsr#16		@ manner...
+	mov	$t3,$s0,lsr#8
+	strb	$t1,[$rounds,#0]
+	strb	$t2,[$rounds,#1]
+	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
+	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
+	mov	$t3,$s1,lsr#8
+	strb	$t1,[$rounds,#4]
+	strb	$t2,[$rounds,#5]
+	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
+	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
+	mov	$t3,$s2,lsr#8
+	strb	$t1,[$rounds,#8]
+	strb	$t2,[$rounds,#9]
+	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
+	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
+	mov	$t3,$s3,lsr#8
+	strb	$t1,[$rounds,#12]
+	strb	$t2,[$rounds,#13]
+	strb	$t3,[$rounds,#14]
+	strb	$s3,[$rounds,#15]
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	asm_AES_decrypt,.-asm_AES_decrypt
+
+.type   _armv4_AES_decrypt,%function
+.align	2
+_armv4_AES_decrypt:
+	str	lr,[sp,#-4]!		@ push lr
+	ldmia	$key!,{$t1-$i1}
+	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
+	eor	$s1,$s1,$t2
+	eor	$s2,$s2,$t3
+	eor	$s3,$s3,$i1
+	sub	$rounds,$rounds,#1
+	mov	lr,#255
+
+	and	$i1,lr,$s0,lsr#16
+	and	$i2,lr,$s0,lsr#8
+	and	$i3,lr,$s0
+	mov	$s0,$s0,lsr#24
+.Ldec_loop:
+	ldr	$t1,[$tbl,$i1,lsl#2]	@ Td1[s0>>16]
+	and	$i1,lr,$s1		@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Td2[s0>>8]
+	and	$i2,lr,$s1,lsr#16
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Td3[s0>>0]
+	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Td0[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td3[s1>>0]
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td1[s1>>16]
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td2[s1>>8]
+	eor	$s0,$s0,$i1,ror#24
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Td0[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,ror#8
+	and	$i2,lr,$s2		@ i1
+	eor	$t3,$i3,$t3,ror#8
+	and	$i3,lr,$s2,lsr#16
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td2[s2>>8]
+	eor	$s1,$s1,$t1,ror#8
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td3[s2>>0]
+	mov	$s2,$s2,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td1[s2>>16]
+	eor	$s0,$s0,$i1,ror#16
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Td0[s2>>24]
+	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s1,$s1,$i2,ror#24
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,ror#8
+	and	$i3,lr,$s3		@ i2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td1[s3>>16]
+	eor	$s2,$s2,$t2,ror#8
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td3[s3>>0]
+	eor	$s0,$s0,$i1,ror#8
+	ldr	$i1,[$key],#16
+	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Td0[s3>>24]
+	eor	$s2,$s2,$i3,ror#24
+
+	ldr	$t1,[$key,#-12]
+	eor	$s0,$s0,$i1
+	ldr	$t2,[$key,#-8]
+	eor	$s3,$s3,$t3,ror#8
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0,lsr#16
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
+
+	subs	$rounds,$rounds,#1
+	bne	.Ldec_loop
+
+	add	$tbl,$tbl,#1024
+
+	ldr	$t2,[$tbl,#0]		@ prefetch Td4
+	ldr	$t3,[$tbl,#32]
+	ldr	$t1,[$tbl,#64]
+	ldr	$t2,[$tbl,#96]
+	ldr	$t3,[$tbl,#128]
+	ldr	$t1,[$tbl,#160]
+	ldr	$t2,[$tbl,#192]
+	ldr	$t3,[$tbl,#224]
+
+	ldrb	$s0,[$tbl,$s0]		@ Td4[s0>>24]
+	ldrb	$t1,[$tbl,$i1]		@ Td4[s0>>16]
+	and	$i1,lr,$s1		@ i0
+	ldrb	$t2,[$tbl,$i2]		@ Td4[s0>>8]
+	and	$i2,lr,$s1,lsr#16
+	ldrb	$t3,[$tbl,$i3]		@ Td4[s0>>0]
+	and	$i3,lr,$s1,lsr#8
+
+	add	$s1,$tbl,$s1,lsr#24
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s1>>0]
+	ldrb	$s1,[$s1]		@ Td4[s1>>24]
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s1>>16]
+	eor	$s0,$i1,$s0,lsl#24
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s1>>8]
+	eor	$s1,$t1,$s1,lsl#8
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,lsl#8
+	and	$i2,lr,$s2		@ i1
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s2>>8]
+	eor	$t3,$t3,$i3,lsl#8
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s2>>0]
+	and	$i3,lr,$s2,lsr#16
+
+	add	$s2,$tbl,$s2,lsr#24
+	ldrb	$s2,[$s2]		@ Td4[s2>>24]
+	eor	$s0,$s0,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s2>>16]
+	eor	$s1,$i2,$s1,lsl#16
+	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s2,$t2,$s2,lsl#16
+	and	$i2,lr,$s3,lsr#8	@ i1
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s3>>16]
+	eor	$t3,$t3,$i3,lsl#16
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s3>>8]
+	and	$i3,lr,$s3		@ i2
+
+	add	$s3,$tbl,$s3,lsr#24
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s3>>0]
+	ldrb	$s3,[$s3]		@ Td4[s3>>24]
+	eor	$s0,$s0,$i1,lsl#16
+	ldr	$i1,[$key,#0]
+	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
+	eor	$s2,$i3,$s2,lsl#8
+	ldr	$t2,[$key,#8]
+	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
+
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
+
+	sub	$tbl,$tbl,#1024
+	ldr	pc,[sp],#4		@ pop and return
+.size	_armv4_AES_decrypt,.-_armv4_AES_decrypt
+.asciz	"AES for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
+.align	2
+
+#endif
+___
+
+$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+	last if (!s/^#/@/ and !/^$/);
+	print;
+}
+close SELF;
+
+print $code;
+close STDOUT;	# enforce flush
diff --git a/src/crypto/aes/asm/aes-x86_64.pl b/src/crypto/aes/asm/aes-x86_64.pl
new file mode 100644
index 0000000..4b6e1b4
--- /dev/null
+++ b/src/crypto/aes/asm/aes-x86_64.pl
@@ -0,0 +1,2805 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# Version 2.1.
+#
+# aes-*-cbc benchmarks are improved by >70% [compared to gcc 3.3.2 on
+# Opteron 240 CPU] plus all the bells-n-whistles from 32-bit version
+# [you'll notice a lot of resemblance], such as compressed S-boxes
+# in little-endian byte order, prefetch of these tables in CBC mode,
+# as well as avoiding L1 cache aliasing between stack frame and key
+# schedule and already mentioned tables, compressed Td4...
+#
+# Performance in number of cycles per processed byte for 128-bit key:
+#
+#		ECB encrypt	ECB decrypt	CBC large chunk
+# AMD64		33		43		13.0
+# EM64T		38		56		18.6(*)
+# Core 2	30		42		14.5(*)
+# Atom		65		86		32.1(*)
+#
+# (*) with hyper-threading off
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+$verticalspin=1;	# unlike 32-bit version $verticalspin performs
+			# ~15% better on both AMD and Intel cores
+$speed_limit=512;	# see aes-586.pl for details
+
+$code=".text\n";
+
+$s0="%eax";
+$s1="%ebx";
+$s2="%ecx";
+$s3="%edx";
+$acc0="%esi";	$mask80="%rsi";
+$acc1="%edi";	$maskfe="%rdi";
+$acc2="%ebp";	$mask1b="%rbp";
+$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 LO() { my $r=shift; $r =~ s/%r([a-z]+)/%e\1/;
+			$r =~ s/%r([0-9]+)/%r\1d/;	$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")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	movzb	`&lo("$s1")`,$acc2
+	xor	1($sbox,$acc0,8),$t0
+	xor	1($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t3
+
+	mov	12($key),$s3
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	mov	0($key),$s0
+	xor	1($sbox,$acc1,8),$t2
+	xor	1($sbox,$acc2,8),$t3
+
+	mov	4($key),$s1
+	mov	8($key),$s2
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub enclastvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	movzb	2($sbox,$acc0,8),$t0
+	movzb	2($sbox,$acc1,8),$t1
+	movzb	2($sbox,$acc2,8),$t2
+
+	movzb	`&lo("$s3")`,$acc0
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	movzb	2($sbox,$acc0,8),$t3
+	mov	0($sbox,$acc1,8),$acc1	#$t0
+	mov	0($sbox,$acc2,8),$acc2	#$t1
+
+	and	\$0x0000ff00,$acc1
+	and	\$0x0000ff00,$acc2
+
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+	shr	\$16,$s2
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	shr	\$16,$s3
+	mov	0($sbox,$acc0,8),$acc0	#$t2
+	mov	0($sbox,$acc1,8),$acc1	#$t3
+
+	and	\$0x0000ff00,$acc0
+	and	\$0x0000ff00,$acc1
+	shr	\$16,$s1
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+	shr	\$16,$s0
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	mov	0($sbox,$acc0,8),$acc0	#$t0
+	mov	0($sbox,$acc1,8),$acc1	#$t1
+	mov	0($sbox,$acc2,8),$acc2	#$t2
+
+	and	\$0x00ff0000,$acc0
+	and	\$0x00ff0000,$acc1
+	and	\$0x00ff0000,$acc2
+
+	xor	$acc0,$t0
+	xor	$acc1,$t1
+	xor	$acc2,$t2
+
+	movzb	`&lo("$s1")`,$acc0
+	movzb	`&hi("$s3")`,$acc1
+	movzb	`&hi("$s0")`,$acc2
+	mov	0($sbox,$acc0,8),$acc0	#$t3
+	mov	2($sbox,$acc1,8),$acc1	#$t0
+	mov	2($sbox,$acc2,8),$acc2	#$t1
+
+	and	\$0x00ff0000,$acc0
+	and	\$0xff000000,$acc1
+	and	\$0xff000000,$acc2
+
+	xor	$acc0,$t3
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	mov	16+12($key),$s3
+	mov	2($sbox,$acc0,8),$acc0	#$t2
+	mov	2($sbox,$acc1,8),$acc1	#$t3
+	mov	16+0($key),$s0
+
+	and	\$0xff000000,$acc0
+	and	\$0xff000000,$acc1
+
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+
+	mov	16+4($key),$s1
+	mov	16+8($key),$s2
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub encstep()
+{ my ($i,@s) = @_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	if ($i==3) {
+		$tmp0=$s[1];
+		$tmp1=$s[2];
+		$tmp2=$s[3];
+	}
+	$code.="	movzb	".&lo($s[0]).",$out\n";
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	lea	16($key),$key\n"	if ($i==0);
+
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	mov	0($sbox,$out,8),$out\n";
+
+	$code.="	shr	\$16,$tmp1\n";
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+	$code.="	xor	3($sbox,$tmp0,8),$out\n";
+
+	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+	$code.="	xor	4*$i($key),$out\n";
+
+	$code.="	xor	2($sbox,$tmp1,8),$out\n";
+	$code.="	xor	1($sbox,$tmp2,8),$out\n";
+
+	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+sub enclast()
+{ my ($i,@s)=@_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	if ($i==3) {
+		$tmp0=$s[1];
+		$tmp1=$s[2];
+		$tmp2=$s[3];
+	}
+	$code.="	movzb	".&lo($s[0]).",$out\n";
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+
+	$code.="	mov	2($sbox,$out,8),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+	$code.="	and	\$0x000000ff,$out\n";
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	mov	0($sbox,$tmp0,8),$tmp0\n";
+	$code.="	mov	0($sbox,$tmp1,8),$tmp1\n";
+	$code.="	mov	2($sbox,$tmp2,8),$tmp2\n";
+
+	$code.="	and	\$0x0000ff00,$tmp0\n";
+	$code.="	and	\$0x00ff0000,$tmp1\n";
+	$code.="	and	\$0xff000000,$tmp2\n";
+
+	$code.="	xor	$tmp0,$out\n";
+	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
+	$code.="	xor	$tmp1,$out\n";
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	xor	$tmp2,$out\n";
+	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+$code.=<<___;
+.type	_x86_64_AES_encrypt,\@abi-omnipotent
+.align	16
+_x86_64_AES_encrypt:
+	xor	0($key),$s0			# xor with key
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+
+	mov	240($key),$rnds			# load key->rounds
+	sub	\$1,$rnds
+	jmp	.Lenc_loop
+.align	16
+.Lenc_loop:
+___
+	if ($verticalspin) { &encvert(); }
+	else {	&encstep(0,$s0,$s1,$s2,$s3);
+		&encstep(1,$s1,$s2,$s3,$s0);
+		&encstep(2,$s2,$s3,$s0,$s1);
+		&encstep(3,$s3,$s0,$s1,$s2);
+	}
+$code.=<<___;
+	sub	\$1,$rnds
+	jnz	.Lenc_loop
+___
+	if ($verticalspin) { &enclastvert(); }
+	else {	&enclast(0,$s0,$s1,$s2,$s3);
+		&enclast(1,$s1,$s2,$s3,$s0);
+		&enclast(2,$s2,$s3,$s0,$s1);
+		&enclast(3,$s3,$s0,$s1,$s2);
+		$code.=<<___;
+		xor	16+0($key),$s0		# xor with key
+		xor	16+4($key),$s1
+		xor	16+8($key),$s2
+		xor	16+12($key),$s3
+___
+	}
+$code.=<<___;
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_encrypt,.-_x86_64_AES_encrypt
+___
+
+# it's possible to implement this by shifting tN by 8, filling least
+# significant byte with byte load and finally bswap-ing at the end,
+# but such partial register load kills Core 2...
+sub enccompactvert()
+{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$t0
+	movzb	`&lo("$s1")`,$t1
+	movzb	`&lo("$s2")`,$t2
+	movzb	`&lo("$s3")`,$t3
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	shr	\$16,$s2
+	movzb	`&hi("$s3")`,$acc2
+	movzb	($sbox,$t0,1),$t0
+	movzb	($sbox,$t1,1),$t1
+	movzb	($sbox,$t2,1),$t2
+	movzb	($sbox,$t3,1),$t3
+
+	movzb	($sbox,$acc0,1),$t4	#$t0
+	movzb	`&hi("$s0")`,$acc0
+	movzb	($sbox,$acc1,1),$t5	#$t1
+	movzb	`&lo("$s2")`,$acc1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+
+	shl	\$8,$t4
+	shr	\$16,$s3
+	shl	\$8,$t5
+	xor	$t4,$t0
+	shr	\$16,$s0
+	movzb	`&lo("$s3")`,$t4
+	shr	\$16,$s1
+	xor	$t5,$t1
+	shl	\$8,$acc2
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	xor	$acc2,$t2
+
+	shl	\$8,$acc0
+	movzb	`&lo("$s1")`,$acc2
+	shl	\$16,$acc1
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s3")`,$acc0
+	movzb	($sbox,$t5,1),$t5	#$t2
+	xor	$acc1,$t0
+
+	shr	\$8,$s2
+	movzb	`&hi("$s0")`,$acc1
+	shl	\$16,$t4
+	shr	\$8,$s1
+	shl	\$16,$t5
+	xor	$t4,$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t3
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$acc1	#$t1
+	movzb	($sbox,$s2,1),$s3	#$t3
+	movzb	($sbox,$s1,1),$s2	#$t2
+
+	shl	\$16,$acc2
+	xor	$t5,$t2
+	shl	\$24,$acc0
+	xor	$acc2,$t3
+	shl	\$24,$acc1
+	xor	$acc0,$t0
+	shl	\$24,$s3
+	xor	$acc1,$t1
+	shl	\$24,$s2
+	mov	$t0,$s0
+	mov	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub enctransform_ref()
+{ my $sn = shift;
+  my ($acc,$r2,$tmp)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	mov	$sn,$acc
+	and	\$0x80808080,$acc
+	mov	$acc,$tmp
+	shr	\$7,$tmp
+	lea	($sn,$sn),$r2
+	sub	$tmp,$acc
+	and	\$0xfefefefe,$r2
+	and	\$0x1b1b1b1b,$acc
+	mov	$sn,$tmp
+	xor	$acc,$r2
+
+	xor	$r2,$sn
+	rol	\$24,$sn
+	xor	$r2,$sn
+	ror	\$16,$tmp
+	xor	$tmp,$sn
+	ror	\$8,$tmp
+	xor	$tmp,$sn
+___
+}
+
+# unlike decrypt case it does not pay off to parallelize enctransform
+sub enctransform()
+{ my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
+
+$code.=<<___;
+	mov	\$0x80808080,$t0
+	mov	\$0x80808080,$t1
+	and	$s0,$t0
+	and	$s1,$t1
+	mov	$t0,$acc0
+	mov	$t1,$acc1
+	shr	\$7,$t0
+	lea	($s0,$s0),$r20
+	shr	\$7,$t1
+	lea	($s1,$s1),$r21
+	sub	$t0,$acc0
+	sub	$t1,$acc1
+	and	\$0xfefefefe,$r20
+	and	\$0xfefefefe,$r21
+	and	\$0x1b1b1b1b,$acc0
+	and	\$0x1b1b1b1b,$acc1
+	mov	$s0,$t0
+	mov	$s1,$t1
+	xor	$acc0,$r20
+	xor	$acc1,$r21
+
+	xor	$r20,$s0
+	xor	$r21,$s1
+	 mov	\$0x80808080,$t2
+	rol	\$24,$s0
+	 mov	\$0x80808080,$t3
+	rol	\$24,$s1
+	 and	$s2,$t2
+	 and	$s3,$t3
+	xor	$r20,$s0
+	xor	$r21,$s1
+	 mov	$t2,$acc0
+	ror	\$16,$t0
+	 mov	$t3,$acc1
+	ror	\$16,$t1
+	 lea	($s2,$s2),$r20
+	 shr	\$7,$t2
+	xor	$t0,$s0
+	 shr	\$7,$t3
+	xor	$t1,$s1
+	ror	\$8,$t0
+	 lea	($s3,$s3),$r21
+	ror	\$8,$t1
+	 sub	$t2,$acc0
+	 sub	$t3,$acc1
+	xor	$t0,$s0
+	xor	$t1,$s1
+
+	and	\$0xfefefefe,$r20
+	and	\$0xfefefefe,$r21
+	and	\$0x1b1b1b1b,$acc0
+	and	\$0x1b1b1b1b,$acc1
+	mov	$s2,$t2
+	mov	$s3,$t3
+	xor	$acc0,$r20
+	xor	$acc1,$r21
+
+	ror	\$16,$t2
+	xor	$r20,$s2
+	ror	\$16,$t3
+	xor	$r21,$s3
+	rol	\$24,$s2
+	mov	0($sbox),$acc0			# prefetch Te4
+	rol	\$24,$s3
+	xor	$r20,$s2
+	mov	64($sbox),$acc1
+	xor	$r21,$s3
+	mov	128($sbox),$r20
+	xor	$t2,$s2
+	ror	\$8,$t2
+	xor	$t3,$s3
+	ror	\$8,$t3
+	xor	$t2,$s2
+	mov	192($sbox),$r21
+	xor	$t3,$s3
+___
+}
+
+$code.=<<___;
+.type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
+.align	16
+_x86_64_AES_encrypt_compact:
+	lea	128($sbox),$inp			# size optimization
+	mov	0-128($inp),$acc1		# prefetch Te4
+	mov	32-128($inp),$acc2
+	mov	64-128($inp),$t0
+	mov	96-128($inp),$t1
+	mov	128-128($inp),$acc1
+	mov	160-128($inp),$acc2
+	mov	192-128($inp),$t0
+	mov	224-128($inp),$t1
+	jmp	.Lenc_loop_compact
+.align	16
+.Lenc_loop_compact:
+		xor	0($key),$s0		# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+		lea	16($key),$key
+___
+		&enccompactvert();
+$code.=<<___;
+		cmp	16(%rsp),$key
+		je	.Lenc_compact_done
+___
+		&enctransform();
+$code.=<<___;
+	jmp	.Lenc_loop_compact
+.align	16
+.Lenc_compact_done:
+	xor	0($key),$s0
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
+___
+
+# void asm_AES_encrypt (const void *inp,void *out,const AES_KEY *key);
+$code.=<<___;
+.align	16
+.globl	asm_AES_encrypt
+.type	asm_AES_encrypt,\@function,3
+.hidden	asm_AES_encrypt
+asm_AES_encrypt:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+
+	# allocate frame "above" key schedule
+	mov	%rsp,%r10
+	lea	-63(%rdx),%rcx	# %rdx is key argument
+	and	\$-64,%rsp
+	sub	%rsp,%rcx
+	neg	%rcx
+	and	\$0x3c0,%rcx
+	sub	%rcx,%rsp
+	sub	\$32,%rsp
+
+	mov	%rsi,16(%rsp)	# save out
+	mov	%r10,24(%rsp)	# save real stack pointer
+.Lenc_prologue:
+
+	mov	%rdx,$key
+	mov	240($key),$rnds	# load rounds
+
+	mov	0(%rdi),$s0	# load input vector
+	mov	4(%rdi),$s1
+	mov	8(%rdi),$s2
+	mov	12(%rdi),$s3
+
+	shl	\$4,$rnds
+	lea	($key,$rnds),%rbp
+	mov	$key,(%rsp)	# key schedule
+	mov	%rbp,8(%rsp)	# end of key schedule
+
+	# pick Te4 copy which can't "overlap" with stack frame or key schedule
+	lea	.LAES_Te+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
+	sub	$sbox,%rbp
+	and	\$0x300,%rbp
+	lea	($sbox,%rbp),$sbox
+
+	call	_x86_64_AES_encrypt_compact
+
+	mov	16(%rsp),$out	# restore out
+	mov	24(%rsp),%rsi	# restore saved stack pointer
+	mov	$s0,0($out)	# write output vector
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	mov	(%rsi),%r15
+	mov	8(%rsi),%r14
+	mov	16(%rsi),%r13
+	mov	24(%rsi),%r12
+	mov	32(%rsi),%rbp
+	mov	40(%rsi),%rbx
+	lea	48(%rsi),%rsp
+.Lenc_epilogue:
+	ret
+.size	asm_AES_encrypt,.-asm_AES_encrypt
+___
+
+#------------------------------------------------------------------#
+
+sub decvert()
+{ 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),$t2
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	movzb	`&lo("$s3")`,$acc2
+	xor	3($sbox,$acc0,8),$t0
+	xor	3($sbox,$acc1,8),$t1
+	mov	0($sbox,$acc2,8),$t3
+
+	movzb	`&hi("$s1")`,$acc0
+	shr	\$16,$s0
+	movzb	`&hi("$s2")`,$acc2
+	xor	3($sbox,$acc0,8),$t2
+	shr	\$16,$s3
+	xor	3($sbox,$acc2,8),$t3
+
+	shr	\$16,$s1
+	lea	16($key),$key
+	shr	\$16,$s2
+
+	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),$t2
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	movzb	`&lo("$s1")`,$acc2
+	xor	1($sbox,$acc0,8),$t0
+	xor	1($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t3
+
+	movzb	`&hi("$s3")`,$acc0
+	mov	12($key),$s3
+	movzb	`&hi("$s0")`,$acc2
+	xor	1($sbox,$acc0,8),$t2
+	mov	0($key),$s0
+	xor	1($sbox,$acc2,8),$t3
+
+	xor	$t0,$s0
+	mov	4($key),$s1
+	mov	8($key),$s2
+	xor	$t2,$s2
+	xor	$t1,$s1
+	xor	$t3,$s3
+___
+}
+
+sub declastvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	lea	2048($sbox),$sbox	# size optimization
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	movzb	($sbox,$acc0,1),$t0
+	movzb	($sbox,$acc1,1),$t1
+	movzb	($sbox,$acc2,1),$t2
+
+	movzb	`&lo("$s3")`,$acc0
+	movzb	`&hi("$s3")`,$acc1
+	movzb	`&hi("$s0")`,$acc2
+	movzb	($sbox,$acc0,1),$t3
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	movzb	($sbox,$acc2,1),$acc2	#$t1
+
+	shl	\$8,$acc1
+	shl	\$8,$acc2
+
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+	shr	\$16,$s3
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	shr	\$16,$s0
+	movzb	($sbox,$acc0,1),$acc0	#$t2
+	movzb	($sbox,$acc1,1),$acc1	#$t3
+
+	shl	\$8,$acc0
+	shl	\$8,$acc1
+	shr	\$16,$s1
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+	shr	\$16,$s2
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$acc1	#$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+
+	shl	\$16,$acc0
+	shl	\$16,$acc1
+	shl	\$16,$acc2
+
+	xor	$acc0,$t0
+	xor	$acc1,$t1
+	xor	$acc2,$t2
+
+	movzb	`&lo("$s1")`,$acc0
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	movzb	($sbox,$acc2,1),$acc2	#$t1
+
+	shl	\$16,$acc0
+	shl	\$24,$acc1
+	shl	\$24,$acc2
+
+	xor	$acc0,$t3
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	mov	16+12($key),$s3
+	movzb	($sbox,$acc0,1),$acc0	#$t2
+	movzb	($sbox,$acc1,1),$acc1	#$t3
+	mov	16+0($key),$s0
+
+	shl	\$24,$acc0
+	shl	\$24,$acc1
+
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+
+	mov	16+4($key),$s1
+	mov	16+8($key),$s2
+	lea	-2048($sbox),$sbox
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub decstep()
+{ my ($i,@s) = @_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	$code.="	mov	$s[0],$out\n"		if ($i!=3);
+			$tmp1=$s[2]			if ($i==3);
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	and	\$0xFF,$out\n";
+
+	$code.="	mov	0($sbox,$out,8),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+			$tmp2=$s[3]			if ($i==3);
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+			$tmp0=$s[1]			if ($i==3);
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	and	\$0xFF,$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	xor	3($sbox,$tmp0,8),$out\n";
+	$code.="	xor	2($sbox,$tmp1,8),$out\n";
+	$code.="	xor	1($sbox,$tmp2,8),$out\n";
+
+	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+sub declast()
+{ my ($i,@s)=@_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	$code.="	mov	$s[0],$out\n"		if ($i!=3);
+			$tmp1=$s[2]			if ($i==3);
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	and	\$0xFF,$out\n";
+
+	$code.="	movzb	2048($sbox,$out,1),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+			$tmp2=$s[3]			if ($i==3);
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+			$tmp0=$s[1]			if ($i==3);
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	and	\$0xFF,$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
+	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
+	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
+
+	$code.="	shl	\$8,$tmp0\n";
+	$code.="	shl	\$16,$tmp1\n";
+	$code.="	shl	\$24,$tmp2\n";
+
+	$code.="	xor	$tmp0,$out\n";
+	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
+	$code.="	xor	$tmp1,$out\n";
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	xor	$tmp2,$out\n";
+	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+$code.=<<___;
+.type	_x86_64_AES_decrypt,\@abi-omnipotent
+.align	16
+_x86_64_AES_decrypt:
+	xor	0($key),$s0			# xor with key
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+
+	mov	240($key),$rnds			# load key->rounds
+	sub	\$1,$rnds
+	jmp	.Ldec_loop
+.align	16
+.Ldec_loop:
+___
+	if ($verticalspin) { &decvert(); }
+	else {	&decstep(0,$s0,$s3,$s2,$s1);
+		&decstep(1,$s1,$s0,$s3,$s2);
+		&decstep(2,$s2,$s1,$s0,$s3);
+		&decstep(3,$s3,$s2,$s1,$s0);
+		$code.=<<___;
+		lea	16($key),$key
+		xor	0($key),$s0			# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+___
+	}
+$code.=<<___;
+	sub	\$1,$rnds
+	jnz	.Ldec_loop
+___
+	if ($verticalspin) { &declastvert(); }
+	else {	&declast(0,$s0,$s3,$s2,$s1);
+		&declast(1,$s1,$s0,$s3,$s2);
+		&declast(2,$s2,$s1,$s0,$s3);
+		&declast(3,$s3,$s2,$s1,$s0);
+		$code.=<<___;
+		xor	16+0($key),$s0			# xor with key
+		xor	16+4($key),$s1
+		xor	16+8($key),$s2
+		xor	16+12($key),$s3
+___
+	}
+$code.=<<___;
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
+___
+
+sub deccompactvert()
+{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$t0
+	movzb	`&lo("$s1")`,$t1
+	movzb	`&lo("$s2")`,$t2
+	movzb	`&lo("$s3")`,$t3
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	shr	\$16,$s3
+	movzb	`&hi("$s1")`,$acc2
+	movzb	($sbox,$t0,1),$t0
+	movzb	($sbox,$t1,1),$t1
+	movzb	($sbox,$t2,1),$t2
+	movzb	($sbox,$t3,1),$t3
+
+	movzb	($sbox,$acc0,1),$t4	#$t0
+	movzb	`&hi("$s2")`,$acc0
+	movzb	($sbox,$acc1,1),$t5	#$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+
+	shr	\$16,$s2
+	shl	\$8,$t5
+	shl	\$8,$t4
+	movzb	`&lo("$s2")`,$acc1
+	shr	\$16,$s0
+	xor	$t4,$t0
+	shr	\$16,$s1
+	movzb	`&lo("$s3")`,$t4
+
+	shl	\$8,$acc2
+	xor	$t5,$t1
+	shl	\$8,$acc0
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	xor	$acc2,$t2
+	movzb	`&lo("$s1")`,$acc2
+
+	shl	\$16,$acc1
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s1")`,$acc0
+	movzb	($sbox,$acc2,1),$acc2	#$t3
+	xor	$acc1,$t0
+	movzb	($sbox,$t5,1),$t5	#$t2
+	movzb	`&hi("$s2")`,$acc1
+
+	shl	\$16,$acc2
+	shl	\$16,$t4
+	shl	\$16,$t5
+	xor	$acc2,$t3
+	movzb	`&hi("$s3")`,$acc2
+	xor	$t4,$t1
+	shr	\$8,$s0
+	xor	$t5,$t2
+
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$s1	#$t1
+	movzb	($sbox,$acc2,1),$s2	#$t2
+	movzb	($sbox,$s0,1),$s3	#$t3
+
+	mov	$t0,$s0
+	shl	\$24,$acc0
+	shl	\$24,$s1
+	shl	\$24,$s2
+	xor	$acc0,$s0
+	shl	\$24,$s3
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+# parallelized version! input is pair of 64-bit values: %rax=s1.s0
+# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
+# %ecx=s2 and %edx=s3.
+sub dectransform()
+{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
+  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
+  my $prefetch = shift;
+
+$code.=<<___;
+	mov	$mask80,$tp40
+	mov	$mask80,$tp48
+	and	$tp10,$tp40
+	and	$tp18,$tp48
+	mov	$tp40,$acc0
+	mov	$tp48,$acc8
+	shr	\$7,$tp40
+	lea	($tp10,$tp10),$tp20
+	shr	\$7,$tp48
+	lea	($tp18,$tp18),$tp28
+	sub	$tp40,$acc0
+	sub	$tp48,$acc8
+	and	$maskfe,$tp20
+	and	$maskfe,$tp28
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp20
+	xor	$acc8,$tp28
+	mov	$mask80,$tp80
+	mov	$mask80,$tp88
+
+	and	$tp20,$tp80
+	and	$tp28,$tp88
+	mov	$tp80,$acc0
+	mov	$tp88,$acc8
+	shr	\$7,$tp80
+	lea	($tp20,$tp20),$tp40
+	shr	\$7,$tp88
+	lea	($tp28,$tp28),$tp48
+	sub	$tp80,$acc0
+	sub	$tp88,$acc8
+	and	$maskfe,$tp40
+	and	$maskfe,$tp48
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp40
+	xor	$acc8,$tp48
+	mov	$mask80,$tp80
+	mov	$mask80,$tp88
+
+	and	$tp40,$tp80
+	and	$tp48,$tp88
+	mov	$tp80,$acc0
+	mov	$tp88,$acc8
+	shr	\$7,$tp80
+	 xor	$tp10,$tp20		# tp2^=tp1
+	shr	\$7,$tp88
+	 xor	$tp18,$tp28		# tp2^=tp1
+	sub	$tp80,$acc0
+	sub	$tp88,$acc8
+	lea	($tp40,$tp40),$tp80
+	lea	($tp48,$tp48),$tp88
+	 xor	$tp10,$tp40		# tp4^=tp1
+	 xor	$tp18,$tp48		# tp4^=tp1
+	and	$maskfe,$tp80
+	and	$maskfe,$tp88
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp80
+	xor	$acc8,$tp88
+
+	xor	$tp80,$tp10		# tp1^=tp8
+	xor	$tp88,$tp18		# tp1^=tp8
+	xor	$tp80,$tp20		# tp2^tp1^=tp8
+	xor	$tp88,$tp28		# tp2^tp1^=tp8
+	mov	$tp10,$acc0
+	mov	$tp18,$acc8
+	xor	$tp80,$tp40		# tp4^tp1^=tp8
+	shr	\$32,$acc0
+	xor	$tp88,$tp48		# tp4^tp1^=tp8
+	shr	\$32,$acc8
+	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
+	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
+	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
+	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
+	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
+	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
+	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
+
+	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
+	xor	`&LO("$tp80")`,`&LO("$tp10")`
+	shr	\$32,$tp80
+	xor	`&LO("$tp88")`,`&LO("$tp18")`
+	shr	\$32,$tp88
+	xor	`&LO("$tp80")`,`&LO("$acc0")`
+	xor	`&LO("$tp88")`,`&LO("$acc8")`
+
+	mov	$tp20,$tp80
+	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp28,$tp88
+	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
+	shr	\$32,$tp80
+	xor	`&LO("$tp20")`,`&LO("$tp10")`
+	shr	\$32,$tp88
+	xor	`&LO("$tp28")`,`&LO("$tp18")`
+	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp40,$tp20
+	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp48,$tp28
+	shr	\$32,$tp20
+	xor	`&LO("$tp80")`,`&LO("$acc0")`
+	shr	\$32,$tp28
+	xor	`&LO("$tp88")`,`&LO("$acc8")`
+
+	`"mov	0($sbox),$mask80"	if ($prefetch)`
+	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	64($sbox),$maskfe"	if ($prefetch)`
+	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	128($sbox),$mask1b"	if ($prefetch)`
+	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	192($sbox),$tp80"	if ($prefetch)`
+	xor	`&LO("$tp40")`,`&LO("$tp10")`
+	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
+	xor	`&LO("$tp48")`,`&LO("$tp18")`
+	`"mov	256($sbox),$tp88"	if ($prefetch)`
+	xor	`&LO("$tp20")`,`&LO("$acc0")`
+	xor	`&LO("$tp28")`,`&LO("$acc8")`
+___
+}
+
+$code.=<<___;
+.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
+.align	16
+_x86_64_AES_decrypt_compact:
+	lea	128($sbox),$inp			# size optimization
+	mov	0-128($inp),$acc1		# prefetch Td4
+	mov	32-128($inp),$acc2
+	mov	64-128($inp),$t0
+	mov	96-128($inp),$t1
+	mov	128-128($inp),$acc1
+	mov	160-128($inp),$acc2
+	mov	192-128($inp),$t0
+	mov	224-128($inp),$t1
+	jmp	.Ldec_loop_compact
+
+.align	16
+.Ldec_loop_compact:
+		xor	0($key),$s0		# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+		lea	16($key),$key
+___
+		&deccompactvert();
+$code.=<<___;
+		cmp	16(%rsp),$key
+		je	.Ldec_compact_done
+
+		mov	256+0($sbox),$mask80
+		shl	\$32,%rbx
+		shl	\$32,%rdx
+		mov	256+8($sbox),$maskfe
+		or	%rbx,%rax
+		or	%rdx,%rcx
+		mov	256+16($sbox),$mask1b
+___
+		&dectransform(1);
+$code.=<<___;
+	jmp	.Ldec_loop_compact
+.align	16
+.Ldec_compact_done:
+	xor	0($key),$s0
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
+___
+
+# void asm_AES_decrypt (const void *inp,void *out,const AES_KEY *key);
+$code.=<<___;
+.align	16
+.globl	asm_AES_decrypt
+.type	asm_AES_decrypt,\@function,3
+.hidden	asm_AES_decrypt
+asm_AES_decrypt:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+
+	# allocate frame "above" key schedule
+	mov	%rsp,%r10
+	lea	-63(%rdx),%rcx	# %rdx is key argument
+	and	\$-64,%rsp
+	sub	%rsp,%rcx
+	neg	%rcx
+	and	\$0x3c0,%rcx
+	sub	%rcx,%rsp
+	sub	\$32,%rsp
+
+	mov	%rsi,16(%rsp)	# save out
+	mov	%r10,24(%rsp)	# save real stack pointer
+.Ldec_prologue:
+
+	mov	%rdx,$key
+	mov	240($key),$rnds	# load rounds
+
+	mov	0(%rdi),$s0	# load input vector
+	mov	4(%rdi),$s1
+	mov	8(%rdi),$s2
+	mov	12(%rdi),$s3
+
+	shl	\$4,$rnds
+	lea	($key,$rnds),%rbp
+	mov	$key,(%rsp)	# key schedule
+	mov	%rbp,8(%rsp)	# end of key schedule
+
+	# pick Td4 copy which can't "overlap" with stack frame or key schedule
+	lea	.LAES_Td+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
+	sub	$sbox,%rbp
+	and	\$0x300,%rbp
+	lea	($sbox,%rbp),$sbox
+	shr	\$3,%rbp	# recall "magic" constants!
+	add	%rbp,$sbox
+
+	call	_x86_64_AES_decrypt_compact
+
+	mov	16(%rsp),$out	# restore out
+	mov	24(%rsp),%rsi	# restore saved stack pointer
+	mov	$s0,0($out)	# write output vector
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	mov	(%rsi),%r15
+	mov	8(%rsi),%r14
+	mov	16(%rsi),%r13
+	mov	24(%rsi),%r12
+	mov	32(%rsi),%rbp
+	mov	40(%rsi),%rbx
+	lea	48(%rsi),%rsp
+.Ldec_epilogue:
+	ret
+.size	asm_AES_decrypt,.-asm_AES_decrypt
+___
+#------------------------------------------------------------------#
+
+sub enckey()
+{
+$code.=<<___;
+	movz	%dl,%esi		# rk[i]>>0
+	movzb	-128(%rbp,%rsi),%ebx
+	movz	%dh,%esi		# rk[i]>>8
+	shl	\$24,%ebx
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	shr	\$16,%edx
+	movz	%dl,%esi		# rk[i]>>16
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	movz	%dh,%esi		# rk[i]>>24
+	shl	\$8,%ebx
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	shl	\$16,%ebx
+	xor	%ebx,%eax
+
+	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
+___
+}
+
+# int asm_AES_set_encrypt_key(const unsigned char *userKey, const int bits, AES_KEY *key)
+$code.=<<___;
+.align	16
+.globl asm_AES_set_encrypt_key
+.type  asm_AES_set_encrypt_key,\@function,3
+asm_AES_set_encrypt_key:
+	push	%rbx
+	push	%rbp
+	push	%r12			# redundant, but allows to share 
+	push	%r13			# exception handler...
+	push	%r14
+	push	%r15
+	sub	\$8,%rsp
+.Lenc_key_prologue:
+
+	call	_x86_64_AES_set_encrypt_key
+
+	mov	40(%rsp),%rbp
+	mov	48(%rsp),%rbx
+	add	\$56,%rsp
+.Lenc_key_epilogue:
+	ret
+.size asm_AES_set_encrypt_key,.-asm_AES_set_encrypt_key
+
+.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
+.align	16
+_x86_64_AES_set_encrypt_key:
+	mov	%esi,%ecx			# %ecx=bits
+	mov	%rdi,%rsi			# %rsi=userKey
+	mov	%rdx,%rdi			# %rdi=key
+
+	test	\$-1,%rsi
+	jz	.Lbadpointer
+	test	\$-1,%rdi
+	jz	.Lbadpointer
+
+	lea	.LAES_Te(%rip),%rbp
+	lea	2048+128(%rbp),%rbp
+
+	# prefetch Te4
+	mov	0-128(%rbp),%eax
+	mov	32-128(%rbp),%ebx
+	mov	64-128(%rbp),%r8d
+	mov	96-128(%rbp),%edx
+	mov	128-128(%rbp),%eax
+	mov	160-128(%rbp),%ebx
+	mov	192-128(%rbp),%r8d
+	mov	224-128(%rbp),%edx
+
+	cmp	\$128,%ecx
+	je	.L10rounds
+	cmp	\$192,%ecx
+	je	.L12rounds
+	cmp	\$256,%ecx
+	je	.L14rounds
+	mov	\$-2,%rax			# invalid number of bits
+	jmp	.Lexit
+
+.L10rounds:
+	mov	0(%rsi),%rax			# copy first 4 dwords
+	mov	8(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rdx,8(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L10shortcut
+.align	4
+.L10loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	12(%rdi),%edx			# rk[3]
+.L10shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,16(%rdi)			# rk[4]
+		xor	4(%rdi),%eax
+		mov	%eax,20(%rdi)			# rk[5]
+		xor	8(%rdi),%eax
+		mov	%eax,24(%rdi)			# rk[6]
+		xor	12(%rdi),%eax
+		mov	%eax,28(%rdi)			# rk[7]
+		add	\$1,%ecx
+		lea	16(%rdi),%rdi
+		cmp	\$10,%ecx
+	jl	.L10loop
+
+	movl	\$10,80(%rdi)			# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.L12rounds:
+	mov	0(%rsi),%rax			# copy first 6 dwords
+	mov	8(%rsi),%rbx
+	mov	16(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rbx,8(%rdi)
+	mov	%rdx,16(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L12shortcut
+.align	4
+.L12loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	20(%rdi),%edx			# rk[5]
+.L12shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,24(%rdi)			# rk[6]
+		xor	4(%rdi),%eax
+		mov	%eax,28(%rdi)			# rk[7]
+		xor	8(%rdi),%eax
+		mov	%eax,32(%rdi)			# rk[8]
+		xor	12(%rdi),%eax
+		mov	%eax,36(%rdi)			# rk[9]
+
+		cmp	\$7,%ecx
+		je	.L12break
+		add	\$1,%ecx
+
+		xor	16(%rdi),%eax
+		mov	%eax,40(%rdi)			# rk[10]
+		xor	20(%rdi),%eax
+		mov	%eax,44(%rdi)			# rk[11]
+
+		lea	24(%rdi),%rdi
+	jmp	.L12loop
+.L12break:
+	movl	\$12,72(%rdi)		# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.L14rounds:		
+	mov	0(%rsi),%rax			# copy first 8 dwords
+	mov	8(%rsi),%rbx
+	mov	16(%rsi),%rcx
+	mov	24(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rbx,8(%rdi)
+	mov	%rcx,16(%rdi)
+	mov	%rdx,24(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L14shortcut
+.align	4
+.L14loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	28(%rdi),%edx			# rk[4]
+.L14shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,32(%rdi)			# rk[8]
+		xor	4(%rdi),%eax
+		mov	%eax,36(%rdi)			# rk[9]
+		xor	8(%rdi),%eax
+		mov	%eax,40(%rdi)			# rk[10]
+		xor	12(%rdi),%eax
+		mov	%eax,44(%rdi)			# rk[11]
+
+		cmp	\$6,%ecx
+		je	.L14break
+		add	\$1,%ecx
+
+		mov	%eax,%edx
+		mov	16(%rdi),%eax			# rk[4]
+		movz	%dl,%esi			# rk[11]>>0
+		movzb	-128(%rbp,%rsi),%ebx
+		movz	%dh,%esi			# rk[11]>>8
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		shr	\$16,%edx
+		shl	\$8,%ebx
+		movz	%dl,%esi			# rk[11]>>16
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		movz	%dh,%esi			# rk[11]>>24
+		shl	\$16,%ebx
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		shl	\$24,%ebx
+		xor	%ebx,%eax
+
+		mov	%eax,48(%rdi)			# rk[12]
+		xor	20(%rdi),%eax
+		mov	%eax,52(%rdi)			# rk[13]
+		xor	24(%rdi),%eax
+		mov	%eax,56(%rdi)			# rk[14]
+		xor	28(%rdi),%eax
+		mov	%eax,60(%rdi)			# rk[15]
+
+		lea	32(%rdi),%rdi
+	jmp	.L14loop
+.L14break:
+	movl	\$14,48(%rdi)		# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.Lbadpointer:
+	mov	\$-1,%rax
+.Lexit:
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
+___
+
+sub deckey_ref()
+{ my ($i,$ptr,$te,$td) = @_;
+  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
+$code.=<<___;
+	mov	$i($ptr),$tp1
+	mov	$tp1,$acc
+	and	\$0x80808080,$acc
+	mov	$acc,$tp4
+	shr	\$7,$tp4
+	lea	0($tp1,$tp1),$tp2
+	sub	$tp4,$acc
+	and	\$0xfefefefe,$tp2
+	and	\$0x1b1b1b1b,$acc
+	xor	$tp2,$acc
+	mov	$acc,$tp2
+
+	and	\$0x80808080,$acc
+	mov	$acc,$tp8
+	shr	\$7,$tp8
+	lea	0($tp2,$tp2),$tp4
+	sub	$tp8,$acc
+	and	\$0xfefefefe,$tp4
+	and	\$0x1b1b1b1b,$acc
+	 xor	$tp1,$tp2		# tp2^tp1
+	xor	$tp4,$acc
+	mov	$acc,$tp4
+
+	and	\$0x80808080,$acc
+	mov	$acc,$tp8
+	shr	\$7,$tp8
+	sub	$tp8,$acc
+	lea	0($tp4,$tp4),$tp8
+	 xor	$tp1,$tp4		# tp4^tp1
+	and	\$0xfefefefe,$tp8
+	and	\$0x1b1b1b1b,$acc
+	xor	$acc,$tp8
+
+	xor	$tp8,$tp1		# tp1^tp8
+	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
+	xor	$tp8,$tp2		# tp2^tp1^tp8
+	xor	$tp8,$tp4		# tp4^tp1^tp8
+	xor	$tp2,$tp8
+	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
+
+	xor	$tp8,$tp1
+	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
+	xor	$tp2,$tp1
+	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
+	xor	$tp4,$tp1
+
+	mov	$tp1,$i($ptr)
+___
+}
+
+# int asm_AES_set_decrypt_key(const unsigned char *userKey, const int bits, AES_KEY *key)
+$code.=<<___;
+.align	16
+.globl asm_AES_set_decrypt_key
+.type  asm_AES_set_decrypt_key,\@function,3
+asm_AES_set_decrypt_key:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	push	%rdx			# save key schedule
+.Ldec_key_prologue:
+
+	call	_x86_64_AES_set_encrypt_key
+	mov	(%rsp),%r8		# restore key schedule
+	cmp	\$0,%eax
+	jne	.Labort
+
+	mov	240(%r8),%r14d		# pull number of rounds
+	xor	%rdi,%rdi
+	lea	(%rdi,%r14d,4),%rcx
+	mov	%r8,%rsi
+	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
+.align	4
+.Linvert:
+		mov	0(%rsi),%rax
+		mov	8(%rsi),%rbx
+		mov	0(%rdi),%rcx
+		mov	8(%rdi),%rdx
+		mov	%rax,0(%rdi)
+		mov	%rbx,8(%rdi)
+		mov	%rcx,0(%rsi)
+		mov	%rdx,8(%rsi)
+		lea	16(%rsi),%rsi
+		lea	-16(%rdi),%rdi
+		cmp	%rsi,%rdi
+	jne	.Linvert
+
+	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
+
+	mov	40(%rax),$mask80
+	mov	48(%rax),$maskfe
+	mov	56(%rax),$mask1b
+
+	mov	%r8,$key
+	sub	\$1,%r14d
+.align	4
+.Lpermute:
+		lea	16($key),$key
+		mov	0($key),%rax
+		mov	8($key),%rcx
+___
+		&dectransform ();
+$code.=<<___;
+		mov	%eax,0($key)
+		mov	%ebx,4($key)
+		mov	%ecx,8($key)
+		mov	%edx,12($key)
+		sub	\$1,%r14d
+	jnz	.Lpermute
+
+	xor	%rax,%rax
+.Labort:
+	mov	8(%rsp),%r15
+	mov	16(%rsp),%r14
+	mov	24(%rsp),%r13
+	mov	32(%rsp),%r12
+	mov	40(%rsp),%rbp
+	mov	48(%rsp),%rbx
+	add	\$56,%rsp
+.Ldec_key_epilogue:
+	ret
+.size	asm_AES_set_decrypt_key,.-asm_AES_set_decrypt_key
+___
+
+# void asm_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
+# -8(%rsp)		return address
+my $keyp="0(%rsp)";		# one to pass as $key
+my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
+my $_rsp="16(%rsp)";		# saved %rsp
+my $_inp="24(%rsp)";		# copy of 1st parameter, inp
+my $_out="32(%rsp)";		# copy of 2nd parameter, out
+my $_len="40(%rsp)";		# copy of 3rd parameter, length
+my $_key="48(%rsp)";		# copy of 4th parameter, key
+my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
+my $ivec="64(%rsp)";		# ivec[16]
+my $aes_key="80(%rsp)";		# copy of aes_key
+my $mark="80+240(%rsp)";	# copy of aes_key->rounds
+
+$code.=<<___;
+.align	16
+.globl	asm_AES_cbc_encrypt
+.type	asm_AES_cbc_encrypt,\@function,6
+.extern	OPENSSL_ia32cap_P
+.hidden	asm_AES_cbc_encrypt
+asm_AES_cbc_encrypt:
+	cmp	\$0,%rdx	# check length
+	je	.Lcbc_epilogue
+	pushfq
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lcbc_prologue:
+
+	cld
+	mov	%r9d,%r9d	# clear upper half of enc
+
+	lea	.LAES_Te(%rip),$sbox
+	cmp	\$0,%r9
+	jne	.Lcbc_picked_te
+	lea	.LAES_Td(%rip),$sbox
+.Lcbc_picked_te:
+
+	mov	OPENSSL_ia32cap_P(%rip),%r10d
+	cmp	\$$speed_limit,%rdx
+	jb	.Lcbc_slow_prologue
+	test	\$15,%rdx
+	jnz	.Lcbc_slow_prologue
+	bt	\$28,%r10d
+	jc	.Lcbc_slow_prologue
+
+	# allocate aligned stack frame...
+	lea	-88-248(%rsp),$key
+	and	\$-64,$key
+
+	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
+	mov	$sbox,%r10
+	lea	2304($sbox),%r11
+	mov	$key,%r12
+	and	\$0xFFF,%r10	# s = $sbox&0xfff
+	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
+	and	\$0xFFF,%r12	# p = %rsp&0xfff
+
+	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
+	jb	.Lcbc_te_break_out
+	sub	%r11,%r12
+	sub	%r12,$key
+	jmp	.Lcbc_te_ok
+.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
+	sub	%r10,%r12
+	and	\$0xFFF,%r12
+	add	\$320,%r12
+	sub	%r12,$key
+.align	4
+.Lcbc_te_ok:
+
+	xchg	%rsp,$key
+	#add	\$8,%rsp	# reserve for return address!
+	mov	$key,$_rsp	# save %rsp
+.Lcbc_fast_body:
+	mov	%rdi,$_inp	# save copy of inp
+	mov	%rsi,$_out	# save copy of out
+	mov	%rdx,$_len	# save copy of len
+	mov	%rcx,$_key	# save copy of key
+	mov	%r8,$_ivp	# save copy of ivp
+	movl	\$0,$mark	# copy of aes_key->rounds = 0;
+	mov	%r8,%rbp	# rearrange input arguments
+	mov	%r9,%rbx
+	mov	%rsi,$out
+	mov	%rdi,$inp
+	mov	%rcx,$key
+
+	mov	240($key),%eax		# key->rounds
+	# do we copy key schedule to stack?
+	mov	$key,%r10
+	sub	$sbox,%r10
+	and	\$0xfff,%r10
+	cmp	\$2304,%r10
+	jb	.Lcbc_do_ecopy
+	cmp	\$4096-248,%r10
+	jb	.Lcbc_skip_ecopy
+.align	4
+.Lcbc_do_ecopy:
+		mov	$key,%rsi
+		lea	$aes_key,%rdi
+		lea	$aes_key,$key
+		mov	\$240/8,%ecx
+		.long	0x90A548F3	# rep movsq
+		mov	%eax,(%rdi)	# copy aes_key->rounds
+.Lcbc_skip_ecopy:
+	mov	$key,$keyp	# save key pointer
+
+	mov	\$18,%ecx
+.align	4
+.Lcbc_prefetch_te:
+		mov	0($sbox),%r10
+		mov	32($sbox),%r11
+		mov	64($sbox),%r12
+		mov	96($sbox),%r13
+		lea	128($sbox),$sbox
+		sub	\$1,%ecx
+	jnz	.Lcbc_prefetch_te
+	lea	-2304($sbox),$sbox
+
+	cmp	\$0,%rbx
+	je	.LFAST_DECRYPT
+
+#----------------------------- ENCRYPT -----------------------------#
+	mov	0(%rbp),$s0		# load iv
+	mov	4(%rbp),$s1
+	mov	8(%rbp),$s2
+	mov	12(%rbp),$s3
+
+.align	4
+.Lcbc_fast_enc_loop:
+		xor	0($inp),$s0
+		xor	4($inp),$s1
+		xor	8($inp),$s2
+		xor	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_encrypt
+
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10
+		mov	$s0,0($out)
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		sub	\$16,%r10
+		test	\$-16,%r10
+		mov	%r10,$_len
+	jnz	.Lcbc_fast_enc_loop
+	mov	$_ivp,%rbp	# restore ivp
+	mov	$s0,0(%rbp)	# save ivec
+	mov	$s1,4(%rbp)
+	mov	$s2,8(%rbp)
+	mov	$s3,12(%rbp)
+
+	jmp	.Lcbc_fast_cleanup
+
+#----------------------------- DECRYPT -----------------------------#
+.align	16
+.LFAST_DECRYPT:
+	cmp	$inp,$out
+	je	.Lcbc_fast_dec_in_place
+
+	mov	%rbp,$ivec
+.align	4
+.Lcbc_fast_dec_loop:
+		mov	0($inp),$s0	# read input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_decrypt
+
+		mov	$ivec,%rbp	# load ivp
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10	# load len
+		xor	0(%rbp),$s0	# xor iv
+		xor	4(%rbp),$s1
+		xor	8(%rbp),$s2
+		xor	12(%rbp),$s3
+		mov	$inp,%rbp	# current input, next iv
+
+		sub	\$16,%r10
+		mov	%r10,$_len	# update len
+		mov	%rbp,$ivec	# update ivp
+
+		mov	$s0,0($out)	# write output
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+	jnz	.Lcbc_fast_dec_loop
+	mov	$_ivp,%r12		# load user ivp
+	mov	0(%rbp),%r10		# load iv
+	mov	8(%rbp),%r11
+	mov	%r10,0(%r12)		# copy back to user
+	mov	%r11,8(%r12)
+	jmp	.Lcbc_fast_cleanup
+
+.align	16
+.Lcbc_fast_dec_in_place:
+	mov	0(%rbp),%r10		# copy iv to stack
+	mov	8(%rbp),%r11
+	mov	%r10,0+$ivec
+	mov	%r11,8+$ivec
+.align	4
+.Lcbc_fast_dec_in_place_loop:
+		mov	0($inp),$s0	# load input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_decrypt
+
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10
+		xor	0+$ivec,$s0
+		xor	4+$ivec,$s1
+		xor	8+$ivec,$s2
+		xor	12+$ivec,$s3
+
+		mov	0($inp),%r11	# load input
+		mov	8($inp),%r12
+		sub	\$16,%r10
+		jz	.Lcbc_fast_dec_in_place_done
+
+		mov	%r11,0+$ivec	# copy input to iv
+		mov	%r12,8+$ivec
+
+		mov	$s0,0($out)	# save output [zaps input]
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		mov	%r10,$_len
+	jmp	.Lcbc_fast_dec_in_place_loop
+.Lcbc_fast_dec_in_place_done:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)	# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0($out)	# save output [zaps input]
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+.align	4
+.Lcbc_fast_cleanup:
+	cmpl	\$0,$mark	# was the key schedule copied?
+	lea	$aes_key,%rdi
+	je	.Lcbc_exit
+		mov	\$240/8,%ecx
+		xor	%rax,%rax
+		.long	0x90AB48F3	# rep stosq
+
+	jmp	.Lcbc_exit
+
+#--------------------------- SLOW ROUTINE ---------------------------#
+.align	16
+.Lcbc_slow_prologue:
+	# allocate aligned stack frame...
+	lea	-88(%rsp),%rbp
+	and	\$-64,%rbp
+	# ... just "above" key schedule
+	lea	-88-63(%rcx),%r10
+	sub	%rbp,%r10
+	neg	%r10
+	and	\$0x3c0,%r10
+	sub	%r10,%rbp
+
+	xchg	%rsp,%rbp
+	#add	\$8,%rsp	# reserve for return address!
+	mov	%rbp,$_rsp	# save %rsp
+.Lcbc_slow_body:
+	#mov	%rdi,$_inp	# save copy of inp
+	#mov	%rsi,$_out	# save copy of out
+	#mov	%rdx,$_len	# save copy of len
+	#mov	%rcx,$_key	# save copy of key
+	mov	%r8,$_ivp	# save copy of ivp
+	mov	%r8,%rbp	# rearrange input arguments
+	mov	%r9,%rbx
+	mov	%rsi,$out
+	mov	%rdi,$inp
+	mov	%rcx,$key
+	mov	%rdx,%r10
+
+	mov	240($key),%eax
+	mov	$key,$keyp	# save key pointer
+	shl	\$4,%eax
+	lea	($key,%rax),%rax
+	mov	%rax,$keyend
+
+	# pick Te4 copy which can't "overlap" with stack frame or key scdedule
+	lea	2048($sbox),$sbox
+	lea	768-8(%rsp),%rax
+	sub	$sbox,%rax
+	and	\$0x300,%rax
+	lea	($sbox,%rax),$sbox
+
+	cmp	\$0,%rbx
+	je	.LSLOW_DECRYPT
+
+#--------------------------- SLOW ENCRYPT ---------------------------#
+	test	\$-16,%r10		# check upon length
+	mov	0(%rbp),$s0		# load iv
+	mov	4(%rbp),$s1
+	mov	8(%rbp),$s2
+	mov	12(%rbp),$s3
+	jz	.Lcbc_slow_enc_tail	# short input...
+
+.align	4
+.Lcbc_slow_enc_loop:
+		xor	0($inp),$s0
+		xor	4($inp),$s1
+		xor	8($inp),$s2
+		xor	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# save inp
+		mov	$out,$_out	# save out
+		mov	%r10,$_len	# save len
+
+		call	_x86_64_AES_encrypt_compact
+
+		mov	$_inp,$inp	# restore inp
+		mov	$_out,$out	# restore out
+		mov	$_len,%r10	# restore len
+		mov	$s0,0($out)
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		sub	\$16,%r10
+		test	\$-16,%r10
+	jnz	.Lcbc_slow_enc_loop
+	test	\$15,%r10
+	jnz	.Lcbc_slow_enc_tail
+	mov	$_ivp,%rbp	# restore ivp
+	mov	$s0,0(%rbp)	# save ivec
+	mov	$s1,4(%rbp)
+	mov	$s2,8(%rbp)
+	mov	$s3,12(%rbp)
+
+	jmp	.Lcbc_exit
+
+.align	4
+.Lcbc_slow_enc_tail:
+	mov	%rax,%r11
+	mov	%rcx,%r12
+	mov	%r10,%rcx
+	mov	$inp,%rsi
+	mov	$out,%rdi
+	.long	0x9066A4F3		# rep movsb
+	mov	\$16,%rcx		# zero tail
+	sub	%r10,%rcx
+	xor	%rax,%rax
+	.long	0x9066AAF3		# rep stosb
+	mov	$out,$inp		# this is not a mistake!
+	mov	\$16,%r10		# len=16
+	mov	%r11,%rax
+	mov	%r12,%rcx
+	jmp	.Lcbc_slow_enc_loop	# one more spin...
+#--------------------------- SLOW DECRYPT ---------------------------#
+.align	16
+.LSLOW_DECRYPT:
+	shr	\$3,%rax
+	add	%rax,$sbox		# recall "magic" constants!
+
+	mov	0(%rbp),%r11		# copy iv to stack
+	mov	8(%rbp),%r12
+	mov	%r11,0+$ivec
+	mov	%r12,8+$ivec
+
+.align	4
+.Lcbc_slow_dec_loop:
+		mov	0($inp),$s0	# load input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# save inp
+		mov	$out,$_out	# save out
+		mov	%r10,$_len	# save len
+
+		call	_x86_64_AES_decrypt_compact
+
+		mov	$_inp,$inp	# restore inp
+		mov	$_out,$out	# restore out
+		mov	$_len,%r10
+		xor	0+$ivec,$s0
+		xor	4+$ivec,$s1
+		xor	8+$ivec,$s2
+		xor	12+$ivec,$s3
+
+		mov	0($inp),%r11	# load input
+		mov	8($inp),%r12
+		sub	\$16,%r10
+		jc	.Lcbc_slow_dec_partial
+		jz	.Lcbc_slow_dec_done
+
+		mov	%r11,0+$ivec	# copy input to iv
+		mov	%r12,8+$ivec
+
+		mov	$s0,0($out)	# save output [can zap input]
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+	jmp	.Lcbc_slow_dec_loop
+.Lcbc_slow_dec_done:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)		# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0($out)		# save output [can zap input]
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	jmp	.Lcbc_exit
+
+.align	4
+.Lcbc_slow_dec_partial:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)		# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0+$ivec		# save output to stack
+	mov	$s1,4+$ivec
+	mov	$s2,8+$ivec
+	mov	$s3,12+$ivec
+
+	mov	$out,%rdi
+	lea	$ivec,%rsi
+	lea	16(%r10),%rcx
+	.long	0x9066A4F3	# rep movsb
+	jmp	.Lcbc_exit
+
+.align	16
+.Lcbc_exit:
+	mov	$_rsp,%rsi
+	mov	(%rsi),%r15
+	mov	8(%rsi),%r14
+	mov	16(%rsi),%r13
+	mov	24(%rsi),%r12
+	mov	32(%rsi),%rbp
+	mov	40(%rsi),%rbx
+	lea	48(%rsi),%rsp
+.Lcbc_popfq:
+	popfq
+.Lcbc_epilogue:
+	ret
+.size	asm_AES_cbc_encrypt,.-asm_AES_cbc_encrypt
+___
+}
+
+$code.=<<___;
+.align	64
+.LAES_Te:
+___
+	&_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);
+
+#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+#rcon:
+$code.=<<___;
+	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
+	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
+	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
+	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
+___
+$code.=<<___;
+.align	64
+.LAES_Td:
+___
+	&_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:	# four copies of Td4 to choose from to avoid L1 aliasing
+	&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);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&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);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&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);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&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);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+.asciz  "AES for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align	64
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	block_se_handler,\@abi-omnipotent
+.align	16
+block_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_block_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_block_prologue
+
+	mov	24(%rax),%rax		# pull saved real stack pointer
+	lea	48(%rax),%rax		# adjust...
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_block_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	jmp	.Lcommon_seh_exit
+.size	block_se_handler,.-block_se_handler
+
+.type	key_se_handler,\@abi-omnipotent
+.align	16
+key_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_key_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_key_prologue
+
+	lea	56(%rax),%rax
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_key_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	jmp	.Lcommon_seh_exit
+.size	key_se_handler,.-key_se_handler
+
+.type	cbc_se_handler,\@abi-omnipotent
+.align	16
+cbc_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lcbc_prologue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
+	jb	.Lin_cbc_prologue
+
+	lea	.Lcbc_fast_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
+	jb	.Lin_cbc_frame_setup
+
+	lea	.Lcbc_slow_prologue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
+	jb	.Lin_cbc_body
+
+	lea	.Lcbc_slow_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
+	jb	.Lin_cbc_frame_setup
+
+.Lin_cbc_body:
+	mov	152($context),%rax	# pull context->Rsp
+
+	lea	.Lcbc_epilogue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
+	jae	.Lin_cbc_prologue
+
+	lea	8(%rax),%rax
+
+	lea	.Lcbc_popfq(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
+	jae	.Lin_cbc_prologue
+
+	mov	`16-8`(%rax),%rax	# biased $_rsp
+	lea	56(%rax),%rax
+
+.Lin_cbc_frame_setup:
+	mov	-16(%rax),%rbx
+	mov	-24(%rax),%rbp
+	mov	-32(%rax),%r12
+	mov	-40(%rax),%r13
+	mov	-48(%rax),%r14
+	mov	-56(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_cbc_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+.Lcommon_seh_exit:
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	cbc_se_handler,.-cbc_se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_asm_AES_encrypt
+	.rva	.LSEH_end_asm_AES_encrypt
+	.rva	.LSEH_info_asm_AES_encrypt
+
+	.rva	.LSEH_begin_asm_AES_decrypt
+	.rva	.LSEH_end_asm_AES_decrypt
+	.rva	.LSEH_info_asm_AES_decrypt
+
+	.rva	.LSEH_begin_asm_AES_set_encrypt_key
+	.rva	.LSEH_end_asm_AES_set_encrypt_key
+	.rva	.LSEH_info_asm_AES_set_encrypt_key
+
+	.rva	.LSEH_begin_asm_AES_set_decrypt_key
+	.rva	.LSEH_end_asm_AES_set_decrypt_key
+	.rva	.LSEH_info_asm_AES_set_decrypt_key
+
+	.rva	.LSEH_begin_asm_AES_cbc_encrypt
+	.rva	.LSEH_end_asm_AES_cbc_encrypt
+	.rva	.LSEH_info_asm_AES_cbc_encrypt
+
+.section	.xdata
+.align	8
+.LSEH_info_asm_AES_encrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
+.LSEH_info_asm_AES_decrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
+.LSEH_info_asm_AES_set_encrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
+.LSEH_info_asm_AES_set_decrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
+.LSEH_info_asm_AES_cbc_encrypt:
+	.byte	9,0,0,0
+	.rva	cbc_se_handler
+___
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+
+print $code;
+
+close STDOUT;
diff --git a/src/crypto/aes/asm/aesni-x86.pl b/src/crypto/aes/asm/aesni-x86.pl
new file mode 100644
index 0000000..3deb86a
--- /dev/null
+++ b/src/crypto/aes/asm/aesni-x86.pl
@@ -0,0 +1,2232 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> 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.
+
+######################################################################
+# Current large-block performance in cycles per byte processed with
+# 128-bit key (less is better).
+#
+#		CBC en-/decrypt	CTR	XTS	ECB
+# Westmere	3.77/1.37	1.37	1.52	1.27
+# * Bridge	5.07/0.98	0.99	1.09	0.91
+# Haswell	4.44/0.80	0.97	1.03	0.72
+# Atom		5.77/3.56	3.67	4.03	3.46
+# Bulldozer	5.80/0.98	1.05	1.24	0.93
+
+$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 originally made no sense to implement 2x subroutine.
+# But times change and it became appropriate to spend extra 192 bytes
+# on 2x subroutine on Atom Silvermont account. For processors that
+# can schedule aes[enc|dec] every cycle optimal interleave factor
+# equals to corresponding instructions latency. 8x is optimal for
+# * Bridge, but it's unfeasible to accommodate such implementation
+# in XMM registers addreassable in 32-bit mode and therefore maximum
+# of 6x is used instead...
+
+sub aesni_generate2
+{ my $p=shift;
+
+    &function_begin_B("_aesni_${p}rypt2");
+	&$movekey	($rndkey0,&QWP(0,$key));
+	&shl		($rounds,4);
+	&$movekey	($rndkey1,&QWP(16,$key));
+	&xorps		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);
+	&$movekey	($rndkey0,&QWP(32,$key));
+	&lea		($key,&DWP(32,$key,$rounds));
+	&neg		($rounds);
+	&add		($rounds,16);
+
+    &set_label("${p}2_loop");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	eval"&aes${p}	($inout1,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	eval"&aes${p}	($inout0,$rndkey0)";
+	eval"&aes${p}	($inout1,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&jnz		(&label("${p}2_loop"));
+    eval"&aes${p}	($inout0,$rndkey1)";
+    eval"&aes${p}	($inout1,$rndkey1)";
+    eval"&aes${p}last	($inout0,$rndkey0)";
+    eval"&aes${p}last	($inout1,$rndkey0)";
+    &ret();
+    &function_end_B("_aesni_${p}rypt2");
+}
+
+sub aesni_generate3
+{ my $p=shift;
+
+    &function_begin_B("_aesni_${p}rypt3");
+	&$movekey	($rndkey0,&QWP(0,$key));
+	&shl		($rounds,4);
+	&$movekey	($rndkey1,&QWP(16,$key));
+	&xorps		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&$movekey	($rndkey0,&QWP(32,$key));
+	&lea		($key,&DWP(32,$key,$rounds));
+	&neg		($rounds);
+	&add		($rounds,16);
+
+    &set_label("${p}3_loop");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	eval"&aes${p}	($inout1,$rndkey1)";
+	eval"&aes${p}	($inout2,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	eval"&aes${p}	($inout0,$rndkey0)";
+	eval"&aes${p}	($inout1,$rndkey0)";
+	eval"&aes${p}	($inout2,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&jnz		(&label("${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));
+	&shl		($rounds,4);
+	&xorps		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($inout3,$rndkey0);
+	&$movekey	($rndkey0,&QWP(32,$key));
+	&lea		($key,&DWP(32,$key,$rounds));
+	&neg		($rounds);
+	&data_byte	(0x0f,0x1f,0x40,0x00);
+	&add		($rounds,16);
+
+    &set_label("${p}4_loop");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	eval"&aes${p}	($inout1,$rndkey1)";
+	eval"&aes${p}	($inout2,$rndkey1)";
+	eval"&aes${p}	($inout3,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	eval"&aes${p}	($inout0,$rndkey0)";
+	eval"&aes${p}	($inout1,$rndkey0)";
+	eval"&aes${p}	($inout2,$rndkey0)";
+	eval"&aes${p}	($inout3,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+    &jnz		(&label("${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));
+	&shl		($rounds,4);
+	&$movekey	($rndkey1,&QWP(16,$key));
+	&xorps		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);	# pxor does better here
+	&pxor		($inout2,$rndkey0);
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&pxor		($inout3,$rndkey0);
+	&pxor		($inout4,$rndkey0);
+	eval"&aes${p}	($inout1,$rndkey1)";
+	&lea		($key,&DWP(32,$key,$rounds));
+	&neg		($rounds);
+	eval"&aes${p}	($inout2,$rndkey1)";
+	&pxor		($inout5,$rndkey0);
+	&add		($rounds,16);
+	eval"&aes${p}	($inout3,$rndkey1)";
+	eval"&aes${p}	($inout4,$rndkey1)";
+	eval"&aes${p}	($inout5,$rndkey1)";
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&jmp		(&label("_aesni_${p}rypt6_enter"));
+
+    &set_label("${p}6_loop",16);
+	eval"&aes${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,$rndkey1)";
+    &set_label("_aesni_${p}rypt6_enter");
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	eval"&aes${p}	($inout0,$rndkey0)";
+	eval"&aes${p}	($inout1,$rndkey0)";
+	eval"&aes${p}	($inout2,$rndkey0)";
+	eval"&aes${p}	($inout3,$rndkey0)";
+	eval"&aes${p}	($inout4,$rndkey0)";
+	eval"&aes${p}	($inout5,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+    &jnz		(&label("${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_${p}rypt6");
+}
+&aesni_generate2("enc") if ($PREFIX eq "aesni");
+&aesni_generate2("dec");
+&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);
+	&call	("_aesni_encrypt2");
+	&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);
+	&call	("_aesni_decrypt2");
+	&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"),$key_);
+
+	&shl	($rounds,4);
+	&mov	($rounds_,16);
+	&lea	($key_,&DWP(0,$key));
+	&movdqa	($inout3,&QWP(0,"esp"));
+	&movdqa	($inout0,$ivec);
+	&lea	($key,&DWP(32,$key,$rounds));
+	&sub	($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,$in0);
+	&xorps		($cmac,$rndkey0);		# cmac^=inp
+	&$movekey	($rndkey0,&QWP(32,$key_));
+
+&set_label("ccm64_enc2_loop");
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($cmac,$rndkey1);
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	&aesenc		($inout0,$rndkey0);
+	&aesenc		($cmac,$rndkey0);
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&jnz		(&label("ccm64_enc2_loop"));
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($cmac,$rndkey1);
+	&paddq		($ivec,&QWP(16,"esp"));
+	&dec		($len);
+	&aesenclast	($inout0,$rndkey0);
+	&aesenclast	($cmac,$rndkey0);
+
+	&lea	($inp,&DWP(16,$inp));
+	&xorps	($in0,$inout0);			# inp^=E(ivec)
+	&movdqa	($inout0,$ivec);
+	&movups	(&QWP(0,$out),$in0);		# save output
+	&pshufb	($inout0,$inout3);
+	&lea	($out,&DWP(16,$out));
+	&jnz	(&label("ccm64_enc_outer"));
+
+	&mov	("esp",&DWP(48,"esp"));
+	&mov	($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 ($inline)
+	{   &aesni_inline_generate1("enc");	}
+	else
+	{   &call	("_aesni_encrypt1");	}
+	&shl	($rounds_,4);
+	&mov	($rounds,16);
+	&movups	($in0,&QWP(0,$inp));		# load inp
+	&paddq	($ivec,&QWP(16,"esp"));
+	&lea	($inp,&QWP(16,$inp));
+	&sub	($rounds,$rounds_);
+	&lea	($key,&DWP(32,$key_,$rounds_));
+	&mov	($rounds_,$rounds);
+	&jmp	(&label("ccm64_dec_outer"));
+
+&set_label("ccm64_dec_outer",16);
+	&xorps	($in0,$inout0);			# inp ^= E(ivec)
+	&movdqa	($inout0,$ivec);
+	&movups	(&QWP(0,$out),$in0);		# save output
+	&lea	($out,&DWP(16,$out));
+	&pshufb	($inout0,$inout3);
+
+	&sub	($len,1);
+	&jz	(&label("ccm64_dec_break"));
+
+	&$movekey	($rndkey0,&QWP(0,$key_));
+	&mov		($rounds,$rounds_);
+	&$movekey	($rndkey1,&QWP(16,$key_));
+	&xorps		($in0,$rndkey0);
+	&xorps		($inout0,$rndkey0);
+	&xorps		($cmac,$in0);		# cmac^=out
+	&$movekey	($rndkey0,&QWP(32,$key_));
+
+&set_label("ccm64_dec2_loop");
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($cmac,$rndkey1);
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
+	&aesenc		($inout0,$rndkey0);
+	&aesenc		($cmac,$rndkey0);
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&jnz		(&label("ccm64_dec2_loop"));
+	&movups		($in0,&QWP(0,$inp));	# load inp
+	&paddq		($ivec,&QWP(16,"esp"));
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($cmac,$rndkey1);
+	&aesenclast	($inout0,$rndkey0);
+	&aesenclast	($cmac,$rndkey0);
+	&lea		($inp,&QWP(16,$inp));
+	&jmp	(&label("ccm64_dec_outer"));
+
+&set_label("ccm64_dec_break",16);
+	&mov	($rounds,&DWP(240,$key_));
+	&mov	($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 crypto/modes/ctr128.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	($rndkey0,$rndkey0);
+	&pxor	($rndkey1,$rndkey1);
+	&movdqa	($inout0,&QWP(0,"esp"));	# load byte-swap mask
+	&pinsrd	($rndkey0,$rounds_,0);
+	&lea	($key_,&DWP(3,$rounds_));
+	&pinsrd	($rndkey1,$key_,0);
+	&inc	($rounds_);
+	&pinsrd	($rndkey0,$rounds_,1);
+	&inc	($key_);
+	&pinsrd	($rndkey1,$key_,1);
+	&inc	($rounds_);
+	&pinsrd	($rndkey0,$rounds_,2);
+	&inc	($key_);
+	&pinsrd	($rndkey1,$key_,2);
+	&movdqa	(&QWP(48,"esp"),$rndkey0);	# save 1st triplet
+	&pshufb	($rndkey0,$inout0);		# byte swap
+	&movdqu	($inout4,&QWP(0,$key));		# key[0]
+	&movdqa	(&QWP(64,"esp"),$rndkey1);	# save 2nd triplet
+	&pshufb	($rndkey1,$inout0);		# byte swap
+
+	&pshufd	($inout0,$rndkey0,3<<6);	# place counter to upper dword
+	&pshufd	($inout1,$rndkey0,2<<6);
+	&cmp	($len,6);
+	&jb	(&label("ctr32_tail"));
+	&pxor	($inout5,$inout4);		# counter-less ivec^key[0]
+	&shl	($rounds,4);
+	&mov	($rounds_,16);
+	&movdqa	(&QWP(32,"esp"),$inout5);	# save counter-less ivec^key[0]
+	&mov	($key_,$key);			# backup $key
+	&sub	($rounds_,$rounds);		# backup twisted $rounds
+	&lea	($key,&DWP(32,$key,$rounds));
+	&sub	($len,6);
+	&jmp	(&label("ctr32_loop6"));
+
+&set_label("ctr32_loop6",16);
+	# inlining _aesni_encrypt6's prologue gives ~6% improvement...
+	&pshufd	($inout2,$rndkey0,1<<6);
+	&movdqa	($rndkey0,&QWP(32,"esp"));	# pull counter-less ivec
+	&pshufd	($inout3,$rndkey1,3<<6);
+	&pxor		($inout0,$rndkey0);	# merge counter-less ivec
+	&pshufd	($inout4,$rndkey1,2<<6);
+	&pxor		($inout1,$rndkey0);
+	&pshufd	($inout5,$rndkey1,1<<6);
+	&$movekey	($rndkey1,&QWP(16,$key_));
+	&pxor		($inout2,$rndkey0);
+	&pxor		($inout3,$rndkey0);
+	&aesenc		($inout0,$rndkey1);
+	&pxor		($inout4,$rndkey0);
+	&pxor		($inout5,$rndkey0);
+	&aesenc		($inout1,$rndkey1);
+	&$movekey	($rndkey0,&QWP(32,$key_));
+	&mov		($rounds,$rounds_);
+	&aesenc		($inout2,$rndkey1);
+	&aesenc		($inout3,$rndkey1);
+	&aesenc		($inout4,$rndkey1);
+	&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(64,"esp"));	# load 2nd triplet
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+
+	&paddd	($rndkey1,$rndkey0);		# 2nd triplet increment
+	&paddd	($rndkey0,&QWP(48,"esp"));	# 1st triplet increment
+	&movdqa	($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"),$rndkey0);	# save 1st triplet
+	&pshufb	($rndkey0,$inout0);		# byte swap
+	&xorps	($inout4,$inout2);
+	&movups	(&QWP(0x30,$out),$inout3);
+	&xorps	($inout5,$inout1);
+	&movdqa	(&QWP(64,"esp"),$rndkey1);	# save 2nd triplet
+	&pshufb	($rndkey1,$inout0);		# byte swap
+	&movups	(&QWP(0x40,$out),$inout4);
+	&pshufd	($inout0,$rndkey0,3<<6);
+	&movups	(&QWP(0x50,$out),$inout5);
+	&lea	($out,&DWP(0x60,$out));
+
+	&pshufd	($inout1,$rndkey0,2<<6);
+	&sub	($len,6);
+	&jnc	(&label("ctr32_loop6"));
+
+	&add	($len,6);
+	&jz	(&label("ctr32_ret"));
+	&movdqu	($inout5,&QWP(0,$key_));
+	&mov	($key,$key_);
+	&pxor	($inout5,&QWP(32,"esp"));	# restore count-less ivec
+	&mov	($rounds,&DWP(240,$key_));	# restore $rounds
+
+&set_label("ctr32_tail");
+	&por	($inout0,$inout5);
+	&cmp	($len,2);
+	&jb	(&label("ctr32_one"));
+
+	&pshufd	($inout2,$rndkey0,1<<6);
+	&por	($inout1,$inout5);
+	&je	(&label("ctr32_two"));
+
+	&pshufd	($inout3,$rndkey1,3<<6);
+	&por	($inout2,$inout5);
+	&cmp	($len,4);
+	&jb	(&label("ctr32_three"));
+
+	&pshufd	($inout4,$rndkey1,2<<6);
+	&por	($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_encrypt2");
+	&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"));
+
+	&shl	($rounds,4);
+	&mov	($rounds_,16);
+	&sub	($rounds_,$rounds);
+	&lea	($key,&DWP(32,$key,$rounds));
+	&jmp	(&label("xts_enc_loop6"));
+
+&set_label("xts_enc_loop6",16);
+	for ($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,$tweak);
+
+	# inline _aesni_encrypt6 prologue and flip xor with tweak and key[0]
+	&mov	($rounds,$rounds_);		# restore $rounds
+	&movdqu	($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,$key_));
+	&pxor	($inout1,&QWP(16*1,"esp"));
+	&pxor	($inout2,&QWP(16*2,"esp"));
+	 &aesenc	($inout0,$rndkey1);
+	&pxor	($inout3,&QWP(16*3,"esp"));
+	&pxor	($inout4,&QWP(16*4,"esp"));
+	 &aesenc	($inout1,$rndkey1);
+	&pxor		($inout5,$rndkey0);
+	 &$movekey	($rndkey0,&QWP(32,$key_));
+	 &aesenc	($inout2,$rndkey1);
+	 &aesenc	($inout3,$rndkey1);
+	 &aesenc	($inout4,$rndkey1);
+	 &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
+	&pxor	($tweak,$twres);
+
+	&sub	($len,16*6);
+	&jnc	(&label("xts_enc_loop6"));
+
+	&mov	($rounds,&DWP(240,$key_));	# 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,$inout4);
+
+	&call	("_aesni_encrypt2");
+
+	&xorps	($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"));
+
+	&shl	($rounds,4);
+	&mov	($rounds_,16);
+	&sub	($rounds_,$rounds);
+	&lea	($key,&DWP(32,$key,$rounds));
+	&jmp	(&label("xts_dec_loop6"));
+
+&set_label("xts_dec_loop6",16);
+	for ($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,$tweak);
+
+	# inline _aesni_encrypt6 prologue and flip xor with tweak and key[0]
+	&mov	($rounds,$rounds_);
+	&movdqu	($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,$key_));
+	&pxor	($inout1,&QWP(16*1,"esp"));
+	&pxor	($inout2,&QWP(16*2,"esp"));
+	 &aesdec	($inout0,$rndkey1);
+	&pxor	($inout3,&QWP(16*3,"esp"));
+	&pxor	($inout4,&QWP(16*4,"esp"));
+	 &aesdec	($inout1,$rndkey1);
+	&pxor		($inout5,$rndkey0);
+	 &$movekey	($rndkey0,&QWP(32,$key_));
+	 &aesdec	($inout2,$rndkey1);
+	 &aesdec	($inout3,$rndkey1);
+	 &aesdec	($inout4,$rndkey1);
+	 &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
+	&pxor	($tweak,$twres);
+
+	&sub	($len,16*6);
+	&jnc	(&label("xts_dec_loop6"));
+
+	&mov	($rounds,&DWP(240,$key_));	# 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,$inout4);
+
+	&call	("_aesni_decrypt2");
+
+	&xorps	($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);
+	&call	("_aesni_decrypt2");
+	&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();
diff --git a/src/crypto/aes/asm/aesni-x86_64.pl b/src/crypto/aes/asm/aesni-x86_64.pl
new file mode 100644
index 0000000..5f61746
--- /dev/null
+++ b/src/crypto/aes/asm/aesni-x86_64.pl
@@ -0,0 +1,3555 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> 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-x86_64.pl [see below for
+# details].
+#
+# Performance.
+#
+# Given aes(enc|dec) instructions' latency asymptotic performance for
+# non-parallelizable modes such as CBC encrypt is 3.75 cycles per byte
+# processed with 128-bit key. And given their throughput asymptotic
+# performance for parallelizable modes is 1.25 cycles per byte. Being
+# asymptotic limit it's not something you commonly achieve in reality,
+# but how close does one get? Below are results collected for
+# different modes and block sized. Pairs of numbers are for en-/
+# decryption.
+#
+#	16-byte     64-byte     256-byte    1-KB        8-KB
+# ECB	4.25/4.25   1.38/1.38   1.28/1.28   1.26/1.26	1.26/1.26
+# CTR	5.42/5.42   1.92/1.92   1.44/1.44   1.28/1.28   1.26/1.26
+# CBC	4.38/4.43   4.15/1.43   4.07/1.32   4.07/1.29   4.06/1.28
+# CCM	5.66/9.42   4.42/5.41   4.16/4.40   4.09/4.15   4.06/4.07   
+# OFB	5.42/5.42   4.64/4.64   4.44/4.44   4.39/4.39   4.38/4.38
+# CFB	5.73/5.85   5.56/5.62   5.48/5.56   5.47/5.55   5.47/5.55
+#
+# ECB, CTR, CBC and CCM results are free from EVP overhead. This means
+# that otherwise used 'openssl speed -evp aes-128-??? -engine aesni
+# [-decrypt]' will exhibit 10-15% worse results for smaller blocks.
+# The results were collected with specially crafted speed.c benchmark
+# in order to compare them with results reported in "Intel Advanced
+# Encryption Standard (AES) New Instruction Set" White Paper Revision
+# 3.0 dated May 2010. All above results are consistently better. This
+# module also provides better performance for block sizes smaller than
+# 128 bytes in points *not* represented in the above table.
+#
+# Looking at the results for 8-KB buffer.
+#
+# CFB and OFB results are far from the limit, because implementation
+# uses "generic" CRYPTO_[c|o]fb128_encrypt interfaces relying on
+# single-block aesni_encrypt, which is not the most optimal way to go.
+# CBC encrypt result is unexpectedly high and there is no documented
+# explanation for it. Seemingly there is a small penalty for feeding
+# the result back to AES unit the way it's done in CBC mode. There is
+# nothing one can do and the result appears optimal. CCM result is
+# identical to CBC, because CBC-MAC is essentially CBC encrypt without
+# saving output. CCM CTR "stays invisible," because it's neatly
+# interleaved wih CBC-MAC. This provides ~30% improvement over
+# "straghtforward" CCM implementation with CTR and CBC-MAC performed
+# disjointly. Parallelizable modes practically achieve the theoretical
+# limit.
+#
+# Looking at how results vary with buffer size.
+#
+# Curves are practically saturated at 1-KB buffer size. In most cases
+# "256-byte" performance is >95%, and "64-byte" is ~90% of "8-KB" one.
+# CTR curve doesn't follow this pattern and is "slowest" changing one
+# with "256-byte" result being 87% of "8-KB." This is because overhead
+# in CTR mode is most computationally intensive. Small-block CCM
+# decrypt is slower than encrypt, because first CTR and last CBC-MAC
+# iterations can't be interleaved.
+#
+# Results for 192- and 256-bit keys.
+#
+# EVP-free results were observed to scale perfectly with number of
+# rounds for larger block sizes, i.e. 192-bit result being 10/12 times
+# lower and 256-bit one - 10/14. Well, in CBC encrypt case differences
+# are a tad smaller, because the above mentioned penalty biases all
+# results by same constant value. In similar way function call
+# overhead affects small-block performance, as well as OFB and CFB
+# results. Differences are not large, most common coefficients are
+# 10/11.7 and 10/13.4 (as opposite to 10/12.0 and 10/14.0), but one
+# observe even 10/11.2 and 10/12.4 (CTR, OFB, CFB)...
+
+# January 2011
+#
+# While Westmere processor features 6 cycles latency for aes[enc|dec]
+# instructions, which can be scheduled every second cycle, Sandy
+# Bridge spends 8 cycles per instruction, but it can schedule them
+# every cycle. This means that code targeting Westmere would perform
+# suboptimally on Sandy Bridge. Therefore this update.
+#
+# In addition, non-parallelizable CBC encrypt (as well as CCM) is
+# optimized. Relative improvement might appear modest, 8% on Westmere,
+# but in absolute terms it's 3.77 cycles per byte encrypted with
+# 128-bit key on Westmere, and 5.07 - on Sandy Bridge. These numbers
+# should be compared to asymptotic limits of 3.75 for Westmere and
+# 5.00 for Sandy Bridge. Actually, the fact that they get this close
+# to asymptotic limits is quite amazing. Indeed, the limit is
+# calculated as latency times number of rounds, 10 for 128-bit key,
+# and divided by 16, the number of bytes in block, or in other words
+# it accounts *solely* for aesenc instructions. But there are extra
+# instructions, and numbers so close to the asymptotic limits mean
+# that it's as if it takes as little as *one* additional cycle to
+# execute all of them. How is it possible? It is possible thanks to
+# out-of-order execution logic, which manages to overlap post-
+# processing of previous block, things like saving the output, with
+# actual encryption of current block, as well as pre-processing of
+# current block, things like fetching input and xor-ing it with
+# 0-round element of the key schedule, with actual encryption of
+# previous block. Keep this in mind...
+#
+# For parallelizable modes, such as ECB, CBC decrypt, CTR, higher
+# performance is achieved by interleaving instructions working on
+# independent blocks. In which case asymptotic limit for such modes
+# can be obtained by dividing above mentioned numbers by AES
+# instructions' interleave factor. Westmere can execute at most 3 
+# instructions at a time, meaning that optimal interleave factor is 3,
+# and that's where the "magic" number of 1.25 come from. "Optimal
+# interleave factor" means that increase of interleave factor does
+# not improve performance. The formula has proven to reflect reality
+# pretty well on Westmere... Sandy Bridge on the other hand can
+# execute up to 8 AES instructions at a time, so how does varying
+# interleave factor affect the performance? Here is table for ECB
+# (numbers are cycles per byte processed with 128-bit key):
+#
+# instruction interleave factor		3x	6x	8x
+# theoretical asymptotic limit		1.67	0.83	0.625
+# measured performance for 8KB block	1.05	0.86	0.84
+#
+# "as if" interleave factor		4.7x	5.8x	6.0x
+#
+# Further data for other parallelizable modes:
+#
+# CBC decrypt				1.16	0.93	0.74
+# CTR					1.14	0.91	0.74
+#
+# Well, given 3x column it's probably inappropriate to call the limit
+# asymptotic, if it can be surpassed, isn't it? What happens there?
+# Rewind to CBC paragraph for the answer. Yes, out-of-order execution
+# magic is responsible for this. Processor overlaps not only the
+# additional instructions with AES ones, but even AES instuctions
+# processing adjacent triplets of independent blocks. In the 6x case
+# additional instructions  still claim disproportionally small amount
+# of additional cycles, but in 8x case number of instructions must be
+# a tad too high for out-of-order logic to cope with, and AES unit
+# remains underutilized... As you can see 8x interleave is hardly
+# justifiable, so there no need to feel bad that 32-bit aesni-x86.pl
+# utilizies 6x interleave because of limited register bank capacity.
+#
+# Higher interleave factors do have negative impact on Westmere
+# performance. While for ECB mode it's negligible ~1.5%, other
+# parallelizables perform ~5% worse, which is outweighed by ~25%
+# improvement on Sandy Bridge. To balance regression on Westmere
+# CTR mode was implemented with 6x aesenc interleave factor.
+
+# April 2011
+#
+# Add aesni_xts_[en|de]crypt. Westmere spends 1.25 cycles processing
+# one byte out of 8KB with 128-bit key, Sandy Bridge - 0.90. Just like
+# in CTR mode AES instruction interleave factor was chosen to be 6x.
+
+######################################################################
+# Current large-block performance in cycles per byte processed with
+# 128-bit key (less is better).
+#
+#		CBC en-/decrypt	CTR	XTS	ECB
+# Westmere	3.77/1.25	1.25	1.25	1.26
+# * Bridge	5.07/0.74	0.75	0.90	0.85
+# Haswell	4.44/0.63	0.63	0.73	0.63
+# Atom		5.75/3.54	3.56	4.12	3.87(*)
+# Bulldozer	5.77/0.70	0.72	0.90	0.70
+#
+# (*)	Atom ECB result is suboptimal because of penalties incurred
+#	by operations on %xmm8-15. As ECB is not considered
+#	critical, nothing was done to mitigate the problem.
+
+$PREFIX="aesni";	# if $PREFIX is set to "AES", the script
+			# generates drop-in replacement for
+			# crypto/aes/asm/aes-x86_64.pl:-)
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+$movkey = $PREFIX eq "aesni" ? "movups" : "movups";
+@_4args=$win64?	("%rcx","%rdx","%r8", "%r9") :	# Win64 order
+		("%rdi","%rsi","%rdx","%rcx");	# Unix order
+
+$code=".text\n";
+$code.=".extern	OPENSSL_ia32cap_P\n";
+
+$rounds="%eax";	# input to and changed by aesni_[en|de]cryptN !!!
+# this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ...
+$inp="%rdi";
+$out="%rsi";
+$len="%rdx";
+$key="%rcx";	# input to and changed by aesni_[en|de]cryptN !!!
+$ivp="%r8";	# cbc, ctr, ...
+
+$rnds_="%r10d";	# backup copy for $rounds
+$key_="%r11";	# backup copy for $key
+
+# %xmm register layout
+$rndkey0="%xmm0";	$rndkey1="%xmm1";
+$inout0="%xmm2";	$inout1="%xmm3";
+$inout2="%xmm4";	$inout3="%xmm5";
+$inout4="%xmm6";	$inout5="%xmm7";
+$inout6="%xmm8";	$inout7="%xmm9";
+
+$in2="%xmm6";		$in1="%xmm7";	# used in CBC decrypt, CTR, ...
+$in0="%xmm8";		$iv="%xmm9";
+
+# Inline version of internal aesni_[en|de]crypt1.
+#
+# Why folded loop? Because aes[enc|dec] is slow enough to accommodate
+# cycles which take care of loop variables...
+{ my $sn;
+sub aesni_generate1 {
+my ($p,$key,$rounds,$inout,$ivec)=@_;	$inout=$inout0 if (!defined($inout));
+++$sn;
+$code.=<<___;
+	$movkey	($key),$rndkey0
+	$movkey	16($key),$rndkey1
+___
+$code.=<<___ if (defined($ivec));
+	xorps	$rndkey0,$ivec
+	lea	32($key),$key
+	xorps	$ivec,$inout
+___
+$code.=<<___ if (!defined($ivec));
+	lea	32($key),$key
+	xorps	$rndkey0,$inout
+___
+$code.=<<___;
+.Loop_${p}1_$sn:
+	aes${p}	$rndkey1,$inout
+	dec	$rounds
+	$movkey	($key),$rndkey1
+	lea	16($key),$key
+	jnz	.Loop_${p}1_$sn	# loop body is 16 bytes
+	aes${p}last	$rndkey1,$inout
+___
+}}
+# void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key);
+#
+{ my ($inp,$out,$key) = @_4args;
+
+$code.=<<___;
+.globl	${PREFIX}_encrypt
+.type	${PREFIX}_encrypt,\@abi-omnipotent
+.align	16
+${PREFIX}_encrypt:
+	movups	($inp),$inout0		# load input
+	mov	240($key),$rounds	# key->rounds
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)		# output
+	ret
+.size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
+
+.globl	${PREFIX}_decrypt
+.type	${PREFIX}_decrypt,\@abi-omnipotent
+.align	16
+${PREFIX}_decrypt:
+	movups	($inp),$inout0		# load input
+	mov	240($key),$rounds	# key->rounds
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)		# output
+	ret
+.size	${PREFIX}_decrypt, .-${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 originally made no sense to implement 2x subroutine.
+# But times change and it became appropriate to spend extra 192 bytes
+# on 2x subroutine on Atom Silvermont account. For processors that
+# can schedule aes[enc|dec] every cycle optimal interleave factor
+# equals to corresponding instructions latency. 8x is optimal for
+# * Bridge and "super-optimal" for other Intel CPUs... 
+
+sub aesni_generate2 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-1] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt2,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt2:
+	$movkey	($key),$rndkey0
+	shl	\$4,$rounds
+	$movkey	16($key),$rndkey1
+	xorps	$rndkey0,$inout0
+	xorps	$rndkey0,$inout1
+	$movkey	32($key),$rndkey0
+	lea	32($key,$rounds),$key
+	neg	%rax				# $rounds
+	add	\$16,%rax
+
+.L${dir}_loop2:
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+	aes${dir}	$rndkey0,$inout0
+	aes${dir}	$rndkey0,$inout1
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.L${dir}_loop2
+
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	ret
+.size	_aesni_${dir}rypt2,.-_aesni_${dir}rypt2
+___
+}
+sub aesni_generate3 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-2] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt3,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt3:
+	$movkey	($key),$rndkey0
+	shl	\$4,$rounds
+	$movkey	16($key),$rndkey1
+	xorps	$rndkey0,$inout0
+	xorps	$rndkey0,$inout1
+	xorps	$rndkey0,$inout2
+	$movkey	32($key),$rndkey0
+	lea	32($key,$rounds),$key
+	neg	%rax				# $rounds
+	add	\$16,%rax
+
+.L${dir}_loop3:
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+	aes${dir}	$rndkey0,$inout0
+	aes${dir}	$rndkey0,$inout1
+	aes${dir}	$rndkey0,$inout2
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.L${dir}_loop3
+
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	ret
+.size	_aesni_${dir}rypt3,.-_aesni_${dir}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 $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-3] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt4,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt4:
+	$movkey	($key),$rndkey0
+	shl	\$4,$rounds
+	$movkey	16($key),$rndkey1
+	xorps	$rndkey0,$inout0
+	xorps	$rndkey0,$inout1
+	xorps	$rndkey0,$inout2
+	xorps	$rndkey0,$inout3
+	$movkey	32($key),$rndkey0
+	lea	32($key,$rounds),$key
+	neg	%rax				# $rounds
+	.byte	0x0f,0x1f,0x00
+	add	\$16,%rax
+
+.L${dir}_loop4:
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+	aes${dir}	$rndkey0,$inout0
+	aes${dir}	$rndkey0,$inout1
+	aes${dir}	$rndkey0,$inout2
+	aes${dir}	$rndkey0,$inout3
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.L${dir}_loop4
+
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	aes${dir}last	$rndkey0,$inout3
+	ret
+.size	_aesni_${dir}rypt4,.-_aesni_${dir}rypt4
+___
+}
+sub aesni_generate6 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-5] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt6,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt6:
+	$movkey		($key),$rndkey0
+	shl		\$4,$rounds
+	$movkey		16($key),$rndkey1
+	xorps		$rndkey0,$inout0
+	pxor		$rndkey0,$inout1
+	pxor		$rndkey0,$inout2
+	aes${dir}	$rndkey1,$inout0
+	lea		32($key,$rounds),$key
+	neg		%rax			# $rounds
+	aes${dir}	$rndkey1,$inout1
+	pxor		$rndkey0,$inout3
+	pxor		$rndkey0,$inout4
+	aes${dir}	$rndkey1,$inout2
+	pxor		$rndkey0,$inout5
+	add		\$16,%rax
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+	$movkey		-16($key,%rax),$rndkey0
+	jmp		.L${dir}_loop6_enter
+.align	16
+.L${dir}_loop6:
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+.L${dir}_loop6_enter:
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+	aes${dir}	$rndkey0,$inout0
+	aes${dir}	$rndkey0,$inout1
+	aes${dir}	$rndkey0,$inout2
+	aes${dir}	$rndkey0,$inout3
+	aes${dir}	$rndkey0,$inout4
+	aes${dir}	$rndkey0,$inout5
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.L${dir}_loop6
+
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	aes${dir}last	$rndkey0,$inout3
+	aes${dir}last	$rndkey0,$inout4
+	aes${dir}last	$rndkey0,$inout5
+	ret
+.size	_aesni_${dir}rypt6,.-_aesni_${dir}rypt6
+___
+}
+sub aesni_generate8 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-7] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt8,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt8:
+	$movkey		($key),$rndkey0
+	shl		\$4,$rounds
+	$movkey		16($key),$rndkey1
+	xorps		$rndkey0,$inout0
+	xorps		$rndkey0,$inout1
+	pxor		$rndkey0,$inout2
+	pxor		$rndkey0,$inout3
+	pxor		$rndkey0,$inout4
+	lea		32($key,$rounds),$key
+	neg		%rax			# $rounds
+	aes${dir}	$rndkey1,$inout0
+	add		\$16,%rax
+	pxor		$rndkey0,$inout5
+	aes${dir}	$rndkey1,$inout1
+	pxor		$rndkey0,$inout6
+	pxor		$rndkey0,$inout7
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+	aes${dir}	$rndkey1,$inout6
+	aes${dir}	$rndkey1,$inout7
+	$movkey		-16($key,%rax),$rndkey0
+	jmp		.L${dir}_loop8_enter
+.align	16
+.L${dir}_loop8:
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+	aes${dir}	$rndkey1,$inout6
+	aes${dir}	$rndkey1,$inout7
+.L${dir}_loop8_enter:
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+	aes${dir}	$rndkey0,$inout0
+	aes${dir}	$rndkey0,$inout1
+	aes${dir}	$rndkey0,$inout2
+	aes${dir}	$rndkey0,$inout3
+	aes${dir}	$rndkey0,$inout4
+	aes${dir}	$rndkey0,$inout5
+	aes${dir}	$rndkey0,$inout6
+	aes${dir}	$rndkey0,$inout7
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.L${dir}_loop8
+
+	aes${dir}	$rndkey1,$inout0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout5
+	aes${dir}	$rndkey1,$inout6
+	aes${dir}	$rndkey1,$inout7
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	aes${dir}last	$rndkey0,$inout3
+	aes${dir}last	$rndkey0,$inout4
+	aes${dir}last	$rndkey0,$inout5
+	aes${dir}last	$rndkey0,$inout6
+	aes${dir}last	$rndkey0,$inout7
+	ret
+.size	_aesni_${dir}rypt8,.-_aesni_${dir}rypt8
+___
+}
+&aesni_generate2("enc") if ($PREFIX eq "aesni");
+&aesni_generate2("dec");
+&aesni_generate3("enc") if ($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");
+&aesni_generate8("enc") if ($PREFIX eq "aesni");
+&aesni_generate8("dec");
+
+if ($PREFIX eq "aesni") {
+########################################################################
+# void aesni_ecb_encrypt (const void *in, void *out,
+#			  size_t length, const AES_KEY *key,
+#			  int enc);
+$code.=<<___;
+.globl	aesni_ecb_encrypt
+.type	aesni_ecb_encrypt,\@function,5
+.align	16
+aesni_ecb_encrypt:
+___
+$code.=<<___ if ($win64);
+	lea	-0x58(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+.Lecb_enc_body:
+___
+$code.=<<___;
+	and	\$-16,$len
+	jz	.Lecb_ret
+
+	mov	240($key),$rounds	# key->rounds
+	$movkey	($key),$rndkey0
+	mov	$key,$key_		# backup $key
+	mov	$rounds,$rnds_		# backup $rounds
+	test	%r8d,%r8d		# 5th argument
+	jz	.Lecb_decrypt
+#--------------------------- ECB ENCRYPT ------------------------------#
+	cmp	\$0x80,$len
+	jb	.Lecb_enc_tail
+
+	movdqu	($inp),$inout0
+	movdqu	0x10($inp),$inout1
+	movdqu	0x20($inp),$inout2
+	movdqu	0x30($inp),$inout3
+	movdqu	0x40($inp),$inout4
+	movdqu	0x50($inp),$inout5
+	movdqu	0x60($inp),$inout6
+	movdqu	0x70($inp),$inout7
+	lea	0x80($inp),$inp
+	sub	\$0x80,$len
+	jmp	.Lecb_enc_loop8_enter
+.align 16
+.Lecb_enc_loop8:
+	movups	$inout0,($out)
+	mov	$key_,$key		# restore $key
+	movdqu	($inp),$inout0
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout1,0x10($out)
+	movdqu	0x10($inp),$inout1
+	movups	$inout2,0x20($out)
+	movdqu	0x20($inp),$inout2
+	movups	$inout3,0x30($out)
+	movdqu	0x30($inp),$inout3
+	movups	$inout4,0x40($out)
+	movdqu	0x40($inp),$inout4
+	movups	$inout5,0x50($out)
+	movdqu	0x50($inp),$inout5
+	movups	$inout6,0x60($out)
+	movdqu	0x60($inp),$inout6
+	movups	$inout7,0x70($out)
+	lea	0x80($out),$out
+	movdqu	0x70($inp),$inout7
+	lea	0x80($inp),$inp
+.Lecb_enc_loop8_enter:
+
+	call	_aesni_encrypt8
+
+	sub	\$0x80,$len
+	jnc	.Lecb_enc_loop8
+
+	movups	$inout0,($out)
+	mov	$key_,$key		# restore $key
+	movups	$inout1,0x10($out)
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+	movups	$inout6,0x60($out)
+	movups	$inout7,0x70($out)
+	lea	0x80($out),$out
+	add	\$0x80,$len
+	jz	.Lecb_ret
+
+.Lecb_enc_tail:
+	movups	($inp),$inout0
+	cmp	\$0x20,$len
+	jb	.Lecb_enc_one
+	movups	0x10($inp),$inout1
+	je	.Lecb_enc_two
+	movups	0x20($inp),$inout2
+	cmp	\$0x40,$len
+	jb	.Lecb_enc_three
+	movups	0x30($inp),$inout3
+	je	.Lecb_enc_four
+	movups	0x40($inp),$inout4
+	cmp	\$0x60,$len
+	jb	.Lecb_enc_five
+	movups	0x50($inp),$inout5
+	je	.Lecb_enc_six
+	movdqu	0x60($inp),$inout6
+	call	_aesni_encrypt8
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+	movups	$inout6,0x60($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_one:
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_two:
+	call	_aesni_encrypt2
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_three:
+	call	_aesni_encrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_four:
+	call	_aesni_encrypt4
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_five:
+	xorps	$inout5,$inout5
+	call	_aesni_encrypt6
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_six:
+	call	_aesni_encrypt6
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+	jmp	.Lecb_ret
+#--------------------------- ECB DECRYPT ------------------------------#
+.align	16
+.Lecb_decrypt:
+	cmp	\$0x80,$len
+	jb	.Lecb_dec_tail
+
+	movdqu	($inp),$inout0
+	movdqu	0x10($inp),$inout1
+	movdqu	0x20($inp),$inout2
+	movdqu	0x30($inp),$inout3
+	movdqu	0x40($inp),$inout4
+	movdqu	0x50($inp),$inout5
+	movdqu	0x60($inp),$inout6
+	movdqu	0x70($inp),$inout7
+	lea	0x80($inp),$inp
+	sub	\$0x80,$len
+	jmp	.Lecb_dec_loop8_enter
+.align 16
+.Lecb_dec_loop8:
+	movups	$inout0,($out)
+	mov	$key_,$key		# restore $key
+	movdqu	($inp),$inout0
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout1,0x10($out)
+	movdqu	0x10($inp),$inout1
+	movups	$inout2,0x20($out)
+	movdqu	0x20($inp),$inout2
+	movups	$inout3,0x30($out)
+	movdqu	0x30($inp),$inout3
+	movups	$inout4,0x40($out)
+	movdqu	0x40($inp),$inout4
+	movups	$inout5,0x50($out)
+	movdqu	0x50($inp),$inout5
+	movups	$inout6,0x60($out)
+	movdqu	0x60($inp),$inout6
+	movups	$inout7,0x70($out)
+	lea	0x80($out),$out
+	movdqu	0x70($inp),$inout7
+	lea	0x80($inp),$inp
+.Lecb_dec_loop8_enter:
+
+	call	_aesni_decrypt8
+
+	$movkey	($key_),$rndkey0
+	sub	\$0x80,$len
+	jnc	.Lecb_dec_loop8
+
+	movups	$inout0,($out)
+	mov	$key_,$key		# restore $key
+	movups	$inout1,0x10($out)
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+	movups	$inout6,0x60($out)
+	movups	$inout7,0x70($out)
+	lea	0x80($out),$out
+	add	\$0x80,$len
+	jz	.Lecb_ret
+
+.Lecb_dec_tail:
+	movups	($inp),$inout0
+	cmp	\$0x20,$len
+	jb	.Lecb_dec_one
+	movups	0x10($inp),$inout1
+	je	.Lecb_dec_two
+	movups	0x20($inp),$inout2
+	cmp	\$0x40,$len
+	jb	.Lecb_dec_three
+	movups	0x30($inp),$inout3
+	je	.Lecb_dec_four
+	movups	0x40($inp),$inout4
+	cmp	\$0x60,$len
+	jb	.Lecb_dec_five
+	movups	0x50($inp),$inout5
+	je	.Lecb_dec_six
+	movups	0x60($inp),$inout6
+	$movkey	($key),$rndkey0
+	call	_aesni_decrypt8
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+	movups	$inout6,0x60($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_one:
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_two:
+	call	_aesni_decrypt2
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_three:
+	call	_aesni_decrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_four:
+	call	_aesni_decrypt4
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_five:
+	xorps	$inout5,$inout5
+	call	_aesni_decrypt6
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_six:
+	call	_aesni_decrypt6
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	movups	$inout4,0x40($out)
+	movups	$inout5,0x50($out)
+
+.Lecb_ret:
+___
+$code.=<<___ if ($win64);
+	movaps	(%rsp),%xmm6
+	movaps	0x10(%rsp),%xmm7
+	movaps	0x20(%rsp),%xmm8
+	movaps	0x30(%rsp),%xmm9
+	lea	0x58(%rsp),%rsp
+.Lecb_enc_ret:
+___
+$code.=<<___;
+	ret
+.size	aesni_ecb_encrypt,.-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="%r9";	# 6th argument
+
+my $increment="%xmm9";
+my $iv="%xmm6";
+my $bswap_mask="%xmm7";
+
+$code.=<<___;
+.globl	aesni_ccm64_encrypt_blocks
+.type	aesni_ccm64_encrypt_blocks,\@function,6
+.align	16
+aesni_ccm64_encrypt_blocks:
+___
+$code.=<<___ if ($win64);
+	lea	-0x58(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+.Lccm64_enc_body:
+___
+$code.=<<___;
+	mov	240($key),$rounds		# key->rounds
+	movdqu	($ivp),$iv
+	movdqa	.Lincrement64(%rip),$increment
+	movdqa	.Lbswap_mask(%rip),$bswap_mask
+
+	shl	\$4,$rounds
+	mov	\$16,$rnds_
+	lea	0($key),$key_
+	movdqu	($cmac),$inout1
+	movdqa	$iv,$inout0
+	lea	32($key,$rounds),$key		# end of key schedule
+	pshufb	$bswap_mask,$iv
+	sub	%rax,%r10			# twisted $rounds
+	jmp	.Lccm64_enc_outer
+.align	16
+.Lccm64_enc_outer:
+	$movkey	($key_),$rndkey0
+	mov	%r10,%rax
+	movups	($inp),$in0			# load inp
+
+	xorps	$rndkey0,$inout0		# counter
+	$movkey	16($key_),$rndkey1
+	xorps	$in0,$rndkey0
+	xorps	$rndkey0,$inout1		# cmac^=inp
+	$movkey	32($key_),$rndkey0
+
+.Lccm64_enc2_loop:
+	aesenc	$rndkey1,$inout0
+	aesenc	$rndkey1,$inout1
+	$movkey	($key,%rax),$rndkey1
+	add	\$32,%rax
+	aesenc	$rndkey0,$inout0
+	aesenc	$rndkey0,$inout1
+	$movkey	-16($key,%rax),$rndkey0
+	jnz	.Lccm64_enc2_loop
+	aesenc	$rndkey1,$inout0
+	aesenc	$rndkey1,$inout1
+	paddq	$increment,$iv
+	dec	$len
+	aesenclast	$rndkey0,$inout0
+	aesenclast	$rndkey0,$inout1
+
+	lea	16($inp),$inp
+	xorps	$inout0,$in0			# inp ^= E(iv)
+	movdqa	$iv,$inout0
+	movups	$in0,($out)			# save output
+	pshufb	$bswap_mask,$inout0
+	lea	16($out),$out
+	jnz	.Lccm64_enc_outer
+
+	movups	$inout1,($cmac)
+___
+$code.=<<___ if ($win64);
+	movaps	(%rsp),%xmm6
+	movaps	0x10(%rsp),%xmm7
+	movaps	0x20(%rsp),%xmm8
+	movaps	0x30(%rsp),%xmm9
+	lea	0x58(%rsp),%rsp
+.Lccm64_enc_ret:
+___
+$code.=<<___;
+	ret
+.size	aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks
+___
+######################################################################
+$code.=<<___;
+.globl	aesni_ccm64_decrypt_blocks
+.type	aesni_ccm64_decrypt_blocks,\@function,6
+.align	16
+aesni_ccm64_decrypt_blocks:
+___
+$code.=<<___ if ($win64);
+	lea	-0x58(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+.Lccm64_dec_body:
+___
+$code.=<<___;
+	mov	240($key),$rounds		# key->rounds
+	movups	($ivp),$iv
+	movdqu	($cmac),$inout1
+	movdqa	.Lincrement64(%rip),$increment
+	movdqa	.Lbswap_mask(%rip),$bswap_mask
+
+	movaps	$iv,$inout0
+	mov	$rounds,$rnds_
+	mov	$key,$key_
+	pshufb	$bswap_mask,$iv
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	shl	\$4,$rnds_
+	mov	\$16,$rounds
+	movups	($inp),$in0			# load inp
+	paddq	$increment,$iv
+	lea	16($inp),$inp
+	sub	%r10,%rax			# twisted $rounds
+	lea	32($key_,$rnds_),$key		# end of key schedule
+	mov	%rax,%r10
+	jmp	.Lccm64_dec_outer
+.align	16
+.Lccm64_dec_outer:
+	xorps	$inout0,$in0			# inp ^= E(iv)
+	movdqa	$iv,$inout0
+	movups	$in0,($out)			# save output
+	lea	16($out),$out
+	pshufb	$bswap_mask,$inout0
+
+	sub	\$1,$len
+	jz	.Lccm64_dec_break
+
+	$movkey	($key_),$rndkey0
+	mov	%r10,%rax
+	$movkey	16($key_),$rndkey1
+	xorps	$rndkey0,$in0
+	xorps	$rndkey0,$inout0
+	xorps	$in0,$inout1			# cmac^=out
+	$movkey	32($key_),$rndkey0
+	jmp	.Lccm64_dec2_loop
+.align	16
+.Lccm64_dec2_loop:
+	aesenc	$rndkey1,$inout0
+	aesenc	$rndkey1,$inout1
+	$movkey	($key,%rax),$rndkey1
+	add	\$32,%rax
+	aesenc	$rndkey0,$inout0
+	aesenc	$rndkey0,$inout1
+	$movkey	-16($key,%rax),$rndkey0
+	jnz	.Lccm64_dec2_loop
+	movups	($inp),$in0			# load inp
+	paddq	$increment,$iv
+	aesenc	$rndkey1,$inout0
+	aesenc	$rndkey1,$inout1
+	aesenclast	$rndkey0,$inout0
+	aesenclast	$rndkey0,$inout1
+	lea	16($inp),$inp
+	jmp	.Lccm64_dec_outer
+
+.align	16
+.Lccm64_dec_break:
+	#xorps	$in0,$inout1			# cmac^=out
+	mov	240($key_),$rounds
+___
+	&aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
+$code.=<<___;
+	movups	$inout1,($cmac)
+___
+$code.=<<___ if ($win64);
+	movaps	(%rsp),%xmm6
+	movaps	0x10(%rsp),%xmm7
+	movaps	0x20(%rsp),%xmm8
+	movaps	0x30(%rsp),%xmm9
+	lea	0x58(%rsp),%rsp
+.Lccm64_dec_ret:
+___
+$code.=<<___;
+	ret
+.size	aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks
+___
+}
+######################################################################
+# void aesni_ctr32_encrypt_blocks (const void *in, void *out,
+#                         size_t blocks, const AES_KEY *key,
+#                         const char *ivec);
+#
+# Handles only complete blocks, operates on 32-bit counter and
+# does not update *ivec! (see crypto/modes/ctr128.c for details)
+#
+# Overhaul based on suggestions from Shay Gueron and Vlad Krasnov,
+# http://rt.openssl.org/Ticket/Display.html?id=3021&user=guest&pass=guest.
+# Keywords are full unroll and modulo-schedule counter calculations
+# with zero-round key xor.
+{
+my ($in0,$in1,$in2,$in3,$in4,$in5)=map("%xmm$_",(10..15));
+my ($key0,$ctr)=("${key_}d","${ivp}d");
+my $frame_size = 0x80 + ($win64?160:0);
+
+$code.=<<___;
+.globl	aesni_ctr32_encrypt_blocks
+.type	aesni_ctr32_encrypt_blocks,\@function,5
+.align	16
+aesni_ctr32_encrypt_blocks:
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,-0xa8(%rax)
+	movaps	%xmm7,-0x98(%rax)
+	movaps	%xmm8,-0x88(%rax)
+	movaps	%xmm9,-0x78(%rax)
+	movaps	%xmm10,-0x68(%rax)
+	movaps	%xmm11,-0x58(%rax)
+	movaps	%xmm12,-0x48(%rax)
+	movaps	%xmm13,-0x38(%rax)
+	movaps	%xmm14,-0x28(%rax)
+	movaps	%xmm15,-0x18(%rax)
+.Lctr32_body:
+___
+$code.=<<___;
+	lea	-8(%rax),%rbp
+
+	cmp	\$1,$len
+	je	.Lctr32_one_shortcut
+
+	movdqu	($ivp),$inout0
+	movdqu	($key),$rndkey0
+	mov	12($ivp),$ctr			# counter LSB
+	pxor	$rndkey0,$inout0
+	mov	12($key),$key0			# 0-round key LSB
+	movdqa	$inout0,0x00(%rsp)		# populate counter block
+	bswap	$ctr
+	movdqa	$inout0,$inout1
+	movdqa	$inout0,$inout2
+	movdqa	$inout0,$inout3
+	movdqa	$inout0,0x40(%rsp)
+	movdqa	$inout0,0x50(%rsp)
+	movdqa	$inout0,0x60(%rsp)
+	mov	%rdx,%r10			# borrow %rdx
+	movdqa	$inout0,0x70(%rsp)
+
+	lea	1($ctr),%rax
+	 lea	2($ctr),%rdx
+	bswap	%eax
+	 bswap	%edx
+	xor	$key0,%eax
+	 xor	$key0,%edx
+	pinsrd	\$3,%eax,$inout1
+	lea	3($ctr),%rax
+	movdqa	$inout1,0x10(%rsp)
+	 pinsrd	\$3,%edx,$inout2
+	bswap	%eax
+	 mov	%r10,%rdx			# restore %rdx
+	 lea	4($ctr),%r10
+	 movdqa	$inout2,0x20(%rsp)
+	xor	$key0,%eax
+	 bswap	%r10d
+	pinsrd	\$3,%eax,$inout3
+	 xor	$key0,%r10d
+	movdqa	$inout3,0x30(%rsp)
+	lea	5($ctr),%r9
+	 mov	%r10d,0x40+12(%rsp)
+	bswap	%r9d
+	 lea	6($ctr),%r10
+	mov	240($key),$rounds		# key->rounds
+	xor	$key0,%r9d
+	 bswap	%r10d
+	mov	%r9d,0x50+12(%rsp)
+	 xor	$key0,%r10d
+	lea	7($ctr),%r9
+	 mov	%r10d,0x60+12(%rsp)
+	bswap	%r9d
+	 mov	OPENSSL_ia32cap_P+4(%rip),%r10d 
+	xor	$key0,%r9d
+	 and	\$`1<<26|1<<22`,%r10d		# isolate XSAVE+MOVBE
+	mov	%r9d,0x70+12(%rsp)
+
+	$movkey	0x10($key),$rndkey1
+
+	movdqa	0x40(%rsp),$inout4
+	movdqa	0x50(%rsp),$inout5
+
+	cmp	\$8,$len
+	jb	.Lctr32_tail
+
+	sub	\$6,$len
+	cmp	\$`1<<22`,%r10d		# check for MOVBE without XSAVE
+	je	.Lctr32_6x
+
+	lea	0x80($key),$key		# size optimization
+	sub	\$2,$len
+	jmp	.Lctr32_loop8
+
+.align	16
+.Lctr32_6x:
+	shl	\$4,$rounds
+	mov	\$48,$rnds_
+	bswap	$key0
+	lea	32($key,$rounds),$key	# end of key schedule
+	sub	%rax,%r10		# twisted $rounds
+	jmp	.Lctr32_loop6
+
+.align	16
+.Lctr32_loop6:
+	 add	\$6,$ctr
+	$movkey	-48($key,$rnds_),$rndkey0
+	aesenc	$rndkey1,$inout0
+	 mov	$ctr,%eax
+	 xor	$key0,%eax
+	aesenc	$rndkey1,$inout1
+	 movbe	%eax,`0x00+12`(%rsp)
+	 lea	1($ctr),%eax
+	aesenc	$rndkey1,$inout2
+	 xor	$key0,%eax
+	 movbe	%eax,`0x10+12`(%rsp)
+	aesenc	$rndkey1,$inout3
+	 lea	2($ctr),%eax
+	 xor	$key0,%eax
+	aesenc	$rndkey1,$inout4
+	 movbe	%eax,`0x20+12`(%rsp)
+	 lea	3($ctr),%eax
+	aesenc	$rndkey1,$inout5
+	$movkey	-32($key,$rnds_),$rndkey1
+	 xor	$key0,%eax
+
+	aesenc	$rndkey0,$inout0
+	 movbe	%eax,`0x30+12`(%rsp)
+	 lea	4($ctr),%eax
+	aesenc	$rndkey0,$inout1
+	 xor	$key0,%eax
+	 movbe	%eax,`0x40+12`(%rsp)
+	aesenc	$rndkey0,$inout2
+	 lea	5($ctr),%eax
+	 xor	$key0,%eax
+	aesenc	$rndkey0,$inout3
+	 movbe	%eax,`0x50+12`(%rsp)
+	 mov	%r10,%rax		# mov	$rnds_,$rounds
+	aesenc	$rndkey0,$inout4
+	aesenc	$rndkey0,$inout5
+	$movkey	-16($key,$rnds_),$rndkey0
+
+	call	.Lenc_loop6
+
+	movdqu	($inp),$inout6
+	movdqu	0x10($inp),$inout7
+	movdqu	0x20($inp),$in0
+	movdqu	0x30($inp),$in1
+	movdqu	0x40($inp),$in2
+	movdqu	0x50($inp),$in3
+	lea	0x60($inp),$inp
+	$movkey	-64($key,$rnds_),$rndkey1
+	pxor	$inout0,$inout6
+	movaps	0x00(%rsp),$inout0
+	pxor	$inout1,$inout7
+	movaps	0x10(%rsp),$inout1
+	pxor	$inout2,$in0
+	movaps	0x20(%rsp),$inout2
+	pxor	$inout3,$in1
+	movaps	0x30(%rsp),$inout3
+	pxor	$inout4,$in2
+	movaps	0x40(%rsp),$inout4
+	pxor	$inout5,$in3
+	movaps	0x50(%rsp),$inout5
+	movdqu	$inout6,($out)
+	movdqu	$inout7,0x10($out)
+	movdqu	$in0,0x20($out)
+	movdqu	$in1,0x30($out)
+	movdqu	$in2,0x40($out)
+	movdqu	$in3,0x50($out)
+	lea	0x60($out),$out
+	
+	sub	\$6,$len
+	jnc	.Lctr32_loop6
+
+	add	\$6,$len
+	jz	.Lctr32_done
+
+	lea	-48($rnds_),$rounds
+	lea	-80($key,$rnds_),$key	# restore $key
+	neg	$rounds
+	shr	\$4,$rounds		# restore $rounds
+	jmp	.Lctr32_tail
+
+.align	32
+.Lctr32_loop8:
+	 add		\$8,$ctr
+	movdqa		0x60(%rsp),$inout6
+	aesenc		$rndkey1,$inout0
+	 mov		$ctr,%r9d
+	movdqa		0x70(%rsp),$inout7
+	aesenc		$rndkey1,$inout1
+	 bswap		%r9d
+	$movkey		0x20-0x80($key),$rndkey0
+	aesenc		$rndkey1,$inout2
+	 xor		$key0,%r9d
+	 nop
+	aesenc		$rndkey1,$inout3
+	 mov		%r9d,0x00+12(%rsp)
+	 lea		1($ctr),%r9
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	$movkey		0x30-0x80($key),$rndkey1
+___
+for($i=2;$i<8;$i++) {
+my $rndkeyx = ($i&1)?$rndkey1:$rndkey0;
+$code.=<<___;
+	 bswap		%r9d
+	aesenc		$rndkeyx,$inout0
+	aesenc		$rndkeyx,$inout1
+	 xor		$key0,%r9d
+	 .byte		0x66,0x90
+	aesenc		$rndkeyx,$inout2
+	aesenc		$rndkeyx,$inout3
+	 mov		%r9d,`0x10*($i-1)`+12(%rsp)
+	 lea		$i($ctr),%r9
+	aesenc		$rndkeyx,$inout4
+	aesenc		$rndkeyx,$inout5
+	aesenc		$rndkeyx,$inout6
+	aesenc		$rndkeyx,$inout7
+	$movkey		`0x20+0x10*$i`-0x80($key),$rndkeyx
+___
+}
+$code.=<<___;
+	 bswap		%r9d
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	 xor		$key0,%r9d
+	 movdqu		0x00($inp),$in0
+	aesenc		$rndkey0,$inout3
+	 mov		%r9d,0x70+12(%rsp)
+	 cmp		\$11,$rounds
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	aesenc		$rndkey0,$inout6
+	aesenc		$rndkey0,$inout7
+	$movkey		0xa0-0x80($key),$rndkey0
+
+	jb		.Lctr32_enc_done
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	$movkey		0xb0-0x80($key),$rndkey1
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	aesenc		$rndkey0,$inout6
+	aesenc		$rndkey0,$inout7
+	$movkey		0xc0-0x80($key),$rndkey0
+	je		.Lctr32_enc_done
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	$movkey		0xd0-0x80($key),$rndkey1
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	aesenc		$rndkey0,$inout6
+	aesenc		$rndkey0,$inout7
+	$movkey		0xe0-0x80($key),$rndkey0
+	jmp		.Lctr32_enc_done
+
+.align	16
+.Lctr32_enc_done:
+	movdqu		0x10($inp),$in1
+	pxor		$rndkey0,$in0
+	movdqu		0x20($inp),$in2
+	pxor		$rndkey0,$in1
+	movdqu		0x30($inp),$in3
+	pxor		$rndkey0,$in2
+	movdqu		0x40($inp),$in4
+	pxor		$rndkey0,$in3
+	movdqu		0x50($inp),$in5
+	pxor		$rndkey0,$in4
+	pxor		$rndkey0,$in5
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	movdqu		0x60($inp),$rndkey1
+	lea		0x80($inp),$inp
+
+	aesenclast	$in0,$inout0
+	pxor		$rndkey0,$rndkey1
+	movdqu		0x70-0x80($inp),$in0
+	aesenclast	$in1,$inout1
+	pxor		$rndkey0,$in0
+	movdqa		0x00(%rsp),$in1		# load next counter block
+	aesenclast	$in2,$inout2
+	aesenclast	$in3,$inout3
+	movdqa		0x10(%rsp),$in2
+	movdqa		0x20(%rsp),$in3
+	aesenclast	$in4,$inout4
+	aesenclast	$in5,$inout5
+	movdqa		0x30(%rsp),$in4
+	movdqa		0x40(%rsp),$in5
+	aesenclast	$rndkey1,$inout6
+	movdqa		0x50(%rsp),$rndkey0
+	$movkey		0x10-0x80($key),$rndkey1
+	aesenclast	$in0,$inout7
+
+	movups		$inout0,($out)		# store output
+	movdqa		$in1,$inout0
+	movups		$inout1,0x10($out)
+	movdqa		$in2,$inout1
+	movups		$inout2,0x20($out)
+	movdqa		$in3,$inout2
+	movups		$inout3,0x30($out)
+	movdqa		$in4,$inout3
+	movups		$inout4,0x40($out)
+	movdqa		$in5,$inout4
+	movups		$inout5,0x50($out)
+	movdqa		$rndkey0,$inout5
+	movups		$inout6,0x60($out)
+	movups		$inout7,0x70($out)
+	lea		0x80($out),$out
+	
+	sub	\$8,$len
+	jnc	.Lctr32_loop8
+
+	add	\$8,$len
+	jz	.Lctr32_done
+	lea	-0x80($key),$key
+
+.Lctr32_tail:
+	lea	16($key),$key
+	cmp	\$4,$len
+	jb	.Lctr32_loop3
+	je	.Lctr32_loop4
+
+	shl		\$4,$rounds
+	movdqa		0x60(%rsp),$inout6
+	pxor		$inout7,$inout7
+
+	$movkey		16($key),$rndkey0
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	lea		32-16($key,$rounds),$key
+	neg		%rax
+	aesenc		$rndkey1,$inout2
+	add		\$16,%rax
+	 movups		($inp),$in0
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	 movups		0x10($inp),$in1
+	 movups		0x20($inp),$in2
+	aesenc		$rndkey1,$inout5
+	aesenc		$rndkey1,$inout6
+
+	call            .Lenc_loop8_enter
+
+	movdqu	0x30($inp),$in3
+	pxor	$in0,$inout0
+	movdqu	0x40($inp),$in0
+	pxor	$in1,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in2,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in3,$inout3
+	movdqu	$inout2,0x20($out)
+	pxor	$in0,$inout4
+	movdqu	$inout3,0x30($out)
+	movdqu	$inout4,0x40($out)
+	cmp	\$6,$len
+	jb	.Lctr32_done
+
+	movups	0x50($inp),$in1
+	xorps	$in1,$inout5
+	movups	$inout5,0x50($out)
+	je	.Lctr32_done
+
+	movups	0x60($inp),$in2
+	xorps	$in2,$inout6
+	movups	$inout6,0x60($out)
+	jmp	.Lctr32_done
+
+.align	32
+.Lctr32_loop4:
+	aesenc		$rndkey1,$inout0
+	lea		16($key),$key
+	dec		$rounds
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	$movkey		($key),$rndkey1
+	jnz		.Lctr32_loop4
+	aesenclast	$rndkey1,$inout0
+	aesenclast	$rndkey1,$inout1
+	 movups		($inp),$in0
+	 movups		0x10($inp),$in1
+	aesenclast	$rndkey1,$inout2
+	aesenclast	$rndkey1,$inout3
+	 movups		0x20($inp),$in2
+	 movups		0x30($inp),$in3
+
+	xorps	$in0,$inout0
+	movups	$inout0,($out)
+	xorps	$in1,$inout1
+	movups	$inout1,0x10($out)
+	pxor	$in2,$inout2
+	movdqu	$inout2,0x20($out)
+	pxor	$in3,$inout3
+	movdqu	$inout3,0x30($out)
+	jmp	.Lctr32_done
+
+.align	32
+.Lctr32_loop3:
+	aesenc		$rndkey1,$inout0
+	lea		16($key),$key
+	dec		$rounds
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	$movkey		($key),$rndkey1
+	jnz		.Lctr32_loop3
+	aesenclast	$rndkey1,$inout0
+	aesenclast	$rndkey1,$inout1
+	aesenclast	$rndkey1,$inout2
+
+	movups	($inp),$in0
+	xorps	$in0,$inout0
+	movups	$inout0,($out)
+	cmp	\$2,$len
+	jb	.Lctr32_done
+
+	movups	0x10($inp),$in1
+	xorps	$in1,$inout1
+	movups	$inout1,0x10($out)
+	je	.Lctr32_done
+
+	movups	0x20($inp),$in2
+	xorps	$in2,$inout2
+	movups	$inout2,0x20($out)
+	jmp	.Lctr32_done
+
+.align	16
+.Lctr32_one_shortcut:
+	movups	($ivp),$inout0
+	movups	($inp),$in0
+	mov	240($key),$rounds		# key->rounds
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	xorps	$in0,$inout0
+	movups	$inout0,($out)
+	jmp	.Lctr32_done
+
+.align	16
+.Lctr32_done:
+___
+$code.=<<___ if ($win64);
+	movaps	-0xa0(%rbp),%xmm6
+	movaps	-0x90(%rbp),%xmm7
+	movaps	-0x80(%rbp),%xmm8
+	movaps	-0x70(%rbp),%xmm9
+	movaps	-0x60(%rbp),%xmm10
+	movaps	-0x50(%rbp),%xmm11
+	movaps	-0x40(%rbp),%xmm12
+	movaps	-0x30(%rbp),%xmm13
+	movaps	-0x20(%rbp),%xmm14
+	movaps	-0x10(%rbp),%xmm15
+___
+$code.=<<___;
+	lea	(%rbp),%rsp
+	pop	%rbp
+.Lctr32_epilogue:
+	ret
+.size	aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
+___
+}
+
+######################################################################
+# void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
+#	const AES_KEY *key1, const AES_KEY *key2
+#	const unsigned char iv[16]);
+#
+{
+my @tweak=map("%xmm$_",(10..15));
+my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
+my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
+my $frame_size = 0x70 + ($win64?160:0);
+
+$code.=<<___;
+.globl	aesni_xts_encrypt
+.type	aesni_xts_encrypt,\@function,6
+.align	16
+aesni_xts_encrypt:
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,-0xa8(%rax)
+	movaps	%xmm7,-0x98(%rax)
+	movaps	%xmm8,-0x88(%rax)
+	movaps	%xmm9,-0x78(%rax)
+	movaps	%xmm10,-0x68(%rax)
+	movaps	%xmm11,-0x58(%rax)
+	movaps	%xmm12,-0x48(%rax)
+	movaps	%xmm13,-0x38(%rax)
+	movaps	%xmm14,-0x28(%rax)
+	movaps	%xmm15,-0x18(%rax)
+.Lxts_enc_body:
+___
+$code.=<<___;
+	lea	-8(%rax),%rbp
+	movups	($ivp),$inout0			# load clear-text tweak
+	mov	240(%r8),$rounds		# key2->rounds
+	mov	240($key),$rnds_		# key1->rounds
+___
+	# generate the tweak
+	&aesni_generate1("enc",$key2,$rounds,$inout0);
+$code.=<<___;
+	$movkey	($key),$rndkey0			# zero round key
+	mov	$key,$key_			# backup $key
+	mov	$rnds_,$rounds			# backup $rounds
+	shl	\$4,$rnds_
+	mov	$len,$len_			# backup $len
+	and	\$-16,$len
+
+	$movkey	16($key,$rnds_),$rndkey1	# last round key
+
+	movdqa	.Lxts_magic(%rip),$twmask
+	movdqa	$inout0,@tweak[5]
+	pshufd	\$0x5f,$inout0,$twres
+	pxor	$rndkey0,$rndkey1
+___
+    # alternative tweak calculation algorithm is based on suggestions
+    # by Shay Gueron. psrad doesn't conflict with AES-NI instructions
+    # and should help in the future...
+    for ($i=0;$i<4;$i++) {
+    $code.=<<___;
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
+	movdqa	@tweak[5],@tweak[$i]
+	psrad	\$31,$twtmp			# broadcast upper bits
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	pxor	$rndkey0,@tweak[$i]
+	pxor	$twtmp,@tweak[5]
+___
+    }
+$code.=<<___;
+	movdqa	@tweak[5],@tweak[4]
+	psrad	\$31,$twres
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twres
+	pxor	$rndkey0,@tweak[4]
+	pxor	$twres,@tweak[5]
+	movaps	$rndkey1,0x60(%rsp)		# save round[0]^round[last]
+
+	sub	\$16*6,$len
+	jc	.Lxts_enc_short
+
+	mov	\$16+96,$rounds
+	lea	32($key_,$rnds_),$key		# end of key schedule
+	sub	%r10,%rax			# twisted $rounds
+	$movkey	16($key_),$rndkey1
+	mov	%rax,%r10			# backup twisted $rounds
+	lea	.Lxts_magic(%rip),%r8
+	jmp	.Lxts_enc_grandloop
+
+.align	32
+.Lxts_enc_grandloop:
+	movdqu	`16*0`($inp),$inout0		# load input
+	movdqa	$rndkey0,$twmask
+	movdqu	`16*1`($inp),$inout1
+	pxor	@tweak[0],$inout0
+	movdqu	`16*2`($inp),$inout2
+	pxor	@tweak[1],$inout1
+	 aesenc		$rndkey1,$inout0
+	movdqu	`16*3`($inp),$inout3
+	pxor	@tweak[2],$inout2
+	 aesenc		$rndkey1,$inout1
+	movdqu	`16*4`($inp),$inout4
+	pxor	@tweak[3],$inout3
+	 aesenc		$rndkey1,$inout2
+	movdqu	`16*5`($inp),$inout5
+	pxor	@tweak[5],$twmask		# round[0]^=tweak[5]
+	 movdqa	0x60(%rsp),$twres		# load round[0]^round[last]
+	pxor	@tweak[4],$inout4
+	 aesenc		$rndkey1,$inout3
+	$movkey	32($key_),$rndkey0
+	lea	`16*6`($inp),$inp
+	pxor	$twmask,$inout5
+
+	 pxor	$twres,@tweak[0]
+	aesenc		$rndkey1,$inout4
+	 pxor	$twres,@tweak[1]
+	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks^last round key
+	aesenc		$rndkey1,$inout5
+	$movkey		48($key_),$rndkey1
+	 pxor	$twres,@tweak[2]
+
+	aesenc		$rndkey0,$inout0
+	 pxor	$twres,@tweak[3]
+	 movdqa	@tweak[1],`16*1`(%rsp)
+	aesenc		$rndkey0,$inout1
+	 pxor	$twres,@tweak[4]
+	 movdqa	@tweak[2],`16*2`(%rsp)
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	 pxor	$twres,$twmask
+	 movdqa	@tweak[4],`16*4`(%rsp)
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	$movkey		64($key_),$rndkey0
+	 movdqa	$twmask,`16*5`(%rsp)
+	pshufd	\$0x5f,@tweak[5],$twres
+	jmp	.Lxts_enc_loop6
+.align	32
+.Lxts_enc_loop6:
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	$movkey		-64($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	$movkey		-80($key,%rax),$rndkey0
+	jnz		.Lxts_enc_loop6
+
+	movdqa	(%r8),$twmask
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
+	 aesenc		$rndkey1,$inout0
+	paddq	@tweak[5],@tweak[5]
+	psrad	\$31,$twtmp
+	 aesenc		$rndkey1,$inout1
+	pand	$twmask,$twtmp
+	$movkey	($key_),@tweak[0]		# load round[0]
+	 aesenc		$rndkey1,$inout2
+	 aesenc		$rndkey1,$inout3
+	 aesenc		$rndkey1,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[0],@tweak[1]		# copy round[0]
+	 aesenc		$rndkey1,$inout5
+	 $movkey	-64($key),$rndkey1
+
+	movdqa	$twres,$twtmp
+	 aesenc		$rndkey0,$inout0
+	paddd	$twres,$twres
+	pxor	@tweak[5],@tweak[0]
+	 aesenc		$rndkey0,$inout1
+	psrad	\$31,$twtmp
+	paddq	@tweak[5],@tweak[5]
+	 aesenc		$rndkey0,$inout2
+	 aesenc		$rndkey0,$inout3
+	pand	$twmask,$twtmp
+	movaps	@tweak[1],@tweak[2]
+	 aesenc		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movdqa	$twres,$twtmp
+	 aesenc		$rndkey0,$inout5
+	 $movkey	-48($key),$rndkey0
+
+	paddd	$twres,$twres
+	 aesenc		$rndkey1,$inout0
+	pxor	@tweak[5],@tweak[1]
+	psrad	\$31,$twtmp
+	 aesenc		$rndkey1,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	 aesenc		$rndkey1,$inout2
+	 aesenc		$rndkey1,$inout3
+	 movdqa	@tweak[3],`16*3`(%rsp)
+	pxor	$twtmp,@tweak[5]
+	 aesenc		$rndkey1,$inout4
+	movaps	@tweak[2],@tweak[3]
+	movdqa	$twres,$twtmp
+	 aesenc		$rndkey1,$inout5
+	 $movkey	-32($key),$rndkey1
+
+	paddd	$twres,$twres
+	 aesenc		$rndkey0,$inout0
+	pxor	@tweak[5],@tweak[2]
+	psrad	\$31,$twtmp
+	 aesenc		$rndkey0,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	 aesenc		$rndkey0,$inout2
+	 aesenc		$rndkey0,$inout3
+	 aesenc		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[3],@tweak[4]
+	 aesenc		$rndkey0,$inout5
+
+	movdqa	$twres,$rndkey0
+	paddd	$twres,$twres
+	 aesenc		$rndkey1,$inout0
+	pxor	@tweak[5],@tweak[3]
+	psrad	\$31,$rndkey0
+	 aesenc		$rndkey1,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$rndkey0
+	 aesenc		$rndkey1,$inout2
+	 aesenc		$rndkey1,$inout3
+	pxor	$rndkey0,@tweak[5]
+	$movkey		($key_),$rndkey0
+	 aesenc		$rndkey1,$inout4
+	 aesenc		$rndkey1,$inout5
+	$movkey		16($key_),$rndkey1
+
+	pxor	@tweak[5],@tweak[4]
+	 aesenclast	`16*0`(%rsp),$inout0
+	psrad	\$31,$twres
+	paddq	@tweak[5],@tweak[5]
+	 aesenclast	`16*1`(%rsp),$inout1
+	 aesenclast	`16*2`(%rsp),$inout2
+	pand	$twmask,$twres
+	mov	%r10,%rax			# restore $rounds
+	 aesenclast	`16*3`(%rsp),$inout3
+	 aesenclast	`16*4`(%rsp),$inout4
+	 aesenclast	`16*5`(%rsp),$inout5
+	pxor	$twres,@tweak[5]
+
+	lea	`16*6`($out),$out
+	movups	$inout0,`-16*6`($out)		# write output
+	movups	$inout1,`-16*5`($out)
+	movups	$inout2,`-16*4`($out)
+	movups	$inout3,`-16*3`($out)
+	movups	$inout4,`-16*2`($out)
+	movups	$inout5,`-16*1`($out)
+	sub	\$16*6,$len
+	jnc	.Lxts_enc_grandloop
+
+	mov	\$16+96,$rounds
+	sub	$rnds_,$rounds
+	mov	$key_,$key			# restore $key
+	shr	\$4,$rounds			# restore original value
+
+.Lxts_enc_short:
+	mov	$rounds,$rnds_			# backup $rounds
+	pxor	$rndkey0,@tweak[0]
+	add	\$16*6,$len
+	jz	.Lxts_enc_done
+
+	pxor	$rndkey0,@tweak[1]
+	cmp	\$0x20,$len
+	jb	.Lxts_enc_one
+	pxor	$rndkey0,@tweak[2]
+	je	.Lxts_enc_two
+
+	pxor	$rndkey0,@tweak[3]
+	cmp	\$0x40,$len
+	jb	.Lxts_enc_three
+	pxor	$rndkey0,@tweak[4]
+	je	.Lxts_enc_four
+
+	movdqu	($inp),$inout0
+	movdqu	16*1($inp),$inout1
+	movdqu	16*2($inp),$inout2
+	pxor	@tweak[0],$inout0
+	movdqu	16*3($inp),$inout3
+	pxor	@tweak[1],$inout1
+	movdqu	16*4($inp),$inout4
+	lea	16*5($inp),$inp
+	pxor	@tweak[2],$inout2
+	pxor	@tweak[3],$inout3
+	pxor	@tweak[4],$inout4
+
+	call	_aesni_encrypt6
+
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[5],@tweak[0]
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+	movdqu	$inout0,($out)
+	xorps	@tweak[3],$inout3
+	movdqu	$inout1,16*1($out)
+	xorps	@tweak[4],$inout4
+	movdqu	$inout2,16*2($out)
+	movdqu	$inout3,16*3($out)
+	movdqu	$inout4,16*4($out)
+	lea	16*5($out),$out
+	jmp	.Lxts_enc_done
+
+.align	16
+.Lxts_enc_one:
+	movups	($inp),$inout0
+	lea	16*1($inp),$inp
+	xorps	@tweak[0],$inout0
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[1],@tweak[0]
+	movups	$inout0,($out)
+	lea	16*1($out),$out
+	jmp	.Lxts_enc_done
+
+.align	16
+.Lxts_enc_two:
+	movups	($inp),$inout0
+	movups	16($inp),$inout1
+	lea	32($inp),$inp
+	xorps	@tweak[0],$inout0
+	xorps	@tweak[1],$inout1
+
+	call	_aesni_encrypt2
+
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[2],@tweak[0]
+	xorps	@tweak[1],$inout1
+	movups	$inout0,($out)
+	movups	$inout1,16*1($out)
+	lea	16*2($out),$out
+	jmp	.Lxts_enc_done
+
+.align	16
+.Lxts_enc_three:
+	movups	($inp),$inout0
+	movups	16*1($inp),$inout1
+	movups	16*2($inp),$inout2
+	lea	16*3($inp),$inp
+	xorps	@tweak[0],$inout0
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+
+	call	_aesni_encrypt3
+
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[3],@tweak[0]
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+	movups	$inout0,($out)
+	movups	$inout1,16*1($out)
+	movups	$inout2,16*2($out)
+	lea	16*3($out),$out
+	jmp	.Lxts_enc_done
+
+.align	16
+.Lxts_enc_four:
+	movups	($inp),$inout0
+	movups	16*1($inp),$inout1
+	movups	16*2($inp),$inout2
+	xorps	@tweak[0],$inout0
+	movups	16*3($inp),$inout3
+	lea	16*4($inp),$inp
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+	xorps	@tweak[3],$inout3
+
+	call	_aesni_encrypt4
+
+	pxor	@tweak[0],$inout0
+	movdqa	@tweak[4],@tweak[0]
+	pxor	@tweak[1],$inout1
+	pxor	@tweak[2],$inout2
+	movdqu	$inout0,($out)
+	pxor	@tweak[3],$inout3
+	movdqu	$inout1,16*1($out)
+	movdqu	$inout2,16*2($out)
+	movdqu	$inout3,16*3($out)
+	lea	16*4($out),$out
+	jmp	.Lxts_enc_done
+
+.align	16
+.Lxts_enc_done:
+	and	\$15,$len_
+	jz	.Lxts_enc_ret
+	mov	$len_,$len
+
+.Lxts_enc_steal:
+	movzb	($inp),%eax			# borrow $rounds ...
+	movzb	-16($out),%ecx			# ... and $key
+	lea	1($inp),$inp
+	mov	%al,-16($out)
+	mov	%cl,0($out)
+	lea	1($out),$out
+	sub	\$1,$len
+	jnz	.Lxts_enc_steal
+
+	sub	$len_,$out			# rewind $out
+	mov	$key_,$key			# restore $key
+	mov	$rnds_,$rounds			# restore $rounds
+
+	movups	-16($out),$inout0
+	xorps	@tweak[0],$inout0
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	xorps	@tweak[0],$inout0
+	movups	$inout0,-16($out)
+
+.Lxts_enc_ret:
+___
+$code.=<<___ if ($win64);
+	movaps	-0xa0(%rbp),%xmm6
+	movaps	-0x90(%rbp),%xmm7
+	movaps	-0x80(%rbp),%xmm8
+	movaps	-0x70(%rbp),%xmm9
+	movaps	-0x60(%rbp),%xmm10
+	movaps	-0x50(%rbp),%xmm11
+	movaps	-0x40(%rbp),%xmm12
+	movaps	-0x30(%rbp),%xmm13
+	movaps	-0x20(%rbp),%xmm14
+	movaps	-0x10(%rbp),%xmm15
+___
+$code.=<<___;
+	lea	(%rbp),%rsp
+	pop	%rbp
+.Lxts_enc_epilogue:
+	ret
+.size	aesni_xts_encrypt,.-aesni_xts_encrypt
+___
+
+$code.=<<___;
+.globl	aesni_xts_decrypt
+.type	aesni_xts_decrypt,\@function,6
+.align	16
+aesni_xts_decrypt:
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,-0xa8(%rax)
+	movaps	%xmm7,-0x98(%rax)
+	movaps	%xmm8,-0x88(%rax)
+	movaps	%xmm9,-0x78(%rax)
+	movaps	%xmm10,-0x68(%rax)
+	movaps	%xmm11,-0x58(%rax)
+	movaps	%xmm12,-0x48(%rax)
+	movaps	%xmm13,-0x38(%rax)
+	movaps	%xmm14,-0x28(%rax)
+	movaps	%xmm15,-0x18(%rax)
+.Lxts_dec_body:
+___
+$code.=<<___;
+	lea	-8(%rax),%rbp
+	movups	($ivp),$inout0			# load clear-text tweak
+	mov	240($key2),$rounds		# key2->rounds
+	mov	240($key),$rnds_		# key1->rounds
+___
+	# generate the tweak
+	&aesni_generate1("enc",$key2,$rounds,$inout0);
+$code.=<<___;
+	xor	%eax,%eax			# if ($len%16) len-=16;
+	test	\$15,$len
+	setnz	%al
+	shl	\$4,%rax
+	sub	%rax,$len
+
+	$movkey	($key),$rndkey0			# zero round key
+	mov	$key,$key_			# backup $key
+	mov	$rnds_,$rounds			# backup $rounds
+	shl	\$4,$rnds_
+	mov	$len,$len_			# backup $len
+	and	\$-16,$len
+
+	$movkey	16($key,$rnds_),$rndkey1	# last round key
+
+	movdqa	.Lxts_magic(%rip),$twmask
+	movdqa	$inout0,@tweak[5]
+	pshufd	\$0x5f,$inout0,$twres
+	pxor	$rndkey0,$rndkey1
+___
+    for ($i=0;$i<4;$i++) {
+    $code.=<<___;
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
+	movdqa	@tweak[5],@tweak[$i]
+	psrad	\$31,$twtmp			# broadcast upper bits
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	pxor	$rndkey0,@tweak[$i]
+	pxor	$twtmp,@tweak[5]
+___
+    }
+$code.=<<___;
+	movdqa	@tweak[5],@tweak[4]
+	psrad	\$31,$twres
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twres
+	pxor	$rndkey0,@tweak[4]
+	pxor	$twres,@tweak[5]
+	movaps	$rndkey1,0x60(%rsp)		# save round[0]^round[last]
+
+	sub	\$16*6,$len
+	jc	.Lxts_dec_short
+
+	mov	\$16+96,$rounds
+	lea	32($key_,$rnds_),$key		# end of key schedule
+	sub	%r10,%rax			# twisted $rounds
+	$movkey	16($key_),$rndkey1
+	mov	%rax,%r10			# backup twisted $rounds
+	lea	.Lxts_magic(%rip),%r8
+	jmp	.Lxts_dec_grandloop
+
+.align	32
+.Lxts_dec_grandloop:
+	movdqu	`16*0`($inp),$inout0		# load input
+	movdqa	$rndkey0,$twmask
+	movdqu	`16*1`($inp),$inout1
+	pxor	@tweak[0],$inout0
+	movdqu	`16*2`($inp),$inout2
+	pxor	@tweak[1],$inout1
+	 aesdec		$rndkey1,$inout0
+	movdqu	`16*3`($inp),$inout3
+	pxor	@tweak[2],$inout2
+	 aesdec		$rndkey1,$inout1
+	movdqu	`16*4`($inp),$inout4
+	pxor	@tweak[3],$inout3
+	 aesdec		$rndkey1,$inout2
+	movdqu	`16*5`($inp),$inout5
+	pxor	@tweak[5],$twmask		# round[0]^=tweak[5]
+	 movdqa	0x60(%rsp),$twres		# load round[0]^round[last]
+	pxor	@tweak[4],$inout4
+	 aesdec		$rndkey1,$inout3
+	$movkey	32($key_),$rndkey0
+	lea	`16*6`($inp),$inp
+	pxor	$twmask,$inout5
+
+	 pxor	$twres,@tweak[0]
+	aesdec		$rndkey1,$inout4
+	 pxor	$twres,@tweak[1]
+	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks^last round key
+	aesdec		$rndkey1,$inout5
+	$movkey		48($key_),$rndkey1
+	 pxor	$twres,@tweak[2]
+
+	aesdec		$rndkey0,$inout0
+	 pxor	$twres,@tweak[3]
+	 movdqa	@tweak[1],`16*1`(%rsp)
+	aesdec		$rndkey0,$inout1
+	 pxor	$twres,@tweak[4]
+	 movdqa	@tweak[2],`16*2`(%rsp)
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	 pxor	$twres,$twmask
+	 movdqa	@tweak[4],`16*4`(%rsp)
+	aesdec		$rndkey0,$inout4
+	aesdec		$rndkey0,$inout5
+	$movkey		64($key_),$rndkey0
+	 movdqa	$twmask,`16*5`(%rsp)
+	pshufd	\$0x5f,@tweak[5],$twres
+	jmp	.Lxts_dec_loop6
+.align	32
+.Lxts_dec_loop6:
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	aesdec		$rndkey1,$inout4
+	aesdec		$rndkey1,$inout5
+	$movkey		-64($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesdec		$rndkey0,$inout0
+	aesdec		$rndkey0,$inout1
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	aesdec		$rndkey0,$inout4
+	aesdec		$rndkey0,$inout5
+	$movkey		-80($key,%rax),$rndkey0
+	jnz		.Lxts_dec_loop6
+
+	movdqa	(%r8),$twmask
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
+	 aesdec		$rndkey1,$inout0
+	paddq	@tweak[5],@tweak[5]
+	psrad	\$31,$twtmp
+	 aesdec		$rndkey1,$inout1
+	pand	$twmask,$twtmp
+	$movkey	($key_),@tweak[0]		# load round[0]
+	 aesdec		$rndkey1,$inout2
+	 aesdec		$rndkey1,$inout3
+	 aesdec		$rndkey1,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[0],@tweak[1]		# copy round[0]
+	 aesdec		$rndkey1,$inout5
+	 $movkey	-64($key),$rndkey1
+
+	movdqa	$twres,$twtmp
+	 aesdec		$rndkey0,$inout0
+	paddd	$twres,$twres
+	pxor	@tweak[5],@tweak[0]
+	 aesdec		$rndkey0,$inout1
+	psrad	\$31,$twtmp
+	paddq	@tweak[5],@tweak[5]
+	 aesdec		$rndkey0,$inout2
+	 aesdec		$rndkey0,$inout3
+	pand	$twmask,$twtmp
+	movaps	@tweak[1],@tweak[2]
+	 aesdec		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movdqa	$twres,$twtmp
+	 aesdec		$rndkey0,$inout5
+	 $movkey	-48($key),$rndkey0
+
+	paddd	$twres,$twres
+	 aesdec		$rndkey1,$inout0
+	pxor	@tweak[5],@tweak[1]
+	psrad	\$31,$twtmp
+	 aesdec		$rndkey1,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	 aesdec		$rndkey1,$inout2
+	 aesdec		$rndkey1,$inout3
+	 movdqa	@tweak[3],`16*3`(%rsp)
+	pxor	$twtmp,@tweak[5]
+	 aesdec		$rndkey1,$inout4
+	movaps	@tweak[2],@tweak[3]
+	movdqa	$twres,$twtmp
+	 aesdec		$rndkey1,$inout5
+	 $movkey	-32($key),$rndkey1
+
+	paddd	$twres,$twres
+	 aesdec		$rndkey0,$inout0
+	pxor	@tweak[5],@tweak[2]
+	psrad	\$31,$twtmp
+	 aesdec		$rndkey0,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
+	 aesdec		$rndkey0,$inout2
+	 aesdec		$rndkey0,$inout3
+	 aesdec		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[3],@tweak[4]
+	 aesdec		$rndkey0,$inout5
+
+	movdqa	$twres,$rndkey0
+	paddd	$twres,$twres
+	 aesdec		$rndkey1,$inout0
+	pxor	@tweak[5],@tweak[3]
+	psrad	\$31,$rndkey0
+	 aesdec		$rndkey1,$inout1
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$rndkey0
+	 aesdec		$rndkey1,$inout2
+	 aesdec		$rndkey1,$inout3
+	pxor	$rndkey0,@tweak[5]
+	$movkey		($key_),$rndkey0
+	 aesdec		$rndkey1,$inout4
+	 aesdec		$rndkey1,$inout5
+	$movkey		16($key_),$rndkey1
+
+	pxor	@tweak[5],@tweak[4]
+	 aesdeclast	`16*0`(%rsp),$inout0
+	psrad	\$31,$twres
+	paddq	@tweak[5],@tweak[5]
+	 aesdeclast	`16*1`(%rsp),$inout1
+	 aesdeclast	`16*2`(%rsp),$inout2
+	pand	$twmask,$twres
+	mov	%r10,%rax			# restore $rounds
+	 aesdeclast	`16*3`(%rsp),$inout3
+	 aesdeclast	`16*4`(%rsp),$inout4
+	 aesdeclast	`16*5`(%rsp),$inout5
+	pxor	$twres,@tweak[5]
+
+	lea	`16*6`($out),$out
+	movups	$inout0,`-16*6`($out)		# write output
+	movups	$inout1,`-16*5`($out)
+	movups	$inout2,`-16*4`($out)
+	movups	$inout3,`-16*3`($out)
+	movups	$inout4,`-16*2`($out)
+	movups	$inout5,`-16*1`($out)
+	sub	\$16*6,$len
+	jnc	.Lxts_dec_grandloop
+
+	mov	\$16+96,$rounds
+	sub	$rnds_,$rounds
+	mov	$key_,$key			# restore $key
+	shr	\$4,$rounds			# restore original value
+
+.Lxts_dec_short:
+	mov	$rounds,$rnds_			# backup $rounds
+	pxor	$rndkey0,@tweak[0]
+	pxor	$rndkey0,@tweak[1]
+	add	\$16*6,$len
+	jz	.Lxts_dec_done
+
+	pxor	$rndkey0,@tweak[2]
+	cmp	\$0x20,$len
+	jb	.Lxts_dec_one
+	pxor	$rndkey0,@tweak[3]
+	je	.Lxts_dec_two
+
+	pxor	$rndkey0,@tweak[4]
+	cmp	\$0x40,$len
+	jb	.Lxts_dec_three
+	je	.Lxts_dec_four
+
+	movdqu	($inp),$inout0
+	movdqu	16*1($inp),$inout1
+	movdqu	16*2($inp),$inout2
+	pxor	@tweak[0],$inout0
+	movdqu	16*3($inp),$inout3
+	pxor	@tweak[1],$inout1
+	movdqu	16*4($inp),$inout4
+	lea	16*5($inp),$inp
+	pxor	@tweak[2],$inout2
+	pxor	@tweak[3],$inout3
+	pxor	@tweak[4],$inout4
+
+	call	_aesni_decrypt6
+
+	xorps	@tweak[0],$inout0
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+	movdqu	$inout0,($out)
+	xorps	@tweak[3],$inout3
+	movdqu	$inout1,16*1($out)
+	xorps	@tweak[4],$inout4
+	movdqu	$inout2,16*2($out)
+	 pxor		$twtmp,$twtmp
+	movdqu	$inout3,16*3($out)
+	 pcmpgtd	@tweak[5],$twtmp
+	movdqu	$inout4,16*4($out)
+	lea	16*5($out),$out
+	 pshufd		\$0x13,$twtmp,@tweak[1]	# $twres
+	and	\$15,$len_
+	jz	.Lxts_dec_ret
+
+	movdqa	@tweak[5],@tweak[0]
+	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
+	pand	$twmask,@tweak[1]		# isolate carry and residue
+	pxor	@tweak[5],@tweak[1]
+	jmp	.Lxts_dec_done2
+
+.align	16
+.Lxts_dec_one:
+	movups	($inp),$inout0
+	lea	16*1($inp),$inp
+	xorps	@tweak[0],$inout0
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[1],@tweak[0]
+	movups	$inout0,($out)
+	movdqa	@tweak[2],@tweak[1]
+	lea	16*1($out),$out
+	jmp	.Lxts_dec_done
+
+.align	16
+.Lxts_dec_two:
+	movups	($inp),$inout0
+	movups	16($inp),$inout1
+	lea	32($inp),$inp
+	xorps	@tweak[0],$inout0
+	xorps	@tweak[1],$inout1
+
+	call	_aesni_decrypt2
+
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[2],@tweak[0]
+	xorps	@tweak[1],$inout1
+	movdqa	@tweak[3],@tweak[1]
+	movups	$inout0,($out)
+	movups	$inout1,16*1($out)
+	lea	16*2($out),$out
+	jmp	.Lxts_dec_done
+
+.align	16
+.Lxts_dec_three:
+	movups	($inp),$inout0
+	movups	16*1($inp),$inout1
+	movups	16*2($inp),$inout2
+	lea	16*3($inp),$inp
+	xorps	@tweak[0],$inout0
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+
+	call	_aesni_decrypt3
+
+	xorps	@tweak[0],$inout0
+	movdqa	@tweak[3],@tweak[0]
+	xorps	@tweak[1],$inout1
+	movdqa	@tweak[4],@tweak[1]
+	xorps	@tweak[2],$inout2
+	movups	$inout0,($out)
+	movups	$inout1,16*1($out)
+	movups	$inout2,16*2($out)
+	lea	16*3($out),$out
+	jmp	.Lxts_dec_done
+
+.align	16
+.Lxts_dec_four:
+	movups	($inp),$inout0
+	movups	16*1($inp),$inout1
+	movups	16*2($inp),$inout2
+	xorps	@tweak[0],$inout0
+	movups	16*3($inp),$inout3
+	lea	16*4($inp),$inp
+	xorps	@tweak[1],$inout1
+	xorps	@tweak[2],$inout2
+	xorps	@tweak[3],$inout3
+
+	call	_aesni_decrypt4
+
+	pxor	@tweak[0],$inout0
+	movdqa	@tweak[4],@tweak[0]
+	pxor	@tweak[1],$inout1
+	movdqa	@tweak[5],@tweak[1]
+	pxor	@tweak[2],$inout2
+	movdqu	$inout0,($out)
+	pxor	@tweak[3],$inout3
+	movdqu	$inout1,16*1($out)
+	movdqu	$inout2,16*2($out)
+	movdqu	$inout3,16*3($out)
+	lea	16*4($out),$out
+	jmp	.Lxts_dec_done
+
+.align	16
+.Lxts_dec_done:
+	and	\$15,$len_
+	jz	.Lxts_dec_ret
+.Lxts_dec_done2:
+	mov	$len_,$len
+	mov	$key_,$key			# restore $key
+	mov	$rnds_,$rounds			# restore $rounds
+
+	movups	($inp),$inout0
+	xorps	@tweak[1],$inout0
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	xorps	@tweak[1],$inout0
+	movups	$inout0,($out)
+
+.Lxts_dec_steal:
+	movzb	16($inp),%eax			# borrow $rounds ...
+	movzb	($out),%ecx			# ... and $key
+	lea	1($inp),$inp
+	mov	%al,($out)
+	mov	%cl,16($out)
+	lea	1($out),$out
+	sub	\$1,$len
+	jnz	.Lxts_dec_steal
+
+	sub	$len_,$out			# rewind $out
+	mov	$key_,$key			# restore $key
+	mov	$rnds_,$rounds			# restore $rounds
+
+	movups	($out),$inout0
+	xorps	@tweak[0],$inout0
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	xorps	@tweak[0],$inout0
+	movups	$inout0,($out)
+
+.Lxts_dec_ret:
+___
+$code.=<<___ if ($win64);
+	movaps	-0xa0(%rbp),%xmm6
+	movaps	-0x90(%rbp),%xmm7
+	movaps	-0x80(%rbp),%xmm8
+	movaps	-0x70(%rbp),%xmm9
+	movaps	-0x60(%rbp),%xmm10
+	movaps	-0x50(%rbp),%xmm11
+	movaps	-0x40(%rbp),%xmm12
+	movaps	-0x30(%rbp),%xmm13
+	movaps	-0x20(%rbp),%xmm14
+	movaps	-0x10(%rbp),%xmm15
+___
+$code.=<<___;
+	lea	(%rbp),%rsp
+	pop	%rbp
+.Lxts_dec_epilogue:
+	ret
+.size	aesni_xts_decrypt,.-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);
+{
+my $frame_size = 0x10 + ($win64?0xa0:0);	# used in decrypt
+my ($iv,$in0,$in1,$in2,$in3,$in4)=map("%xmm$_",(10..15));
+my $inp_=$key_;
+
+$code.=<<___;
+.globl	${PREFIX}_cbc_encrypt
+.type	${PREFIX}_cbc_encrypt,\@function,6
+.align	16
+${PREFIX}_cbc_encrypt:
+	test	$len,$len		# check length
+	jz	.Lcbc_ret
+
+	mov	240($key),$rnds_	# key->rounds
+	mov	$key,$key_		# backup $key
+	test	%r9d,%r9d		# 6th argument
+	jz	.Lcbc_decrypt
+#--------------------------- CBC ENCRYPT ------------------------------#
+	movups	($ivp),$inout0		# load iv as initial state
+	mov	$rnds_,$rounds
+	cmp	\$16,$len
+	jb	.Lcbc_enc_tail
+	sub	\$16,$len
+	jmp	.Lcbc_enc_loop
+.align	16
+.Lcbc_enc_loop:
+	movups	($inp),$inout1		# load input
+	lea	16($inp),$inp
+	#xorps	$inout1,$inout0
+___
+	&aesni_generate1("enc",$key,$rounds,$inout0,$inout1);
+$code.=<<___;
+	mov	$rnds_,$rounds		# restore $rounds
+	mov	$key_,$key		# restore $key
+	movups	$inout0,0($out)		# store output
+	lea	16($out),$out
+	sub	\$16,$len
+	jnc	.Lcbc_enc_loop
+	add	\$16,$len
+	jnz	.Lcbc_enc_tail
+	movups	$inout0,($ivp)
+	jmp	.Lcbc_ret
+
+.Lcbc_enc_tail:
+	mov	$len,%rcx	# zaps $key
+	xchg	$inp,$out	# $inp is %rsi and $out is %rdi now
+	.long	0x9066A4F3	# rep movsb
+	mov	\$16,%ecx	# zero tail
+	sub	$len,%rcx
+	xor	%eax,%eax
+	.long	0x9066AAF3	# rep stosb
+	lea	-16(%rdi),%rdi	# rewind $out by 1 block
+	mov	$rnds_,$rounds	# restore $rounds
+	mov	%rdi,%rsi	# $inp and $out are the same
+	mov	$key_,$key	# restore $key
+	xor	$len,$len	# len=16
+	jmp	.Lcbc_enc_loop	# one more spin
+#--------------------------- CBC DECRYPT ------------------------------#
+.align	16
+.Lcbc_decrypt:
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Lcbc_decrypt_body:
+___
+$code.=<<___;
+	lea	-8(%rax),%rbp
+	movups	($ivp),$iv
+	mov	$rnds_,$rounds
+	cmp	\$0x50,$len
+	jbe	.Lcbc_dec_tail
+
+	$movkey	($key),$rndkey0
+	movdqu	0x00($inp),$inout0	# load input
+	movdqu	0x10($inp),$inout1
+	movdqa	$inout0,$in0
+	movdqu	0x20($inp),$inout2
+	movdqa	$inout1,$in1
+	movdqu	0x30($inp),$inout3
+	movdqa	$inout2,$in2
+	movdqu	0x40($inp),$inout4
+	movdqa	$inout3,$in3
+	movdqu	0x50($inp),$inout5
+	movdqa	$inout4,$in4
+	mov	OPENSSL_ia32cap_P+4(%rip),%r9d
+	cmp	\$0x70,$len
+	jbe	.Lcbc_dec_six_or_seven
+
+	and	\$`1<<26|1<<22`,%r9d	# isolate XSAVE+MOVBE	
+	sub	\$0x50,$len
+	cmp	\$`1<<22`,%r9d		# check for MOVBE without XSAVE
+	je	.Lcbc_dec_loop6_enter
+	sub	\$0x20,$len
+	lea	0x70($key),$key		# size optimization
+	jmp	.Lcbc_dec_loop8_enter
+.align	16
+.Lcbc_dec_loop8:
+	movups	$inout7,($out)
+	lea	0x10($out),$out
+.Lcbc_dec_loop8_enter:
+	movdqu		0x60($inp),$inout6
+	pxor		$rndkey0,$inout0
+	movdqu		0x70($inp),$inout7
+	pxor		$rndkey0,$inout1
+	$movkey		0x10-0x70($key),$rndkey1
+	pxor		$rndkey0,$inout2
+	xor		$inp_,$inp_
+	cmp		\$0x70,$len	# is there at least 0x60 bytes ahead?
+	pxor		$rndkey0,$inout3
+	pxor		$rndkey0,$inout4
+	pxor		$rndkey0,$inout5
+	pxor		$rndkey0,$inout6
+
+	aesdec		$rndkey1,$inout0
+	pxor		$rndkey0,$inout7
+	$movkey		0x20-0x70($key),$rndkey0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	aesdec		$rndkey1,$inout4
+	aesdec		$rndkey1,$inout5
+	aesdec		$rndkey1,$inout6
+	setnc		${inp_}b
+	shl		\$7,$inp_
+	aesdec		$rndkey1,$inout7
+	add		$inp,$inp_
+	$movkey		0x30-0x70($key),$rndkey1
+___
+for($i=1;$i<12;$i++) {
+my $rndkeyx = ($i&1)?$rndkey0:$rndkey1;
+$code.=<<___	if ($i==7);
+	cmp		\$11,$rounds
+___
+$code.=<<___;
+	aesdec		$rndkeyx,$inout0
+	aesdec		$rndkeyx,$inout1
+	aesdec		$rndkeyx,$inout2
+	aesdec		$rndkeyx,$inout3
+	aesdec		$rndkeyx,$inout4
+	aesdec		$rndkeyx,$inout5
+	aesdec		$rndkeyx,$inout6
+	aesdec		$rndkeyx,$inout7
+	$movkey		`0x30+0x10*$i`-0x70($key),$rndkeyx
+___
+$code.=<<___	if ($i<6 || (!($i&1) && $i>7));
+	nop
+___
+$code.=<<___	if ($i==7);
+	jb		.Lcbc_dec_done
+___
+$code.=<<___	if ($i==9);
+	je		.Lcbc_dec_done
+___
+$code.=<<___	if ($i==11);
+	jmp		.Lcbc_dec_done
+___
+}
+$code.=<<___;
+.align	16
+.Lcbc_dec_done:
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	pxor		$rndkey0,$iv
+	pxor		$rndkey0,$in0
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	pxor		$rndkey0,$in1
+	pxor		$rndkey0,$in2
+	aesdec		$rndkey1,$inout4
+	aesdec		$rndkey1,$inout5
+	pxor		$rndkey0,$in3
+	pxor		$rndkey0,$in4
+	aesdec		$rndkey1,$inout6
+	aesdec		$rndkey1,$inout7
+	movdqu		0x50($inp),$rndkey1
+
+	aesdeclast	$iv,$inout0
+	movdqu		0x60($inp),$iv		# borrow $iv
+	pxor		$rndkey0,$rndkey1
+	aesdeclast	$in0,$inout1
+	pxor		$rndkey0,$iv
+	movdqu		0x70($inp),$rndkey0	# next IV
+	aesdeclast	$in1,$inout2
+	lea		0x80($inp),$inp
+	movdqu		0x00($inp_),$in0
+	aesdeclast	$in2,$inout3
+	aesdeclast	$in3,$inout4
+	movdqu		0x10($inp_),$in1
+	movdqu		0x20($inp_),$in2
+	aesdeclast	$in4,$inout5
+	aesdeclast	$rndkey1,$inout6
+	movdqu		0x30($inp_),$in3
+	movdqu		0x40($inp_),$in4
+	aesdeclast	$iv,$inout7
+	movdqa		$rndkey0,$iv		# return $iv
+	movdqu		0x50($inp_),$rndkey1
+	$movkey		-0x70($key),$rndkey0
+
+	movups		$inout0,($out)		# store output
+	movdqa		$in0,$inout0
+	movups		$inout1,0x10($out)
+	movdqa		$in1,$inout1
+	movups		$inout2,0x20($out)
+	movdqa		$in2,$inout2
+	movups		$inout3,0x30($out)
+	movdqa		$in3,$inout3
+	movups		$inout4,0x40($out)
+	movdqa		$in4,$inout4
+	movups		$inout5,0x50($out)
+	movdqa		$rndkey1,$inout5
+	movups		$inout6,0x60($out)
+	lea		0x70($out),$out
+
+	sub	\$0x80,$len
+	ja	.Lcbc_dec_loop8
+
+	movaps	$inout7,$inout0
+	lea	-0x70($key),$key
+	add	\$0x70,$len
+	jle	.Lcbc_dec_tail_collected
+	movups	$inout7,($out)
+	lea	0x10($out),$out
+	cmp	\$0x50,$len
+	jbe	.Lcbc_dec_tail
+
+	movaps	$in0,$inout0
+.Lcbc_dec_six_or_seven:
+	cmp	\$0x60,$len
+	ja	.Lcbc_dec_seven
+
+	movaps	$inout5,$inout6
+	call	_aesni_decrypt6
+	pxor	$iv,$inout0		# ^= IV
+	movaps	$inout6,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movdqu	$inout2,0x20($out)
+	pxor	$in3,$inout4
+	movdqu	$inout3,0x30($out)
+	pxor	$in4,$inout5
+	movdqu	$inout4,0x40($out)
+	lea	0x50($out),$out
+	movdqa	$inout5,$inout0
+	jmp	.Lcbc_dec_tail_collected
+
+.align	16
+.Lcbc_dec_seven:
+	movups	0x60($inp),$inout6
+	xorps	$inout7,$inout7
+	call	_aesni_decrypt8
+	movups	0x50($inp),$inout7
+	pxor	$iv,$inout0		# ^= IV
+	movups	0x60($inp),$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movdqu	$inout2,0x20($out)
+	pxor	$in3,$inout4
+	movdqu	$inout3,0x30($out)
+	pxor	$in4,$inout5
+	movdqu	$inout4,0x40($out)
+	pxor	$inout7,$inout6
+	movdqu	$inout5,0x50($out)
+	lea	0x60($out),$out
+	movdqa	$inout6,$inout0
+	jmp	.Lcbc_dec_tail_collected
+
+.align	16
+.Lcbc_dec_loop6:
+	movups	$inout5,($out)
+	lea	0x10($out),$out
+	movdqu	0x00($inp),$inout0	# load input
+	movdqu	0x10($inp),$inout1
+	movdqa	$inout0,$in0
+	movdqu	0x20($inp),$inout2
+	movdqa	$inout1,$in1
+	movdqu	0x30($inp),$inout3
+	movdqa	$inout2,$in2
+	movdqu	0x40($inp),$inout4
+	movdqa	$inout3,$in3
+	movdqu	0x50($inp),$inout5
+	movdqa	$inout4,$in4
+.Lcbc_dec_loop6_enter:
+	lea	0x60($inp),$inp
+	movdqa	$inout5,$inout6
+
+	call	_aesni_decrypt6
+
+	pxor	$iv,$inout0		# ^= IV
+	movdqa	$inout6,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movdqu	$inout2,0x20($out)
+	pxor	$in3,$inout4
+	mov	$key_,$key
+	movdqu	$inout3,0x30($out)
+	pxor	$in4,$inout5
+	mov	$rnds_,$rounds
+	movdqu	$inout4,0x40($out)
+	lea	0x50($out),$out
+	sub	\$0x60,$len
+	ja	.Lcbc_dec_loop6
+
+	movdqa	$inout5,$inout0
+	add	\$0x50,$len
+	jle	.Lcbc_dec_tail_collected
+	movups	$inout5,($out)
+	lea	0x10($out),$out
+
+.Lcbc_dec_tail:
+	movups	($inp),$inout0
+	sub	\$0x10,$len
+	jbe	.Lcbc_dec_one
+
+	movups	0x10($inp),$inout1
+	movaps	$inout0,$in0
+	sub	\$0x10,$len
+	jbe	.Lcbc_dec_two
+
+	movups	0x20($inp),$inout2
+	movaps	$inout1,$in1
+	sub	\$0x10,$len
+	jbe	.Lcbc_dec_three
+
+	movups	0x30($inp),$inout3
+	movaps	$inout2,$in2
+	sub	\$0x10,$len
+	jbe	.Lcbc_dec_four
+
+	movups	0x40($inp),$inout4
+	movaps	$inout3,$in3
+	movaps	$inout4,$in4
+	xorps	$inout5,$inout5
+	call	_aesni_decrypt6
+	pxor	$iv,$inout0
+	movaps	$in4,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movdqu	$inout2,0x20($out)
+	pxor	$in3,$inout4
+	movdqu	$inout3,0x30($out)
+	lea	0x40($out),$out
+	movdqa	$inout4,$inout0
+	sub	\$0x10,$len
+	jmp	.Lcbc_dec_tail_collected
+
+.align	16
+.Lcbc_dec_one:
+	movaps	$inout0,$in0
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	xorps	$iv,$inout0
+	movaps	$in0,$iv
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_two:
+	movaps	$inout1,$in1
+	call	_aesni_decrypt2
+	pxor	$iv,$inout0
+	movaps	$in1,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	movdqa	$inout1,$inout0
+	lea	0x10($out),$out
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_three:
+	movaps	$inout2,$in2
+	call	_aesni_decrypt3
+	pxor	$iv,$inout0
+	movaps	$in2,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	movdqa	$inout2,$inout0
+	lea	0x20($out),$out
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_four:
+	movaps	$inout3,$in3
+	call	_aesni_decrypt4
+	pxor	$iv,$inout0
+	movaps	$in3,$iv
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movdqu	$inout2,0x20($out)
+	movdqa	$inout3,$inout0
+	lea	0x30($out),$out
+	jmp	.Lcbc_dec_tail_collected
+
+.align	16
+.Lcbc_dec_tail_collected:
+	movups	$iv,($ivp)
+	and	\$15,$len
+	jnz	.Lcbc_dec_tail_partial
+	movups	$inout0,($out)
+	jmp	.Lcbc_dec_ret
+.align	16
+.Lcbc_dec_tail_partial:
+	movaps	$inout0,(%rsp)
+	mov	\$16,%rcx
+	mov	$out,%rdi
+	sub	$len,%rcx
+	lea	(%rsp),%rsi
+	.long	0x9066A4F3	# rep movsb
+
+.Lcbc_dec_ret:
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+___
+$code.=<<___;
+	lea	(%rbp),%rsp
+	pop	%rbp
+.Lcbc_ret:
+	ret
+.size	${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
+___
+} 
+# int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
+#				int bits, AES_KEY *key)
+{ my ($inp,$bits,$key) = @_4args;
+  $bits =~ s/%r/%e/;
+
+$code.=<<___;
+.globl	${PREFIX}_set_decrypt_key
+.type	${PREFIX}_set_decrypt_key,\@abi-omnipotent
+.align	16
+${PREFIX}_set_decrypt_key:
+	.byte	0x48,0x83,0xEC,0x08	# sub rsp,8
+	call	__aesni_set_encrypt_key
+	shl	\$4,$bits		# rounds-1 after _aesni_set_encrypt_key
+	test	%eax,%eax
+	jnz	.Ldec_key_ret
+	lea	16($key,$bits),$inp	# points at the end of key schedule
+
+	$movkey	($key),%xmm0		# just swap
+	$movkey	($inp),%xmm1
+	$movkey	%xmm0,($inp)
+	$movkey	%xmm1,($key)
+	lea	16($key),$key
+	lea	-16($inp),$inp
+
+.Ldec_key_inverse:
+	$movkey	($key),%xmm0		# swap and inverse
+	$movkey	($inp),%xmm1
+	aesimc	%xmm0,%xmm0
+	aesimc	%xmm1,%xmm1
+	lea	16($key),$key
+	lea	-16($inp),$inp
+	$movkey	%xmm0,16($inp)
+	$movkey	%xmm1,-16($key)
+	cmp	$key,$inp
+	ja	.Ldec_key_inverse
+
+	$movkey	($key),%xmm0		# inverse middle
+	aesimc	%xmm0,%xmm0
+	$movkey	%xmm0,($inp)
+.Ldec_key_ret:
+	add	\$8,%rsp
+	ret
+.LSEH_end_set_decrypt_key:
+.size	${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
+___
+
+# This is based on submission by
+#
+#	Huang Ying <ying.huang@intel.com>
+#	Vinodh Gopal <vinodh.gopal@intel.com>
+#	Kahraman Akdemir
+#
+# Agressively optimized in respect to aeskeygenassist's critical path
+# and is contained in %xmm0-5 to meet Win64 ABI requirement.
+#
+$code.=<<___;
+.globl	${PREFIX}_set_encrypt_key
+.type	${PREFIX}_set_encrypt_key,\@abi-omnipotent
+.align	16
+${PREFIX}_set_encrypt_key:
+__aesni_set_encrypt_key:
+	.byte	0x48,0x83,0xEC,0x08	# sub rsp,8
+	mov	\$-1,%rax
+	test	$inp,$inp
+	jz	.Lenc_key_ret
+	test	$key,$key
+	jz	.Lenc_key_ret
+
+	movups	($inp),%xmm0		# pull first 128 bits of *userKey
+	xorps	%xmm4,%xmm4		# low dword of xmm4 is assumed 0
+	lea	16($key),%rax
+	cmp	\$256,$bits
+	je	.L14rounds
+	cmp	\$192,$bits
+	je	.L12rounds
+	cmp	\$128,$bits
+	jne	.Lbad_keybits
+
+.L10rounds:
+	mov	\$9,$bits			# 10 rounds for 128-bit key
+	$movkey	%xmm0,($key)			# round 0
+	aeskeygenassist	\$0x1,%xmm0,%xmm1	# round 1
+	call		.Lkey_expansion_128_cold
+	aeskeygenassist	\$0x2,%xmm0,%xmm1	# round 2
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x4,%xmm0,%xmm1	# round 3
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x8,%xmm0,%xmm1	# round 4
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x10,%xmm0,%xmm1	# round 5
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x20,%xmm0,%xmm1	# round 6
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x40,%xmm0,%xmm1	# round 7
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x80,%xmm0,%xmm1	# round 8
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x1b,%xmm0,%xmm1	# round 9
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x36,%xmm0,%xmm1	# round 10
+	call		.Lkey_expansion_128
+	$movkey	%xmm0,(%rax)
+	mov	$bits,80(%rax)	# 240(%rdx)
+	xor	%eax,%eax
+	jmp	.Lenc_key_ret
+
+.align	16
+.L12rounds:
+	movq	16($inp),%xmm2			# remaining 1/3 of *userKey
+	mov	\$11,$bits			# 12 rounds for 192
+	$movkey	%xmm0,($key)			# round 0
+	aeskeygenassist	\$0x1,%xmm2,%xmm1	# round 1,2
+	call		.Lkey_expansion_192a_cold
+	aeskeygenassist	\$0x2,%xmm2,%xmm1	# round 2,3
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x4,%xmm2,%xmm1	# round 4,5
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x8,%xmm2,%xmm1	# round 5,6
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x10,%xmm2,%xmm1	# round 7,8
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x20,%xmm2,%xmm1	# round 8,9
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x40,%xmm2,%xmm1	# round 10,11
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x80,%xmm2,%xmm1	# round 11,12
+	call		.Lkey_expansion_192b
+	$movkey	%xmm0,(%rax)
+	mov	$bits,48(%rax)	# 240(%rdx)
+	xor	%rax, %rax
+	jmp	.Lenc_key_ret
+
+.align	16
+.L14rounds:
+	movups	16($inp),%xmm2			# remaning half of *userKey
+	mov	\$13,$bits			# 14 rounds for 256
+	lea	16(%rax),%rax
+	$movkey	%xmm0,($key)			# round 0
+	$movkey	%xmm2,16($key)			# round 1
+	aeskeygenassist	\$0x1,%xmm2,%xmm1	# round 2
+	call		.Lkey_expansion_256a_cold
+	aeskeygenassist	\$0x1,%xmm0,%xmm1	# round 3
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x2,%xmm2,%xmm1	# round 4
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x2,%xmm0,%xmm1	# round 5
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x4,%xmm2,%xmm1	# round 6
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x4,%xmm0,%xmm1	# round 7
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x8,%xmm2,%xmm1	# round 8
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x8,%xmm0,%xmm1	# round 9
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x10,%xmm2,%xmm1	# round 10
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x10,%xmm0,%xmm1	# round 11
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x20,%xmm2,%xmm1	# round 12
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x20,%xmm0,%xmm1	# round 13
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x40,%xmm2,%xmm1	# round 14
+	call		.Lkey_expansion_256a
+	$movkey	%xmm0,(%rax)
+	mov	$bits,16(%rax)	# 240(%rdx)
+	xor	%rax,%rax
+	jmp	.Lenc_key_ret
+
+.align	16
+.Lbad_keybits:
+	mov	\$-2,%rax
+.Lenc_key_ret:
+	add	\$8,%rsp
+	ret
+.LSEH_end_set_encrypt_key:
+
+.align	16
+.Lkey_expansion_128:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_128_cold:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	xorps	%xmm4, %xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	xorps	%xmm4, %xmm0
+	shufps	\$0b11111111,%xmm1,%xmm1	# critical path
+	xorps	%xmm1,%xmm0
+	ret
+
+.align 16
+.Lkey_expansion_192a:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_192a_cold:
+	movaps	%xmm2, %xmm5
+.Lkey_expansion_192b_warm:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	movdqa	%xmm2,%xmm3
+	xorps	%xmm4,%xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	pslldq	\$4,%xmm3
+	xorps	%xmm4,%xmm0
+	pshufd	\$0b01010101,%xmm1,%xmm1	# critical path
+	pxor	%xmm3,%xmm2
+	pxor	%xmm1,%xmm0
+	pshufd	\$0b11111111,%xmm0,%xmm3
+	pxor	%xmm3,%xmm2
+	ret
+
+.align 16
+.Lkey_expansion_192b:
+	movaps	%xmm0,%xmm3
+	shufps	\$0b01000100,%xmm0,%xmm5
+	$movkey	%xmm5,(%rax)
+	shufps	\$0b01001110,%xmm2,%xmm3
+	$movkey	%xmm3,16(%rax)
+	lea	32(%rax),%rax
+	jmp	.Lkey_expansion_192b_warm
+
+.align	16
+.Lkey_expansion_256a:
+	$movkey	%xmm2,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_256a_cold:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	xorps	%xmm4,%xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	xorps	%xmm4,%xmm0
+	shufps	\$0b11111111,%xmm1,%xmm1	# critical path
+	xorps	%xmm1,%xmm0
+	ret
+
+.align 16
+.Lkey_expansion_256b:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+
+	shufps	\$0b00010000,%xmm2,%xmm4
+	xorps	%xmm4,%xmm2
+	shufps	\$0b10001100,%xmm2,%xmm4
+	xorps	%xmm4,%xmm2
+	shufps	\$0b10101010,%xmm1,%xmm1	# critical path
+	xorps	%xmm1,%xmm2
+	ret
+.size	${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
+.size	__aesni_set_encrypt_key,.-__aesni_set_encrypt_key
+___
+}
+
+$code.=<<___;
+.align	64
+.Lbswap_mask:
+	.byte	15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
+.Lincrement32:
+	.long	6,6,6,0
+.Lincrement64:
+	.long	1,0,0,0
+.Lxts_magic:
+	.long	0x87,0,1,0
+.Lincrement1:
+	.byte	0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
+
+.asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
+.align	64
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+___
+$code.=<<___ if ($PREFIX eq "aesni");
+.type	ecb_ccm64_se_handler,\@abi-omnipotent
+.align	16
+ecb_ccm64_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lcommon_seh_tail
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lcommon_seh_tail
+
+	lea	0(%rax),%rsi		# %xmm save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$8,%ecx		# 4*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	0x58(%rax),%rax		# adjust stack pointer
+
+	jmp	.Lcommon_seh_tail
+.size	ecb_ccm64_se_handler,.-ecb_ccm64_se_handler
+
+.type	ctr_xts_se_handler,\@abi-omnipotent
+.align	16
+ctr_xts_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue lable
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lcommon_seh_tail
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lcommon_seh_tail
+
+	mov	160($context),%rax	# pull context->Rbp
+	lea	-0xa0(%rax),%rsi	# %xmm save area
+	lea	512($context),%rdi	# & context.Xmm6
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	jmp	.Lcommon_rbp_tail
+.size	ctr_xts_se_handler,.-ctr_xts_se_handler
+___
+$code.=<<___;
+.type	cbc_se_handler,\@abi-omnipotent
+.align	16
+cbc_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	152($context),%rax	# pull context->Rsp
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lcbc_decrypt(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<"prologue" label
+	jb	.Lcommon_seh_tail
+
+	lea	.Lcbc_decrypt_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<cbc_decrypt_body
+	jb	.Lrestore_cbc_rax
+
+	lea	.Lcbc_ret(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>="epilogue" label
+	jae	.Lcommon_seh_tail
+
+	lea	16(%rax),%rsi		# %xmm save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+
+.Lcommon_rbp_tail:
+	mov	160($context),%rax	# pull context->Rbp
+	mov	(%rax),%rbp		# restore saved %rbp
+	lea	8(%rax),%rax		# adjust stack pointer
+	mov	%rbp,160($context)	# restore context->Rbp
+	jmp	.Lcommon_seh_tail
+
+.Lrestore_cbc_rax:
+	mov	120($context),%rax
+
+.Lcommon_seh_tail:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	cbc_se_handler,.-cbc_se_handler
+
+.section	.pdata
+.align	4
+___
+$code.=<<___ if ($PREFIX eq "aesni");
+	.rva	.LSEH_begin_aesni_ecb_encrypt
+	.rva	.LSEH_end_aesni_ecb_encrypt
+	.rva	.LSEH_info_ecb
+
+	.rva	.LSEH_begin_aesni_ccm64_encrypt_blocks
+	.rva	.LSEH_end_aesni_ccm64_encrypt_blocks
+	.rva	.LSEH_info_ccm64_enc
+
+	.rva	.LSEH_begin_aesni_ccm64_decrypt_blocks
+	.rva	.LSEH_end_aesni_ccm64_decrypt_blocks
+	.rva	.LSEH_info_ccm64_dec
+
+	.rva	.LSEH_begin_aesni_ctr32_encrypt_blocks
+	.rva	.LSEH_end_aesni_ctr32_encrypt_blocks
+	.rva	.LSEH_info_ctr32
+
+	.rva	.LSEH_begin_aesni_xts_encrypt
+	.rva	.LSEH_end_aesni_xts_encrypt
+	.rva	.LSEH_info_xts_enc
+
+	.rva	.LSEH_begin_aesni_xts_decrypt
+	.rva	.LSEH_end_aesni_xts_decrypt
+	.rva	.LSEH_info_xts_dec
+___
+$code.=<<___;
+	.rva	.LSEH_begin_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_end_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_info_cbc
+
+	.rva	${PREFIX}_set_decrypt_key
+	.rva	.LSEH_end_set_decrypt_key
+	.rva	.LSEH_info_key
+
+	.rva	${PREFIX}_set_encrypt_key
+	.rva	.LSEH_end_set_encrypt_key
+	.rva	.LSEH_info_key
+.section	.xdata
+.align	8
+___
+$code.=<<___ if ($PREFIX eq "aesni");
+.LSEH_info_ecb:
+	.byte	9,0,0,0
+	.rva	ecb_ccm64_se_handler
+	.rva	.Lecb_enc_body,.Lecb_enc_ret		# HandlerData[]
+.LSEH_info_ccm64_enc:
+	.byte	9,0,0,0
+	.rva	ecb_ccm64_se_handler
+	.rva	.Lccm64_enc_body,.Lccm64_enc_ret	# HandlerData[]
+.LSEH_info_ccm64_dec:
+	.byte	9,0,0,0
+	.rva	ecb_ccm64_se_handler
+	.rva	.Lccm64_dec_body,.Lccm64_dec_ret	# HandlerData[]
+.LSEH_info_ctr32:
+	.byte	9,0,0,0
+	.rva	ctr_xts_se_handler
+	.rva	.Lctr32_body,.Lctr32_epilogue		# HandlerData[]
+.LSEH_info_xts_enc:
+	.byte	9,0,0,0
+	.rva	ctr_xts_se_handler
+	.rva	.Lxts_enc_body,.Lxts_enc_epilogue	# HandlerData[]
+.LSEH_info_xts_dec:
+	.byte	9,0,0,0
+	.rva	ctr_xts_se_handler
+	.rva	.Lxts_dec_body,.Lxts_dec_epilogue	# HandlerData[]
+___
+$code.=<<___;
+.LSEH_info_cbc:
+	.byte	9,0,0,0
+	.rva	cbc_se_handler
+.LSEH_info_key:
+	.byte	0x01,0x04,0x01,0x00
+	.byte	0x04,0x02,0x00,0x00	# sub rsp,8
+___
+}
+
+sub rex {
+  local *opcode=shift;
+  my ($dst,$src)=@_;
+  my $rex=0;
+
+    $rex|=0x04			if($dst>=8);
+    $rex|=0x01			if($src>=8);
+    push @opcode,$rex|0x40	if($rex);
+}
+
+sub aesni {
+  my $line=shift;
+  my @opcode=(0x66);
+
+    if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+	rex(\@opcode,$4,$3);
+	push @opcode,0x0f,0x3a,0xdf;
+	push @opcode,0xc0|($3&7)|(($4&7)<<3);	# ModR/M
+	my $c=$2;
+	push @opcode,$c=~/^0/?oct($c):$c;
+	return ".byte\t".join(',',@opcode);
+    }
+    elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+	my %opcodelet = (
+		"aesimc" => 0xdb,
+		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
+		"aesdec" => 0xde,	"aesdeclast" => 0xdf
+	);
+	return undef if (!defined($opcodelet{$1}));
+	rex(\@opcode,$3,$2);
+	push @opcode,0x0f,0x38,$opcodelet{$1};
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);	# ModR/M
+	return ".byte\t".join(',',@opcode);
+    }
+    elsif ($line=~/(aes[a-z]+)\s+([0x1-9a-fA-F]*)\(%rsp\),\s*%xmm([0-9]+)/) {
+	my %opcodelet = (
+		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
+		"aesdec" => 0xde,	"aesdeclast" => 0xdf
+	);
+	return undef if (!defined($opcodelet{$1}));
+	my $off = $2;
+	push @opcode,0x44 if ($3>=8);
+	push @opcode,0x0f,0x38,$opcodelet{$1};
+	push @opcode,0x44|(($3&7)<<3),0x24;	# ModR/M
+	push @opcode,($off=~/^0/?oct($off):$off)&0xff;
+	return ".byte\t".join(',',@opcode);
+    }
+    return $line;
+}
+
+sub movbe {
+	".byte	0x0f,0x38,0xf1,0x44,0x24,".shift;
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
+#$code =~ s/\bmovbe\s+%eax/bswap %eax; mov %eax/gm;	# debugging artefact
+$code =~ s/\bmovbe\s+%eax,\s*([0-9]+)\(%rsp\)/movbe($1)/gem;
+
+print $code;
+
+close STDOUT;
diff --git a/src/crypto/aes/asm/aesv8-armx.pl b/src/crypto/aes/asm/aesv8-armx.pl
new file mode 100644
index 0000000..1e93f86
--- /dev/null
+++ b/src/crypto/aes/asm/aesv8-armx.pl
@@ -0,0 +1,962 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements support for ARMv8 AES instructions. The
+# module is endian-agnostic in sense that it supports both big- and
+# little-endian cases. As does it support both 32- and 64-bit modes
+# of operation. Latter is achieved by limiting amount of utilized
+# registers to 16, which implies additional NEON load and integer
+# instructions. This has no effect on mighty Apple A7, where results
+# are literally equal to the theoretical estimates based on AES
+# instruction latencies and issue rates. On Cortex-A53, an in-order
+# execution core, this costs up to 10-15%, which is partially
+# compensated by implementing dedicated code path for 128-bit
+# CBC encrypt case. On Cortex-A57 parallelizable mode performance
+# seems to be limited by sheer amount of NEON instructions...
+#
+# Performance in cycles per byte processed with 128-bit key:
+#
+#		CBC enc		CBC dec		CTR
+# Apple A7	2.39		1.20		1.20
+# Cortex-A53	2.45		1.87		1.94
+# Cortex-A57	3.64		1.34		1.32
+
+$flavour = shift;
+open STDOUT,">".shift;
+
+$prefix="aes_v8";
+
+$code=<<___;
+#include "arm_arch.h"
+
+#if __ARM_MAX_ARCH__>=7
+.text
+___
+$code.=".arch	armv8-a+crypto\n"			if ($flavour =~ /64/);
+$code.=".arch	armv7-a\n.fpu	neon\n.code	32\n"	if ($flavour !~ /64/);
+		#^^^^^^ this is done to simplify adoption by not depending
+		#	on latest binutils.
+
+# Assembler mnemonics are an eclectic mix of 32- and 64-bit syntax,
+# NEON is mostly 32-bit mnemonics, integer - mostly 64. Goal is to
+# maintain both 32- and 64-bit codes within single module and
+# transliterate common code to either flavour with regex vodoo.
+#
+{{{
+my ($inp,$bits,$out,$ptr,$rounds)=("x0","w1","x2","x3","w12");
+my ($zero,$rcon,$mask,$in0,$in1,$tmp,$key)=
+	$flavour=~/64/? map("q$_",(0..6)) : map("q$_",(0..3,8..10));
+
+
+$code.=<<___;
+.align	5
+rcon:
+.long	0x01,0x01,0x01,0x01
+.long	0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d	// rotate-n-splat
+.long	0x1b,0x1b,0x1b,0x1b
+
+.globl	${prefix}_set_encrypt_key
+.type	${prefix}_set_encrypt_key,%function
+.align	5
+${prefix}_set_encrypt_key:
+.Lenc_key:
+___
+$code.=<<___	if ($flavour =~ /64/);
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+___
+$code.=<<___;
+	mov	$ptr,#-1
+	cmp	$inp,#0
+	b.eq	.Lenc_key_abort
+	cmp	$out,#0
+	b.eq	.Lenc_key_abort
+	mov	$ptr,#-2
+	cmp	$bits,#128
+	b.lt	.Lenc_key_abort
+	cmp	$bits,#256
+	b.gt	.Lenc_key_abort
+	tst	$bits,#0x3f
+	b.ne	.Lenc_key_abort
+
+	adr	$ptr,rcon
+	cmp	$bits,#192
+
+	veor	$zero,$zero,$zero
+	vld1.8	{$in0},[$inp],#16
+	mov	$bits,#8		// reuse $bits
+	vld1.32	{$rcon,$mask},[$ptr],#32
+
+	b.lt	.Loop128
+	b.eq	.L192
+	b	.L256
+
+.align	4
+.Loop128:
+	vtbl.8	$key,{$in0},$mask
+	vext.8	$tmp,$zero,$in0,#12
+	vst1.32	{$in0},[$out],#16
+	aese	$key,$zero
+	subs	$bits,$bits,#1
+
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	 veor	$key,$key,$rcon
+	veor	$in0,$in0,$tmp
+	vshl.u8	$rcon,$rcon,#1
+	veor	$in0,$in0,$key
+	b.ne	.Loop128
+
+	vld1.32	{$rcon},[$ptr]
+
+	vtbl.8	$key,{$in0},$mask
+	vext.8	$tmp,$zero,$in0,#12
+	vst1.32	{$in0},[$out],#16
+	aese	$key,$zero
+
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	 veor	$key,$key,$rcon
+	veor	$in0,$in0,$tmp
+	vshl.u8	$rcon,$rcon,#1
+	veor	$in0,$in0,$key
+
+	vtbl.8	$key,{$in0},$mask
+	vext.8	$tmp,$zero,$in0,#12
+	vst1.32	{$in0},[$out],#16
+	aese	$key,$zero
+
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	 veor	$key,$key,$rcon
+	veor	$in0,$in0,$tmp
+	veor	$in0,$in0,$key
+	vst1.32	{$in0},[$out]
+	add	$out,$out,#0x50
+
+	mov	$rounds,#10
+	b	.Ldone
+
+.align	4
+.L192:
+	vld1.8	{$in1},[$inp],#8
+	vmov.i8	$key,#8			// borrow $key
+	vst1.32	{$in0},[$out],#16
+	vsub.i8	$mask,$mask,$key	// adjust the mask
+
+.Loop192:
+	vtbl.8	$key,{$in1},$mask
+	vext.8	$tmp,$zero,$in0,#12
+	vst1.32	{$in1},[$out],#8
+	aese	$key,$zero
+	subs	$bits,$bits,#1
+
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+
+	vdup.32	$tmp,${in0}[3]
+	veor	$tmp,$tmp,$in1
+	 veor	$key,$key,$rcon
+	vext.8	$in1,$zero,$in1,#12
+	vshl.u8	$rcon,$rcon,#1
+	veor	$in1,$in1,$tmp
+	veor	$in0,$in0,$key
+	veor	$in1,$in1,$key
+	vst1.32	{$in0},[$out],#16
+	b.ne	.Loop192
+
+	mov	$rounds,#12
+	add	$out,$out,#0x20
+	b	.Ldone
+
+.align	4
+.L256:
+	vld1.8	{$in1},[$inp]
+	mov	$bits,#7
+	mov	$rounds,#14
+	vst1.32	{$in0},[$out],#16
+
+.Loop256:
+	vtbl.8	$key,{$in1},$mask
+	vext.8	$tmp,$zero,$in0,#12
+	vst1.32	{$in1},[$out],#16
+	aese	$key,$zero
+	subs	$bits,$bits,#1
+
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in0,$in0,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	 veor	$key,$key,$rcon
+	veor	$in0,$in0,$tmp
+	vshl.u8	$rcon,$rcon,#1
+	veor	$in0,$in0,$key
+	vst1.32	{$in0},[$out],#16
+	b.eq	.Ldone
+
+	vdup.32	$key,${in0}[3]		// just splat
+	vext.8	$tmp,$zero,$in1,#12
+	aese	$key,$zero
+
+	veor	$in1,$in1,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in1,$in1,$tmp
+	vext.8	$tmp,$zero,$tmp,#12
+	veor	$in1,$in1,$tmp
+
+	veor	$in1,$in1,$key
+	b	.Loop256
+
+.Ldone:
+	str	$rounds,[$out]
+	mov	$ptr,#0
+
+.Lenc_key_abort:
+	mov	x0,$ptr			// return value
+	`"ldr	x29,[sp],#16"		if ($flavour =~ /64/)`
+	ret
+.size	${prefix}_set_encrypt_key,.-${prefix}_set_encrypt_key
+
+.globl	${prefix}_set_decrypt_key
+.type	${prefix}_set_decrypt_key,%function
+.align	5
+${prefix}_set_decrypt_key:
+___
+$code.=<<___	if ($flavour =~ /64/);
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+___
+$code.=<<___	if ($flavour !~ /64/);
+	stmdb	sp!,{r4,lr}
+___
+$code.=<<___;
+	bl	.Lenc_key
+
+	cmp	x0,#0
+	b.ne	.Ldec_key_abort
+
+	sub	$out,$out,#240		// restore original $out
+	mov	x4,#-16
+	add	$inp,$out,x12,lsl#4	// end of key schedule
+
+	vld1.32	{v0.16b},[$out]
+	vld1.32	{v1.16b},[$inp]
+	vst1.32	{v0.16b},[$inp],x4
+	vst1.32	{v1.16b},[$out],#16
+
+.Loop_imc:
+	vld1.32	{v0.16b},[$out]
+	vld1.32	{v1.16b},[$inp]
+	aesimc	v0.16b,v0.16b
+	aesimc	v1.16b,v1.16b
+	vst1.32	{v0.16b},[$inp],x4
+	vst1.32	{v1.16b},[$out],#16
+	cmp	$inp,$out
+	b.hi	.Loop_imc
+
+	vld1.32	{v0.16b},[$out]
+	aesimc	v0.16b,v0.16b
+	vst1.32	{v0.16b},[$inp]
+
+	eor	x0,x0,x0		// return value
+.Ldec_key_abort:
+___
+$code.=<<___	if ($flavour !~ /64/);
+	ldmia	sp!,{r4,pc}
+___
+$code.=<<___	if ($flavour =~ /64/);
+	ldp	x29,x30,[sp],#16
+	ret
+___
+$code.=<<___;
+.size	${prefix}_set_decrypt_key,.-${prefix}_set_decrypt_key
+___
+}}}
+{{{
+sub gen_block () {
+my $dir = shift;
+my ($e,$mc) = $dir eq "en" ? ("e","mc") : ("d","imc");
+my ($inp,$out,$key)=map("x$_",(0..2));
+my $rounds="w3";
+my ($rndkey0,$rndkey1,$inout)=map("q$_",(0..3));
+
+$code.=<<___;
+.globl	${prefix}_${dir}crypt
+.type	${prefix}_${dir}crypt,%function
+.align	5
+${prefix}_${dir}crypt:
+	ldr	$rounds,[$key,#240]
+	vld1.32	{$rndkey0},[$key],#16
+	vld1.8	{$inout},[$inp]
+	sub	$rounds,$rounds,#2
+	vld1.32	{$rndkey1},[$key],#16
+
+.Loop_${dir}c:
+	aes$e	$inout,$rndkey0
+	vld1.32	{$rndkey0},[$key],#16
+	aes$mc	$inout,$inout
+	subs	$rounds,$rounds,#2
+	aes$e	$inout,$rndkey1
+	vld1.32	{$rndkey1},[$key],#16
+	aes$mc	$inout,$inout
+	b.gt	.Loop_${dir}c
+
+	aes$e	$inout,$rndkey0
+	vld1.32	{$rndkey0},[$key]
+	aes$mc	$inout,$inout
+	aes$e	$inout,$rndkey1
+	veor	$inout,$inout,$rndkey0
+
+	vst1.8	{$inout},[$out]
+	ret
+.size	${prefix}_${dir}crypt,.-${prefix}_${dir}crypt
+___
+}
+&gen_block("en");
+&gen_block("de");
+}}}
+{{{
+my ($inp,$out,$len,$key,$ivp)=map("x$_",(0..4)); my $enc="w5";
+my ($rounds,$cnt,$key_,$step,$step1)=($enc,"w6","x7","x8","x12");
+my ($dat0,$dat1,$in0,$in1,$tmp0,$tmp1,$ivec,$rndlast)=map("q$_",(0..7));
+
+my ($dat,$tmp,$rndzero_n_last)=($dat0,$tmp0,$tmp1);
+
+### q8-q15	preloaded key schedule
+
+$code.=<<___;
+.globl	${prefix}_cbc_encrypt
+.type	${prefix}_cbc_encrypt,%function
+.align	5
+${prefix}_cbc_encrypt:
+___
+$code.=<<___	if ($flavour =~ /64/);
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+___
+$code.=<<___	if ($flavour !~ /64/);
+	mov	ip,sp
+	stmdb	sp!,{r4-r8,lr}
+	vstmdb	sp!,{d8-d15}            @ ABI specification says so
+	ldmia	ip,{r4-r5}		@ load remaining args
+___
+$code.=<<___;
+	subs	$len,$len,#16
+	mov	$step,#16
+	b.lo	.Lcbc_abort
+	cclr	$step,eq
+
+	cmp	$enc,#0			// en- or decrypting?
+	ldr	$rounds,[$key,#240]
+	and	$len,$len,#-16
+	vld1.8	{$ivec},[$ivp]
+	vld1.8	{$dat},[$inp],$step
+
+	vld1.32	{q8-q9},[$key]		// load key schedule...
+	sub	$rounds,$rounds,#6
+	add	$key_,$key,x5,lsl#4	// pointer to last 7 round keys
+	sub	$rounds,$rounds,#2
+	vld1.32	{q10-q11},[$key_],#32
+	vld1.32	{q12-q13},[$key_],#32
+	vld1.32	{q14-q15},[$key_],#32
+	vld1.32	{$rndlast},[$key_]
+
+	add	$key_,$key,#32
+	mov	$cnt,$rounds
+	b.eq	.Lcbc_dec
+
+	cmp	$rounds,#2
+	veor	$dat,$dat,$ivec
+	veor	$rndzero_n_last,q8,$rndlast
+	b.eq	.Lcbc_enc128
+
+.Loop_cbc_enc:
+	aese	$dat,q8
+	vld1.32	{q8},[$key_],#16
+	aesmc	$dat,$dat
+	subs	$cnt,$cnt,#2
+	aese	$dat,q9
+	vld1.32	{q9},[$key_],#16
+	aesmc	$dat,$dat
+	b.gt	.Loop_cbc_enc
+
+	aese	$dat,q8
+	aesmc	$dat,$dat
+	 subs	$len,$len,#16
+	aese	$dat,q9
+	aesmc	$dat,$dat
+	 cclr	$step,eq
+	aese	$dat,q10
+	aesmc	$dat,$dat
+	 add	$key_,$key,#16
+	aese	$dat,q11
+	aesmc	$dat,$dat
+	 vld1.8	{q8},[$inp],$step
+	aese	$dat,q12
+	aesmc	$dat,$dat
+	 veor	q8,q8,$rndzero_n_last
+	aese	$dat,q13
+	aesmc	$dat,$dat
+	 vld1.32 {q9},[$key_],#16	// re-pre-load rndkey[1]
+	aese	$dat,q14
+	aesmc	$dat,$dat
+	aese	$dat,q15
+
+	 mov	$cnt,$rounds
+	veor	$ivec,$dat,$rndlast
+	vst1.8	{$ivec},[$out],#16
+	b.hs	.Loop_cbc_enc
+
+	b	.Lcbc_done
+
+.align	5
+.Lcbc_enc128:
+	vld1.32	{$in0-$in1},[$key_]
+	aese	$dat,q8
+	aesmc	$dat,$dat
+	b	.Lenter_cbc_enc128
+.Loop_cbc_enc128:
+	aese	$dat,q8
+	aesmc	$dat,$dat
+	 vst1.8	{$ivec},[$out],#16
+.Lenter_cbc_enc128:
+	aese	$dat,q9
+	aesmc	$dat,$dat
+	 subs	$len,$len,#16
+	aese	$dat,$in0
+	aesmc	$dat,$dat
+	 cclr	$step,eq
+	aese	$dat,$in1
+	aesmc	$dat,$dat
+	aese	$dat,q10
+	aesmc	$dat,$dat
+	aese	$dat,q11
+	aesmc	$dat,$dat
+	 vld1.8	{q8},[$inp],$step
+	aese	$dat,q12
+	aesmc	$dat,$dat
+	aese	$dat,q13
+	aesmc	$dat,$dat
+	aese	$dat,q14
+	aesmc	$dat,$dat
+	 veor	q8,q8,$rndzero_n_last
+	aese	$dat,q15
+	veor	$ivec,$dat,$rndlast
+	b.hs	.Loop_cbc_enc128
+
+	vst1.8	{$ivec},[$out],#16
+	b	.Lcbc_done
+___
+{
+my ($dat2,$in2,$tmp2)=map("q$_",(10,11,9));
+$code.=<<___;
+.align	5
+.Lcbc_dec:
+	vld1.8	{$dat2},[$inp],#16
+	subs	$len,$len,#32		// bias
+	add	$cnt,$rounds,#2
+	vorr	$in1,$dat,$dat
+	vorr	$dat1,$dat,$dat
+	vorr	$in2,$dat2,$dat2
+	b.lo	.Lcbc_dec_tail
+
+	vorr	$dat1,$dat2,$dat2
+	vld1.8	{$dat2},[$inp],#16
+	vorr	$in0,$dat,$dat
+	vorr	$in1,$dat1,$dat1
+	vorr	$in2,$dat2,$dat2
+
+.Loop3x_cbc_dec:
+	aesd	$dat0,q8
+	aesd	$dat1,q8
+	aesd	$dat2,q8
+	vld1.32	{q8},[$key_],#16
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	subs	$cnt,$cnt,#2
+	aesd	$dat0,q9
+	aesd	$dat1,q9
+	aesd	$dat2,q9
+	vld1.32	{q9},[$key_],#16
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	b.gt	.Loop3x_cbc_dec
+
+	aesd	$dat0,q8
+	aesd	$dat1,q8
+	aesd	$dat2,q8
+	 veor	$tmp0,$ivec,$rndlast
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 veor	$tmp1,$in0,$rndlast
+	aesd	$dat0,q9
+	aesd	$dat1,q9
+	aesd	$dat2,q9
+	 veor	$tmp2,$in1,$rndlast
+	 subs	$len,$len,#0x30
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 vorr	$ivec,$in2,$in2
+	 mov.lo	x6,$len			// x6, $cnt, is zero at this point
+	aesd	$dat0,q12
+	aesd	$dat1,q12
+	aesd	$dat2,q12
+	 add	$inp,$inp,x6		// $inp is adjusted in such way that
+					// at exit from the loop $dat1-$dat2
+					// are loaded with last "words"
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 mov	$key_,$key
+	aesd	$dat0,q13
+	aesd	$dat1,q13
+	aesd	$dat2,q13
+	 vld1.8	{$in0},[$inp],#16
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 vld1.8	{$in1},[$inp],#16
+	aesd	$dat0,q14
+	aesd	$dat1,q14
+	aesd	$dat2,q14
+	 vld1.8	{$in2},[$inp],#16
+	aesimc	$dat0,$dat0
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 vld1.32 {q8},[$key_],#16	// re-pre-load rndkey[0]
+	aesd	$dat0,q15
+	aesd	$dat1,q15
+	aesd	$dat2,q15
+
+	 add	$cnt,$rounds,#2
+	veor	$tmp0,$tmp0,$dat0
+	veor	$tmp1,$tmp1,$dat1
+	veor	$dat2,$dat2,$tmp2
+	 vld1.32 {q9},[$key_],#16	// re-pre-load rndkey[1]
+	 vorr	$dat0,$in0,$in0
+	vst1.8	{$tmp0},[$out],#16
+	 vorr	$dat1,$in1,$in1
+	vst1.8	{$tmp1},[$out],#16
+	vst1.8	{$dat2},[$out],#16
+	 vorr	$dat2,$in2,$in2
+	b.hs	.Loop3x_cbc_dec
+
+	cmn	$len,#0x30
+	b.eq	.Lcbc_done
+	nop
+
+.Lcbc_dec_tail:
+	aesd	$dat1,q8
+	aesd	$dat2,q8
+	vld1.32	{q8},[$key_],#16
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	subs	$cnt,$cnt,#2
+	aesd	$dat1,q9
+	aesd	$dat2,q9
+	vld1.32	{q9},[$key_],#16
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	b.gt	.Lcbc_dec_tail
+
+	aesd	$dat1,q8
+	aesd	$dat2,q8
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	aesd	$dat1,q9
+	aesd	$dat2,q9
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	aesd	$dat1,q12
+	aesd	$dat2,q12
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 cmn	$len,#0x20
+	aesd	$dat1,q13
+	aesd	$dat2,q13
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 veor	$tmp1,$ivec,$rndlast
+	aesd	$dat1,q14
+	aesd	$dat2,q14
+	aesimc	$dat1,$dat1
+	aesimc	$dat2,$dat2
+	 veor	$tmp2,$in1,$rndlast
+	aesd	$dat1,q15
+	aesd	$dat2,q15
+	b.eq	.Lcbc_dec_one
+	veor	$tmp1,$tmp1,$dat1
+	veor	$tmp2,$tmp2,$dat2
+	 vorr	$ivec,$in2,$in2
+	vst1.8	{$tmp1},[$out],#16
+	vst1.8	{$tmp2},[$out],#16
+	b	.Lcbc_done
+
+.Lcbc_dec_one:
+	veor	$tmp1,$tmp1,$dat2
+	 vorr	$ivec,$in2,$in2
+	vst1.8	{$tmp1},[$out],#16
+
+.Lcbc_done:
+	vst1.8	{$ivec},[$ivp]
+.Lcbc_abort:
+___
+}
+$code.=<<___	if ($flavour !~ /64/);
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r8,pc}
+___
+$code.=<<___	if ($flavour =~ /64/);
+	ldr	x29,[sp],#16
+	ret
+___
+$code.=<<___;
+.size	${prefix}_cbc_encrypt,.-${prefix}_cbc_encrypt
+___
+}}}
+{{{
+my ($inp,$out,$len,$key,$ivp)=map("x$_",(0..4));
+my ($rounds,$cnt,$key_)=("w5","w6","x7");
+my ($ctr,$tctr0,$tctr1,$tctr2)=map("w$_",(8..10,12));
+my $step="x12";		# aliases with $tctr2
+
+my ($dat0,$dat1,$in0,$in1,$tmp0,$tmp1,$ivec,$rndlast)=map("q$_",(0..7));
+my ($dat2,$in2,$tmp2)=map("q$_",(10,11,9));
+
+my ($dat,$tmp)=($dat0,$tmp0);
+
+### q8-q15	preloaded key schedule
+
+$code.=<<___;
+.globl	${prefix}_ctr32_encrypt_blocks
+.type	${prefix}_ctr32_encrypt_blocks,%function
+.align	5
+${prefix}_ctr32_encrypt_blocks:
+___
+$code.=<<___	if ($flavour =~ /64/);
+	stp		x29,x30,[sp,#-16]!
+	add		x29,sp,#0
+___
+$code.=<<___	if ($flavour !~ /64/);
+	mov		ip,sp
+	stmdb		sp!,{r4-r10,lr}
+	vstmdb		sp!,{d8-d15}            @ ABI specification says so
+	ldr		r4, [ip]		@ load remaining arg
+___
+$code.=<<___;
+	ldr		$rounds,[$key,#240]
+
+	ldr		$ctr, [$ivp, #12]
+	vld1.32		{$dat0},[$ivp]
+
+	vld1.32		{q8-q9},[$key]		// load key schedule...
+	sub		$rounds,$rounds,#4
+	mov		$step,#16
+	cmp		$len,#2
+	add		$key_,$key,x5,lsl#4	// pointer to last 5 round keys
+	sub		$rounds,$rounds,#2
+	vld1.32		{q12-q13},[$key_],#32
+	vld1.32		{q14-q15},[$key_],#32
+	vld1.32		{$rndlast},[$key_]
+	add		$key_,$key,#32
+	mov		$cnt,$rounds
+	cclr		$step,lo
+#ifndef __ARMEB__
+	rev		$ctr, $ctr
+#endif
+	vorr		$dat1,$dat0,$dat0
+	add		$tctr1, $ctr, #1
+	vorr		$dat2,$dat0,$dat0
+	add		$ctr, $ctr, #2
+	vorr		$ivec,$dat0,$dat0
+	rev		$tctr1, $tctr1
+	vmov.32		${dat1}[3],$tctr1
+	b.ls		.Lctr32_tail
+	rev		$tctr2, $ctr
+	sub		$len,$len,#3		// bias
+	vmov.32		${dat2}[3],$tctr2
+	b		.Loop3x_ctr32
+
+.align	4
+.Loop3x_ctr32:
+	aese		$dat0,q8
+	aese		$dat1,q8
+	aese		$dat2,q8
+	vld1.32		{q8},[$key_],#16
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	aesmc		$dat2,$dat2
+	subs		$cnt,$cnt,#2
+	aese		$dat0,q9
+	aese		$dat1,q9
+	aese		$dat2,q9
+	vld1.32		{q9},[$key_],#16
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	aesmc		$dat2,$dat2
+	b.gt		.Loop3x_ctr32
+
+	aese		$dat0,q8
+	aese		$dat1,q8
+	aese		$dat2,q8
+	 mov		$key_,$key
+	aesmc		$tmp0,$dat0
+	 vld1.8		{$in0},[$inp],#16
+	aesmc		$tmp1,$dat1
+	aesmc		$dat2,$dat2
+	 vorr		$dat0,$ivec,$ivec
+	aese		$tmp0,q9
+	 vld1.8		{$in1},[$inp],#16
+	aese		$tmp1,q9
+	aese		$dat2,q9
+	 vorr		$dat1,$ivec,$ivec
+	aesmc		$tmp0,$tmp0
+	 vld1.8		{$in2},[$inp],#16
+	aesmc		$tmp1,$tmp1
+	aesmc		$tmp2,$dat2
+	 vorr		$dat2,$ivec,$ivec
+	 add		$tctr0,$ctr,#1
+	aese		$tmp0,q12
+	aese		$tmp1,q12
+	aese		$tmp2,q12
+	 veor		$in0,$in0,$rndlast
+	 add		$tctr1,$ctr,#2
+	aesmc		$tmp0,$tmp0
+	aesmc		$tmp1,$tmp1
+	aesmc		$tmp2,$tmp2
+	 veor		$in1,$in1,$rndlast
+	 add		$ctr,$ctr,#3
+	aese		$tmp0,q13
+	aese		$tmp1,q13
+	aese		$tmp2,q13
+	 veor		$in2,$in2,$rndlast
+	 rev		$tctr0,$tctr0
+	aesmc		$tmp0,$tmp0
+	 vld1.32	 {q8},[$key_],#16	// re-pre-load rndkey[0]
+	aesmc		$tmp1,$tmp1
+	aesmc		$tmp2,$tmp2
+	 vmov.32	${dat0}[3], $tctr0
+	 rev		$tctr1,$tctr1
+	aese		$tmp0,q14
+	aese		$tmp1,q14
+	aese		$tmp2,q14
+	 vmov.32	${dat1}[3], $tctr1
+	 rev		$tctr2,$ctr
+	aesmc		$tmp0,$tmp0
+	aesmc		$tmp1,$tmp1
+	aesmc		$tmp2,$tmp2
+	 vmov.32	${dat2}[3], $tctr2
+	 subs		$len,$len,#3
+	aese		$tmp0,q15
+	aese		$tmp1,q15
+	aese		$tmp2,q15
+
+	 mov		$cnt,$rounds
+	veor		$in0,$in0,$tmp0
+	veor		$in1,$in1,$tmp1
+	veor		$in2,$in2,$tmp2
+	 vld1.32	 {q9},[$key_],#16	// re-pre-load rndkey[1]
+	vst1.8		{$in0},[$out],#16
+	vst1.8		{$in1},[$out],#16
+	vst1.8		{$in2},[$out],#16
+	b.hs		.Loop3x_ctr32
+
+	adds		$len,$len,#3
+	b.eq		.Lctr32_done
+	cmp		$len,#1
+	mov		$step,#16
+	cclr		$step,eq
+
+.Lctr32_tail:
+	aese		$dat0,q8
+	aese		$dat1,q8
+	vld1.32		{q8},[$key_],#16
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	subs		$cnt,$cnt,#2
+	aese		$dat0,q9
+	aese		$dat1,q9
+	vld1.32		{q9},[$key_],#16
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	b.gt		.Lctr32_tail
+
+	aese		$dat0,q8
+	aese		$dat1,q8
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	aese		$dat0,q9
+	aese		$dat1,q9
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	 vld1.8		{$in0},[$inp],$step
+	aese		$dat0,q12
+	aese		$dat1,q12
+	 vld1.8		{$in1},[$inp]
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	aese		$dat0,q13
+	aese		$dat1,q13
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	aese		$dat0,q14
+	aese		$dat1,q14
+	 veor		$in0,$in0,$rndlast
+	aesmc		$dat0,$dat0
+	aesmc		$dat1,$dat1
+	 veor		$in1,$in1,$rndlast
+	aese		$dat0,q15
+	aese		$dat1,q15
+
+	cmp		$len,#1
+	veor		$in0,$in0,$dat0
+	veor		$in1,$in1,$dat1
+	vst1.8		{$in0},[$out],#16
+	b.eq		.Lctr32_done
+	vst1.8		{$in1},[$out]
+
+.Lctr32_done:
+___
+$code.=<<___	if ($flavour !~ /64/);
+	vldmia		sp!,{d8-d15}
+	ldmia		sp!,{r4-r10,pc}
+___
+$code.=<<___	if ($flavour =~ /64/);
+	ldr		x29,[sp],#16
+	ret
+___
+$code.=<<___;
+.size	${prefix}_ctr32_encrypt_blocks,.-${prefix}_ctr32_encrypt_blocks
+___
+}}}
+$code.=<<___;
+#endif
+___
+########################################
+if ($flavour =~ /64/) {			######## 64-bit code
+    my %opcode = (
+	"aesd"	=>	0x4e285800,	"aese"	=>	0x4e284800,
+	"aesimc"=>	0x4e287800,	"aesmc"	=>	0x4e286800	);
+
+    local *unaes = sub {
+	my ($mnemonic,$arg)=@_;
+
+	$arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)/o	&&
+	sprintf ".inst\t0x%08x\t//%s %s",
+			$opcode{$mnemonic}|$1|($2<<5),
+			$mnemonic,$arg;
+    };
+
+    foreach(split("\n",$code)) {
+	s/\`([^\`]*)\`/eval($1)/geo;
+
+	s/\bq([0-9]+)\b/"v".($1<8?$1:$1+8).".16b"/geo;	# old->new registers
+	s/@\s/\/\//o;			# old->new style commentary
+
+	#s/[v]?(aes\w+)\s+([qv].*)/unaes($1,$2)/geo	or
+	s/cclr\s+([wx])([^,]+),\s*([a-z]+)/csel	$1$2,$1zr,$1$2,$3/o	or
+	s/mov\.([a-z]+)\s+([wx][0-9]+),\s*([wx][0-9]+)/csel	$2,$3,$2,$1/o	or
+	s/vmov\.i8/movi/o	or	# fix up legacy mnemonics
+	s/vext\.8/ext/o		or
+	s/vrev32\.8/rev32/o	or
+	s/vtst\.8/cmtst/o	or
+	s/vshr/ushr/o		or
+	s/^(\s+)v/$1/o		or	# strip off v prefix
+	s/\bbx\s+lr\b/ret/o;
+
+	# fix up remainig legacy suffixes
+	s/\.[ui]?8//o;
+	m/\],#8/o and s/\.16b/\.8b/go;
+	s/\.[ui]?32//o and s/\.16b/\.4s/go;
+	s/\.[ui]?64//o and s/\.16b/\.2d/go;
+	s/\.[42]([sd])\[([0-3])\]/\.$1\[$2\]/o;
+
+	print $_,"\n";
+    }
+} else {				######## 32-bit code
+    my %opcode = (
+	"aesd"	=>	0xf3b00340,	"aese"	=>	0xf3b00300,
+	"aesimc"=>	0xf3b003c0,	"aesmc"	=>	0xf3b00380	);
+
+    local *unaes = sub {
+	my ($mnemonic,$arg)=@_;
+
+	if ($arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)/o) {
+	    my $word = $opcode{$mnemonic}|(($1&7)<<13)|(($1&8)<<19)
+					 |(($2&7)<<1) |(($2&8)<<2);
+	    # since ARMv7 instructions are always encoded little-endian.
+	    # correct solution is to use .inst directive, but older
+	    # assemblers don't implement it:-(
+	    sprintf ".byte\t0x%02x,0x%02x,0x%02x,0x%02x\t@ %s %s",
+			$word&0xff,($word>>8)&0xff,
+			($word>>16)&0xff,($word>>24)&0xff,
+			$mnemonic,$arg;
+	}
+    };
+
+    sub unvtbl {
+	my $arg=shift;
+
+	$arg =~ m/q([0-9]+),\s*\{q([0-9]+)\},\s*q([0-9]+)/o &&
+	sprintf	"vtbl.8	d%d,{q%d},d%d\n\t".
+		"vtbl.8	d%d,{q%d},d%d", 2*$1,$2,2*$3, 2*$1+1,$2,2*$3+1;	
+    }
+
+    sub unvdup32 {
+	my $arg=shift;
+
+	$arg =~ m/q([0-9]+),\s*q([0-9]+)\[([0-3])\]/o &&
+	sprintf	"vdup.32	q%d,d%d[%d]",$1,2*$2+($3>>1),$3&1;	
+    }
+
+    sub unvmov32 {
+	my $arg=shift;
+
+	$arg =~ m/q([0-9]+)\[([0-3])\],(.*)/o &&
+	sprintf	"vmov.32	d%d[%d],%s",2*$1+($2>>1),$2&1,$3;	
+    }
+
+    foreach(split("\n",$code)) {
+	s/\`([^\`]*)\`/eval($1)/geo;
+
+	s/\b[wx]([0-9]+)\b/r$1/go;		# new->old registers
+	s/\bv([0-9])\.[12468]+[bsd]\b/q$1/go;	# new->old registers
+	s/\/\/\s?/@ /o;				# new->old style commentary
+
+	# fix up remainig new-style suffixes
+	s/\{q([0-9]+)\},\s*\[(.+)\],#8/sprintf "{d%d},[$2]!",2*$1/eo	or
+	s/\],#[0-9]+/]!/o;
+
+	s/[v]?(aes\w+)\s+([qv].*)/unaes($1,$2)/geo	or
+	s/cclr\s+([^,]+),\s*([a-z]+)/mov$2	$1,#0/o	or
+	s/vtbl\.8\s+(.*)/unvtbl($1)/geo			or
+	s/vdup\.32\s+(.*)/unvdup32($1)/geo		or
+	s/vmov\.32\s+(.*)/unvmov32($1)/geo		or
+	s/^(\s+)b\./$1b/o				or
+	s/^(\s+)mov\./$1mov/o				or
+	s/^(\s+)ret/$1bx\tlr/o;
+
+	print $_,"\n";
+    }
+}
+
+close STDOUT;
diff --git a/src/crypto/aes/asm/bsaes-armv7.pl b/src/crypto/aes/asm/bsaes-armv7.pl
new file mode 100644
index 0000000..d70f3ea
--- /dev/null
+++ b/src/crypto/aes/asm/bsaes-armv7.pl
@@ -0,0 +1,2477 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> 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/.
+#
+# Specific modes and adaptation for Linux kernel by Ard Biesheuvel
+# <ard.biesheuvel@linaro.org>. Permission to use under GPL terms is
+# granted.
+# ====================================================================
+
+# Bit-sliced AES for ARM NEON
+#
+# February 2012.
+#
+# This implementation is direct adaptation of bsaes-x86_64 module for
+# ARM NEON. Except that this module is endian-neutral [in sense that
+# it can be compiled for either endianness] by courtesy of vld1.8's
+# neutrality. Initial version doesn't implement interface to OpenSSL,
+# only low-level primitives and unsupported entry points, just enough
+# to collect performance results, which for Cortex-A8 core are:
+#
+# encrypt	19.5 cycles per byte processed with 128-bit key
+# decrypt	22.1 cycles per byte processed with 128-bit key
+# key conv.	440  cycles per 128-bit key/0.18 of 8x block
+#
+# Snapdragon S4 encrypts byte in 17.6 cycles and decrypts in 19.7,
+# which is [much] worse than anticipated (for further details see
+# http://www.openssl.org/~appro/Snapdragon-S4.html).
+#
+# Cortex-A15 manages in 14.2/16.1 cycles [when integer-only code
+# manages in 20.0 cycles].
+#
+# When comparing to x86_64 results keep in mind that NEON unit is
+# [mostly] single-issue and thus can't [fully] benefit from
+# instruction-level parallelism. And when comparing to aes-armv4
+# results keep in mind key schedule conversion overhead (see
+# bsaes-x86_64.pl for further details)...
+#
+#						<appro@openssl.org>
+
+# April-August 2013
+#
+# Add CBC, CTR and XTS subroutines, adapt for kernel use.
+#
+#					<ard.biesheuvel@linaro.org>
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
+my ($inp,$out,$len,$key)=("r0","r1","r2","r3");
+my @XMM=map("q$_",(0..15));
+
+{
+my ($key,$rounds,$const)=("r4","r5","r6");
+
+sub Dlo()   { shift=~m|q([1]?[0-9])|?"d".($1*2):"";     }
+sub Dhi()   { shift=~m|q([1]?[0-9])|?"d".($1*2+1):"";   }
+
+sub Sbox {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b0, b1, b4, b6, b3, b7, b2, b5] < msb
+my @b=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+	&InBasisChange	(@b);
+	&Inv_GF256	(@b[6,5,0,3,7,1,4,2],@t,@s);
+	&OutBasisChange	(@b[7,1,4,2,6,5,0,3]);
+}
+
+sub InBasisChange {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b6, b5, b0, b3, b7, b1, b4, b2] < msb 
+my @b=@_[0..7];
+$code.=<<___;
+	veor	@b[2], @b[2], @b[1]
+	veor	@b[5], @b[5], @b[6]
+	veor	@b[3], @b[3], @b[0]
+	veor	@b[6], @b[6], @b[2]
+	veor	@b[5], @b[5], @b[0]
+
+	veor	@b[6], @b[6], @b[3]
+	veor	@b[3], @b[3], @b[7]
+	veor	@b[7], @b[7], @b[5]
+	veor	@b[3], @b[3], @b[4]
+	veor	@b[4], @b[4], @b[5]
+
+	veor	@b[2], @b[2], @b[7]
+	veor	@b[3], @b[3], @b[1]
+	veor	@b[1], @b[1], @b[5]
+___
+}
+
+sub OutBasisChange {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b6, b1, b2, b4, b7, b0, b3, b5] < msb
+my @b=@_[0..7];
+$code.=<<___;
+	veor	@b[0], @b[0], @b[6]
+	veor	@b[1], @b[1], @b[4]
+	veor	@b[4], @b[4], @b[6]
+	veor	@b[2], @b[2], @b[0]
+	veor	@b[6], @b[6], @b[1]
+
+	veor	@b[1], @b[1], @b[5]
+	veor	@b[5], @b[5], @b[3]
+	veor	@b[3], @b[3], @b[7]
+	veor	@b[7], @b[7], @b[5]
+	veor	@b[2], @b[2], @b[5]
+
+	veor	@b[4], @b[4], @b[7]
+___
+}
+
+sub InvSbox {
+# input in lsb 	> [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb	> [b0, b1, b6, b4, b2, b7, b3, b5] < msb
+my @b=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+	&InvInBasisChange	(@b);
+	&Inv_GF256		(@b[5,1,2,6,3,7,0,4],@t,@s);
+	&InvOutBasisChange	(@b[3,7,0,4,5,1,2,6]);
+}
+
+sub InvInBasisChange {		# OutBasisChange in reverse (with twist)
+my @b=@_[5,1,2,6,3,7,0,4];
+$code.=<<___
+	 veor	@b[1], @b[1], @b[7]
+	veor	@b[4], @b[4], @b[7]
+
+	veor	@b[7], @b[7], @b[5]
+	 veor	@b[1], @b[1], @b[3]
+	veor	@b[2], @b[2], @b[5]
+	veor	@b[3], @b[3], @b[7]
+
+	veor	@b[6], @b[6], @b[1]
+	veor	@b[2], @b[2], @b[0]
+	 veor	@b[5], @b[5], @b[3]
+	veor	@b[4], @b[4], @b[6]
+	veor	@b[0], @b[0], @b[6]
+	veor	@b[1], @b[1], @b[4]
+___
+}
+
+sub InvOutBasisChange {		# InBasisChange in reverse
+my @b=@_[2,5,7,3,6,1,0,4];
+$code.=<<___;
+	veor	@b[1], @b[1], @b[5]
+	veor	@b[2], @b[2], @b[7]
+
+	veor	@b[3], @b[3], @b[1]
+	veor	@b[4], @b[4], @b[5]
+	veor	@b[7], @b[7], @b[5]
+	veor	@b[3], @b[3], @b[4]
+	 veor 	@b[5], @b[5], @b[0]
+	veor	@b[3], @b[3], @b[7]
+	 veor	@b[6], @b[6], @b[2]
+	 veor	@b[2], @b[2], @b[1]
+	veor	@b[6], @b[6], @b[3]
+
+	veor	@b[3], @b[3], @b[0]
+	veor	@b[5], @b[5], @b[6]
+___
+}
+
+sub Mul_GF4 {
+#;*************************************************************
+#;* Mul_GF4: Input x0-x1,y0-y1 Output x0-x1 Temp t0 (8) *
+#;*************************************************************
+my ($x0,$x1,$y0,$y1,$t0,$t1)=@_;
+$code.=<<___;
+	veor 	$t0, $y0, $y1
+	vand	$t0, $t0, $x0
+	veor	$x0, $x0, $x1
+	vand	$t1, $x1, $y0
+	vand	$x0, $x0, $y1
+	veor	$x1, $t1, $t0
+	veor	$x0, $x0, $t1
+___
+}
+
+sub Mul_GF4_N {				# not used, see next subroutine
+# multiply and scale by N
+my ($x0,$x1,$y0,$y1,$t0)=@_;
+$code.=<<___;
+	veor	$t0, $y0, $y1
+	vand	$t0, $t0, $x0
+	veor	$x0, $x0, $x1
+	vand	$x1, $x1, $y0
+	vand	$x0, $x0, $y1
+	veor	$x1, $x1, $x0
+	veor	$x0, $x0, $t0
+___
+}
+
+sub Mul_GF4_N_GF4 {
+# interleaved Mul_GF4_N and Mul_GF4
+my ($x0,$x1,$y0,$y1,$t0,
+    $x2,$x3,$y2,$y3,$t1)=@_;
+$code.=<<___;
+	veor	$t0, $y0, $y1
+	 veor 	$t1, $y2, $y3
+	vand	$t0, $t0, $x0
+	 vand	$t1, $t1, $x2
+	veor	$x0, $x0, $x1
+	 veor	$x2, $x2, $x3
+	vand	$x1, $x1, $y0
+	 vand	$x3, $x3, $y2
+	vand	$x0, $x0, $y1
+	 vand	$x2, $x2, $y3
+	veor	$x1, $x1, $x0
+	 veor	$x2, $x2, $x3
+	veor	$x0, $x0, $t0
+	 veor	$x3, $x3, $t1
+___
+}
+sub Mul_GF16_2 {
+my @x=@_[0..7];
+my @y=@_[8..11];
+my @t=@_[12..15];
+$code.=<<___;
+	veor	@t[0], @x[0], @x[2]
+	veor	@t[1], @x[1], @x[3]
+___
+	&Mul_GF4  	(@x[0], @x[1], @y[0], @y[1], @t[2..3]);
+$code.=<<___;
+	veor	@y[0], @y[0], @y[2]
+	veor	@y[1], @y[1], @y[3]
+___
+	Mul_GF4_N_GF4	(@t[0], @t[1], @y[0], @y[1], @t[3],
+			 @x[2], @x[3], @y[2], @y[3], @t[2]);
+$code.=<<___;
+	veor	@x[0], @x[0], @t[0]
+	veor	@x[2], @x[2], @t[0]
+	veor	@x[1], @x[1], @t[1]
+	veor	@x[3], @x[3], @t[1]
+
+	veor	@t[0], @x[4], @x[6]
+	veor	@t[1], @x[5], @x[7]
+___
+	&Mul_GF4_N_GF4	(@t[0], @t[1], @y[0], @y[1], @t[3],
+			 @x[6], @x[7], @y[2], @y[3], @t[2]);
+$code.=<<___;
+	veor	@y[0], @y[0], @y[2]
+	veor	@y[1], @y[1], @y[3]
+___
+	&Mul_GF4  	(@x[4], @x[5], @y[0], @y[1], @t[2..3]);
+$code.=<<___;
+	veor	@x[4], @x[4], @t[0]
+	veor	@x[6], @x[6], @t[0]
+	veor	@x[5], @x[5], @t[1]
+	veor	@x[7], @x[7], @t[1]
+___
+}
+sub Inv_GF256 {
+#;********************************************************************
+#;* Inv_GF256: Input x0-x7 Output x0-x7 Temp t0-t3,s0-s3 (144)       *
+#;********************************************************************
+my @x=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+# direct optimizations from hardware
+$code.=<<___;
+	veor	@t[3], @x[4], @x[6]
+	veor	@t[2], @x[5], @x[7]
+	veor	@t[1], @x[1], @x[3]
+	veor	@s[1], @x[7], @x[6]
+	 vmov	@t[0], @t[2]
+	veor	@s[0], @x[0], @x[2]
+
+	vorr	@t[2], @t[2], @t[1]
+	veor	@s[3], @t[3], @t[0]
+	vand	@s[2], @t[3], @s[0]
+	vorr	@t[3], @t[3], @s[0]
+	veor	@s[0], @s[0], @t[1]
+	vand	@t[0], @t[0], @t[1]
+	veor	@t[1], @x[3], @x[2]
+	vand	@s[3], @s[3], @s[0]
+	vand	@s[1], @s[1], @t[1]
+	veor	@t[1], @x[4], @x[5]
+	veor	@s[0], @x[1], @x[0]
+	veor	@t[3], @t[3], @s[1]
+	veor	@t[2], @t[2], @s[1]
+	vand	@s[1], @t[1], @s[0]
+	vorr	@t[1], @t[1], @s[0]
+	veor	@t[3], @t[3], @s[3]
+	veor	@t[0], @t[0], @s[1]
+	veor	@t[2], @t[2], @s[2]
+	veor	@t[1], @t[1], @s[3]
+	veor	@t[0], @t[0], @s[2]
+	vand	@s[0], @x[7], @x[3]
+	veor	@t[1], @t[1], @s[2]
+	vand	@s[1], @x[6], @x[2]
+	vand	@s[2], @x[5], @x[1]
+	vorr	@s[3], @x[4], @x[0]
+	veor	@t[3], @t[3], @s[0]
+	veor	@t[1], @t[1], @s[2]
+	veor	@t[0], @t[0], @s[3]
+	veor	@t[2], @t[2], @s[1]
+
+	@ Inv_GF16 \t0, \t1, \t2, \t3, \s0, \s1, \s2, \s3
+
+	@ new smaller inversion
+
+	vand	@s[2], @t[3], @t[1]
+	vmov	@s[0], @t[0]
+
+	veor	@s[1], @t[2], @s[2]
+	veor	@s[3], @t[0], @s[2]
+	veor	@s[2], @t[0], @s[2]	@ @s[2]=@s[3]
+
+	vbsl	@s[1], @t[1], @t[0]
+	vbsl	@s[3], @t[3], @t[2]
+	veor	@t[3], @t[3], @t[2]
+
+	vbsl	@s[0], @s[1], @s[2]
+	vbsl	@t[0], @s[2], @s[1]
+
+	vand	@s[2], @s[0], @s[3]
+	veor	@t[1], @t[1], @t[0]
+
+	veor	@s[2], @s[2], @t[3]
+___
+# output in s3, s2, s1, t1
+
+# Mul_GF16_2 \x0, \x1, \x2, \x3, \x4, \x5, \x6, \x7, \t2, \t3, \t0, \t1, \s0, \s1, \s2, \s3
+
+# Mul_GF16_2 \x0, \x1, \x2, \x3, \x4, \x5, \x6, \x7, \s3, \s2, \s1, \t1, \s0, \t0, \t2, \t3
+	&Mul_GF16_2(@x,@s[3,2,1],@t[1],@s[0],@t[0,2,3]);
+
+### output msb > [x3,x2,x1,x0,x7,x6,x5,x4] < lsb
+}
+
+# AES linear components
+
+sub ShiftRows {
+my @x=@_[0..7];
+my @t=@_[8..11];
+my $mask=pop;
+$code.=<<___;
+	vldmia	$key!, {@t[0]-@t[3]}
+	veor	@t[0], @t[0], @x[0]
+	veor	@t[1], @t[1], @x[1]
+	vtbl.8	`&Dlo(@x[0])`, {@t[0]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[0])`, {@t[0]}, `&Dhi($mask)`
+	vldmia	$key!, {@t[0]}
+	veor	@t[2], @t[2], @x[2]
+	vtbl.8	`&Dlo(@x[1])`, {@t[1]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[1])`, {@t[1]}, `&Dhi($mask)`
+	vldmia	$key!, {@t[1]}
+	veor	@t[3], @t[3], @x[3]
+	vtbl.8	`&Dlo(@x[2])`, {@t[2]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[2])`, {@t[2]}, `&Dhi($mask)`
+	vldmia	$key!, {@t[2]}
+	vtbl.8	`&Dlo(@x[3])`, {@t[3]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[3])`, {@t[3]}, `&Dhi($mask)`
+	vldmia	$key!, {@t[3]}
+	veor	@t[0], @t[0], @x[4]
+	veor	@t[1], @t[1], @x[5]
+	vtbl.8	`&Dlo(@x[4])`, {@t[0]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[4])`, {@t[0]}, `&Dhi($mask)`
+	veor	@t[2], @t[2], @x[6]
+	vtbl.8	`&Dlo(@x[5])`, {@t[1]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[5])`, {@t[1]}, `&Dhi($mask)`
+	veor	@t[3], @t[3], @x[7]
+	vtbl.8	`&Dlo(@x[6])`, {@t[2]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[6])`, {@t[2]}, `&Dhi($mask)`
+	vtbl.8	`&Dlo(@x[7])`, {@t[3]}, `&Dlo($mask)`
+	vtbl.8	`&Dhi(@x[7])`, {@t[3]}, `&Dhi($mask)`
+___
+}
+
+sub MixColumns {
+# modified to emit output in order suitable for feeding back to aesenc[last]
+my @x=@_[0..7];
+my @t=@_[8..15];
+my $inv=@_[16];	# optional
+$code.=<<___;
+	vext.8	@t[0], @x[0], @x[0], #12	@ x0 <<< 32
+	vext.8	@t[1], @x[1], @x[1], #12
+	 veor	@x[0], @x[0], @t[0]		@ x0 ^ (x0 <<< 32)
+	vext.8	@t[2], @x[2], @x[2], #12
+	 veor	@x[1], @x[1], @t[1]
+	vext.8	@t[3], @x[3], @x[3], #12
+	 veor	@x[2], @x[2], @t[2]
+	vext.8	@t[4], @x[4], @x[4], #12
+	 veor	@x[3], @x[3], @t[3]
+	vext.8	@t[5], @x[5], @x[5], #12
+	 veor	@x[4], @x[4], @t[4]
+	vext.8	@t[6], @x[6], @x[6], #12
+	 veor	@x[5], @x[5], @t[5]
+	vext.8	@t[7], @x[7], @x[7], #12
+	 veor	@x[6], @x[6], @t[6]
+
+	veor	@t[1], @t[1], @x[0]
+	 veor	@x[7], @x[7], @t[7]
+	 vext.8	@x[0], @x[0], @x[0], #8		@ (x0 ^ (x0 <<< 32)) <<< 64)
+	veor	@t[2], @t[2], @x[1]
+	veor	@t[0], @t[0], @x[7]
+	veor	@t[1], @t[1], @x[7]
+	 vext.8	@x[1], @x[1], @x[1], #8
+	veor	@t[5], @t[5], @x[4]
+	 veor	@x[0], @x[0], @t[0]
+	veor	@t[6], @t[6], @x[5]
+	 veor	@x[1], @x[1], @t[1]
+	 vext.8	@t[0], @x[4], @x[4], #8
+	veor	@t[4], @t[4], @x[3]
+	 vext.8	@t[1], @x[5], @x[5], #8
+	veor	@t[7], @t[7], @x[6]
+	 vext.8	@x[4], @x[3], @x[3], #8
+	veor	@t[3], @t[3], @x[2]
+	 vext.8	@x[5], @x[7], @x[7], #8
+	veor	@t[4], @t[4], @x[7]
+	 vext.8	@x[3], @x[6], @x[6], #8
+	veor	@t[3], @t[3], @x[7]
+	 vext.8	@x[6], @x[2], @x[2], #8
+	veor	@x[7], @t[1], @t[5]
+___
+$code.=<<___ if (!$inv);
+	veor	@x[2], @t[0], @t[4]
+	veor	@x[4], @x[4], @t[3]
+	veor	@x[5], @x[5], @t[7]
+	veor	@x[3], @x[3], @t[6]
+	 @ vmov	@x[2], @t[0]
+	veor	@x[6], @x[6], @t[2]
+	 @ vmov	@x[7], @t[1]
+___
+$code.=<<___ if ($inv);
+	veor	@t[3], @t[3], @x[4]
+	veor	@x[5], @x[5], @t[7]
+	veor	@x[2], @x[3], @t[6]
+	veor	@x[3], @t[0], @t[4]
+	veor	@x[4], @x[6], @t[2]
+	vmov	@x[6], @t[3]
+	 @ vmov	@x[7], @t[1]
+___
+}
+
+sub InvMixColumns_orig {
+my @x=@_[0..7];
+my @t=@_[8..15];
+
+$code.=<<___;
+	@ multiplication by 0x0e
+	vext.8	@t[7], @x[7], @x[7], #12
+	vmov	@t[2], @x[2]
+	veor	@x[2], @x[2], @x[5]		@ 2 5
+	veor	@x[7], @x[7], @x[5]		@ 7 5
+	vext.8	@t[0], @x[0], @x[0], #12
+	vmov	@t[5], @x[5]
+	veor	@x[5], @x[5], @x[0]		@ 5 0		[1]
+	veor	@x[0], @x[0], @x[1]		@ 0 1
+	vext.8	@t[1], @x[1], @x[1], #12
+	veor	@x[1], @x[1], @x[2]		@ 1 25
+	veor	@x[0], @x[0], @x[6]		@ 01 6		[2]
+	vext.8	@t[3], @x[3], @x[3], #12
+	veor	@x[1], @x[1], @x[3]		@ 125 3		[4]
+	veor	@x[2], @x[2], @x[0]		@ 25 016	[3]
+	veor	@x[3], @x[3], @x[7]		@ 3 75
+	veor	@x[7], @x[7], @x[6]		@ 75 6		[0]
+	vext.8	@t[6], @x[6], @x[6], #12
+	vmov	@t[4], @x[4]
+	veor	@x[6], @x[6], @x[4]		@ 6 4
+	veor	@x[4], @x[4], @x[3]		@ 4 375		[6]
+	veor	@x[3], @x[3], @x[7]		@ 375 756=36
+	veor	@x[6], @x[6], @t[5]		@ 64 5		[7]
+	veor	@x[3], @x[3], @t[2]		@ 36 2
+	vext.8	@t[5], @t[5], @t[5], #12
+	veor	@x[3], @x[3], @t[4]		@ 362 4		[5]
+___
+					my @y = @x[7,5,0,2,1,3,4,6];
+$code.=<<___;
+	@ multiplication by 0x0b
+	veor	@y[1], @y[1], @y[0]
+	veor	@y[0], @y[0], @t[0]
+	vext.8	@t[2], @t[2], @t[2], #12
+	veor	@y[1], @y[1], @t[1]
+	veor	@y[0], @y[0], @t[5]
+	vext.8	@t[4], @t[4], @t[4], #12
+	veor	@y[1], @y[1], @t[6]
+	veor	@y[0], @y[0], @t[7]
+	veor	@t[7], @t[7], @t[6]		@ clobber t[7]
+
+	veor	@y[3], @y[3], @t[0]
+	 veor	@y[1], @y[1], @y[0]
+	vext.8	@t[0], @t[0], @t[0], #12
+	veor	@y[2], @y[2], @t[1]
+	veor	@y[4], @y[4], @t[1]
+	vext.8	@t[1], @t[1], @t[1], #12
+	veor	@y[2], @y[2], @t[2]
+	veor	@y[3], @y[3], @t[2]
+	veor	@y[5], @y[5], @t[2]
+	veor	@y[2], @y[2], @t[7]
+	vext.8	@t[2], @t[2], @t[2], #12
+	veor	@y[3], @y[3], @t[3]
+	veor	@y[6], @y[6], @t[3]
+	veor	@y[4], @y[4], @t[3]
+	veor	@y[7], @y[7], @t[4]
+	vext.8	@t[3], @t[3], @t[3], #12
+	veor	@y[5], @y[5], @t[4]
+	veor	@y[7], @y[7], @t[7]
+	veor	@t[7], @t[7], @t[5]		@ clobber t[7] even more
+	veor	@y[3], @y[3], @t[5]
+	veor	@y[4], @y[4], @t[4]
+
+	veor	@y[5], @y[5], @t[7]
+	vext.8	@t[4], @t[4], @t[4], #12
+	veor	@y[6], @y[6], @t[7]
+	veor	@y[4], @y[4], @t[7]
+
+	veor	@t[7], @t[7], @t[5]
+	vext.8	@t[5], @t[5], @t[5], #12
+
+	@ multiplication by 0x0d
+	veor	@y[4], @y[4], @y[7]
+	 veor	@t[7], @t[7], @t[6]		@ restore t[7]
+	veor	@y[7], @y[7], @t[4]
+	vext.8	@t[6], @t[6], @t[6], #12
+	veor	@y[2], @y[2], @t[0]
+	veor	@y[7], @y[7], @t[5]
+	vext.8	@t[7], @t[7], @t[7], #12
+	veor	@y[2], @y[2], @t[2]
+
+	veor	@y[3], @y[3], @y[1]
+	veor	@y[1], @y[1], @t[1]
+	veor	@y[0], @y[0], @t[0]
+	veor	@y[3], @y[3], @t[0]
+	veor	@y[1], @y[1], @t[5]
+	veor	@y[0], @y[0], @t[5]
+	vext.8	@t[0], @t[0], @t[0], #12
+	veor	@y[1], @y[1], @t[7]
+	veor	@y[0], @y[0], @t[6]
+	veor	@y[3], @y[3], @y[1]
+	veor	@y[4], @y[4], @t[1]
+	vext.8	@t[1], @t[1], @t[1], #12
+
+	veor	@y[7], @y[7], @t[7]
+	veor	@y[4], @y[4], @t[2]
+	veor	@y[5], @y[5], @t[2]
+	veor	@y[2], @y[2], @t[6]
+	veor	@t[6], @t[6], @t[3]		@ clobber t[6]
+	vext.8	@t[2], @t[2], @t[2], #12
+	veor	@y[4], @y[4], @y[7]
+	veor	@y[3], @y[3], @t[6]
+
+	veor	@y[6], @y[6], @t[6]
+	veor	@y[5], @y[5], @t[5]
+	vext.8	@t[5], @t[5], @t[5], #12
+	veor	@y[6], @y[6], @t[4]
+	vext.8	@t[4], @t[4], @t[4], #12
+	veor	@y[5], @y[5], @t[6]
+	veor	@y[6], @y[6], @t[7]
+	vext.8	@t[7], @t[7], @t[7], #12
+	veor	@t[6], @t[6], @t[3]		@ restore t[6]
+	vext.8	@t[3], @t[3], @t[3], #12
+
+	@ multiplication by 0x09
+	veor	@y[4], @y[4], @y[1]
+	veor	@t[1], @t[1], @y[1]		@ t[1]=y[1]
+	veor	@t[0], @t[0], @t[5]		@ clobber t[0]
+	vext.8	@t[6], @t[6], @t[6], #12
+	veor	@t[1], @t[1], @t[5]
+	veor	@y[3], @y[3], @t[0]
+	veor	@t[0], @t[0], @y[0]		@ t[0]=y[0]
+	veor	@t[1], @t[1], @t[6]
+	veor	@t[6], @t[6], @t[7]		@ clobber t[6]
+	veor	@y[4], @y[4], @t[1]
+	veor	@y[7], @y[7], @t[4]
+	veor	@y[6], @y[6], @t[3]
+	veor	@y[5], @y[5], @t[2]
+	veor	@t[4], @t[4], @y[4]		@ t[4]=y[4]
+	veor	@t[3], @t[3], @y[3]		@ t[3]=y[3]
+	veor	@t[5], @t[5], @y[5]		@ t[5]=y[5]
+	veor	@t[2], @t[2], @y[2]		@ t[2]=y[2]
+	veor	@t[3], @t[3], @t[7]
+	veor	@XMM[5], @t[5], @t[6]
+	veor	@XMM[6], @t[6], @y[6]		@ t[6]=y[6]
+	veor	@XMM[2], @t[2], @t[6]
+	veor	@XMM[7], @t[7], @y[7]		@ t[7]=y[7]
+
+	vmov	@XMM[0], @t[0]
+	vmov	@XMM[1], @t[1]
+	@ vmov	@XMM[2], @t[2]
+	vmov	@XMM[3], @t[3]
+	vmov	@XMM[4], @t[4]
+	@ vmov	@XMM[5], @t[5]
+	@ vmov	@XMM[6], @t[6]
+	@ vmov	@XMM[7], @t[7]
+___
+}
+
+sub InvMixColumns {
+my @x=@_[0..7];
+my @t=@_[8..15];
+
+# Thanks to Jussi Kivilinna for providing pointer to
+#
+# | 0e 0b 0d 09 |   | 02 03 01 01 |   | 05 00 04 00 |
+# | 09 0e 0b 0d | = | 01 02 03 01 | x | 00 05 00 04 |
+# | 0d 09 0e 0b |   | 01 01 02 03 |   | 04 00 05 00 |
+# | 0b 0d 09 0e |   | 03 01 01 02 |   | 00 04 00 05 |
+
+$code.=<<___;
+	@ multiplication by 0x05-0x00-0x04-0x00
+	vext.8	@t[0], @x[0], @x[0], #8
+	vext.8	@t[6], @x[6], @x[6], #8
+	vext.8	@t[7], @x[7], @x[7], #8
+	veor	@t[0], @t[0], @x[0]
+	vext.8	@t[1], @x[1], @x[1], #8
+	veor	@t[6], @t[6], @x[6]
+	vext.8	@t[2], @x[2], @x[2], #8
+	veor	@t[7], @t[7], @x[7]
+	vext.8	@t[3], @x[3], @x[3], #8
+	veor	@t[1], @t[1], @x[1]
+	vext.8	@t[4], @x[4], @x[4], #8
+	veor	@t[2], @t[2], @x[2]
+	vext.8	@t[5], @x[5], @x[5], #8
+	veor	@t[3], @t[3], @x[3]
+	veor	@t[4], @t[4], @x[4]
+	veor	@t[5], @t[5], @x[5]
+
+	 veor	@x[0], @x[0], @t[6]
+	 veor	@x[1], @x[1], @t[6]
+	 veor	@x[2], @x[2], @t[0]
+	 veor	@x[4], @x[4], @t[2]
+	 veor	@x[3], @x[3], @t[1]
+	 veor	@x[1], @x[1], @t[7]
+	 veor	@x[2], @x[2], @t[7]
+	 veor	@x[4], @x[4], @t[6]
+	 veor	@x[5], @x[5], @t[3]
+	 veor	@x[3], @x[3], @t[6]
+	 veor	@x[6], @x[6], @t[4]
+	 veor	@x[4], @x[4], @t[7]
+	 veor	@x[5], @x[5], @t[7]
+	 veor	@x[7], @x[7], @t[5]
+___
+	&MixColumns	(@x,@t,1);	# flipped 2<->3 and 4<->6
+}
+
+sub swapmove {
+my ($a,$b,$n,$mask,$t)=@_;
+$code.=<<___;
+	vshr.u64	$t, $b, #$n
+	veor		$t, $t, $a
+	vand		$t, $t, $mask
+	veor		$a, $a, $t
+	vshl.u64	$t, $t, #$n
+	veor		$b, $b, $t
+___
+}
+sub swapmove2x {
+my ($a0,$b0,$a1,$b1,$n,$mask,$t0,$t1)=@_;
+$code.=<<___;
+	vshr.u64	$t0, $b0, #$n
+	 vshr.u64	$t1, $b1, #$n
+	veor		$t0, $t0, $a0
+	 veor		$t1, $t1, $a1
+	vand		$t0, $t0, $mask
+	 vand		$t1, $t1, $mask
+	veor		$a0, $a0, $t0
+	vshl.u64	$t0, $t0, #$n
+	 veor		$a1, $a1, $t1
+	 vshl.u64	$t1, $t1, #$n
+	veor		$b0, $b0, $t0
+	 veor		$b1, $b1, $t1
+___
+}
+
+sub bitslice {
+my @x=reverse(@_[0..7]);
+my ($t0,$t1,$t2,$t3)=@_[8..11];
+$code.=<<___;
+	vmov.i8	$t0,#0x55			@ compose .LBS0
+	vmov.i8	$t1,#0x33			@ compose .LBS1
+___
+	&swapmove2x(@x[0,1,2,3],1,$t0,$t2,$t3);
+	&swapmove2x(@x[4,5,6,7],1,$t0,$t2,$t3);
+$code.=<<___;
+	vmov.i8	$t0,#0x0f			@ compose .LBS2
+___
+	&swapmove2x(@x[0,2,1,3],2,$t1,$t2,$t3);
+	&swapmove2x(@x[4,6,5,7],2,$t1,$t2,$t3);
+
+	&swapmove2x(@x[0,4,1,5],4,$t0,$t2,$t3);
+	&swapmove2x(@x[2,6,3,7],4,$t0,$t2,$t3);
+}
+
+$code.=<<___;
+#if defined(__arm__)
+#ifndef __KERNEL__
+# include "arm_arch.h"
+
+# define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
+# define VFP_ABI_POP	vldmia	sp!,{d8-d15}
+# define VFP_ABI_FRAME	0x40
+#else
+# define VFP_ABI_PUSH
+# define VFP_ABI_POP
+# define VFP_ABI_FRAME	0
+# define BSAES_ASM_EXTENDED_KEY
+# define XTS_CHAIN_TWEAK
+# define __ARM_ARCH__ __LINUX_ARM_ARCH__
+#endif
+
+#ifdef __thumb__
+# define adrl adr
+#endif
+
+#if __ARM_ARCH__>=7
+.text
+.syntax	unified 	@ ARMv7-capable assembler is expected to handle this
+#ifdef __thumb2__
+.thumb
+#else
+.code   32
+#endif
+
+.fpu	neon
+
+.type	_bsaes_decrypt8,%function
+.align	4
+_bsaes_decrypt8:
+	adr	$const,_bsaes_decrypt8
+	vldmia	$key!, {@XMM[9]}		@ round 0 key
+	add	$const,$const,#.LM0ISR-_bsaes_decrypt8
+
+	vldmia	$const!, {@XMM[8]}		@ .LM0ISR
+	veor	@XMM[10], @XMM[0], @XMM[9]	@ xor with round0 key
+	veor	@XMM[11], @XMM[1], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[0])`, {@XMM[10]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[0])`, {@XMM[10]}, `&Dhi(@XMM[8])`
+	veor	@XMM[12], @XMM[2], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[1])`, {@XMM[11]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[1])`, {@XMM[11]}, `&Dhi(@XMM[8])`
+	veor	@XMM[13], @XMM[3], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[2])`, {@XMM[12]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[2])`, {@XMM[12]}, `&Dhi(@XMM[8])`
+	veor	@XMM[14], @XMM[4], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[3])`, {@XMM[13]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[3])`, {@XMM[13]}, `&Dhi(@XMM[8])`
+	veor	@XMM[15], @XMM[5], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[4])`, {@XMM[14]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[4])`, {@XMM[14]}, `&Dhi(@XMM[8])`
+	veor	@XMM[10], @XMM[6], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[5])`, {@XMM[15]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[5])`, {@XMM[15]}, `&Dhi(@XMM[8])`
+	veor	@XMM[11], @XMM[7], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[6])`, {@XMM[10]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[6])`, {@XMM[10]}, `&Dhi(@XMM[8])`
+	 vtbl.8	`&Dlo(@XMM[7])`, {@XMM[11]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[7])`, {@XMM[11]}, `&Dhi(@XMM[8])`
+___
+	&bitslice	(@XMM[0..7, 8..11]);
+$code.=<<___;
+	sub	$rounds,$rounds,#1
+	b	.Ldec_sbox
+.align	4
+.Ldec_loop:
+___
+	&ShiftRows	(@XMM[0..7, 8..12]);
+$code.=".Ldec_sbox:\n";
+	&InvSbox	(@XMM[0..7, 8..15]);
+$code.=<<___;
+	subs	$rounds,$rounds,#1
+	bcc	.Ldec_done
+___
+	&InvMixColumns	(@XMM[0,1,6,4,2,7,3,5, 8..15]);
+$code.=<<___;
+	vldmia	$const, {@XMM[12]}		@ .LISR
+	ite	eq				@ Thumb2 thing, sanity check in ARM
+	addeq	$const,$const,#0x10
+	bne	.Ldec_loop
+	vldmia	$const, {@XMM[12]}		@ .LISRM0
+	b	.Ldec_loop
+.align	4
+.Ldec_done:
+___
+	&bitslice	(@XMM[0,1,6,4,2,7,3,5, 8..11]);
+$code.=<<___;
+	vldmia	$key, {@XMM[8]}			@ last round key
+	veor	@XMM[6], @XMM[6], @XMM[8]
+	veor	@XMM[4], @XMM[4], @XMM[8]
+	veor	@XMM[2], @XMM[2], @XMM[8]
+	veor	@XMM[7], @XMM[7], @XMM[8]
+	veor	@XMM[3], @XMM[3], @XMM[8]
+	veor	@XMM[5], @XMM[5], @XMM[8]
+	veor	@XMM[0], @XMM[0], @XMM[8]
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	bx	lr
+.size	_bsaes_decrypt8,.-_bsaes_decrypt8
+
+.type	_bsaes_const,%object
+.align	6
+_bsaes_const:
+.LM0ISR:	@ InvShiftRows constants
+	.quad	0x0a0e0206070b0f03, 0x0004080c0d010509
+.LISR:
+	.quad	0x0504070602010003, 0x0f0e0d0c080b0a09
+.LISRM0:
+	.quad	0x01040b0e0205080f, 0x0306090c00070a0d
+.LM0SR:		@ ShiftRows constants
+	.quad	0x0a0e02060f03070b, 0x0004080c05090d01
+.LSR:
+	.quad	0x0504070600030201, 0x0f0e0d0c0a09080b
+.LSRM0:
+	.quad	0x0304090e00050a0f, 0x01060b0c0207080d
+.LM0:
+	.quad	0x02060a0e03070b0f, 0x0004080c0105090d
+.LREVM0SR:
+	.quad	0x090d01050c000408, 0x03070b0f060a0e02
+.asciz	"Bit-sliced AES for NEON, CRYPTOGAMS by <appro\@openssl.org>"
+.align	6
+.size	_bsaes_const,.-_bsaes_const
+
+.type	_bsaes_encrypt8,%function
+.align	4
+_bsaes_encrypt8:
+	adr	$const,_bsaes_encrypt8
+	vldmia	$key!, {@XMM[9]}		@ round 0 key
+	sub	$const,$const,#_bsaes_encrypt8-.LM0SR
+
+	vldmia	$const!, {@XMM[8]}		@ .LM0SR
+_bsaes_encrypt8_alt:
+	veor	@XMM[10], @XMM[0], @XMM[9]	@ xor with round0 key
+	veor	@XMM[11], @XMM[1], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[0])`, {@XMM[10]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[0])`, {@XMM[10]}, `&Dhi(@XMM[8])`
+	veor	@XMM[12], @XMM[2], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[1])`, {@XMM[11]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[1])`, {@XMM[11]}, `&Dhi(@XMM[8])`
+	veor	@XMM[13], @XMM[3], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[2])`, {@XMM[12]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[2])`, {@XMM[12]}, `&Dhi(@XMM[8])`
+	veor	@XMM[14], @XMM[4], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[3])`, {@XMM[13]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[3])`, {@XMM[13]}, `&Dhi(@XMM[8])`
+	veor	@XMM[15], @XMM[5], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[4])`, {@XMM[14]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[4])`, {@XMM[14]}, `&Dhi(@XMM[8])`
+	veor	@XMM[10], @XMM[6], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[5])`, {@XMM[15]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[5])`, {@XMM[15]}, `&Dhi(@XMM[8])`
+	veor	@XMM[11], @XMM[7], @XMM[9]
+	 vtbl.8	`&Dlo(@XMM[6])`, {@XMM[10]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[6])`, {@XMM[10]}, `&Dhi(@XMM[8])`
+	 vtbl.8	`&Dlo(@XMM[7])`, {@XMM[11]}, `&Dlo(@XMM[8])`
+	 vtbl.8	`&Dhi(@XMM[7])`, {@XMM[11]}, `&Dhi(@XMM[8])`
+_bsaes_encrypt8_bitslice:
+___
+	&bitslice	(@XMM[0..7, 8..11]);
+$code.=<<___;
+	sub	$rounds,$rounds,#1
+	b	.Lenc_sbox
+.align	4
+.Lenc_loop:
+___
+	&ShiftRows	(@XMM[0..7, 8..12]);
+$code.=".Lenc_sbox:\n";
+	&Sbox		(@XMM[0..7, 8..15]);
+$code.=<<___;
+	subs	$rounds,$rounds,#1
+	bcc	.Lenc_done
+___
+	&MixColumns	(@XMM[0,1,4,6,3,7,2,5, 8..15]);
+$code.=<<___;
+	vldmia	$const, {@XMM[12]}		@ .LSR
+	ite	eq				@ Thumb2 thing, samity check in ARM
+	addeq	$const,$const,#0x10
+	bne	.Lenc_loop
+	vldmia	$const, {@XMM[12]}		@ .LSRM0
+	b	.Lenc_loop
+.align	4
+.Lenc_done:
+___
+	# output in lsb > [t0, t1, t4, t6, t3, t7, t2, t5] < msb
+	&bitslice	(@XMM[0,1,4,6,3,7,2,5, 8..11]);
+$code.=<<___;
+	vldmia	$key, {@XMM[8]}			@ last round key
+	veor	@XMM[4], @XMM[4], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[8]
+	veor	@XMM[3], @XMM[3], @XMM[8]
+	veor	@XMM[7], @XMM[7], @XMM[8]
+	veor	@XMM[2], @XMM[2], @XMM[8]
+	veor	@XMM[5], @XMM[5], @XMM[8]
+	veor	@XMM[0], @XMM[0], @XMM[8]
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	bx	lr
+.size	_bsaes_encrypt8,.-_bsaes_encrypt8
+___
+}
+{
+my ($out,$inp,$rounds,$const)=("r12","r4","r5","r6");
+
+sub bitslice_key {
+my @x=reverse(@_[0..7]);
+my ($bs0,$bs1,$bs2,$t2,$t3)=@_[8..12];
+
+	&swapmove	(@x[0,1],1,$bs0,$t2,$t3);
+$code.=<<___;
+	@ &swapmove(@x[2,3],1,$t0,$t2,$t3);
+	vmov	@x[2], @x[0]
+	vmov	@x[3], @x[1]
+___
+	#&swapmove2x(@x[4,5,6,7],1,$t0,$t2,$t3);
+
+	&swapmove2x	(@x[0,2,1,3],2,$bs1,$t2,$t3);
+$code.=<<___;
+	@ &swapmove2x(@x[4,6,5,7],2,$t1,$t2,$t3);
+	vmov	@x[4], @x[0]
+	vmov	@x[6], @x[2]
+	vmov	@x[5], @x[1]
+	vmov	@x[7], @x[3]
+___
+	&swapmove2x	(@x[0,4,1,5],4,$bs2,$t2,$t3);
+	&swapmove2x	(@x[2,6,3,7],4,$bs2,$t2,$t3);
+}
+
+$code.=<<___;
+.type	_bsaes_key_convert,%function
+.align	4
+_bsaes_key_convert:
+	adr	$const,_bsaes_key_convert
+	vld1.8	{@XMM[7]},  [$inp]!		@ load round 0 key
+	sub	$const,$const,#_bsaes_key_convert-.LM0
+	vld1.8	{@XMM[15]}, [$inp]!		@ load round 1 key
+
+	vmov.i8	@XMM[8],  #0x01			@ bit masks
+	vmov.i8	@XMM[9],  #0x02
+	vmov.i8	@XMM[10], #0x04
+	vmov.i8	@XMM[11], #0x08
+	vmov.i8	@XMM[12], #0x10
+	vmov.i8	@XMM[13], #0x20
+	vldmia	$const, {@XMM[14]}		@ .LM0
+
+#ifdef __ARMEL__
+	vrev32.8	@XMM[7],  @XMM[7]
+	vrev32.8	@XMM[15], @XMM[15]
+#endif
+	sub	$rounds,$rounds,#1
+	vstmia	$out!, {@XMM[7]}		@ save round 0 key
+	b	.Lkey_loop
+
+.align	4
+.Lkey_loop:
+	vtbl.8	`&Dlo(@XMM[7])`,{@XMM[15]},`&Dlo(@XMM[14])`
+	vtbl.8	`&Dhi(@XMM[7])`,{@XMM[15]},`&Dhi(@XMM[14])`
+	vmov.i8	@XMM[6],  #0x40
+	vmov.i8	@XMM[15], #0x80
+
+	vtst.8	@XMM[0], @XMM[7], @XMM[8]
+	vtst.8	@XMM[1], @XMM[7], @XMM[9]
+	vtst.8	@XMM[2], @XMM[7], @XMM[10]
+	vtst.8	@XMM[3], @XMM[7], @XMM[11]
+	vtst.8	@XMM[4], @XMM[7], @XMM[12]
+	vtst.8	@XMM[5], @XMM[7], @XMM[13]
+	vtst.8	@XMM[6], @XMM[7], @XMM[6]
+	vtst.8	@XMM[7], @XMM[7], @XMM[15]
+	vld1.8	{@XMM[15]}, [$inp]!		@ load next round key
+	vmvn	@XMM[0], @XMM[0]		@ "pnot"
+	vmvn	@XMM[1], @XMM[1]
+	vmvn	@XMM[5], @XMM[5]
+	vmvn	@XMM[6], @XMM[6]
+#ifdef __ARMEL__
+	vrev32.8	@XMM[15], @XMM[15]
+#endif
+	subs	$rounds,$rounds,#1
+	vstmia	$out!,{@XMM[0]-@XMM[7]}		@ write bit-sliced round key
+	bne	.Lkey_loop
+
+	vmov.i8	@XMM[7],#0x63			@ compose .L63
+	@ don't save last round key
+	bx	lr
+.size	_bsaes_key_convert,.-_bsaes_key_convert
+___
+}
+
+if (0) {		# following four functions are unsupported interface
+			# used for benchmarking...
+$code.=<<___;
+.globl	bsaes_enc_key_convert
+.hidden	bsaes_enc_key_convert
+.type	bsaes_enc_key_convert,%function
+.align	4
+bsaes_enc_key_convert:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+
+	ldr	r5,[$inp,#240]			@ pass rounds
+	mov	r4,$inp				@ pass key
+	mov	r12,$out			@ pass key schedule
+	bl	_bsaes_key_convert
+	veor	@XMM[7],@XMM[7],@XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}			@ save last round key
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_enc_key_convert,.-bsaes_enc_key_convert
+
+.globl	bsaes_encrypt_128
+.hidden	bsaes_encrypt_128
+.type	bsaes_encrypt_128,%function
+.align	4
+bsaes_encrypt_128:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+.Lenc128_loop:
+	vld1.8	{@XMM[0]-@XMM[1]}, [$inp]!	@ load input
+	vld1.8	{@XMM[2]-@XMM[3]}, [$inp]!
+	mov	r4,$key				@ pass the key
+	vld1.8	{@XMM[4]-@XMM[5]}, [$inp]!
+	mov	r5,#10				@ pass rounds
+	vld1.8	{@XMM[6]-@XMM[7]}, [$inp]!
+
+	bl	_bsaes_encrypt8
+
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[3]}, [$out]!
+	vst1.8	{@XMM[7]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	subs	$len,$len,#0x80
+	vst1.8	{@XMM[5]}, [$out]!
+	bhi	.Lenc128_loop
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_encrypt_128,.-bsaes_encrypt_128
+
+.globl	bsaes_dec_key_convert
+.hidden	bsaes_dec_key_convert
+.type	bsaes_dec_key_convert,%function
+.align	4
+bsaes_dec_key_convert:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+
+	ldr	r5,[$inp,#240]			@ pass rounds
+	mov	r4,$inp				@ pass key
+	mov	r12,$out			@ pass key schedule
+	bl	_bsaes_key_convert
+	vldmia	$out, {@XMM[6]}
+	vstmia	r12,  {@XMM[15]}		@ save last round key
+	veor	@XMM[7], @XMM[7], @XMM[6]	@ fix up round 0 key
+	vstmia	$out, {@XMM[7]}
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_dec_key_convert,.-bsaes_dec_key_convert
+
+.globl	bsaes_decrypt_128
+.hidden	bsaes_decrypt_128
+.type	bsaes_decrypt_128,%function
+.align	4
+bsaes_decrypt_128:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+.Ldec128_loop:
+	vld1.8	{@XMM[0]-@XMM[1]}, [$inp]!	@ load input
+	vld1.8	{@XMM[2]-@XMM[3]}, [$inp]!
+	mov	r4,$key				@ pass the key
+	vld1.8	{@XMM[4]-@XMM[5]}, [$inp]!
+	mov	r5,#10				@ pass rounds
+	vld1.8	{@XMM[6]-@XMM[7]}, [$inp]!
+
+	bl	_bsaes_decrypt8
+
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	vst1.8	{@XMM[7]}, [$out]!
+	vst1.8	{@XMM[3]}, [$out]!
+	subs	$len,$len,#0x80
+	vst1.8	{@XMM[5]}, [$out]!
+	bhi	.Ldec128_loop
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_decrypt_128,.-bsaes_decrypt_128
+___
+}
+{
+my ($inp,$out,$len,$key, $ivp,$fp,$rounds)=map("r$_",(0..3,8..10));
+my ($keysched)=("sp");
+
+$code.=<<___;
+.extern AES_cbc_encrypt
+.extern AES_decrypt
+
+.global	bsaes_cbc_encrypt
+.hidden	bsaes_cbc_encrypt
+.type	bsaes_cbc_encrypt,%function
+.align	5
+bsaes_cbc_encrypt:
+#ifndef	__KERNEL__
+	cmp	$len, #128
+#ifndef	__thumb__
+	blo	AES_cbc_encrypt
+#else
+	bhs	1f
+	b	AES_cbc_encrypt
+1:
+#endif
+#endif
+
+	@ it is up to the caller to make sure we are called with enc == 0
+
+	mov	ip, sp
+	stmdb	sp!, {r4-r10, lr}
+	VFP_ABI_PUSH
+	ldr	$ivp, [ip]			@ IV is 1st arg on the stack
+	mov	$len, $len, lsr#4		@ len in 16 byte blocks
+	sub	sp, #0x10			@ scratch space to carry over the IV
+	mov	$fp, sp				@ save sp
+
+	ldr	$rounds, [$key, #240]		@ get # of rounds
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	@ allocate the key schedule on the stack
+	sub	r12, sp, $rounds, lsl#7		@ 128 bytes per inner round key
+	add	r12, #`128-32`			@ sifze of bit-slices key schedule
+
+	@ populate the key schedule
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	mov	sp, r12				@ sp is $keysched
+	bl	_bsaes_key_convert
+	vldmia	$keysched, {@XMM[6]}
+	vstmia	r12,  {@XMM[15]}		@ save last round key
+	veor	@XMM[7], @XMM[7], @XMM[6]	@ fix up round 0 key
+	vstmia	$keysched, {@XMM[7]}
+#else
+	ldr	r12, [$key, #244]
+	eors	r12, #1
+	beq	0f
+
+	@ populate the key schedule
+	str	r12, [$key, #244]
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	add	r12, $key, #248			@ pass key schedule
+	bl	_bsaes_key_convert
+	add	r4, $key, #248
+	vldmia	r4, {@XMM[6]}
+	vstmia	r12, {@XMM[15]}			@ save last round key
+	veor	@XMM[7], @XMM[7], @XMM[6]	@ fix up round 0 key
+	vstmia	r4, {@XMM[7]}
+
+.align	2
+0:
+#endif
+
+	vld1.8	{@XMM[15]}, [$ivp]		@ load IV
+	b	.Lcbc_dec_loop
+
+.align	4
+.Lcbc_dec_loop:
+	subs	$len, $len, #0x8
+	bmi	.Lcbc_dec_loop_finish
+
+	vld1.8	{@XMM[0]-@XMM[1]}, [$inp]!	@ load input
+	vld1.8	{@XMM[2]-@XMM[3]}, [$inp]!
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	mov	r4, $keysched			@ pass the key
+#else
+	add	r4, $key, #248
+#endif
+	vld1.8	{@XMM[4]-@XMM[5]}, [$inp]!
+	mov	r5, $rounds
+	vld1.8	{@XMM[6]-@XMM[7]}, [$inp]
+	sub	$inp, $inp, #0x60
+	vstmia	$fp, {@XMM[15]}			@ put aside IV
+
+	bl	_bsaes_decrypt8
+
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[10]-@XMM[11]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vld1.8	{@XMM[12]-@XMM[13]}, [$inp]!
+	veor	@XMM[4], @XMM[4], @XMM[10]
+	veor	@XMM[2], @XMM[2], @XMM[11]
+	vld1.8	{@XMM[14]-@XMM[15]}, [$inp]!
+	veor	@XMM[7], @XMM[7], @XMM[12]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	veor	@XMM[3], @XMM[3], @XMM[13]
+	vst1.8	{@XMM[6]}, [$out]!
+	veor	@XMM[5], @XMM[5], @XMM[14]
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	vst1.8	{@XMM[7]}, [$out]!
+	vst1.8	{@XMM[3]}, [$out]!
+	vst1.8	{@XMM[5]}, [$out]!
+
+	b	.Lcbc_dec_loop
+
+.Lcbc_dec_loop_finish:
+	adds	$len, $len, #8
+	beq	.Lcbc_dec_done
+
+	vld1.8	{@XMM[0]}, [$inp]!		@ load input
+	cmp	$len, #2
+	blo	.Lcbc_dec_one
+	vld1.8	{@XMM[1]}, [$inp]!
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	mov	r4, $keysched			@ pass the key
+#else
+	add	r4, $key, #248
+#endif
+	mov	r5, $rounds
+	vstmia	$fp, {@XMM[15]}			@ put aside IV
+	beq	.Lcbc_dec_two
+	vld1.8	{@XMM[2]}, [$inp]!
+	cmp	$len, #4
+	blo	.Lcbc_dec_three
+	vld1.8	{@XMM[3]}, [$inp]!
+	beq	.Lcbc_dec_four
+	vld1.8	{@XMM[4]}, [$inp]!
+	cmp	$len, #6
+	blo	.Lcbc_dec_five
+	vld1.8	{@XMM[5]}, [$inp]!
+	beq	.Lcbc_dec_six
+	vld1.8	{@XMM[6]}, [$inp]!
+	sub	$inp, $inp, #0x70
+
+	bl	_bsaes_decrypt8
+
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[10]-@XMM[11]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vld1.8	{@XMM[12]-@XMM[13]}, [$inp]!
+	veor	@XMM[4], @XMM[4], @XMM[10]
+	veor	@XMM[2], @XMM[2], @XMM[11]
+	vld1.8	{@XMM[15]}, [$inp]!
+	veor	@XMM[7], @XMM[7], @XMM[12]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	veor	@XMM[3], @XMM[3], @XMM[13]
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	vst1.8	{@XMM[7]}, [$out]!
+	vst1.8	{@XMM[3]}, [$out]!
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_six:
+	sub	$inp, $inp, #0x60
+	bl	_bsaes_decrypt8
+	vldmia	$fp,{@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[10]-@XMM[11]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vld1.8	{@XMM[12]}, [$inp]!
+	veor	@XMM[4], @XMM[4], @XMM[10]
+	veor	@XMM[2], @XMM[2], @XMM[11]
+	vld1.8	{@XMM[15]}, [$inp]!
+	veor	@XMM[7], @XMM[7], @XMM[12]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	vst1.8	{@XMM[7]}, [$out]!
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_five:
+	sub	$inp, $inp, #0x50
+	bl	_bsaes_decrypt8
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[10]-@XMM[11]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vld1.8	{@XMM[15]}, [$inp]!
+	veor	@XMM[4], @XMM[4], @XMM[10]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	veor	@XMM[2], @XMM[2], @XMM[11]
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[4]}, [$out]!
+	vst1.8	{@XMM[2]}, [$out]!
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_four:
+	sub	$inp, $inp, #0x40
+	bl	_bsaes_decrypt8
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[10]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vld1.8	{@XMM[15]}, [$inp]!
+	veor	@XMM[4], @XMM[4], @XMM[10]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	vst1.8	{@XMM[6]}, [$out]!
+	vst1.8	{@XMM[4]}, [$out]!
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_three:
+	sub	$inp, $inp, #0x30
+	bl	_bsaes_decrypt8
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]-@XMM[9]}, [$inp]!	@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[15]}, [$inp]!
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	veor	@XMM[6], @XMM[6], @XMM[9]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	vst1.8	{@XMM[6]}, [$out]!
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_two:
+	sub	$inp, $inp, #0x20
+	bl	_bsaes_decrypt8
+	vldmia	$fp, {@XMM[14]}			@ reload IV
+	vld1.8	{@XMM[8]}, [$inp]!		@ reload input
+	veor	@XMM[0], @XMM[0], @XMM[14]	@ ^= IV
+	vld1.8	{@XMM[15]}, [$inp]!		@ reload input
+	veor	@XMM[1], @XMM[1], @XMM[8]
+	vst1.8	{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	b	.Lcbc_dec_done
+.align	4
+.Lcbc_dec_one:
+	sub	$inp, $inp, #0x10
+	mov	$rounds, $out			@ save original out pointer
+	mov	$out, $fp			@ use the iv scratch space as out buffer
+	mov	r2, $key
+	vmov	@XMM[4],@XMM[15]		@ just in case ensure that IV
+	vmov	@XMM[5],@XMM[0]			@ and input are preserved
+	bl	AES_decrypt
+	vld1.8	{@XMM[0]}, [$fp,:64]		@ load result
+	veor	@XMM[0], @XMM[0], @XMM[4]	@ ^= IV
+	vmov	@XMM[15], @XMM[5]		@ @XMM[5] holds input
+	vst1.8	{@XMM[0]}, [$rounds]		@ write output
+
+.Lcbc_dec_done:
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	vmov.i32	q0, #0
+	vmov.i32	q1, #0
+.Lcbc_dec_bzero:				@ wipe key schedule [if any]
+	vstmia		$keysched!, {q0-q1}
+	cmp		$keysched, $fp
+	bne		.Lcbc_dec_bzero
+#endif
+
+	mov	sp, $fp
+	add	sp, #0x10			@ add sp,$fp,#0x10 is no good for thumb
+	vst1.8	{@XMM[15]}, [$ivp]		@ return IV
+	VFP_ABI_POP
+	ldmia	sp!, {r4-r10, pc}
+.size	bsaes_cbc_encrypt,.-bsaes_cbc_encrypt
+___
+}
+{
+my ($inp,$out,$len,$key, $ctr,$fp,$rounds)=(map("r$_",(0..3,8..10)));
+my $const = "r6";	# shared with _bsaes_encrypt8_alt
+my $keysched = "sp";
+
+$code.=<<___;
+.extern	AES_encrypt
+.global	bsaes_ctr32_encrypt_blocks
+.hidden	bsaes_ctr32_encrypt_blocks
+.type	bsaes_ctr32_encrypt_blocks,%function
+.align	5
+bsaes_ctr32_encrypt_blocks:
+	cmp	$len, #8			@ use plain AES for
+	blo	.Lctr_enc_short			@ small sizes
+
+	mov	ip, sp
+	stmdb	sp!, {r4-r10, lr}
+	VFP_ABI_PUSH
+	ldr	$ctr, [ip]			@ ctr is 1st arg on the stack
+	sub	sp, sp, #0x10			@ scratch space to carry over the ctr
+	mov	$fp, sp				@ save sp
+
+	ldr	$rounds, [$key, #240]		@ get # of rounds
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	@ allocate the key schedule on the stack
+	sub	r12, sp, $rounds, lsl#7		@ 128 bytes per inner round key
+	add	r12, #`128-32`			@ size of bit-sliced key schedule
+
+	@ populate the key schedule
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	mov	sp, r12				@ sp is $keysched
+	bl	_bsaes_key_convert
+	veor	@XMM[7],@XMM[7],@XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}			@ save last round key
+
+	vld1.8	{@XMM[0]}, [$ctr]		@ load counter
+	add	$ctr, $const, #.LREVM0SR-.LM0	@ borrow $ctr
+	vldmia	$keysched, {@XMM[4]}		@ load round0 key
+#else
+	ldr	r12, [$key, #244]
+	eors	r12, #1
+	beq	0f
+
+	@ populate the key schedule
+	str	r12, [$key, #244]
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	add	r12, $key, #248			@ pass key schedule
+	bl	_bsaes_key_convert
+	veor	@XMM[7],@XMM[7],@XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}			@ save last round key
+
+.align	2
+0:	add	r12, $key, #248
+	vld1.8	{@XMM[0]}, [$ctr]		@ load counter
+	adrl	$ctr, .LREVM0SR			@ borrow $ctr
+	vldmia	r12, {@XMM[4]}			@ load round0 key
+	sub	sp, #0x10			@ place for adjusted round0 key
+#endif
+
+	vmov.i32	@XMM[8],#1		@ compose 1<<96
+	veor		@XMM[9],@XMM[9],@XMM[9]
+	vrev32.8	@XMM[0],@XMM[0]
+	vext.8		@XMM[8],@XMM[9],@XMM[8],#4
+	vrev32.8	@XMM[4],@XMM[4]
+	vadd.u32	@XMM[9],@XMM[8],@XMM[8]	@ compose 2<<96
+	vstmia	$keysched, {@XMM[4]}		@ save adjusted round0 key
+	b	.Lctr_enc_loop
+
+.align	4
+.Lctr_enc_loop:
+	vadd.u32	@XMM[10], @XMM[8], @XMM[9]	@ compose 3<<96
+	vadd.u32	@XMM[1], @XMM[0], @XMM[8]	@ +1
+	vadd.u32	@XMM[2], @XMM[0], @XMM[9]	@ +2
+	vadd.u32	@XMM[3], @XMM[0], @XMM[10]	@ +3
+	vadd.u32	@XMM[4], @XMM[1], @XMM[10]
+	vadd.u32	@XMM[5], @XMM[2], @XMM[10]
+	vadd.u32	@XMM[6], @XMM[3], @XMM[10]
+	vadd.u32	@XMM[7], @XMM[4], @XMM[10]
+	vadd.u32	@XMM[10], @XMM[5], @XMM[10]	@ next counter
+
+	@ Borrow prologue from _bsaes_encrypt8 to use the opportunity
+	@ to flip byte order in 32-bit counter
+
+	vldmia		$keysched, {@XMM[9]}		@ load round0 key
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, $keysched, #0x10		@ pass next round key
+#else
+	add		r4, $key, #`248+16`
+#endif
+	vldmia		$ctr, {@XMM[8]}			@ .LREVM0SR
+	mov		r5, $rounds			@ pass rounds
+	vstmia		$fp, {@XMM[10]}			@ save next counter
+	sub		$const, $ctr, #.LREVM0SR-.LSR	@ pass constants
+
+	bl		_bsaes_encrypt8_alt
+
+	subs		$len, $len, #8
+	blo		.Lctr_enc_loop_done
+
+	vld1.8		{@XMM[8]-@XMM[9]}, [$inp]!	@ load input
+	vld1.8		{@XMM[10]-@XMM[11]}, [$inp]!
+	veor		@XMM[0], @XMM[8]
+	veor		@XMM[1], @XMM[9]
+	vld1.8		{@XMM[12]-@XMM[13]}, [$inp]!
+	veor		@XMM[4], @XMM[10]
+	veor		@XMM[6], @XMM[11]
+	vld1.8		{@XMM[14]-@XMM[15]}, [$inp]!
+	veor		@XMM[3], @XMM[12]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!	@ write output
+	veor		@XMM[7], @XMM[13]
+	veor		@XMM[2], @XMM[14]
+	vst1.8		{@XMM[4]}, [$out]!
+	veor		@XMM[5], @XMM[15]
+	vst1.8		{@XMM[6]}, [$out]!
+	vmov.i32	@XMM[8], #1			@ compose 1<<96
+	vst1.8		{@XMM[3]}, [$out]!
+	veor		@XMM[9], @XMM[9], @XMM[9]
+	vst1.8		{@XMM[7]}, [$out]!
+	vext.8		@XMM[8], @XMM[9], @XMM[8], #4
+	vst1.8		{@XMM[2]}, [$out]!
+	vadd.u32	@XMM[9],@XMM[8],@XMM[8]		@ compose 2<<96
+	vst1.8		{@XMM[5]}, [$out]!
+	vldmia		$fp, {@XMM[0]}			@ load counter
+
+	bne		.Lctr_enc_loop
+	b		.Lctr_enc_done
+
+.align	4
+.Lctr_enc_loop_done:
+	add		$len, $len, #8
+	vld1.8		{@XMM[8]}, [$inp]!	@ load input
+	veor		@XMM[0], @XMM[8]
+	vst1.8		{@XMM[0]}, [$out]!	@ write output
+	cmp		$len, #2
+	blo		.Lctr_enc_done
+	vld1.8		{@XMM[9]}, [$inp]!
+	veor		@XMM[1], @XMM[9]
+	vst1.8		{@XMM[1]}, [$out]!
+	beq		.Lctr_enc_done
+	vld1.8		{@XMM[10]}, [$inp]!
+	veor		@XMM[4], @XMM[10]
+	vst1.8		{@XMM[4]}, [$out]!
+	cmp		$len, #4
+	blo		.Lctr_enc_done
+	vld1.8		{@XMM[11]}, [$inp]!
+	veor		@XMM[6], @XMM[11]
+	vst1.8		{@XMM[6]}, [$out]!
+	beq		.Lctr_enc_done
+	vld1.8		{@XMM[12]}, [$inp]!
+	veor		@XMM[3], @XMM[12]
+	vst1.8		{@XMM[3]}, [$out]!
+	cmp		$len, #6
+	blo		.Lctr_enc_done
+	vld1.8		{@XMM[13]}, [$inp]!
+	veor		@XMM[7], @XMM[13]
+	vst1.8		{@XMM[7]}, [$out]!
+	beq		.Lctr_enc_done
+	vld1.8		{@XMM[14]}, [$inp]
+	veor		@XMM[2], @XMM[14]
+	vst1.8		{@XMM[2]}, [$out]!
+
+.Lctr_enc_done:
+	vmov.i32	q0, #0
+	vmov.i32	q1, #0
+#ifndef	BSAES_ASM_EXTENDED_KEY
+.Lctr_enc_bzero:			@ wipe key schedule [if any]
+	vstmia		$keysched!, {q0-q1}
+	cmp		$keysched, $fp
+	bne		.Lctr_enc_bzero
+#else
+	vstmia		$keysched, {q0-q1}
+#endif
+
+	mov	sp, $fp
+	add	sp, #0x10		@ add sp,$fp,#0x10 is no good for thumb
+	VFP_ABI_POP
+	ldmia	sp!, {r4-r10, pc}	@ return
+
+.align	4
+.Lctr_enc_short:
+	ldr	ip, [sp]		@ ctr pointer is passed on stack
+	stmdb	sp!, {r4-r8, lr}
+
+	mov	r4, $inp		@ copy arguments
+	mov	r5, $out
+	mov	r6, $len
+	mov	r7, $key
+	ldr	r8, [ip, #12]		@ load counter LSW
+	vld1.8	{@XMM[1]}, [ip]		@ load whole counter value
+#ifdef __ARMEL__
+	rev	r8, r8
+#endif
+	sub	sp, sp, #0x10
+	vst1.8	{@XMM[1]}, [sp,:64]	@ copy counter value
+	sub	sp, sp, #0x10
+
+.Lctr_enc_short_loop:
+	add	r0, sp, #0x10		@ input counter value
+	mov	r1, sp			@ output on the stack
+	mov	r2, r7			@ key
+
+	bl	AES_encrypt
+
+	vld1.8	{@XMM[0]}, [r4]!	@ load input
+	vld1.8	{@XMM[1]}, [sp,:64]	@ load encrypted counter
+	add	r8, r8, #1
+#ifdef __ARMEL__
+	rev	r0, r8
+	str	r0, [sp, #0x1c]		@ next counter value
+#else
+	str	r8, [sp, #0x1c]		@ next counter value
+#endif
+	veor	@XMM[0],@XMM[0],@XMM[1]
+	vst1.8	{@XMM[0]}, [r5]!	@ store output
+	subs	r6, r6, #1
+	bne	.Lctr_enc_short_loop
+
+	vmov.i32	q0, #0
+	vmov.i32	q1, #0
+	vstmia		sp!, {q0-q1}
+
+	ldmia	sp!, {r4-r8, pc}
+.size	bsaes_ctr32_encrypt_blocks,.-bsaes_ctr32_encrypt_blocks
+___
+}
+{
+######################################################################
+# void bsaes_xts_[en|de]crypt(const char *inp,char *out,size_t len,
+#	const AES_KEY *key1, const AES_KEY *key2,
+#	const unsigned char iv[16]);
+#
+my ($inp,$out,$len,$key,$rounds,$magic,$fp)=(map("r$_",(7..10,1..3)));
+my $const="r6";		# returned by _bsaes_key_convert
+my $twmask=@XMM[5];
+my @T=@XMM[6..7];
+
+$code.=<<___;
+.globl	bsaes_xts_encrypt
+.hidden	bsaes_xts_encrypt
+.type	bsaes_xts_encrypt,%function
+.align	4
+bsaes_xts_encrypt:
+	mov	ip, sp
+	stmdb	sp!, {r4-r10, lr}		@ 0x20
+	VFP_ABI_PUSH
+	mov	r6, sp				@ future $fp
+
+	mov	$inp, r0
+	mov	$out, r1
+	mov	$len, r2
+	mov	$key, r3
+
+	sub	r0, sp, #0x10			@ 0x10
+	bic	r0, #0xf			@ align at 16 bytes
+	mov	sp, r0
+
+#ifdef	XTS_CHAIN_TWEAK
+	ldr	r0, [ip]			@ pointer to input tweak
+#else
+	@ generate initial tweak
+	ldr	r0, [ip, #4]			@ iv[]
+	mov	r1, sp
+	ldr	r2, [ip, #0]			@ key2
+	bl	AES_encrypt
+	mov	r0,sp				@ pointer to initial tweak
+#endif
+
+	ldr	$rounds, [$key, #240]		@ get # of rounds
+	mov	$fp, r6
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	@ allocate the key schedule on the stack
+	sub	r12, sp, $rounds, lsl#7		@ 128 bytes per inner round key
+	@ add	r12, #`128-32`			@ size of bit-sliced key schedule
+	sub	r12, #`32+16`			@ place for tweak[9]
+
+	@ populate the key schedule
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	mov	sp, r12
+	add	r12, #0x90			@ pass key schedule
+	bl	_bsaes_key_convert
+	veor	@XMM[7], @XMM[7], @XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}			@ save last round key
+#else
+	ldr	r12, [$key, #244]
+	eors	r12, #1
+	beq	0f
+
+	str	r12, [$key, #244]
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	add	r12, $key, #248			@ pass key schedule
+	bl	_bsaes_key_convert
+	veor	@XMM[7], @XMM[7], @XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}
+
+.align	2
+0:	sub	sp, #0x90			@ place for tweak[9]
+#endif
+
+	vld1.8	{@XMM[8]}, [r0]			@ initial tweak
+	adr	$magic, .Lxts_magic
+
+	subs	$len, #0x80
+	blo	.Lxts_enc_short
+	b	.Lxts_enc_loop
+
+.align	4
+.Lxts_enc_loop:
+	vldmia		$magic, {$twmask}	@ load XTS magic
+	vshr.s64	@T[0], @XMM[8], #63
+	mov		r0, sp
+	vand		@T[0], @T[0], $twmask
+___
+for($i=9;$i<16;$i++) {
+$code.=<<___;
+	vadd.u64	@XMM[$i], @XMM[$i-1], @XMM[$i-1]
+	vst1.64		{@XMM[$i-1]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	vshr.s64	@T[1], @XMM[$i], #63
+	veor		@XMM[$i], @XMM[$i], @T[0]
+	vand		@T[1], @T[1], $twmask
+___
+	@T=reverse(@T);
+
+$code.=<<___ if ($i>=10);
+	vld1.8		{@XMM[$i-10]}, [$inp]!
+___
+$code.=<<___ if ($i>=11);
+	veor		@XMM[$i-11], @XMM[$i-11], @XMM[$i-3]
+___
+}
+$code.=<<___;
+	vadd.u64	@XMM[8], @XMM[15], @XMM[15]
+	vst1.64		{@XMM[15]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	veor		@XMM[8], @XMM[8], @T[0]
+	vst1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+
+	vld1.8		{@XMM[6]-@XMM[7]}, [$inp]!
+	veor		@XMM[5], @XMM[5], @XMM[13]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[6], @XMM[6], @XMM[14]
+	mov		r5, $rounds			@ pass rounds
+	veor		@XMM[7], @XMM[7], @XMM[15]
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[6], @XMM[11]
+	vld1.64		{@XMM[14]-@XMM[15]}, [r0,:128]!
+	veor		@XMM[10], @XMM[3], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	veor		@XMM[12], @XMM[2], @XMM[14]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+	veor		@XMM[13], @XMM[5], @XMM[15]
+	vst1.8		{@XMM[12]-@XMM[13]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+
+	subs		$len, #0x80
+	bpl		.Lxts_enc_loop
+
+.Lxts_enc_short:
+	adds		$len, #0x70
+	bmi		.Lxts_enc_done
+
+	vldmia		$magic, {$twmask}	@ load XTS magic
+	vshr.s64	@T[0], @XMM[8], #63
+	mov		r0, sp
+	vand		@T[0], @T[0], $twmask
+___
+for($i=9;$i<16;$i++) {
+$code.=<<___;
+	vadd.u64	@XMM[$i], @XMM[$i-1], @XMM[$i-1]
+	vst1.64		{@XMM[$i-1]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	vshr.s64	@T[1], @XMM[$i], #63
+	veor		@XMM[$i], @XMM[$i], @T[0]
+	vand		@T[1], @T[1], $twmask
+___
+	@T=reverse(@T);
+
+$code.=<<___ if ($i>=10);
+	vld1.8		{@XMM[$i-10]}, [$inp]!
+	subs		$len, #0x10
+	bmi		.Lxts_enc_`$i-9`
+___
+$code.=<<___ if ($i>=11);
+	veor		@XMM[$i-11], @XMM[$i-11], @XMM[$i-3]
+___
+}
+$code.=<<___;
+	sub		$len, #0x10
+	vst1.64		{@XMM[15]}, [r0,:128]		@ next round tweak
+
+	vld1.8		{@XMM[6]}, [$inp]!
+	veor		@XMM[5], @XMM[5], @XMM[13]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[6], @XMM[6], @XMM[14]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[6], @XMM[11]
+	vld1.64		{@XMM[14]}, [r0,:128]!
+	veor		@XMM[10], @XMM[3], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	veor		@XMM[12], @XMM[2], @XMM[14]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+	vst1.8		{@XMM[12]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+.align	4
+.Lxts_enc_6:
+	vst1.64		{@XMM[14]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[4], @XMM[4], @XMM[12]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[5], @XMM[5], @XMM[13]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[6], @XMM[11]
+	veor		@XMM[10], @XMM[3], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+
+@ put this in range for both ARM and Thumb mode adr instructions
+.align	5
+.Lxts_magic:
+	.quad	1, 0x87
+
+.align	5
+.Lxts_enc_5:
+	vst1.64		{@XMM[13]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[3], @XMM[3], @XMM[11]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[4], @XMM[4], @XMM[12]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[6], @XMM[11]
+	veor		@XMM[10], @XMM[3], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	vst1.8		{@XMM[10]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+.align	4
+.Lxts_enc_4:
+	vst1.64		{@XMM[12]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[2], @XMM[2], @XMM[10]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[3], @XMM[3], @XMM[11]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[6], @XMM[11]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+.align	4
+.Lxts_enc_3:
+	vst1.64		{@XMM[11]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[1], @XMM[1], @XMM[9]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[2], @XMM[2], @XMM[10]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[8]-@XMM[9]}, [r0,:128]!
+	vld1.64		{@XMM[10]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[4], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	vst1.8		{@XMM[8]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+.align	4
+.Lxts_enc_2:
+	vst1.64		{@XMM[10]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[0], @XMM[0], @XMM[8]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[1], @XMM[1], @XMM[9]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_encrypt8
+
+	vld1.64		{@XMM[8]-@XMM[9]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_enc_done
+.align	4
+.Lxts_enc_1:
+	mov		r0, sp
+	veor		@XMM[0], @XMM[8]
+	mov		r1, sp
+	vst1.8		{@XMM[0]}, [sp,:128]
+	mov		r2, $key
+	mov		r4, $fp				@ preserve fp
+
+	bl		AES_encrypt
+
+	vld1.8		{@XMM[0]}, [sp,:128]
+	veor		@XMM[0], @XMM[0], @XMM[8]
+	vst1.8		{@XMM[0]}, [$out]!
+	mov		$fp, r4
+
+	vmov		@XMM[8], @XMM[9]		@ next round tweak
+
+.Lxts_enc_done:
+#ifndef	XTS_CHAIN_TWEAK
+	adds		$len, #0x10
+	beq		.Lxts_enc_ret
+	sub		r6, $out, #0x10
+
+.Lxts_enc_steal:
+	ldrb		r0, [$inp], #1
+	ldrb		r1, [$out, #-0x10]
+	strb		r0, [$out, #-0x10]
+	strb		r1, [$out], #1
+
+	subs		$len, #1
+	bhi		.Lxts_enc_steal
+
+	vld1.8		{@XMM[0]}, [r6]
+	mov		r0, sp
+	veor		@XMM[0], @XMM[0], @XMM[8]
+	mov		r1, sp
+	vst1.8		{@XMM[0]}, [sp,:128]
+	mov		r2, $key
+	mov		r4, $fp			@ preserve fp
+
+	bl		AES_encrypt
+
+	vld1.8		{@XMM[0]}, [sp,:128]
+	veor		@XMM[0], @XMM[0], @XMM[8]
+	vst1.8		{@XMM[0]}, [r6]
+	mov		$fp, r4
+#endif
+
+.Lxts_enc_ret:
+	bic		r0, $fp, #0xf
+	vmov.i32	q0, #0
+	vmov.i32	q1, #0
+#ifdef	XTS_CHAIN_TWEAK
+	ldr		r1, [$fp, #0x20+VFP_ABI_FRAME]	@ chain tweak
+#endif
+.Lxts_enc_bzero:				@ wipe key schedule [if any]
+	vstmia		sp!, {q0-q1}
+	cmp		sp, r0
+	bne		.Lxts_enc_bzero
+
+	mov		sp, $fp
+#ifdef	XTS_CHAIN_TWEAK
+	vst1.8		{@XMM[8]}, [r1]
+#endif
+	VFP_ABI_POP
+	ldmia		sp!, {r4-r10, pc}	@ return
+
+.size	bsaes_xts_encrypt,.-bsaes_xts_encrypt
+
+.globl	bsaes_xts_decrypt
+.hidden	bsaes_xts_decrypt
+.type	bsaes_xts_decrypt,%function
+.align	4
+bsaes_xts_decrypt:
+	mov	ip, sp
+	stmdb	sp!, {r4-r10, lr}		@ 0x20
+	VFP_ABI_PUSH
+	mov	r6, sp				@ future $fp
+
+	mov	$inp, r0
+	mov	$out, r1
+	mov	$len, r2
+	mov	$key, r3
+
+	sub	r0, sp, #0x10			@ 0x10
+	bic	r0, #0xf			@ align at 16 bytes
+	mov	sp, r0
+
+#ifdef	XTS_CHAIN_TWEAK
+	ldr	r0, [ip]			@ pointer to input tweak
+#else
+	@ generate initial tweak
+	ldr	r0, [ip, #4]			@ iv[]
+	mov	r1, sp
+	ldr	r2, [ip, #0]			@ key2
+	bl	AES_encrypt
+	mov	r0, sp				@ pointer to initial tweak
+#endif
+
+	ldr	$rounds, [$key, #240]		@ get # of rounds
+	mov	$fp, r6
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	@ allocate the key schedule on the stack
+	sub	r12, sp, $rounds, lsl#7		@ 128 bytes per inner round key
+	@ add	r12, #`128-32`			@ size of bit-sliced key schedule
+	sub	r12, #`32+16`			@ place for tweak[9]
+
+	@ populate the key schedule
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	mov	sp, r12
+	add	r12, #0x90			@ pass key schedule
+	bl	_bsaes_key_convert
+	add	r4, sp, #0x90
+	vldmia	r4, {@XMM[6]}
+	vstmia	r12,  {@XMM[15]}		@ save last round key
+	veor	@XMM[7], @XMM[7], @XMM[6]	@ fix up round 0 key
+	vstmia	r4, {@XMM[7]}
+#else
+	ldr	r12, [$key, #244]
+	eors	r12, #1
+	beq	0f
+
+	str	r12, [$key, #244]
+	mov	r4, $key			@ pass key
+	mov	r5, $rounds			@ pass # of rounds
+	add	r12, $key, #248			@ pass key schedule
+	bl	_bsaes_key_convert
+	add	r4, $key, #248
+	vldmia	r4, {@XMM[6]}
+	vstmia	r12,  {@XMM[15]}		@ save last round key
+	veor	@XMM[7], @XMM[7], @XMM[6]	@ fix up round 0 key
+	vstmia	r4, {@XMM[7]}
+
+.align	2
+0:	sub	sp, #0x90			@ place for tweak[9]
+#endif
+	vld1.8	{@XMM[8]}, [r0]			@ initial tweak
+	adr	$magic, .Lxts_magic
+
+	tst	$len, #0xf			@ if not multiple of 16
+	it	ne				@ Thumb2 thing, sanity check in ARM
+	subne	$len, #0x10			@ subtract another 16 bytes
+	subs	$len, #0x80
+
+	blo	.Lxts_dec_short
+	b	.Lxts_dec_loop
+
+.align	4
+.Lxts_dec_loop:
+	vldmia		$magic, {$twmask}	@ load XTS magic
+	vshr.s64	@T[0], @XMM[8], #63
+	mov		r0, sp
+	vand		@T[0], @T[0], $twmask
+___
+for($i=9;$i<16;$i++) {
+$code.=<<___;
+	vadd.u64	@XMM[$i], @XMM[$i-1], @XMM[$i-1]
+	vst1.64		{@XMM[$i-1]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	vshr.s64	@T[1], @XMM[$i], #63
+	veor		@XMM[$i], @XMM[$i], @T[0]
+	vand		@T[1], @T[1], $twmask
+___
+	@T=reverse(@T);
+
+$code.=<<___ if ($i>=10);
+	vld1.8		{@XMM[$i-10]}, [$inp]!
+___
+$code.=<<___ if ($i>=11);
+	veor		@XMM[$i-11], @XMM[$i-11], @XMM[$i-3]
+___
+}
+$code.=<<___;
+	vadd.u64	@XMM[8], @XMM[15], @XMM[15]
+	vst1.64		{@XMM[15]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	veor		@XMM[8], @XMM[8], @T[0]
+	vst1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+
+	vld1.8		{@XMM[6]-@XMM[7]}, [$inp]!
+	veor		@XMM[5], @XMM[5], @XMM[13]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[6], @XMM[6], @XMM[14]
+	mov		r5, $rounds			@ pass rounds
+	veor		@XMM[7], @XMM[7], @XMM[15]
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[4], @XMM[11]
+	vld1.64		{@XMM[14]-@XMM[15]}, [r0,:128]!
+	veor		@XMM[10], @XMM[2], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	veor		@XMM[12], @XMM[3], @XMM[14]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+	veor		@XMM[13], @XMM[5], @XMM[15]
+	vst1.8		{@XMM[12]-@XMM[13]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+
+	subs		$len, #0x80
+	bpl		.Lxts_dec_loop
+
+.Lxts_dec_short:
+	adds		$len, #0x70
+	bmi		.Lxts_dec_done
+
+	vldmia		$magic, {$twmask}	@ load XTS magic
+	vshr.s64	@T[0], @XMM[8], #63
+	mov		r0, sp
+	vand		@T[0], @T[0], $twmask
+___
+for($i=9;$i<16;$i++) {
+$code.=<<___;
+	vadd.u64	@XMM[$i], @XMM[$i-1], @XMM[$i-1]
+	vst1.64		{@XMM[$i-1]}, [r0,:128]!
+	vswp		`&Dhi("@T[0]")`,`&Dlo("@T[0]")`
+	vshr.s64	@T[1], @XMM[$i], #63
+	veor		@XMM[$i], @XMM[$i], @T[0]
+	vand		@T[1], @T[1], $twmask
+___
+	@T=reverse(@T);
+
+$code.=<<___ if ($i>=10);
+	vld1.8		{@XMM[$i-10]}, [$inp]!
+	subs		$len, #0x10
+	bmi		.Lxts_dec_`$i-9`
+___
+$code.=<<___ if ($i>=11);
+	veor		@XMM[$i-11], @XMM[$i-11], @XMM[$i-3]
+___
+}
+$code.=<<___;
+	sub		$len, #0x10
+	vst1.64		{@XMM[15]}, [r0,:128]		@ next round tweak
+
+	vld1.8		{@XMM[6]}, [$inp]!
+	veor		@XMM[5], @XMM[5], @XMM[13]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[6], @XMM[6], @XMM[14]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[4], @XMM[11]
+	vld1.64		{@XMM[14]}, [r0,:128]!
+	veor		@XMM[10], @XMM[2], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	veor		@XMM[12], @XMM[3], @XMM[14]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+	vst1.8		{@XMM[12]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_6:
+	vst1.64		{@XMM[14]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[4], @XMM[4], @XMM[12]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[5], @XMM[5], @XMM[13]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]-@XMM[13]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[4], @XMM[11]
+	veor		@XMM[10], @XMM[2], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	veor		@XMM[11], @XMM[7], @XMM[13]
+	vst1.8		{@XMM[10]-@XMM[11]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_5:
+	vst1.64		{@XMM[13]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[3], @XMM[3], @XMM[11]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[4], @XMM[4], @XMM[12]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	vld1.64		{@XMM[12]}, [r0,:128]!
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[4], @XMM[11]
+	veor		@XMM[10], @XMM[2], @XMM[12]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+	vst1.8		{@XMM[10]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_4:
+	vst1.64		{@XMM[12]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[2], @XMM[2], @XMM[10]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[3], @XMM[3], @XMM[11]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[ 8]-@XMM[ 9]}, [r0,:128]!
+	vld1.64		{@XMM[10]-@XMM[11]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	veor		@XMM[9], @XMM[4], @XMM[11]
+	vst1.8		{@XMM[8]-@XMM[9]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_3:
+	vst1.64		{@XMM[11]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[1], @XMM[1], @XMM[9]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[2], @XMM[2], @XMM[10]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[8]-@XMM[9]}, [r0,:128]!
+	vld1.64		{@XMM[10]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	veor		@XMM[8], @XMM[6], @XMM[10]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+	vst1.8		{@XMM[8]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_2:
+	vst1.64		{@XMM[10]}, [r0,:128]		@ next round tweak
+
+	veor		@XMM[0], @XMM[0], @XMM[8]
+#ifndef	BSAES_ASM_EXTENDED_KEY
+	add		r4, sp, #0x90			@ pass key schedule
+#else
+	add		r4, $key, #248			@ pass key schedule
+#endif
+	veor		@XMM[1], @XMM[1], @XMM[9]
+	mov		r5, $rounds			@ pass rounds
+	mov		r0, sp
+
+	bl		_bsaes_decrypt8
+
+	vld1.64		{@XMM[8]-@XMM[9]}, [r0,:128]!
+	veor		@XMM[0], @XMM[0], @XMM[ 8]
+	veor		@XMM[1], @XMM[1], @XMM[ 9]
+	vst1.8		{@XMM[0]-@XMM[1]}, [$out]!
+
+	vld1.64		{@XMM[8]}, [r0,:128]		@ next round tweak
+	b		.Lxts_dec_done
+.align	4
+.Lxts_dec_1:
+	mov		r0, sp
+	veor		@XMM[0], @XMM[8]
+	mov		r1, sp
+	vst1.8		{@XMM[0]}, [sp,:128]
+	mov		r2, $key
+	mov		r4, $fp				@ preserve fp
+	mov		r5, $magic			@ preserve magic
+
+	bl		AES_decrypt
+
+	vld1.8		{@XMM[0]}, [sp,:128]
+	veor		@XMM[0], @XMM[0], @XMM[8]
+	vst1.8		{@XMM[0]}, [$out]!
+	mov		$fp, r4
+	mov		$magic, r5
+
+	vmov		@XMM[8], @XMM[9]		@ next round tweak
+
+.Lxts_dec_done:
+#ifndef	XTS_CHAIN_TWEAK
+	adds		$len, #0x10
+	beq		.Lxts_dec_ret
+
+	@ calculate one round of extra tweak for the stolen ciphertext
+	vldmia		$magic, {$twmask}
+	vshr.s64	@XMM[6], @XMM[8], #63
+	vand		@XMM[6], @XMM[6], $twmask
+	vadd.u64	@XMM[9], @XMM[8], @XMM[8]
+	vswp		`&Dhi("@XMM[6]")`,`&Dlo("@XMM[6]")`
+	veor		@XMM[9], @XMM[9], @XMM[6]
+
+	@ perform the final decryption with the last tweak value
+	vld1.8		{@XMM[0]}, [$inp]!
+	mov		r0, sp
+	veor		@XMM[0], @XMM[0], @XMM[9]
+	mov		r1, sp
+	vst1.8		{@XMM[0]}, [sp,:128]
+	mov		r2, $key
+	mov		r4, $fp			@ preserve fp
+
+	bl		AES_decrypt
+
+	vld1.8		{@XMM[0]}, [sp,:128]
+	veor		@XMM[0], @XMM[0], @XMM[9]
+	vst1.8		{@XMM[0]}, [$out]
+
+	mov		r6, $out
+.Lxts_dec_steal:
+	ldrb		r1, [$out]
+	ldrb		r0, [$inp], #1
+	strb		r1, [$out, #0x10]
+	strb		r0, [$out], #1
+
+	subs		$len, #1
+	bhi		.Lxts_dec_steal
+
+	vld1.8		{@XMM[0]}, [r6]
+	mov		r0, sp
+	veor		@XMM[0], @XMM[8]
+	mov		r1, sp
+	vst1.8		{@XMM[0]}, [sp,:128]
+	mov		r2, $key
+
+	bl		AES_decrypt
+
+	vld1.8		{@XMM[0]}, [sp,:128]
+	veor		@XMM[0], @XMM[0], @XMM[8]
+	vst1.8		{@XMM[0]}, [r6]
+	mov		$fp, r4
+#endif
+
+.Lxts_dec_ret:
+	bic		r0, $fp, #0xf
+	vmov.i32	q0, #0
+	vmov.i32	q1, #0
+#ifdef	XTS_CHAIN_TWEAK
+	ldr		r1, [$fp, #0x20+VFP_ABI_FRAME]	@ chain tweak
+#endif
+.Lxts_dec_bzero:				@ wipe key schedule [if any]
+	vstmia		sp!, {q0-q1}
+	cmp		sp, r0
+	bne		.Lxts_dec_bzero
+
+	mov		sp, $fp
+#ifdef	XTS_CHAIN_TWEAK
+	vst1.8		{@XMM[8]}, [r1]
+#endif
+	VFP_ABI_POP
+	ldmia		sp!, {r4-r10, pc}	@ return
+
+.size	bsaes_xts_decrypt,.-bsaes_xts_decrypt
+___
+}
+$code.=<<___;
+#endif
+#endif
+___
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+        last if (!s/^#/@/ and !/^$/);
+        print;
+}
+close SELF;
+
+print $code;
+
+close STDOUT;
diff --git a/src/crypto/aes/asm/bsaes-x86_64.pl b/src/crypto/aes/asm/bsaes-x86_64.pl
new file mode 100644
index 0000000..3f7d33c
--- /dev/null
+++ b/src/crypto/aes/asm/bsaes-x86_64.pl
@@ -0,0 +1,3102 @@
+#!/usr/bin/env perl
+
+###################################################################
+### AES-128 [originally in CTR mode]				###
+### bitsliced implementation for Intel Core 2 processors	###
+### requires support of SSE extensions up to SSSE3		###
+### Author: Emilia Käsper and Peter Schwabe			###
+### Date: 2009-03-19						###
+### Public domain						###
+###								###
+### See http://homes.esat.kuleuven.be/~ekasper/#software for	###
+### further information.					###
+###################################################################
+#
+# September 2011.
+#
+# Started as transliteration to "perlasm" the original code has
+# undergone following changes:
+#
+# - code was made position-independent;
+# - rounds were folded into a loop resulting in >5x size reduction
+#   from 12.5KB to 2.2KB;
+# - above was possibile thanks to mixcolumns() modification that
+#   allowed to feed its output back to aesenc[last], this was
+#   achieved at cost of two additional inter-registers moves;
+# - some instruction reordering and interleaving;
+# - this module doesn't implement key setup subroutine, instead it
+#   relies on conversion of "conventional" key schedule as returned
+#   by AES_set_encrypt_key (see discussion below);
+# - first and last round keys are treated differently, which allowed
+#   to skip one shiftrows(), reduce bit-sliced key schedule and
+#   speed-up conversion by 22%;
+# - support for 192- and 256-bit keys was added;
+#
+# Resulting performance in CPU cycles spent to encrypt one byte out
+# of 4096-byte buffer with 128-bit key is:
+#
+#		Emilia's	this(*)		difference
+#
+# Core 2    	9.30		8.69		+7%
+# Nehalem(**) 	7.63		6.88		+11%
+# Atom	    	17.1		16.4		+4%
+# Silvermont	-		12.9
+#
+# (*)	Comparison is not completely fair, because "this" is ECB,
+#	i.e. no extra processing such as counter values calculation
+#	and xor-ing input as in Emilia's CTR implementation is
+#	performed. However, the CTR calculations stand for not more
+#	than 1% of total time, so comparison is *rather* fair.
+#
+# (**)	Results were collected on Westmere, which is considered to
+#	be equivalent to Nehalem for this code.
+#
+# As for key schedule conversion subroutine. Interface to OpenSSL
+# relies on per-invocation on-the-fly conversion. This naturally
+# has impact on performance, especially for short inputs. Conversion
+# time in CPU cycles and its ratio to CPU cycles spent in 8x block
+# function is:
+#
+# 		conversion	conversion/8x block
+# Core 2	240		0.22
+# Nehalem	180		0.20
+# Atom		430		0.20
+#
+# The ratio values mean that 128-byte blocks will be processed
+# 16-18% slower, 256-byte blocks - 9-10%, 384-byte blocks - 6-7%,
+# etc. Then keep in mind that input sizes not divisible by 128 are
+# *effectively* slower, especially shortest ones, e.g. consecutive
+# 144-byte blocks are processed 44% slower than one would expect,
+# 272 - 29%, 400 - 22%, etc. Yet, despite all these "shortcomings"
+# it's still faster than ["hyper-threading-safe" code path in]
+# aes-x86_64.pl on all lengths above 64 bytes...
+#
+# October 2011.
+#
+# Add decryption procedure. Performance in CPU cycles spent to decrypt
+# one byte out of 4096-byte buffer with 128-bit key is:
+#
+# Core 2	9.98
+# Nehalem	7.80
+# Atom		17.9
+# Silvermont	14.0
+#
+# November 2011.
+#
+# Add bsaes_xts_[en|de]crypt. Less-than-80-bytes-block performance is
+# suboptimal, but XTS is meant to be used with larger blocks...
+#
+#						<appro@openssl.org>
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+my ($inp,$out,$len,$key,$ivp)=("%rdi","%rsi","%rdx","%rcx");
+my @XMM=map("%xmm$_",(15,0..14));	# best on Atom, +10% over (0..15)
+my $ecb=0;	# suppress unreferenced ECB subroutines, spare some space...
+
+{
+my ($key,$rounds,$const)=("%rax","%r10d","%r11");
+
+sub Sbox {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b0, b1, b4, b6, b3, b7, b2, b5] < msb
+my @b=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+	&InBasisChange	(@b);
+	&Inv_GF256	(@b[6,5,0,3,7,1,4,2],@t,@s);
+	&OutBasisChange	(@b[7,1,4,2,6,5,0,3]);
+}
+
+sub InBasisChange {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b6, b5, b0, b3, b7, b1, b4, b2] < msb 
+my @b=@_[0..7];
+$code.=<<___;
+	pxor	@b[6], @b[5]
+	pxor	@b[1], @b[2]
+	pxor	@b[0], @b[3]
+	pxor	@b[2], @b[6]
+	pxor 	@b[0], @b[5]
+
+	pxor	@b[3], @b[6]
+	pxor	@b[7], @b[3]
+	pxor	@b[5], @b[7]
+	pxor	@b[4], @b[3]
+	pxor	@b[5], @b[4]
+	pxor	@b[1], @b[3]
+
+	pxor	@b[7], @b[2]
+	pxor	@b[5], @b[1]
+___
+}
+
+sub OutBasisChange {
+# input in  lsb > [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb > [b6, b1, b2, b4, b7, b0, b3, b5] < msb
+my @b=@_[0..7];
+$code.=<<___;
+	pxor	@b[6], @b[0]
+	pxor	@b[4], @b[1]
+	pxor	@b[0], @b[2]
+	pxor	@b[6], @b[4]
+	pxor	@b[1], @b[6]
+
+	pxor	@b[5], @b[1]
+	pxor	@b[3], @b[5]
+	pxor	@b[7], @b[3]
+	pxor	@b[5], @b[7]
+	pxor	@b[5], @b[2]
+
+	pxor	@b[7], @b[4]
+___
+}
+
+sub InvSbox {
+# input in lsb 	> [b0, b1, b2, b3, b4, b5, b6, b7] < msb
+# output in lsb	> [b0, b1, b6, b4, b2, b7, b3, b5] < msb
+my @b=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+	&InvInBasisChange	(@b);
+	&Inv_GF256		(@b[5,1,2,6,3,7,0,4],@t,@s);
+	&InvOutBasisChange	(@b[3,7,0,4,5,1,2,6]);
+}
+
+sub InvInBasisChange {		# OutBasisChange in reverse
+my @b=@_[5,1,2,6,3,7,0,4];
+$code.=<<___
+	pxor	@b[7], @b[4]
+
+	pxor	@b[5], @b[7]
+	pxor	@b[5], @b[2]
+	pxor	@b[7], @b[3]
+	pxor	@b[3], @b[5]
+	pxor	@b[5], @b[1]
+
+	pxor	@b[1], @b[6]
+	pxor	@b[0], @b[2]
+	pxor	@b[6], @b[4]
+	pxor	@b[6], @b[0]
+	pxor	@b[4], @b[1]
+___
+}
+
+sub InvOutBasisChange {		# InBasisChange in reverse
+my @b=@_[2,5,7,3,6,1,0,4];
+$code.=<<___;
+	pxor	@b[5], @b[1]
+	pxor	@b[7], @b[2]
+
+	pxor	@b[1], @b[3]
+	pxor	@b[5], @b[4]
+	pxor	@b[5], @b[7]
+	pxor	@b[4], @b[3]
+	 pxor 	@b[0], @b[5]
+	pxor	@b[7], @b[3]
+	 pxor	@b[2], @b[6]
+	 pxor	@b[1], @b[2]
+	pxor	@b[3], @b[6]
+
+	pxor	@b[0], @b[3]
+	pxor	@b[6], @b[5]
+___
+}
+
+sub Mul_GF4 {
+#;*************************************************************
+#;* Mul_GF4: Input x0-x1,y0-y1 Output x0-x1 Temp t0 (8) *
+#;*************************************************************
+my ($x0,$x1,$y0,$y1,$t0)=@_;
+$code.=<<___;
+	movdqa	$y0, $t0
+	pxor 	$y1, $t0
+	pand	$x0, $t0
+	pxor	$x1, $x0
+	pand	$y0, $x1
+	pand	$y1, $x0
+	pxor	$x1, $x0
+	pxor	$t0, $x1
+___
+}
+
+sub Mul_GF4_N {				# not used, see next subroutine
+# multiply and scale by N
+my ($x0,$x1,$y0,$y1,$t0)=@_;
+$code.=<<___;
+	movdqa	$y0, $t0
+	pxor	$y1, $t0
+	pand	$x0, $t0
+	pxor	$x1, $x0
+	pand	$y0, $x1
+	pand	$y1, $x0
+	pxor	$x0, $x1
+	pxor	$t0, $x0
+___
+}
+
+sub Mul_GF4_N_GF4 {
+# interleaved Mul_GF4_N and Mul_GF4
+my ($x0,$x1,$y0,$y1,$t0,
+    $x2,$x3,$y2,$y3,$t1)=@_;
+$code.=<<___;
+	movdqa	$y0, $t0
+	 movdqa	$y2, $t1
+	pxor	$y1, $t0
+	 pxor 	$y3, $t1
+	pand	$x0, $t0
+	 pand	$x2, $t1
+	pxor	$x1, $x0
+	 pxor	$x3, $x2
+	pand	$y0, $x1
+	 pand	$y2, $x3
+	pand	$y1, $x0
+	 pand	$y3, $x2
+	pxor	$x0, $x1
+	 pxor	$x3, $x2
+	pxor	$t0, $x0
+	 pxor	$t1, $x3
+___
+}
+sub Mul_GF16_2 {
+my @x=@_[0..7];
+my @y=@_[8..11];
+my @t=@_[12..15];
+$code.=<<___;
+	movdqa	@x[0], @t[0]
+	movdqa	@x[1], @t[1]
+___
+	&Mul_GF4  	(@x[0], @x[1], @y[0], @y[1], @t[2]);
+$code.=<<___;
+	pxor	@x[2], @t[0]
+	pxor	@x[3], @t[1]
+	pxor	@y[2], @y[0]
+	pxor	@y[3], @y[1]
+___
+	Mul_GF4_N_GF4	(@t[0], @t[1], @y[0], @y[1], @t[3],
+			 @x[2], @x[3], @y[2], @y[3], @t[2]);
+$code.=<<___;
+	pxor	@t[0], @x[0]
+	pxor	@t[0], @x[2]
+	pxor	@t[1], @x[1]
+	pxor	@t[1], @x[3]
+
+	movdqa	@x[4], @t[0]
+	movdqa	@x[5], @t[1]
+	pxor	@x[6], @t[0]
+	pxor	@x[7], @t[1]
+___
+	&Mul_GF4_N_GF4	(@t[0], @t[1], @y[0], @y[1], @t[3],
+			 @x[6], @x[7], @y[2], @y[3], @t[2]);
+$code.=<<___;
+	pxor	@y[2], @y[0]
+	pxor	@y[3], @y[1]
+___
+	&Mul_GF4  	(@x[4], @x[5], @y[0], @y[1], @t[3]);
+$code.=<<___;
+	pxor	@t[0], @x[4]
+	pxor	@t[0], @x[6]
+	pxor	@t[1], @x[5]
+	pxor	@t[1], @x[7]
+___
+}
+sub Inv_GF256 {
+#;********************************************************************
+#;* Inv_GF256: Input x0-x7 Output x0-x7 Temp t0-t3,s0-s3 (144)       *
+#;********************************************************************
+my @x=@_[0..7];
+my @t=@_[8..11];
+my @s=@_[12..15];
+# direct optimizations from hardware
+$code.=<<___;
+	movdqa	@x[4], @t[3]
+	movdqa	@x[5], @t[2]
+	movdqa	@x[1], @t[1]
+	movdqa	@x[7], @s[1]
+	movdqa	@x[0], @s[0]
+
+	pxor	@x[6], @t[3]
+	pxor	@x[7], @t[2]
+	pxor	@x[3], @t[1]
+	 movdqa	@t[3], @s[2]
+	pxor	@x[6], @s[1]
+	 movdqa	@t[2], @t[0]
+	pxor	@x[2], @s[0]
+	 movdqa	@t[3], @s[3]
+
+	por	@t[1], @t[2]
+	por	@s[0], @t[3]
+	pxor	@t[0], @s[3]
+	pand	@s[0], @s[2]
+	pxor	@t[1], @s[0]
+	pand	@t[1], @t[0]
+	pand	@s[0], @s[3]
+	movdqa	@x[3], @s[0]
+	pxor	@x[2], @s[0]
+	pand	@s[0], @s[1]
+	pxor	@s[1], @t[3]
+	pxor	@s[1], @t[2]
+	movdqa	@x[4], @s[1]
+	movdqa	@x[1], @s[0]
+	pxor	@x[5], @s[1]
+	pxor	@x[0], @s[0]
+	movdqa	@s[1], @t[1]
+	pand	@s[0], @s[1]
+	por	@s[0], @t[1]
+	pxor	@s[1], @t[0]
+	pxor	@s[3], @t[3]
+	pxor	@s[2], @t[2]
+	pxor	@s[3], @t[1]
+	movdqa	@x[7], @s[0]
+	pxor	@s[2], @t[0]
+	movdqa	@x[6], @s[1]
+	pxor	@s[2], @t[1]
+	movdqa	@x[5], @s[2]
+	pand	@x[3], @s[0]
+	movdqa	@x[4], @s[3]
+	pand	@x[2], @s[1]
+	pand	@x[1], @s[2]
+	por	@x[0], @s[3]
+	pxor	@s[0], @t[3]
+	pxor	@s[1], @t[2]
+	pxor	@s[2], @t[1]
+	pxor	@s[3], @t[0] 
+
+	#Inv_GF16 \t0, \t1, \t2, \t3, \s0, \s1, \s2, \s3
+
+	# new smaller inversion
+
+	movdqa	@t[3], @s[0]
+	pand	@t[1], @t[3]
+	pxor	@t[2], @s[0]
+
+	movdqa	@t[0], @s[2]
+	movdqa	@s[0], @s[3]
+	pxor	@t[3], @s[2]
+	pand	@s[2], @s[3]
+
+	movdqa	@t[1], @s[1]
+	pxor	@t[2], @s[3]
+	pxor	@t[0], @s[1]
+
+	pxor	@t[2], @t[3]
+
+	pand	@t[3], @s[1]
+
+	movdqa	@s[2], @t[2]
+	pxor	@t[0], @s[1]
+
+	pxor	@s[1], @t[2]
+	pxor	@s[1], @t[1]
+
+	pand	@t[0], @t[2]
+
+	pxor	@t[2], @s[2]
+	pxor	@t[2], @t[1]
+
+	pand	@s[3], @s[2]
+
+	pxor	@s[0], @s[2]
+___
+# output in s3, s2, s1, t1
+
+# Mul_GF16_2 \x0, \x1, \x2, \x3, \x4, \x5, \x6, \x7, \t2, \t3, \t0, \t1, \s0, \s1, \s2, \s3
+
+# Mul_GF16_2 \x0, \x1, \x2, \x3, \x4, \x5, \x6, \x7, \s3, \s2, \s1, \t1, \s0, \t0, \t2, \t3
+	&Mul_GF16_2(@x,@s[3,2,1],@t[1],@s[0],@t[0,2,3]);
+
+### output msb > [x3,x2,x1,x0,x7,x6,x5,x4] < lsb
+}
+
+# AES linear components
+
+sub ShiftRows {
+my @x=@_[0..7];
+my $mask=pop;
+$code.=<<___;
+	pxor	0x00($key),@x[0]
+	pxor	0x10($key),@x[1]
+	pxor	0x20($key),@x[2]
+	pxor	0x30($key),@x[3]
+	pshufb	$mask,@x[0]
+	pshufb	$mask,@x[1]
+	pxor	0x40($key),@x[4]
+	pxor	0x50($key),@x[5]
+	pshufb	$mask,@x[2]
+	pshufb	$mask,@x[3]
+	pxor	0x60($key),@x[6]
+	pxor	0x70($key),@x[7]
+	pshufb	$mask,@x[4]
+	pshufb	$mask,@x[5]
+	pshufb	$mask,@x[6]
+	pshufb	$mask,@x[7]
+	lea	0x80($key),$key
+___
+}
+
+sub MixColumns {
+# modified to emit output in order suitable for feeding back to aesenc[last]
+my @x=@_[0..7];
+my @t=@_[8..15];
+my $inv=@_[16];	# optional
+$code.=<<___;
+	pshufd	\$0x93, @x[0], @t[0]	# x0 <<< 32
+	pshufd	\$0x93, @x[1], @t[1]
+	 pxor	@t[0], @x[0]		# x0 ^ (x0 <<< 32)
+	pshufd	\$0x93, @x[2], @t[2]
+	 pxor	@t[1], @x[1]
+	pshufd	\$0x93, @x[3], @t[3]
+	 pxor	@t[2], @x[2]
+	pshufd	\$0x93, @x[4], @t[4]
+	 pxor	@t[3], @x[3]
+	pshufd	\$0x93, @x[5], @t[5]
+	 pxor	@t[4], @x[4]
+	pshufd	\$0x93, @x[6], @t[6]
+	 pxor	@t[5], @x[5]
+	pshufd	\$0x93, @x[7], @t[7]
+	 pxor	@t[6], @x[6]
+	 pxor	@t[7], @x[7]
+
+	pxor	@x[0], @t[1]
+	pxor	@x[7], @t[0]
+	pxor	@x[7], @t[1]
+	 pshufd	\$0x4E, @x[0], @x[0] 	# (x0 ^ (x0 <<< 32)) <<< 64)
+	pxor	@x[1], @t[2]
+	 pshufd	\$0x4E, @x[1], @x[1]
+	pxor	@x[4], @t[5]
+	 pxor	@t[0], @x[0]
+	pxor	@x[5], @t[6]
+	 pxor	@t[1], @x[1]
+	pxor	@x[3], @t[4]
+	 pshufd	\$0x4E, @x[4], @t[0]
+	pxor	@x[6], @t[7]
+	 pshufd	\$0x4E, @x[5], @t[1]
+	pxor	@x[2], @t[3]
+	 pshufd	\$0x4E, @x[3], @x[4]
+	pxor	@x[7], @t[3]
+	 pshufd	\$0x4E, @x[7], @x[5]
+	pxor	@x[7], @t[4]
+	 pshufd	\$0x4E, @x[6], @x[3]
+	pxor	@t[4], @t[0]
+	 pshufd	\$0x4E, @x[2], @x[6]
+	pxor	@t[5], @t[1]
+___
+$code.=<<___ if (!$inv);
+	pxor	@t[3], @x[4]
+	pxor	@t[7], @x[5]
+	pxor	@t[6], @x[3]
+	 movdqa	@t[0], @x[2]
+	pxor	@t[2], @x[6]
+	 movdqa	@t[1], @x[7]
+___
+$code.=<<___ if ($inv);
+	pxor	@x[4], @t[3]
+	pxor	@t[7], @x[5]
+	pxor	@x[3], @t[6]
+	 movdqa	@t[0], @x[3]
+	pxor	@t[2], @x[6]
+	 movdqa	@t[6], @x[2]
+	 movdqa	@t[1], @x[7]
+	 movdqa	@x[6], @x[4]
+	 movdqa	@t[3], @x[6]
+___
+}
+
+sub InvMixColumns_orig {
+my @x=@_[0..7];
+my @t=@_[8..15];
+
+$code.=<<___;
+	# multiplication by 0x0e
+	pshufd	\$0x93, @x[7], @t[7]
+	movdqa	@x[2], @t[2]
+	pxor	@x[5], @x[7]		# 7 5
+	pxor	@x[5], @x[2]		# 2 5
+	pshufd	\$0x93, @x[0], @t[0]
+	movdqa	@x[5], @t[5]
+	pxor	@x[0], @x[5]		# 5 0		[1]
+	pxor	@x[1], @x[0]		# 0 1
+	pshufd	\$0x93, @x[1], @t[1]
+	pxor	@x[2], @x[1]		# 1 25
+	pxor	@x[6], @x[0]		# 01 6		[2]
+	pxor	@x[3], @x[1]		# 125 3		[4]
+	pshufd	\$0x93, @x[3], @t[3]
+	pxor	@x[0], @x[2]		# 25 016	[3]
+	pxor	@x[7], @x[3]		# 3 75
+	pxor	@x[6], @x[7]		# 75 6		[0]
+	pshufd	\$0x93, @x[6], @t[6]
+	movdqa	@x[4], @t[4]
+	pxor	@x[4], @x[6]		# 6 4
+	pxor	@x[3], @x[4]		# 4 375		[6]
+	pxor	@x[7], @x[3]		# 375 756=36
+	pxor	@t[5], @x[6]		# 64 5		[7]
+	pxor	@t[2], @x[3]		# 36 2
+	pxor	@t[4], @x[3]		# 362 4		[5]
+	pshufd	\$0x93, @t[5], @t[5]
+___
+					my @y = @x[7,5,0,2,1,3,4,6];
+$code.=<<___;
+	# multiplication by 0x0b
+	pxor	@y[0], @y[1]
+	pxor	@t[0], @y[0]
+	pxor	@t[1], @y[1]
+	pshufd	\$0x93, @t[2], @t[2]
+	pxor	@t[5], @y[0]
+	pxor	@t[6], @y[1]
+	pxor	@t[7], @y[0]
+	pshufd	\$0x93, @t[4], @t[4]
+	pxor	@t[6], @t[7]		# clobber t[7]
+	pxor	@y[0], @y[1]
+
+	pxor	@t[0], @y[3]
+	pshufd	\$0x93, @t[0], @t[0]
+	pxor	@t[1], @y[2]
+	pxor	@t[1], @y[4]
+	pxor	@t[2], @y[2]
+	pshufd	\$0x93, @t[1], @t[1]
+	pxor	@t[2], @y[3]
+	pxor	@t[2], @y[5]
+	pxor	@t[7], @y[2]
+	pshufd	\$0x93, @t[2], @t[2]
+	pxor	@t[3], @y[3]
+	pxor	@t[3], @y[6]
+	pxor	@t[3], @y[4]
+	pshufd	\$0x93, @t[3], @t[3]
+	pxor	@t[4], @y[7]
+	pxor	@t[4], @y[5]
+	pxor	@t[7], @y[7]
+	pxor	@t[5], @y[3]
+	pxor	@t[4], @y[4]
+	pxor	@t[5], @t[7]		# clobber t[7] even more
+
+	pxor	@t[7], @y[5]
+	pshufd	\$0x93, @t[4], @t[4]
+	pxor	@t[7], @y[6]
+	pxor	@t[7], @y[4]
+
+	pxor	@t[5], @t[7]
+	pshufd	\$0x93, @t[5], @t[5]
+	pxor	@t[6], @t[7]		# restore t[7]
+
+	# multiplication by 0x0d
+	pxor	@y[7], @y[4]
+	pxor	@t[4], @y[7]
+	pshufd	\$0x93, @t[6], @t[6]
+	pxor	@t[0], @y[2]
+	pxor	@t[5], @y[7]
+	pxor	@t[2], @y[2]
+	pshufd	\$0x93, @t[7], @t[7]
+
+	pxor	@y[1], @y[3]
+	pxor	@t[1], @y[1]
+	pxor	@t[0], @y[0]
+	pxor	@t[0], @y[3]
+	pxor	@t[5], @y[1]
+	pxor	@t[5], @y[0]
+	pxor	@t[7], @y[1]
+	pshufd	\$0x93, @t[0], @t[0]
+	pxor	@t[6], @y[0]
+	pxor	@y[1], @y[3]
+	pxor	@t[1], @y[4]
+	pshufd	\$0x93, @t[1], @t[1]
+
+	pxor	@t[7], @y[7]
+	pxor	@t[2], @y[4]
+	pxor	@t[2], @y[5]
+	pshufd	\$0x93, @t[2], @t[2]
+	pxor	@t[6], @y[2]
+	pxor	@t[3], @t[6]		# clobber t[6]
+	pxor	@y[7], @y[4]
+	pxor	@t[6], @y[3]
+
+	pxor	@t[6], @y[6]
+	pxor	@t[5], @y[5]
+	pxor	@t[4], @y[6]
+	pshufd	\$0x93, @t[4], @t[4]
+	pxor	@t[6], @y[5]
+	pxor	@t[7], @y[6]
+	pxor	@t[3], @t[6]		# restore t[6]
+
+	pshufd	\$0x93, @t[5], @t[5]
+	pshufd	\$0x93, @t[6], @t[6]
+	pshufd	\$0x93, @t[7], @t[7]
+	pshufd	\$0x93, @t[3], @t[3]
+
+	# multiplication by 0x09
+	pxor	@y[1], @y[4]
+	pxor	@y[1], @t[1]		# t[1]=y[1]
+	pxor	@t[5], @t[0]		# clobber t[0]
+	pxor	@t[5], @t[1]
+	pxor	@t[0], @y[3]
+	pxor	@y[0], @t[0]		# t[0]=y[0]
+	pxor	@t[6], @t[1]
+	pxor	@t[7], @t[6]		# clobber t[6]
+	pxor	@t[1], @y[4]
+	pxor	@t[4], @y[7]
+	pxor	@y[4], @t[4]		# t[4]=y[4]
+	pxor	@t[3], @y[6]
+	pxor	@y[3], @t[3]		# t[3]=y[3]
+	pxor	@t[2], @y[5]
+	pxor	@y[2], @t[2]		# t[2]=y[2]
+	pxor	@t[7], @t[3]
+	pxor	@y[5], @t[5]		# t[5]=y[5]
+	pxor	@t[6], @t[2]
+	pxor	@t[6], @t[5]
+	pxor	@y[6], @t[6]		# t[6]=y[6]
+	pxor	@y[7], @t[7]		# t[7]=y[7]
+
+	movdqa	@t[0],@XMM[0]
+	movdqa	@t[1],@XMM[1]
+	movdqa	@t[2],@XMM[2]
+	movdqa	@t[3],@XMM[3]
+	movdqa	@t[4],@XMM[4]
+	movdqa	@t[5],@XMM[5]
+	movdqa	@t[6],@XMM[6]
+	movdqa	@t[7],@XMM[7]
+___
+}
+
+sub InvMixColumns {
+my @x=@_[0..7];
+my @t=@_[8..15];
+
+# Thanks to Jussi Kivilinna for providing pointer to
+#
+# | 0e 0b 0d 09 |   | 02 03 01 01 |   | 05 00 04 00 |
+# | 09 0e 0b 0d | = | 01 02 03 01 | x | 00 05 00 04 |
+# | 0d 09 0e 0b |   | 01 01 02 03 |   | 04 00 05 00 |
+# | 0b 0d 09 0e |   | 03 01 01 02 |   | 00 04 00 05 |
+
+$code.=<<___;
+	# multiplication by 0x05-0x00-0x04-0x00
+	pshufd	\$0x4E, @x[0], @t[0]
+	pshufd	\$0x4E, @x[6], @t[6]
+	pxor	@x[0], @t[0]
+	pshufd	\$0x4E, @x[7], @t[7]
+	pxor	@x[6], @t[6]
+	pshufd	\$0x4E, @x[1], @t[1]
+	pxor	@x[7], @t[7]
+	pshufd	\$0x4E, @x[2], @t[2]
+	pxor	@x[1], @t[1]
+	pshufd	\$0x4E, @x[3], @t[3]
+	pxor	@x[2], @t[2]
+	 pxor	@t[6], @x[0]
+	 pxor	@t[6], @x[1]
+	pshufd	\$0x4E, @x[4], @t[4]
+	pxor	@x[3], @t[3]
+	 pxor	@t[0], @x[2]
+	 pxor	@t[1], @x[3]
+	pshufd	\$0x4E, @x[5], @t[5]
+	pxor	@x[4], @t[4]
+	 pxor	@t[7], @x[1]
+	 pxor	@t[2], @x[4]
+	pxor	@x[5], @t[5]
+
+	 pxor	@t[7], @x[2]
+	 pxor	@t[6], @x[3]
+	 pxor	@t[6], @x[4]
+	 pxor	@t[3], @x[5]
+	 pxor	@t[4], @x[6]
+	 pxor	@t[7], @x[4]
+	 pxor	@t[7], @x[5]
+	 pxor	@t[5], @x[7]
+___
+	&MixColumns	(@x,@t,1);	# flipped 2<->3 and 4<->6
+}
+
+sub aesenc {				# not used
+my @b=@_[0..7];
+my @t=@_[8..15];
+$code.=<<___;
+	movdqa	0x30($const),@t[0]	# .LSR
+___
+	&ShiftRows	(@b,@t[0]);
+	&Sbox		(@b,@t);
+	&MixColumns	(@b[0,1,4,6,3,7,2,5],@t);
+}
+
+sub aesenclast {			# not used
+my @b=@_[0..7];
+my @t=@_[8..15];
+$code.=<<___;
+	movdqa	0x40($const),@t[0]	# .LSRM0
+___
+	&ShiftRows	(@b,@t[0]);
+	&Sbox		(@b,@t);
+$code.=<<___
+	pxor	0x00($key),@b[0]
+	pxor	0x10($key),@b[1]
+	pxor	0x20($key),@b[4]
+	pxor	0x30($key),@b[6]
+	pxor	0x40($key),@b[3]
+	pxor	0x50($key),@b[7]
+	pxor	0x60($key),@b[2]
+	pxor	0x70($key),@b[5]
+___
+}
+
+sub swapmove {
+my ($a,$b,$n,$mask,$t)=@_;
+$code.=<<___;
+	movdqa	$b,$t
+	psrlq	\$$n,$b
+	pxor  	$a,$b
+	pand	$mask,$b
+	pxor	$b,$a
+	psllq	\$$n,$b
+	pxor	$t,$b
+___
+}
+sub swapmove2x {
+my ($a0,$b0,$a1,$b1,$n,$mask,$t0,$t1)=@_;
+$code.=<<___;
+	movdqa	$b0,$t0
+	psrlq	\$$n,$b0
+	 movdqa	$b1,$t1
+	 psrlq	\$$n,$b1
+	pxor  	$a0,$b0
+	 pxor  	$a1,$b1
+	pand	$mask,$b0
+	 pand	$mask,$b1
+	pxor	$b0,$a0
+	psllq	\$$n,$b0
+	 pxor	$b1,$a1
+	 psllq	\$$n,$b1
+	pxor	$t0,$b0
+	 pxor	$t1,$b1
+___
+}
+
+sub bitslice {
+my @x=reverse(@_[0..7]);
+my ($t0,$t1,$t2,$t3)=@_[8..11];
+$code.=<<___;
+	movdqa	0x00($const),$t0	# .LBS0
+	movdqa	0x10($const),$t1	# .LBS1
+___
+	&swapmove2x(@x[0,1,2,3],1,$t0,$t2,$t3);
+	&swapmove2x(@x[4,5,6,7],1,$t0,$t2,$t3);
+$code.=<<___;
+	movdqa	0x20($const),$t0	# .LBS2
+___
+	&swapmove2x(@x[0,2,1,3],2,$t1,$t2,$t3);
+	&swapmove2x(@x[4,6,5,7],2,$t1,$t2,$t3);
+
+	&swapmove2x(@x[0,4,1,5],4,$t0,$t2,$t3);
+	&swapmove2x(@x[2,6,3,7],4,$t0,$t2,$t3);
+}
+
+$code.=<<___;
+.text
+
+.extern	asm_AES_encrypt
+.extern	asm_AES_decrypt
+
+.type	_bsaes_encrypt8,\@abi-omnipotent
+.align	64
+_bsaes_encrypt8:
+	lea	.LBS0(%rip), $const	# constants table
+
+	movdqa	($key), @XMM[9]		# round 0 key
+	lea	0x10($key), $key
+	movdqa	0x50($const), @XMM[8]	# .LM0SR
+	pxor	@XMM[9], @XMM[0]	# xor with round0 key
+	pxor	@XMM[9], @XMM[1]
+	pxor	@XMM[9], @XMM[2]
+	pxor	@XMM[9], @XMM[3]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
+	pxor	@XMM[9], @XMM[4]
+	pxor	@XMM[9], @XMM[5]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
+	pxor	@XMM[9], @XMM[6]
+	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
+	 pshufb	@XMM[8], @XMM[6]
+	 pshufb	@XMM[8], @XMM[7]
+_bsaes_encrypt8_bitslice:
+___
+	&bitslice	(@XMM[0..7, 8..11]);
+$code.=<<___;
+	dec	$rounds
+	jmp	.Lenc_sbox
+.align	16
+.Lenc_loop:
+___
+	&ShiftRows	(@XMM[0..7, 8]);
+$code.=".Lenc_sbox:\n";
+	&Sbox		(@XMM[0..7, 8..15]);
+$code.=<<___;
+	dec	$rounds
+	jl	.Lenc_done
+___
+	&MixColumns	(@XMM[0,1,4,6,3,7,2,5, 8..15]);
+$code.=<<___;
+	movdqa	0x30($const), @XMM[8]	# .LSR
+	jnz	.Lenc_loop
+	movdqa	0x40($const), @XMM[8]	# .LSRM0
+	jmp	.Lenc_loop
+.align	16
+.Lenc_done:
+___
+	# output in lsb > [t0, t1, t4, t6, t3, t7, t2, t5] < msb
+	&bitslice	(@XMM[0,1,4,6,3,7,2,5, 8..11]);
+$code.=<<___;
+	movdqa	($key), @XMM[8]		# last round key
+	pxor	@XMM[8], @XMM[4]
+	pxor	@XMM[8], @XMM[6]
+	pxor	@XMM[8], @XMM[3]
+	pxor	@XMM[8], @XMM[7]
+	pxor	@XMM[8], @XMM[2]
+	pxor	@XMM[8], @XMM[5]
+	pxor	@XMM[8], @XMM[0]
+	pxor	@XMM[8], @XMM[1]
+	ret
+.size	_bsaes_encrypt8,.-_bsaes_encrypt8
+
+.type	_bsaes_decrypt8,\@abi-omnipotent
+.align	64
+_bsaes_decrypt8:
+	lea	.LBS0(%rip), $const	# constants table
+
+	movdqa	($key), @XMM[9]		# round 0 key
+	lea	0x10($key), $key
+	movdqa	-0x30($const), @XMM[8]	# .LM0ISR
+	pxor	@XMM[9], @XMM[0]	# xor with round0 key
+	pxor	@XMM[9], @XMM[1]
+	pxor	@XMM[9], @XMM[2]
+	pxor	@XMM[9], @XMM[3]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
+	pxor	@XMM[9], @XMM[4]
+	pxor	@XMM[9], @XMM[5]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
+	pxor	@XMM[9], @XMM[6]
+	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
+	 pshufb	@XMM[8], @XMM[6]
+	 pshufb	@XMM[8], @XMM[7]
+___
+	&bitslice	(@XMM[0..7, 8..11]);
+$code.=<<___;
+	dec	$rounds
+	jmp	.Ldec_sbox
+.align	16
+.Ldec_loop:
+___
+	&ShiftRows	(@XMM[0..7, 8]);
+$code.=".Ldec_sbox:\n";
+	&InvSbox	(@XMM[0..7, 8..15]);
+$code.=<<___;
+	dec	$rounds
+	jl	.Ldec_done
+___
+	&InvMixColumns	(@XMM[0,1,6,4,2,7,3,5, 8..15]);
+$code.=<<___;
+	movdqa	-0x10($const), @XMM[8]	# .LISR
+	jnz	.Ldec_loop
+	movdqa	-0x20($const), @XMM[8]	# .LISRM0
+	jmp	.Ldec_loop
+.align	16
+.Ldec_done:
+___
+	&bitslice	(@XMM[0,1,6,4,2,7,3,5, 8..11]);
+$code.=<<___;
+	movdqa	($key), @XMM[8]		# last round key
+	pxor	@XMM[8], @XMM[6]
+	pxor	@XMM[8], @XMM[4]
+	pxor	@XMM[8], @XMM[2]
+	pxor	@XMM[8], @XMM[7]
+	pxor	@XMM[8], @XMM[3]
+	pxor	@XMM[8], @XMM[5]
+	pxor	@XMM[8], @XMM[0]
+	pxor	@XMM[8], @XMM[1]
+	ret
+.size	_bsaes_decrypt8,.-_bsaes_decrypt8
+___
+}
+{
+my ($out,$inp,$rounds,$const)=("%rax","%rcx","%r10d","%r11");
+
+sub bitslice_key {
+my @x=reverse(@_[0..7]);
+my ($bs0,$bs1,$bs2,$t2,$t3)=@_[8..12];
+
+	&swapmove	(@x[0,1],1,$bs0,$t2,$t3);
+$code.=<<___;
+	#&swapmove(@x[2,3],1,$t0,$t2,$t3);
+	movdqa	@x[0], @x[2]
+	movdqa	@x[1], @x[3]
+___
+	#&swapmove2x(@x[4,5,6,7],1,$t0,$t2,$t3);
+
+	&swapmove2x	(@x[0,2,1,3],2,$bs1,$t2,$t3);
+$code.=<<___;
+	#&swapmove2x(@x[4,6,5,7],2,$t1,$t2,$t3);
+	movdqa	@x[0], @x[4]
+	movdqa	@x[2], @x[6]
+	movdqa	@x[1], @x[5]
+	movdqa	@x[3], @x[7]
+___
+	&swapmove2x	(@x[0,4,1,5],4,$bs2,$t2,$t3);
+	&swapmove2x	(@x[2,6,3,7],4,$bs2,$t2,$t3);
+}
+
+$code.=<<___;
+.type	_bsaes_key_convert,\@abi-omnipotent
+.align	16
+_bsaes_key_convert:
+	lea	.Lmasks(%rip), $const
+	movdqu	($inp), %xmm7		# load round 0 key
+	lea	0x10($inp), $inp
+	movdqa	0x00($const), %xmm0	# 0x01...
+	movdqa	0x10($const), %xmm1	# 0x02...
+	movdqa	0x20($const), %xmm2	# 0x04...
+	movdqa	0x30($const), %xmm3	# 0x08...
+	movdqa	0x40($const), %xmm4	# .LM0
+	pcmpeqd	%xmm5, %xmm5		# .LNOT
+
+	movdqu	($inp), %xmm6		# load round 1 key
+	movdqa	%xmm7, ($out)		# save round 0 key
+	lea	0x10($out), $out
+	dec	$rounds
+	jmp	.Lkey_loop
+.align	16
+.Lkey_loop:
+	pshufb	%xmm4, %xmm6		# .LM0
+
+	movdqa	%xmm0,	%xmm8
+	movdqa	%xmm1,	%xmm9
+
+	pand	%xmm6,	%xmm8
+	pand	%xmm6,	%xmm9
+	movdqa	%xmm2,	%xmm10
+	pcmpeqb	%xmm0,	%xmm8
+	psllq	\$4,	%xmm0		# 0x10...
+	movdqa	%xmm3,	%xmm11
+	pcmpeqb	%xmm1,	%xmm9
+	psllq	\$4,	%xmm1		# 0x20...
+
+	pand	%xmm6,	%xmm10
+	pand	%xmm6,	%xmm11
+	movdqa	%xmm0,	%xmm12
+	pcmpeqb	%xmm2,	%xmm10
+	psllq	\$4,	%xmm2		# 0x40...
+	movdqa	%xmm1,	%xmm13
+	pcmpeqb	%xmm3,	%xmm11
+	psllq	\$4,	%xmm3		# 0x80...
+
+	movdqa	%xmm2,	%xmm14
+	movdqa	%xmm3,	%xmm15
+	 pxor	%xmm5,	%xmm8		# "pnot"
+	 pxor	%xmm5,	%xmm9
+
+	pand	%xmm6,	%xmm12
+	pand	%xmm6,	%xmm13
+	 movdqa	%xmm8, 0x00($out)	# write bit-sliced round key
+	pcmpeqb	%xmm0,	%xmm12
+	psrlq	\$4,	%xmm0		# 0x01...
+	 movdqa	%xmm9, 0x10($out)
+	pcmpeqb	%xmm1,	%xmm13
+	psrlq	\$4,	%xmm1		# 0x02...
+	 lea	0x10($inp), $inp
+
+	pand	%xmm6,	%xmm14
+	pand	%xmm6,	%xmm15
+	 movdqa	%xmm10, 0x20($out)
+	pcmpeqb	%xmm2,	%xmm14
+	psrlq	\$4,	%xmm2		# 0x04...
+	 movdqa	%xmm11, 0x30($out)
+	pcmpeqb	%xmm3,	%xmm15
+	psrlq	\$4,	%xmm3		# 0x08...
+	 movdqu	($inp), %xmm6		# load next round key
+
+	pxor	%xmm5, %xmm13		# "pnot"
+	pxor	%xmm5, %xmm14
+	movdqa	%xmm12, 0x40($out)
+	movdqa	%xmm13, 0x50($out)
+	movdqa	%xmm14, 0x60($out)
+	movdqa	%xmm15, 0x70($out)
+	lea	0x80($out),$out
+	dec	$rounds
+	jnz	.Lkey_loop
+
+	movdqa	0x50($const), %xmm7	# .L63
+	#movdqa	%xmm6, ($out)		# don't save last round key
+	ret
+.size	_bsaes_key_convert,.-_bsaes_key_convert
+___
+}
+
+if (0 && !$win64) {	# following four functions are unsupported interface
+			# used for benchmarking...
+$code.=<<___;
+.globl	bsaes_enc_key_convert
+.type	bsaes_enc_key_convert,\@function,2
+.align	16
+bsaes_enc_key_convert:
+	mov	240($inp),%r10d		# pass rounds
+	mov	$inp,%rcx		# pass key
+	mov	$out,%rax		# pass key schedule
+	call	_bsaes_key_convert
+	pxor	%xmm6,%xmm7		# fix up last round key
+	movdqa	%xmm7,(%rax)		# save last round key
+	ret
+.size	bsaes_enc_key_convert,.-bsaes_enc_key_convert
+
+.globl	bsaes_encrypt_128
+.type	bsaes_encrypt_128,\@function,4
+.align	16
+bsaes_encrypt_128:
+.Lenc128_loop:
+	movdqu	0x00($inp), @XMM[0]	# load input
+	movdqu	0x10($inp), @XMM[1]
+	movdqu	0x20($inp), @XMM[2]
+	movdqu	0x30($inp), @XMM[3]
+	movdqu	0x40($inp), @XMM[4]
+	movdqu	0x50($inp), @XMM[5]
+	movdqu	0x60($inp), @XMM[6]
+	movdqu	0x70($inp), @XMM[7]
+	mov	$key, %rax		# pass the $key
+	lea	0x80($inp), $inp
+	mov	\$10,%r10d
+
+	call	_bsaes_encrypt8
+
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[2], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	sub	\$0x80,$len
+	ja	.Lenc128_loop
+	ret
+.size	bsaes_encrypt_128,.-bsaes_encrypt_128
+
+.globl	bsaes_dec_key_convert
+.type	bsaes_dec_key_convert,\@function,2
+.align	16
+bsaes_dec_key_convert:
+	mov	240($inp),%r10d		# pass rounds
+	mov	$inp,%rcx		# pass key
+	mov	$out,%rax		# pass key schedule
+	call	_bsaes_key_convert
+	pxor	($out),%xmm7		# fix up round 0 key
+	movdqa	%xmm6,(%rax)		# save last round key
+	movdqa	%xmm7,($out)
+	ret
+.size	bsaes_dec_key_convert,.-bsaes_dec_key_convert
+
+.globl	bsaes_decrypt_128
+.type	bsaes_decrypt_128,\@function,4
+.align	16
+bsaes_decrypt_128:
+.Ldec128_loop:
+	movdqu	0x00($inp), @XMM[0]	# load input
+	movdqu	0x10($inp), @XMM[1]
+	movdqu	0x20($inp), @XMM[2]
+	movdqu	0x30($inp), @XMM[3]
+	movdqu	0x40($inp), @XMM[4]
+	movdqu	0x50($inp), @XMM[5]
+	movdqu	0x60($inp), @XMM[6]
+	movdqu	0x70($inp), @XMM[7]
+	mov	$key, %rax		# pass the $key
+	lea	0x80($inp), $inp
+	mov	\$10,%r10d
+
+	call	_bsaes_decrypt8
+
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	sub	\$0x80,$len
+	ja	.Ldec128_loop
+	ret
+.size	bsaes_decrypt_128,.-bsaes_decrypt_128
+___
+}
+{
+######################################################################
+#
+# OpenSSL interface
+#
+my ($arg1,$arg2,$arg3,$arg4,$arg5,$arg6)=$win64	? ("%rcx","%rdx","%r8","%r9","%r10","%r11d")
+						: ("%rdi","%rsi","%rdx","%rcx","%r8","%r9d");
+my ($inp,$out,$len,$key)=("%r12","%r13","%r14","%r15");
+
+if ($ecb) {
+$code.=<<___;
+.globl	bsaes_ecb_encrypt_blocks
+.type	bsaes_ecb_encrypt_blocks,\@abi-omnipotent
+.align	16
+bsaes_ecb_encrypt_blocks:
+	mov	%rsp, %rax
+.Lecb_enc_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp),%rsp
+___
+$code.=<<___ if ($win64);
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lecb_enc_body:
+___
+$code.=<<___;
+	mov	%rsp,%rbp		# backup %rsp
+	mov	240($arg4),%eax		# rounds
+	mov	$arg1,$inp		# backup arguments
+	mov	$arg2,$out
+	mov	$arg3,$len
+	mov	$arg4,$key
+	cmp	\$8,$arg3
+	jb	.Lecb_enc_short
+
+	mov	%eax,%ebx		# backup rounds
+	shl	\$7,%rax		# 128 bytes per inner round key
+	sub	\$`128-32`,%rax		# size of bit-sliced key schedule
+	sub	%rax,%rsp
+	mov	%rsp,%rax		# pass key schedule
+	mov	$key,%rcx		# pass key
+	mov	%ebx,%r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	%xmm6,%xmm7		# fix up last round key
+	movdqa	%xmm7,(%rax)		# save last round key
+
+	sub	\$8,$len
+.Lecb_enc_loop:
+	movdqu	0x00($inp), @XMM[0]	# load input
+	movdqu	0x10($inp), @XMM[1]
+	movdqu	0x20($inp), @XMM[2]
+	movdqu	0x30($inp), @XMM[3]
+	movdqu	0x40($inp), @XMM[4]
+	movdqu	0x50($inp), @XMM[5]
+	mov	%rsp, %rax		# pass key schedule
+	movdqu	0x60($inp), @XMM[6]
+	mov	%ebx,%r10d		# pass rounds
+	movdqu	0x70($inp), @XMM[7]
+	lea	0x80($inp), $inp
+
+	call	_bsaes_encrypt8
+
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[2], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	sub	\$8,$len
+	jnc	.Lecb_enc_loop
+
+	add	\$8,$len
+	jz	.Lecb_enc_done
+
+	movdqu	0x00($inp), @XMM[0]	# load input
+	mov	%rsp, %rax		# pass key schedule
+	mov	%ebx,%r10d		# pass rounds
+	cmp	\$2,$len
+	jb	.Lecb_enc_one
+	movdqu	0x10($inp), @XMM[1]
+	je	.Lecb_enc_two
+	movdqu	0x20($inp), @XMM[2]
+	cmp	\$4,$len
+	jb	.Lecb_enc_three
+	movdqu	0x30($inp), @XMM[3]
+	je	.Lecb_enc_four
+	movdqu	0x40($inp), @XMM[4]
+	cmp	\$6,$len
+	jb	.Lecb_enc_five
+	movdqu	0x50($inp), @XMM[5]
+	je	.Lecb_enc_six
+	movdqu	0x60($inp), @XMM[6]
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[2], 0x60($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_six:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_five:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_four:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_three:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_two:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_one:
+	call	_bsaes_encrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	jmp	.Lecb_enc_done
+.align	16
+.Lecb_enc_short:
+	lea	($inp), $arg1
+	lea	($out), $arg2
+	lea	($key), $arg3
+	call	asm_AES_encrypt
+	lea	16($inp), $inp
+	lea	16($out), $out
+	dec	$len
+	jnz	.Lecb_enc_short
+
+.Lecb_enc_done:
+	lea	(%rsp),%rax
+	pxor	%xmm0, %xmm0
+.Lecb_enc_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	jb	.Lecb_enc_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lecb_enc_epilogue:
+	ret
+.size	bsaes_ecb_encrypt_blocks,.-bsaes_ecb_encrypt_blocks
+
+.globl	bsaes_ecb_decrypt_blocks
+.type	bsaes_ecb_decrypt_blocks,\@abi-omnipotent
+.align	16
+bsaes_ecb_decrypt_blocks:
+	mov	%rsp, %rax
+.Lecb_dec_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp),%rsp
+___
+$code.=<<___ if ($win64);
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lecb_dec_body:
+___
+$code.=<<___;
+	mov	%rsp,%rbp		# backup %rsp
+	mov	240($arg4),%eax		# rounds
+	mov	$arg1,$inp		# backup arguments
+	mov	$arg2,$out
+	mov	$arg3,$len
+	mov	$arg4,$key
+	cmp	\$8,$arg3
+	jb	.Lecb_dec_short
+
+	mov	%eax,%ebx		# backup rounds
+	shl	\$7,%rax		# 128 bytes per inner round key
+	sub	\$`128-32`,%rax		# size of bit-sliced key schedule
+	sub	%rax,%rsp
+	mov	%rsp,%rax		# pass key schedule
+	mov	$key,%rcx		# pass key
+	mov	%ebx,%r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	(%rsp),%xmm7		# fix up 0 round key
+	movdqa	%xmm6,(%rax)		# save last round key
+	movdqa	%xmm7,(%rsp)
+
+	sub	\$8,$len
+.Lecb_dec_loop:
+	movdqu	0x00($inp), @XMM[0]	# load input
+	movdqu	0x10($inp), @XMM[1]
+	movdqu	0x20($inp), @XMM[2]
+	movdqu	0x30($inp), @XMM[3]
+	movdqu	0x40($inp), @XMM[4]
+	movdqu	0x50($inp), @XMM[5]
+	mov	%rsp, %rax		# pass key schedule
+	movdqu	0x60($inp), @XMM[6]
+	mov	%ebx,%r10d		# pass rounds
+	movdqu	0x70($inp), @XMM[7]
+	lea	0x80($inp), $inp
+
+	call	_bsaes_decrypt8
+
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	sub	\$8,$len
+	jnc	.Lecb_dec_loop
+
+	add	\$8,$len
+	jz	.Lecb_dec_done
+
+	movdqu	0x00($inp), @XMM[0]	# load input
+	mov	%rsp, %rax		# pass key schedule
+	mov	%ebx,%r10d		# pass rounds
+	cmp	\$2,$len
+	jb	.Lecb_dec_one
+	movdqu	0x10($inp), @XMM[1]
+	je	.Lecb_dec_two
+	movdqu	0x20($inp), @XMM[2]
+	cmp	\$4,$len
+	jb	.Lecb_dec_three
+	movdqu	0x30($inp), @XMM[3]
+	je	.Lecb_dec_four
+	movdqu	0x40($inp), @XMM[4]
+	cmp	\$6,$len
+	jb	.Lecb_dec_five
+	movdqu	0x50($inp), @XMM[5]
+	je	.Lecb_dec_six
+	movdqu	0x60($inp), @XMM[6]
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_six:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_five:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_four:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_three:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_two:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_one:
+	call	_bsaes_decrypt8
+	movdqu	@XMM[0], 0x00($out)	# write output
+	jmp	.Lecb_dec_done
+.align	16
+.Lecb_dec_short:
+	lea	($inp), $arg1
+	lea	($out), $arg2
+	lea	($key), $arg3
+	call	asm_AES_decrypt
+	lea	16($inp), $inp
+	lea	16($out), $out
+	dec	$len
+	jnz	.Lecb_dec_short
+
+.Lecb_dec_done:
+	lea	(%rsp),%rax
+	pxor	%xmm0, %xmm0
+.Lecb_dec_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	jb	.Lecb_dec_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lecb_dec_epilogue:
+	ret
+.size	bsaes_ecb_decrypt_blocks,.-bsaes_ecb_decrypt_blocks
+___
+}
+$code.=<<___;
+.extern	asm_AES_cbc_encrypt
+.globl	bsaes_cbc_encrypt
+.type	bsaes_cbc_encrypt,\@abi-omnipotent
+.align	16
+bsaes_cbc_encrypt:
+___
+$code.=<<___ if ($win64);
+	mov	48(%rsp),$arg6		# pull direction flag
+___
+$code.=<<___;
+	cmp	\$0,$arg6
+	jne	asm_AES_cbc_encrypt
+	cmp	\$128,$arg3
+	jb	asm_AES_cbc_encrypt
+
+	mov	%rsp, %rax
+.Lcbc_dec_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp), %rsp
+___
+$code.=<<___ if ($win64);
+	mov	0xa0(%rsp),$arg5	# pull ivp
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lcbc_dec_body:
+___
+$code.=<<___;
+	mov	%rsp, %rbp		# backup %rsp
+	mov	240($arg4), %eax	# rounds
+	mov	$arg1, $inp		# backup arguments
+	mov	$arg2, $out
+	mov	$arg3, $len
+	mov	$arg4, $key
+	mov	$arg5, %rbx
+	shr	\$4, $len		# bytes to blocks
+
+	mov	%eax, %edx		# rounds
+	shl	\$7, %rax		# 128 bytes per inner round key
+	sub	\$`128-32`, %rax	# size of bit-sliced key schedule
+	sub	%rax, %rsp
+
+	mov	%rsp, %rax		# pass key schedule
+	mov	$key, %rcx		# pass key
+	mov	%edx, %r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	(%rsp),%xmm7		# fix up 0 round key
+	movdqa	%xmm6,(%rax)		# save last round key
+	movdqa	%xmm7,(%rsp)
+
+	movdqu	(%rbx), @XMM[15]	# load IV
+	sub	\$8,$len
+.Lcbc_dec_loop:
+	movdqu	0x00($inp), @XMM[0]	# load input
+	movdqu	0x10($inp), @XMM[1]
+	movdqu	0x20($inp), @XMM[2]
+	movdqu	0x30($inp), @XMM[3]
+	movdqu	0x40($inp), @XMM[4]
+	movdqu	0x50($inp), @XMM[5]
+	mov	%rsp, %rax		# pass key schedule
+	movdqu	0x60($inp), @XMM[6]
+	mov	%edx,%r10d		# pass rounds
+	movdqu	0x70($inp), @XMM[7]
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+
+	call	_bsaes_decrypt8
+
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[9], @XMM[6]
+	movdqu	0x30($inp), @XMM[11]
+	pxor	@XMM[10], @XMM[4]
+	movdqu	0x40($inp), @XMM[12]
+	pxor	@XMM[11], @XMM[2]
+	movdqu	0x50($inp), @XMM[13]
+	pxor	@XMM[12], @XMM[7]
+	movdqu	0x60($inp), @XMM[14]
+	pxor	@XMM[13], @XMM[3]
+	movdqu	0x70($inp), @XMM[15]	# IV
+	pxor	@XMM[14], @XMM[5]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	lea	0x80($inp), $inp
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	sub	\$8,$len
+	jnc	.Lcbc_dec_loop
+
+	add	\$8,$len
+	jz	.Lcbc_dec_done
+
+	movdqu	0x00($inp), @XMM[0]	# load input
+	mov	%rsp, %rax		# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+	cmp	\$2,$len
+	jb	.Lcbc_dec_one
+	movdqu	0x10($inp), @XMM[1]
+	je	.Lcbc_dec_two
+	movdqu	0x20($inp), @XMM[2]
+	cmp	\$4,$len
+	jb	.Lcbc_dec_three
+	movdqu	0x30($inp), @XMM[3]
+	je	.Lcbc_dec_four
+	movdqu	0x40($inp), @XMM[4]
+	cmp	\$6,$len
+	jb	.Lcbc_dec_five
+	movdqu	0x50($inp), @XMM[5]
+	je	.Lcbc_dec_six
+	movdqu	0x60($inp), @XMM[6]
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[9], @XMM[6]
+	movdqu	0x30($inp), @XMM[11]
+	pxor	@XMM[10], @XMM[4]
+	movdqu	0x40($inp), @XMM[12]
+	pxor	@XMM[11], @XMM[2]
+	movdqu	0x50($inp), @XMM[13]
+	pxor	@XMM[12], @XMM[7]
+	movdqu	0x60($inp), @XMM[15]	# IV
+	pxor	@XMM[13], @XMM[3]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_six:
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[9], @XMM[6]
+	movdqu	0x30($inp), @XMM[11]
+	pxor	@XMM[10], @XMM[4]
+	movdqu	0x40($inp), @XMM[12]
+	pxor	@XMM[11], @XMM[2]
+	movdqu	0x50($inp), @XMM[15]	# IV
+	pxor	@XMM[12], @XMM[7]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_five:
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[9], @XMM[6]
+	movdqu	0x30($inp), @XMM[11]
+	pxor	@XMM[10], @XMM[4]
+	movdqu	0x40($inp), @XMM[15]	# IV
+	pxor	@XMM[11], @XMM[2]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_four:
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[9], @XMM[6]
+	movdqu	0x30($inp), @XMM[15]	# IV
+	pxor	@XMM[10], @XMM[4]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_three:
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[8], @XMM[1]
+	movdqu	0x20($inp), @XMM[15]	# IV
+	pxor	@XMM[9], @XMM[6]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_two:
+	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
+	call	_bsaes_decrypt8
+	pxor	0x20(%rbp), @XMM[0]	# ^= IV
+	movdqu	0x00($inp), @XMM[8]	# re-load input
+	movdqu	0x10($inp), @XMM[15]	# IV
+	pxor	@XMM[8], @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	jmp	.Lcbc_dec_done
+.align	16
+.Lcbc_dec_one:
+	lea	($inp), $arg1
+	lea	0x20(%rbp), $arg2	# buffer output
+	lea	($key), $arg3
+	call	asm_AES_decrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[15]	# ^= IV
+	movdqu	@XMM[15], ($out)	# write output
+	movdqa	@XMM[0], @XMM[15]	# IV
+
+.Lcbc_dec_done:
+	movdqu	@XMM[15], (%rbx)	# return IV
+	lea	(%rsp), %rax
+	pxor	%xmm0, %xmm0
+.Lcbc_dec_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	ja	.Lcbc_dec_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lcbc_dec_epilogue:
+	ret
+.size	bsaes_cbc_encrypt,.-bsaes_cbc_encrypt
+
+.globl	bsaes_ctr32_encrypt_blocks
+.type	bsaes_ctr32_encrypt_blocks,\@abi-omnipotent
+.align	16
+bsaes_ctr32_encrypt_blocks:
+	mov	%rsp, %rax
+.Lctr_enc_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp), %rsp
+___
+$code.=<<___ if ($win64);
+	mov	0xa0(%rsp),$arg5	# pull ivp
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lctr_enc_body:
+___
+$code.=<<___;
+	mov	%rsp, %rbp		# backup %rsp
+	movdqu	($arg5), %xmm0		# load counter
+	mov	240($arg4), %eax	# rounds
+	mov	$arg1, $inp		# backup arguments
+	mov	$arg2, $out
+	mov	$arg3, $len
+	mov	$arg4, $key
+	movdqa	%xmm0, 0x20(%rbp)	# copy counter
+	cmp	\$8, $arg3
+	jb	.Lctr_enc_short
+
+	mov	%eax, %ebx		# rounds
+	shl	\$7, %rax		# 128 bytes per inner round key
+	sub	\$`128-32`, %rax	# size of bit-sliced key schedule
+	sub	%rax, %rsp
+
+	mov	%rsp, %rax		# pass key schedule
+	mov	$key, %rcx		# pass key
+	mov	%ebx, %r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	%xmm6,%xmm7		# fix up last round key
+	movdqa	%xmm7,(%rax)		# save last round key
+
+	movdqa	(%rsp), @XMM[9]		# load round0 key
+	lea	.LADD1(%rip), %r11
+	movdqa	0x20(%rbp), @XMM[0]	# counter copy
+	movdqa	-0x20(%r11), @XMM[8]	# .LSWPUP
+	pshufb	@XMM[8], @XMM[9]	# byte swap upper part
+	pshufb	@XMM[8], @XMM[0]
+	movdqa	@XMM[9], (%rsp)		# save adjusted round0 key
+	jmp	.Lctr_enc_loop
+.align	16
+.Lctr_enc_loop:
+	movdqa	@XMM[0], 0x20(%rbp)	# save counter
+	movdqa	@XMM[0], @XMM[1]	# prepare 8 counter values
+	movdqa	@XMM[0], @XMM[2]
+	paddd	0x00(%r11), @XMM[1]	# .LADD1
+	movdqa	@XMM[0], @XMM[3]
+	paddd	0x10(%r11), @XMM[2]	# .LADD2
+	movdqa	@XMM[0], @XMM[4]
+	paddd	0x20(%r11), @XMM[3]	# .LADD3
+	movdqa	@XMM[0], @XMM[5]
+	paddd	0x30(%r11), @XMM[4]	# .LADD4
+	movdqa	@XMM[0], @XMM[6]
+	paddd	0x40(%r11), @XMM[5]	# .LADD5
+	movdqa	@XMM[0], @XMM[7]
+	paddd	0x50(%r11), @XMM[6]	# .LADD6
+	paddd	0x60(%r11), @XMM[7]	# .LADD7
+
+	# Borrow prologue from _bsaes_encrypt8 to use the opportunity
+	# to flip byte order in 32-bit counter
+	movdqa	(%rsp), @XMM[9]		# round 0 key
+	lea	0x10(%rsp), %rax	# pass key schedule
+	movdqa	-0x10(%r11), @XMM[8]	# .LSWPUPM0SR
+	pxor	@XMM[9], @XMM[0]	# xor with round0 key
+	pxor	@XMM[9], @XMM[1]
+	pxor	@XMM[9], @XMM[2]
+	pxor	@XMM[9], @XMM[3]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
+	pxor	@XMM[9], @XMM[4]
+	pxor	@XMM[9], @XMM[5]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
+	pxor	@XMM[9], @XMM[6]
+	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
+	 pshufb	@XMM[8], @XMM[6]
+	 pshufb	@XMM[8], @XMM[7]
+	lea	.LBS0(%rip), %r11	# constants table
+	mov	%ebx,%r10d		# pass rounds
+
+	call	_bsaes_encrypt8_bitslice
+
+	sub	\$8,$len
+	jc	.Lctr_enc_loop_done
+
+	movdqu	0x00($inp), @XMM[8]	# load input
+	movdqu	0x10($inp), @XMM[9]
+	movdqu	0x20($inp), @XMM[10]
+	movdqu	0x30($inp), @XMM[11]
+	movdqu	0x40($inp), @XMM[12]
+	movdqu	0x50($inp), @XMM[13]
+	movdqu	0x60($inp), @XMM[14]
+	movdqu	0x70($inp), @XMM[15]
+	lea	0x80($inp),$inp
+	pxor	@XMM[0], @XMM[8]
+	movdqa	0x20(%rbp), @XMM[0]	# load counter
+	pxor	@XMM[9], @XMM[1]
+	movdqu	@XMM[8], 0x00($out)	# write output
+	pxor	@XMM[10], @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	@XMM[11], @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	pxor	@XMM[12], @XMM[3]
+	movdqu	@XMM[6], 0x30($out)
+	pxor	@XMM[13], @XMM[7]
+	movdqu	@XMM[3], 0x40($out)
+	pxor	@XMM[14], @XMM[2]
+	movdqu	@XMM[7], 0x50($out)
+	pxor	@XMM[15], @XMM[5]
+	movdqu	@XMM[2], 0x60($out)
+	lea	.LADD1(%rip), %r11
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+	paddd	0x70(%r11), @XMM[0]	# .LADD8
+	jnz	.Lctr_enc_loop
+
+	jmp	.Lctr_enc_done
+.align	16
+.Lctr_enc_loop_done:
+	add	\$8, $len
+	movdqu	0x00($inp), @XMM[8]	# load input
+	pxor	@XMM[8], @XMM[0]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	cmp	\$2,$len
+	jb	.Lctr_enc_done
+	movdqu	0x10($inp), @XMM[9]
+	pxor	@XMM[9], @XMM[1]
+	movdqu	@XMM[1], 0x10($out)
+	je	.Lctr_enc_done
+	movdqu	0x20($inp), @XMM[10]
+	pxor	@XMM[10], @XMM[4]
+	movdqu	@XMM[4], 0x20($out)
+	cmp	\$4,$len
+	jb	.Lctr_enc_done
+	movdqu	0x30($inp), @XMM[11]
+	pxor	@XMM[11], @XMM[6]
+	movdqu	@XMM[6], 0x30($out)
+	je	.Lctr_enc_done
+	movdqu	0x40($inp), @XMM[12]
+	pxor	@XMM[12], @XMM[3]
+	movdqu	@XMM[3], 0x40($out)
+	cmp	\$6,$len
+	jb	.Lctr_enc_done
+	movdqu	0x50($inp), @XMM[13]
+	pxor	@XMM[13], @XMM[7]
+	movdqu	@XMM[7], 0x50($out)
+	je	.Lctr_enc_done
+	movdqu	0x60($inp), @XMM[14]
+	pxor	@XMM[14], @XMM[2]
+	movdqu	@XMM[2], 0x60($out)
+	jmp	.Lctr_enc_done
+
+.align	16
+.Lctr_enc_short:
+	lea	0x20(%rbp), $arg1
+	lea	0x30(%rbp), $arg2
+	lea	($key), $arg3
+	call	asm_AES_encrypt
+	movdqu	($inp), @XMM[1]
+	lea	16($inp), $inp
+	mov	0x2c(%rbp), %eax	# load 32-bit counter
+	bswap	%eax
+	pxor	0x30(%rbp), @XMM[1]
+	inc	%eax			# increment
+	movdqu	@XMM[1], ($out)
+	bswap	%eax
+	lea	16($out), $out
+	mov	%eax, 0x2c(%rsp)	# save 32-bit counter
+	dec	$len
+	jnz	.Lctr_enc_short
+
+.Lctr_enc_done:
+	lea	(%rsp), %rax
+	pxor	%xmm0, %xmm0
+.Lctr_enc_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	ja	.Lctr_enc_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lctr_enc_epilogue:
+	ret
+.size	bsaes_ctr32_encrypt_blocks,.-bsaes_ctr32_encrypt_blocks
+___
+######################################################################
+# void bsaes_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 ($twmask,$twres,$twtmp)=@XMM[13..15];
+$arg6=~s/d$//;
+
+$code.=<<___;
+.globl	bsaes_xts_encrypt
+.type	bsaes_xts_encrypt,\@abi-omnipotent
+.align	16
+bsaes_xts_encrypt:
+	mov	%rsp, %rax
+.Lxts_enc_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp), %rsp
+___
+$code.=<<___ if ($win64);
+	mov	0xa0(%rsp),$arg5	# pull key2
+	mov	0xa8(%rsp),$arg6	# pull ivp
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lxts_enc_body:
+___
+$code.=<<___;
+	mov	%rsp, %rbp		# backup %rsp
+	mov	$arg1, $inp		# backup arguments
+	mov	$arg2, $out
+	mov	$arg3, $len
+	mov	$arg4, $key
+
+	lea	($arg6), $arg1
+	lea	0x20(%rbp), $arg2
+	lea	($arg5), $arg3
+	call	asm_AES_encrypt		# generate initial tweak
+
+	mov	240($key), %eax		# rounds
+	mov	$len, %rbx		# backup $len
+
+	mov	%eax, %edx		# rounds
+	shl	\$7, %rax		# 128 bytes per inner round key
+	sub	\$`128-32`, %rax	# size of bit-sliced key schedule
+	sub	%rax, %rsp
+
+	mov	%rsp, %rax		# pass key schedule
+	mov	$key, %rcx		# pass key
+	mov	%edx, %r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	%xmm6, %xmm7		# fix up last round key
+	movdqa	%xmm7, (%rax)		# save last round key
+
+	and	\$-16, $len
+	sub	\$0x80, %rsp		# place for tweak[8]
+	movdqa	0x20(%rbp), @XMM[7]	# initial tweak
+
+	pxor	$twtmp, $twtmp
+	movdqa	.Lxts_magic(%rip), $twmask
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+
+	sub	\$0x80, $len
+	jc	.Lxts_enc_short
+	jmp	.Lxts_enc_loop
+
+.align	16
+.Lxts_enc_loop:
+___
+    for ($i=0;$i<7;$i++) {
+    $code.=<<___;
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	movdqa	@XMM[7], @XMM[$i]
+	movdqa	@XMM[7], `0x10*$i`(%rsp)# save tweak[$i]
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+___
+    $code.=<<___ if ($i>=1);
+	movdqu	`0x10*($i-1)`($inp), @XMM[8+$i-1]
+___
+    $code.=<<___ if ($i>=2);
+	pxor	@XMM[8+$i-2], @XMM[$i-2]# input[] ^ tweak[]
+___
+    }
+$code.=<<___;
+	movdqu	0x60($inp), @XMM[8+6]
+	pxor	@XMM[8+5], @XMM[5]
+	movdqu	0x70($inp), @XMM[8+7]
+	lea	0x80($inp), $inp
+	movdqa	@XMM[7], 0x70(%rsp)
+	pxor	@XMM[8+6], @XMM[6]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	pxor	@XMM[8+7], @XMM[7]
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	pxor	0x40(%rsp), @XMM[3]
+	movdqu	@XMM[6], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[3], 0x40($out)
+	pxor	0x60(%rsp), @XMM[2]
+	movdqu	@XMM[7], 0x50($out)
+	pxor	0x70(%rsp), @XMM[5]
+	movdqu	@XMM[2], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+
+	movdqa	0x70(%rsp), @XMM[7]	# prepare next iteration tweak
+	pxor	$twtmp, $twtmp
+	movdqa	.Lxts_magic(%rip), $twmask
+	pcmpgtd	@XMM[7], $twtmp
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+
+	sub	\$0x80,$len
+	jnc	.Lxts_enc_loop
+
+.Lxts_enc_short:
+	add	\$0x80, $len
+	jz	.Lxts_enc_done
+___
+    for ($i=0;$i<7;$i++) {
+    $code.=<<___;
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	movdqa	@XMM[7], @XMM[$i]
+	movdqa	@XMM[7], `0x10*$i`(%rsp)# save tweak[$i]
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+___
+    $code.=<<___ if ($i>=1);
+	movdqu	`0x10*($i-1)`($inp), @XMM[8+$i-1]
+	cmp	\$`0x10*$i`,$len
+	je	.Lxts_enc_$i
+___
+    $code.=<<___ if ($i>=2);
+	pxor	@XMM[8+$i-2], @XMM[$i-2]# input[] ^ tweak[]
+___
+    }
+$code.=<<___;
+	movdqu	0x60($inp), @XMM[8+6]
+	pxor	@XMM[8+5], @XMM[5]
+	movdqa	@XMM[7], 0x70(%rsp)
+	lea	0x70($inp), $inp
+	pxor	@XMM[8+6], @XMM[6]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	pxor	0x40(%rsp), @XMM[3]
+	movdqu	@XMM[6], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[3], 0x40($out)
+	pxor	0x60(%rsp), @XMM[2]
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[2], 0x60($out)
+	lea	0x70($out), $out
+
+	movdqa	0x70(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_6:
+	pxor	@XMM[8+4], @XMM[4]
+	lea	0x60($inp), $inp
+	pxor	@XMM[8+5], @XMM[5]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	pxor	0x40(%rsp), @XMM[3]
+	movdqu	@XMM[6], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[3], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	lea	0x60($out), $out
+
+	movdqa	0x60(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_5:
+	pxor	@XMM[8+3], @XMM[3]
+	lea	0x50($inp), $inp
+	pxor	@XMM[8+4], @XMM[4]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	pxor	0x40(%rsp), @XMM[3]
+	movdqu	@XMM[6], 0x30($out)
+	movdqu	@XMM[3], 0x40($out)
+	lea	0x50($out), $out
+
+	movdqa	0x50(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_4:
+	pxor	@XMM[8+2], @XMM[2]
+	lea	0x40($inp), $inp
+	pxor	@XMM[8+3], @XMM[3]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[6]
+	movdqu	@XMM[4], 0x20($out)
+	movdqu	@XMM[6], 0x30($out)
+	lea	0x40($out), $out
+
+	movdqa	0x40(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_3:
+	pxor	@XMM[8+1], @XMM[1]
+	lea	0x30($inp), $inp
+	pxor	@XMM[8+2], @XMM[2]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[4]
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[4], 0x20($out)
+	lea	0x30($out), $out
+
+	movdqa	0x30(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_2:
+	pxor	@XMM[8+0], @XMM[0]
+	lea	0x20($inp), $inp
+	pxor	@XMM[8+1], @XMM[1]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_encrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	lea	0x20($out), $out
+
+	movdqa	0x20(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_enc_done
+.align	16
+.Lxts_enc_1:
+	pxor	@XMM[0], @XMM[8]
+	lea	0x10($inp), $inp
+	movdqa	@XMM[8], 0x20(%rbp)
+	lea	0x20(%rbp), $arg1
+	lea	0x20(%rbp), $arg2
+	lea	($key), $arg3
+	call	asm_AES_encrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[0]	# ^= tweak[]
+	#pxor	@XMM[8], @XMM[0]
+	#lea	0x80(%rsp), %rax	# pass key schedule
+	#mov	%edx, %r10d		# pass rounds
+	#call	_bsaes_encrypt8
+	#pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	lea	0x10($out), $out
+
+	movdqa	0x10(%rsp), @XMM[7]	# next iteration tweak
+
+.Lxts_enc_done:
+	and	\$15, %ebx
+	jz	.Lxts_enc_ret
+	mov	$out, %rdx
+
+.Lxts_enc_steal:
+	movzb	($inp), %eax
+	movzb	-16(%rdx), %ecx
+	lea	1($inp), $inp
+	mov	%al, -16(%rdx)
+	mov	%cl, 0(%rdx)
+	lea	1(%rdx), %rdx
+	sub	\$1,%ebx
+	jnz	.Lxts_enc_steal
+
+	movdqu	-16($out), @XMM[0]
+	lea	0x20(%rbp), $arg1
+	pxor	@XMM[7], @XMM[0]
+	lea	0x20(%rbp), $arg2
+	movdqa	@XMM[0], 0x20(%rbp)
+	lea	($key), $arg3
+	call	asm_AES_encrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[7]
+	movdqu	@XMM[7], -16($out)
+
+.Lxts_enc_ret:
+	lea	(%rsp), %rax
+	pxor	%xmm0, %xmm0
+.Lxts_enc_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	ja	.Lxts_enc_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lxts_enc_epilogue:
+	ret
+.size	bsaes_xts_encrypt,.-bsaes_xts_encrypt
+
+.globl	bsaes_xts_decrypt
+.type	bsaes_xts_decrypt,\@abi-omnipotent
+.align	16
+bsaes_xts_decrypt:
+	mov	%rsp, %rax
+.Lxts_dec_prologue:
+	push	%rbp
+	push	%rbx
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	-0x48(%rsp), %rsp
+___
+$code.=<<___ if ($win64);
+	mov	0xa0(%rsp),$arg5	# pull key2
+	mov	0xa8(%rsp),$arg6	# pull ivp
+	lea	-0xa0(%rsp), %rsp
+	movaps	%xmm6, 0x40(%rsp)
+	movaps	%xmm7, 0x50(%rsp)
+	movaps	%xmm8, 0x60(%rsp)
+	movaps	%xmm9, 0x70(%rsp)
+	movaps	%xmm10, 0x80(%rsp)
+	movaps	%xmm11, 0x90(%rsp)
+	movaps	%xmm12, 0xa0(%rsp)
+	movaps	%xmm13, 0xb0(%rsp)
+	movaps	%xmm14, 0xc0(%rsp)
+	movaps	%xmm15, 0xd0(%rsp)
+.Lxts_dec_body:
+___
+$code.=<<___;
+	mov	%rsp, %rbp		# backup %rsp
+	mov	$arg1, $inp		# backup arguments
+	mov	$arg2, $out
+	mov	$arg3, $len
+	mov	$arg4, $key
+
+	lea	($arg6), $arg1
+	lea	0x20(%rbp), $arg2
+	lea	($arg5), $arg3
+	call	asm_AES_encrypt		# generate initial tweak
+
+	mov	240($key), %eax		# rounds
+	mov	$len, %rbx		# backup $len
+
+	mov	%eax, %edx		# rounds
+	shl	\$7, %rax		# 128 bytes per inner round key
+	sub	\$`128-32`, %rax	# size of bit-sliced key schedule
+	sub	%rax, %rsp
+
+	mov	%rsp, %rax		# pass key schedule
+	mov	$key, %rcx		# pass key
+	mov	%edx, %r10d		# pass rounds
+	call	_bsaes_key_convert
+	pxor	(%rsp), %xmm7		# fix up round 0 key
+	movdqa	%xmm6, (%rax)		# save last round key
+	movdqa	%xmm7, (%rsp)
+
+	xor	%eax, %eax		# if ($len%16) len-=16;
+	and	\$-16, $len
+	test	\$15, %ebx
+	setnz	%al
+	shl	\$4, %rax
+	sub	%rax, $len
+
+	sub	\$0x80, %rsp		# place for tweak[8]
+	movdqa	0x20(%rbp), @XMM[7]	# initial tweak
+
+	pxor	$twtmp, $twtmp
+	movdqa	.Lxts_magic(%rip), $twmask
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+
+	sub	\$0x80, $len
+	jc	.Lxts_dec_short
+	jmp	.Lxts_dec_loop
+
+.align	16
+.Lxts_dec_loop:
+___
+    for ($i=0;$i<7;$i++) {
+    $code.=<<___;
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	movdqa	@XMM[7], @XMM[$i]
+	movdqa	@XMM[7], `0x10*$i`(%rsp)# save tweak[$i]
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+___
+    $code.=<<___ if ($i>=1);
+	movdqu	`0x10*($i-1)`($inp), @XMM[8+$i-1]
+___
+    $code.=<<___ if ($i>=2);
+	pxor	@XMM[8+$i-2], @XMM[$i-2]# input[] ^ tweak[]
+___
+    }
+$code.=<<___;
+	movdqu	0x60($inp), @XMM[8+6]
+	pxor	@XMM[8+5], @XMM[5]
+	movdqu	0x70($inp), @XMM[8+7]
+	lea	0x80($inp), $inp
+	movdqa	@XMM[7], 0x70(%rsp)
+	pxor	@XMM[8+6], @XMM[6]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	pxor	@XMM[8+7], @XMM[7]
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[4]
+	movdqu	@XMM[6], 0x20($out)
+	pxor	0x40(%rsp), @XMM[2]
+	movdqu	@XMM[4], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[2], 0x40($out)
+	pxor	0x60(%rsp), @XMM[3]
+	movdqu	@XMM[7], 0x50($out)
+	pxor	0x70(%rsp), @XMM[5]
+	movdqu	@XMM[3], 0x60($out)
+	movdqu	@XMM[5], 0x70($out)
+	lea	0x80($out), $out
+
+	movdqa	0x70(%rsp), @XMM[7]	# prepare next iteration tweak
+	pxor	$twtmp, $twtmp
+	movdqa	.Lxts_magic(%rip), $twmask
+	pcmpgtd	@XMM[7], $twtmp
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+
+	sub	\$0x80,$len
+	jnc	.Lxts_dec_loop
+
+.Lxts_dec_short:
+	add	\$0x80, $len
+	jz	.Lxts_dec_done
+___
+    for ($i=0;$i<7;$i++) {
+    $code.=<<___;
+	pshufd	\$0x13, $twtmp, $twres
+	pxor	$twtmp, $twtmp
+	movdqa	@XMM[7], @XMM[$i]
+	movdqa	@XMM[7], `0x10*$i`(%rsp)# save tweak[$i]
+	paddq	@XMM[7], @XMM[7]	# psllq	1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	pcmpgtd	@XMM[7], $twtmp		# broadcast upper bits
+	pxor	$twres, @XMM[7]
+___
+    $code.=<<___ if ($i>=1);
+	movdqu	`0x10*($i-1)`($inp), @XMM[8+$i-1]
+	cmp	\$`0x10*$i`,$len
+	je	.Lxts_dec_$i
+___
+    $code.=<<___ if ($i>=2);
+	pxor	@XMM[8+$i-2], @XMM[$i-2]# input[] ^ tweak[]
+___
+    }
+$code.=<<___;
+	movdqu	0x60($inp), @XMM[8+6]
+	pxor	@XMM[8+5], @XMM[5]
+	movdqa	@XMM[7], 0x70(%rsp)
+	lea	0x70($inp), $inp
+	pxor	@XMM[8+6], @XMM[6]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[4]
+	movdqu	@XMM[6], 0x20($out)
+	pxor	0x40(%rsp), @XMM[2]
+	movdqu	@XMM[4], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[2], 0x40($out)
+	pxor	0x60(%rsp), @XMM[3]
+	movdqu	@XMM[7], 0x50($out)
+	movdqu	@XMM[3], 0x60($out)
+	lea	0x70($out), $out
+
+	movdqa	0x70(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_6:
+	pxor	@XMM[8+4], @XMM[4]
+	lea	0x60($inp), $inp
+	pxor	@XMM[8+5], @XMM[5]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[4]
+	movdqu	@XMM[6], 0x20($out)
+	pxor	0x40(%rsp), @XMM[2]
+	movdqu	@XMM[4], 0x30($out)
+	pxor	0x50(%rsp), @XMM[7]
+	movdqu	@XMM[2], 0x40($out)
+	movdqu	@XMM[7], 0x50($out)
+	lea	0x60($out), $out
+
+	movdqa	0x60(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_5:
+	pxor	@XMM[8+3], @XMM[3]
+	lea	0x50($inp), $inp
+	pxor	@XMM[8+4], @XMM[4]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[4]
+	movdqu	@XMM[6], 0x20($out)
+	pxor	0x40(%rsp), @XMM[2]
+	movdqu	@XMM[4], 0x30($out)
+	movdqu	@XMM[2], 0x40($out)
+	lea	0x50($out), $out
+
+	movdqa	0x50(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_4:
+	pxor	@XMM[8+2], @XMM[2]
+	lea	0x40($inp), $inp
+	pxor	@XMM[8+3], @XMM[3]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	pxor	0x30(%rsp), @XMM[4]
+	movdqu	@XMM[6], 0x20($out)
+	movdqu	@XMM[4], 0x30($out)
+	lea	0x40($out), $out
+
+	movdqa	0x40(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_3:
+	pxor	@XMM[8+1], @XMM[1]
+	lea	0x30($inp), $inp
+	pxor	@XMM[8+2], @XMM[2]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	pxor	0x20(%rsp), @XMM[6]
+	movdqu	@XMM[1], 0x10($out)
+	movdqu	@XMM[6], 0x20($out)
+	lea	0x30($out), $out
+
+	movdqa	0x30(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_2:
+	pxor	@XMM[8+0], @XMM[0]
+	lea	0x20($inp), $inp
+	pxor	@XMM[8+1], @XMM[1]
+	lea	0x80(%rsp), %rax	# pass key schedule
+	mov	%edx, %r10d		# pass rounds
+
+	call	_bsaes_decrypt8
+
+	pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	pxor	0x10(%rsp), @XMM[1]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	movdqu	@XMM[1], 0x10($out)
+	lea	0x20($out), $out
+
+	movdqa	0x20(%rsp), @XMM[7]	# next iteration tweak
+	jmp	.Lxts_dec_done
+.align	16
+.Lxts_dec_1:
+	pxor	@XMM[0], @XMM[8]
+	lea	0x10($inp), $inp
+	movdqa	@XMM[8], 0x20(%rbp)
+	lea	0x20(%rbp), $arg1
+	lea	0x20(%rbp), $arg2
+	lea	($key), $arg3
+	call	asm_AES_decrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[0]	# ^= tweak[]
+	#pxor	@XMM[8], @XMM[0]
+	#lea	0x80(%rsp), %rax	# pass key schedule
+	#mov	%edx, %r10d		# pass rounds
+	#call	_bsaes_decrypt8
+	#pxor	0x00(%rsp), @XMM[0]	# ^= tweak[]
+	movdqu	@XMM[0], 0x00($out)	# write output
+	lea	0x10($out), $out
+
+	movdqa	0x10(%rsp), @XMM[7]	# next iteration tweak
+
+.Lxts_dec_done:
+	and	\$15, %ebx
+	jz	.Lxts_dec_ret
+
+	pxor	$twtmp, $twtmp
+	movdqa	.Lxts_magic(%rip), $twmask
+	pcmpgtd	@XMM[7], $twtmp
+	pshufd	\$0x13, $twtmp, $twres
+	movdqa	@XMM[7], @XMM[6]
+	paddq	@XMM[7], @XMM[7]	# psllq 1,$tweak
+	pand	$twmask, $twres		# isolate carry and residue
+	movdqu	($inp), @XMM[0]
+	pxor	$twres, @XMM[7]
+
+	lea	0x20(%rbp), $arg1
+	pxor	@XMM[7], @XMM[0]
+	lea	0x20(%rbp), $arg2
+	movdqa	@XMM[0], 0x20(%rbp)
+	lea	($key), $arg3
+	call	asm_AES_decrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[7]
+	mov	$out, %rdx
+	movdqu	@XMM[7], ($out)
+
+.Lxts_dec_steal:
+	movzb	16($inp), %eax
+	movzb	(%rdx), %ecx
+	lea	1($inp), $inp
+	mov	%al, (%rdx)
+	mov	%cl, 16(%rdx)
+	lea	1(%rdx), %rdx
+	sub	\$1,%ebx
+	jnz	.Lxts_dec_steal
+
+	movdqu	($out), @XMM[0]
+	lea	0x20(%rbp), $arg1
+	pxor	@XMM[6], @XMM[0]
+	lea	0x20(%rbp), $arg2
+	movdqa	@XMM[0], 0x20(%rbp)
+	lea	($key), $arg3
+	call	asm_AES_decrypt		# doesn't touch %xmm
+	pxor	0x20(%rbp), @XMM[6]
+	movdqu	@XMM[6], ($out)
+
+.Lxts_dec_ret:
+	lea	(%rsp), %rax
+	pxor	%xmm0, %xmm0
+.Lxts_dec_bzero:			# wipe key schedule [if any]
+	movdqa	%xmm0, 0x00(%rax)
+	movdqa	%xmm0, 0x10(%rax)
+	lea	0x20(%rax), %rax
+	cmp	%rax, %rbp
+	ja	.Lxts_dec_bzero
+
+	lea	(%rbp),%rsp		# restore %rsp
+___
+$code.=<<___ if ($win64);
+	movaps	0x40(%rbp), %xmm6
+	movaps	0x50(%rbp), %xmm7
+	movaps	0x60(%rbp), %xmm8
+	movaps	0x70(%rbp), %xmm9
+	movaps	0x80(%rbp), %xmm10
+	movaps	0x90(%rbp), %xmm11
+	movaps	0xa0(%rbp), %xmm12
+	movaps	0xb0(%rbp), %xmm13
+	movaps	0xc0(%rbp), %xmm14
+	movaps	0xd0(%rbp), %xmm15
+	lea	0xa0(%rbp), %rsp
+___
+$code.=<<___;
+	mov	0x48(%rsp), %r15
+	mov	0x50(%rsp), %r14
+	mov	0x58(%rsp), %r13
+	mov	0x60(%rsp), %r12
+	mov	0x68(%rsp), %rbx
+	mov	0x70(%rsp), %rax
+	lea	0x78(%rsp), %rsp
+	mov	%rax, %rbp
+.Lxts_dec_epilogue:
+	ret
+.size	bsaes_xts_decrypt,.-bsaes_xts_decrypt
+___
+}
+$code.=<<___;
+.type	_bsaes_const,\@object
+.align	64
+_bsaes_const:
+.LM0ISR:	# InvShiftRows constants
+	.quad	0x0a0e0206070b0f03, 0x0004080c0d010509
+.LISRM0:
+	.quad	0x01040b0e0205080f, 0x0306090c00070a0d
+.LISR:
+	.quad	0x0504070602010003, 0x0f0e0d0c080b0a09
+.LBS0:		# bit-slice constants
+	.quad	0x5555555555555555, 0x5555555555555555
+.LBS1:
+	.quad	0x3333333333333333, 0x3333333333333333
+.LBS2:
+	.quad	0x0f0f0f0f0f0f0f0f, 0x0f0f0f0f0f0f0f0f
+.LSR:		# shiftrows constants
+	.quad	0x0504070600030201, 0x0f0e0d0c0a09080b
+.LSRM0:
+	.quad	0x0304090e00050a0f, 0x01060b0c0207080d
+.LM0SR:
+	.quad	0x0a0e02060f03070b, 0x0004080c05090d01
+.LSWPUP:	# byte-swap upper dword
+	.quad	0x0706050403020100, 0x0c0d0e0f0b0a0908
+.LSWPUPM0SR:
+	.quad	0x0a0d02060c03070b, 0x0004080f05090e01
+.LADD1:		# counter increment constants
+	.quad	0x0000000000000000, 0x0000000100000000
+.LADD2:
+	.quad	0x0000000000000000, 0x0000000200000000
+.LADD3:
+	.quad	0x0000000000000000, 0x0000000300000000
+.LADD4:
+	.quad	0x0000000000000000, 0x0000000400000000
+.LADD5:
+	.quad	0x0000000000000000, 0x0000000500000000
+.LADD6:
+	.quad	0x0000000000000000, 0x0000000600000000
+.LADD7:
+	.quad	0x0000000000000000, 0x0000000700000000
+.LADD8:
+	.quad	0x0000000000000000, 0x0000000800000000
+.Lxts_magic:
+	.long	0x87,0,1,0
+.Lmasks:
+	.quad	0x0101010101010101, 0x0101010101010101
+	.quad	0x0202020202020202, 0x0202020202020202
+	.quad	0x0404040404040404, 0x0404040404040404
+	.quad	0x0808080808080808, 0x0808080808080808
+.LM0:
+	.quad	0x02060a0e03070b0f, 0x0004080c0105090d
+.L63:
+	.quad	0x6363636363636363, 0x6363636363636363
+.asciz	"Bit-sliced AES for x86_64/SSSE3, Emilia Käsper, Peter Schwabe, Andy Polyakov"
+.align	64
+.size	_bsaes_const,.-_bsaes_const
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	se_handler,\@abi-omnipotent
+.align	16
+se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_prologue
+
+	mov	160($context),%rax	# pull context->Rbp
+
+	lea	0x40(%rax),%rsi		# %xmm save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	0xa0(%rax),%rax		# adjust stack pointer
+
+	mov	0x70(%rax),%rbp
+	mov	0x68(%rax),%rbx
+	mov	0x60(%rax),%r12
+	mov	0x58(%rax),%r13
+	mov	0x50(%rax),%r14
+	mov	0x48(%rax),%r15
+	lea	0x78(%rax),%rax		# adjust stack pointer
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_prologue:
+	mov	%rax,152($context)	# restore context->Rsp
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	se_handler,.-se_handler
+
+.section	.pdata
+.align	4
+___
+$code.=<<___ if ($ecb);
+	.rva	.Lecb_enc_prologue
+	.rva	.Lecb_enc_epilogue
+	.rva	.Lecb_enc_info
+
+	.rva	.Lecb_dec_prologue
+	.rva	.Lecb_dec_epilogue
+	.rva	.Lecb_dec_info
+___
+$code.=<<___;
+	.rva	.Lcbc_dec_prologue
+	.rva	.Lcbc_dec_epilogue
+	.rva	.Lcbc_dec_info
+
+	.rva	.Lctr_enc_prologue
+	.rva	.Lctr_enc_epilogue
+	.rva	.Lctr_enc_info
+
+	.rva	.Lxts_enc_prologue
+	.rva	.Lxts_enc_epilogue
+	.rva	.Lxts_enc_info
+
+	.rva	.Lxts_dec_prologue
+	.rva	.Lxts_dec_epilogue
+	.rva	.Lxts_dec_info
+
+.section	.xdata
+.align	8
+___
+$code.=<<___ if ($ecb);
+.Lecb_enc_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lecb_enc_body,.Lecb_enc_epilogue	# HandlerData[]
+.Lecb_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lecb_dec_body,.Lecb_dec_epilogue	# HandlerData[]
+___
+$code.=<<___;
+.Lcbc_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lcbc_dec_body,.Lcbc_dec_epilogue	# HandlerData[]
+.Lctr_enc_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lctr_enc_body,.Lctr_enc_epilogue	# HandlerData[]
+.Lxts_enc_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lxts_enc_body,.Lxts_enc_epilogue	# HandlerData[]
+.Lxts_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lxts_dec_body,.Lxts_dec_epilogue	# HandlerData[]
+___
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+
+print $code;
+
+close STDOUT;
diff --git a/src/crypto/aes/asm/vpaes-x86.pl b/src/crypto/aes/asm/vpaes-x86.pl
new file mode 100644
index 0000000..2ba149c
--- /dev/null
+++ b/src/crypto/aes/asm/vpaes-x86.pl
@@ -0,0 +1,903 @@
+#!/usr/bin/env perl
+
+######################################################################
+## Constant-time SSSE3 AES core implementation.
+## version 0.1
+##
+## By Mike Hamburg (Stanford University), 2009
+## Public domain.
+##
+## For details see http://shiftleft.org/papers/vector_aes/ and
+## http://crypto.stanford.edu/vpaes/.
+
+######################################################################
+# September 2011.
+#
+# Port vpaes-x86_64.pl as 32-bit "almost" drop-in replacement for
+# aes-586.pl. "Almost" refers to the fact that AES_cbc_encrypt
+# doesn't handle partial vectors (doesn't have to if called from
+# EVP only). "Drop-in" implies that this module doesn't share key
+# schedule structure with the original nor does it make assumption
+# about its alignment...
+#
+# Performance summary. aes-586.pl column lists large-block CBC
+# encrypt/decrypt/with-hyper-threading-off(*) results in cycles per
+# byte processed with 128-bit key, and vpaes-x86.pl column - [also
+# large-block CBC] encrypt/decrypt.
+#
+#		aes-586.pl		vpaes-x86.pl
+#
+# Core 2(**)	28.1/41.4/18.3		21.9/25.2(***)
+# Nehalem	27.9/40.4/18.1		10.2/11.9
+# Atom		70.7/92.1/60.1		61.1/75.4(***)
+# Silvermont	45.4/62.9/24.1		49.2/61.1(***)
+#
+# (*)	"Hyper-threading" in the context refers rather to cache shared
+#	among multiple cores, than to specifically Intel HTT. As vast
+#	majority of contemporary cores share cache, slower code path
+#	is common place. In other words "with-hyper-threading-off"
+#	results are presented mostly for reference purposes.
+#
+# (**)	"Core 2" refers to initial 65nm design, a.k.a. Conroe.
+#
+# (***)	Less impressive improvement on Core 2 and Atom is due to slow
+#	pshufb,	yet it's respectable +28%/64%  improvement on Core 2
+#	and +15% on Atom (as implied, over "hyper-threading-safe"
+#	code path).
+#
+#						<appro@openssl.org>
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],"vpaes-x86.pl",$x86only = $ARGV[$#ARGV] eq "386");
+
+$PREFIX="vpaes";
+
+my  ($round, $base, $magic, $key, $const, $inp, $out)=
+    ("eax",  "ebx", "ecx",  "edx","ebp",  "esi","edi");
+
+&static_label("_vpaes_consts");
+&static_label("_vpaes_schedule_low_round");
+
+&set_label("_vpaes_consts",64);
+$k_inv=-0x30;		# inv, inva
+	&data_word(0x0D080180,0x0E05060F,0x0A0B0C02,0x04070309);
+	&data_word(0x0F0B0780,0x01040A06,0x02050809,0x030D0E0C);
+
+$k_s0F=-0x10;		# s0F
+	&data_word(0x0F0F0F0F,0x0F0F0F0F,0x0F0F0F0F,0x0F0F0F0F);
+
+$k_ipt=0x00;		# input transform (lo, hi)
+	&data_word(0x5A2A7000,0xC2B2E898,0x52227808,0xCABAE090);
+	&data_word(0x317C4D00,0x4C01307D,0xB0FDCC81,0xCD80B1FC);
+
+$k_sb1=0x20;		# sb1u, sb1t
+	&data_word(0xCB503E00,0xB19BE18F,0x142AF544,0xA5DF7A6E);
+	&data_word(0xFAE22300,0x3618D415,0x0D2ED9EF,0x3BF7CCC1);
+$k_sb2=0x40;		# sb2u, sb2t
+	&data_word(0x0B712400,0xE27A93C6,0xBC982FCD,0x5EB7E955);
+	&data_word(0x0AE12900,0x69EB8840,0xAB82234A,0xC2A163C8);
+$k_sbo=0x60;		# sbou, sbot
+	&data_word(0x6FBDC700,0xD0D26D17,0xC502A878,0x15AABF7A);
+	&data_word(0x5FBB6A00,0xCFE474A5,0x412B35FA,0x8E1E90D1);
+
+$k_mc_forward=0x80;	# mc_forward
+	&data_word(0x00030201,0x04070605,0x080B0A09,0x0C0F0E0D);
+	&data_word(0x04070605,0x080B0A09,0x0C0F0E0D,0x00030201);
+	&data_word(0x080B0A09,0x0C0F0E0D,0x00030201,0x04070605);
+	&data_word(0x0C0F0E0D,0x00030201,0x04070605,0x080B0A09);
+
+$k_mc_backward=0xc0;	# mc_backward
+	&data_word(0x02010003,0x06050407,0x0A09080B,0x0E0D0C0F);
+	&data_word(0x0E0D0C0F,0x02010003,0x06050407,0x0A09080B);
+	&data_word(0x0A09080B,0x0E0D0C0F,0x02010003,0x06050407);
+	&data_word(0x06050407,0x0A09080B,0x0E0D0C0F,0x02010003);
+
+$k_sr=0x100;		# sr
+	&data_word(0x03020100,0x07060504,0x0B0A0908,0x0F0E0D0C);
+	&data_word(0x0F0A0500,0x030E0904,0x07020D08,0x0B06010C);
+	&data_word(0x0B020900,0x0F060D04,0x030A0108,0x070E050C);
+	&data_word(0x070A0D00,0x0B0E0104,0x0F020508,0x0306090C);
+
+$k_rcon=0x140;		# rcon
+	&data_word(0xAF9DEEB6,0x1F8391B9,0x4D7C7D81,0x702A9808);
+
+$k_s63=0x150;		# s63: all equal to 0x63 transformed
+	&data_word(0x5B5B5B5B,0x5B5B5B5B,0x5B5B5B5B,0x5B5B5B5B);
+
+$k_opt=0x160;		# output transform
+	&data_word(0xD6B66000,0xFF9F4929,0xDEBE6808,0xF7974121);
+	&data_word(0x50BCEC00,0x01EDBD51,0xB05C0CE0,0xE10D5DB1);
+
+$k_deskew=0x180;	# deskew tables: inverts the sbox's "skew"
+	&data_word(0x47A4E300,0x07E4A340,0x5DBEF91A,0x1DFEB95A);
+	&data_word(0x83EA6900,0x5F36B5DC,0xF49D1E77,0x2841C2AB);
+##
+##  Decryption stuff
+##  Key schedule constants
+##
+$k_dksd=0x1a0;		# decryption key schedule: invskew x*D
+	&data_word(0xA3E44700,0xFEB91A5D,0x5A1DBEF9,0x0740E3A4);
+	&data_word(0xB5368300,0x41C277F4,0xAB289D1E,0x5FDC69EA);
+$k_dksb=0x1c0;		# decryption key schedule: invskew x*B
+	&data_word(0x8550D500,0x9A4FCA1F,0x1CC94C99,0x03D65386);
+	&data_word(0xB6FC4A00,0x115BEDA7,0x7E3482C8,0xD993256F);
+$k_dkse=0x1e0;		# decryption key schedule: invskew x*E + 0x63
+	&data_word(0x1FC9D600,0xD5031CCA,0x994F5086,0x53859A4C);
+	&data_word(0x4FDC7BE8,0xA2319605,0x20B31487,0xCD5EF96A);
+$k_dks9=0x200;		# decryption key schedule: invskew x*9
+	&data_word(0x7ED9A700,0xB6116FC8,0x82255BFC,0x4AED9334);
+	&data_word(0x27143300,0x45765162,0xE9DAFDCE,0x8BB89FAC);
+
+##
+##  Decryption stuff
+##  Round function constants
+##
+$k_dipt=0x220;		# decryption input transform
+	&data_word(0x0B545F00,0x0F505B04,0x114E451A,0x154A411E);
+	&data_word(0x60056500,0x86E383E6,0xF491F194,0x12771772);
+
+$k_dsb9=0x240;		# decryption sbox output *9*u, *9*t
+	&data_word(0x9A86D600,0x851C0353,0x4F994CC9,0xCAD51F50);
+	&data_word(0xECD74900,0xC03B1789,0xB2FBA565,0x725E2C9E);
+$k_dsbd=0x260;		# decryption sbox output *D*u, *D*t
+	&data_word(0xE6B1A200,0x7D57CCDF,0x882A4439,0xF56E9B13);
+	&data_word(0x24C6CB00,0x3CE2FAF7,0x15DEEFD3,0x2931180D);
+$k_dsbb=0x280;		# decryption sbox output *B*u, *B*t
+	&data_word(0x96B44200,0xD0226492,0xB0F2D404,0x602646F6);
+	&data_word(0xCD596700,0xC19498A6,0x3255AA6B,0xF3FF0C3E);
+$k_dsbe=0x2a0;		# decryption sbox output *E*u, *E*t
+	&data_word(0x26D4D000,0x46F29296,0x64B4F6B0,0x22426004);
+	&data_word(0xFFAAC100,0x0C55A6CD,0x98593E32,0x9467F36B);
+$k_dsbo=0x2c0;		# decryption sbox final output
+	&data_word(0x7EF94000,0x1387EA53,0xD4943E2D,0xC7AA6DB9);
+	&data_word(0x93441D00,0x12D7560F,0xD8C58E9C,0xCA4B8159);
+&asciz	("Vector Permutation AES for x86/SSSE3, Mike Hamburg (Stanford University)");
+&align	(64);
+
+&function_begin_B("_vpaes_preheat");
+	&add	($const,&DWP(0,"esp"));
+	&movdqa	("xmm7",&QWP($k_inv,$const));
+	&movdqa	("xmm6",&QWP($k_s0F,$const));
+	&ret	();
+&function_end_B("_vpaes_preheat");
+
+##
+##  _aes_encrypt_core
+##
+##  AES-encrypt %xmm0.
+##
+##  Inputs:
+##     %xmm0 = input
+##     %xmm6-%xmm7 as in _vpaes_preheat
+##    (%edx) = scheduled keys
+##
+##  Output in %xmm0
+##  Clobbers  %xmm1-%xmm5, %eax, %ebx, %ecx, %edx
+##
+##
+&function_begin_B("_vpaes_encrypt_core");
+	&mov	($magic,16);
+	&mov	($round,&DWP(240,$key));
+	&movdqa	("xmm1","xmm6")
+	&movdqa	("xmm2",&QWP($k_ipt,$const));
+	&pandn	("xmm1","xmm0");
+	&pand	("xmm0","xmm6");
+	&movdqu	("xmm5",&QWP(0,$key));
+	&pshufb	("xmm2","xmm0");
+	&movdqa	("xmm0",&QWP($k_ipt+16,$const));
+	&pxor	("xmm2","xmm5");
+	&psrld	("xmm1",4);
+	&add	($key,16);
+	&pshufb	("xmm0","xmm1");
+	&lea	($base,&DWP($k_mc_backward,$const));
+	&pxor	("xmm0","xmm2");
+	&jmp	(&label("enc_entry"));
+
+
+&set_label("enc_loop",16);
+	# middle of middle round
+	&movdqa	("xmm4",&QWP($k_sb1,$const));	# 4 : sb1u
+	&movdqa	("xmm0",&QWP($k_sb1+16,$const));# 0 : sb1t
+	&pshufb	("xmm4","xmm2");		# 4 = sb1u
+	&pshufb	("xmm0","xmm3");		# 0 = sb1t
+	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
+	&movdqa	("xmm5",&QWP($k_sb2,$const));	# 4 : sb2u
+	&pxor	("xmm0","xmm4");		# 0 = A
+	&movdqa	("xmm1",&QWP(-0x40,$base,$magic));# .Lk_mc_forward[]
+	&pshufb	("xmm5","xmm2");		# 4 = sb2u
+	&movdqa	("xmm2",&QWP($k_sb2+16,$const));# 2 : sb2t
+	&movdqa	("xmm4",&QWP(0,$base,$magic));	# .Lk_mc_backward[]
+	&pshufb	("xmm2","xmm3");		# 2 = sb2t
+	&movdqa	("xmm3","xmm0");		# 3 = A
+	&pxor	("xmm2","xmm5");		# 2 = 2A
+	&pshufb	("xmm0","xmm1");		# 0 = B
+	&add	($key,16);			# next key
+	&pxor	("xmm0","xmm2");		# 0 = 2A+B
+	&pshufb	("xmm3","xmm4");		# 3 = D
+	&add	($magic,16);			# next mc
+	&pxor	("xmm3","xmm0");		# 3 = 2A+B+D
+	&pshufb	("xmm0","xmm1");		# 0 = 2B+C
+	&and	($magic,0x30);			# ... mod 4
+	&sub	($round,1);			# nr--
+	&pxor	("xmm0","xmm3");		# 0 = 2A+3B+C+D
+
+&set_label("enc_entry");
+	# top of round
+	&movdqa	("xmm1","xmm6");		# 1 : i
+	&movdqa	("xmm5",&QWP($k_inv+16,$const));# 2 : a/k
+	&pandn	("xmm1","xmm0");		# 1 = i<<4
+	&psrld	("xmm1",4);			# 1 = i
+	&pand	("xmm0","xmm6");		# 0 = k
+	&pshufb	("xmm5","xmm0");		# 2 = a/k
+	&movdqa	("xmm3","xmm7");		# 3 : 1/i
+	&pxor	("xmm0","xmm1");		# 0 = j
+	&pshufb	("xmm3","xmm1");		# 3 = 1/i
+	&movdqa	("xmm4","xmm7");		# 4 : 1/j
+	&pxor	("xmm3","xmm5");		# 3 = iak = 1/i + a/k
+	&pshufb	("xmm4","xmm0");		# 4 = 1/j
+	&movdqa	("xmm2","xmm7");		# 2 : 1/iak
+	&pxor	("xmm4","xmm5");		# 4 = jak = 1/j + a/k
+	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
+	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
+	&pxor	("xmm2","xmm0");		# 2 = io
+	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
+	&movdqu	("xmm5",&QWP(0,$key));
+	&pxor	("xmm3","xmm1");		# 3 = jo
+	&jnz	(&label("enc_loop"));
+
+	# middle of last round
+	&movdqa	("xmm4",&QWP($k_sbo,$const));	# 3 : sbou      .Lk_sbo
+	&movdqa	("xmm0",&QWP($k_sbo+16,$const));# 3 : sbot      .Lk_sbo+16
+	&pshufb	("xmm4","xmm2");		# 4 = sbou
+	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
+	&pshufb	("xmm0","xmm3");		# 0 = sb1t
+	&movdqa	("xmm1",&QWP(0x40,$base,$magic));# .Lk_sr[]
+	&pxor	("xmm0","xmm4");		# 0 = A
+	&pshufb	("xmm0","xmm1");
+	&ret	();
+&function_end_B("_vpaes_encrypt_core");
+
+##
+##  Decryption core
+##
+##  Same API as encryption core.
+##
+&function_begin_B("_vpaes_decrypt_core");
+	&lea	($base,&DWP($k_dsbd,$const));
+	&mov	($round,&DWP(240,$key));
+	&movdqa	("xmm1","xmm6");
+	&movdqa	("xmm2",&QWP($k_dipt-$k_dsbd,$base));
+	&pandn	("xmm1","xmm0");
+	&mov	($magic,$round);
+	&psrld	("xmm1",4)
+	&movdqu	("xmm5",&QWP(0,$key));
+	&shl	($magic,4);
+	&pand	("xmm0","xmm6");
+	&pshufb	("xmm2","xmm0");
+	&movdqa	("xmm0",&QWP($k_dipt-$k_dsbd+16,$base));
+	&xor	($magic,0x30);
+	&pshufb	("xmm0","xmm1");
+	&and	($magic,0x30);
+	&pxor	("xmm2","xmm5");
+	&movdqa	("xmm5",&QWP($k_mc_forward+48,$const));
+	&pxor	("xmm0","xmm2");
+	&add	($key,16);
+	&lea	($magic,&DWP($k_sr-$k_dsbd,$base,$magic));
+	&jmp	(&label("dec_entry"));
+
+&set_label("dec_loop",16);
+##
+##  Inverse mix columns
+##
+	&movdqa	("xmm4",&QWP(-0x20,$base));	# 4 : sb9u
+	&movdqa	("xmm1",&QWP(-0x10,$base));	# 0 : sb9t
+	&pshufb	("xmm4","xmm2");		# 4 = sb9u
+	&pshufb	("xmm1","xmm3");		# 0 = sb9t
+	&pxor	("xmm0","xmm4");
+	&movdqa	("xmm4",&QWP(0,$base));		# 4 : sbdu
+	&pxor	("xmm0","xmm1");		# 0 = ch
+	&movdqa	("xmm1",&QWP(0x10,$base));	# 0 : sbdt
+
+	&pshufb	("xmm4","xmm2");		# 4 = sbdu
+	&pshufb	("xmm0","xmm5");		# MC ch
+	&pshufb	("xmm1","xmm3");		# 0 = sbdt
+	&pxor	("xmm0","xmm4");		# 4 = ch
+	&movdqa	("xmm4",&QWP(0x20,$base));	# 4 : sbbu
+	&pxor	("xmm0","xmm1");		# 0 = ch
+	&movdqa	("xmm1",&QWP(0x30,$base));	# 0 : sbbt
+
+	&pshufb	("xmm4","xmm2");		# 4 = sbbu
+	&pshufb	("xmm0","xmm5");		# MC ch
+	&pshufb	("xmm1","xmm3");		# 0 = sbbt
+	&pxor	("xmm0","xmm4");		# 4 = ch
+	&movdqa	("xmm4",&QWP(0x40,$base));	# 4 : sbeu
+	&pxor	("xmm0","xmm1");		# 0 = ch
+	&movdqa	("xmm1",&QWP(0x50,$base));	# 0 : sbet
+
+	&pshufb	("xmm4","xmm2");		# 4 = sbeu
+	&pshufb	("xmm0","xmm5");		# MC ch
+	&pshufb	("xmm1","xmm3");		# 0 = sbet
+	&pxor	("xmm0","xmm4");		# 4 = ch
+	&add	($key,16);			# next round key
+	&palignr("xmm5","xmm5",12);
+	&pxor	("xmm0","xmm1");		# 0 = ch
+	&sub	($round,1);			# nr--
+
+&set_label("dec_entry");
+	# top of round
+	&movdqa	("xmm1","xmm6");		# 1 : i
+	&movdqa	("xmm2",&QWP($k_inv+16,$const));# 2 : a/k
+	&pandn	("xmm1","xmm0");		# 1 = i<<4
+	&pand	("xmm0","xmm6");		# 0 = k
+	&psrld	("xmm1",4);			# 1 = i
+	&pshufb	("xmm2","xmm0");		# 2 = a/k
+	&movdqa	("xmm3","xmm7");		# 3 : 1/i
+	&pxor	("xmm0","xmm1");		# 0 = j
+	&pshufb	("xmm3","xmm1");		# 3 = 1/i
+	&movdqa	("xmm4","xmm7");		# 4 : 1/j
+	&pxor	("xmm3","xmm2");		# 3 = iak = 1/i + a/k
+	&pshufb	("xmm4","xmm0");		# 4 = 1/j
+	&pxor	("xmm4","xmm2");		# 4 = jak = 1/j + a/k
+	&movdqa	("xmm2","xmm7");		# 2 : 1/iak
+	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
+	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
+	&pxor	("xmm2","xmm0");		# 2 = io
+	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
+	&movdqu	("xmm0",&QWP(0,$key));
+	&pxor	("xmm3","xmm1");		# 3 = jo
+	&jnz	(&label("dec_loop"));
+
+	# middle of last round
+	&movdqa	("xmm4",&QWP(0x60,$base));	# 3 : sbou
+	&pshufb	("xmm4","xmm2");		# 4 = sbou
+	&pxor	("xmm4","xmm0");		# 4 = sb1u + k
+	&movdqa	("xmm0",&QWP(0x70,$base));	# 0 : sbot
+	&movdqa	("xmm2",&QWP(0,$magic));
+	&pshufb	("xmm0","xmm3");		# 0 = sb1t
+	&pxor	("xmm0","xmm4");		# 0 = A
+	&pshufb	("xmm0","xmm2");
+	&ret	();
+&function_end_B("_vpaes_decrypt_core");
+
+########################################################
+##                                                    ##
+##                  AES key schedule                  ##
+##                                                    ##
+########################################################
+&function_begin_B("_vpaes_schedule_core");
+	&add	($const,&DWP(0,"esp"));
+	&movdqu	("xmm0",&QWP(0,$inp));		# load key (unaligned)
+	&movdqa	("xmm2",&QWP($k_rcon,$const));	# load rcon
+
+	# input transform
+	&movdqa	("xmm3","xmm0");
+	&lea	($base,&DWP($k_ipt,$const));
+	&movdqa	(&QWP(4,"esp"),"xmm2");		# xmm8
+	&call	("_vpaes_schedule_transform");
+	&movdqa	("xmm7","xmm0");
+
+	&test	($out,$out);
+	&jnz	(&label("schedule_am_decrypting"));
+
+	# encrypting, output zeroth round key after transform
+	&movdqu	(&QWP(0,$key),"xmm0");
+	&jmp	(&label("schedule_go"));
+
+&set_label("schedule_am_decrypting");
+	# decrypting, output zeroth round key after shiftrows
+	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
+	&pshufb	("xmm3","xmm1");
+	&movdqu	(&QWP(0,$key),"xmm3");
+	&xor	($magic,0x30);
+
+&set_label("schedule_go");
+	&cmp	($round,192);
+	&ja	(&label("schedule_256"));
+	&je	(&label("schedule_192"));
+	# 128: fall though
+
+##
+##  .schedule_128
+##
+##  128-bit specific part of key schedule.
+##
+##  This schedule is really simple, because all its parts
+##  are accomplished by the subroutines.
+##
+&set_label("schedule_128");
+	&mov	($round,10);
+
+&set_label("loop_schedule_128");
+	&call	("_vpaes_schedule_round");
+	&dec	($round);
+	&jz	(&label("schedule_mangle_last"));
+	&call	("_vpaes_schedule_mangle");	# write output
+	&jmp	(&label("loop_schedule_128"));
+
+##
+##  .aes_schedule_192
+##
+##  192-bit specific part of key schedule.
+##
+##  The main body of this schedule is the same as the 128-bit
+##  schedule, but with more smearing.  The long, high side is
+##  stored in %xmm7 as before, and the short, low side is in
+##  the high bits of %xmm6.
+##
+##  This schedule is somewhat nastier, however, because each
+##  round produces 192 bits of key material, or 1.5 round keys.
+##  Therefore, on each cycle we do 2 rounds and produce 3 round
+##  keys.
+##
+&set_label("schedule_192",16);
+	&movdqu	("xmm0",&QWP(8,$inp));		# load key part 2 (very unaligned)
+	&call	("_vpaes_schedule_transform");	# input transform	
+	&movdqa	("xmm6","xmm0");		# save short part
+	&pxor	("xmm4","xmm4");		# clear 4
+	&movhlps("xmm6","xmm4");		# clobber low side with zeros
+	&mov	($round,4);
+
+&set_label("loop_schedule_192");
+	&call	("_vpaes_schedule_round");
+	&palignr("xmm0","xmm6",8);
+	&call	("_vpaes_schedule_mangle");	# save key n
+	&call	("_vpaes_schedule_192_smear");
+	&call	("_vpaes_schedule_mangle");	# save key n+1
+	&call	("_vpaes_schedule_round");
+	&dec	($round);
+	&jz	(&label("schedule_mangle_last"));
+	&call	("_vpaes_schedule_mangle");	# save key n+2
+	&call	("_vpaes_schedule_192_smear");
+	&jmp	(&label("loop_schedule_192"));
+
+##
+##  .aes_schedule_256
+##
+##  256-bit specific part of key schedule.
+##
+##  The structure here is very similar to the 128-bit
+##  schedule, but with an additional "low side" in
+##  %xmm6.  The low side's rounds are the same as the
+##  high side's, except no rcon and no rotation.
+##
+&set_label("schedule_256",16);
+	&movdqu	("xmm0",&QWP(16,$inp));		# load key part 2 (unaligned)
+	&call	("_vpaes_schedule_transform");	# input transform	
+	&mov	($round,7);
+
+&set_label("loop_schedule_256");
+	&call	("_vpaes_schedule_mangle");	# output low result
+	&movdqa	("xmm6","xmm0");		# save cur_lo in xmm6
+
+	# high round
+	&call	("_vpaes_schedule_round");
+	&dec	($round);
+	&jz	(&label("schedule_mangle_last"));
+	&call	("_vpaes_schedule_mangle");	
+
+	# low round. swap xmm7 and xmm6
+	&pshufd	("xmm0","xmm0",0xFF);
+	&movdqa	(&QWP(20,"esp"),"xmm7");
+	&movdqa	("xmm7","xmm6");
+	&call	("_vpaes_schedule_low_round");
+	&movdqa	("xmm7",&QWP(20,"esp"));
+
+	&jmp	(&label("loop_schedule_256"));
+
+##
+##  .aes_schedule_mangle_last
+##
+##  Mangler for last round of key schedule
+##  Mangles %xmm0
+##    when encrypting, outputs out(%xmm0) ^ 63
+##    when decrypting, outputs unskew(%xmm0)
+##
+##  Always called right before return... jumps to cleanup and exits
+##
+&set_label("schedule_mangle_last",16);
+	# schedule last round key from xmm0
+	&lea	($base,&DWP($k_deskew,$const));
+	&test	($out,$out);
+	&jnz	(&label("schedule_mangle_last_dec"));
+
+	# encrypting
+	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
+	&pshufb	("xmm0","xmm1");		# output permute
+	&lea	($base,&DWP($k_opt,$const));	# prepare to output transform
+	&add	($key,32);
+
+&set_label("schedule_mangle_last_dec");
+	&add	($key,-16);
+	&pxor	("xmm0",&QWP($k_s63,$const));
+	&call	("_vpaes_schedule_transform");	# output transform
+	&movdqu	(&QWP(0,$key),"xmm0");		# save last key
+
+	# cleanup
+	&pxor	("xmm0","xmm0");
+	&pxor	("xmm1","xmm1");
+	&pxor	("xmm2","xmm2");
+	&pxor	("xmm3","xmm3");
+	&pxor	("xmm4","xmm4");
+	&pxor	("xmm5","xmm5");
+	&pxor	("xmm6","xmm6");
+	&pxor	("xmm7","xmm7");
+	&ret	();
+&function_end_B("_vpaes_schedule_core");
+
+##
+##  .aes_schedule_192_smear
+##
+##  Smear the short, low side in the 192-bit key schedule.
+##
+##  Inputs:
+##    %xmm7: high side, b  a  x  y
+##    %xmm6:  low side, d  c  0  0
+##    %xmm13: 0
+##
+##  Outputs:
+##    %xmm6: b+c+d  b+c  0  0
+##    %xmm0: b+c+d  b+c  b  a
+##
+&function_begin_B("_vpaes_schedule_192_smear");
+	&pshufd	("xmm1","xmm6",0x80);		# d c 0 0 -> c 0 0 0
+	&pshufd	("xmm0","xmm7",0xFE);		# b a _ _ -> b b b a
+	&pxor	("xmm6","xmm1");		# -> c+d c 0 0
+	&pxor	("xmm1","xmm1");
+	&pxor	("xmm6","xmm0");		# -> b+c+d b+c b a
+	&movdqa	("xmm0","xmm6");
+	&movhlps("xmm6","xmm1");		# clobber low side with zeros
+	&ret	();
+&function_end_B("_vpaes_schedule_192_smear");
+
+##
+##  .aes_schedule_round
+##
+##  Runs one main round of the key schedule on %xmm0, %xmm7
+##
+##  Specifically, runs subbytes on the high dword of %xmm0
+##  then rotates it by one byte and xors into the low dword of
+##  %xmm7.
+##
+##  Adds rcon from low byte of %xmm8, then rotates %xmm8 for
+##  next rcon.
+##
+##  Smears the dwords of %xmm7 by xoring the low into the
+##  second low, result into third, result into highest.
+##
+##  Returns results in %xmm7 = %xmm0.
+##  Clobbers %xmm1-%xmm5.
+##
+&function_begin_B("_vpaes_schedule_round");
+	# extract rcon from xmm8
+	&movdqa	("xmm2",&QWP(8,"esp"));		# xmm8
+	&pxor	("xmm1","xmm1");
+	&palignr("xmm1","xmm2",15);
+	&palignr("xmm2","xmm2",15);
+	&pxor	("xmm7","xmm1");
+
+	# rotate
+	&pshufd	("xmm0","xmm0",0xFF);
+	&palignr("xmm0","xmm0",1);
+
+	# fall through...
+	&movdqa	(&QWP(8,"esp"),"xmm2");		# xmm8
+
+	# low round: same as high round, but no rotation and no rcon.
+&set_label("_vpaes_schedule_low_round");
+	# smear xmm7
+	&movdqa	("xmm1","xmm7");
+	&pslldq	("xmm7",4);
+	&pxor	("xmm7","xmm1");
+	&movdqa	("xmm1","xmm7");
+	&pslldq	("xmm7",8);
+	&pxor	("xmm7","xmm1");
+	&pxor	("xmm7",&QWP($k_s63,$const));
+
+	# subbyte
+	&movdqa	("xmm4",&QWP($k_s0F,$const));
+	&movdqa	("xmm5",&QWP($k_inv,$const));	# 4 : 1/j
+	&movdqa	("xmm1","xmm4");	
+	&pandn	("xmm1","xmm0");
+	&psrld	("xmm1",4);			# 1 = i
+	&pand	("xmm0","xmm4");		# 0 = k
+	&movdqa	("xmm2",&QWP($k_inv+16,$const));# 2 : a/k
+	&pshufb	("xmm2","xmm0");		# 2 = a/k
+	&pxor	("xmm0","xmm1");		# 0 = j
+	&movdqa	("xmm3","xmm5");		# 3 : 1/i
+	&pshufb	("xmm3","xmm1");		# 3 = 1/i
+	&pxor	("xmm3","xmm2");		# 3 = iak = 1/i + a/k
+	&movdqa	("xmm4","xmm5");		# 4 : 1/j
+	&pshufb	("xmm4","xmm0");		# 4 = 1/j
+	&pxor	("xmm4","xmm2");		# 4 = jak = 1/j + a/k
+	&movdqa	("xmm2","xmm5");		# 2 : 1/iak
+	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
+	&pxor	("xmm2","xmm0");		# 2 = io
+	&movdqa	("xmm3","xmm5");		# 3 : 1/jak
+	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
+	&pxor	("xmm3","xmm1");		# 3 = jo
+	&movdqa	("xmm4",&QWP($k_sb1,$const));	# 4 : sbou
+	&pshufb	("xmm4","xmm2");		# 4 = sbou
+	&movdqa	("xmm0",&QWP($k_sb1+16,$const));# 0 : sbot
+	&pshufb	("xmm0","xmm3");		# 0 = sb1t
+	&pxor	("xmm0","xmm4");		# 0 = sbox output
+
+	# add in smeared stuff
+	&pxor	("xmm0","xmm7");
+	&movdqa	("xmm7","xmm0");
+	&ret	();
+&function_end_B("_vpaes_schedule_round");
+
+##
+##  .aes_schedule_transform
+##
+##  Linear-transform %xmm0 according to tables at (%ebx)
+##
+##  Output in %xmm0
+##  Clobbers %xmm1, %xmm2
+##
+&function_begin_B("_vpaes_schedule_transform");
+	&movdqa	("xmm2",&QWP($k_s0F,$const));
+	&movdqa	("xmm1","xmm2");
+	&pandn	("xmm1","xmm0");
+	&psrld	("xmm1",4);
+	&pand	("xmm0","xmm2");
+	&movdqa	("xmm2",&QWP(0,$base));
+	&pshufb	("xmm2","xmm0");
+	&movdqa	("xmm0",&QWP(16,$base));
+	&pshufb	("xmm0","xmm1");
+	&pxor	("xmm0","xmm2");
+	&ret	();
+&function_end_B("_vpaes_schedule_transform");
+
+##
+##  .aes_schedule_mangle
+##
+##  Mangle xmm0 from (basis-transformed) standard version
+##  to our version.
+##
+##  On encrypt,
+##    xor with 0x63
+##    multiply by circulant 0,1,1,1
+##    apply shiftrows transform
+##
+##  On decrypt,
+##    xor with 0x63
+##    multiply by "inverse mixcolumns" circulant E,B,D,9
+##    deskew
+##    apply shiftrows transform
+##
+##
+##  Writes out to (%edx), and increments or decrements it
+##  Keeps track of round number mod 4 in %ecx
+##  Preserves xmm0
+##  Clobbers xmm1-xmm5
+##
+&function_begin_B("_vpaes_schedule_mangle");
+	&movdqa	("xmm4","xmm0");	# save xmm0 for later
+	&movdqa	("xmm5",&QWP($k_mc_forward,$const));
+	&test	($out,$out);
+	&jnz	(&label("schedule_mangle_dec"));
+
+	# encrypting
+	&add	($key,16);
+	&pxor	("xmm4",&QWP($k_s63,$const));
+	&pshufb	("xmm4","xmm5");
+	&movdqa	("xmm3","xmm4");
+	&pshufb	("xmm4","xmm5");
+	&pxor	("xmm3","xmm4");
+	&pshufb	("xmm4","xmm5");
+	&pxor	("xmm3","xmm4");
+
+	&jmp	(&label("schedule_mangle_both"));
+
+&set_label("schedule_mangle_dec",16);
+	# inverse mix columns
+	&movdqa	("xmm2",&QWP($k_s0F,$const));
+	&lea	($inp,&DWP($k_dksd,$const));
+	&movdqa	("xmm1","xmm2");
+	&pandn	("xmm1","xmm4");
+	&psrld	("xmm1",4);			# 1 = hi
+	&pand	("xmm4","xmm2");		# 4 = lo
+
+	&movdqa	("xmm2",&QWP(0,$inp));
+	&pshufb	("xmm2","xmm4");
+	&movdqa	("xmm3",&QWP(0x10,$inp));
+	&pshufb	("xmm3","xmm1");
+	&pxor	("xmm3","xmm2");
+	&pshufb	("xmm3","xmm5");
+
+	&movdqa	("xmm2",&QWP(0x20,$inp));
+	&pshufb	("xmm2","xmm4");
+	&pxor	("xmm2","xmm3");
+	&movdqa	("xmm3",&QWP(0x30,$inp));
+	&pshufb	("xmm3","xmm1");
+	&pxor	("xmm3","xmm2");
+	&pshufb	("xmm3","xmm5");
+
+	&movdqa	("xmm2",&QWP(0x40,$inp));
+	&pshufb	("xmm2","xmm4");
+	&pxor	("xmm2","xmm3");
+	&movdqa	("xmm3",&QWP(0x50,$inp));
+	&pshufb	("xmm3","xmm1");
+	&pxor	("xmm3","xmm2");
+	&pshufb	("xmm3","xmm5");
+
+	&movdqa	("xmm2",&QWP(0x60,$inp));
+	&pshufb	("xmm2","xmm4");
+	&pxor	("xmm2","xmm3");
+	&movdqa	("xmm3",&QWP(0x70,$inp));
+	&pshufb	("xmm3","xmm1");
+	&pxor	("xmm3","xmm2");
+
+	&add	($key,-16);
+
+&set_label("schedule_mangle_both");
+	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
+	&pshufb	("xmm3","xmm1");
+	&add	($magic,-16);
+	&and	($magic,0x30);
+	&movdqu	(&QWP(0,$key),"xmm3");
+	&ret	();
+&function_end_B("_vpaes_schedule_mangle");
+
+#
+# Interface to OpenSSL
+#
+&function_begin("${PREFIX}_set_encrypt_key");
+	&mov	($inp,&wparam(0));		# inp
+	&lea	($base,&DWP(-56,"esp"));
+	&mov	($round,&wparam(1));		# bits
+	&and	($base,-16);
+	&mov	($key,&wparam(2));		# key
+	&xchg	($base,"esp");			# alloca
+	&mov	(&DWP(48,"esp"),$base);
+
+	&mov	($base,$round);
+	&shr	($base,5);
+	&add	($base,5);
+	&mov	(&DWP(240,$key),$base);		# AES_KEY->rounds = nbits/32+5;
+	&mov	($magic,0x30);
+	&mov	($out,0);
+
+	&lea	($const,&DWP(&label("_vpaes_consts")."+0x30-".&label("pic_point")));
+	&call	("_vpaes_schedule_core");
+&set_label("pic_point");
+
+	&mov	("esp",&DWP(48,"esp"));
+	&xor	("eax","eax");
+&function_end("${PREFIX}_set_encrypt_key");
+
+&function_begin("${PREFIX}_set_decrypt_key");
+	&mov	($inp,&wparam(0));		# inp
+	&lea	($base,&DWP(-56,"esp"));
+	&mov	($round,&wparam(1));		# bits
+	&and	($base,-16);
+	&mov	($key,&wparam(2));		# key
+	&xchg	($base,"esp");			# alloca
+	&mov	(&DWP(48,"esp"),$base);
+
+	&mov	($base,$round);
+	&shr	($base,5);
+	&add	($base,5);
+	&mov	(&DWP(240,$key),$base);	# AES_KEY->rounds = nbits/32+5;
+	&shl	($base,4);
+	&lea	($key,&DWP(16,$key,$base));
+
+	&mov	($out,1);
+	&mov	($magic,$round);
+	&shr	($magic,1);
+	&and	($magic,32);
+	&xor	($magic,32);			# nbist==192?0:32;
+
+	&lea	($const,&DWP(&label("_vpaes_consts")."+0x30-".&label("pic_point")));
+	&call	("_vpaes_schedule_core");
+&set_label("pic_point");
+
+	&mov	("esp",&DWP(48,"esp"));
+	&xor	("eax","eax");
+&function_end("${PREFIX}_set_decrypt_key");
+
+&function_begin("${PREFIX}_encrypt");
+	&lea	($const,&DWP(&label("_vpaes_consts")."+0x30-".&label("pic_point")));
+	&call	("_vpaes_preheat");
+&set_label("pic_point");
+	&mov	($inp,&wparam(0));		# inp
+	&lea	($base,&DWP(-56,"esp"));
+	&mov	($out,&wparam(1));		# out
+	&and	($base,-16);
+	&mov	($key,&wparam(2));		# key
+	&xchg	($base,"esp");			# alloca
+	&mov	(&DWP(48,"esp"),$base);
+
+	&movdqu	("xmm0",&QWP(0,$inp));
+	&call	("_vpaes_encrypt_core");
+	&movdqu	(&QWP(0,$out),"xmm0");
+
+	&mov	("esp",&DWP(48,"esp"));
+&function_end("${PREFIX}_encrypt");
+
+&function_begin("${PREFIX}_decrypt");
+	&lea	($const,&DWP(&label("_vpaes_consts")."+0x30-".&label("pic_point")));
+	&call	("_vpaes_preheat");
+&set_label("pic_point");
+	&mov	($inp,&wparam(0));		# inp
+	&lea	($base,&DWP(-56,"esp"));
+	&mov	($out,&wparam(1));		# out
+	&and	($base,-16);
+	&mov	($key,&wparam(2));		# key
+	&xchg	($base,"esp");			# alloca
+	&mov	(&DWP(48,"esp"),$base);
+
+	&movdqu	("xmm0",&QWP(0,$inp));
+	&call	("_vpaes_decrypt_core");
+	&movdqu	(&QWP(0,$out),"xmm0");
+
+	&mov	("esp",&DWP(48,"esp"));
+&function_end("${PREFIX}_decrypt");
+
+&function_begin("${PREFIX}_cbc_encrypt");
+	&mov	($inp,&wparam(0));		# inp
+	&mov	($out,&wparam(1));		# out
+	&mov	($round,&wparam(2));		# len
+	&mov	($key,&wparam(3));		# key
+	&sub	($round,16);
+	&jc	(&label("cbc_abort"));
+	&lea	($base,&DWP(-56,"esp"));
+	&mov	($const,&wparam(4));		# ivp
+	&and	($base,-16);
+	&mov	($magic,&wparam(5));		# enc
+	&xchg	($base,"esp");			# alloca
+	&movdqu	("xmm1",&QWP(0,$const));	# load IV
+	&sub	($out,$inp);
+	&mov	(&DWP(48,"esp"),$base);
+
+	&mov	(&DWP(0,"esp"),$out);		# save out
+	&mov	(&DWP(4,"esp"),$key)		# save key
+	&mov	(&DWP(8,"esp"),$const);		# save ivp
+	&mov	($out,$round);			# $out works as $len
+
+	&lea	($const,&DWP(&label("_vpaes_consts")."+0x30-".&label("pic_point")));
+	&call	("_vpaes_preheat");
+&set_label("pic_point");
+	&cmp	($magic,0);
+	&je	(&label("cbc_dec_loop"));
+	&jmp	(&label("cbc_enc_loop"));
+
+&set_label("cbc_enc_loop",16);
+	&movdqu	("xmm0",&QWP(0,$inp));		# load input
+	&pxor	("xmm0","xmm1");		# inp^=iv
+	&call	("_vpaes_encrypt_core");
+	&mov	($base,&DWP(0,"esp"));		# restore out
+	&mov	($key,&DWP(4,"esp"));		# restore key
+	&movdqa	("xmm1","xmm0");
+	&movdqu	(&QWP(0,$base,$inp),"xmm0");	# write output
+	&lea	($inp,&DWP(16,$inp));
+	&sub	($out,16);
+	&jnc	(&label("cbc_enc_loop"));
+	&jmp	(&label("cbc_done"));
+
+&set_label("cbc_dec_loop",16);
+	&movdqu	("xmm0",&QWP(0,$inp));		# load input
+	&movdqa	(&QWP(16,"esp"),"xmm1");	# save IV
+	&movdqa	(&QWP(32,"esp"),"xmm0");	# save future IV
+	&call	("_vpaes_decrypt_core");
+	&mov	($base,&DWP(0,"esp"));		# restore out
+	&mov	($key,&DWP(4,"esp"));		# restore key
+	&pxor	("xmm0",&QWP(16,"esp"));	# out^=iv
+	&movdqa	("xmm1",&QWP(32,"esp"));	# load next IV
+	&movdqu	(&QWP(0,$base,$inp),"xmm0");	# write output
+	&lea	($inp,&DWP(16,$inp));
+	&sub	($out,16);
+	&jnc	(&label("cbc_dec_loop"));
+
+&set_label("cbc_done");
+	&mov	($base,&DWP(8,"esp"));		# restore ivp
+	&mov	("esp",&DWP(48,"esp"));
+	&movdqu	(&QWP(0,$base),"xmm1");		# write IV
+&set_label("cbc_abort");
+&function_end("${PREFIX}_cbc_encrypt");
+
+&asm_finish();
diff --git a/src/crypto/aes/asm/vpaes-x86_64.pl b/src/crypto/aes/asm/vpaes-x86_64.pl
new file mode 100644
index 0000000..f2ef318
--- /dev/null
+++ b/src/crypto/aes/asm/vpaes-x86_64.pl
@@ -0,0 +1,1207 @@
+#!/usr/bin/env perl
+
+######################################################################
+## Constant-time SSSE3 AES core implementation.
+## version 0.1
+##
+## By Mike Hamburg (Stanford University), 2009
+## Public domain.
+##
+## For details see http://shiftleft.org/papers/vector_aes/ and
+## http://crypto.stanford.edu/vpaes/.
+
+######################################################################
+# September 2011.
+#
+# Interface to OpenSSL as "almost" drop-in replacement for
+# aes-x86_64.pl. "Almost" refers to the fact that AES_cbc_encrypt
+# doesn't handle partial vectors (doesn't have to if called from
+# EVP only). "Drop-in" implies that this module doesn't share key
+# schedule structure with the original nor does it make assumption
+# about its alignment...
+#
+# Performance summary. aes-x86_64.pl column lists large-block CBC
+# encrypt/decrypt/with-hyper-threading-off(*) results in cycles per
+# byte processed with 128-bit key, and vpaes-x86_64.pl column -
+# [also large-block CBC] encrypt/decrypt.
+#
+#		aes-x86_64.pl		vpaes-x86_64.pl
+#
+# Core 2(**)	29.6/41.1/14.3		21.9/25.2(***)
+# Nehalem	29.6/40.3/14.6		10.0/11.8
+# Atom		57.3/74.2/32.1		60.9/77.2(***)
+# Silvermont	52.7/64.0/19.5		48.8/60.8(***)
+#
+# (*)	"Hyper-threading" in the context refers rather to cache shared
+#	among multiple cores, than to specifically Intel HTT. As vast
+#	majority of contemporary cores share cache, slower code path
+#	is common place. In other words "with-hyper-threading-off"
+#	results are presented mostly for reference purposes.
+#
+# (**)	"Core 2" refers to initial 65nm design, a.k.a. Conroe.
+#
+# (***)	Less impressive improvement on Core 2 and Atom is due to slow
+#	pshufb,	yet it's respectable +36%/62% improvement on Core 2
+#	(as implied, over "hyper-threading-safe" code path).
+#
+#						<appro@openssl.org>
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+$PREFIX="vpaes";
+
+$code.=<<___;
+.text
+
+##
+##  _aes_encrypt_core
+##
+##  AES-encrypt %xmm0.
+##
+##  Inputs:
+##     %xmm0 = input
+##     %xmm9-%xmm15 as in _vpaes_preheat
+##    (%rdx) = scheduled keys
+##
+##  Output in %xmm0
+##  Clobbers  %xmm1-%xmm5, %r9, %r10, %r11, %rax
+##  Preserves %xmm6 - %xmm8 so you get some local vectors
+##
+##
+.type	_vpaes_encrypt_core,\@abi-omnipotent
+.align 16
+_vpaes_encrypt_core:
+	mov	%rdx,	%r9
+	mov	\$16,	%r11
+	mov	240(%rdx),%eax
+	movdqa	%xmm9,	%xmm1
+	movdqa	.Lk_ipt(%rip), %xmm2	# iptlo
+	pandn	%xmm0,	%xmm1
+	movdqu	(%r9),	%xmm5		# round0 key
+	psrld	\$4,	%xmm1
+	pand	%xmm9,	%xmm0
+	pshufb	%xmm0,	%xmm2
+	movdqa	.Lk_ipt+16(%rip), %xmm0	# ipthi
+	pshufb	%xmm1,	%xmm0
+	pxor	%xmm5,	%xmm2
+	add	\$16,	%r9
+	pxor	%xmm2,	%xmm0
+	lea	.Lk_mc_backward(%rip),%r10
+	jmp	.Lenc_entry
+
+.align 16
+.Lenc_loop:
+	# middle of middle round
+	movdqa  %xmm13,	%xmm4	# 4 : sb1u
+	movdqa  %xmm12,	%xmm0	# 0 : sb1t
+	pshufb  %xmm2,	%xmm4	# 4 = sb1u
+	pshufb  %xmm3,	%xmm0	# 0 = sb1t
+	pxor	%xmm5,	%xmm4	# 4 = sb1u + k
+	movdqa  %xmm15,	%xmm5	# 4 : sb2u
+	pxor	%xmm4,	%xmm0	# 0 = A
+	movdqa	-0x40(%r11,%r10), %xmm1		# .Lk_mc_forward[]
+	pshufb	%xmm2,	%xmm5	# 4 = sb2u
+	movdqa	(%r11,%r10), %xmm4		# .Lk_mc_backward[]
+	movdqa	%xmm14, %xmm2	# 2 : sb2t
+	pshufb	%xmm3,  %xmm2	# 2 = sb2t
+	movdqa	%xmm0,  %xmm3	# 3 = A
+	pxor	%xmm5,	%xmm2	# 2 = 2A
+	pshufb  %xmm1,  %xmm0	# 0 = B
+	add	\$16,	%r9	# next key
+	pxor	%xmm2,  %xmm0	# 0 = 2A+B
+	pshufb	%xmm4,	%xmm3	# 3 = D
+	add	\$16,	%r11	# next mc
+	pxor	%xmm0,	%xmm3	# 3 = 2A+B+D
+	pshufb  %xmm1,	%xmm0	# 0 = 2B+C
+	and	\$0x30,	%r11	# ... mod 4
+	sub	\$1,%rax	# nr--
+	pxor	%xmm3,	%xmm0	# 0 = 2A+3B+C+D
+
+.Lenc_entry:
+	# top of round
+	movdqa  %xmm9, 	%xmm1	# 1 : i
+	movdqa	%xmm11, %xmm5	# 2 : a/k
+	pandn	%xmm0, 	%xmm1	# 1 = i<<4
+	psrld	\$4,   	%xmm1   # 1 = i
+	pand	%xmm9, 	%xmm0   # 0 = k
+	pshufb  %xmm0,  %xmm5	# 2 = a/k
+	movdqa	%xmm10,	%xmm3  	# 3 : 1/i
+	pxor	%xmm1,	%xmm0	# 0 = j
+	pshufb  %xmm1, 	%xmm3  	# 3 = 1/i
+	movdqa	%xmm10,	%xmm4  	# 4 : 1/j
+	pxor	%xmm5, 	%xmm3  	# 3 = iak = 1/i + a/k
+	pshufb	%xmm0, 	%xmm4  	# 4 = 1/j
+	movdqa	%xmm10,	%xmm2  	# 2 : 1/iak
+	pxor	%xmm5, 	%xmm4  	# 4 = jak = 1/j + a/k
+	pshufb  %xmm3,	%xmm2  	# 2 = 1/iak
+	movdqa	%xmm10, %xmm3   # 3 : 1/jak
+	pxor	%xmm0, 	%xmm2  	# 2 = io
+	pshufb  %xmm4,  %xmm3   # 3 = 1/jak
+	movdqu	(%r9),	%xmm5
+	pxor	%xmm1,  %xmm3   # 3 = jo
+	jnz	.Lenc_loop
+
+	# middle of last round
+	movdqa	-0x60(%r10), %xmm4	# 3 : sbou	.Lk_sbo
+	movdqa	-0x50(%r10), %xmm0	# 0 : sbot	.Lk_sbo+16
+	pshufb  %xmm2,  %xmm4	# 4 = sbou
+	pxor	%xmm5,  %xmm4	# 4 = sb1u + k
+	pshufb  %xmm3,	%xmm0	# 0 = sb1t
+	movdqa	0x40(%r11,%r10), %xmm1		# .Lk_sr[]
+	pxor	%xmm4,	%xmm0	# 0 = A
+	pshufb	%xmm1,	%xmm0
+	ret
+.size	_vpaes_encrypt_core,.-_vpaes_encrypt_core
+	
+##
+##  Decryption core
+##
+##  Same API as encryption core.
+##
+.type	_vpaes_decrypt_core,\@abi-omnipotent
+.align	16
+_vpaes_decrypt_core:
+	mov	%rdx,	%r9		# load key
+	mov	240(%rdx),%eax
+	movdqa	%xmm9,	%xmm1
+	movdqa	.Lk_dipt(%rip), %xmm2	# iptlo
+	pandn	%xmm0,	%xmm1
+	mov	%rax,	%r11
+	psrld	\$4,	%xmm1
+	movdqu	(%r9),	%xmm5		# round0 key
+	shl	\$4,	%r11
+	pand	%xmm9,	%xmm0
+	pshufb	%xmm0,	%xmm2
+	movdqa	.Lk_dipt+16(%rip), %xmm0 # ipthi
+	xor	\$0x30,	%r11
+	lea	.Lk_dsbd(%rip),%r10
+	pshufb	%xmm1,	%xmm0
+	and	\$0x30,	%r11
+	pxor	%xmm5,	%xmm2
+	movdqa	.Lk_mc_forward+48(%rip), %xmm5
+	pxor	%xmm2,	%xmm0
+	add	\$16,	%r9
+	add	%r10,	%r11
+	jmp	.Ldec_entry
+
+.align 16
+.Ldec_loop:
+##
+##  Inverse mix columns
+##
+	movdqa  -0x20(%r10),%xmm4	# 4 : sb9u
+	movdqa  -0x10(%r10),%xmm1	# 0 : sb9t
+	pshufb	%xmm2,	%xmm4		# 4 = sb9u
+	pshufb	%xmm3,	%xmm1		# 0 = sb9t
+	pxor	%xmm4,	%xmm0
+	movdqa  0x00(%r10),%xmm4	# 4 : sbdu
+	pxor	%xmm1,	%xmm0		# 0 = ch
+	movdqa  0x10(%r10),%xmm1	# 0 : sbdt
+
+	pshufb	%xmm2,	%xmm4		# 4 = sbdu
+	pshufb	%xmm5,	%xmm0		# MC ch
+	pshufb	%xmm3,	%xmm1		# 0 = sbdt
+	pxor	%xmm4,	%xmm0		# 4 = ch
+	movdqa  0x20(%r10),%xmm4	# 4 : sbbu
+	pxor	%xmm1,	%xmm0		# 0 = ch
+	movdqa  0x30(%r10),%xmm1	# 0 : sbbt
+
+	pshufb	%xmm2,	%xmm4		# 4 = sbbu
+	pshufb	%xmm5,	%xmm0		# MC ch
+	pshufb	%xmm3,	%xmm1		# 0 = sbbt
+	pxor	%xmm4,	%xmm0		# 4 = ch
+	movdqa  0x40(%r10),%xmm4	# 4 : sbeu
+	pxor	%xmm1,	%xmm0		# 0 = ch
+	movdqa  0x50(%r10),%xmm1	# 0 : sbet
+
+	pshufb	%xmm2,	%xmm4		# 4 = sbeu
+	pshufb	%xmm5,	%xmm0		# MC ch
+	pshufb	%xmm3,	%xmm1		# 0 = sbet
+	pxor	%xmm4,	%xmm0		# 4 = ch
+	add	\$16, %r9		# next round key
+	palignr	\$12,	%xmm5,	%xmm5
+	pxor	%xmm1,	%xmm0		# 0 = ch
+	sub	\$1,%rax		# nr--
+
+.Ldec_entry:
+	# top of round
+	movdqa  %xmm9, 	%xmm1	# 1 : i
+	pandn	%xmm0, 	%xmm1	# 1 = i<<4
+	movdqa	%xmm11, %xmm2	# 2 : a/k
+	psrld	\$4,    %xmm1	# 1 = i
+	pand	%xmm9, 	%xmm0	# 0 = k
+	pshufb  %xmm0,  %xmm2	# 2 = a/k
+	movdqa	%xmm10,	%xmm3	# 3 : 1/i
+	pxor	%xmm1,	%xmm0	# 0 = j
+	pshufb  %xmm1, 	%xmm3	# 3 = 1/i
+	movdqa	%xmm10,	%xmm4	# 4 : 1/j
+	pxor	%xmm2, 	%xmm3	# 3 = iak = 1/i + a/k
+	pshufb	%xmm0, 	%xmm4	# 4 = 1/j
+	pxor	%xmm2, 	%xmm4	# 4 = jak = 1/j + a/k
+	movdqa	%xmm10,	%xmm2	# 2 : 1/iak
+	pshufb  %xmm3,	%xmm2	# 2 = 1/iak
+	movdqa	%xmm10, %xmm3	# 3 : 1/jak
+	pxor	%xmm0, 	%xmm2	# 2 = io
+	pshufb  %xmm4,  %xmm3	# 3 = 1/jak
+	movdqu	(%r9),	%xmm0
+	pxor	%xmm1,  %xmm3	# 3 = jo
+	jnz	.Ldec_loop
+
+	# middle of last round
+	movdqa	0x60(%r10), %xmm4	# 3 : sbou
+	pshufb  %xmm2,  %xmm4	# 4 = sbou
+	pxor	%xmm0,  %xmm4	# 4 = sb1u + k
+	movdqa	0x70(%r10), %xmm0	# 0 : sbot
+	movdqa	-0x160(%r11), %xmm2	# .Lk_sr-.Lk_dsbd=-0x160
+	pshufb  %xmm3,	%xmm0	# 0 = sb1t
+	pxor	%xmm4,	%xmm0	# 0 = A
+	pshufb	%xmm2,	%xmm0
+	ret
+.size	_vpaes_decrypt_core,.-_vpaes_decrypt_core
+
+########################################################
+##                                                    ##
+##                  AES key schedule                  ##
+##                                                    ##
+########################################################
+.type	_vpaes_schedule_core,\@abi-omnipotent
+.align	16
+_vpaes_schedule_core:
+	# rdi = key
+	# rsi = size in bits
+	# rdx = buffer
+	# rcx = direction.  0=encrypt, 1=decrypt
+
+	call	_vpaes_preheat		# load the tables
+	movdqa	.Lk_rcon(%rip), %xmm8	# load rcon
+	movdqu	(%rdi),	%xmm0		# load key (unaligned)
+
+	# input transform
+	movdqa	%xmm0,	%xmm3
+	lea	.Lk_ipt(%rip), %r11
+	call	_vpaes_schedule_transform
+	movdqa	%xmm0,	%xmm7
+
+	lea	.Lk_sr(%rip),%r10
+	test	%rcx,	%rcx
+	jnz	.Lschedule_am_decrypting
+
+	# encrypting, output zeroth round key after transform
+	movdqu	%xmm0,	(%rdx)
+	jmp	.Lschedule_go
+
+.Lschedule_am_decrypting:
+	# decrypting, output zeroth round key after shiftrows
+	movdqa	(%r8,%r10),%xmm1
+	pshufb  %xmm1,	%xmm3
+	movdqu	%xmm3,	(%rdx)
+	xor	\$0x30, %r8
+
+.Lschedule_go:
+	cmp	\$192,	%esi
+	ja	.Lschedule_256
+	je	.Lschedule_192
+	# 128: fall though
+
+##
+##  .schedule_128
+##
+##  128-bit specific part of key schedule.
+##
+##  This schedule is really simple, because all its parts
+##  are accomplished by the subroutines.
+##
+.Lschedule_128:
+	mov	\$10, %esi
+	
+.Loop_schedule_128:
+	call 	_vpaes_schedule_round
+	dec	%rsi
+	jz 	.Lschedule_mangle_last
+	call	_vpaes_schedule_mangle	# write output
+	jmp 	.Loop_schedule_128
+
+##
+##  .aes_schedule_192
+##
+##  192-bit specific part of key schedule.
+##
+##  The main body of this schedule is the same as the 128-bit
+##  schedule, but with more smearing.  The long, high side is
+##  stored in %xmm7 as before, and the short, low side is in
+##  the high bits of %xmm6.
+##
+##  This schedule is somewhat nastier, however, because each
+##  round produces 192 bits of key material, or 1.5 round keys.
+##  Therefore, on each cycle we do 2 rounds and produce 3 round
+##  keys.
+##
+.align	16
+.Lschedule_192:
+	movdqu	8(%rdi),%xmm0		# load key part 2 (very unaligned)
+	call	_vpaes_schedule_transform	# input transform
+	movdqa	%xmm0,	%xmm6		# save short part
+	pxor	%xmm4,	%xmm4		# clear 4
+	movhlps	%xmm4,	%xmm6		# clobber low side with zeros
+	mov	\$4,	%esi
+
+.Loop_schedule_192:
+	call	_vpaes_schedule_round
+	palignr	\$8,%xmm6,%xmm0	
+	call	_vpaes_schedule_mangle	# save key n
+	call	_vpaes_schedule_192_smear
+	call	_vpaes_schedule_mangle	# save key n+1
+	call	_vpaes_schedule_round
+	dec	%rsi
+	jz 	.Lschedule_mangle_last
+	call	_vpaes_schedule_mangle	# save key n+2
+	call	_vpaes_schedule_192_smear
+	jmp	.Loop_schedule_192
+
+##
+##  .aes_schedule_256
+##
+##  256-bit specific part of key schedule.
+##
+##  The structure here is very similar to the 128-bit
+##  schedule, but with an additional "low side" in
+##  %xmm6.  The low side's rounds are the same as the
+##  high side's, except no rcon and no rotation.
+##
+.align	16
+.Lschedule_256:
+	movdqu	16(%rdi),%xmm0		# load key part 2 (unaligned)
+	call	_vpaes_schedule_transform	# input transform
+	mov	\$7, %esi
+	
+.Loop_schedule_256:
+	call	_vpaes_schedule_mangle	# output low result
+	movdqa	%xmm0,	%xmm6		# save cur_lo in xmm6
+
+	# high round
+	call	_vpaes_schedule_round
+	dec	%rsi
+	jz 	.Lschedule_mangle_last
+	call	_vpaes_schedule_mangle	
+
+	# low round. swap xmm7 and xmm6
+	pshufd	\$0xFF,	%xmm0,	%xmm0
+	movdqa	%xmm7,	%xmm5
+	movdqa	%xmm6,	%xmm7
+	call	_vpaes_schedule_low_round
+	movdqa	%xmm5,	%xmm7
+	
+	jmp	.Loop_schedule_256
+
+	
+##
+##  .aes_schedule_mangle_last
+##
+##  Mangler for last round of key schedule
+##  Mangles %xmm0
+##    when encrypting, outputs out(%xmm0) ^ 63
+##    when decrypting, outputs unskew(%xmm0)
+##
+##  Always called right before return... jumps to cleanup and exits
+##
+.align	16
+.Lschedule_mangle_last:
+	# schedule last round key from xmm0
+	lea	.Lk_deskew(%rip),%r11	# prepare to deskew
+	test	%rcx, 	%rcx
+	jnz	.Lschedule_mangle_last_dec
+
+	# encrypting
+	movdqa	(%r8,%r10),%xmm1
+	pshufb	%xmm1,	%xmm0		# output permute
+	lea	.Lk_opt(%rip),	%r11	# prepare to output transform
+	add	\$32,	%rdx
+
+.Lschedule_mangle_last_dec:
+	add	\$-16,	%rdx
+	pxor	.Lk_s63(%rip),	%xmm0
+	call	_vpaes_schedule_transform # output transform
+	movdqu	%xmm0,	(%rdx)		# save last key
+
+	# cleanup
+	pxor	%xmm0,  %xmm0
+	pxor	%xmm1,  %xmm1
+	pxor	%xmm2,  %xmm2
+	pxor	%xmm3,  %xmm3
+	pxor	%xmm4,  %xmm4
+	pxor	%xmm5,  %xmm5
+	pxor	%xmm6,  %xmm6
+	pxor	%xmm7,  %xmm7
+	ret
+.size	_vpaes_schedule_core,.-_vpaes_schedule_core
+
+##
+##  .aes_schedule_192_smear
+##
+##  Smear the short, low side in the 192-bit key schedule.
+##
+##  Inputs:
+##    %xmm7: high side, b  a  x  y
+##    %xmm6:  low side, d  c  0  0
+##    %xmm13: 0
+##
+##  Outputs:
+##    %xmm6: b+c+d  b+c  0  0
+##    %xmm0: b+c+d  b+c  b  a
+##
+.type	_vpaes_schedule_192_smear,\@abi-omnipotent
+.align	16
+_vpaes_schedule_192_smear:
+	pshufd	\$0x80,	%xmm6,	%xmm1	# d c 0 0 -> c 0 0 0
+	pshufd	\$0xFE,	%xmm7,	%xmm0	# b a _ _ -> b b b a
+	pxor	%xmm1,	%xmm6		# -> c+d c 0 0
+	pxor	%xmm1,	%xmm1
+	pxor	%xmm0,	%xmm6		# -> b+c+d b+c b a
+	movdqa	%xmm6,	%xmm0
+	movhlps	%xmm1,	%xmm6		# clobber low side with zeros
+	ret
+.size	_vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear
+
+##
+##  .aes_schedule_round
+##
+##  Runs one main round of the key schedule on %xmm0, %xmm7
+##
+##  Specifically, runs subbytes on the high dword of %xmm0
+##  then rotates it by one byte and xors into the low dword of
+##  %xmm7.
+##
+##  Adds rcon from low byte of %xmm8, then rotates %xmm8 for
+##  next rcon.
+##
+##  Smears the dwords of %xmm7 by xoring the low into the
+##  second low, result into third, result into highest.
+##
+##  Returns results in %xmm7 = %xmm0.
+##  Clobbers %xmm1-%xmm4, %r11.
+##
+.type	_vpaes_schedule_round,\@abi-omnipotent
+.align	16
+_vpaes_schedule_round:
+	# extract rcon from xmm8
+	pxor	%xmm1,	%xmm1
+	palignr	\$15,	%xmm8,	%xmm1
+	palignr	\$15,	%xmm8,	%xmm8
+	pxor	%xmm1,	%xmm7
+
+	# rotate
+	pshufd	\$0xFF,	%xmm0,	%xmm0
+	palignr	\$1,	%xmm0,	%xmm0
+	
+	# fall through...
+	
+	# low round: same as high round, but no rotation and no rcon.
+_vpaes_schedule_low_round:
+	# smear xmm7
+	movdqa	%xmm7,	%xmm1
+	pslldq	\$4,	%xmm7
+	pxor	%xmm1,	%xmm7
+	movdqa	%xmm7,	%xmm1
+	pslldq	\$8,	%xmm7
+	pxor	%xmm1,	%xmm7
+	pxor	.Lk_s63(%rip), %xmm7
+
+	# subbytes
+	movdqa  %xmm9, 	%xmm1
+	pandn	%xmm0, 	%xmm1
+	psrld	\$4,    %xmm1		# 1 = i
+	pand	%xmm9, 	%xmm0		# 0 = k
+	movdqa	%xmm11, %xmm2		# 2 : a/k
+	pshufb  %xmm0,  %xmm2		# 2 = a/k
+	pxor	%xmm1,	%xmm0		# 0 = j
+	movdqa	%xmm10,	%xmm3		# 3 : 1/i
+	pshufb  %xmm1, 	%xmm3		# 3 = 1/i
+	pxor	%xmm2, 	%xmm3		# 3 = iak = 1/i + a/k
+	movdqa	%xmm10,	%xmm4		# 4 : 1/j
+	pshufb	%xmm0, 	%xmm4		# 4 = 1/j
+	pxor	%xmm2, 	%xmm4		# 4 = jak = 1/j + a/k
+	movdqa	%xmm10,	%xmm2		# 2 : 1/iak
+	pshufb  %xmm3,	%xmm2		# 2 = 1/iak
+	pxor	%xmm0, 	%xmm2		# 2 = io
+	movdqa	%xmm10, %xmm3		# 3 : 1/jak
+	pshufb  %xmm4,  %xmm3		# 3 = 1/jak
+	pxor	%xmm1,  %xmm3		# 3 = jo
+	movdqa	%xmm13, %xmm4		# 4 : sbou
+	pshufb  %xmm2,  %xmm4		# 4 = sbou
+	movdqa	%xmm12, %xmm0		# 0 : sbot
+	pshufb  %xmm3,	%xmm0		# 0 = sb1t
+	pxor	%xmm4, 	%xmm0		# 0 = sbox output
+
+	# add in smeared stuff
+	pxor	%xmm7,	%xmm0	
+	movdqa	%xmm0,	%xmm7
+	ret
+.size	_vpaes_schedule_round,.-_vpaes_schedule_round
+
+##
+##  .aes_schedule_transform
+##
+##  Linear-transform %xmm0 according to tables at (%r11)
+##
+##  Requires that %xmm9 = 0x0F0F... as in preheat
+##  Output in %xmm0
+##  Clobbers %xmm1, %xmm2
+##
+.type	_vpaes_schedule_transform,\@abi-omnipotent
+.align	16
+_vpaes_schedule_transform:
+	movdqa	%xmm9,	%xmm1
+	pandn	%xmm0,	%xmm1
+	psrld	\$4,	%xmm1
+	pand	%xmm9,	%xmm0
+	movdqa	(%r11), %xmm2 	# lo
+	pshufb	%xmm0,	%xmm2
+	movdqa	16(%r11), %xmm0 # hi
+	pshufb	%xmm1,	%xmm0
+	pxor	%xmm2,	%xmm0
+	ret
+.size	_vpaes_schedule_transform,.-_vpaes_schedule_transform
+
+##
+##  .aes_schedule_mangle
+##
+##  Mangle xmm0 from (basis-transformed) standard version
+##  to our version.
+##
+##  On encrypt,
+##    xor with 0x63
+##    multiply by circulant 0,1,1,1
+##    apply shiftrows transform
+##
+##  On decrypt,
+##    xor with 0x63
+##    multiply by "inverse mixcolumns" circulant E,B,D,9
+##    deskew
+##    apply shiftrows transform
+##
+##
+##  Writes out to (%rdx), and increments or decrements it
+##  Keeps track of round number mod 4 in %r8
+##  Preserves xmm0
+##  Clobbers xmm1-xmm5
+##
+.type	_vpaes_schedule_mangle,\@abi-omnipotent
+.align	16
+_vpaes_schedule_mangle:
+	movdqa	%xmm0,	%xmm4	# save xmm0 for later
+	movdqa	.Lk_mc_forward(%rip),%xmm5
+	test	%rcx, 	%rcx
+	jnz	.Lschedule_mangle_dec
+
+	# encrypting
+	add	\$16,	%rdx
+	pxor	.Lk_s63(%rip),%xmm4
+	pshufb	%xmm5,	%xmm4
+	movdqa	%xmm4,	%xmm3
+	pshufb	%xmm5,	%xmm4
+	pxor	%xmm4,	%xmm3
+	pshufb	%xmm5,	%xmm4
+	pxor	%xmm4,	%xmm3
+
+	jmp	.Lschedule_mangle_both
+.align	16
+.Lschedule_mangle_dec:
+	# inverse mix columns
+	lea	.Lk_dksd(%rip),%r11
+	movdqa	%xmm9,	%xmm1
+	pandn	%xmm4,	%xmm1
+	psrld	\$4,	%xmm1	# 1 = hi
+	pand	%xmm9,	%xmm4	# 4 = lo
+
+	movdqa	0x00(%r11), %xmm2
+	pshufb	%xmm4,	%xmm2
+	movdqa	0x10(%r11), %xmm3
+	pshufb	%xmm1,	%xmm3
+	pxor	%xmm2,	%xmm3
+	pshufb	%xmm5,	%xmm3
+
+	movdqa	0x20(%r11), %xmm2
+	pshufb	%xmm4,	%xmm2
+	pxor	%xmm3,	%xmm2
+	movdqa	0x30(%r11), %xmm3
+	pshufb	%xmm1,	%xmm3
+	pxor	%xmm2,	%xmm3
+	pshufb	%xmm5,	%xmm3
+
+	movdqa	0x40(%r11), %xmm2
+	pshufb	%xmm4,	%xmm2
+	pxor	%xmm3,	%xmm2
+	movdqa	0x50(%r11), %xmm3
+	pshufb	%xmm1,	%xmm3
+	pxor	%xmm2,	%xmm3
+	pshufb	%xmm5,	%xmm3
+
+	movdqa	0x60(%r11), %xmm2
+	pshufb	%xmm4,	%xmm2
+	pxor	%xmm3,	%xmm2
+	movdqa	0x70(%r11), %xmm3
+	pshufb	%xmm1,	%xmm3
+	pxor	%xmm2,	%xmm3
+
+	add	\$-16,	%rdx
+
+.Lschedule_mangle_both:
+	movdqa	(%r8,%r10),%xmm1
+	pshufb	%xmm1,%xmm3
+	add	\$-16,	%r8
+	and	\$0x30,	%r8
+	movdqu	%xmm3,	(%rdx)
+	ret
+.size	_vpaes_schedule_mangle,.-_vpaes_schedule_mangle
+
+#
+# Interface to OpenSSL
+#
+.globl	${PREFIX}_set_encrypt_key
+.type	${PREFIX}_set_encrypt_key,\@function,3
+.align	16
+${PREFIX}_set_encrypt_key:
+___
+$code.=<<___ if ($win64);
+	lea	-0xb8(%rsp),%rsp
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Lenc_key_body:
+___
+$code.=<<___;
+	mov	%esi,%eax
+	shr	\$5,%eax
+	add	\$5,%eax
+	mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
+
+	mov	\$0,%ecx
+	mov	\$0x30,%r8d
+	call	_vpaes_schedule_core
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+	lea	0xb8(%rsp),%rsp
+.Lenc_key_epilogue:
+___
+$code.=<<___;
+	xor	%eax,%eax
+	ret
+.size	${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
+
+.globl	${PREFIX}_set_decrypt_key
+.type	${PREFIX}_set_decrypt_key,\@function,3
+.align	16
+${PREFIX}_set_decrypt_key:
+___
+$code.=<<___ if ($win64);
+	lea	-0xb8(%rsp),%rsp
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Ldec_key_body:
+___
+$code.=<<___;
+	mov	%esi,%eax
+	shr	\$5,%eax
+	add	\$5,%eax
+	mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
+	shl	\$4,%eax
+	lea	16(%rdx,%rax),%rdx
+
+	mov	\$1,%ecx
+	mov	%esi,%r8d
+	shr	\$1,%r8d
+	and	\$32,%r8d
+	xor	\$32,%r8d	# nbits==192?0:32
+	call	_vpaes_schedule_core
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+	lea	0xb8(%rsp),%rsp
+.Ldec_key_epilogue:
+___
+$code.=<<___;
+	xor	%eax,%eax
+	ret
+.size	${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
+
+.globl	${PREFIX}_encrypt
+.type	${PREFIX}_encrypt,\@function,3
+.align	16
+${PREFIX}_encrypt:
+___
+$code.=<<___ if ($win64);
+	lea	-0xb8(%rsp),%rsp
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Lenc_body:
+___
+$code.=<<___;
+	movdqu	(%rdi),%xmm0
+	call	_vpaes_preheat
+	call	_vpaes_encrypt_core
+	movdqu	%xmm0,(%rsi)
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+	lea	0xb8(%rsp),%rsp
+.Lenc_epilogue:
+___
+$code.=<<___;
+	ret
+.size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
+
+.globl	${PREFIX}_decrypt
+.type	${PREFIX}_decrypt,\@function,3
+.align	16
+${PREFIX}_decrypt:
+___
+$code.=<<___ if ($win64);
+	lea	-0xb8(%rsp),%rsp
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Ldec_body:
+___
+$code.=<<___;
+	movdqu	(%rdi),%xmm0
+	call	_vpaes_preheat
+	call	_vpaes_decrypt_core
+	movdqu	%xmm0,(%rsi)
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+	lea	0xb8(%rsp),%rsp
+.Ldec_epilogue:
+___
+$code.=<<___;
+	ret
+.size	${PREFIX}_decrypt,.-${PREFIX}_decrypt
+___
+{
+my ($inp,$out,$len,$key,$ivp,$enc)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
+# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
+#                       size_t length, const AES_KEY *key,
+#                       unsigned char *ivp,const int enc);
+$code.=<<___;
+.globl	${PREFIX}_cbc_encrypt
+.type	${PREFIX}_cbc_encrypt,\@function,6
+.align	16
+${PREFIX}_cbc_encrypt:
+	xchg	$key,$len
+___
+($len,$key)=($key,$len);
+$code.=<<___;
+	sub	\$16,$len
+	jc	.Lcbc_abort
+___
+$code.=<<___ if ($win64);
+	lea	-0xb8(%rsp),%rsp
+	movaps	%xmm6,0x10(%rsp)
+	movaps	%xmm7,0x20(%rsp)
+	movaps	%xmm8,0x30(%rsp)
+	movaps	%xmm9,0x40(%rsp)
+	movaps	%xmm10,0x50(%rsp)
+	movaps	%xmm11,0x60(%rsp)
+	movaps	%xmm12,0x70(%rsp)
+	movaps	%xmm13,0x80(%rsp)
+	movaps	%xmm14,0x90(%rsp)
+	movaps	%xmm15,0xa0(%rsp)
+.Lcbc_body:
+___
+$code.=<<___;
+	movdqu	($ivp),%xmm6		# load IV
+	sub	$inp,$out
+	call	_vpaes_preheat
+	cmp	\$0,${enc}d
+	je	.Lcbc_dec_loop
+	jmp	.Lcbc_enc_loop
+.align	16
+.Lcbc_enc_loop:
+	movdqu	($inp),%xmm0
+	pxor	%xmm6,%xmm0
+	call	_vpaes_encrypt_core
+	movdqa	%xmm0,%xmm6
+	movdqu	%xmm0,($out,$inp)
+	lea	16($inp),$inp
+	sub	\$16,$len
+	jnc	.Lcbc_enc_loop
+	jmp	.Lcbc_done
+.align	16
+.Lcbc_dec_loop:
+	movdqu	($inp),%xmm0
+	movdqa	%xmm0,%xmm7
+	call	_vpaes_decrypt_core
+	pxor	%xmm6,%xmm0
+	movdqa	%xmm7,%xmm6
+	movdqu	%xmm0,($out,$inp)
+	lea	16($inp),$inp
+	sub	\$16,$len
+	jnc	.Lcbc_dec_loop
+.Lcbc_done:
+	movdqu	%xmm6,($ivp)		# save IV
+___
+$code.=<<___ if ($win64);
+	movaps	0x10(%rsp),%xmm6
+	movaps	0x20(%rsp),%xmm7
+	movaps	0x30(%rsp),%xmm8
+	movaps	0x40(%rsp),%xmm9
+	movaps	0x50(%rsp),%xmm10
+	movaps	0x60(%rsp),%xmm11
+	movaps	0x70(%rsp),%xmm12
+	movaps	0x80(%rsp),%xmm13
+	movaps	0x90(%rsp),%xmm14
+	movaps	0xa0(%rsp),%xmm15
+	lea	0xb8(%rsp),%rsp
+.Lcbc_epilogue:
+___
+$code.=<<___;
+.Lcbc_abort:
+	ret
+.size	${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
+___
+}
+$code.=<<___;
+##
+##  _aes_preheat
+##
+##  Fills register %r10 -> .aes_consts (so you can -fPIC)
+##  and %xmm9-%xmm15 as specified below.
+##
+.type	_vpaes_preheat,\@abi-omnipotent
+.align	16
+_vpaes_preheat:
+	lea	.Lk_s0F(%rip), %r10
+	movdqa	-0x20(%r10), %xmm10	# .Lk_inv
+	movdqa	-0x10(%r10), %xmm11	# .Lk_inv+16
+	movdqa	0x00(%r10), %xmm9	# .Lk_s0F
+	movdqa	0x30(%r10), %xmm13	# .Lk_sb1
+	movdqa	0x40(%r10), %xmm12	# .Lk_sb1+16
+	movdqa	0x50(%r10), %xmm15	# .Lk_sb2
+	movdqa	0x60(%r10), %xmm14	# .Lk_sb2+16
+	ret
+.size	_vpaes_preheat,.-_vpaes_preheat
+########################################################
+##                                                    ##
+##                     Constants                      ##
+##                                                    ##
+########################################################
+.type	_vpaes_consts,\@object
+.align	64
+_vpaes_consts:
+.Lk_inv:	# inv, inva
+	.quad	0x0E05060F0D080180, 0x040703090A0B0C02
+	.quad	0x01040A060F0B0780, 0x030D0E0C02050809
+
+.Lk_s0F:	# s0F
+	.quad	0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F
+
+.Lk_ipt:	# input transform (lo, hi)
+	.quad	0xC2B2E8985A2A7000, 0xCABAE09052227808
+	.quad	0x4C01307D317C4D00, 0xCD80B1FCB0FDCC81
+
+.Lk_sb1:	# sb1u, sb1t
+	.quad	0xB19BE18FCB503E00, 0xA5DF7A6E142AF544
+	.quad	0x3618D415FAE22300, 0x3BF7CCC10D2ED9EF
+.Lk_sb2:	# sb2u, sb2t
+	.quad	0xE27A93C60B712400, 0x5EB7E955BC982FCD
+	.quad	0x69EB88400AE12900, 0xC2A163C8AB82234A
+.Lk_sbo:	# sbou, sbot
+	.quad	0xD0D26D176FBDC700, 0x15AABF7AC502A878
+	.quad	0xCFE474A55FBB6A00, 0x8E1E90D1412B35FA
+
+.Lk_mc_forward:	# mc_forward
+	.quad	0x0407060500030201, 0x0C0F0E0D080B0A09
+	.quad	0x080B0A0904070605, 0x000302010C0F0E0D
+	.quad	0x0C0F0E0D080B0A09, 0x0407060500030201
+	.quad	0x000302010C0F0E0D, 0x080B0A0904070605
+
+.Lk_mc_backward:# mc_backward
+	.quad	0x0605040702010003, 0x0E0D0C0F0A09080B
+	.quad	0x020100030E0D0C0F, 0x0A09080B06050407
+	.quad	0x0E0D0C0F0A09080B, 0x0605040702010003
+	.quad	0x0A09080B06050407, 0x020100030E0D0C0F
+
+.Lk_sr:		# sr
+	.quad	0x0706050403020100, 0x0F0E0D0C0B0A0908
+	.quad	0x030E09040F0A0500, 0x0B06010C07020D08
+	.quad	0x0F060D040B020900, 0x070E050C030A0108
+	.quad	0x0B0E0104070A0D00, 0x0306090C0F020508
+
+.Lk_rcon:	# rcon
+	.quad	0x1F8391B9AF9DEEB6, 0x702A98084D7C7D81
+
+.Lk_s63:	# s63: all equal to 0x63 transformed
+	.quad	0x5B5B5B5B5B5B5B5B, 0x5B5B5B5B5B5B5B5B
+
+.Lk_opt:	# output transform
+	.quad	0xFF9F4929D6B66000, 0xF7974121DEBE6808
+	.quad	0x01EDBD5150BCEC00, 0xE10D5DB1B05C0CE0
+
+.Lk_deskew:	# deskew tables: inverts the sbox's "skew"
+	.quad	0x07E4A34047A4E300, 0x1DFEB95A5DBEF91A
+	.quad	0x5F36B5DC83EA6900, 0x2841C2ABF49D1E77
+
+##
+##  Decryption stuff
+##  Key schedule constants
+##
+.Lk_dksd:	# decryption key schedule: invskew x*D
+	.quad	0xFEB91A5DA3E44700, 0x0740E3A45A1DBEF9
+	.quad	0x41C277F4B5368300, 0x5FDC69EAAB289D1E
+.Lk_dksb:	# decryption key schedule: invskew x*B
+	.quad	0x9A4FCA1F8550D500, 0x03D653861CC94C99
+	.quad	0x115BEDA7B6FC4A00, 0xD993256F7E3482C8
+.Lk_dkse:	# decryption key schedule: invskew x*E + 0x63
+	.quad	0xD5031CCA1FC9D600, 0x53859A4C994F5086
+	.quad	0xA23196054FDC7BE8, 0xCD5EF96A20B31487
+.Lk_dks9:	# decryption key schedule: invskew x*9
+	.quad	0xB6116FC87ED9A700, 0x4AED933482255BFC
+	.quad	0x4576516227143300, 0x8BB89FACE9DAFDCE
+
+##
+##  Decryption stuff
+##  Round function constants
+##
+.Lk_dipt:	# decryption input transform
+	.quad	0x0F505B040B545F00, 0x154A411E114E451A
+	.quad	0x86E383E660056500, 0x12771772F491F194
+
+.Lk_dsb9:	# decryption sbox output *9*u, *9*t
+	.quad	0x851C03539A86D600, 0xCAD51F504F994CC9
+	.quad	0xC03B1789ECD74900, 0x725E2C9EB2FBA565
+.Lk_dsbd:	# decryption sbox output *D*u, *D*t
+	.quad	0x7D57CCDFE6B1A200, 0xF56E9B13882A4439
+	.quad	0x3CE2FAF724C6CB00, 0x2931180D15DEEFD3
+.Lk_dsbb:	# decryption sbox output *B*u, *B*t
+	.quad	0xD022649296B44200, 0x602646F6B0F2D404
+	.quad	0xC19498A6CD596700, 0xF3FF0C3E3255AA6B
+.Lk_dsbe:	# decryption sbox output *E*u, *E*t
+	.quad	0x46F2929626D4D000, 0x2242600464B4F6B0
+	.quad	0x0C55A6CDFFAAC100, 0x9467F36B98593E32
+.Lk_dsbo:	# decryption sbox final output
+	.quad	0x1387EA537EF94000, 0xC7AA6DB9D4943E2D
+	.quad	0x12D7560F93441D00, 0xCA4B8159D8C58E9C
+.asciz	"Vector Permutation AES for x86_64/SSSE3, Mike Hamburg (Stanford University)"
+.align	64
+.size	_vpaes_consts,.-_vpaes_consts
+___
+
+if ($win64) {
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	se_handler,\@abi-omnipotent
+.align	16
+se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_prologue
+
+	lea	16(%rax),%rsi		# %xmm save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	0xb8(%rax),%rax		# adjust stack pointer
+
+.Lin_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	se_handler,.-se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_${PREFIX}_set_encrypt_key
+	.rva	.LSEH_end_${PREFIX}_set_encrypt_key
+	.rva	.LSEH_info_${PREFIX}_set_encrypt_key
+
+	.rva	.LSEH_begin_${PREFIX}_set_decrypt_key
+	.rva	.LSEH_end_${PREFIX}_set_decrypt_key
+	.rva	.LSEH_info_${PREFIX}_set_decrypt_key
+
+	.rva	.LSEH_begin_${PREFIX}_encrypt
+	.rva	.LSEH_end_${PREFIX}_encrypt
+	.rva	.LSEH_info_${PREFIX}_encrypt
+
+	.rva	.LSEH_begin_${PREFIX}_decrypt
+	.rva	.LSEH_end_${PREFIX}_decrypt
+	.rva	.LSEH_info_${PREFIX}_decrypt
+
+	.rva	.LSEH_begin_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_end_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_info_${PREFIX}_cbc_encrypt
+
+.section	.xdata
+.align	8
+.LSEH_info_${PREFIX}_set_encrypt_key:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lenc_key_body,.Lenc_key_epilogue	# HandlerData[]
+.LSEH_info_${PREFIX}_set_decrypt_key:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Ldec_key_body,.Ldec_key_epilogue	# HandlerData[]
+.LSEH_info_${PREFIX}_encrypt:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lenc_body,.Lenc_epilogue		# HandlerData[]
+.LSEH_info_${PREFIX}_decrypt:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Ldec_body,.Ldec_epilogue		# HandlerData[]
+.LSEH_info_${PREFIX}_cbc_encrypt:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lcbc_body,.Lcbc_epilogue		# HandlerData[]
+___
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+
+print $code;
+
+close STDOUT;