X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Faes%2Fasm%2Fbsaes-x86_64.pl;h=2c79c2b67c897d60eca6ea0a5d28af75e4ca3f84;hp=67799cebc09fec5e93d5c23e64b689035dcc28b8;hb=b84460ad3a3e4fcb22efaa0a8365b826f4264ecf;hpb=a75a52a43e32e3ab3b7640872a2d65cfa6c22426

diff --git a/crypto/aes/asm/bsaes-x86_64.pl b/crypto/aes/asm/bsaes-x86_64.pl
index 67799cebc0..2c79c2b67c 100644
--- a/crypto/aes/asm/bsaes-x86_64.pl
+++ b/crypto/aes/asm/bsaes-x86_64.pl
@@ -1,4 +1,11 @@
-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
 
 ###################################################################
 ### AES-128 [originally in CTR mode]				###
@@ -38,8 +45,10 @@
 #		Emilia's	this(*)		difference
 #
 # Core 2    	9.30		8.69		+7%
-# Nehalem(**) 	7.63		6.98		+9%
-# Atom	    	17.1		17.4		-2%(***)
+# Nehalem(**) 	7.63		6.88		+11%
+# Atom	    	17.1		16.4		+4%
+# Silvermont	-		12.9
+# Goldmont	-		8.85
 #
 # (*)	Comparison is not completely fair, because "this" is ECB,
 #	i.e. no extra processing such as counter values calculation
@@ -50,14 +59,6 @@
 # (**)	Results were collected on Westmere, which is considered to
 #	be equivalent to Nehalem for this code.
 #
-# (***)	Slowdown on Atom is rather strange per se, because original
-#	implementation has a number of 9+-bytes instructions, which
-#	are bad for Atom front-end, and which I eliminated completely.
-#	In attempt to address deterioration sbox() was tested in FP
-#	SIMD "domain" (movaps instead of movdqa, xorps instead of
-#	pxor, etc.). While it resulted in nominal 4% improvement on
-#	Atom, it hurted Westmere by more than 2x factor.
-#
 # 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
@@ -65,12 +66,12 @@
 # function is:
 #
 # 		conversion	conversion/8x block
-# Core 2	410		0.37
-# Nehalem	310		0.35
-# Atom		570		0.26
+# Core 2	240		0.22
+# Nehalem	180		0.20
+# Atom		430		0.20
 #
 # The ratio values mean that 128-byte blocks will be processed
-# 21-27% slower, 256-byte blocks - 12-16%, 384-byte blocks - 8-11%,
+# 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,
@@ -83,8 +84,16 @@
 # Add decryption procedure. Performance in CPU cycles spent to decrypt
 # one byte out of 4096-byte buffer with 128-bit key is:
 #
-# Core 2	11.0
-# Nehalem	9.16
+# Core 2	9.98
+# Nehalem	7.80
+# Atom		17.9
+# Silvermont	14.0
+# Goldmont	10.2
+#
+# 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>
 
@@ -99,10 +108,12 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
 die "can't locate x86_64-xlate.pl";
 
-open STDOUT,"| $^X $xlate $flavour $output";
+open OUT,"| \"$^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");
@@ -120,7 +131,7 @@ my @s=@_[12..15];
 
 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 
+# output in lsb > [b6, b5, b0, b3, b7, b1, b4, b2] < msb
 my @b=@_[0..7];
 $code.=<<___;
 	pxor	@b[6], @b[5]
@@ -370,7 +381,7 @@ $code.=<<___;
 	pxor	@s[0], @t[3]
 	pxor	@s[1], @t[2]
 	pxor	@s[2], @t[1]
-	pxor	@s[3], @t[0] 
+	pxor	@s[3], @t[0]
 
 	#Inv_GF16 \t0, \t1, \t2, \t3, \s0, \s1, \s2, \s3
 
@@ -426,21 +437,21 @@ my $mask=pop;
 $code.=<<___;
 	pxor	0x00($key),@x[0]
 	pxor	0x10($key),@x[1]
-	pshufb	$mask,@x[0]
 	pxor	0x20($key),@x[2]
-	pshufb	$mask,@x[1]
 	pxor	0x30($key),@x[3]
-	pshufb	$mask,@x[2]
+	pshufb	$mask,@x[0]
+	pshufb	$mask,@x[1]
 	pxor	0x40($key),@x[4]
-	pshufb	$mask,@x[3]
 	pxor	0x50($key),@x[5]
-	pshufb	$mask,@x[4]
+	pshufb	$mask,@x[2]
+	pshufb	$mask,@x[3]
 	pxor	0x60($key),@x[6]
-	pshufb	$mask,@x[5]
 	pxor	0x70($key),@x[7]
+	pshufb	$mask,@x[4]
+	pshufb	$mask,@x[5]
 	pshufb	$mask,@x[6]
-	lea	0x80($key),$key
 	pshufb	$mask,@x[7]
+	lea	0x80($key),$key
 ___
 }
 
@@ -448,6 +459,7 @@ 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]
@@ -489,7 +501,8 @@ $code.=<<___;
 	pxor	@t[4], @t[0]
 	 pshufd	\$0x4E, @x[2], @x[6]
 	pxor	@t[5], @t[1]
-
+___
+$code.=<<___ if (!$inv);
 	pxor	@t[3], @x[4]
 	pxor	@t[7], @x[5]
 	pxor	@t[6], @x[3]
@@ -497,9 +510,20 @@ $code.=<<___;
 	pxor	@t[2], @x[6]
 	 movdqa	@t[1], @x[7]
 ___
+$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 {
+sub InvMixColumns_orig {
 my @x=@_[0..7];
 my @t=@_[8..15];
 
@@ -653,6 +677,54 @@ $code.=<<___;
 ___
 }
 
+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];
@@ -738,8 +810,8 @@ ___
 $code.=<<___;
 .text
 
-.extern	AES_encrypt
-.extern	AES_decrypt
+.extern	asm_AES_encrypt
+.extern	asm_AES_decrypt
 
 .type	_bsaes_encrypt8,\@abi-omnipotent
 .align	64
@@ -748,21 +820,21 @@ _bsaes_encrypt8:
 
 	movdqa	($key), @XMM[9]		# round 0 key
 	lea	0x10($key), $key
-	movdqa	0x60($const), @XMM[8]	# .LM0SR
+	movdqa	0x50($const), @XMM[8]	# .LM0SR
 	pxor	@XMM[9], @XMM[0]	# xor with round0 key
 	pxor	@XMM[9], @XMM[1]
-	 pshufb	@XMM[8], @XMM[0]
 	pxor	@XMM[9], @XMM[2]
-	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[3]
-	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[4]
-	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[5]
-	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[6]
-	 pshufb	@XMM[8], @XMM[5]
 	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
 	 pshufb	@XMM[8], @XMM[6]
 	 pshufb	@XMM[8], @XMM[7]
 _bsaes_encrypt8_bitslice:
@@ -815,18 +887,18 @@ _bsaes_decrypt8:
 	movdqa	-0x30($const), @XMM[8]	# .LM0ISR
 	pxor	@XMM[9], @XMM[0]	# xor with round0 key
 	pxor	@XMM[9], @XMM[1]
-	 pshufb	@XMM[8], @XMM[0]
 	pxor	@XMM[9], @XMM[2]
-	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[3]
-	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[4]
-	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[5]
-	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[6]
-	 pshufb	@XMM[8], @XMM[5]
 	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]
 ___
@@ -899,53 +971,89 @@ $code.=<<___;
 .type	_bsaes_key_convert,\@abi-omnipotent
 .align	16
 _bsaes_key_convert:
-	lea	.LBS1(%rip), $const
+	lea	.Lmasks(%rip), $const
 	movdqu	($inp), %xmm7		# load round 0 key
-	movdqa	-0x10($const), %xmm8	# .LBS0
-	movdqa	0x00($const), %xmm9	# .LBS1
-	movdqa	0x10($const), %xmm10	# .LBS2
-	movdqa	0x40($const), %xmm13	# .LM0
-	movdqa	0x60($const), %xmm14	# .LNOT
-
-	movdqu	0x10($inp), %xmm6	# load round 1 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	$rounds
 	jmp	.Lkey_loop
 .align	16
 .Lkey_loop:
-	pshufb	%xmm13, %xmm6		# .LM0
-	movdqa	%xmm6, %xmm7
-___
-	&bitslice_key	(map("%xmm$_",(0..7, 8..12)));
-$code.=<<___;
-	pxor	%xmm14, %xmm5		# "pnot"
-	pxor	%xmm14, %xmm6
-	pxor	%xmm14, %xmm0
-	pxor	%xmm14, %xmm1
-	lea	0x10($inp), $inp
-	movdqa	%xmm0, 0x00($out)	# write bit-sliced round key
-	movdqa	%xmm1, 0x10($out)
-	movdqa	%xmm2, 0x20($out)
-	movdqa	%xmm3, 0x30($out)
-	movdqa	%xmm4, 0x40($out)
-	movdqa	%xmm5, 0x50($out)
-	movdqa	%xmm6, 0x60($out)
-	movdqa	%xmm7, 0x70($out)
+	pshufb	%xmm4, %xmm6		# .LM0
+
+	movdqa	%xmm0,	%xmm8
+	movdqa	%xmm1,	%xmm9
+
+	pand	%xmm6,	%xmm8
+	pand	%xmm6,	%xmm9
+	movdqa	%xmm2,	%xmm10
+	pcmpeqb	%xmm0,	%xmm8
+	psllq	\$4,	%xmm0		# 0x10...
+	movdqa	%xmm3,	%xmm11
+	pcmpeqb	%xmm1,	%xmm9
+	psllq	\$4,	%xmm1		# 0x20...
+
+	pand	%xmm6,	%xmm10
+	pand	%xmm6,	%xmm11
+	movdqa	%xmm0,	%xmm12
+	pcmpeqb	%xmm2,	%xmm10
+	psllq	\$4,	%xmm2		# 0x40...
+	movdqa	%xmm1,	%xmm13
+	pcmpeqb	%xmm3,	%xmm11
+	psllq	\$4,	%xmm3		# 0x80...
+
+	movdqa	%xmm2,	%xmm14
+	movdqa	%xmm3,	%xmm15
+	 pxor	%xmm5,	%xmm8		# "pnot"
+	 pxor	%xmm5,	%xmm9
+
+	pand	%xmm6,	%xmm12
+	pand	%xmm6,	%xmm13
+	 movdqa	%xmm8, 0x00($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
-	movdqu	($inp), %xmm6		# load next round key
 	dec	$rounds
 	jnz	.Lkey_loop
 
-	movdqa	0x70($const), %xmm7	# .L63
+	movdqa	0x50($const), %xmm7	# .L63
 	#movdqa	%xmm6, ($out)		# don't save last round key
 	ret
 .size	_bsaes_key_convert,.-_bsaes_key_convert
 ___
 }
 
-if (1 && !$win64) {	# following four functions are unsupported interface
+if (0 && !$win64) {	# following four functions are unsupported interface
 			# used for benchmarking...
 $code.=<<___;
 .globl	bsaes_enc_key_convert
@@ -1051,19 +1159,29 @@ my ($arg1,$arg2,$arg3,$arg4,$arg5,$arg6)=$win64	? ("%rcx","%rdx","%r8","%r9","%r
 						: ("%rdi","%rsi","%rdx","%rcx","%r8","%r9d");
 my ($inp,$out,$len,$key)=("%r12","%r13","%r14","%r15");
 
-if (0) {	# suppress unreferenced ECB subroutines, spare some space...
+if ($ecb) {
 $code.=<<___;
 .globl	bsaes_ecb_encrypt_blocks
 .type	bsaes_ecb_encrypt_blocks,\@abi-omnipotent
 .align	16
 bsaes_ecb_encrypt_blocks:
+.cfi_startproc
+	mov	%rsp, %rax
+.Lecb_enc_prologue:
 	push	%rbp
+.cfi_push	%rbp
 	push	%rbx
+.cfi_push	%rbx
 	push	%r12
+.cfi_push	%r12
 	push	%r13
+.cfi_push	%r13
 	push	%r14
+.cfi_push	%r14
 	push	%r15
+.cfi_push	%r15
 	lea	-0x48(%rsp),%rsp
+.cfi_adjust_cfa_offset	0x48
 ___
 $code.=<<___ if ($win64);
 	lea	-0xa0(%rsp), %rsp
@@ -1081,6 +1199,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	mov	%rsp,%rbp		# backup %rsp
+.cfi_def_cfa_register	%rbp
 	mov	240($arg4),%eax		# rounds
 	mov	$arg1,$inp		# backup arguments
 	mov	$arg2,$out
@@ -1208,7 +1327,7 @@ $code.=<<___;
 	lea	($inp), $arg1
 	lea	($out), $arg2
 	lea	($key), $arg3
-	call	AES_encrypt
+	call	asm_AES_encrypt
 	lea	16($inp), $inp
 	lea	16($out), $out
 	dec	$len
@@ -1224,7 +1343,8 @@ $code.=<<___;
 	cmp	%rax, %rbp
 	jb	.Lecb_enc_bzero
 
-	lea	(%rbp),%rsp		# restore %rsp
+	lea	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
 ___
 $code.=<<___ if ($win64);
 	movaps	0x40(%rbp), %xmm6
@@ -1237,31 +1357,50 @@ $code.=<<___ if ($win64);
 	movaps	0xb0(%rbp), %xmm13
 	movaps	0xc0(%rbp), %xmm14
 	movaps	0xd0(%rbp), %xmm15
-	lea	0xa0(%rbp), %rsp
+	lea	0xa0(%rax), %rax
+.Lecb_enc_tail:
 ___
 $code.=<<___;
-	mov	0x48(%rsp), %r15
-	mov	0x50(%rsp), %r14
-	mov	0x58(%rsp), %r13
-	mov	0x60(%rsp), %r12
-	mov	0x68(%rsp), %rbx
-	mov	0x70(%rsp), %rbp
-	lea	0x78(%rsp), %rsp
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
 .Lecb_enc_epilogue:
 	ret
+.cfi_endproc
 .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:
+.cfi_startproc
+	mov	%rsp, %rax
+.Lecb_dec_prologue:
 	push	%rbp
+.cfi_push	%rbp
 	push	%rbx
+.cfi_push	%rbx
 	push	%r12
+.cfi_push	%r12
 	push	%r13
+.cfi_push	%r13
 	push	%r14
+.cfi_push	%r14
 	push	%r15
+.cfi_push	%r15
 	lea	-0x48(%rsp),%rsp
+.cfi_adjust_cfa_offset	0x48
 ___
 $code.=<<___ if ($win64);
 	lea	-0xa0(%rsp), %rsp
@@ -1279,6 +1418,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	mov	%rsp,%rbp		# backup %rsp
+.cfi_def_cfa_register	%rbp
 	mov	240($arg4),%eax		# rounds
 	mov	$arg1,$inp		# backup arguments
 	mov	$arg2,$out
@@ -1407,7 +1547,7 @@ $code.=<<___;
 	lea	($inp), $arg1
 	lea	($out), $arg2
 	lea	($key), $arg3
-	call	AES_decrypt
+	call	asm_AES_decrypt
 	lea	16($inp), $inp
 	lea	16($out), $out
 	dec	$len
@@ -1423,7 +1563,8 @@ $code.=<<___;
 	cmp	%rax, %rbp
 	jb	.Lecb_dec_bzero
 
-	lea	(%rbp),%rsp		# restore %rsp
+	lea	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
 ___
 $code.=<<___ if ($win64);
 	movaps	0x40(%rbp), %xmm6
@@ -1436,44 +1577,63 @@ $code.=<<___ if ($win64);
 	movaps	0xb0(%rbp), %xmm13
 	movaps	0xc0(%rbp), %xmm14
 	movaps	0xd0(%rbp), %xmm15
-	lea	0xa0(%rbp), %rsp
+	lea	0xa0(%rax), %rax
+.Lecb_dec_tail:
 ___
 $code.=<<___;
-	mov	0x48(%rsp), %r15
-	mov	0x50(%rsp), %r14
-	mov	0x58(%rsp), %r13
-	mov	0x60(%rsp), %r12
-	mov	0x68(%rsp), %rbx
-	mov	0x70(%rsp), %rbp
-	lea	0x78(%rsp), %rsp
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
 .Lecb_dec_epilogue:
 	ret
+.cfi_endproc
 .size	bsaes_ecb_decrypt_blocks,.-bsaes_ecb_decrypt_blocks
 ___
 }
 $code.=<<___;
-.extern	AES_cbc_encrypt
+.extern	asm_AES_cbc_encrypt
 .globl	bsaes_cbc_encrypt
 .type	bsaes_cbc_encrypt,\@abi-omnipotent
 .align	16
 bsaes_cbc_encrypt:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	mov	48(%rsp),$arg6		# pull direction flag
 ___
 $code.=<<___;
 	cmp	\$0,$arg6
-	jne	AES_cbc_encrypt
+	jne	asm_AES_cbc_encrypt
 	cmp	\$128,$arg3
-	jb	AES_cbc_encrypt
+	jb	asm_AES_cbc_encrypt
 
+	mov	%rsp, %rax
+.Lcbc_dec_prologue:
 	push	%rbp
+.cfi_push	%rbp
 	push	%rbx
+.cfi_push	%rbx
 	push	%r12
+.cfi_push	%r12
 	push	%r13
+.cfi_push	%r13
 	push	%r14
+.cfi_push	%r14
 	push	%r15
+.cfi_push	%r15
 	lea	-0x48(%rsp), %rsp
+.cfi_adjust_cfa_offset	0x48
 ___
 $code.=<<___ if ($win64);
 	mov	0xa0(%rsp),$arg5	# pull ivp
@@ -1492,28 +1652,29 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	mov	%rsp, %rbp		# backup %rsp
+.cfi_def_cfa_register	%rbp
 	mov	240($arg4), %eax	# rounds
 	mov	$arg1, $inp		# backup arguments
 	mov	$arg2, $out
 	mov	$arg3, $len
 	mov	$arg4, $key
-	mov	$arg5, %rdx
+	mov	$arg5, %rbx
 	shr	\$4, $len		# bytes to blocks
 
-	mov	%eax, %ebx		# rounds
+	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	%ebx, %r10d		# pass rounds
+	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	(%rdx), @XMM[15]	# load IV
+	movdqu	(%rbx), @XMM[15]	# load IV
 	sub	\$8,$len
 .Lcbc_dec_loop:
 	movdqu	0x00($inp), @XMM[0]	# load input
@@ -1524,7 +1685,7 @@ $code.=<<___;
 	movdqu	0x50($inp), @XMM[5]
 	mov	%rsp, %rax		# pass key schedule
 	movdqu	0x60($inp), @XMM[6]
-	mov	%ebx,%r10d		# pass rounds
+	mov	%edx,%r10d		# pass rounds
 	movdqu	0x70($inp), @XMM[7]
 	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
 
@@ -1564,7 +1725,7 @@ $code.=<<___;
 
 	movdqu	0x00($inp), @XMM[0]	# load input
 	mov	%rsp, %rax		# pass key schedule
-	mov	%ebx, %r10d		# pass rounds
+	mov	%edx, %r10d		# pass rounds
 	cmp	\$2,$len
 	jb	.Lcbc_dec_one
 	movdqu	0x10($inp), @XMM[1]
@@ -1691,14 +1852,16 @@ $code.=<<___;
 	jmp	.Lcbc_dec_done
 .align	16
 .Lcbc_dec_one:
-	movdqa	@XMM[15], 0x20(%rbp)	# put aside IV
-	call	_bsaes_decrypt8
-	pxor	0x20(%rbp), @XMM[0]	# ^= IV
-	movdqu	0x00($inp), @XMM[15]	# IV
-	movdqu	@XMM[0], 0x00($out)	# write output
+	lea	($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], (%rdx)	# return IV
+	movdqu	@XMM[15], (%rbx)	# return IV
 	lea	(%rsp), %rax
 	pxor	%xmm0, %xmm0
 .Lcbc_dec_bzero:			# wipe key schedule [if any]
@@ -1708,7 +1871,8 @@ $code.=<<___;
 	cmp	%rax, %rbp
 	ja	.Lcbc_dec_bzero
 
-	lea	(%rbp),%rsp		# restore %rsp
+	lea	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
 ___
 $code.=<<___ if ($win64);
 	movaps	0x40(%rbp), %xmm6
@@ -1721,31 +1885,50 @@ $code.=<<___ if ($win64);
 	movaps	0xb0(%rbp), %xmm13
 	movaps	0xc0(%rbp), %xmm14
 	movaps	0xd0(%rbp), %xmm15
-	lea	0xa0(%rbp), %rsp
+	lea	0xa0(%rax), %rax
+.Lcbc_dec_tail:
 ___
 $code.=<<___;
-	mov	0x48(%rsp), %r15
-	mov	0x50(%rsp), %r14
-	mov	0x58(%rsp), %r13
-	mov	0x60(%rsp), %r12
-	mov	0x68(%rsp), %rbx
-	mov	0x70(%rsp), %rbp
-	lea	0x78(%rsp), %rsp
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
 .Lcbc_dec_epilogue:
 	ret
+.cfi_endproc
 .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:
+.cfi_startproc
+	mov	%rsp, %rax
+.Lctr_enc_prologue:
 	push	%rbp
+.cfi_push	%rbp
 	push	%rbx
+.cfi_push	%rbx
 	push	%r12
+.cfi_push	%r12
 	push	%r13
+.cfi_push	%r13
 	push	%r14
+.cfi_push	%r14
 	push	%r15
+.cfi_push	%r15
 	lea	-0x48(%rsp), %rsp
+.cfi_adjust_cfa_offset	0x48
 ___
 $code.=<<___ if ($win64);
 	mov	0xa0(%rsp),$arg5	# pull ivp
@@ -1764,6 +1947,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	mov	%rsp, %rbp		# backup %rsp
+.cfi_def_cfa_register	%rbp
 	movdqu	($arg5), %xmm0		# load counter
 	mov	240($arg4), %eax	# rounds
 	mov	$arg1, $inp		# backup arguments
@@ -1819,21 +2003,21 @@ $code.=<<___;
 	movdqa	-0x10(%r11), @XMM[8]	# .LSWPUPM0SR
 	pxor	@XMM[9], @XMM[0]	# xor with round0 key
 	pxor	@XMM[9], @XMM[1]
-	 pshufb	@XMM[8], @XMM[0]
 	pxor	@XMM[9], @XMM[2]
-	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[3]
-	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[4]
-	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[5]
-	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[6]
-	 pshufb	@XMM[8], @XMM[5]
 	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
 	 pshufb	@XMM[8], @XMM[6]
-	lea	.LBS0(%rip), %r11	# constants table
 	 pshufb	@XMM[8], @XMM[7]
+	lea	.LBS0(%rip), %r11	# constants table
 	mov	%ebx,%r10d		# pass rounds
 
 	call	_bsaes_encrypt8_bitslice
@@ -1875,6 +2059,7 @@ $code.=<<___;
 	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
@@ -1912,7 +2097,7 @@ $code.=<<___;
 	lea	0x20(%rbp), $arg1
 	lea	0x30(%rbp), $arg2
 	lea	($key), $arg3
-	call	AES_encrypt
+	call	asm_AES_encrypt
 	movdqu	($inp), @XMM[1]
 	lea	16($inp), $inp
 	mov	0x2c(%rbp), %eax	# load 32-bit counter
@@ -1936,7 +2121,8 @@ $code.=<<___;
 	cmp	%rax, %rbp
 	ja	.Lctr_enc_bzero
 
-	lea	(%rbp),%rsp		# restore %rsp
+	lea	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
 ___
 $code.=<<___ if ($win64);
 	movaps	0x40(%rbp), %xmm6
@@ -1949,20 +2135,861 @@ $code.=<<___ if ($win64);
 	movaps	0xb0(%rbp), %xmm13
 	movaps	0xc0(%rbp), %xmm14
 	movaps	0xd0(%rbp), %xmm15
-	lea	0xa0(%rbp), %rsp
+	lea	0xa0(%rax), %rax
+.Lctr_enc_tail:
 ___
 $code.=<<___;
-	mov	0x48(%rsp), %r15
-	mov	0x50(%rsp), %r14
-	mov	0x58(%rsp), %r13
-	mov	0x60(%rsp), %r12
-	mov	0x68(%rsp), %rbx
-	mov	0x70(%rsp), %rbp
-	lea	0x78(%rsp), %rsp
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
 .Lctr_enc_epilogue:
 	ret
+.cfi_endproc
 .size	bsaes_ctr32_encrypt_blocks,.-bsaes_ctr32_encrypt_blocks
 ___
+######################################################################
+# void bsaes_xts_[en|de]crypt(const char *inp,char *out,size_t len,
+#	const AES_KEY *key1, const AES_KEY *key2,
+#	const unsigned char iv[16]);
+#
+my ($twmask,$twres,$twtmp)=@XMM[13..15];
+$arg6=~s/d$//;
+
+$code.=<<___;
+.globl	bsaes_xts_encrypt
+.type	bsaes_xts_encrypt,\@abi-omnipotent
+.align	16
+bsaes_xts_encrypt:
+.cfi_startproc
+	mov	%rsp, %rax
+.Lxts_enc_prologue:
+	push	%rbp
+.cfi_push	%rbp
+	push	%rbx
+.cfi_push	%rbx
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+	lea	-0x48(%rsp), %rsp
+.cfi_adjust_cfa_offset	0x48
+___
+$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
+.cfi_def_cfa_register	%rbp
+	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	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
+___
+$code.=<<___ if ($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(%rax), %rax
+.Lxts_enc_tail:
+___
+$code.=<<___;
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
+.Lxts_enc_epilogue:
+	ret
+.cfi_endproc
+.size	bsaes_xts_encrypt,.-bsaes_xts_encrypt
+
+.globl	bsaes_xts_decrypt
+.type	bsaes_xts_decrypt,\@abi-omnipotent
+.align	16
+bsaes_xts_decrypt:
+.cfi_startproc
+	mov	%rsp, %rax
+.Lxts_dec_prologue:
+	push	%rbp
+.cfi_push	%rbp
+	push	%rbx
+.cfi_push	%rbx
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+	lea	-0x48(%rsp), %rsp
+.cfi_adjust_cfa_offset	0x48
+___
+$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	0x78(%rbp),%rax
+.cfi_def_cfa	%rax,8
+___
+$code.=<<___ if ($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(%rax), %rax
+.Lxts_dec_tail:
+___
+$code.=<<___;
+	mov	-48(%rax), %r15
+.cfi_restore	%r15
+	mov	-40(%rax), %r14
+.cfi_restore	%r14
+	mov	-32(%rax), %r13
+.cfi_restore	%r13
+	mov	-24(%rax), %r12
+.cfi_restore	%r12
+	mov	-16(%rax), %rbx
+.cfi_restore	%rbx
+	mov	-8(%rax), %rbp
+.cfi_restore	%rbp
+	lea	(%rax), %rsp		# restore %rsp
+.cfi_def_cfa_register	%rsp
+.Lxts_dec_epilogue:
+	ret
+.cfi_endproc
+.size	bsaes_xts_decrypt,.-bsaes_xts_decrypt
+___
 }
 $code.=<<___;
 .type	_bsaes_const,\@object
@@ -1984,14 +3011,8 @@ _bsaes_const:
 	.quad	0x0504070600030201, 0x0f0e0d0c0a09080b
 .LSRM0:
 	.quad	0x0304090e00050a0f, 0x01060b0c0207080d
-.LM0:
-	.quad	0x02060a0e03070b0f, 0x0004080c0105090d
 .LM0SR:
 	.quad	0x0a0e02060f03070b, 0x0004080c05090d01
-.LNOT:		# magic constants
-	.quad	0xffffffffffffffff, 0xffffffffffffffff
-.L63:
-	.quad	0x6363636363636363, 0x6363636363636363
 .LSWPUP:	# byte-swap upper dword
 	.quad	0x0706050403020100, 0x0c0d0e0f0b0a0908
 .LSWPUPM0SR:
@@ -2012,11 +3033,199 @@ _bsaes_const:
 	.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
+	jbe	.Lin_prologue
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_prologue
+
+	mov	8(%r11),%r10d		# HandlerData[2]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=tail label
+	jae	.Lin_tail
+
+	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+0x78(%rax),%rax	# adjust stack pointer
+
+.Lin_tail:
+	mov	-48(%rax),%rbp
+	mov	-40(%rax),%rbx
+	mov	-32(%rax),%r12
+	mov	-24(%rax),%r13
+	mov	-16(%rax),%r14
+	mov	-8(%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_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[]
+	.rva	.Lecb_enc_tail
+	.long	0
+.Lecb_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lecb_dec_body,.Lecb_dec_epilogue	# HandlerData[]
+	.rva	.Lecb_dec_tail
+	.long	0
+___
+$code.=<<___;
+.Lcbc_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lcbc_dec_body,.Lcbc_dec_epilogue	# HandlerData[]
+	.rva	.Lcbc_dec_tail
+	.long	0
+.Lctr_enc_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lctr_enc_body,.Lctr_enc_epilogue	# HandlerData[]
+	.rva	.Lctr_enc_tail
+	.long	0
+.Lxts_enc_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lxts_enc_body,.Lxts_enc_epilogue	# HandlerData[]
+	.rva	.Lxts_enc_tail
+	.long	0
+.Lxts_dec_info:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lxts_dec_body,.Lxts_dec_epilogue	# HandlerData[]
+	.rva	.Lxts_dec_tail
+	.long	0
+___
+}
+
 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 
 print $code;