X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Faes%2Fasm%2Faes-x86_64.pl;h=a757b47146790bf1a675a508184c65d031f64550;hp=91a19545de2507859229374f28a04b9dcce12e07;hb=0e3ef596e55654a7782ae7ce302f8933b4817f12;hpb=e59f992be61a028e1c20d1937420befdcaa5422a;ds=sidebyside

diff --git a/crypto/aes/asm/aes-x86_64.pl b/crypto/aes/asm/aes-x86_64.pl
index 91a19545de..a757b47146 100755
--- a/crypto/aes/asm/aes-x86_64.pl
+++ b/crypto/aes/asm/aes-x86_64.pl
@@ -7,7 +7,7 @@
 # details see http://www.openssl.org/~appro/cryptogams/.
 # ====================================================================
 #
-# Version 2.0.
+# 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
@@ -20,21 +20,30 @@
 #
 #		ECB encrypt	ECB decrypt	CBC large chunk
 # AMD64		33		41		13.0
-# EM64T		38		59		18.6
-# Core 2	30		43		14.5
+# EM64T		38		59		18.6(*)
+# Core 2	30		43		14.5(*)
+#
+# (*) with hyper-threading off
 
-$verticalspin=1;	# unlike 32-bit version $verticalspin performs
-			# ~15% better on both AMD and Intel cores
-$output=shift;
+$flavour = shift;
+$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 STDOUT,"| $^X $xlate $output";
+open STDOUT,"| $^X $xlate $flavour $output";
+
+$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
 
-$code=".text\n";
+$code=".section .note.GNU-stack,\"\",\@progbits\n";
+$code.=".text\n";
 
 $s0="%eax";
 $s1="%ebx";
@@ -145,22 +154,17 @@ $code.=<<___;
 	movzb	`&lo("$s0")`,$acc0
 	movzb	`&lo("$s1")`,$acc1
 	movzb	`&lo("$s2")`,$acc2
-	mov	2($sbox,$acc0,8),$t0
-	mov	2($sbox,$acc1,8),$t1
-	mov	2($sbox,$acc2,8),$t2
-
-	and	\$0x000000ff,$t0
-	and	\$0x000000ff,$t1
-	and	\$0x000000ff,$t2
+	movzb	2($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
-	mov	2($sbox,$acc0,8),$t3
+	movzb	2($sbox,$acc0,8),$t3
 	mov	0($sbox,$acc1,8),$acc1	#$t0
 	mov	0($sbox,$acc2,8),$acc2	#$t1
 
-	and	\$0x000000ff,$t3
 	and	\$0x0000ff00,$acc1
 	and	\$0x0000ff00,$acc2
 
@@ -594,18 +598,20 @@ AES_encrypt:
 	push	%r15
 
 	# allocate frame "above" key schedule
-	mov	%rsp,%rax
-	mov	%rdx,$key
-	lea	-63(%rdx),%rcx
+	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
 
-	push	%rax		# save real stack pointer
-	push	%rsi		# save out
+	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
@@ -615,33 +621,33 @@ AES_encrypt:
 
 	shl	\$4,$rnds
 	lea	($key,$rnds),%rbp
-	push	%rbp
-	push	$key
+	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
-	.picmeup	$sbox
-	lea	AES_Te+2048-.($sbox),$sbox
-	lea	768-32(%rsp),%rbp
+	lea	.LAES_Te+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
 	sub	$sbox,%rbp
 	and	\$0x300,%rbp
 	lea	($sbox,%rbp),$sbox
 
 	call	_x86_64_AES_encrypt_compact
 
-	lea	16(%rsp),%rsp
-	pop	$out		# restore out
+	mov	16(%rsp),$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	(%rsp),%rsp
-	pop	%r15
-	pop	%r14
-	pop	%r13
-	pop	%r12
-	pop	%rbp
-	pop	%rbx
+	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	AES_encrypt,.-AES_encrypt
 ___
@@ -1188,18 +1194,20 @@ AES_decrypt:
 	push	%r15
 
 	# allocate frame "above" key schedule
-	mov	%rsp,%rax
-	mov	%rdx,$key
-	lea	-63(%rdx),%rcx
+	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
 
-	push	%rax		# save real stack pointer
-	push	%rsi		# save out
+	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
@@ -1209,35 +1217,35 @@ AES_decrypt:
 
 	shl	\$4,$rnds
 	lea	($key,$rnds),%rbp
-	push	%rbp
-	push	$key
+	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
-	.picmeup	$sbox
-	lea	AES_Td+2048-.($sbox),$sbox
-	lea	768-32(%rsp),%rbp
+	lea	.LAES_Td+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
 	sub	$sbox,%rbp
 	and	\$0x300,%rbp
 	lea	($sbox,%rbp),$sbox
-	shr	\$3,%rbp		# recall "magic" constants!
+	shr	\$3,%rbp	# recall "magic" constants!
 	add	%rbp,$sbox
 
 	call	_x86_64_AES_decrypt_compact
 
-	lea	16(%rsp),%rsp
-	pop	$out	# restore out
-	mov	$s0,0($out)
+	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	(%rsp),%rsp
-	pop	%r15
-	pop	%r14
-	pop	%r13
-	pop	%r12
-	pop	%rbp
-	pop	%rbx
+	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	AES_decrypt,.-AES_decrypt
 ___
@@ -1279,7 +1287,29 @@ $code.=<<___;
 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	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
+.Lenc_key_epilogue:
+	ret
+.size	AES_set_encrypt_key,.-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
@@ -1289,8 +1319,7 @@ AES_set_encrypt_key:
 	test	\$-1,%rdi
 	jz	.Lbadpointer
 
-	.picmeup %rbp
-	lea	AES_Te-.(%rbp),%rbp
+	lea	.LAES_Te(%rip),%rbp
 	lea	2048+128(%rbp),%rbp
 
 	# prefetch Te4
@@ -1461,10 +1490,8 @@ $code.=<<___;
 .Lbadpointer:
 	mov	\$-1,%rax
 .Lexit:
-	pop	%rbp
-	pop	%rbx
-	ret
-.size	AES_set_encrypt_key,.-AES_set_encrypt_key
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
 ___
 
 sub deckey_ref()
@@ -1528,18 +1555,19 @@ $code.=<<___;
 .type	AES_set_decrypt_key,\@function,3
 .align	16
 AES_set_decrypt_key:
-	push	%rdx			# save key schedule
-	call	AES_set_encrypt_key
-	cmp	\$0,%eax
-	pop	%r8			# restore key schedule
-	jne	.Labort
-
 	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
@@ -1561,8 +1589,7 @@ AES_set_decrypt_key:
 		cmp	%rsi,%rdi
 	jne	.Linvert
 
-	.picmeup %rax
-	lea	AES_Te+2048+1024-.(%rax),%rax	# rcon
+	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
 
 	mov	40(%rax),$mask80
 	mov	48(%rax),$maskfe
@@ -1586,13 +1613,15 @@ $code.=<<___;
 	jnz	.Lpermute
 
 	xor	%rax,%rax
-	pop	%r15
-	pop	%r14
-	pop	%r13
-	pop	%r12
-	pop	%rbp
-	pop	%rbx
 .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	AES_set_decrypt_key,.-AES_set_decrypt_key
 ___
@@ -1603,46 +1632,59 @@ ___
 {
 # stack frame layout
 # -8(%rsp)		return address
-my $_rsp="0(%rsp)";		# saved %rsp
-my $_len="8(%rsp)";		# copy of 3rd parameter, length
-my $_key="16(%rsp)";		# copy of 4th parameter, key
-my $_ivp="24(%rsp)";		# copy of 5th parameter, ivp
-my $keyp="32(%rsp)";		# one to pass as $key
-my $ivec="40(%rsp)";		# ivec[16]
-my $aes_key="56(%rsp)";		# copy of aes_key
-my $mark="56+240(%rsp)";	# copy of aes_key->rounds
+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.=<<___;
 .globl	AES_cbc_encrypt
 .type	AES_cbc_encrypt,\@function,6
 .align	16
+.extern	OPENSSL_ia32cap_P
 AES_cbc_encrypt:
 	cmp	\$0,%rdx	# check length
-	je	.Lcbc_just_ret
+	je	.Lcbc_epilogue
+	pushfq
 	push	%rbx
 	push	%rbp
 	push	%r12
 	push	%r13
 	push	%r14
 	push	%r15
-	pushfq
-	cld
+.Lcbc_prologue:
 
-	.picmeup $sbox
-.Lcbc_pic_point:
+	cld
+	mov	%r9d,%r9d	# clear upper half of enc
 
+	lea	.LAES_Te(%rip),$sbox
 	cmp	\$0,%r9
-	je	.LDECRYPT
-
-	lea	AES_Te-.Lcbc_pic_point($sbox),$sbox
+	jne	.Lcbc_picked_te
+	lea	.LAES_Td(%rip),$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	-64-248(%rsp),$key
+	lea	-88-248(%rsp),$key
 	and	\$-64,$key
 
-	# ... and make it doesn't alias with AES_Te modulo 4096
+	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
 	mov	$sbox,%r10
-	lea	2048($sbox),%r11
+	lea	2304($sbox),%r11
 	mov	$key,%r12
 	and	\$0xFFF,%r10	# s = $sbox&0xfff
 	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
@@ -1662,22 +1704,27 @@ AES_cbc_encrypt:
 .Lcbc_te_ok:
 
 	xchg	%rsp,$key
-	add	\$8,%rsp	# reserve for return address!
+	#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	\$2048,%r10
+	cmp	\$2304,%r10
 	jb	.Lcbc_do_ecopy
 	cmp	\$4096-248,%r10
 	jb	.Lcbc_skip_ecopy
@@ -1688,12 +1735,11 @@ AES_cbc_encrypt:
 		lea	$aes_key,$key
 		mov	\$240/8,%ecx
 		.long	0x90A548F3	# rep movsq
-		mov	(%rsi),%eax	# copy aes_key->rounds
-		mov	%eax,(%rdi)
+		mov	%eax,(%rdi)	# copy aes_key->rounds
 .Lcbc_skip_ecopy:
 	mov	$key,$keyp	# save key pointer
 
-	mov	\$16,%ecx
+	mov	\$18,%ecx
 .align	4
 .Lcbc_prefetch_te:
 		mov	0($sbox),%r10
@@ -1703,183 +1749,77 @@ AES_cbc_encrypt:
 		lea	128($sbox),$sbox
 		sub	\$1,%ecx
 	jnz	.Lcbc_prefetch_te
-	sub	\$2048,$sbox
+	lea	-2304($sbox),$sbox
 
-	test	\$-16,%rdx		# check upon length
-	mov	%rdx,%r10
+	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
-	jz	.Lcbc_enc_tail		# short input...
 
 .align	4
-.Lcbc_enc_loop:
+.Lcbc_fast_enc_loop:
 		xor	0($inp),$s0
 		xor	4($inp),$s1
 		xor	8($inp),$s2
 		xor	12($inp),$s3
-		mov	$inp,$ivec	# if ($verticalspin) save inp
-
 		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
 		call	_x86_64_AES_encrypt
 
-		mov	$ivec,$inp	# if ($verticalspin) restore inp
+		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)
 
-		mov	$_len,%r10
 		lea	16($inp),$inp
 		lea	16($out),$out
 		sub	\$16,%r10
 		test	\$-16,%r10
 		mov	%r10,$_len
-	jnz	.Lcbc_enc_loop
-	test	\$15,%r10
-	jnz	.Lcbc_enc_tail
+	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)
 
-.align	4
-.Lcbc_cleanup:
-	cmpl	\$0,$mark	# was the key schedule copied?
-	lea	$aes_key,%rdi
-	mov	$_rsp,%rsp
-	je	.Lcbc_exit
-		mov	\$240/8,%ecx
-		xor	%rax,%rax
-		.long	0x90AB48F3	# rep stosq
-.Lcbc_exit:
-	popfq
-	pop	%r15
-	pop	%r14
-	pop	%r13
-	pop	%r12
-	pop	%rbp
-	pop	%rbx
-.Lcbc_just_ret:
-	ret
-.align	4
-.Lcbc_enc_tail:
-	cmp	$inp,$out
-	je	.Lcbc_enc_in_place
-	mov	%r10,%rcx
-	mov	$inp,%rsi
-	mov	$out,%rdi
-	.long	0xF689A4F3		# rep movsb
-.Lcbc_enc_in_place:
-	mov	\$16,%rcx		# zero tail
-	sub	%r10,%rcx
-	xor	%rax,%rax
-	.long	0xF689AAF3		# rep stosb
-	mov	$out,$inp		# this is not a mistake!
-	movq	\$16,$_len		# len=16
-	jmp	.Lcbc_enc_loop		# one more spin...
+	jmp	.Lcbc_fast_cleanup
+
 #----------------------------- DECRYPT -----------------------------#
 .align	16
-.LDECRYPT:
-	lea	AES_Td-.Lcbc_pic_point($sbox),$sbox
-
-	# allocate aligned stack frame...
-	lea	-64-248(%rsp),$key
-	and	\$-64,$key
-
-	# ... and make it doesn't alias with AES_Td modulo 4096
-	mov	$sbox,%r10
-	lea	2304($sbox),%r11
-	mov	$key,%r12
-	and	\$0xFFF,%r10	# s = $sbox&0xfff
-	and	\$0xFFF,%r11	# e = ($sbox+2048+256)&0xfff
-	and	\$0xFFF,%r12	# p = %rsp&0xfff
-
-	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
-	jb	.Lcbc_td_break_out
-	sub	%r11,%r12
-	sub	%r12,$key
-	jmp	.Lcbc_td_ok
-.Lcbc_td_break_out:		# else %rsp -= (p-s)&0xfff + framesz
-	sub	%r10,%r12
-	and	\$0xFFF,%r12
-	add	\$320,%r12
-	sub	%r12,$key
-.align	4
-.Lcbc_td_ok:
-
-	xchg	%rsp,$key
-	add	\$8,%rsp	# reserve for return address!
-	mov	$key,$_rsp	# save %rsp
-	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	%rsi,$out
-	mov	%rdi,$inp
-	mov	%rcx,$key
-
-	# do we copy key schedule to stack?
-	mov	$key,%r10
-	sub	$sbox,%r10
-	and	\$0xfff,%r10
-	cmp	\$2304,%r10
-	jb	.Lcbc_do_dcopy
-	cmp	\$4096-248,%r10
-	jb	.Lcbc_skip_dcopy
-.align	4
-.Lcbc_do_dcopy:
-		mov	$key,%rsi
-		lea	$aes_key,%rdi
-		lea	$aes_key,$key
-		mov	\$240/8,%ecx
-		.long	0x90A548F3	# rep movsq
-		mov	(%rsi),%eax	# copy aes_key->rounds
-		mov	%eax,(%rdi)
-.Lcbc_skip_dcopy:
-	mov	$key,$keyp	# save key pointer
-
-	mov	\$18,%ecx
-.align	4
-.Lcbc_prefetch_td:
-		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_td
-	sub	\$2304,$sbox
-
+.LFAST_DECRYPT:
 	cmp	$inp,$out
-	je	.Lcbc_dec_in_place
+	je	.Lcbc_fast_dec_in_place
 
 	mov	%rbp,$ivec
 .align	4
-.Lcbc_dec_loop:
-		mov	0($inp),$s0		# read input
+.Lcbc_fast_dec_loop:
+		mov	0($inp),$s0	# read input
 		mov	4($inp),$s1
 		mov	8($inp),$s2
 		mov	12($inp),$s3
-		mov	$inp,8+$ivec	# if ($verticalspin) save inp
-
 		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
 		call	_x86_64_AES_decrypt
 
 		mov	$ivec,%rbp	# load ivp
-		mov	8+$ivec,$inp	# if ($verticalspin) restore inp
+		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
 
-		mov	$_len,%r10	# load len
 		sub	\$16,%r10
-		jc	.Lcbc_dec_partial
 		mov	%r10,$_len	# update len
 		mov	%rbp,$ivec	# update ivp
 
@@ -1890,81 +1830,281 @@ AES_cbc_encrypt:
 
 		lea	16($inp),$inp
 		lea	16($out),$out
-	jnz	.Lcbc_dec_loop
-.Lcbc_dec_end:
+	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_cleanup
-
-.align	4
-.Lcbc_dec_partial:
-	mov	$s0,0+$ivec		# dump output to stack
-	mov	$s1,4+$ivec
-	mov	$s2,8+$ivec
-	mov	$s3,12+$ivec
-	mov	$out,%rdi
-	lea	$ivec,%rsi
-	mov	\$16,%rcx
-	add	%r10,%rcx		# number of bytes to copy
-	.long	0xF689A4F3		# rep movsb
-	jmp	.Lcbc_dec_end
+	jmp	.Lcbc_fast_cleanup
 
 .align	16
-.Lcbc_dec_in_place:
+.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
 
-		mov	$inp,$ivec	# if ($verticalspin) save inp
-		mov	$keyp,$key
 		call	_x86_64_AES_decrypt
 
-		mov	$ivec,$inp	# if ($verticalspin) restore inp
-		mov	$_ivp,%rbp
-		xor	0(%rbp),$s0
-		xor	4(%rbp),$s1
-		xor	8(%rbp),$s2
-		xor	12(%rbp),$s3
+		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	0($inp),%r10	# copy input to iv
-		mov	8($inp),%r11
-		mov	%r10,0(%rbp)
-		mov	%r11,8(%rbp)
+		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)
 
-		mov	$_len,%rcx
 		lea	16($inp),$inp
 		lea	16($out),$out
-		sub	\$16,%rcx
-		jc	.Lcbc_dec_in_place_partial
-		mov	%rcx,$_len
-	jnz	.Lcbc_dec_in_place
-	jmp	.Lcbc_cleanup
+		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_dec_in_place_partial:
-	# one can argue if this is actually required
-	lea	($out,%rcx),%rdi
-	lea	(%rbp,%rcx),%rsi
-	neg	%rcx
-	.long	0xF689A4F3	# rep movsb	# restore tail
-	jmp	.Lcbc_cleanup
+.Lcbc_fast_cleanup:
+	cmpl	\$0,$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	AES_cbc_encrypt,.-AES_cbc_encrypt
 ___
 }
 
 $code.=<<___;
-.globl	AES_Te
 .align	64
-AES_Te:
+.LAES_Te:
 ___
 	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
 	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
@@ -2171,9 +2311,8 @@ $code.=<<___;
 	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
 ___
 $code.=<<___;
-.globl	AES_Td
 .align	64
-AES_Td:
+.LAES_Td:
 ___
 	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
 	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
@@ -2385,8 +2524,285 @@ $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_AES_encrypt
+	.rva	.LSEH_end_AES_encrypt
+	.rva	.LSEH_info_AES_encrypt
+
+	.rva	.LSEH_begin_AES_decrypt
+	.rva	.LSEH_end_AES_decrypt
+	.rva	.LSEH_info_AES_decrypt
+
+	.rva	.LSEH_begin_AES_set_encrypt_key
+	.rva	.LSEH_end_AES_set_encrypt_key
+	.rva	.LSEH_info_AES_set_encrypt_key
+
+	.rva	.LSEH_begin_AES_set_decrypt_key
+	.rva	.LSEH_end_AES_set_decrypt_key
+	.rva	.LSEH_info_AES_set_decrypt_key
+
+	.rva	.LSEH_begin_AES_cbc_encrypt
+	.rva	.LSEH_end_AES_cbc_encrypt
+	.rva	.LSEH_info_AES_cbc_encrypt
+
+.section	.xdata
+.align	8
+.LSEH_info_AES_encrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
+.LSEH_info_AES_decrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
+.LSEH_info_AES_set_encrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
+.LSEH_info_AES_set_decrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
+.LSEH_info_AES_cbc_encrypt:
+	.byte	9,0,0,0
+	.rva	cbc_se_handler
+___
+}
+
 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 
 print $code;