Upgraded to openssl-1.0.2

18 files changed, 14649 insertions, 1732 deletions

diff --git a/openssl/crypto/aes/asm/aes-586.pl b/openssl/crypto/aes/asm/aes-586.pl
index 687ed811b..451d0e0ed 100644
--- a/openssl/crypto/aes/asm/aes-586.pl
+++ b/openssl/crypto/aes/asm/aes-586.pl
@@ -39,7 +39,7 @@
 # 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 private_AES_set_[de|en]cryption_key.
+# 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
@@ -103,11 +103,12 @@
 # byte for 128-bit key.
 #
 #		ECB encrypt	ECB decrypt	CBC large chunk
-# P4		56[60]		84[100]		23
-# AMD K8	48[44]		70[79]		18
-# PIII		41[50]		61[91]		24
-# Core 2	32[38]		45[70]		18.5
-# Pentium	120		160		77
+# 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.
 #
@@ -242,7 +243,7 @@ $vertical_spin=0;	# shift "verticaly" defaults to 0, because of
 
 sub encvert()
 { my ($te,@s) = @_;
-  my $v0 = $acc, $v1 = $key;
+  my ($v0,$v1) = ($acc,$key);
 
 	&mov	($v0,$s[3]);				# copy s3
 	&mov	(&DWP(4,"esp"),$s[2]);			# save s2
@@ -299,7 +300,7 @@ sub encvert()
 # Another experimental routine, which features "horizontal spin," but
 # eliminates one reference to stack. Strangely enough runs slower...
 sub enchoriz()
-{ my $v0 = $key, $v1 = $acc;
+{ my ($v0,$v1) = ($key,$acc);
 
 	&movz	($v0,&LB($s0));			#  3, 2, 1, 0*
 	&rotr	($s2,8);			#  8,11,10, 9
@@ -427,7 +428,7 @@ sub sse_encbody()
 ######################################################################
 
 sub enccompact()
-{ my $Fn = mov;
+{ my $Fn = \&mov;
   while ($#_>5) { pop(@_); $Fn=sub{}; }
   my ($i,$te,@s)=@_;
   my $tmp = $key;
@@ -476,24 +477,25 @@ sub enctransform()
   my $tmp = $tbl;
   my $r2  = $key ;
 
-	&mov	($acc,$s[$i]);
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
-	&shr	($tmp,7);
+	&and	($tmp,$s[$i]);
 	&lea	($r2,&DWP(0,$s[$i],$s[$i]));
-	&sub	($acc,$tmp);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
 	&and	($r2,0xfefefefe);
-	&and	($acc,0x1b1b1b1b);
+	&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	($s[$i],$acc)	# ROTATE(r2^r0,24) ^ r2
-	&rotr	($tmp,16);
-	&xor	($s[$i],$tmp);
-	&rotr	($tmp,8);
-	&xor	($s[$i],$tmp);
+	&xor	($acc,$r2);
+	&mov	($tmp,0x80808080)	if ($i!=1);
+	&xor	($s[$i],$acc);	# ROTATE(r2^r0,24) ^ r2
 }
 
 &function_begin_B("_x86_AES_encrypt_compact");
@@ -526,6 +528,7 @@ sub enctransform()
 		&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);
@@ -607,82 +610,84 @@ sub sse_enccompact()
 	&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	($acc,&LB("eax"));		#  0
-	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
-	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
 	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
+	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
+	&movz	($key,&LB("ebx"));		# 10
 	&movz	("edx",&BP(-128,$tbl,"edx",1));	#  1
-	&shl	("edx",8);			#  1
 	&shr	("eax",16);			#  5, 4
+	&shl	("edx",8);			#  1
 
-	&movz	($acc,&LB("ebx"));		# 10
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 10
+	&movz	($key,&HB("ebx"));		# 11
 	&shl	($acc,16);			# 10
-	&or	("ecx",$acc);			# 10
 	&pshufw	("mm6","mm4",0x08);		# 13,12, 9, 8
-	&movz	($acc,&HB("ebx"));		# 11
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 11
+	&or	("ecx",$acc);			# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  5
 	&shl	($acc,24);			# 11
-	&or	("edx",$acc);			# 11
 	&shr	("ebx",16);			# 15,14
+	&or	("edx",$acc);			# 11
 
-	&movz	($acc,&HB("eax"));		#  5
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  5
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  5
+	&movz	($key,&HB("ebx"));		# 15
 	&shl	($acc,8);			#  5
 	&or	("ecx",$acc);			#  5
-	&movz	($acc,&HB("ebx"));		# 15
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 15
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 15
+	&movz	($key,&LB("eax"));		#  4
 	&shl	($acc,24);			# 15
 	&or	("ecx",$acc);			# 15
-	&movd	("mm0","ecx");			# t[0] collected
 
-	&movz	($acc,&LB("eax"));		#  4
-	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  4
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
 	&movd	("eax","mm2");			#  7, 6, 3, 2
-	&movz	($acc,&LB("ebx"));		# 14
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 14
-	&shl	($acc,16);			# 14
+	&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
 
-	&movd	("ebx","mm6");			# 13,12, 9, 8
-	&movz	($acc,&HB("eax"));		#  3
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  3
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  3
+	&movz	($key,&HB("ebx"));		#  9
 	&shl	($acc,24);			#  3
 	&or	("ecx",$acc);			#  3
-	&movz	($acc,&HB("ebx"));		#  9
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  9
+	&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
-	&movd	("mm1","ecx");			# t[1] collected
 
-	&movz	($acc,&LB("ebx"));		#  8
-	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  8
-	&shr	("ebx",16);			# 13,12
-	&movz	($acc,&LB("eax"));		#  2
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  2
-	&shl	($acc,16);			#  2
-	&or	("ecx",$acc);			#  2
+	&movz	($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	($acc,&HB("eax"));		#  7
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  7
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
 	&shl	($acc,24);			#  7
-	&or	("ecx",$acc);			#  7
-	&and	("eax",0xff);			#  6
+	&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,$key,1));	# 13
 	&or	("edx","eax");			#  6
-	&movz	($acc,&HB("ebx"));		# 13
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 13
 	&shl	($acc,8);			# 13
-	&or	("ecx",$acc);			# 13
-	&movd	("mm4","ecx");			# t[2] collected
-	&and	("ebx",0xff);			# 12
 	&movz	("ebx",&BP(-128,$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
@@ -1222,7 +1227,7 @@ sub enclast()
 ######################################################################
 
 sub deccompact()
-{ my $Fn = mov;
+{ my $Fn = \&mov;
   while ($#_>5) { pop(@_); $Fn=sub{}; }
   my ($i,$td,@s)=@_;
   my $tmp = $key;
@@ -1270,30 +1275,30 @@ sub dectransform()
   my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
   my $tp8 = $tbl;
 
-	&mov	($acc,$s[$i]);
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
+	&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	($acc,$tp2);
-	&mov	($tp2,$acc);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
 
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
+	&and	($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	($acc,$tp4);
-	&mov	($tp4,$acc);
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
 
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
+	&and	($tmp,$tp4);
+	&mov	($acc,$tmp);
 	&shr	($tmp,7);
 	&lea	($tp8,&DWP(0,$tp4,$tp4));
 	&sub	($acc,$tmp);
@@ -1305,13 +1310,13 @@ sub dectransform()
 
 	&xor	($s[$i],$tp2);
 	&xor	($tp2,$tp8);
-	&rotl	($tp2,24);
 	&xor	($s[$i],$tp4);
 	&xor	($tp4,$tp8);
-	&rotl	($tp4,16);
+	&rotl	($tp2,24);
 	&xor	($s[$i],$tp8);	# ^= tp8^(tp4^tp1)^(tp2^tp1)
-	&rotl	($tp8,8);
+	&rotl	($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 $s1
@@ -1389,85 +1394,87 @@ sub dectransform()
 sub sse_deccompact()
 {
 	&pshufw	("mm1","mm0",0x0c);		#  7, 6, 1, 0
+	&pshufw	("mm5","mm4",0x09);		# 13,12,11,10
 	&movd	("eax","mm1");			#  7, 6, 1, 0
+	&movd	("ebx","mm5");			# 13,12,11,10
+	&mov	($__key,$key);
 
-	&pshufw	("mm5","mm4",0x09);		# 13,12,11,10
 	&movz	($acc,&LB("eax"));		#  0
-	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
-	&movd	("ebx","mm5");			# 13,12,11,10
 	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x06);		#  3, 2, 5, 4
+	&movz	("ecx",&BP(-128,$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
 
-	&pshufw	("mm2","mm0",0x06);		#  3, 2, 5, 4
-	&movz	($acc,&LB("ebx"));		# 10
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 10
+	&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
-	&shr	("eax",16);			#  7, 6
-	&movz	($acc,&HB("ebx"));		# 11
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 11
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  7
 	&shl	($acc,24);			# 11
-	&or	("edx",$acc);			# 11
 	&shr	("ebx",16);			# 13,12
+	&or	("edx",$acc);			# 11
 
-	&pshufw	("mm6","mm4",0x03);		# 9, 8,15,14
-	&movz	($acc,&HB("eax"));		#  7
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  7
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
 	&shl	($acc,24);			#  7
 	&or	("ecx",$acc);			#  7
-	&movz	($acc,&HB("ebx"));		# 13
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 13
+	&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
-	&movd	("mm0","ecx");			# t[0] collected
 
-	&movz	($acc,&LB("eax"));		#  6
-	&movd	("eax","mm2");			#  3, 2, 5, 4
-	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  6
-	&shl	("ecx",16);			#  6
-	&movz	($acc,&LB("ebx"));		# 12
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  6
+	&movz	($key,&LB("ebx"));		# 12
+	&shl	($acc,16);			#  6
 	&movd	("ebx","mm6");			#  9, 8,15,14
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 12
+	&movd	("mm0","ecx");			# t[0] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	# 12
+	&movz	($key,&LB("eax"));		#  4
 	&or	("ecx",$acc);			# 12
 
-	&movz	($acc,&LB("eax"));		#  4
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  4
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
 	&or	("edx",$acc);			#  4
-	&movz	($acc,&LB("ebx"));		# 14
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 14
+	&movz	($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
-	&movd	("mm1","edx");			# t[1] collected
 
-	&movz	($acc,&HB("eax"));		#  5
-	&movz	("edx",&BP(-128,$tbl,$acc,1));	#  5
-	&shl	("edx",8);			#  5
-	&movz	($acc,&HB("ebx"));		# 15
-	&shr	("eax",16);			#  3, 2
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	# 15
-	&shl	($acc,24);			# 15
-	&or	("edx",$acc);			# 15
+	&movz	($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	($acc,&HB("ebx"));		#  9
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  9
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  9
+	&movz	($key,&LB("eax"));		#  2
 	&shl	($acc,8);			#  9
-	&or	("ecx",$acc);			#  9
-	&and	("ebx",0xff);			#  8
+	&movz	("eax",&HB("eax"));		#  3
 	&movz	("ebx",&BP(-128,$tbl,"ebx",1));	#  8
+	&or	("ecx",$acc);			#  9
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  2
 	&or	("edx","ebx");			#  8
-	&movz	($acc,&LB("eax"));		#  2
-	&movz	($acc,&BP(-128,$tbl,$acc,1));	#  2
 	&shl	($acc,16);			#  2
-	&or	("edx",$acc);			#  2
-	&movd	("mm4","edx");			# t[2] collected
-	&movz	("eax",&HB("eax"));		#  3
 	&movz	("eax",&BP(-128,$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
@@ -2181,8 +2188,8 @@ my $mark=&DWP(76+240,"esp");	# copy of aes_key->rounds
 	&mov	("ecx",240/4);
 	&xor	("eax","eax");
 	&align	(4);
-	&data_word(0xABF3F689);	# rep stosd
-	&set_label("skip_ezero")
+	&data_word(0xABF3F689);		# rep stosd
+	&set_label("skip_ezero");
 	&mov	("esp",$_esp);
 	&popf	();
     &set_label("drop_out");
@@ -2301,8 +2308,8 @@ my $mark=&DWP(76+240,"esp");	# copy of aes_key->rounds
 	&mov	("ecx",240/4);
 	&xor	("eax","eax");
 	&align	(4);
-	&data_word(0xABF3F689);	# rep stosd
-	&set_label("skip_dzero")
+	&data_word(0xABF3F689);		# rep stosd
+	&set_label("skip_dzero");
 	&mov	("esp",$_esp);
 	&popf	();
 	&function_end_A();
@@ -2865,32 +2872,32 @@ sub deckey()
 { my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
   my $tmp = $tbl;
 
-	&mov	($acc,$tp1);
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
-	&shr	($tmp,7);
+	&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	($acc,0x1b1b1b1b);
-	&xor	($acc,$tp2);
-	&mov	($tp2,$acc);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
 
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
-	&shr	($tmp,7);
+	&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	($acc,$tp4);
-	&mov	($tp4,$acc);
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
 
-	&and	($acc,0x80808080);
-	&mov	($tmp,$acc);
-	&shr	($tmp,7);
+	&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);
diff --git a/openssl/crypto/aes/asm/aes-armv4.pl b/openssl/crypto/aes/asm/aes-armv4.pl
index 86b86c4a0..4f8917089 100644
--- a/openssl/crypto/aes/asm/aes-armv4.pl
+++ b/openssl/crypto/aes/asm/aes-armv4.pl
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 
 # ====================================================================
-# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# 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/.
@@ -51,9 +51,23 @@ $key="r11";
 $rounds="r12";
 
 $code=<<___;
-#include "arm_arch.h"
+#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
@@ -167,7 +181,11 @@ AES_Te:
 .type   AES_encrypt,%function
 .align	5
 AES_encrypt:
+#if __ARM_ARCH__<7
 	sub	r3,pc,#8		@ AES_encrypt
+#else
+	adr	r3,AES_encrypt
+#endif
 	stmdb   sp!,{r1,r4-r12,lr}
 	mov	$rounds,r0		@ inp
 	mov	$key,r2
@@ -409,11 +427,21 @@ _armv4_AES_encrypt:
 .align	5
 private_AES_set_encrypt_key:
 _armv4_AES_set_encrypt_key:
+#if __ARM_ARCH__<7
 	sub	r3,pc,#8		@ AES_set_encrypt_key
+#else
+	adr	r3,private_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
 
@@ -422,6 +450,9 @@ _armv4_AES_set_encrypt_key:
 	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
 
@@ -576,6 +607,9 @@ _armv4_AES_set_encrypt_key:
 	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
 
@@ -645,6 +679,9 @@ _armv4_AES_set_encrypt_key:
 	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
 
@@ -674,11 +711,17 @@ _armv4_AES_set_encrypt_key:
 	str	$i3,[$key,#-4]
 	b	.L256_loop
 
+.align	2
 .Ldone:	mov	r0,#0
 	ldmia   sp!,{r4-r12,lr}
-.Labrt:	tst	lr,#1
+.Labrt:
+#if __ARM_ARCH__>=5
+	ret				@ bx lr
+#else
+	tst	lr,#1
 	moveq	pc,lr			@ be binary compatible with V4, yet
 	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
 .size	private_AES_set_encrypt_key,.-private_AES_set_encrypt_key
 
 .global private_AES_set_decrypt_key
@@ -688,34 +731,57 @@ private_AES_set_decrypt_key:
 	str	lr,[sp,#-4]!            @ push lr
 	bl	_armv4_AES_set_encrypt_key
 	teq	r0,#0
-	ldrne	lr,[sp],#4              @ pop lr
+	ldr	lr,[sp],#4              @ pop lr
 	bne	.Labrt
 
-	stmdb   sp!,{r4-r12}
+	mov	r0,r2			@ AES_set_encrypt_key preserves r2,
+	mov	r1,r2			@ which is AES_KEY *key
+	b	_armv4_AES_set_enc2dec_key
+.size	private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
 
-	ldr	$rounds,[r2,#240]	@ AES_set_encrypt_key preserves r2,
-	mov	$key,r2			@ which is AES_KEY *key
-	mov	$i1,r2
-	add	$i2,r2,$rounds,lsl#4
+@ void AES_set_enc2dec_key(const AES_KEY *inp,AES_KEY *out)
+.global	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
 
-.Linv:	ldr	$s0,[$i1]
+	ldr	$s0,[$i1]
 	ldr	$s1,[$i1,#4]
 	ldr	$s2,[$i1,#8]
 	ldr	$s3,[$i1,#12]
-	ldr	$t1,[$i2]
-	ldr	$t2,[$i2,#4]
-	ldr	$t3,[$i2,#8]
-	ldr	$i3,[$i2,#12]
-	str	$s0,[$i2],#-16
-	str	$s1,[$i2,#16+4]
-	str	$s2,[$i2,#16+8]
-	str	$s3,[$i2,#16+12]
-	str	$t1,[$i1],#16
-	str	$t2,[$i1,#-12]
-	str	$t3,[$i1,#-8]
-	str	$i3,[$i1,#-4]
-	teq	$i1,$i2
-	bne	.Linv
+	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;
@@ -773,7 +839,7 @@ $code.=<<___;
 	moveq	pc,lr			@ be binary compatible with V4, yet
 	bx	lr			@ interoperable with Thumb ISA:-)
 #endif
-.size	private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
+.size	AES_set_enc2dec_key,.-AES_set_enc2dec_key
 
 .type	AES_Td,%object
 .align	5
@@ -883,7 +949,11 @@ AES_Td:
 .type   AES_decrypt,%function
 .align	5
 AES_decrypt:
+#if __ARM_ARCH__<7
 	sub	r3,pc,#8		@ AES_decrypt
+#else
+	adr	r3,AES_decrypt
+#endif
 	stmdb   sp!,{r1,r4-r12,lr}
 	mov	$rounds,r0		@ inp
 	mov	$key,r2
@@ -1080,8 +1150,9 @@ _armv4_AES_decrypt:
 	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	$s1,[$tbl,$s1,lsr#24]	@ Td4[s1>>24]
+	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]
@@ -1094,7 +1165,8 @@ _armv4_AES_decrypt:
 	ldrb	$i2,[$tbl,$i2]		@ Td4[s2>>0]
 	and	$i3,lr,$s2,lsr#16
 
-	ldrb	$s2,[$tbl,$s2,lsr#24]	@ Td4[s2>>24]
+	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
@@ -1106,8 +1178,9 @@ _armv4_AES_decrypt:
 	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	$s3,[$tbl,$s3,lsr#24]	@ Td4[s3>>24]
+	ldrb	$s3,[$s3]		@ Td4[s3>>24]
 	eor	$s0,$s0,$i1,lsl#16
 	ldr	$i1,[$key,#0]
 	eor	$s1,$s1,$i2,lsl#8
@@ -1130,5 +1203,15 @@ _armv4_AES_decrypt:
 ___
 
 $code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
+$code =~ s/\bret\b/bx\tlr/gm;
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+	last if (!s/^#/@/ and !/^$/);
+	print;
+}
+close SELF;
+
 print $code;
 close STDOUT;	# enforce flush
diff --git a/openssl/crypto/aes/asm/aes-mips.pl b/openssl/crypto/aes/asm/aes-mips.pl
index e52395421..4de3ee26b 100644
--- a/openssl/crypto/aes/asm/aes-mips.pl
+++ b/openssl/crypto/aes/asm/aes-mips.pl
@@ -20,6 +20,13 @@
 # thing about this module is its endian neutrality, which means that
 # it processes data without ever changing byte order...
 
+# September 2012
+#
+# Add MIPS32R2 (~10% less instructions) and SmartMIPS ASE (further
+# ~25% less instructions) code. Note that there is no run-time switch,
+# instead, code path is chosen upon pre-process time, pass -mips32r2
+# or/and -msmartmips.
+
 ######################################################################
 # There is a number of MIPS ABI in use, O32 and N32/64 are most
 # widely used. Then there is a new contender: NUBI. It appears that if
@@ -47,11 +54,12 @@
 # ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
 # ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
 #
-$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
+$flavour = shift || "o32"; # supported flavours are o32,n32,64,nubi32,nubi64
 
 if ($flavour =~ /64|n32/i) {
 	$PTR_ADD="dadd";	# incidentally works even on n32
 	$PTR_SUB="dsub";	# incidentally works even on n32
+	$PTR_INS="dins";
 	$REG_S="sd";
 	$REG_L="ld";
 	$PTR_SLL="dsll";	# incidentally works even on n32
@@ -59,6 +67,7 @@ if ($flavour =~ /64|n32/i) {
 } else {
 	$PTR_ADD="add";
 	$PTR_SUB="sub";
+	$PTR_INS="ins";
 	$REG_S="sw";
 	$REG_L="lw";
 	$PTR_SLL="sll";
@@ -70,7 +79,7 @@ $pf = ($flavour =~ /nubi/i) ? $t0 : $t2;
 #
 ######################################################################
 
-$big_endian=(`echo MIPSEL | $ENV{CC} -E -P -`=~/MIPSEL/)?1:0;
+$big_endian=(`echo MIPSEL | $ENV{CC} -E -`=~/MIPSEL/)?1:0 if ($ENV{CC});
 
 for (@ARGV) {	$output=$_ if (/^\w[\w\-]*\.\w+$/);	}
 open STDOUT,">$output";
@@ -89,7 +98,11 @@ $code.=<<___;
 # include <openssl/fipssyms.h>
 #endif
 
-#if !defined(__vxworks) || defined(__pic__)
+#if defined(__mips_smartmips) && !defined(_MIPS_ARCH_MIPS32R2)
+#define _MIPS_ARCH_MIPS32R2
+#endif
+
+#if !defined(__mips_eabi) && (!defined(__vxworks) || defined(__pic__))
 .option	pic2
 #endif
 .set	noat
@@ -125,6 +138,89 @@ _mips_AES_encrypt:
 	xor	$s3,$t3
 
 	sub	$cnt,1
+#if defined(__mips_smartmips)
+	ext	$i0,$s1,16,8
+.Loop_enc:
+	ext	$i1,$s2,16,8
+	ext	$i2,$s3,16,8
+	ext	$i3,$s0,16,8
+	lwxs	$t0,$i0($Tbl)		# Te1[s1>>16]
+	ext	$i0,$s2,8,8
+	lwxs	$t1,$i1($Tbl)		# Te1[s2>>16]
+	ext	$i1,$s3,8,8
+	lwxs	$t2,$i2($Tbl)		# Te1[s3>>16]
+	ext	$i2,$s0,8,8
+	lwxs	$t3,$i3($Tbl)		# Te1[s0>>16]
+	ext	$i3,$s1,8,8
+
+	lwxs	$t4,$i0($Tbl)		# Te2[s2>>8]
+	ext	$i0,$s3,0,8
+	lwxs	$t5,$i1($Tbl)		# Te2[s3>>8]
+	ext	$i1,$s0,0,8
+	lwxs	$t6,$i2($Tbl)		# Te2[s0>>8]
+	ext	$i2,$s1,0,8
+	lwxs	$t7,$i3($Tbl)		# Te2[s1>>8]
+	ext	$i3,$s2,0,8
+
+	lwxs	$t8,$i0($Tbl)		# Te3[s3]
+	ext	$i0,$s0,24,8
+	lwxs	$t9,$i1($Tbl)		# Te3[s0]
+	ext	$i1,$s1,24,8
+	lwxs	$t10,$i2($Tbl)		# Te3[s1]
+	ext	$i2,$s2,24,8
+	lwxs	$t11,$i3($Tbl)		# Te3[s2]
+	ext	$i3,$s3,24,8
+
+	rotr	$t0,$t0,8
+	rotr	$t1,$t1,8
+	rotr	$t2,$t2,8
+	rotr	$t3,$t3,8
+
+	rotr	$t4,$t4,16
+	rotr	$t5,$t5,16
+	rotr	$t6,$t6,16
+	rotr	$t7,$t7,16
+
+	xor	$t0,$t4
+	lwxs	$t4,$i0($Tbl)		# Te0[s0>>24]
+	xor	$t1,$t5
+	lwxs	$t5,$i1($Tbl)		# Te0[s1>>24]
+	xor	$t2,$t6
+	lwxs	$t6,$i2($Tbl)		# Te0[s2>>24]
+	xor	$t3,$t7
+	lwxs	$t7,$i3($Tbl)		# Te0[s3>>24]
+
+	rotr	$t8,$t8,24
+	lw	$s0,0($key0)
+	rotr	$t9,$t9,24
+	lw	$s1,4($key0)
+	rotr	$t10,$t10,24
+	lw	$s2,8($key0)
+	rotr	$t11,$t11,24
+	lw	$s3,12($key0)
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	sub	$cnt,1
+	$PTR_ADD $key0,16
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+	.set	noreorder
+	bnez	$cnt,.Loop_enc
+	ext	$i0,$s1,16,8
+
+	_xtr	$i0,$s1,16-2
+#else
 	_xtr	$i0,$s1,16-2
 .Loop_enc:
 	_xtr	$i1,$s2,16-2
@@ -138,19 +234,29 @@ _mips_AES_encrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+	lw	$t0,0($i0)		# Te1[s1>>16]
+	_xtr	$i0,$s2,8-2
+	lw	$t1,0($i1)		# Te1[s2>>16]
+	_xtr	$i1,$s3,8-2
+	lw	$t2,0($i2)		# Te1[s3>>16]
+	_xtr	$i2,$s0,8-2
+	lw	$t3,0($i3)		# Te1[s0>>16]
+	_xtr	$i3,$s1,8-2
+#else
 	lwl	$t0,3($i0)		# Te1[s1>>16]
 	lwl	$t1,3($i1)		# Te1[s2>>16]
 	lwl	$t2,3($i2)		# Te1[s3>>16]
 	lwl	$t3,3($i3)		# Te1[s0>>16]
 	lwr	$t0,2($i0)		# Te1[s1>>16]
-	lwr	$t1,2($i1)		# Te1[s2>>16]
-	lwr	$t2,2($i2)		# Te1[s3>>16]
-	lwr	$t3,2($i3)		# Te1[s0>>16]
-
 	_xtr	$i0,$s2,8-2
+	lwr	$t1,2($i1)		# Te1[s2>>16]
 	_xtr	$i1,$s3,8-2
+	lwr	$t2,2($i2)		# Te1[s3>>16]
 	_xtr	$i2,$s0,8-2
+	lwr	$t3,2($i3)		# Te1[s0>>16]
 	_xtr	$i3,$s1,8-2
+#endif
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -159,19 +265,88 @@ _mips_AES_encrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+	rotr	$t0,$t0,8
+	rotr	$t1,$t1,8
+	rotr	$t2,$t2,8
+	rotr	$t3,$t3,8
+# if defined(_MIPSEL)
+	lw	$t4,0($i0)		# Te2[s2>>8]
+	_xtr	$i0,$s3,0-2
+	lw	$t5,0($i1)		# Te2[s3>>8]
+	_xtr	$i1,$s0,0-2
+	lw	$t6,0($i2)		# Te2[s0>>8]
+	_xtr	$i2,$s1,0-2
+	lw	$t7,0($i3)		# Te2[s1>>8]
+	_xtr	$i3,$s2,0-2
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lw	$t8,0($i0)		# Te3[s3]
+	$PTR_INS $i0,$s0,2,8
+	lw	$t9,0($i1)		# Te3[s0]
+	$PTR_INS $i1,$s1,2,8
+	lw	$t10,0($i2)		# Te3[s1]
+	$PTR_INS $i2,$s2,2,8
+	lw	$t11,0($i3)		# Te3[s2]
+	$PTR_INS $i3,$s3,2,8
+# else
+	lw	$t4,0($i0)		# Te2[s2>>8]
+	$PTR_INS $i0,$s3,2,8
+	lw	$t5,0($i1)		# Te2[s3>>8]
+	$PTR_INS $i1,$s0,2,8
+	lw	$t6,0($i2)		# Te2[s0>>8]
+	$PTR_INS $i2,$s1,2,8
+	lw	$t7,0($i3)		# Te2[s1>>8]
+	$PTR_INS $i3,$s2,2,8
+
+	lw	$t8,0($i0)		# Te3[s3]
+	_xtr	$i0,$s0,24-2
+	lw	$t9,0($i1)		# Te3[s0]
+	_xtr	$i1,$s1,24-2
+	lw	$t10,0($i2)		# Te3[s1]
+	_xtr	$i2,$s2,24-2
+	lw	$t11,0($i3)		# Te3[s2]
+	_xtr	$i3,$s3,24-2
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+# endif
+	rotr	$t4,$t4,16
+	rotr	$t5,$t5,16
+	rotr	$t6,$t6,16
+	rotr	$t7,$t7,16
+
+	rotr	$t8,$t8,24
+	rotr	$t9,$t9,24
+	rotr	$t10,$t10,24
+	rotr	$t11,$t11,24
+#else
 	lwl	$t4,2($i0)		# Te2[s2>>8]
 	lwl	$t5,2($i1)		# Te2[s3>>8]
 	lwl	$t6,2($i2)		# Te2[s0>>8]
 	lwl	$t7,2($i3)		# Te2[s1>>8]
 	lwr	$t4,1($i0)		# Te2[s2>>8]
-	lwr	$t5,1($i1)		# Te2[s3>>8]
-	lwr	$t6,1($i2)		# Te2[s0>>8]
-	lwr	$t7,1($i3)		# Te2[s1>>8]
-
 	_xtr	$i0,$s3,0-2
+	lwr	$t5,1($i1)		# Te2[s3>>8]
 	_xtr	$i1,$s0,0-2
+	lwr	$t6,1($i2)		# Te2[s0>>8]
 	_xtr	$i2,$s1,0-2
+	lwr	$t7,1($i3)		# Te2[s1>>8]
 	_xtr	$i3,$s2,0-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -185,14 +360,14 @@ _mips_AES_encrypt:
 	lwl	$t10,1($i2)		# Te3[s1]
 	lwl	$t11,1($i3)		# Te3[s2]
 	lwr	$t8,0($i0)		# Te3[s3]
-	lwr	$t9,0($i1)		# Te3[s0]
-	lwr	$t10,0($i2)		# Te3[s1]
-	lwr	$t11,0($i3)		# Te3[s2]
-
 	_xtr	$i0,$s0,24-2
+	lwr	$t9,0($i1)		# Te3[s0]
 	_xtr	$i1,$s1,24-2
+	lwr	$t10,0($i2)		# Te3[s1]
 	_xtr	$i2,$s2,24-2
+	lwr	$t11,0($i3)		# Te3[s2]
 	_xtr	$i3,$s3,24-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -201,24 +376,24 @@ _mips_AES_encrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#endif
 	xor	$t0,$t4
-	xor	$t1,$t5
-	xor	$t2,$t6
-	xor	$t3,$t7
 	lw	$t4,0($i0)		# Te0[s0>>24]
+	xor	$t1,$t5
 	lw	$t5,0($i1)		# Te0[s1>>24]
+	xor	$t2,$t6
 	lw	$t6,0($i2)		# Te0[s2>>24]
+	xor	$t3,$t7
 	lw	$t7,0($i3)		# Te0[s3>>24]
 
-	lw	$s0,0($key0)
-	lw	$s1,4($key0)
-	lw	$s2,8($key0)
-	lw	$s3,12($key0)
-
 	xor	$t0,$t8
+	lw	$s0,0($key0)
 	xor	$t1,$t9
+	lw	$s1,4($key0)
 	xor	$t2,$t10
+	lw	$s2,8($key0)
 	xor	$t3,$t11
+	lw	$s3,12($key0)
 
 	xor	$t0,$t4
 	xor	$t1,$t5
@@ -234,6 +409,7 @@ _mips_AES_encrypt:
 	.set	noreorder
 	bnez	$cnt,.Loop_enc
 	_xtr	$i0,$s1,16-2
+#endif
 
 	.set	reorder
 	_xtr	$i1,$s2,16-2
@@ -248,14 +424,14 @@ _mips_AES_encrypt:
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
 	lbu	$t0,2($i0)		# Te4[s1>>16]
-	lbu	$t1,2($i1)		# Te4[s2>>16]
-	lbu	$t2,2($i2)		# Te4[s3>>16]
-	lbu	$t3,2($i3)		# Te4[s0>>16]
-
 	_xtr	$i0,$s2,8-2
+	lbu	$t1,2($i1)		# Te4[s2>>16]
 	_xtr	$i1,$s3,8-2
+	lbu	$t2,2($i2)		# Te4[s3>>16]
 	_xtr	$i2,$s0,8-2
+	lbu	$t3,2($i3)		# Te4[s0>>16]
 	_xtr	$i3,$s1,8-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -264,15 +440,44 @@ _mips_AES_encrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+# if defined(_MIPSEL)
 	lbu	$t4,2($i0)		# Te4[s2>>8]
+	$PTR_INS $i0,$s0,2,8
 	lbu	$t5,2($i1)		# Te4[s3>>8]
+	$PTR_INS $i1,$s1,2,8
 	lbu	$t6,2($i2)		# Te4[s0>>8]
+	$PTR_INS $i2,$s2,2,8
 	lbu	$t7,2($i3)		# Te4[s1>>8]
+	$PTR_INS $i3,$s3,2,8
 
+	lbu	$t8,2($i0)		# Te4[s0>>24]
+	_xtr	$i0,$s3,0-2
+	lbu	$t9,2($i1)		# Te4[s1>>24]
+	_xtr	$i1,$s0,0-2
+	lbu	$t10,2($i2)		# Te4[s2>>24]
+	_xtr	$i2,$s1,0-2
+	lbu	$t11,2($i3)		# Te4[s3>>24]
+	_xtr	$i3,$s2,0-2
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+# else
+	lbu	$t4,2($i0)		# Te4[s2>>8]
 	_xtr	$i0,$s0,24-2
+	lbu	$t5,2($i1)		# Te4[s3>>8]
 	_xtr	$i1,$s1,24-2
+	lbu	$t6,2($i2)		# Te4[s0>>8]
 	_xtr	$i2,$s2,24-2
+	lbu	$t7,2($i3)		# Te4[s1>>8]
 	_xtr	$i3,$s3,24-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -282,18 +487,76 @@ _mips_AES_encrypt:
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
 	lbu	$t8,2($i0)		# Te4[s0>>24]
+	$PTR_INS $i0,$s3,2,8
 	lbu	$t9,2($i1)		# Te4[s1>>24]
+	$PTR_INS $i1,$s0,2,8
 	lbu	$t10,2($i2)		# Te4[s2>>24]
+	$PTR_INS $i2,$s1,2,8
 	lbu	$t11,2($i3)		# Te4[s3>>24]
+	$PTR_INS $i3,$s2,2,8
+# endif
+	_ins	$t0,16
+	_ins	$t1,16
+	_ins	$t2,16
+	_ins	$t3,16
 
+	_ins2	$t0,$t4,8
+	lbu	$t4,2($i0)		# Te4[s3]
+	_ins2	$t1,$t5,8
+	lbu	$t5,2($i1)		# Te4[s0]
+	_ins2	$t2,$t6,8
+	lbu	$t6,2($i2)		# Te4[s1]
+	_ins2	$t3,$t7,8
+	lbu	$t7,2($i3)		# Te4[s2]
+
+	_ins2	$t0,$t8,24
+	lw	$s0,0($key0)
+	_ins2	$t1,$t9,24
+	lw	$s1,4($key0)
+	_ins2	$t2,$t10,24
+	lw	$s2,8($key0)
+	_ins2	$t3,$t11,24
+	lw	$s3,12($key0)
+
+	_ins2	$t0,$t4,0
+	_ins2	$t1,$t5,0
+	_ins2	$t2,$t6,0
+	_ins2	$t3,$t7,0
+#else
+	lbu	$t4,2($i0)		# Te4[s2>>8]
+	_xtr	$i0,$s0,24-2
+	lbu	$t5,2($i1)		# Te4[s3>>8]
+	_xtr	$i1,$s1,24-2
+	lbu	$t6,2($i2)		# Te4[s0>>8]
+	_xtr	$i2,$s2,24-2
+	lbu	$t7,2($i3)		# Te4[s1>>8]
+	_xtr	$i3,$s3,24-2
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t8,2($i0)		# Te4[s0>>24]
 	_xtr	$i0,$s3,0-2
+	lbu	$t9,2($i1)		# Te4[s1>>24]
 	_xtr	$i1,$s0,0-2
+	lbu	$t10,2($i2)		# Te4[s2>>24]
 	_xtr	$i2,$s1,0-2
+	lbu	$t11,2($i3)		# Te4[s3>>24]
 	_xtr	$i3,$s2,0-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
 	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
 
 	_ins	$t0,16
 	_ins	$t1,16
@@ -306,27 +569,21 @@ _mips_AES_encrypt:
 	_ins	$t7,8
 
 	xor	$t0,$t4
-	xor	$t1,$t5
-	xor	$t2,$t6
-	xor	$t3,$t7
-
-	$PTR_ADD $i0,$Tbl
-	$PTR_ADD $i1,$Tbl
-	$PTR_ADD $i2,$Tbl
-	$PTR_ADD $i3,$Tbl
 	lbu	$t4,2($i0)		# Te4[s3]
+	xor	$t1,$t5
 	lbu	$t5,2($i1)		# Te4[s0]
+	xor	$t2,$t6
 	lbu	$t6,2($i2)		# Te4[s1]
+	xor	$t3,$t7
 	lbu	$t7,2($i3)		# Te4[s2]
 
 	_ins	$t8,24
-	_ins	$t9,24
-	_ins	$t10,24
-	_ins	$t11,24
-
 	lw	$s0,0($key0)
+	_ins	$t9,24
 	lw	$s1,4($key0)
+	_ins	$t10,24
 	lw	$s2,8($key0)
+	_ins	$t11,24
 	lw	$s3,12($key0)
 
 	xor	$t0,$t8
@@ -343,7 +600,7 @@ _mips_AES_encrypt:
 	xor	$t1,$t5
 	xor	$t2,$t6
 	xor	$t3,$t7
-
+#endif
 	xor	$s0,$t0
 	xor	$s1,$t1
 	xor	$s2,$t2
@@ -455,6 +712,89 @@ _mips_AES_decrypt:
 	xor	$s3,$t3
 
 	sub	$cnt,1
+#if defined(__mips_smartmips)
+	ext	$i0,$s3,16,8
+.Loop_dec:
+	ext	$i1,$s0,16,8
+	ext	$i2,$s1,16,8
+	ext	$i3,$s2,16,8
+	lwxs	$t0,$i0($Tbl)		# Td1[s3>>16]
+	ext	$i0,$s2,8,8
+	lwxs	$t1,$i1($Tbl)		# Td1[s0>>16]
+	ext	$i1,$s3,8,8
+	lwxs	$t2,$i2($Tbl)		# Td1[s1>>16]
+	ext	$i2,$s0,8,8
+	lwxs	$t3,$i3($Tbl)		# Td1[s2>>16]
+	ext	$i3,$s1,8,8
+
+	lwxs	$t4,$i0($Tbl)		# Td2[s2>>8]
+	ext	$i0,$s1,0,8
+	lwxs	$t5,$i1($Tbl)		# Td2[s3>>8]
+	ext	$i1,$s2,0,8
+	lwxs	$t6,$i2($Tbl)		# Td2[s0>>8]
+	ext	$i2,$s3,0,8
+	lwxs	$t7,$i3($Tbl)		# Td2[s1>>8]
+	ext	$i3,$s0,0,8
+
+	lwxs	$t8,$i0($Tbl)		# Td3[s1]
+	ext	$i0,$s0,24,8
+	lwxs	$t9,$i1($Tbl)		# Td3[s2]
+	ext	$i1,$s1,24,8
+	lwxs	$t10,$i2($Tbl)		# Td3[s3]
+	ext	$i2,$s2,24,8
+	lwxs	$t11,$i3($Tbl)		# Td3[s0]
+	ext	$i3,$s3,24,8
+
+	rotr	$t0,$t0,8
+	rotr	$t1,$t1,8
+	rotr	$t2,$t2,8
+	rotr	$t3,$t3,8
+
+	rotr	$t4,$t4,16
+	rotr	$t5,$t5,16
+	rotr	$t6,$t6,16
+	rotr	$t7,$t7,16
+
+	xor	$t0,$t4
+	lwxs	$t4,$i0($Tbl)		# Td0[s0>>24]
+	xor	$t1,$t5
+	lwxs	$t5,$i1($Tbl)		# Td0[s1>>24]
+	xor	$t2,$t6
+	lwxs	$t6,$i2($Tbl)		# Td0[s2>>24]
+	xor	$t3,$t7
+	lwxs	$t7,$i3($Tbl)		# Td0[s3>>24]
+
+	rotr	$t8,$t8,24
+	lw	$s0,0($key0)
+	rotr	$t9,$t9,24
+	lw	$s1,4($key0)
+	rotr	$t10,$t10,24
+	lw	$s2,8($key0)
+	rotr	$t11,$t11,24
+	lw	$s3,12($key0)
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	sub	$cnt,1
+	$PTR_ADD $key0,16
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+	.set	noreorder
+	bnez	$cnt,.Loop_dec
+	ext	$i0,$s3,16,8
+
+	_xtr	$i0,$s3,16-2
+#else
 	_xtr	$i0,$s3,16-2
 .Loop_dec:
 	_xtr	$i1,$s0,16-2
@@ -468,19 +808,88 @@ _mips_AES_decrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+	lw	$t0,0($i0)		# Td1[s3>>16]
+	_xtr	$i0,$s2,8-2
+	lw	$t1,0($i1)		# Td1[s0>>16]
+	_xtr	$i1,$s3,8-2
+	lw	$t2,0($i2)		# Td1[s1>>16]
+	_xtr	$i2,$s0,8-2
+	lw	$t3,0($i3)		# Td1[s2>>16]
+	_xtr	$i3,$s1,8-2
+#else
 	lwl	$t0,3($i0)		# Td1[s3>>16]
 	lwl	$t1,3($i1)		# Td1[s0>>16]
 	lwl	$t2,3($i2)		# Td1[s1>>16]
 	lwl	$t3,3($i3)		# Td1[s2>>16]
 	lwr	$t0,2($i0)		# Td1[s3>>16]
-	lwr	$t1,2($i1)		# Td1[s0>>16]
-	lwr	$t2,2($i2)		# Td1[s1>>16]
-	lwr	$t3,2($i3)		# Td1[s2>>16]
-
 	_xtr	$i0,$s2,8-2
+	lwr	$t1,2($i1)		# Td1[s0>>16]
 	_xtr	$i1,$s3,8-2
+	lwr	$t2,2($i2)		# Td1[s1>>16]
 	_xtr	$i2,$s0,8-2
+	lwr	$t3,2($i3)		# Td1[s2>>16]
 	_xtr	$i3,$s1,8-2
+#endif
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+	rotr	$t0,$t0,8
+	rotr	$t1,$t1,8
+	rotr	$t2,$t2,8
+	rotr	$t3,$t3,8
+# if defined(_MIPSEL)
+	lw	$t4,0($i0)		# Td2[s2>>8]
+	_xtr	$i0,$s1,0-2
+	lw	$t5,0($i1)		# Td2[s3>>8]
+	_xtr	$i1,$s2,0-2
+	lw	$t6,0($i2)		# Td2[s0>>8]
+	_xtr	$i2,$s3,0-2
+	lw	$t7,0($i3)		# Td2[s1>>8]
+	_xtr	$i3,$s0,0-2
+
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lw	$t8,0($i0)		# Td3[s1]
+	$PTR_INS $i0,$s0,2,8
+	lw	$t9,0($i1)		# Td3[s2]
+	$PTR_INS $i1,$s1,2,8
+	lw	$t10,0($i2)		# Td3[s3]
+	$PTR_INS $i2,$s2,2,8
+	lw	$t11,0($i3)		# Td3[s0]
+	$PTR_INS $i3,$s3,2,8
+#else
+	lw	$t4,0($i0)		# Td2[s2>>8]
+	$PTR_INS $i0,$s1,2,8
+	lw	$t5,0($i1)		# Td2[s3>>8]
+	$PTR_INS $i1,$s2,2,8
+	lw	$t6,0($i2)		# Td2[s0>>8]
+	$PTR_INS $i2,$s3,2,8
+	lw	$t7,0($i3)		# Td2[s1>>8]
+	$PTR_INS $i3,$s0,2,8
+
+	lw	$t8,0($i0)		# Td3[s1]
+	_xtr	$i0,$s0,24-2
+	lw	$t9,0($i1)		# Td3[s2]
+	_xtr	$i1,$s1,24-2
+	lw	$t10,0($i2)		# Td3[s3]
+	_xtr	$i2,$s2,24-2
+	lw	$t11,0($i3)		# Td3[s0]
+	_xtr	$i3,$s3,24-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -489,19 +898,30 @@ _mips_AES_decrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#endif
+	rotr	$t4,$t4,16
+	rotr	$t5,$t5,16
+	rotr	$t6,$t6,16
+	rotr	$t7,$t7,16
+
+	rotr	$t8,$t8,24
+	rotr	$t9,$t9,24
+	rotr	$t10,$t10,24
+	rotr	$t11,$t11,24
+#else
 	lwl	$t4,2($i0)		# Td2[s2>>8]
 	lwl	$t5,2($i1)		# Td2[s3>>8]
 	lwl	$t6,2($i2)		# Td2[s0>>8]
 	lwl	$t7,2($i3)		# Td2[s1>>8]
 	lwr	$t4,1($i0)		# Td2[s2>>8]
-	lwr	$t5,1($i1)		# Td2[s3>>8]
-	lwr	$t6,1($i2)		# Td2[s0>>8]
-	lwr	$t7,1($i3)		# Td2[s1>>8]
-
 	_xtr	$i0,$s1,0-2
+	lwr	$t5,1($i1)		# Td2[s3>>8]
 	_xtr	$i1,$s2,0-2
+	lwr	$t6,1($i2)		# Td2[s0>>8]
 	_xtr	$i2,$s3,0-2
+	lwr	$t7,1($i3)		# Td2[s1>>8]
 	_xtr	$i3,$s0,0-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -515,14 +935,14 @@ _mips_AES_decrypt:
 	lwl	$t10,1($i2)		# Td3[s3]
 	lwl	$t11,1($i3)		# Td3[s0]
 	lwr	$t8,0($i0)		# Td3[s1]
-	lwr	$t9,0($i1)		# Td3[s2]
-	lwr	$t10,0($i2)		# Td3[s3]
-	lwr	$t11,0($i3)		# Td3[s0]
-
 	_xtr	$i0,$s0,24-2
+	lwr	$t9,0($i1)		# Td3[s2]
 	_xtr	$i1,$s1,24-2
+	lwr	$t10,0($i2)		# Td3[s3]
 	_xtr	$i2,$s2,24-2
+	lwr	$t11,0($i3)		# Td3[s0]
 	_xtr	$i3,$s3,24-2
+
 	and	$i0,0x3fc
 	and	$i1,0x3fc
 	and	$i2,0x3fc
@@ -531,27 +951,25 @@ _mips_AES_decrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#endif
 
 	xor	$t0,$t4
-	xor	$t1,$t5
-	xor	$t2,$t6
-	xor	$t3,$t7
-
-
 	lw	$t4,0($i0)		# Td0[s0>>24]
+	xor	$t1,$t5
 	lw	$t5,0($i1)		# Td0[s1>>24]
+	xor	$t2,$t6
 	lw	$t6,0($i2)		# Td0[s2>>24]
+	xor	$t3,$t7
 	lw	$t7,0($i3)		# Td0[s3>>24]
 
-	lw	$s0,0($key0)
-	lw	$s1,4($key0)
-	lw	$s2,8($key0)
-	lw	$s3,12($key0)
-
 	xor	$t0,$t8
+	lw	$s0,0($key0)
 	xor	$t1,$t9
+	lw	$s1,4($key0)
 	xor	$t2,$t10
+	lw	$s2,8($key0)
 	xor	$t3,$t11
+	lw	$s3,12($key0)
 
 	xor	$t0,$t4
 	xor	$t1,$t5
@@ -567,38 +985,39 @@ _mips_AES_decrypt:
 	.set	noreorder
 	bnez	$cnt,.Loop_dec
 	_xtr	$i0,$s3,16-2
+#endif
 
 	.set	reorder
 	lw	$t4,1024($Tbl)		# prefetch Td4
-	lw	$t5,1024+32($Tbl)
-	lw	$t6,1024+64($Tbl)
-	lw	$t7,1024+96($Tbl)
-	lw	$t8,1024+128($Tbl)
-	lw	$t9,1024+160($Tbl)
-	lw	$t10,1024+192($Tbl)
-	lw	$t11,1024+224($Tbl)
-
 	_xtr	$i0,$s3,16
+	lw	$t5,1024+32($Tbl)
 	_xtr	$i1,$s0,16
+	lw	$t6,1024+64($Tbl)
 	_xtr	$i2,$s1,16
+	lw	$t7,1024+96($Tbl)
 	_xtr	$i3,$s2,16
+	lw	$t8,1024+128($Tbl)
 	and	$i0,0xff
+	lw	$t9,1024+160($Tbl)
 	and	$i1,0xff
+	lw	$t10,1024+192($Tbl)
 	and	$i2,0xff
+	lw	$t11,1024+224($Tbl)
 	and	$i3,0xff
+
 	$PTR_ADD $i0,$Tbl
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
 	lbu	$t0,1024($i0)		# Td4[s3>>16]
-	lbu	$t1,1024($i1)		# Td4[s0>>16]
-	lbu	$t2,1024($i2)		# Td4[s1>>16]
-	lbu	$t3,1024($i3)		# Td4[s2>>16]
-
 	_xtr	$i0,$s2,8
+	lbu	$t1,1024($i1)		# Td4[s0>>16]
 	_xtr	$i1,$s3,8
+	lbu	$t2,1024($i2)		# Td4[s1>>16]
 	_xtr	$i2,$s0,8
+	lbu	$t3,1024($i3)		# Td4[s2>>16]
 	_xtr	$i3,$s1,8
+
 	and	$i0,0xff
 	and	$i1,0xff
 	and	$i2,0xff
@@ -607,29 +1026,108 @@ _mips_AES_decrypt:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+# if defined(_MIPSEL)
 	lbu	$t4,1024($i0)		# Td4[s2>>8]
+	$PTR_INS $i0,$s0,0,8
 	lbu	$t5,1024($i1)		# Td4[s3>>8]
+	$PTR_INS $i1,$s1,0,8
 	lbu	$t6,1024($i2)		# Td4[s0>>8]
+	$PTR_INS $i2,$s2,0,8
 	lbu	$t7,1024($i3)		# Td4[s1>>8]
+	$PTR_INS $i3,$s3,0,8
 
+	lbu	$t8,1024($i0)		# Td4[s0>>24]
+	_xtr	$i0,$s1,0
+	lbu	$t9,1024($i1)		# Td4[s1>>24]
+	_xtr	$i1,$s2,0
+	lbu	$t10,1024($i2)		# Td4[s2>>24]
+	_xtr	$i2,$s3,0
+	lbu	$t11,1024($i3)		# Td4[s3>>24]
+	_xtr	$i3,$s0,0
+
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+# else
+	lbu	$t4,1024($i0)		# Td4[s2>>8]
 	_xtr	$i0,$s0,24
+	lbu	$t5,1024($i1)		# Td4[s3>>8]
 	_xtr	$i1,$s1,24
+	lbu	$t6,1024($i2)		# Td4[s0>>8]
 	_xtr	$i2,$s2,24
+	lbu	$t7,1024($i3)		# Td4[s1>>8]
 	_xtr	$i3,$s3,24
+
 	$PTR_ADD $i0,$Tbl
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
 	lbu	$t8,1024($i0)		# Td4[s0>>24]
+	$PTR_INS $i0,$s1,0,8
 	lbu	$t9,1024($i1)		# Td4[s1>>24]
+	$PTR_INS $i1,$s2,0,8
 	lbu	$t10,1024($i2)		# Td4[s2>>24]
+	$PTR_INS $i2,$s3,0,8
 	lbu	$t11,1024($i3)		# Td4[s3>>24]
+	$PTR_INS $i3,$s0,0,8
+# endif
+	_ins	$t0,16
+	_ins	$t1,16
+	_ins	$t2,16
+	_ins	$t3,16
+
+	_ins2	$t0,$t4,8
+	lbu	$t4,1024($i0)		# Td4[s1]
+	_ins2	$t1,$t5,8
+	lbu	$t5,1024($i1)		# Td4[s2]
+	_ins2	$t2,$t6,8
+	lbu	$t6,1024($i2)		# Td4[s3]
+	_ins2	$t3,$t7,8
+	lbu	$t7,1024($i3)		# Td4[s0]
+
+	_ins2	$t0,$t8,24
+	lw	$s0,0($key0)
+	_ins2	$t1,$t9,24
+	lw	$s1,4($key0)
+	_ins2	$t2,$t10,24
+	lw	$s2,8($key0)
+	_ins2	$t3,$t11,24
+	lw	$s3,12($key0)
 
+	_ins2	$t0,$t4,0
+	_ins2	$t1,$t5,0
+	_ins2	$t2,$t6,0
+	_ins2	$t3,$t7,0
+#else
+	lbu	$t4,1024($i0)		# Td4[s2>>8]
+	_xtr	$i0,$s0,24
+	lbu	$t5,1024($i1)		# Td4[s3>>8]
+	_xtr	$i1,$s1,24
+	lbu	$t6,1024($i2)		# Td4[s0>>8]
+	_xtr	$i2,$s2,24
+	lbu	$t7,1024($i3)		# Td4[s1>>8]
+	_xtr	$i3,$s3,24
+
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t8,1024($i0)		# Td4[s0>>24]
 	_xtr	$i0,$s1,0
+	lbu	$t9,1024($i1)		# Td4[s1>>24]
 	_xtr	$i1,$s2,0
+	lbu	$t10,1024($i2)		# Td4[s2>>24]
 	_xtr	$i2,$s3,0
+	lbu	$t11,1024($i3)		# Td4[s3>>24]
 	_xtr	$i3,$s0,0
 
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+
 	_ins	$t0,16
 	_ins	$t1,16
 	_ins	$t2,16
@@ -641,44 +1139,38 @@ _mips_AES_decrypt:
 	_ins	$t7,8
 
 	xor	$t0,$t4
-	xor	$t1,$t5
-	xor	$t2,$t6
-	xor	$t3,$t7
-
-	$PTR_ADD $i0,$Tbl
-	$PTR_ADD $i1,$Tbl
-	$PTR_ADD $i2,$Tbl
-	$PTR_ADD $i3,$Tbl
 	lbu	$t4,1024($i0)		# Td4[s1]
+	xor	$t1,$t5
 	lbu	$t5,1024($i1)		# Td4[s2]
+	xor	$t2,$t6
 	lbu	$t6,1024($i2)		# Td4[s3]
+	xor	$t3,$t7
 	lbu	$t7,1024($i3)		# Td4[s0]
 
 	_ins	$t8,24
-	_ins	$t9,24
-	_ins	$t10,24
-	_ins	$t11,24
-
 	lw	$s0,0($key0)
+	_ins	$t9,24
 	lw	$s1,4($key0)
+	_ins	$t10,24
 	lw	$s2,8($key0)
+	_ins	$t11,24
 	lw	$s3,12($key0)
 
-	_ins	$t4,0
-	_ins	$t5,0
-	_ins	$t6,0
-	_ins	$t7,0
-
-
 	xor	$t0,$t8
 	xor	$t1,$t9
 	xor	$t2,$t10
 	xor	$t3,$t11
 
+	_ins	$t4,0
+	_ins	$t5,0
+	_ins	$t6,0
+	_ins	$t7,0
+
 	xor	$t0,$t4
 	xor	$t1,$t5
 	xor	$t2,$t6
 	xor	$t3,$t7
+#endif
 
 	xor	$s0,$t0
 	xor	$s1,$t1
@@ -791,7 +1283,7 @@ _mips_AES_set_encrypt_key:
 	beqz	$inp,.Lekey_done
 	li	$t0,-1
 	beqz	$key,.Lekey_done
-	$PTR_ADD $rcon,$Tbl,1024+256
+	$PTR_ADD $rcon,$Tbl,256
 
 	.set	reorder
 	lwl	$rk0,0+$MSB($inp)	# load 128 bits
@@ -843,10 +1335,10 @@ _mips_AES_set_encrypt_key:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
-	lbu	$i0,1024($i0)
-	lbu	$i1,1024($i1)
-	lbu	$i2,1024($i2)
-	lbu	$i3,1024($i3)
+	lbu	$i0,0($i0)
+	lbu	$i1,0($i1)
+	lbu	$i2,0($i2)
+	lbu	$i3,0($i3)
 
 	sw	$rk0,0($key)
 	sw	$rk1,4($key)
@@ -898,10 +1390,10 @@ _mips_AES_set_encrypt_key:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
-	lbu	$i0,1024($i0)
-	lbu	$i1,1024($i1)
-	lbu	$i2,1024($i2)
-	lbu	$i3,1024($i3)
+	lbu	$i0,0($i0)
+	lbu	$i1,0($i1)
+	lbu	$i2,0($i2)
+	lbu	$i3,0($i3)
 
 	sw	$rk0,0($key)
 	sw	$rk1,4($key)
@@ -957,10 +1449,10 @@ _mips_AES_set_encrypt_key:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
-	lbu	$i0,1024($i0)
-	lbu	$i1,1024($i1)
-	lbu	$i2,1024($i2)
-	lbu	$i3,1024($i3)
+	lbu	$i0,0($i0)
+	lbu	$i1,0($i1)
+	lbu	$i2,0($i2)
+	lbu	$i3,0($i3)
 
 	sw	$rk0,0($key)
 	sw	$rk1,4($key)
@@ -999,10 +1491,10 @@ _mips_AES_set_encrypt_key:
 	$PTR_ADD $i1,$Tbl
 	$PTR_ADD $i2,$Tbl
 	$PTR_ADD $i3,$Tbl
-	lbu	$i0,1024($i0)
-	lbu	$i1,1024($i1)
-	lbu	$i2,1024($i2)
-	lbu	$i3,1024($i3)
+	lbu	$i0,0($i0)
+	lbu	$i1,0($i1)
+	lbu	$i2,0($i2)
+	lbu	$i3,0($i3)
 	sll	$i0,24
 	sll	$i1,16
 	sll	$i2,8
@@ -1064,7 +1556,7 @@ $code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
 ___
 $code.=<<___;
 	.set	reorder
-	la	$Tbl,AES_Te		# PIC-ified 'load address'
+	la	$Tbl,AES_Te4		# PIC-ified 'load address'
 
 	bal	_mips_AES_set_encrypt_key
 
@@ -1119,7 +1611,7 @@ $code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
 ___
 $code.=<<___;
 	.set	reorder
-	la	$Tbl,AES_Te		# PIC-ified 'load address'
+	la	$Tbl,AES_Te4		# PIC-ified 'load address'
 
 	bal	_mips_AES_set_encrypt_key
 
@@ -1190,6 +1682,16 @@ $code.=<<___;
 	xor	$tpb,$tp9,$tp2
 	xor	$tpd,$tp9,$tp4
 
+#if defined(_MIPS_ARCH_MIPS32R2) || defined(_MIPS_ARCH_MIPS64R2)
+	rotr	$tp1,$tpd,16
+	 xor	$tpe,$tp2
+	rotr	$tp2,$tp9,8
+	xor	$tpe,$tp1
+	rotr	$tp4,$tpb,24
+	xor	$tpe,$tp2
+	lw	$tp1,4($key)		# modulo-scheduled
+	xor	$tpe,$tp4
+#else
 	_ror	$tp1,$tpd,16
 	 xor	$tpe,$tp2
 	_ror	$tp2,$tpd,-16
@@ -1204,6 +1706,7 @@ $code.=<<___;
 	xor	$tpe,$tp1
 	lw	$tp1,4($key)		# modulo-scheduled
 	xor	$tpe,$tp2
+#endif
 	sub	$cnt,1
 	sw	$tpe,0($key)
 	$PTR_ADD $key,4
@@ -1234,7 +1737,7 @@ ___
 # Tables are kept in endian-neutral manner
 $code.=<<___;
 .rdata
-.align	6
+.align	10
 AES_Te:
 .byte	0xc6,0x63,0x63,0xa5,	0xf8,0x7c,0x7c,0x84	# Te0
 .byte	0xee,0x77,0x77,0x99,	0xf6,0x7b,0x7b,0x8d
@@ -1365,46 +1868,6 @@ AES_Te:
 .byte	0x7b,0xb0,0xb0,0xcb,	0xa8,0x54,0x54,0xfc
 .byte	0x6d,0xbb,0xbb,0xd6,	0x2c,0x16,0x16,0x3a
 
-.byte	0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5	# Te4
-.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
-
-.byte	0x01,0x00,0x00,0x00,	0x02,0x00,0x00,0x00	# rcon
-.byte	0x04,0x00,0x00,0x00,	0x08,0x00,0x00,0x00
-.byte	0x10,0x00,0x00,0x00,	0x20,0x00,0x00,0x00
-.byte	0x40,0x00,0x00,0x00,	0x80,0x00,0x00,0x00
-.byte	0x1B,0x00,0x00,0x00,	0x36,0x00,0x00,0x00
-
-.align	6
 AES_Td:
 .byte	0x51,0xf4,0xa7,0x50,	0x7e,0x41,0x65,0x53	# Td0
 .byte	0x1a,0x17,0xa4,0xc3,	0x3a,0x27,0x5e,0x96
@@ -1567,6 +2030,46 @@ AES_Td:
 .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
+
+AES_Te4:
+.byte	0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5	# Te4
+.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
+
+.byte	0x01,0x00,0x00,0x00,	0x02,0x00,0x00,0x00	# rcon
+.byte	0x04,0x00,0x00,0x00,	0x08,0x00,0x00,0x00
+.byte	0x10,0x00,0x00,0x00,	0x20,0x00,0x00,0x00
+.byte	0x40,0x00,0x00,0x00,	0x80,0x00,0x00,0x00
+.byte	0x1B,0x00,0x00,0x00,	0x36,0x00,0x00,0x00
 ___
 
 foreach (split("\n",$code)) {
@@ -1583,6 +2086,9 @@ foreach (split("\n",$code)) {
 	    s/_ins\s+(\$[0-9]+),(\$[0-9]+),([0-9]+)/
 		sprintf("sll\t$1,$2,%d",$big_endian ?	eval($3)
 					:		eval("24-$3"))/e or
+	    s/_ins2\s+(\$[0-9]+),(\$[0-9]+),([0-9]+)/
+		sprintf("ins\t$1,$2,%d,8",$big_endian ?	eval($3)
+					:		eval("24-$3"))/e or
 	    s/_ror\s+(\$[0-9]+),(\$[0-9]+),(\-?[0-9]+)/
 		sprintf("srl\t$1,$2,%d",$big_endian ?	eval($3)
 					:		eval("$3*-1"))/e or
@@ -1605,6 +2111,11 @@ foreach (split("\n",$code)) {
 		sprintf("$1%d($3)",eval("$2-$2%4+($2%4+1)&3"))/e;
 	}
 
+	if (!$big_endian) {
+	    s/(rotr\s+\$[0-9]+,\$[0-9]+),([0-9]+)/sprintf("$1,%d",32-$2)/e;
+	    s/(ext\s+\$[0-9]+,\$[0-9]+),([0-9]+),8/sprintf("$1,%d,8",24-$2)/e;
+	}
+
 	print $_,"\n";
 }
 
diff --git a/openssl/crypto/aes/asm/aes-ppc.pl b/openssl/crypto/aes/asm/aes-ppc.pl
index 7c52cbe5f..7a99fc3d0 100644
--- a/openssl/crypto/aes/asm/aes-ppc.pl
+++ b/openssl/crypto/aes/asm/aes-ppc.pl
@@ -45,6 +45,8 @@ if ($flavour =~ /64/) {
 	$PUSH	="stw";
 } else { die "nonsense $flavour"; }
 
+$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
+
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@@ -68,7 +70,7 @@ $key="r5";
 $Tbl0="r3";
 $Tbl1="r6";
 $Tbl2="r7";
-$Tbl3="r2";
+$Tbl3=$out;	# stay away from "r2"; $out is offloaded to stack
 
 $s0="r8";
 $s1="r9";
@@ -76,7 +78,7 @@ $s2="r10";
 $s3="r11";
 
 $t0="r12";
-$t1="r13";
+$t1="r0";	# stay away from "r13";
 $t2="r14";
 $t3="r15";
 
@@ -100,9 +102,6 @@ $acc13="r29";
 $acc14="r30";
 $acc15="r31";
 
-# stay away from TLS pointer
-if ($SIZE_T==8)	{ die if ($t1 ne "r13");  $t1="r0";		}
-else		{ die if ($Tbl3 ne "r2"); $Tbl3=$t0; $t0="r0";	}
 $mask80=$Tbl2;
 $mask1b=$Tbl3;
 
@@ -337,8 +336,7 @@ $code.=<<___;
 	$STU	$sp,-$FRAME($sp)
 	mflr	r0
 
-	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
-	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
+	$PUSH	$out,`$FRAME-$SIZE_T*19`($sp)
 	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
 	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
 	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
@@ -365,16 +363,61 @@ $code.=<<___;
 	bne	Lenc_unaligned
 
 Lenc_unaligned_ok:
+___
+$code.=<<___ if (!$LITTLE_ENDIAN);
 	lwz	$s0,0($inp)
 	lwz	$s1,4($inp)
 	lwz	$s2,8($inp)
 	lwz	$s3,12($inp)
+___
+$code.=<<___ if ($LITTLE_ENDIAN);
+	lwz	$t0,0($inp)
+	lwz	$t1,4($inp)
+	lwz	$t2,8($inp)
+	lwz	$t3,12($inp)
+	rotlwi	$s0,$t0,8
+	rotlwi	$s1,$t1,8
+	rotlwi	$s2,$t2,8
+	rotlwi	$s3,$t3,8
+	rlwimi	$s0,$t0,24,0,7
+	rlwimi	$s1,$t1,24,0,7
+	rlwimi	$s2,$t2,24,0,7
+	rlwimi	$s3,$t3,24,0,7
+	rlwimi	$s0,$t0,24,16,23
+	rlwimi	$s1,$t1,24,16,23
+	rlwimi	$s2,$t2,24,16,23
+	rlwimi	$s3,$t3,24,16,23
+___
+$code.=<<___;
 	bl	LAES_Te
 	bl	Lppc_AES_encrypt_compact
+	$POP	$out,`$FRAME-$SIZE_T*19`($sp)
+___
+$code.=<<___ if ($LITTLE_ENDIAN);
+	rotlwi	$t0,$s0,8
+	rotlwi	$t1,$s1,8
+	rotlwi	$t2,$s2,8
+	rotlwi	$t3,$s3,8
+	rlwimi	$t0,$s0,24,0,7
+	rlwimi	$t1,$s1,24,0,7
+	rlwimi	$t2,$s2,24,0,7
+	rlwimi	$t3,$s3,24,0,7
+	rlwimi	$t0,$s0,24,16,23
+	rlwimi	$t1,$s1,24,16,23
+	rlwimi	$t2,$s2,24,16,23
+	rlwimi	$t3,$s3,24,16,23
+	stw	$t0,0($out)
+	stw	$t1,4($out)
+	stw	$t2,8($out)
+	stw	$t3,12($out)
+___
+$code.=<<___ if (!$LITTLE_ENDIAN);
 	stw	$s0,0($out)
 	stw	$s1,4($out)
 	stw	$s2,8($out)
 	stw	$s3,12($out)
+___
+$code.=<<___;
 	b	Lenc_done
 
 Lenc_unaligned:
@@ -417,6 +460,7 @@ Lenc_xpage:
 
 	bl	LAES_Te
 	bl	Lppc_AES_encrypt_compact
+	$POP	$out,`$FRAME-$SIZE_T*19`($sp)
 
 	extrwi	$acc00,$s0,8,0
 	extrwi	$acc01,$s0,8,8
@@ -449,8 +493,6 @@ Lenc_xpage:
 
 Lenc_done:
 	$POP	r0,`$FRAME+$LRSAVE`($sp)
-	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
-	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
 	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
 	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
 	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
@@ -764,6 +806,7 @@ Lenc_compact_done:
 	blr
 	.long	0
 	.byte	0,12,0x14,0,0,0,0,0
+.size	.AES_encrypt,.-.AES_encrypt
 
 .globl	.AES_decrypt
 .align	7
@@ -771,8 +814,7 @@ Lenc_compact_done:
 	$STU	$sp,-$FRAME($sp)
 	mflr	r0
 
-	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
-	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
+	$PUSH	$out,`$FRAME-$SIZE_T*19`($sp)
 	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
 	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
 	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
@@ -799,16 +841,61 @@ Lenc_compact_done:
 	bne	Ldec_unaligned
 
 Ldec_unaligned_ok:
+___
+$code.=<<___ if (!$LITTLE_ENDIAN);
 	lwz	$s0,0($inp)
 	lwz	$s1,4($inp)
 	lwz	$s2,8($inp)
 	lwz	$s3,12($inp)
+___
+$code.=<<___ if ($LITTLE_ENDIAN);
+	lwz	$t0,0($inp)
+	lwz	$t1,4($inp)
+	lwz	$t2,8($inp)
+	lwz	$t3,12($inp)
+	rotlwi	$s0,$t0,8
+	rotlwi	$s1,$t1,8
+	rotlwi	$s2,$t2,8
+	rotlwi	$s3,$t3,8
+	rlwimi	$s0,$t0,24,0,7
+	rlwimi	$s1,$t1,24,0,7
+	rlwimi	$s2,$t2,24,0,7
+	rlwimi	$s3,$t3,24,0,7
+	rlwimi	$s0,$t0,24,16,23
+	rlwimi	$s1,$t1,24,16,23
+	rlwimi	$s2,$t2,24,16,23
+	rlwimi	$s3,$t3,24,16,23
+___
+$code.=<<___;
 	bl	LAES_Td
 	bl	Lppc_AES_decrypt_compact
+	$POP	$out,`$FRAME-$SIZE_T*19`($sp)
+___
+$code.=<<___ if ($LITTLE_ENDIAN);
+	rotlwi	$t0,$s0,8
+	rotlwi	$t1,$s1,8
+	rotlwi	$t2,$s2,8
+	rotlwi	$t3,$s3,8
+	rlwimi	$t0,$s0,24,0,7
+	rlwimi	$t1,$s1,24,0,7
+	rlwimi	$t2,$s2,24,0,7
+	rlwimi	$t3,$s3,24,0,7
+	rlwimi	$t0,$s0,24,16,23
+	rlwimi	$t1,$s1,24,16,23
+	rlwimi	$t2,$s2,24,16,23
+	rlwimi	$t3,$s3,24,16,23
+	stw	$t0,0($out)
+	stw	$t1,4($out)
+	stw	$t2,8($out)
+	stw	$t3,12($out)
+___
+$code.=<<___ if (!$LITTLE_ENDIAN);
 	stw	$s0,0($out)
 	stw	$s1,4($out)
 	stw	$s2,8($out)
 	stw	$s3,12($out)
+___
+$code.=<<___;
 	b	Ldec_done
 
 Ldec_unaligned:
@@ -851,6 +938,7 @@ Ldec_xpage:
 
 	bl	LAES_Td
 	bl	Lppc_AES_decrypt_compact
+	$POP	$out,`$FRAME-$SIZE_T*19`($sp)
 
 	extrwi	$acc00,$s0,8,0
 	extrwi	$acc01,$s0,8,8
@@ -883,8 +971,6 @@ Ldec_xpage:
 
 Ldec_done:
 	$POP	r0,`$FRAME+$LRSAVE`($sp)
-	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
-	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
 	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
 	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
 	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
@@ -1355,6 +1441,7 @@ Ldec_compact_done:
 	blr
 	.long	0
 	.byte	0,12,0x14,0,0,0,0,0
+.size	.AES_decrypt,.-.AES_decrypt
 
 .asciz	"AES for PPC, CRYPTOGAMS by <appro\@openssl.org>"
 .align	7
diff --git a/openssl/crypto/aes/asm/aes-x86_64.pl b/openssl/crypto/aes/asm/aes-x86_64.pl
index 34cbb5d84..47f416375 100644
--- a/openssl/crypto/aes/asm/aes-x86_64.pl
+++ b/openssl/crypto/aes/asm/aes-x86_64.pl
@@ -19,9 +19,10 @@
 # Performance in number of cycles per processed byte for 128-bit key:
 #
 #		ECB encrypt	ECB decrypt	CBC large chunk
-# AMD64		33		41		13.0
-# EM64T		38		59		18.6(*)
-# Core 2	30		43		14.5(*)
+# 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
 
@@ -366,68 +367,66 @@ $code.=<<___;
 	movzb	`&lo("$s0")`,$t0
 	movzb	`&lo("$s1")`,$t1
 	movzb	`&lo("$s2")`,$t2
-	movzb	($sbox,$t0,1),$t0
-	movzb	($sbox,$t1,1),$t1
-	movzb	($sbox,$t2,1),$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	#$t0
-	movzb	($sbox,$acc1,1),$t5	#$t1
 
-	movzb	`&hi("$s3")`,$acc2
+	movzb	($sbox,$acc0,1),$t4	#$t0
 	movzb	`&hi("$s0")`,$acc0
-	shr	\$16,$s2
+	movzb	($sbox,$acc1,1),$t5	#$t1
+	movzb	`&lo("$s2")`,$acc1
 	movzb	($sbox,$acc2,1),$acc2	#$t2
 	movzb	($sbox,$acc0,1),$acc0	#$t3
-	shr	\$16,$s3
 
-	movzb	`&lo("$s2")`,$acc1
 	shl	\$8,$t4
+	shr	\$16,$s3
 	shl	\$8,$t5
-	movzb	($sbox,$acc1,1),$acc1	#$t0
 	xor	$t4,$t0
-	xor	$t5,$t1
-
-	movzb	`&lo("$s3")`,$t4
 	shr	\$16,$s0
+	movzb	`&lo("$s3")`,$t4
 	shr	\$16,$s1
-	movzb	`&lo("$s0")`,$t5
+	xor	$t5,$t1
 	shl	\$8,$acc2
-	shl	\$8,$acc0
-	movzb	($sbox,$t4,1),$t4	#$t1
-	movzb	($sbox,$t5,1),$t5	#$t2
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
 	xor	$acc2,$t2
-	xor	$acc0,$t3
 
+	shl	\$8,$acc0
 	movzb	`&lo("$s1")`,$acc2
-	movzb	`&hi("$s3")`,$acc0
 	shl	\$16,$acc1
-	movzb	($sbox,$acc2,1),$acc2	#$t3
-	movzb	($sbox,$acc0,1),$acc0	#$t0
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s3")`,$acc0
+	movzb	($sbox,$t5,1),$t5	#$t2
 	xor	$acc1,$t0
 
-	movzb	`&hi("$s0")`,$acc1
 	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,$t4
-	shl	\$16,$t5
+
 	shl	\$16,$acc2
-	xor	$t4,$t1
 	xor	$t5,$t2
-	xor	$acc2,$t3
-
 	shl	\$24,$acc0
+	xor	$acc2,$t3
 	shl	\$24,$acc1
-	shl	\$24,$s3
 	xor	$acc0,$t0
-	shl	\$24,$s2
+	shl	\$24,$s3
 	xor	$acc1,$t1
+	shl	\$24,$s2
 	mov	$t0,$s0
 	mov	$t1,$s1
 	xor	$t2,$s2
@@ -466,12 +465,12 @@ sub enctransform()
 { my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
 
 $code.=<<___;
-	mov	$s0,$acc0
-	mov	$s1,$acc1
-	and	\$0x80808080,$acc0
-	and	\$0x80808080,$acc1
-	mov	$acc0,$t0
-	mov	$acc1,$t1
+	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
@@ -489,25 +488,25 @@ $code.=<<___;
 
 	xor	$r20,$s0
 	xor	$r21,$s1
-	 mov	$s2,$acc0
-	 mov	$s3,$acc1
+	 mov	\$0x80808080,$t2
 	rol	\$24,$s0
+	 mov	\$0x80808080,$t3
 	rol	\$24,$s1
-	 and	\$0x80808080,$acc0
-	 and	\$0x80808080,$acc1
+	 and	$s2,$t2
+	 and	$s3,$t3
 	xor	$r20,$s0
 	xor	$r21,$s1
-	 mov	$acc0,$t2
-	 mov	$acc1,$t3
+	 mov	$t2,$acc0
 	ror	\$16,$t0
+	 mov	$t3,$acc1
 	ror	\$16,$t1
-	 shr	\$7,$t2
 	 lea	($s2,$s2),$r20
+	 shr	\$7,$t2
 	xor	$t0,$s0
-	xor	$t1,$s1
 	 shr	\$7,$t3
-	 lea	($s3,$s3),$r21
+	xor	$t1,$s1
 	ror	\$8,$t0
+	 lea	($s3,$s3),$r21
 	ror	\$8,$t1
 	 sub	$t2,$acc0
 	 sub	$t3,$acc1
@@ -523,23 +522,23 @@ $code.=<<___;
 	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
-	xor	$r21,$s3
-	mov	0($sbox),$acc0			# prefetch Te4
-	ror	\$16,$t2
-	ror	\$16,$t3
 	mov	64($sbox),$acc1
-	xor	$t2,$s2
-	xor	$t3,$s3
+	xor	$r21,$s3
 	mov	128($sbox),$r20
+	xor	$t2,$s2
 	ror	\$8,$t2
+	xor	$t3,$s3
 	ror	\$8,$t3
-	mov	192($sbox),$r21
 	xor	$t2,$s2
+	mov	192($sbox),$r21
 	xor	$t3,$s3
 ___
 }
@@ -936,70 +935,69 @@ $code.=<<___;
 	movzb	`&lo("$s0")`,$t0
 	movzb	`&lo("$s1")`,$t1
 	movzb	`&lo("$s2")`,$t2
-	movzb	($sbox,$t0,1),$t0
-	movzb	($sbox,$t1,1),$t1
-	movzb	($sbox,$t2,1),$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	#$t0
-	movzb	($sbox,$acc1,1),$t5	#$t1
 
-	movzb	`&hi("$s1")`,$acc2
+	movzb	($sbox,$acc0,1),$t4	#$t0
 	movzb	`&hi("$s2")`,$acc0
-	shr	\$16,$s2
+	movzb	($sbox,$acc1,1),$t5	#$t1
 	movzb	($sbox,$acc2,1),$acc2	#$t2
 	movzb	($sbox,$acc0,1),$acc0	#$t3
-	shr	\$16,$s3
 
-	movzb	`&lo("$s2")`,$acc1
-	shl	\$8,$t4
+	shr	\$16,$s2
 	shl	\$8,$t5
-	movzb	($sbox,$acc1,1),$acc1	#$t0
-	xor	$t4,$t0
-	xor	$t5,$t1
-
-	movzb	`&lo("$s3")`,$t4
+	shl	\$8,$t4
+	movzb	`&lo("$s2")`,$acc1
 	shr	\$16,$s0
+	xor	$t4,$t0
 	shr	\$16,$s1
-	movzb	`&lo("$s0")`,$t5
+	movzb	`&lo("$s3")`,$t4
+
 	shl	\$8,$acc2
+	xor	$t5,$t1
 	shl	\$8,$acc0
-	movzb	($sbox,$t4,1),$t4	#$t1
-	movzb	($sbox,$t5,1),$t5	#$t2
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
 	xor	$acc2,$t2
-	xor	$acc0,$t3
-
 	movzb	`&lo("$s1")`,$acc2
-	movzb	`&hi("$s1")`,$acc0
+
 	shl	\$16,$acc1
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s1")`,$acc0
 	movzb	($sbox,$acc2,1),$acc2	#$t3
-	movzb	($sbox,$acc0,1),$acc0	#$t0
 	xor	$acc1,$t0
-
+	movzb	($sbox,$t5,1),$t5	#$t2
 	movzb	`&hi("$s2")`,$acc1
+
+	shl	\$16,$acc2
 	shl	\$16,$t4
 	shl	\$16,$t5
-	movzb	($sbox,$acc1,1),$s1	#$t1
+	xor	$acc2,$t3
+	movzb	`&hi("$s3")`,$acc2
 	xor	$t4,$t1
+	shr	\$8,$s0
 	xor	$t5,$t2
 
-	movzb	`&hi("$s3")`,$acc1
-	shr	\$8,$s0
-	shl	\$16,$acc2
-	movzb	($sbox,$acc1,1),$s2	#$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
-	xor	$acc2,$t3
 
+	mov	$t0,$s0
 	shl	\$24,$acc0
 	shl	\$24,$s1
 	shl	\$24,$s2
-	xor	$acc0,$t0
+	xor	$acc0,$s0
 	shl	\$24,$s3
 	xor	$t1,$s1
-	mov	$t0,$s0
 	xor	$t2,$s2
 	xor	$t3,$s3
 ___
@@ -1014,12 +1012,12 @@ sub dectransform()
   my $prefetch = shift;
 
 $code.=<<___;
-	mov	$tp10,$acc0
-	mov	$tp18,$acc8
-	and	$mask80,$acc0
-	and	$mask80,$acc8
-	mov	$acc0,$tp40
-	mov	$acc8,$tp48
+	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
@@ -1030,15 +1028,15 @@ $code.=<<___;
 	and	$maskfe,$tp28
 	and	$mask1b,$acc0
 	and	$mask1b,$acc8
-	xor	$tp20,$acc0
-	xor	$tp28,$acc8
-	mov	$acc0,$tp20
-	mov	$acc8,$tp28
-
-	and	$mask80,$acc0
-	and	$mask80,$acc8
-	mov	$acc0,$tp80
-	mov	$acc8,$tp88
+	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
@@ -1049,15 +1047,15 @@ $code.=<<___;
 	and	$maskfe,$tp48
 	and	$mask1b,$acc0
 	and	$mask1b,$acc8
-	xor	$tp40,$acc0
-	xor	$tp48,$acc8
-	mov	$acc0,$tp40
-	mov	$acc8,$tp48
-
-	and	$mask80,$acc0
-	and	$mask80,$acc8
-	mov	$acc0,$tp80
-	mov	$acc8,$tp88
+	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
@@ -1082,51 +1080,51 @@ $code.=<<___;
 	mov	$tp10,$acc0
 	mov	$tp18,$acc8
 	xor	$tp80,$tp40		# tp4^tp1^=tp8
-	xor	$tp88,$tp48		# tp4^tp1^=tp8
 	shr	\$32,$acc0
+	xor	$tp88,$tp48		# tp4^tp1^=tp8
 	shr	\$32,$acc8
 	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
-	xor	$tp28,$tp88		# 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("$acc0")`	# ROTATE(tp1^tp8,8)
 	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
 	xor	`&LO("$tp80")`,`&LO("$tp10")`
-	xor	`&LO("$tp88")`,`&LO("$tp18")`
 	shr	\$32,$tp80
+	xor	`&LO("$tp88")`,`&LO("$tp18")`
 	shr	\$32,$tp88
 	xor	`&LO("$tp80")`,`&LO("$acc0")`
 	xor	`&LO("$tp88")`,`&LO("$acc8")`
 
 	mov	$tp20,$tp80
-	mov	$tp28,$tp88
-	shr	\$32,$tp80
-	shr	\$32,$tp88
 	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp28,$tp88
 	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
-	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
-	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
+	shr	\$32,$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 ($prefetch)`
-	shr	\$32,$tp20
-	shr	\$32,$tp28
-	`"mov	64($sbox),$maskfe"	if ($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("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
-	rol	\$16,`&LO("$tp28")`	# 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")`
@@ -1302,10 +1300,6 @@ private_AES_set_encrypt_key:
 
 	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
diff --git a/openssl/crypto/aes/asm/aesni-mb-x86_64.pl b/openssl/crypto/aes/asm/aesni-mb-x86_64.pl
new file mode 100755
index 000000000..33b1aed3c
--- /dev/null
+++ b/openssl/crypto/aes/asm/aesni-mb-x86_64.pl
@@ -0,0 +1,1395 @@
+#!/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/.
+# ====================================================================
+
+# Multi-buffer AES-NI procedures process several independent buffers
+# in parallel by interleaving independent instructions.
+#
+# Cycles per byte for interleave factor 4:
+#
+#			asymptotic	measured
+#			---------------------------
+# Westmere		5.00/4=1.25	5.13/4=1.28
+# Atom			15.0/4=3.75	?15.7/4=3.93
+# Sandy Bridge		5.06/4=1.27	5.18/4=1.29
+# Ivy Bridge		5.06/4=1.27	5.14/4=1.29
+# Haswell		4.44/4=1.11	4.44/4=1.11
+# Bulldozer		5.75/4=1.44	5.76/4=1.44
+#
+# Cycles per byte for interleave factor 8 (not implemented for
+# pre-AVX processors, where higher interleave factor incidentally
+# doesn't result in improvement):
+#
+#			asymptotic	measured
+#			---------------------------
+# Sandy Bridge		5.06/8=0.64	7.10/8=0.89(*)
+# Ivy Bridge		5.06/8=0.64	7.14/8=0.89(*)
+# Haswell		5.00/8=0.63	5.00/8=0.63
+# Bulldozer		5.75/8=0.72	5.77/8=0.72
+#
+# (*)	Sandy/Ivy Bridge are known to handle high interleave factors
+#	suboptimally;
+
+$flavour = shift;
+$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 $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+$avx=0;
+
+if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.19) + ($1>=2.22);
+}
+
+if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
+	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.09) + ($1>=2.10);
+}
+
+if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
+	   `ml64 2>&1` =~ /Version ([0-9]+)\./) {
+	$avx = ($1>=10) + ($1>=11);
+}
+
+if (!$avx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/) {
+	$avx = ($2>=3.0) + ($2>3.0);
+}
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+# void aesni_multi_cbc_encrypt (
+#     struct {	void *inp,*out; int blocks; double iv[2]; } inp[8];
+#     const AES_KEY *key,
+#     int num);		/* 1 or 2 */
+#
+$inp="%rdi";	# 1st arg
+$key="%rsi";	# 2nd arg
+$num="%edx";
+
+@inptr=map("%r$_",(8..11));
+@outptr=map("%r$_",(12..15));
+
+($rndkey0,$rndkey1)=("%xmm0","%xmm1");
+@out=map("%xmm$_",(2..5));
+@inp=map("%xmm$_",(6..9));
+($counters,$mask,$zero)=map("%xmm$_",(10..12));
+
+($rounds,$one,$sink,$offset)=("%eax","%ecx","%rbp","%rbx");
+
+$code.=<<___;
+.text
+
+.extern	OPENSSL_ia32cap_P
+
+.globl	aesni_multi_cbc_encrypt
+.type	aesni_multi_cbc_encrypt,\@function,3
+.align	32
+aesni_multi_cbc_encrypt:
+___
+$code.=<<___ if ($avx);
+	cmp	\$2,$num
+	jb	.Lenc_non_avx
+	mov	OPENSSL_ia32cap_P+4(%rip),%ecx
+	test	\$`1<<28`,%ecx			# AVX bit
+	jnz	_avx_cbc_enc_shortcut
+	jmp	.Lenc_non_avx
+.align	16
+.Lenc_non_avx:
+___
+$code.=<<___;
+	mov	%rsp,%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+___
+$code.=<<___ if ($win64);
+	lea	-0xa8(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+	movaps	%xmm10,0x40(%rsp)
+	movaps	%xmm11,0x50(%rsp)
+	movaps	%xmm12,0x60(%rsp)
+	movaps	%xmm13,-0x68(%rax)	# not used, saved to share se_handler 
+	movaps	%xmm14,-0x58(%rax)
+	movaps	%xmm15,-0x48(%rax)
+___
+$code.=<<___;
+	# stack layout
+	#
+	# +0	output sink
+	# +16	input sink [original %rsp and $num]
+	# +32	counters
+
+	sub	\$48,%rsp
+	and	\$-64,%rsp
+	mov	%rax,16(%rsp)			# original %rsp
+
+.Lenc4x_body:
+	movdqu	($key),$zero			# 0-round key
+	lea	0x78($key),$key			# size optimization
+	lea	40*2($inp),$inp
+
+.Lenc4x_loop_grande:
+	mov	$num,24(%rsp)			# original $num
+	xor	$num,$num
+___
+for($i=0;$i<4;$i++) {
+    $code.=<<___;
+	mov	`40*$i+16-40*2`($inp),$one	# borrow $one for number of blocks
+	mov	`40*$i+0-40*2`($inp),@inptr[$i]
+	cmp	$num,$one
+	mov	`40*$i+8-40*2`($inp),@outptr[$i]
+	cmovg	$one,$num			# find maximum
+	test	$one,$one
+	movdqu	`40*$i+24-40*2`($inp),@out[$i]	# load IV
+	mov	$one,`32+4*$i`(%rsp)		# initialize counters
+	cmovle	%rsp,@inptr[$i]			# cancel input
+___
+}
+$code.=<<___;
+	test	$num,$num
+	jz	.Lenc4x_done
+
+	movups	0x10-0x78($key),$rndkey1
+	 pxor	$zero,@out[0]
+	movups	0x20-0x78($key),$rndkey0
+	 pxor	$zero,@out[1]
+	mov	0xf0-0x78($key),$rounds
+	 pxor	$zero,@out[2]
+	movdqu	(@inptr[0]),@inp[0]		# load inputs
+	 pxor	$zero,@out[3]
+	movdqu	(@inptr[1]),@inp[1]
+	 pxor	@inp[0],@out[0]
+	movdqu	(@inptr[2]),@inp[2]
+	 pxor	@inp[1],@out[1]
+	movdqu	(@inptr[3]),@inp[3]
+	 pxor	@inp[2],@out[2]
+	 pxor	@inp[3],@out[3]
+	movdqa	32(%rsp),$counters		# load counters
+	xor	$offset,$offset
+	jmp	.Loop_enc4x
+
+.align	32
+.Loop_enc4x:
+	add	\$16,$offset
+	lea	16(%rsp),$sink			# sink pointer
+	mov	\$1,$one			# constant of 1
+	sub	$offset,$sink
+
+	aesenc		$rndkey1,@out[0]
+	prefetcht0	31(@inptr[0],$offset)	# prefetch input
+	prefetcht0	31(@inptr[1],$offset)
+	aesenc		$rndkey1,@out[1]
+	prefetcht0	31(@inptr[2],$offset)
+	prefetcht0	31(@inptr[2],$offset)
+	aesenc		$rndkey1,@out[2]
+	aesenc		$rndkey1,@out[3]
+	movups		0x30-0x78($key),$rndkey1
+___
+for($i=0;$i<4;$i++) {
+my $rndkey = ($i&1) ? $rndkey1 : $rndkey0;
+$code.=<<___;
+	 cmp		`32+4*$i`(%rsp),$one
+	aesenc		$rndkey,@out[0]
+	aesenc		$rndkey,@out[1]
+	aesenc		$rndkey,@out[2]
+	 cmovge		$sink,@inptr[$i]	# cancel input
+	 cmovg		$sink,@outptr[$i]	# sink output
+	aesenc		$rndkey,@out[3]
+	movups		`0x40+16*$i-0x78`($key),$rndkey
+___
+}
+$code.=<<___;
+	 movdqa		$counters,$mask
+	aesenc		$rndkey0,@out[0]
+	prefetcht0	15(@outptr[0],$offset)	# prefetch output
+	prefetcht0	15(@outptr[1],$offset)
+	aesenc		$rndkey0,@out[1]
+	prefetcht0	15(@outptr[2],$offset)
+	prefetcht0	15(@outptr[3],$offset)
+	aesenc		$rndkey0,@out[2]
+	aesenc		$rndkey0,@out[3]
+	movups		0x80-0x78($key),$rndkey0
+	 pxor		$zero,$zero
+
+	aesenc		$rndkey1,@out[0]
+	 pcmpgtd	$zero,$mask
+	 movdqu		-0x78($key),$zero	# reload 0-round key
+	aesenc		$rndkey1,@out[1]
+	 paddd		$mask,$counters		# decrement counters
+	 movdqa		$counters,32(%rsp)	# update counters
+	aesenc		$rndkey1,@out[2]
+	aesenc		$rndkey1,@out[3]
+	movups		0x90-0x78($key),$rndkey1
+
+	cmp	\$11,$rounds
+
+	aesenc		$rndkey0,@out[0]
+	aesenc		$rndkey0,@out[1]
+	aesenc		$rndkey0,@out[2]
+	aesenc		$rndkey0,@out[3]
+	movups		0xa0-0x78($key),$rndkey0
+
+	jb	.Lenc4x_tail
+
+	aesenc		$rndkey1,@out[0]
+	aesenc		$rndkey1,@out[1]
+	aesenc		$rndkey1,@out[2]
+	aesenc		$rndkey1,@out[3]
+	movups		0xb0-0x78($key),$rndkey1
+
+	aesenc		$rndkey0,@out[0]
+	aesenc		$rndkey0,@out[1]
+	aesenc		$rndkey0,@out[2]
+	aesenc		$rndkey0,@out[3]
+	movups		0xc0-0x78($key),$rndkey0
+
+	je	.Lenc4x_tail
+
+	aesenc		$rndkey1,@out[0]
+	aesenc		$rndkey1,@out[1]
+	aesenc		$rndkey1,@out[2]
+	aesenc		$rndkey1,@out[3]
+	movups		0xd0-0x78($key),$rndkey1
+
+	aesenc		$rndkey0,@out[0]
+	aesenc		$rndkey0,@out[1]
+	aesenc		$rndkey0,@out[2]
+	aesenc		$rndkey0,@out[3]
+	movups		0xe0-0x78($key),$rndkey0
+	jmp	.Lenc4x_tail
+
+.align	32
+.Lenc4x_tail:
+	aesenc		$rndkey1,@out[0]
+	aesenc		$rndkey1,@out[1]
+	aesenc		$rndkey1,@out[2]
+	aesenc		$rndkey1,@out[3]
+	 movdqu		(@inptr[0],$offset),@inp[0]
+	movdqu		0x10-0x78($key),$rndkey1
+
+	aesenclast	$rndkey0,@out[0]
+	 movdqu		(@inptr[1],$offset),@inp[1]
+	 pxor		$zero,@inp[0]
+	aesenclast	$rndkey0,@out[1]
+	 movdqu		(@inptr[2],$offset),@inp[2]
+	 pxor		$zero,@inp[1]
+	aesenclast	$rndkey0,@out[2]
+	 movdqu		(@inptr[3],$offset),@inp[3]
+	 pxor		$zero,@inp[2]
+	aesenclast	$rndkey0,@out[3]
+	movdqu		0x20-0x78($key),$rndkey0
+	 pxor		$zero,@inp[3]
+
+	movups		@out[0],-16(@outptr[0],$offset)
+	 pxor		@inp[0],@out[0]
+	movups		@out[1],-16(@outptr[1],$offset)	
+	 pxor		@inp[1],@out[1]
+	movups		@out[2],-16(@outptr[2],$offset)	
+	 pxor		@inp[2],@out[2]
+	movups		@out[3],-16(@outptr[3],$offset)
+	 pxor		@inp[3],@out[3]
+
+	dec	$num
+	jnz	.Loop_enc4x
+
+	mov	16(%rsp),%rax			# original %rsp
+	mov	24(%rsp),$num
+
+	#pxor	@inp[0],@out[0]
+	#pxor	@inp[1],@out[1]
+	#movdqu	@out[0],`40*0+24-40*2`($inp)	# output iv FIX ME!
+	#pxor	@inp[2],@out[2]
+	#movdqu	@out[1],`40*1+24-40*2`($inp)
+	#pxor	@inp[3],@out[3]
+	#movdqu	@out[2],`40*2+24-40*2`($inp)	# won't fix, let caller
+	#movdqu	@out[3],`40*3+24-40*2`($inp)	# figure this out...
+
+	lea	`40*4`($inp),$inp
+	dec	$num
+	jnz	.Lenc4x_loop_grande
+
+.Lenc4x_done:
+___
+$code.=<<___ if ($win64);
+	movaps	-0xd8(%rax),%xmm6
+	movaps	-0xc8(%rax),%xmm7
+	movaps	-0xb8(%rax),%xmm8
+	movaps	-0xa8(%rax),%xmm9
+	movaps	-0x98(%rax),%xmm10
+	movaps	-0x88(%rax),%xmm11
+	movaps	-0x78(%rax),%xmm12
+	#movaps	-0x68(%rax),%xmm13
+	#movaps	-0x58(%rax),%xmm14
+	#movaps	-0x48(%rax),%xmm15
+___
+$code.=<<___;
+	mov	-48(%rax),%r15
+	mov	-40(%rax),%r14
+	mov	-32(%rax),%r13
+	mov	-24(%rax),%r12
+	mov	-16(%rax),%rbp
+	mov	-8(%rax),%rbx
+	lea	(%rax),%rsp
+.Lenc4x_epilogue:
+	ret
+.size	aesni_multi_cbc_encrypt,.-aesni_multi_cbc_encrypt
+
+.globl	aesni_multi_cbc_decrypt
+.type	aesni_multi_cbc_decrypt,\@function,3
+.align	32
+aesni_multi_cbc_decrypt:
+___
+$code.=<<___ if ($avx);
+	cmp	\$2,$num
+	jb	.Ldec_non_avx
+	mov	OPENSSL_ia32cap_P+4(%rip),%ecx
+	test	\$`1<<28`,%ecx			# AVX bit
+	jnz	_avx_cbc_dec_shortcut
+	jmp	.Ldec_non_avx
+.align	16
+.Ldec_non_avx:
+___
+$code.=<<___;
+	mov	%rsp,%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+___
+$code.=<<___ if ($win64);
+	lea	-0xa8(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+	movaps	%xmm10,0x40(%rsp)
+	movaps	%xmm11,0x50(%rsp)
+	movaps	%xmm12,0x60(%rsp)
+	movaps	%xmm13,-0x68(%rax)	# not used, saved to share se_handler 
+	movaps	%xmm14,-0x58(%rax)
+	movaps	%xmm15,-0x48(%rax)
+___
+$code.=<<___;
+	# stack layout
+	#
+	# +0	output sink
+	# +16	input sink [original %rsp and $num]
+	# +32	counters
+
+	sub	\$48,%rsp
+	and	\$-64,%rsp
+	mov	%rax,16(%rsp)			# original %rsp
+
+.Ldec4x_body:
+	movdqu	($key),$zero			# 0-round key
+	lea	0x78($key),$key			# size optimization
+	lea	40*2($inp),$inp
+
+.Ldec4x_loop_grande:
+	mov	$num,24(%rsp)			# original $num
+	xor	$num,$num
+___
+for($i=0;$i<4;$i++) {
+    $code.=<<___;
+	mov	`40*$i+16-40*2`($inp),$one	# borrow $one for number of blocks
+	mov	`40*$i+0-40*2`($inp),@inptr[$i]
+	cmp	$num,$one
+	mov	`40*$i+8-40*2`($inp),@outptr[$i]
+	cmovg	$one,$num			# find maximum
+	test	$one,$one
+	movdqu	`40*$i+24-40*2`($inp),@inp[$i]	# load IV
+	mov	$one,`32+4*$i`(%rsp)		# initialize counters
+	cmovle	%rsp,@inptr[$i]			# cancel input
+___
+}
+$code.=<<___;
+	test	$num,$num
+	jz	.Ldec4x_done
+
+	movups	0x10-0x78($key),$rndkey1
+	movups	0x20-0x78($key),$rndkey0
+	mov	0xf0-0x78($key),$rounds
+	movdqu	(@inptr[0]),@out[0]		# load inputs
+	movdqu	(@inptr[1]),@out[1]
+	 pxor	$zero,@out[0]
+	movdqu	(@inptr[2]),@out[2]
+	 pxor	$zero,@out[1]
+	movdqu	(@inptr[3]),@out[3]
+	 pxor	$zero,@out[2]
+	 pxor	$zero,@out[3]
+	movdqa	32(%rsp),$counters		# load counters
+	xor	$offset,$offset
+	jmp	.Loop_dec4x
+
+.align	32
+.Loop_dec4x:
+	add	\$16,$offset
+	lea	16(%rsp),$sink			# sink pointer
+	mov	\$1,$one			# constant of 1
+	sub	$offset,$sink
+
+	aesdec		$rndkey1,@out[0]
+	prefetcht0	31(@inptr[0],$offset)	# prefetch input
+	prefetcht0	31(@inptr[1],$offset)
+	aesdec		$rndkey1,@out[1]
+	prefetcht0	31(@inptr[2],$offset)
+	prefetcht0	31(@inptr[3],$offset)
+	aesdec		$rndkey1,@out[2]
+	aesdec		$rndkey1,@out[3]
+	movups		0x30-0x78($key),$rndkey1
+___
+for($i=0;$i<4;$i++) {
+my $rndkey = ($i&1) ? $rndkey1 : $rndkey0;
+$code.=<<___;
+	 cmp		`32+4*$i`(%rsp),$one
+	aesdec		$rndkey,@out[0]
+	aesdec		$rndkey,@out[1]
+	aesdec		$rndkey,@out[2]
+	 cmovge		$sink,@inptr[$i]	# cancel input
+	 cmovg		$sink,@outptr[$i]	# sink output
+	aesdec		$rndkey,@out[3]
+	movups		`0x40+16*$i-0x78`($key),$rndkey
+___
+}
+$code.=<<___;
+	 movdqa		$counters,$mask
+	aesdec		$rndkey0,@out[0]
+	prefetcht0	15(@outptr[0],$offset)	# prefetch output
+	prefetcht0	15(@outptr[1],$offset)
+	aesdec		$rndkey0,@out[1]
+	prefetcht0	15(@outptr[2],$offset)
+	prefetcht0	15(@outptr[3],$offset)
+	aesdec		$rndkey0,@out[2]
+	aesdec		$rndkey0,@out[3]
+	movups		0x80-0x78($key),$rndkey0
+	 pxor		$zero,$zero
+
+	aesdec		$rndkey1,@out[0]
+	 pcmpgtd	$zero,$mask
+	 movdqu		-0x78($key),$zero	# reload 0-round key
+	aesdec		$rndkey1,@out[1]
+	 paddd		$mask,$counters		# decrement counters
+	 movdqa		$counters,32(%rsp)	# update counters
+	aesdec		$rndkey1,@out[2]
+	aesdec		$rndkey1,@out[3]
+	movups		0x90-0x78($key),$rndkey1
+
+	cmp	\$11,$rounds
+
+	aesdec		$rndkey0,@out[0]
+	aesdec		$rndkey0,@out[1]
+	aesdec		$rndkey0,@out[2]
+	aesdec		$rndkey0,@out[3]
+	movups		0xa0-0x78($key),$rndkey0
+
+	jb	.Ldec4x_tail
+
+	aesdec		$rndkey1,@out[0]
+	aesdec		$rndkey1,@out[1]
+	aesdec		$rndkey1,@out[2]
+	aesdec		$rndkey1,@out[3]
+	movups		0xb0-0x78($key),$rndkey1
+
+	aesdec		$rndkey0,@out[0]
+	aesdec		$rndkey0,@out[1]
+	aesdec		$rndkey0,@out[2]
+	aesdec		$rndkey0,@out[3]
+	movups		0xc0-0x78($key),$rndkey0
+
+	je	.Ldec4x_tail
+
+	aesdec		$rndkey1,@out[0]
+	aesdec		$rndkey1,@out[1]
+	aesdec		$rndkey1,@out[2]
+	aesdec		$rndkey1,@out[3]
+	movups		0xd0-0x78($key),$rndkey1
+
+	aesdec		$rndkey0,@out[0]
+	aesdec		$rndkey0,@out[1]
+	aesdec		$rndkey0,@out[2]
+	aesdec		$rndkey0,@out[3]
+	movups		0xe0-0x78($key),$rndkey0
+	jmp	.Ldec4x_tail
+
+.align	32
+.Ldec4x_tail:
+	aesdec		$rndkey1,@out[0]
+	aesdec		$rndkey1,@out[1]
+	aesdec		$rndkey1,@out[2]
+	 pxor		$rndkey0,@inp[0]
+	 pxor		$rndkey0,@inp[1]
+	aesdec		$rndkey1,@out[3]
+	movdqu		0x10-0x78($key),$rndkey1
+	 pxor		$rndkey0,@inp[2]
+	 pxor		$rndkey0,@inp[3]
+	movdqu		0x20-0x78($key),$rndkey0
+
+	aesdeclast	@inp[0],@out[0]
+	aesdeclast	@inp[1],@out[1]
+	 movdqu		-16(@inptr[0],$offset),@inp[0]	# load next IV
+	 movdqu		-16(@inptr[1],$offset),@inp[1]
+	aesdeclast	@inp[2],@out[2]
+	aesdeclast	@inp[3],@out[3]
+	 movdqu		-16(@inptr[2],$offset),@inp[2]
+	 movdqu		-16(@inptr[3],$offset),@inp[3]
+
+	movups		@out[0],-16(@outptr[0],$offset)
+	 movdqu		(@inptr[0],$offset),@out[0]
+	movups		@out[1],-16(@outptr[1],$offset)	
+	 movdqu		(@inptr[1],$offset),@out[1]
+	 pxor		$zero,@out[0]
+	movups		@out[2],-16(@outptr[2],$offset)	
+	 movdqu		(@inptr[2],$offset),@out[2]
+	 pxor		$zero,@out[1]
+	movups		@out[3],-16(@outptr[3],$offset)
+	 movdqu		(@inptr[3],$offset),@out[3]
+	 pxor		$zero,@out[2]
+	 pxor		$zero,@out[3]
+
+	dec	$num
+	jnz	.Loop_dec4x
+
+	mov	16(%rsp),%rax			# original %rsp
+	mov	24(%rsp),$num
+
+	lea	`40*4`($inp),$inp
+	dec	$num
+	jnz	.Ldec4x_loop_grande
+
+.Ldec4x_done:
+___
+$code.=<<___ if ($win64);
+	movaps	-0xd8(%rax),%xmm6
+	movaps	-0xc8(%rax),%xmm7
+	movaps	-0xb8(%rax),%xmm8
+	movaps	-0xa8(%rax),%xmm9
+	movaps	-0x98(%rax),%xmm10
+	movaps	-0x88(%rax),%xmm11
+	movaps	-0x78(%rax),%xmm12
+	#movaps	-0x68(%rax),%xmm13
+	#movaps	-0x58(%rax),%xmm14
+	#movaps	-0x48(%rax),%xmm15
+___
+$code.=<<___;
+	mov	-48(%rax),%r15
+	mov	-40(%rax),%r14
+	mov	-32(%rax),%r13
+	mov	-24(%rax),%r12
+	mov	-16(%rax),%rbp
+	mov	-8(%rax),%rbx
+	lea	(%rax),%rsp
+.Ldec4x_epilogue:
+	ret
+.size	aesni_multi_cbc_decrypt,.-aesni_multi_cbc_decrypt
+___
+
+						if ($avx) {{{
+my @ptr=map("%r$_",(8..15));
+my $offload=$sink;
+
+my @out=map("%xmm$_",(2..9));
+my @inp=map("%xmm$_",(10..13));
+my ($counters,$zero)=("%xmm14","%xmm15");
+
+$code.=<<___;
+.type	aesni_multi_cbc_encrypt_avx,\@function,3
+.align	32
+aesni_multi_cbc_encrypt_avx:
+_avx_cbc_enc_shortcut:
+	mov	%rsp,%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+___
+$code.=<<___ if ($win64);
+	lea	-0xa8(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+	movaps	%xmm10,0x40(%rsp)
+	movaps	%xmm11,0x50(%rsp)
+	movaps	%xmm12,-0x78(%rax)
+	movaps	%xmm13,-0x68(%rax)
+	movaps	%xmm14,-0x58(%rax)
+	movaps	%xmm15,-0x48(%rax)
+___
+$code.=<<___;
+	# stack layout
+	#
+	# +0	output sink
+	# +16	input sink [original %rsp and $num]
+	# +32	counters
+	# +64	distances between inputs and outputs
+	# +128	off-load area for @inp[0..3]
+
+	sub	\$192,%rsp
+	and	\$-128,%rsp
+	mov	%rax,16(%rsp)			# original %rsp
+
+.Lenc8x_body:
+	vzeroupper
+	vmovdqu	($key),$zero			# 0-round key
+	lea	0x78($key),$key			# size optimization
+	lea	40*4($inp),$inp
+	shr	\$1,$num
+
+.Lenc8x_loop_grande:
+	#mov	$num,24(%rsp)			# original $num
+	xor	$num,$num
+___
+for($i=0;$i<8;$i++) {
+  my $temp = $i ? $offload : $offset;
+    $code.=<<___;
+	mov	`40*$i+16-40*4`($inp),$one	# borrow $one for number of blocks
+	mov	`40*$i+0-40*4`($inp),@ptr[$i]	# input pointer
+	cmp	$num,$one
+	mov	`40*$i+8-40*4`($inp),$temp	# output pointer
+	cmovg	$one,$num			# find maximum
+	test	$one,$one
+	vmovdqu	`40*$i+24-40*4`($inp),@out[$i]	# load IV
+	mov	$one,`32+4*$i`(%rsp)		# initialize counters
+	cmovle	%rsp,@ptr[$i]			# cancel input
+	sub	@ptr[$i],$temp			# distance between input and output
+	mov	$temp,`64+8*$i`(%rsp)		# initialize distances
+___
+}
+$code.=<<___;
+	test	$num,$num
+	jz	.Lenc8x_done
+
+	vmovups	0x10-0x78($key),$rndkey1
+	vmovups	0x20-0x78($key),$rndkey0
+	mov	0xf0-0x78($key),$rounds
+
+	vpxor	(@ptr[0]),$zero,@inp[0]		# load inputs and xor with 0-round
+	 lea	128(%rsp),$offload		# offload area
+	vpxor	(@ptr[1]),$zero,@inp[1]
+	vpxor	(@ptr[2]),$zero,@inp[2]
+	vpxor	(@ptr[3]),$zero,@inp[3]
+	 vpxor	@inp[0],@out[0],@out[0]
+	vpxor	(@ptr[4]),$zero,@inp[0]
+	 vpxor	@inp[1],@out[1],@out[1]
+	vpxor	(@ptr[5]),$zero,@inp[1]
+	 vpxor	@inp[2],@out[2],@out[2]
+	vpxor	(@ptr[6]),$zero,@inp[2]
+	 vpxor	@inp[3],@out[3],@out[3]
+	vpxor	(@ptr[7]),$zero,@inp[3]
+	 vpxor	@inp[0],@out[4],@out[4]
+	mov	\$1,$one			# constant of 1
+	 vpxor	@inp[1],@out[5],@out[5]
+	 vpxor	@inp[2],@out[6],@out[6]
+	 vpxor	@inp[3],@out[7],@out[7]
+	jmp	.Loop_enc8x
+
+.align	32
+.Loop_enc8x:
+___
+for($i=0;$i<8;$i++) {
+my $rndkey=($i&1)?$rndkey0:$rndkey1;
+$code.=<<___;
+	vaesenc		$rndkey,@out[0],@out[0]
+	 cmp		32+4*$i(%rsp),$one
+___
+$code.=<<___ if ($i);
+	 mov		64+8*$i(%rsp),$offset
+___
+$code.=<<___;
+	vaesenc		$rndkey,@out[1],@out[1]
+	prefetcht0	31(@ptr[$i])			# prefetch input
+	vaesenc		$rndkey,@out[2],@out[2]
+___
+$code.=<<___ if ($i>1);
+	prefetcht0	15(@ptr[$i-2])			# prefetch output
+___
+$code.=<<___;
+	vaesenc		$rndkey,@out[3],@out[3]
+	 lea		(@ptr[$i],$offset),$offset
+	 cmovge		%rsp,@ptr[$i]			# cancel input
+	vaesenc		$rndkey,@out[4],@out[4]
+	 cmovg		%rsp,$offset			# sink output
+	vaesenc		$rndkey,@out[5],@out[5]
+	 sub		@ptr[$i],$offset
+	vaesenc		$rndkey,@out[6],@out[6]
+	 vpxor		16(@ptr[$i]),$zero,@inp[$i%4]	# load input and xor with 0-round
+	 mov		$offset,64+8*$i(%rsp)
+	vaesenc		$rndkey,@out[7],@out[7]
+	vmovups		`16*(3+$i)-0x78`($key),$rndkey
+	 lea		16(@ptr[$i],$offset),@ptr[$i]	# switch to output
+___
+$code.=<<___ if ($i<4)
+	 vmovdqu	@inp[$i%4],`16*$i`($offload)	# off-load
+___
+}
+$code.=<<___;
+	 vmovdqu	32(%rsp),$counters
+	prefetcht0	15(@ptr[$i-2])			# prefetch output
+	prefetcht0	15(@ptr[$i-1])
+	cmp	\$11,$rounds
+	jb	.Lenc8x_tail
+
+	vaesenc		$rndkey1,@out[0],@out[0]
+	vaesenc		$rndkey1,@out[1],@out[1]
+	vaesenc		$rndkey1,@out[2],@out[2]
+	vaesenc		$rndkey1,@out[3],@out[3]
+	vaesenc		$rndkey1,@out[4],@out[4]
+	vaesenc		$rndkey1,@out[5],@out[5]
+	vaesenc		$rndkey1,@out[6],@out[6]
+	vaesenc		$rndkey1,@out[7],@out[7]
+	vmovups		0xb0-0x78($key),$rndkey1
+
+	vaesenc		$rndkey0,@out[0],@out[0]
+	vaesenc		$rndkey0,@out[1],@out[1]
+	vaesenc		$rndkey0,@out[2],@out[2]
+	vaesenc		$rndkey0,@out[3],@out[3]
+	vaesenc		$rndkey0,@out[4],@out[4]
+	vaesenc		$rndkey0,@out[5],@out[5]
+	vaesenc		$rndkey0,@out[6],@out[6]
+	vaesenc		$rndkey0,@out[7],@out[7]
+	vmovups		0xc0-0x78($key),$rndkey0
+	je	.Lenc8x_tail
+
+	vaesenc		$rndkey1,@out[0],@out[0]
+	vaesenc		$rndkey1,@out[1],@out[1]
+	vaesenc		$rndkey1,@out[2],@out[2]
+	vaesenc		$rndkey1,@out[3],@out[3]
+	vaesenc		$rndkey1,@out[4],@out[4]
+	vaesenc		$rndkey1,@out[5],@out[5]
+	vaesenc		$rndkey1,@out[6],@out[6]
+	vaesenc		$rndkey1,@out[7],@out[7]
+	vmovups		0xd0-0x78($key),$rndkey1
+
+	vaesenc		$rndkey0,@out[0],@out[0]
+	vaesenc		$rndkey0,@out[1],@out[1]
+	vaesenc		$rndkey0,@out[2],@out[2]
+	vaesenc		$rndkey0,@out[3],@out[3]
+	vaesenc		$rndkey0,@out[4],@out[4]
+	vaesenc		$rndkey0,@out[5],@out[5]
+	vaesenc		$rndkey0,@out[6],@out[6]
+	vaesenc		$rndkey0,@out[7],@out[7]
+	vmovups		0xe0-0x78($key),$rndkey0
+
+.Lenc8x_tail:
+	vaesenc		$rndkey1,@out[0],@out[0]
+	 vpxor		$zero,$zero,$zero
+	vaesenc		$rndkey1,@out[1],@out[1]
+	vaesenc		$rndkey1,@out[2],@out[2]
+	 vpcmpgtd	$zero,$counters,$zero
+	vaesenc		$rndkey1,@out[3],@out[3]
+	vaesenc		$rndkey1,@out[4],@out[4]
+	 vpaddd		$counters,$zero,$zero		# decrement counters
+	 vmovdqu	48(%rsp),$counters
+	vaesenc		$rndkey1,@out[5],@out[5]
+	 mov		64(%rsp),$offset		# pre-load 1st offset
+	vaesenc		$rndkey1,@out[6],@out[6]
+	vaesenc		$rndkey1,@out[7],@out[7]
+	vmovups		0x10-0x78($key),$rndkey1
+
+	vaesenclast	$rndkey0,@out[0],@out[0]
+	 vmovdqa	$zero,32(%rsp)			# update counters
+	 vpxor		$zero,$zero,$zero
+	vaesenclast	$rndkey0,@out[1],@out[1]
+	vaesenclast	$rndkey0,@out[2],@out[2]
+	 vpcmpgtd	$zero,$counters,$zero
+	vaesenclast	$rndkey0,@out[3],@out[3]
+	vaesenclast	$rndkey0,@out[4],@out[4]
+	 vpaddd		$zero,$counters,$counters	# decrement counters
+	 vmovdqu	-0x78($key),$zero		# 0-round
+	vaesenclast	$rndkey0,@out[5],@out[5]
+	vaesenclast	$rndkey0,@out[6],@out[6]
+	 vmovdqa	$counters,48(%rsp)		# update counters
+	vaesenclast	$rndkey0,@out[7],@out[7]
+	vmovups		0x20-0x78($key),$rndkey0
+
+	vmovups		@out[0],-16(@ptr[0])		# write output
+	 sub		$offset,@ptr[0]			# switch to input
+	 vpxor		0x00($offload),@out[0],@out[0]
+	vmovups		@out[1],-16(@ptr[1])	
+	 sub		`64+1*8`(%rsp),@ptr[1]
+	 vpxor		0x10($offload),@out[1],@out[1]
+	vmovups		@out[2],-16(@ptr[2])	
+	 sub		`64+2*8`(%rsp),@ptr[2]
+	 vpxor		0x20($offload),@out[2],@out[2]
+	vmovups		@out[3],-16(@ptr[3])
+	 sub		`64+3*8`(%rsp),@ptr[3]
+	 vpxor		0x30($offload),@out[3],@out[3]
+	vmovups		@out[4],-16(@ptr[4])
+	 sub		`64+4*8`(%rsp),@ptr[4]
+	 vpxor		@inp[0],@out[4],@out[4]
+	vmovups		@out[5],-16(@ptr[5])	
+	 sub		`64+5*8`(%rsp),@ptr[5]
+	 vpxor		@inp[1],@out[5],@out[5]
+	vmovups		@out[6],-16(@ptr[6])	
+	 sub		`64+6*8`(%rsp),@ptr[6]
+	 vpxor		@inp[2],@out[6],@out[6]
+	vmovups		@out[7],-16(@ptr[7])
+	 sub		`64+7*8`(%rsp),@ptr[7]
+	 vpxor		@inp[3],@out[7],@out[7]
+
+	dec	$num
+	jnz	.Loop_enc8x
+
+	mov	16(%rsp),%rax			# original %rsp
+	#mov	24(%rsp),$num
+	#lea	`40*8`($inp),$inp
+	#dec	$num
+	#jnz	.Lenc8x_loop_grande
+
+.Lenc8x_done:
+	vzeroupper
+___
+$code.=<<___ if ($win64);
+	movaps	-0xd8(%rax),%xmm6
+	movaps	-0xc8(%rax),%xmm7
+	movaps	-0xb8(%rax),%xmm8
+	movaps	-0xa8(%rax),%xmm9
+	movaps	-0x98(%rax),%xmm10
+	movaps	-0x88(%rax),%xmm11
+	movaps	-0x78(%rax),%xmm12
+	movaps	-0x68(%rax),%xmm13
+	movaps	-0x58(%rax),%xmm14
+	movaps	-0x48(%rax),%xmm15
+___
+$code.=<<___;
+	mov	-48(%rax),%r15
+	mov	-40(%rax),%r14
+	mov	-32(%rax),%r13
+	mov	-24(%rax),%r12
+	mov	-16(%rax),%rbp
+	mov	-8(%rax),%rbx
+	lea	(%rax),%rsp
+.Lenc8x_epilogue:
+	ret
+.size	aesni_multi_cbc_encrypt_avx,.-aesni_multi_cbc_encrypt_avx
+
+.type	aesni_multi_cbc_decrypt_avx,\@function,3
+.align	32
+aesni_multi_cbc_decrypt_avx:
+_avx_cbc_dec_shortcut:
+	mov	%rsp,%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+___
+$code.=<<___ if ($win64);
+	lea	-0xa8(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+	movaps	%xmm10,0x40(%rsp)
+	movaps	%xmm11,0x50(%rsp)
+	movaps	%xmm12,-0x78(%rax)
+	movaps	%xmm13,-0x68(%rax)
+	movaps	%xmm14,-0x58(%rax)
+	movaps	%xmm15,-0x48(%rax)
+___
+$code.=<<___;
+	# stack layout
+	#
+	# +0	output sink
+	# +16	input sink [original %rsp and $num]
+	# +32	counters
+	# +64	distances between inputs and outputs
+	# +128	off-load area for @inp[0..3]
+	# +192	IV/input offload
+
+	sub	\$256,%rsp
+	and	\$-256,%rsp
+	sub	\$192,%rsp
+	mov	%rax,16(%rsp)			# original %rsp
+
+.Ldec8x_body:
+	vzeroupper
+	vmovdqu	($key),$zero			# 0-round key
+	lea	0x78($key),$key			# size optimization
+	lea	40*4($inp),$inp
+	shr	\$1,$num
+
+.Ldec8x_loop_grande:
+	#mov	$num,24(%rsp)			# original $num
+	xor	$num,$num
+___
+for($i=0;$i<8;$i++) {
+  my $temp = $i ? $offload : $offset;
+    $code.=<<___;
+	mov	`40*$i+16-40*4`($inp),$one	# borrow $one for number of blocks
+	mov	`40*$i+0-40*4`($inp),@ptr[$i]	# input pointer
+	cmp	$num,$one
+	mov	`40*$i+8-40*4`($inp),$temp	# output pointer
+	cmovg	$one,$num			# find maximum
+	test	$one,$one
+	vmovdqu	`40*$i+24-40*4`($inp),@out[$i]	# load IV
+	mov	$one,`32+4*$i`(%rsp)		# initialize counters
+	cmovle	%rsp,@ptr[$i]			# cancel input
+	sub	@ptr[$i],$temp			# distance between input and output
+	mov	$temp,`64+8*$i`(%rsp)		# initialize distances
+	vmovdqu	@out[$i],`192+16*$i`(%rsp)	# offload IV
+___
+}
+$code.=<<___;
+	test	$num,$num
+	jz	.Ldec8x_done
+
+	vmovups	0x10-0x78($key),$rndkey1
+	vmovups	0x20-0x78($key),$rndkey0
+	mov	0xf0-0x78($key),$rounds
+	 lea	192+128(%rsp),$offload		# offload area
+
+	vmovdqu	(@ptr[0]),@out[0]		# load inputs
+	vmovdqu	(@ptr[1]),@out[1]
+	vmovdqu	(@ptr[2]),@out[2]
+	vmovdqu	(@ptr[3]),@out[3]
+	vmovdqu	(@ptr[4]),@out[4]
+	vmovdqu	(@ptr[5]),@out[5]
+	vmovdqu	(@ptr[6]),@out[6]
+	vmovdqu	(@ptr[7]),@out[7]
+	vmovdqu	@out[0],0x00($offload)		# offload inputs
+	vpxor	$zero,@out[0],@out[0]		# xor inputs with 0-round
+	vmovdqu	@out[1],0x10($offload)
+	vpxor	$zero,@out[1],@out[1]
+	vmovdqu	@out[2],0x20($offload)
+	vpxor	$zero,@out[2],@out[2]
+	vmovdqu	@out[3],0x30($offload)
+	vpxor	$zero,@out[3],@out[3]
+	vmovdqu	@out[4],0x40($offload)
+	vpxor	$zero,@out[4],@out[4]
+	vmovdqu	@out[5],0x50($offload)
+	vpxor	$zero,@out[5],@out[5]
+	vmovdqu	@out[6],0x60($offload)
+	vpxor	$zero,@out[6],@out[6]
+	vmovdqu	@out[7],0x70($offload)
+	vpxor	$zero,@out[7],@out[7]
+	xor	\$0x80,$offload
+	mov	\$1,$one			# constant of 1
+	jmp	.Loop_dec8x
+
+.align	32
+.Loop_dec8x:
+___
+for($i=0;$i<8;$i++) {
+my $rndkey=($i&1)?$rndkey0:$rndkey1;
+$code.=<<___;
+	vaesdec		$rndkey,@out[0],@out[0]
+	 cmp		32+4*$i(%rsp),$one
+___
+$code.=<<___ if ($i);
+	 mov		64+8*$i(%rsp),$offset
+___
+$code.=<<___;
+	vaesdec		$rndkey,@out[1],@out[1]
+	prefetcht0	31(@ptr[$i])			# prefetch input
+	vaesdec		$rndkey,@out[2],@out[2]
+___
+$code.=<<___ if ($i>1);
+	prefetcht0	15(@ptr[$i-2])			# prefetch output
+___
+$code.=<<___;
+	vaesdec		$rndkey,@out[3],@out[3]
+	 lea		(@ptr[$i],$offset),$offset
+	 cmovge		%rsp,@ptr[$i]			# cancel input
+	vaesdec		$rndkey,@out[4],@out[4]
+	 cmovg		%rsp,$offset			# sink output
+	vaesdec		$rndkey,@out[5],@out[5]
+	 sub		@ptr[$i],$offset
+	vaesdec		$rndkey,@out[6],@out[6]
+	 vmovdqu	16(@ptr[$i]),@inp[$i%4]		# load input
+	 mov		$offset,64+8*$i(%rsp)
+	vaesdec		$rndkey,@out[7],@out[7]
+	vmovups		`16*(3+$i)-0x78`($key),$rndkey
+	 lea		16(@ptr[$i],$offset),@ptr[$i]	# switch to output
+___
+$code.=<<___ if ($i<4);
+	 vmovdqu	@inp[$i%4],`128+16*$i`(%rsp)	# off-load
+___
+}
+$code.=<<___;
+	 vmovdqu	32(%rsp),$counters
+	prefetcht0	15(@ptr[$i-2])			# prefetch output
+	prefetcht0	15(@ptr[$i-1])
+	cmp	\$11,$rounds
+	jb	.Ldec8x_tail
+
+	vaesdec		$rndkey1,@out[0],@out[0]
+	vaesdec		$rndkey1,@out[1],@out[1]
+	vaesdec		$rndkey1,@out[2],@out[2]
+	vaesdec		$rndkey1,@out[3],@out[3]
+	vaesdec		$rndkey1,@out[4],@out[4]
+	vaesdec		$rndkey1,@out[5],@out[5]
+	vaesdec		$rndkey1,@out[6],@out[6]
+	vaesdec		$rndkey1,@out[7],@out[7]
+	vmovups		0xb0-0x78($key),$rndkey1
+
+	vaesdec		$rndkey0,@out[0],@out[0]
+	vaesdec		$rndkey0,@out[1],@out[1]
+	vaesdec		$rndkey0,@out[2],@out[2]
+	vaesdec		$rndkey0,@out[3],@out[3]
+	vaesdec		$rndkey0,@out[4],@out[4]
+	vaesdec		$rndkey0,@out[5],@out[5]
+	vaesdec		$rndkey0,@out[6],@out[6]
+	vaesdec		$rndkey0,@out[7],@out[7]
+	vmovups		0xc0-0x78($key),$rndkey0
+	je	.Ldec8x_tail
+
+	vaesdec		$rndkey1,@out[0],@out[0]
+	vaesdec		$rndkey1,@out[1],@out[1]
+	vaesdec		$rndkey1,@out[2],@out[2]
+	vaesdec		$rndkey1,@out[3],@out[3]
+	vaesdec		$rndkey1,@out[4],@out[4]
+	vaesdec		$rndkey1,@out[5],@out[5]
+	vaesdec		$rndkey1,@out[6],@out[6]
+	vaesdec		$rndkey1,@out[7],@out[7]
+	vmovups		0xd0-0x78($key),$rndkey1
+
+	vaesdec		$rndkey0,@out[0],@out[0]
+	vaesdec		$rndkey0,@out[1],@out[1]
+	vaesdec		$rndkey0,@out[2],@out[2]
+	vaesdec		$rndkey0,@out[3],@out[3]
+	vaesdec		$rndkey0,@out[4],@out[4]
+	vaesdec		$rndkey0,@out[5],@out[5]
+	vaesdec		$rndkey0,@out[6],@out[6]
+	vaesdec		$rndkey0,@out[7],@out[7]
+	vmovups		0xe0-0x78($key),$rndkey0
+
+.Ldec8x_tail:
+	vaesdec		$rndkey1,@out[0],@out[0]
+	 vpxor		$zero,$zero,$zero
+	vaesdec		$rndkey1,@out[1],@out[1]
+	vaesdec		$rndkey1,@out[2],@out[2]
+	 vpcmpgtd	$zero,$counters,$zero
+	vaesdec		$rndkey1,@out[3],@out[3]
+	vaesdec		$rndkey1,@out[4],@out[4]
+	 vpaddd		$counters,$zero,$zero		# decrement counters
+	 vmovdqu	48(%rsp),$counters
+	vaesdec		$rndkey1,@out[5],@out[5]
+	 mov		64(%rsp),$offset		# pre-load 1st offset
+	vaesdec		$rndkey1,@out[6],@out[6]
+	vaesdec		$rndkey1,@out[7],@out[7]
+	vmovups		0x10-0x78($key),$rndkey1
+
+	vaesdeclast	$rndkey0,@out[0],@out[0]
+	 vmovdqa	$zero,32(%rsp)			# update counters
+	 vpxor		$zero,$zero,$zero
+	vaesdeclast	$rndkey0,@out[1],@out[1]
+	vpxor		0x00($offload),@out[0],@out[0]	# xor with IV
+	vaesdeclast	$rndkey0,@out[2],@out[2]
+	vpxor		0x10($offload),@out[1],@out[1]
+	 vpcmpgtd	$zero,$counters,$zero
+	vaesdeclast	$rndkey0,@out[3],@out[3]
+	vpxor		0x20($offload),@out[2],@out[2]
+	vaesdeclast	$rndkey0,@out[4],@out[4]
+	vpxor		0x30($offload),@out[3],@out[3]
+	 vpaddd		$zero,$counters,$counters	# decrement counters
+	 vmovdqu	-0x78($key),$zero		# 0-round
+	vaesdeclast	$rndkey0,@out[5],@out[5]
+	vpxor		0x40($offload),@out[4],@out[4]
+	vaesdeclast	$rndkey0,@out[6],@out[6]
+	vpxor		0x50($offload),@out[5],@out[5]
+	 vmovdqa	$counters,48(%rsp)		# update counters
+	vaesdeclast	$rndkey0,@out[7],@out[7]
+	vpxor		0x60($offload),@out[6],@out[6]
+	vmovups		0x20-0x78($key),$rndkey0
+
+	vmovups		@out[0],-16(@ptr[0])		# write output
+	 sub		$offset,@ptr[0]			# switch to input
+	 vmovdqu	128+0(%rsp),@out[0]
+	vpxor		0x70($offload),@out[7],@out[7]
+	vmovups		@out[1],-16(@ptr[1])	
+	 sub		`64+1*8`(%rsp),@ptr[1]
+	 vmovdqu	@out[0],0x00($offload)
+	 vpxor		$zero,@out[0],@out[0]
+	 vmovdqu	128+16(%rsp),@out[1]
+	vmovups		@out[2],-16(@ptr[2])	
+	 sub		`64+2*8`(%rsp),@ptr[2]
+	 vmovdqu	@out[1],0x10($offload)
+	 vpxor		$zero,@out[1],@out[1]
+	 vmovdqu	128+32(%rsp),@out[2]
+	vmovups		@out[3],-16(@ptr[3])
+	 sub		`64+3*8`(%rsp),@ptr[3]
+	 vmovdqu	@out[2],0x20($offload)
+	 vpxor		$zero,@out[2],@out[2]
+	 vmovdqu	128+48(%rsp),@out[3]
+	vmovups		@out[4],-16(@ptr[4])
+	 sub		`64+4*8`(%rsp),@ptr[4]
+	 vmovdqu	@out[3],0x30($offload)
+	 vpxor		$zero,@out[3],@out[3]
+	 vmovdqu	@inp[0],0x40($offload)
+	 vpxor		@inp[0],$zero,@out[4]
+	vmovups		@out[5],-16(@ptr[5])	
+	 sub		`64+5*8`(%rsp),@ptr[5]
+	 vmovdqu	@inp[1],0x50($offload)
+	 vpxor		@inp[1],$zero,@out[5]
+	vmovups		@out[6],-16(@ptr[6])	
+	 sub		`64+6*8`(%rsp),@ptr[6]
+	 vmovdqu	@inp[2],0x60($offload)
+	 vpxor		@inp[2],$zero,@out[6]
+	vmovups		@out[7],-16(@ptr[7])
+	 sub		`64+7*8`(%rsp),@ptr[7]
+	 vmovdqu	@inp[3],0x70($offload)
+	 vpxor		@inp[3],$zero,@out[7]
+
+	xor	\$128,$offload
+	dec	$num
+	jnz	.Loop_dec8x
+
+	mov	16(%rsp),%rax			# original %rsp
+	#mov	24(%rsp),$num
+	#lea	`40*8`($inp),$inp
+	#dec	$num
+	#jnz	.Ldec8x_loop_grande
+
+.Ldec8x_done:
+	vzeroupper
+___
+$code.=<<___ if ($win64);
+	movaps	-0xd8(%rax),%xmm6
+	movaps	-0xc8(%rax),%xmm7
+	movaps	-0xb8(%rax),%xmm8
+	movaps	-0xa8(%rax),%xmm9
+	movaps	-0x98(%rax),%xmm10
+	movaps	-0x88(%rax),%xmm11
+	movaps	-0x78(%rax),%xmm12
+	movaps	-0x68(%rax),%xmm13
+	movaps	-0x58(%rax),%xmm14
+	movaps	-0x48(%rax),%xmm15
+___
+$code.=<<___;
+	mov	-48(%rax),%r15
+	mov	-40(%rax),%r14
+	mov	-32(%rax),%r13
+	mov	-24(%rax),%r12
+	mov	-16(%rax),%rbp
+	mov	-8(%rax),%rbx
+	lea	(%rax),%rsp
+.Ldec8x_epilogue:
+	ret
+.size	aesni_multi_cbc_decrypt_avx,.-aesni_multi_cbc_decrypt_avx
+___
+						}}}
+
+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<.Lprologue
+	jb	.Lin_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
+	jae	.Lin_prologue
+
+	mov	16(%rax),%rax		# pull saved stack pointer
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore cotnext->R12
+	mov	%r13,224($context)	# restore cotnext->R13
+	mov	%r14,232($context)	# restore cotnext->R14
+	mov	%r15,240($context)	# restore cotnext->R15
+
+	lea	-56-10*16(%rax),%rsi
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+
+.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	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	se_handler,.-se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_aesni_multi_cbc_encrypt
+	.rva	.LSEH_end_aesni_multi_cbc_encrypt
+	.rva	.LSEH_info_aesni_multi_cbc_encrypt
+	.rva	.LSEH_begin_aesni_multi_cbc_decrypt
+	.rva	.LSEH_end_aesni_multi_cbc_decrypt
+	.rva	.LSEH_info_aesni_multi_cbc_decrypt
+___
+$code.=<<___ if ($avx);
+	.rva	.LSEH_begin_aesni_multi_cbc_encrypt_avx
+	.rva	.LSEH_end_aesni_multi_cbc_encrypt_avx
+	.rva	.LSEH_info_aesni_multi_cbc_encrypt_avx
+	.rva	.LSEH_begin_aesni_multi_cbc_decrypt_avx
+	.rva	.LSEH_end_aesni_multi_cbc_decrypt_avx
+	.rva	.LSEH_info_aesni_multi_cbc_decrypt_avx
+___
+$code.=<<___;
+.section	.xdata
+.align	8
+.LSEH_info_aesni_multi_cbc_encrypt:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lenc4x_body,.Lenc4x_epilogue		# HandlerData[]
+.LSEH_info_aesni_multi_cbc_decrypt:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Ldec4x_body,.Ldec4x_epilogue		# HandlerData[]
+___
+$code.=<<___ if ($avx);
+.LSEH_info_aesni_multi_cbc_encrypt_avx:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lenc8x_body,.Lenc8x_epilogue		# HandlerData[]
+.LSEH_info_aesni_multi_cbc_decrypt_avx:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Ldec8x_body,.Ldec8x_epilogue		# HandlerData[]
+___
+}
+####################################################################
+
+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 $line;
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
+
+print $code;
+close STDOUT;
diff --git a/openssl/crypto/aes/asm/aesni-sha1-x86_64.pl b/openssl/crypto/aes/asm/aesni-sha1-x86_64.pl
index 3c8f6c19e..97992adca 100644
--- a/openssl/crypto/aes/asm/aesni-sha1-x86_64.pl
+++ b/openssl/crypto/aes/asm/aesni-sha1-x86_64.pl
@@ -21,16 +21,25 @@
 # subroutine:
 #
 #		AES-128-CBC	+SHA1		stitch      gain
-# Westmere	3.77[+5.6]	9.37		6.65	    +41%
-# Sandy Bridge	5.05[+5.2(6.3)]	10.25(11.35)	6.16(7.08)  +67%(+60%)
+# Westmere	3.77[+5.3]	9.07		6.55	    +38%
+# Sandy Bridge	5.05[+5.0(6.1)]	10.06(11.15)	5.98(7.05)  +68%(+58%)
+# Ivy Bridge	5.05[+4.6]	9.65		5.54        +74%
+# Haswell	4.43[+3.6(4.2)]	8.00(8.58)	4.55(5.21)  +75%(+65%)
+# Bulldozer	5.77[+6.0]	11.72		6.37        +84%
 #
 #		AES-192-CBC
-# Westmere	4.51		10.11		6.97	    +45%
-# Sandy Bridge	6.05		11.25(12.35)	6.34(7.27)  +77%(+70%)
+# Westmere	4.51		9.81		6.80	    +44%
+# Sandy Bridge	6.05		11.06(12.15)	6.11(7.19)  +81%(+69%)
+# Ivy Bridge	6.05		10.65		6.07        +75%
+# Haswell	5.29		8.86(9.44)	5.32(5.32)  +67%(+77%)
+# Bulldozer	6.89		12.84		6.96        +84%
 #
 #		AES-256-CBC
-# Westmere	5.25		10.85		7.25	    +50%
-# Sandy Bridge	7.05		12.25(13.35)	7.06(7.70)  +74%(+73%)
+# Westmere	5.25		10.55		7.21	    +46%
+# Sandy Bridge	7.05		12.06(13.15)	7.12(7.72)  +69%(+70%)
+# Ivy Bridge	7.05		11.65		7.12        +64%
+# Haswell	6.19		9.76(10.34)	6.21(6.25)  +57%(+65%)
+# Bulldozer	8.00		13.95		8.25        +69%
 #
 # (*)	There are two code paths: SSSE3 and AVX. See sha1-568.pl for
 #	background information. Above numbers in parentheses are SSSE3
@@ -45,8 +54,25 @@
 # standalone AESNI-CBC decrypt:
 #
 #		AES-128-CBC	AES-192-CBC	AES-256-CBC
-# Westmere	1.31		1.55		1.80
-# Sandy Bridge	0.93		1.06		1.22
+# Westmere	1.25		1.50		1.75
+# Sandy Bridge	0.74		0.91		1.09
+# Ivy Bridge	0.74		0.90		1.11
+# Haswell	0.63		0.76		0.88
+# Bulldozer	0.70		0.85		0.99
+
+# And indeed:
+#
+#		AES-256-CBC	+SHA1		stitch      gain
+# Westmere	1.75		7.20		6.68        +7.8%
+# Sandy Bridge	1.09		6.09(7.22)	5.82(6.95)  +4.6%(+3.9%)
+# Ivy Bridge	1.11		5.70		5.45        +4.6%
+# Haswell	0.88		4.45(5.00)	4.39(4.69)  +1.4%(*)(+6.6%)
+# Bulldozer	0.99		6.95		5.95        +17%(**)
+#
+# (*)	Tiny improvement coefficient on Haswell is because we compare
+#	AVX1 stitch to sum with AVX2 SHA1.
+# (**)	Execution is fully dominated by integer code sequence and
+#	SIMD still hardly shows [in single-process benchmark;-]
 
 $flavour = shift;
 $output  = shift;
@@ -68,6 +94,11 @@ $avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 $avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	   `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
 	   $1>=10);
+$avx=1 if (!$avx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/ && $2>=3.0);
+
+$shaext=1;	### set to zero if compiling for 1.0.1
+
+$stitched_decrypt=0;
 
 open OUT,"| \"$^X\" $xlate $flavour $output";
 *STDOUT=*OUT;
@@ -86,11 +117,15 @@ $code.=<<___;
 
 .globl	aesni_cbc_sha1_enc
 .type	aesni_cbc_sha1_enc,\@abi-omnipotent
-.align	16
+.align	32
 aesni_cbc_sha1_enc:
 	# caller should check for SSSE3 and AES-NI bits
 	mov	OPENSSL_ia32cap_P+0(%rip),%r10d
-	mov	OPENSSL_ia32cap_P+4(%rip),%r11d
+	mov	OPENSSL_ia32cap_P+4(%rip),%r11
+___
+$code.=<<___ if ($shaext);
+	bt	\$61,%r11		# check SHA bit
+	jc	aesni_cbc_sha1_enc_shaext
 ___
 $code.=<<___ if ($avx);
 	and	\$`1<<28`,%r11d		# mask AVX bit
@@ -112,10 +147,21 @@ my @X=map("%xmm$_",(4..7,0..3));
 my @Tx=map("%xmm$_",(8..10));
 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp");	# size optimization
 my @T=("%esi","%edi");
-my $j=0; my $jj=0; my $r=0; my $sn=0;
+my $j=0; my $jj=0; my $r=0; my $sn=0; my $rx=0;
 my $K_XX_XX="%r11";
-my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13));
-my @rndkey=("%xmm14","%xmm15");
+my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13));			# for enc
+my @rndkey=("%xmm14","%xmm15");					# for enc
+my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15));	# for dec
+
+if (1) {	# reassign for Atom Silvermont
+    # The goal is to minimize amount of instructions with more than
+    # 3 prefix bytes. Or in more practical terms to keep AES-NI *and*
+    # SSSE3 instructions to upper half of the register bank.
+    @X=map("%xmm$_",(8..11,4..7));
+    @Tx=map("%xmm$_",(12,13,3));
+    ($iv,$in,$rndkey0)=map("%xmm$_",(2,14,15));
+    @rndkey=("%xmm0","%xmm1");
+}
 
 sub AUTOLOAD()		# thunk [simplified] 32-bit style perlasm
 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
@@ -129,7 +175,7 @@ my $_ror=sub { &ror(@_) };
 
 $code.=<<___;
 .type	aesni_cbc_sha1_enc_ssse3,\@function,6
-.align	16
+.align	32
 aesni_cbc_sha1_enc_ssse3:
 	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
 	#shr	\$6,$len			# debugging artefact
@@ -161,16 +207,16 @@ $code.=<<___;
 	mov	$in0,%r12			# reassign arguments
 	mov	$out,%r13
 	mov	$len,%r14
-	mov	$key,%r15
+	lea	112($key),%r15			# size optimization
 	movdqu	($ivp),$iv			# load IV
 	mov	$ivp,88(%rsp)			# save $ivp
 ___
-my ($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
+($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
 my $rounds="${ivp}d";
 $code.=<<___;
 	shl	\$6,$len
 	sub	$in0,$out
-	mov	240($key),$rounds
+	mov	240-112($key),$rounds
 	add	$inp,$len		# end of input
 
 	lea	K_XX_XX(%rip),$K_XX_XX
@@ -180,19 +226,22 @@ $code.=<<___;
 	mov	12($ctx),$D
 	mov	$B,@T[0]		# magic seed
 	mov	16($ctx),$E
+	mov	$C,@T[1]
+	xor	$D,@T[1]
+	and	@T[1],@T[0]
 
-	movdqa	64($K_XX_XX),@X[2]	# pbswap mask
+	movdqa	64($K_XX_XX),@Tx[2]	# pbswap mask
 	movdqa	0($K_XX_XX),@Tx[1]	# K_00_19
 	movdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
 	movdqu	16($inp),@X[-3&7]
 	movdqu	32($inp),@X[-2&7]
 	movdqu	48($inp),@X[-1&7]
-	pshufb	@X[2],@X[-4&7]		# byte swap
+	pshufb	@Tx[2],@X[-4&7]		# byte swap
+	pshufb	@Tx[2],@X[-3&7]
+	pshufb	@Tx[2],@X[-2&7]
 	add	\$64,$inp
-	pshufb	@X[2],@X[-3&7]
-	pshufb	@X[2],@X[-2&7]
-	pshufb	@X[2],@X[-1&7]
 	paddd	@Tx[1],@X[-4&7]		# add K_00_19
+	pshufb	@Tx[2],@X[-1&7]
 	paddd	@Tx[1],@X[-3&7]
 	paddd	@Tx[1],@X[-2&7]
 	movdqa	@X[-4&7],0(%rsp)	# X[]+K xfer to IALU
@@ -201,8 +250,8 @@ $code.=<<___;
 	psubd	@Tx[1],@X[-3&7]
 	movdqa	@X[-2&7],32(%rsp)
 	psubd	@Tx[1],@X[-2&7]
-	movups	($key),$rndkey0		# $key[0]
-	movups	16($key),$rndkey[0]	# forward reference
+	movups	-112($key),$rndkey0	# $key[0]
+	movups	16-112($key),$rndkey[0]	# forward reference
 	jmp	.Loop_ssse3
 ___
 
@@ -219,31 +268,31 @@ ___
 ___
       $code.=<<___;
 	xorps		$in,$iv
+	movups		`32+16*$k-112`($key),$rndkey[1]
 	aesenc		$rndkey[0],$iv
-	movups		`32+16*$k`($key),$rndkey[1]
 ___
     } elsif ($k==9) {
       $sn++;
       $code.=<<___;
 	cmp		\$11,$rounds
 	jb		.Laesenclast$sn
-	movups		`32+16*($k+0)`($key),$rndkey[1]
+	movups		`32+16*($k+0)-112`($key),$rndkey[1]
 	aesenc		$rndkey[0],$iv
-	movups		`32+16*($k+1)`($key),$rndkey[0]
+	movups		`32+16*($k+1)-112`($key),$rndkey[0]
 	aesenc		$rndkey[1],$iv
 	je		.Laesenclast$sn
-	movups		`32+16*($k+2)`($key),$rndkey[1]
+	movups		`32+16*($k+2)-112`($key),$rndkey[1]
 	aesenc		$rndkey[0],$iv
-	movups		`32+16*($k+3)`($key),$rndkey[0]
+	movups		`32+16*($k+3)-112`($key),$rndkey[0]
 	aesenc		$rndkey[1],$iv
 .Laesenclast$sn:
 	aesenclast	$rndkey[0],$iv
-	movups		16($key),$rndkey[1]		# forward reference
+	movups		16-112($key),$rndkey[1]		# forward reference
 ___
     } else {
       $code.=<<___;
+	movups		`32+16*$k-112`($key),$rndkey[1]
 	aesenc		$rndkey[0],$iv
-	movups		`32+16*$k`($key),$rndkey[1]
 ___
     }
     $r++;	unshift(@rndkey,pop(@rndkey));
@@ -255,61 +304,61 @@ sub Xupdate_ssse3_16_31()		# recall that $Xi starts wtih 4
   my @insns = (&$body,&$body,&$body,&$body);	# 40 instructions
   my ($a,$b,$c,$d,$e);
 
-	&movdqa	(@X[0],@X[-3&7]);
-	 eval(shift(@insns));
+	 eval(shift(@insns));		# ror
+	&pshufd	(@X[0],@X[-4&7],0xee);	# was &movdqa	(@X[0],@X[-3&7]);
 	 eval(shift(@insns));
 	&movdqa	(@Tx[0],@X[-1&7]);
-	&palignr(@X[0],@X[-4&7],8);	# compose "X[-14]" in "X[0]"
+	  &paddd	(@Tx[1],@X[-1&7]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	  &paddd	(@Tx[1],@X[-1&7]);
+	&punpcklqdq(@X[0],@X[-3&7]);	# compose "X[-14]" in "X[0]", was &palignr(@X[0],@X[-4&7],8);
 	 eval(shift(@insns));
+	 eval(shift(@insns));		# rol
 	 eval(shift(@insns));
 	&psrldq	(@Tx[0],4);		# "X[-3]", 3 dwords
 	 eval(shift(@insns));
 	 eval(shift(@insns));
+
 	&pxor	(@X[0],@X[-4&7]);	# "X[0]"^="X[-16]"
 	 eval(shift(@insns));
-	 eval(shift(@insns));
-
+	 eval(shift(@insns));		# ror
 	&pxor	(@Tx[0],@X[-2&7]);	# "X[-3]"^"X[-8]"
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	 eval(shift(@insns));
 
 	&pxor	(@X[0],@Tx[0]);		# "X[0]"^="X[-3]"^"X[-8]"
 	 eval(shift(@insns));
-	 eval(shift(@insns));
+	 eval(shift(@insns));		# rol
 	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
 	&movdqa	(@Tx[2],@X[0]);
-	&movdqa	(@Tx[0],@X[0]);
-	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
+	 eval(shift(@insns));		# ror
+	&movdqa	(@Tx[0],@X[0]);
 	 eval(shift(@insns));
 
 	&pslldq	(@Tx[2],12);		# "X[0]"<<96, extract one dword
 	&paddd	(@X[0],@X[0]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	 eval(shift(@insns));
-	 eval(shift(@insns));
 
 	&psrld	(@Tx[0],31);
 	 eval(shift(@insns));
+	 eval(shift(@insns));		# rol
 	 eval(shift(@insns));
 	&movdqa	(@Tx[1],@Tx[2]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
 	&psrld	(@Tx[2],30);
-	&por	(@X[0],@Tx[0]);		# "X[0]"<<<=1
 	 eval(shift(@insns));
+	 eval(shift(@insns));		# ror
+	&por	(@X[0],@Tx[0]);		# "X[0]"<<<=1
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
@@ -317,12 +366,13 @@ sub Xupdate_ssse3_16_31()		# recall that $Xi starts wtih 4
 	&pslld	(@Tx[1],2);
 	&pxor	(@X[0],@Tx[2]);
 	 eval(shift(@insns));
-	 eval(shift(@insns));
 	  &movdqa	(@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)");	# K_XX_XX
+	 eval(shift(@insns));		# rol
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
 	&pxor	(@X[0],@Tx[1]);		# "X[0]"^=("X[0]">>96)<<<2
+	&pshufd (@Tx[1],@X[-1&7],0xee)	if ($Xi==7);	# was &movdqa	(@Tx[0],@X[-1&7]) in Xupdate_ssse3_32_79
 
 	 foreach (@insns) { eval; }	# remaining instructions [if any]
 
@@ -333,27 +383,30 @@ sub Xupdate_ssse3_16_31()		# recall that $Xi starts wtih 4
 sub Xupdate_ssse3_32_79()
 { use integer;
   my $body = shift;
-  my @insns = (&$body,&$body,&$body,&$body);	# 32 to 48 instructions
+  my @insns = (&$body,&$body,&$body,&$body);	# 32 to 44 instructions
   my ($a,$b,$c,$d,$e);
 
-	&movdqa	(@Tx[0],@X[-1&7])	if ($Xi==8);
-	 eval(shift(@insns));		# body_20_39
+	 eval(shift(@insns))		if ($Xi==8);
 	&pxor	(@X[0],@X[-4&7]);	# "X[0]"="X[-32]"^"X[-16]"
-	&palignr(@Tx[0],@X[-2&7],8);	# compose "X[-6]"
+	 eval(shift(@insns))		if ($Xi==8);
+	 eval(shift(@insns));		# body_20_39
 	 eval(shift(@insns));
+	 eval(shift(@insns))		if (@insns[1] =~ /_ror/);
+	 eval(shift(@insns))		if (@insns[0] =~ /_ror/);
+	&punpcklqdq(@Tx[0],@X[-1&7]);	# compose "X[-6]", was &palignr(@Tx[0],@X[-2&7],8);
 	 eval(shift(@insns));
 	 eval(shift(@insns));		# rol
 
 	&pxor	(@X[0],@X[-7&7]);	# "X[0]"^="X[-28]"
 	 eval(shift(@insns));
-	 eval(shift(@insns))	if (@insns[0] !~ /&ro[rl]/);
+	 eval(shift(@insns));
 	if ($Xi%5) {
 	  &movdqa	(@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
 	} else {			# ... or load next one
 	  &movdqa	(@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
 	}
-	  &paddd	(@Tx[1],@X[-1&7]);
 	 eval(shift(@insns));		# ror
+	  &paddd	(@Tx[1],@X[-1&7]);
 	 eval(shift(@insns));
 
 	&pxor	(@X[0],@Tx[0]);		# "X[0]"^="X[-6]"
@@ -361,29 +414,31 @@ sub Xupdate_ssse3_32_79()
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));		# rol
+	 eval(shift(@insns))		if (@insns[0] =~ /_ror/);
 
 	&movdqa	(@Tx[0],@X[0]);
-	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
 	 eval(shift(@insns));
 	 eval(shift(@insns));
+	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
 	 eval(shift(@insns));		# ror
 	 eval(shift(@insns));
+	 eval(shift(@insns));		# body_20_39
 
 	&pslld	(@X[0],2);
-	 eval(shift(@insns));		# body_20_39
 	 eval(shift(@insns));
-	&psrld	(@Tx[0],30);
 	 eval(shift(@insns));
-	 eval(shift(@insns));		# rol
+	&psrld	(@Tx[0],30);
+	 eval(shift(@insns))		if (@insns[0] =~ /_rol/);# rol
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));		# ror
-	 eval(shift(@insns));
 
 	&por	(@X[0],@Tx[0]);		# "X[0]"<<<=2
-	 eval(shift(@insns));		# body_20_39
 	 eval(shift(@insns));
-	  &movdqa	(@Tx[1],@X[0])	if ($Xi<19);
+	 eval(shift(@insns));		# body_20_39
+	 eval(shift(@insns))		if (@insns[1] =~ /_rol/);
+	 eval(shift(@insns))		if (@insns[0] =~ /_rol/);
+	  &pshufd(@Tx[1],@X[-1&7],0xee)	if ($Xi<19);	# was &movdqa	(@Tx[1],@X[0])
 	 eval(shift(@insns));
 	 eval(shift(@insns));		# rol
 	 eval(shift(@insns));
@@ -404,10 +459,11 @@ sub Xuplast_ssse3_80()
   my ($a,$b,$c,$d,$e);
 
 	 eval(shift(@insns));
-	  &paddd	(@Tx[1],@X[-1&7]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
+	  &paddd	(@Tx[1],@X[-1&7]);
+	 eval(shift(@insns));
 	 eval(shift(@insns));
 
 	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer IALU
@@ -415,17 +471,17 @@ sub Xuplast_ssse3_80()
 	 foreach (@insns) { eval; }		# remaining instructions
 
 	&cmp	($inp,$len);
-	&je	(".Ldone_ssse3");
+	&je	(shift);
 
 	unshift(@Tx,pop(@Tx));
 
-	&movdqa	(@X[2],"64($K_XX_XX)");		# pbswap mask
+	&movdqa	(@Tx[2],"64($K_XX_XX)");	# pbswap mask
 	&movdqa	(@Tx[1],"0($K_XX_XX)");		# K_00_19
 	&movdqu	(@X[-4&7],"0($inp)");		# load input
 	&movdqu	(@X[-3&7],"16($inp)");
 	&movdqu	(@X[-2&7],"32($inp)");
 	&movdqu	(@X[-1&7],"48($inp)");
-	&pshufb	(@X[-4&7],@X[2]);		# byte swap
+	&pshufb	(@X[-4&7],@Tx[2]);		# byte swap
 	&add	($inp,64);
 
   $Xi=0;
@@ -439,7 +495,10 @@ sub Xloop_ssse3()
 
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	&pshufb	(@X[($Xi-3)&7],@X[2]);
+	 eval(shift(@insns));
+	&pshufb	(@X[($Xi-3)&7],@Tx[2]);
+	 eval(shift(@insns));
+	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	&paddd	(@X[($Xi-4)&7],@Tx[1]);
@@ -450,6 +509,8 @@ sub Xloop_ssse3()
 	&movdqa	(eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]);	# X[]+K xfer to IALU
 	 eval(shift(@insns));
 	 eval(shift(@insns));
+	 eval(shift(@insns));
+	 eval(shift(@insns));
 	&psubd	(@X[($Xi-4)&7],@Tx[1]);
 
 	foreach (@insns) { eval; }
@@ -465,76 +526,106 @@ sub Xtail_ssse3()
 	foreach (@insns) { eval; }
 }
 
-sub body_00_19 () {
-  use integer;
-  my ($k,$n);
-  my @r=(
+my @body_00_19 = (
 	'($a,$b,$c,$d,$e)=@V;'.
-	'&add	($e,eval(4*($j&15))."(%rsp)");',	# X[]+K xfer
-	'&xor	($c,$d);',
-	'&mov	(@T[1],$a);',	# $b in next round
-	'&$_rol	($a,5);',
-	'&and	(@T[0],$c);',	# ($b&($c^$d))
-	'&xor	($c,$d);',	# restore $c
-	'&xor	(@T[0],$d);',
-	'&add	($e,$a);',
 	'&$_ror	($b,$j?7:2);',	# $b>>>2
-	'&add	($e,@T[0]);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
+	'&xor	(@T[0],$d);',
+	'&mov	(@T[1],$a);',	# $b for next round
+
+	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
+	'&xor	($b,$c);',	# $c^$d for next round
+
+	'&$_rol	($a,5);',
+	'&add	($e,@T[0]);',
+	'&and	(@T[1],$b);',	# ($b&($c^$d)) for next round
+
+	'&xor	($b,$c);',	# restore $b
+	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
 	);
+
+sub body_00_19 () {	# ((c^d)&b)^d
+    # on start @T[0]=(c^d)&b
+    return &body_20_39() if ($rx==19); $rx++;
+
+    use integer;
+    my ($k,$n);
+    my @r=@body_00_19;
+
 	$n = scalar(@r);
 	$k = (($jj+1)*12/20)*20*$n/12;	# 12 aesencs per these 20 rounds
 	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n);
 	$jj++;
+
     return @r;
 }
 
-sub body_20_39 () {
-  use integer;
-  my ($k,$n);
-  my @r=(
+my @body_20_39 = (
 	'($a,$b,$c,$d,$e)=@V;'.
-	'&add	($e,eval(4*($j++&15))."(%rsp)");',	# X[]+K xfer
-	'&xor	(@T[0],$d);',	# ($b^$d)
-	'&mov	(@T[1],$a);',	# $b in next round
+	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
+	'&xor	(@T[0],$d)	if($j==19);'.
+	'&xor	(@T[0],$c)	if($j> 19);',	# ($b^$d^$c)
+	'&mov	(@T[1],$a);',	# $b for next round
+
 	'&$_rol	($a,5);',
-	'&xor	(@T[0],$c);',	# ($b^$d^$c)
-	'&add	($e,$a);',
+	'&add	($e,@T[0]);',
+	'&xor	(@T[1],$c)	if ($j< 79);',	# $b^$d for next round
+
 	'&$_ror	($b,7);',	# $b>>>2
-	'&add	($e,@T[0]);'	.'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
+	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
 	);
+
+sub body_20_39 () {	# b^d^c
+    # on entry @T[0]=b^d
+    return &body_40_59() if ($rx==39); $rx++;
+
+    use integer;
+    my ($k,$n);
+    my @r=@body_20_39;
+
 	$n = scalar(@r);
 	$k = (($jj+1)*8/20)*20*$n/8;	# 8 aesencs per these 20 rounds
-	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n);
+	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n && $rx!=20);
 	$jj++;
+
     return @r;
 }
 
-sub body_40_59 () {
-  use integer;
-  my ($k,$n);
-  my @r=(
+my @body_40_59 = (
 	'($a,$b,$c,$d,$e)=@V;'.
-	'&mov	(@T[1],$c);',
-	'&xor	($c,$d);',
-	'&add	($e,eval(4*($j++&15))."(%rsp)");',	# X[]+K xfer
-	'&and	(@T[1],$d);',
-	'&and	(@T[0],$c);',	# ($b&($c^$d))
+	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
+	'&and	(@T[0],$c)	if ($j>=40);',	# (b^c)&(c^d)
+	'&xor	($c,$d)		if ($j>=40);',	# restore $c
+
 	'&$_ror	($b,7);',	# $b>>>2
-	'&add	($e,@T[1]);',
-	'&mov	(@T[1],$a);',	# $b in next round
+	'&mov	(@T[1],$a);',	# $b for next round
+	'&xor	(@T[0],$c);',
+
 	'&$_rol	($a,5);',
 	'&add	($e,@T[0]);',
-	'&xor	($c,$d);',	# restore $c
-	'&add	($e,$a);'	.'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
+	'&xor	(@T[1],$c)	if ($j==59);'.
+	'&xor	(@T[1],$b)	if ($j< 59);',	# b^c for next round
+
+	'&xor	($b,$c)		if ($j< 59);',	# c^d for next round
+	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
 	);
+
+sub body_40_59 () {	# ((b^c)&(c^d))^c
+    # on entry @T[0]=(b^c), (c^=d)
+    $rx++;
+
+    use integer;
+    my ($k,$n);
+    my @r=@body_40_59;
+
 	$n = scalar(@r);
 	$k=(($jj+1)*12/20)*20*$n/12;	# 12 aesencs per these 20 rounds
-	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n);
+	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n && $rx!=40);
 	$jj++;
+
     return @r;
 }
 $code.=<<___;
-.align	16
+.align	32
 .Loop_ssse3:
 ___
 	&Xupdate_ssse3_16_31(\&body_00_19);
@@ -553,7 +644,7 @@ ___
 	&Xupdate_ssse3_32_79(\&body_40_59);
 	&Xupdate_ssse3_32_79(\&body_40_59);
 	&Xupdate_ssse3_32_79(\&body_20_39);
-	&Xuplast_ssse3_80(\&body_20_39);	# can jump to "done"
+	&Xuplast_ssse3_80(\&body_20_39,".Ldone_ssse3");	# can jump to "done"
 
 				$saved_j=$j; @saved_V=@V;
 				$saved_r=$r; @saved_rndkey=@rndkey;
@@ -575,11 +666,13 @@ $code.=<<___;
 	mov	@T[0],4($ctx)
 	mov	@T[0],$B			# magic seed
 	mov	$C,8($ctx)
+	mov	$C,@T[1]
 	mov	$D,12($ctx)
+	xor	$D,@T[1]
 	mov	$E,16($ctx)
+	and	@T[1],@T[0]
 	jmp	.Loop_ssse3
 
-.align	16
 .Ldone_ssse3:
 ___
 				$jj=$j=$saved_j; @V=@saved_V;
@@ -631,7 +724,278 @@ $code.=<<___;
 .size	aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
 ___
 
-$j=$jj=$r=$sn=0;
+						if ($stitched_decrypt) {{{
+# reset
+($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+$j=$jj=$r=$rx=0;
+$Xi=4;
+
+# reassign for Atom Silvermont (see above)
+($inout0,$inout1,$inout2,$inout3,$rndkey0)=map("%xmm$_",(0..4));
+@X=map("%xmm$_",(8..13,6,7));
+@Tx=map("%xmm$_",(14,15,5));
+
+my @aes256_dec = (
+	'&movdqu($inout0,"0x00($in0)");',
+	'&movdqu($inout1,"0x10($in0)");	&pxor	($inout0,$rndkey0);',
+	'&movdqu($inout2,"0x20($in0)");	&pxor	($inout1,$rndkey0);',
+	'&movdqu($inout3,"0x30($in0)");	&pxor	($inout2,$rndkey0);',
+
+	'&pxor	($inout3,$rndkey0);	&movups	($rndkey0,"16-112($key)");',
+	'&movaps("64(%rsp)",@X[2]);',	# save IV, originally @X[3]
+	undef,undef
+	);
+for ($i=0;$i<13;$i++) {
+    push (@aes256_dec,(
+	'&aesdec	($inout0,$rndkey0);',
+	'&aesdec	($inout1,$rndkey0);',
+	'&aesdec	($inout2,$rndkey0);',
+	'&aesdec	($inout3,$rndkey0);	&movups($rndkey0,"'.(16*($i+2)-112).'($key)");'
+	));
+    push (@aes256_dec,(undef,undef))	if (($i>=3 && $i<=5) || $i>=11);
+    push (@aes256_dec,(undef,undef))	if ($i==5);
+}
+push(@aes256_dec,(
+	'&aesdeclast	($inout0,$rndkey0);	&movups	(@X[0],"0x00($in0)");',
+	'&aesdeclast	($inout1,$rndkey0);	&movups	(@X[1],"0x10($in0)");',
+	'&aesdeclast	($inout2,$rndkey0);	&movups	(@X[2],"0x20($in0)");',
+	'&aesdeclast	($inout3,$rndkey0);	&movups	(@X[3],"0x30($in0)");',
+
+	'&xorps		($inout0,"64(%rsp)");	&movdqu	($rndkey0,"-112($key)");',
+	'&xorps		($inout1,@X[0]);	&movups	("0x00($out,$in0)",$inout0);',
+	'&xorps		($inout2,@X[1]);	&movups	("0x10($out,$in0)",$inout1);',
+	'&xorps		($inout3,@X[2]);	&movups	("0x20($out,$in0)",$inout2);',
+
+	'&movups	("0x30($out,$in0)",$inout3);'
+	));
+
+sub body_00_19_dec () {	# ((c^d)&b)^d
+    # on start @T[0]=(c^d)&b
+    return &body_20_39_dec() if ($rx==19);
+
+    my @r=@body_00_19;
+
+	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
+	$rx++;
+
+    return @r;
+}
+
+sub body_20_39_dec () {	# b^d^c
+    # on entry @T[0]=b^d
+    return &body_40_59_dec() if ($rx==39);
+  
+    my @r=@body_20_39;
+
+	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
+	$rx++;
+
+    return @r;
+}
+
+sub body_40_59_dec () {	# ((b^c)&(c^d))^c
+    # on entry @T[0]=(b^c), (c^=d)
+
+    my @r=@body_40_59;
+
+	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
+	$rx++;
+
+    return @r;
+}
+
+$code.=<<___;
+.globl	aesni256_cbc_sha1_dec
+.type	aesni256_cbc_sha1_dec,\@abi-omnipotent
+.align	32
+aesni256_cbc_sha1_dec:
+	# caller should check for SSSE3 and AES-NI bits
+	mov	OPENSSL_ia32cap_P+0(%rip),%r10d
+	mov	OPENSSL_ia32cap_P+4(%rip),%r11d
+___
+$code.=<<___ if ($avx);
+	and	\$`1<<28`,%r11d		# mask AVX bit
+	and	\$`1<<30`,%r10d		# mask "Intel CPU" bit
+	or	%r11d,%r10d
+	cmp	\$`1<<28|1<<30`,%r10d
+	je	aesni256_cbc_sha1_dec_avx
+___
+$code.=<<___;
+	jmp	aesni256_cbc_sha1_dec_ssse3
+	ret
+.size	aesni256_cbc_sha1_dec,.-aesni256_cbc_sha1_dec
+
+.type	aesni256_cbc_sha1_dec_ssse3,\@function,6
+.align	32
+aesni256_cbc_sha1_dec_ssse3:
+	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,96+0(%rsp)
+	movaps	%xmm7,96+16(%rsp)
+	movaps	%xmm8,96+32(%rsp)
+	movaps	%xmm9,96+48(%rsp)
+	movaps	%xmm10,96+64(%rsp)
+	movaps	%xmm11,96+80(%rsp)
+	movaps	%xmm12,96+96(%rsp)
+	movaps	%xmm13,96+112(%rsp)
+	movaps	%xmm14,96+128(%rsp)
+	movaps	%xmm15,96+144(%rsp)
+.Lprologue_dec_ssse3:
+___
+$code.=<<___;
+	mov	$in0,%r12			# reassign arguments
+	mov	$out,%r13
+	mov	$len,%r14
+	lea	112($key),%r15			# size optimization
+	movdqu	($ivp),@X[3]			# load IV
+	#mov	$ivp,88(%rsp)			# save $ivp
+___
+($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
+$code.=<<___;
+	shl	\$6,$len
+	sub	$in0,$out
+	add	$inp,$len		# end of input
+
+	lea	K_XX_XX(%rip),$K_XX_XX
+	mov	0($ctx),$A		# load context
+	mov	4($ctx),$B
+	mov	8($ctx),$C
+	mov	12($ctx),$D
+	mov	$B,@T[0]		# magic seed
+	mov	16($ctx),$E
+	mov	$C,@T[1]
+	xor	$D,@T[1]
+	and	@T[1],@T[0]
+
+	movdqa	64($K_XX_XX),@Tx[2]	# pbswap mask
+	movdqa	0($K_XX_XX),@Tx[1]	# K_00_19
+	movdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
+	movdqu	16($inp),@X[-3&7]
+	movdqu	32($inp),@X[-2&7]
+	movdqu	48($inp),@X[-1&7]
+	pshufb	@Tx[2],@X[-4&7]		# byte swap
+	add	\$64,$inp
+	pshufb	@Tx[2],@X[-3&7]
+	pshufb	@Tx[2],@X[-2&7]
+	pshufb	@Tx[2],@X[-1&7]
+	paddd	@Tx[1],@X[-4&7]		# add K_00_19
+	paddd	@Tx[1],@X[-3&7]
+	paddd	@Tx[1],@X[-2&7]
+	movdqa	@X[-4&7],0(%rsp)	# X[]+K xfer to IALU
+	psubd	@Tx[1],@X[-4&7]		# restore X[]
+	movdqa	@X[-3&7],16(%rsp)
+	psubd	@Tx[1],@X[-3&7]
+	movdqa	@X[-2&7],32(%rsp)
+	psubd	@Tx[1],@X[-2&7]
+	movdqu	-112($key),$rndkey0	# $key[0]
+	jmp	.Loop_dec_ssse3
+
+.align	32
+.Loop_dec_ssse3:
+___
+	&Xupdate_ssse3_16_31(\&body_00_19_dec);
+	&Xupdate_ssse3_16_31(\&body_00_19_dec);
+	&Xupdate_ssse3_16_31(\&body_00_19_dec);
+	&Xupdate_ssse3_16_31(\&body_00_19_dec);
+	&Xupdate_ssse3_32_79(\&body_00_19_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xupdate_ssse3_32_79(\&body_40_59_dec);
+	&Xupdate_ssse3_32_79(\&body_40_59_dec);
+	&Xupdate_ssse3_32_79(\&body_40_59_dec);
+	&Xupdate_ssse3_32_79(\&body_40_59_dec);
+	&Xupdate_ssse3_32_79(\&body_40_59_dec);
+	&Xupdate_ssse3_32_79(\&body_20_39_dec);
+	&Xuplast_ssse3_80(\&body_20_39_dec,".Ldone_dec_ssse3");	# can jump to "done"
+
+				$saved_j=$j;   @saved_V=@V;
+				$saved_rx=$rx;
+
+	&Xloop_ssse3(\&body_20_39_dec);
+	&Xloop_ssse3(\&body_20_39_dec);
+	&Xloop_ssse3(\&body_20_39_dec);
+
+	eval(@aes256_dec[-1]);			# last store
+$code.=<<___;
+	lea	64($in0),$in0
+
+	add	0($ctx),$A			# update context
+	add	4($ctx),@T[0]
+	add	8($ctx),$C
+	add	12($ctx),$D
+	mov	$A,0($ctx)
+	add	16($ctx),$E
+	mov	@T[0],4($ctx)
+	mov	@T[0],$B			# magic seed
+	mov	$C,8($ctx)
+	mov	$C,@T[1]
+	mov	$D,12($ctx)
+	xor	$D,@T[1]
+	mov	$E,16($ctx)
+	and	@T[1],@T[0]
+	jmp	.Loop_dec_ssse3
+
+.Ldone_dec_ssse3:
+___
+				$jj=$j=$saved_j; @V=@saved_V;
+				$rx=$saved_rx;
+
+	&Xtail_ssse3(\&body_20_39_dec);
+	&Xtail_ssse3(\&body_20_39_dec);
+	&Xtail_ssse3(\&body_20_39_dec);
+
+	eval(@aes256_dec[-1]);			# last store
+$code.=<<___;
+	add	0($ctx),$A			# update context
+	add	4($ctx),@T[0]
+	add	8($ctx),$C
+	mov	$A,0($ctx)
+	add	12($ctx),$D
+	mov	@T[0],4($ctx)
+	add	16($ctx),$E
+	mov	$C,8($ctx)
+	mov	$D,12($ctx)
+	mov	$E,16($ctx)
+	movups	@X[3],($ivp)			# write IV
+___
+$code.=<<___ if ($win64);
+	movaps	96+0(%rsp),%xmm6
+	movaps	96+16(%rsp),%xmm7
+	movaps	96+32(%rsp),%xmm8
+	movaps	96+48(%rsp),%xmm9
+	movaps	96+64(%rsp),%xmm10
+	movaps	96+80(%rsp),%xmm11
+	movaps	96+96(%rsp),%xmm12
+	movaps	96+112(%rsp),%xmm13
+	movaps	96+128(%rsp),%xmm14
+	movaps	96+144(%rsp),%xmm15
+___
+$code.=<<___;
+	lea	`104+($win64?10*16:0)`(%rsp),%rsi
+	mov	0(%rsi),%r15
+	mov	8(%rsi),%r14
+	mov	16(%rsi),%r13
+	mov	24(%rsi),%r12
+	mov	32(%rsi),%rbp
+	mov	40(%rsi),%rbx
+	lea	48(%rsi),%rsp
+.Lepilogue_dec_ssse3:
+	ret
+.size	aesni256_cbc_sha1_dec_ssse3,.-aesni256_cbc_sha1_dec_ssse3
+___
+						}}}
+$j=$jj=$r=$rx=0;
 
 if ($avx) {
 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
@@ -641,13 +1005,17 @@ my @X=map("%xmm$_",(4..7,0..3));
 my @Tx=map("%xmm$_",(8..10));
 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp");	# size optimization
 my @T=("%esi","%edi");
+my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13));
+my @rndkey=("%xmm14","%xmm15");
+my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15));	# for dec
+my $Kx=@Tx[2];
 
 my $_rol=sub { &shld(@_[0],@_) };
 my $_ror=sub { &shrd(@_[0],@_) };
 
 $code.=<<___;
 .type	aesni_cbc_sha1_enc_avx,\@function,6
-.align	16
+.align	32
 aesni_cbc_sha1_enc_avx:
 	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
 	#shr	\$6,$len			# debugging artefact
@@ -680,17 +1048,16 @@ $code.=<<___;
 	mov	$in0,%r12			# reassign arguments
 	mov	$out,%r13
 	mov	$len,%r14
-	mov	$key,%r15
+	lea	112($key),%r15			# size optimization
 	vmovdqu	($ivp),$iv			# load IV
 	mov	$ivp,88(%rsp)			# save $ivp
 ___
-my ($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
+($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
 my $rounds="${ivp}d";
 $code.=<<___;
 	shl	\$6,$len
 	sub	$in0,$out
-	mov	240($key),$rounds
-	add	\$112,$key		# size optimization
+	mov	240-112($key),$rounds
 	add	$inp,$len		# end of input
 
 	lea	K_XX_XX(%rip),$K_XX_XX
@@ -700,9 +1067,12 @@ $code.=<<___;
 	mov	12($ctx),$D
 	mov	$B,@T[0]		# magic seed
 	mov	16($ctx),$E
+	mov	$C,@T[1]
+	xor	$D,@T[1]
+	and	@T[1],@T[0]
 
 	vmovdqa	64($K_XX_XX),@X[2]	# pbswap mask
-	vmovdqa	0($K_XX_XX),@Tx[1]	# K_00_19
+	vmovdqa	0($K_XX_XX),$Kx		# K_00_19
 	vmovdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
 	vmovdqu	16($inp),@X[-3&7]
 	vmovdqu	32($inp),@X[-2&7]
@@ -712,13 +1082,13 @@ $code.=<<___;
 	vpshufb	@X[2],@X[-3&7],@X[-3&7]
 	vpshufb	@X[2],@X[-2&7],@X[-2&7]
 	vpshufb	@X[2],@X[-1&7],@X[-1&7]
-	vpaddd	@Tx[1],@X[-4&7],@X[0]	# add K_00_19
-	vpaddd	@Tx[1],@X[-3&7],@X[1]
-	vpaddd	@Tx[1],@X[-2&7],@X[2]
+	vpaddd	$Kx,@X[-4&7],@X[0]	# add K_00_19
+	vpaddd	$Kx,@X[-3&7],@X[1]
+	vpaddd	$Kx,@X[-2&7],@X[2]
 	vmovdqa	@X[0],0(%rsp)		# X[]+K xfer to IALU
 	vmovdqa	@X[1],16(%rsp)
 	vmovdqa	@X[2],32(%rsp)
-	vmovups	-112($key),$rndkey0	# $key[0]
+	vmovups	-112($key),$rndkey[1]	# $key[0]
 	vmovups	16-112($key),$rndkey[0]	# forward reference
 	jmp	.Loop_avx
 ___
@@ -728,14 +1098,14 @@ my $aesenc=sub {
   my ($n,$k)=($r/10,$r%10);
     if ($k==0) {
       $code.=<<___;
-	vmovups		`16*$n`($in0),$in		# load input
-	vxorps		$rndkey0,$in,$in
+	vmovdqu		`16*$n`($in0),$in		# load input
+	vpxor		$rndkey[1],$in,$in
 ___
       $code.=<<___ if ($n);
 	vmovups		$iv,`16*($n-1)`($out,$in0)	# write output
 ___
       $code.=<<___;
-	vxorps		$in,$iv,$iv
+	vpxor		$in,$iv,$iv
 	vaesenc		$rndkey[0],$iv,$iv
 	vmovups		`32+16*$k-112`($key),$rndkey[1]
 ___
@@ -755,6 +1125,7 @@ ___
 	vmovups		`32+16*($k+3)-112`($key),$rndkey[0]
 .Lvaesenclast$sn:
 	vaesenclast	$rndkey[0],$iv,$iv
+	vmovups		-112($key),$rndkey[0]
 	vmovups		16-112($key),$rndkey[1]		# forward reference
 ___
     } else {
@@ -778,10 +1149,10 @@ sub Xupdate_avx_16_31()		# recall that $Xi starts wtih 4
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	  &vpaddd	(@Tx[1],@Tx[1],@X[-1&7]);
+	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	&vpsrldq(@Tx[0],@X[-1&7],4);	# "X[-3]", 3 dwords
+	&vpsrldq(@Tx[0],@X[-1&7],4);		# "X[-3]", 3 dwords
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	&vpxor	(@X[0],@X[0],@X[-4&7]);		# "X[0]"^="X[-16]"
@@ -807,31 +1178,31 @@ sub Xupdate_avx_16_31()		# recall that $Xi starts wtih 4
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	&vpslldq(@Tx[2],@X[0],12);		# "X[0]"<<96, extract one dword
+	&vpslldq(@Tx[1],@X[0],12);		# "X[0]"<<96, extract one dword
 	&vpaddd	(@X[0],@X[0],@X[0]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	&vpsrld	(@Tx[1],@Tx[2],30);
 	&vpor	(@X[0],@X[0],@Tx[0]);		# "X[0]"<<<=1
+	&vpsrld	(@Tx[0],@Tx[1],30);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	&vpslld	(@Tx[2],@Tx[2],2);
-	&vpxor	(@X[0],@X[0],@Tx[1]);
+	&vpslld	(@Tx[1],@Tx[1],2);
+	&vpxor	(@X[0],@X[0],@Tx[0]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	&vpxor	(@X[0],@X[0],@Tx[2]);		# "X[0]"^=("X[0]">>96)<<<2
+	&vpxor	(@X[0],@X[0],@Tx[1]);		# "X[0]"^=("X[0]">>96)<<<2
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	  &vmovdqa	(@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)");	# K_XX_XX
+	  &vmovdqa	($Kx,eval(16*(($Xi)/5))."($K_XX_XX)")	if ($Xi%5==0);	# K_XX_XX
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
@@ -839,7 +1210,6 @@ sub Xupdate_avx_16_31()		# recall that $Xi starts wtih 4
 	 foreach (@insns) { eval; }	# remaining instructions [if any]
 
   $Xi++;	push(@X,shift(@X));	# "rotate" X[]
-		push(@Tx,shift(@Tx));
 }
 
 sub Xupdate_avx_32_79()
@@ -858,12 +1228,8 @@ sub Xupdate_avx_32_79()
 	&vpxor	(@X[0],@X[0],@X[-7&7]);		# "X[0]"^="X[-28]"
 	 eval(shift(@insns));
 	 eval(shift(@insns))	if (@insns[0] !~ /&ro[rl]/);
-	if ($Xi%5) {
-	  &vmovdqa	(@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
-	} else {			# ... or load next one
-	  &vmovdqa	(@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
-	}
-	  &vpaddd	(@Tx[1],@Tx[1],@X[-1&7]);
+	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
+	  &vmovdqa	($Kx,eval(16*($Xi/5))."($K_XX_XX)")	if ($Xi%5==0);
 	 eval(shift(@insns));		# ror
 	 eval(shift(@insns));
 
@@ -893,7 +1259,6 @@ sub Xupdate_avx_32_79()
 	&vpor	(@X[0],@X[0],@Tx[0]);		# "X[0]"<<<=2
 	 eval(shift(@insns));		# body_20_39
 	 eval(shift(@insns));
-	  &vmovdqa	(@Tx[1],@X[0])	if ($Xi<19);
 	 eval(shift(@insns));
 	 eval(shift(@insns));		# rol
 	 eval(shift(@insns));
@@ -904,7 +1269,6 @@ sub Xupdate_avx_32_79()
 	 foreach (@insns) { eval; }	# remaining instructions
 
   $Xi++;	push(@X,shift(@X));	# "rotate" X[]
-		push(@Tx,shift(@Tx));
 }
 
 sub Xuplast_avx_80()
@@ -914,28 +1278,26 @@ sub Xuplast_avx_80()
   my ($a,$b,$c,$d,$e);
 
 	 eval(shift(@insns));
-	  &vpaddd	(@Tx[1],@Tx[1],@X[-1&7]);
+	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
-	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer IALU
+	  &vmovdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer IALU
 
 	 foreach (@insns) { eval; }		# remaining instructions
 
 	&cmp	($inp,$len);
-	&je	(".Ldone_avx");
+	&je	(shift);
 
-	unshift(@Tx,pop(@Tx));
-
-	&vmovdqa(@X[2],"64($K_XX_XX)");		# pbswap mask
-	&vmovdqa(@Tx[1],"0($K_XX_XX)");		# K_00_19
+	&vmovdqa(@Tx[1],"64($K_XX_XX)");	# pbswap mask
+	&vmovdqa($Kx,"0($K_XX_XX)");		# K_00_19
 	&vmovdqu(@X[-4&7],"0($inp)");		# load input
 	&vmovdqu(@X[-3&7],"16($inp)");
 	&vmovdqu(@X[-2&7],"32($inp)");
 	&vmovdqu(@X[-1&7],"48($inp)");
-	&vpshufb(@X[-4&7],@X[-4&7],@X[2]);	# byte swap
+	&vpshufb(@X[-4&7],@X[-4&7],@Tx[1]);	# byte swap
 	&add	($inp,64);
 
   $Xi=0;
@@ -949,15 +1311,15 @@ sub Xloop_avx()
 
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	&vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
+	&vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@Tx[1]);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	&vpaddd	(@X[$Xi&7],@X[($Xi-4)&7],@Tx[1]);
+	&vpaddd	(@Tx[0],@X[($Xi-4)&7],$Kx);
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 	 eval(shift(@insns));
-	&vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]);	# X[]+K xfer to IALU
+	&vmovdqa(eval(16*$Xi)."(%rsp)",@Tx[0]);	# X[]+K xfer to IALU
 	 eval(shift(@insns));
 	 eval(shift(@insns));
 
@@ -975,7 +1337,7 @@ sub Xtail_avx()
 }
 
 $code.=<<___;
-.align	16
+.align	32
 .Loop_avx:
 ___
 	&Xupdate_avx_16_31(\&body_00_19);
@@ -994,7 +1356,7 @@ ___
 	&Xupdate_avx_32_79(\&body_40_59);
 	&Xupdate_avx_32_79(\&body_40_59);
 	&Xupdate_avx_32_79(\&body_20_39);
-	&Xuplast_avx_80(\&body_20_39);	# can jump to "done"
+	&Xuplast_avx_80(\&body_20_39,".Ldone_avx");	# can jump to "done"
 
 				$saved_j=$j; @saved_V=@V;
 				$saved_r=$r; @saved_rndkey=@rndkey;
@@ -1016,11 +1378,13 @@ $code.=<<___;
 	mov	@T[0],4($ctx)
 	mov	@T[0],$B			# magic seed
 	mov	$C,8($ctx)
+	mov	$C,@T[1]
 	mov	$D,12($ctx)
+	xor	$D,@T[1]
 	mov	$E,16($ctx)
+	and	@T[1],@T[0]
 	jmp	.Loop_avx
 
-.align	16
 .Ldone_avx:
 ___
 				$jj=$j=$saved_j; @V=@saved_V;
@@ -1072,6 +1436,218 @@ $code.=<<___;
 	ret
 .size	aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
 ___
+
+						if ($stitched_decrypt) {{{
+# reset
+($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+
+$j=$jj=$r=$rx=0;
+$Xi=4;
+
+@aes256_dec = (
+	'&vpxor	($inout0,$rndkey0,"0x00($in0)");',
+	'&vpxor	($inout1,$rndkey0,"0x10($in0)");',
+	'&vpxor	($inout2,$rndkey0,"0x20($in0)");',
+	'&vpxor	($inout3,$rndkey0,"0x30($in0)");',
+
+	'&vmovups($rndkey0,"16-112($key)");',
+	'&vmovups("64(%rsp)",@X[2]);',		# save IV, originally @X[3]
+	undef,undef
+	);
+for ($i=0;$i<13;$i++) {
+    push (@aes256_dec,(
+	'&vaesdec	($inout0,$inout0,$rndkey0);',
+	'&vaesdec	($inout1,$inout1,$rndkey0);',
+	'&vaesdec	($inout2,$inout2,$rndkey0);',
+	'&vaesdec	($inout3,$inout3,$rndkey0);	&vmovups($rndkey0,"'.(16*($i+2)-112).'($key)");'
+	));
+    push (@aes256_dec,(undef,undef))	if (($i>=3 && $i<=5) || $i>=11);
+    push (@aes256_dec,(undef,undef))	if ($i==5);
+}
+push(@aes256_dec,(
+	'&vaesdeclast	($inout0,$inout0,$rndkey0);	&vmovups(@X[0],"0x00($in0)");',
+	'&vaesdeclast	($inout1,$inout1,$rndkey0);	&vmovups(@X[1],"0x10($in0)");',
+	'&vaesdeclast	($inout2,$inout2,$rndkey0);	&vmovups(@X[2],"0x20($in0)");',
+	'&vaesdeclast	($inout3,$inout3,$rndkey0);	&vmovups(@X[3],"0x30($in0)");',
+
+	'&vxorps	($inout0,$inout0,"64(%rsp)");	&vmovdqu($rndkey0,"-112($key)");',
+	'&vxorps	($inout1,$inout1,@X[0]);	&vmovups("0x00($out,$in0)",$inout0);',
+	'&vxorps	($inout2,$inout2,@X[1]);	&vmovups("0x10($out,$in0)",$inout1);',
+	'&vxorps	($inout3,$inout3,@X[2]);	&vmovups("0x20($out,$in0)",$inout2);',
+
+	'&vmovups	("0x30($out,$in0)",$inout3);'
+	));
+
+$code.=<<___;
+.type	aesni256_cbc_sha1_dec_avx,\@function,6
+.align	32
+aesni256_cbc_sha1_dec_avx:
+	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,96+0(%rsp)
+	movaps	%xmm7,96+16(%rsp)
+	movaps	%xmm8,96+32(%rsp)
+	movaps	%xmm9,96+48(%rsp)
+	movaps	%xmm10,96+64(%rsp)
+	movaps	%xmm11,96+80(%rsp)
+	movaps	%xmm12,96+96(%rsp)
+	movaps	%xmm13,96+112(%rsp)
+	movaps	%xmm14,96+128(%rsp)
+	movaps	%xmm15,96+144(%rsp)
+.Lprologue_dec_avx:
+___
+$code.=<<___;
+	vzeroall
+	mov	$in0,%r12			# reassign arguments
+	mov	$out,%r13
+	mov	$len,%r14
+	lea	112($key),%r15			# size optimization
+	vmovdqu	($ivp),@X[3]			# load IV
+___
+($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
+$code.=<<___;
+	shl	\$6,$len
+	sub	$in0,$out
+	add	$inp,$len		# end of input
+
+	lea	K_XX_XX(%rip),$K_XX_XX
+	mov	0($ctx),$A		# load context
+	mov	4($ctx),$B
+	mov	8($ctx),$C
+	mov	12($ctx),$D
+	mov	$B,@T[0]		# magic seed
+	mov	16($ctx),$E
+	mov	$C,@T[1]
+	xor	$D,@T[1]
+	and	@T[1],@T[0]
+
+	vmovdqa	64($K_XX_XX),@X[2]	# pbswap mask
+	vmovdqa	0($K_XX_XX),$Kx		# K_00_19
+	vmovdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
+	vmovdqu	16($inp),@X[-3&7]
+	vmovdqu	32($inp),@X[-2&7]
+	vmovdqu	48($inp),@X[-1&7]
+	vpshufb	@X[2],@X[-4&7],@X[-4&7]	# byte swap
+	add	\$64,$inp
+	vpshufb	@X[2],@X[-3&7],@X[-3&7]
+	vpshufb	@X[2],@X[-2&7],@X[-2&7]
+	vpshufb	@X[2],@X[-1&7],@X[-1&7]
+	vpaddd	$Kx,@X[-4&7],@X[0]	# add K_00_19
+	vpaddd	$Kx,@X[-3&7],@X[1]
+	vpaddd	$Kx,@X[-2&7],@X[2]
+	vmovdqa	@X[0],0(%rsp)		# X[]+K xfer to IALU
+	vmovdqa	@X[1],16(%rsp)
+	vmovdqa	@X[2],32(%rsp)
+	vmovups	-112($key),$rndkey0	# $key[0]
+	jmp	.Loop_dec_avx
+
+.align	32
+.Loop_dec_avx:
+___
+	&Xupdate_avx_16_31(\&body_00_19_dec);
+	&Xupdate_avx_16_31(\&body_00_19_dec);
+	&Xupdate_avx_16_31(\&body_00_19_dec);
+	&Xupdate_avx_16_31(\&body_00_19_dec);
+	&Xupdate_avx_32_79(\&body_00_19_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xupdate_avx_32_79(\&body_40_59_dec);
+	&Xupdate_avx_32_79(\&body_40_59_dec);
+	&Xupdate_avx_32_79(\&body_40_59_dec);
+	&Xupdate_avx_32_79(\&body_40_59_dec);
+	&Xupdate_avx_32_79(\&body_40_59_dec);
+	&Xupdate_avx_32_79(\&body_20_39_dec);
+	&Xuplast_avx_80(\&body_20_39_dec,".Ldone_dec_avx");	# can jump to "done"
+
+				$saved_j=$j; @saved_V=@V;
+				$saved_rx=$rx;
+
+	&Xloop_avx(\&body_20_39_dec);
+	&Xloop_avx(\&body_20_39_dec);
+	&Xloop_avx(\&body_20_39_dec);
+
+	eval(@aes256_dec[-1]);			# last store
+$code.=<<___;
+	lea	64($in0),$in0
+
+	add	0($ctx),$A			# update context
+	add	4($ctx),@T[0]
+	add	8($ctx),$C
+	add	12($ctx),$D
+	mov	$A,0($ctx)
+	add	16($ctx),$E
+	mov	@T[0],4($ctx)
+	mov	@T[0],$B			# magic seed
+	mov	$C,8($ctx)
+	mov	$C,@T[1]
+	mov	$D,12($ctx)
+	xor	$D,@T[1]
+	mov	$E,16($ctx)
+	and	@T[1],@T[0]
+	jmp	.Loop_dec_avx
+
+.Ldone_dec_avx:
+___
+				$jj=$j=$saved_j; @V=@saved_V;
+				$rx=$saved_rx;
+
+	&Xtail_avx(\&body_20_39_dec);
+	&Xtail_avx(\&body_20_39_dec);
+	&Xtail_avx(\&body_20_39_dec);
+
+	eval(@aes256_dec[-1]);			# last store
+$code.=<<___;
+
+	add	0($ctx),$A			# update context
+	add	4($ctx),@T[0]
+	add	8($ctx),$C
+	mov	$A,0($ctx)
+	add	12($ctx),$D
+	mov	@T[0],4($ctx)
+	add	16($ctx),$E
+	mov	$C,8($ctx)
+	mov	$D,12($ctx)
+	mov	$E,16($ctx)
+	vmovups	@X[3],($ivp)			# write IV
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	96+0(%rsp),%xmm6
+	movaps	96+16(%rsp),%xmm7
+	movaps	96+32(%rsp),%xmm8
+	movaps	96+48(%rsp),%xmm9
+	movaps	96+64(%rsp),%xmm10
+	movaps	96+80(%rsp),%xmm11
+	movaps	96+96(%rsp),%xmm12
+	movaps	96+112(%rsp),%xmm13
+	movaps	96+128(%rsp),%xmm14
+	movaps	96+144(%rsp),%xmm15
+___
+$code.=<<___;
+	lea	`104+($win64?10*16:0)`(%rsp),%rsi
+	mov	0(%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
+.Lepilogue_dec_avx:
+	ret
+.size	aesni256_cbc_sha1_dec_avx,.-aesni256_cbc_sha1_dec_avx
+___
+						}}}
 }
 $code.=<<___;
 .align	64
@@ -1081,11 +1657,180 @@ K_XX_XX:
 .long	0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc	# K_40_59
 .long	0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6	# K_60_79
 .long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f	# pbswap mask
+.byte	0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0
 
 .asciz	"AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 .align	64
 ___
+						if ($shaext) {{{
+($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+
+$rounds="%r11d";
+
+($iv,$in,$rndkey0)=map("%xmm$_",(2,14,15));
+@rndkey=("%xmm0","%xmm1");
+$r=0;
+
+my ($BSWAP,$ABCD,$E,$E_,$ABCD_SAVE,$E_SAVE)=map("%xmm$_",(7..12));
+my @MSG=map("%xmm$_",(3..6));
+
+$code.=<<___;
+.type	aesni_cbc_sha1_enc_shaext,\@function,6
+.align	32
+aesni_cbc_sha1_enc_shaext:
+	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
+___
+$code.=<<___ if ($win64);
+	lea	`-8-10*16`(%rsp),%rsp
+	movaps	%xmm6,-8-10*16(%rax)
+	movaps	%xmm7,-8-9*16(%rax)
+	movaps	%xmm8,-8-8*16(%rax)
+	movaps	%xmm9,-8-7*16(%rax)
+	movaps	%xmm10,-8-6*16(%rax)
+	movaps	%xmm11,-8-5*16(%rax)
+	movaps	%xmm12,-8-4*16(%rax)
+	movaps	%xmm13,-8-3*16(%rax)
+	movaps	%xmm14,-8-2*16(%rax)
+	movaps	%xmm15,-8-1*16(%rax)
+.Lprologue_shaext:
+___
+$code.=<<___;
+	movdqu	($ctx),$ABCD
+	movd	16($ctx),$E
+	movdqa	K_XX_XX+0x50(%rip),$BSWAP	# byte-n-word swap
+
+	mov	240($key),$rounds
+	sub	$in0,$out
+	movups	($key),$rndkey0			# $key[0]
+	movups	16($key),$rndkey[0]		# forward reference
+	lea	112($key),$key			# size optimization
+
+	pshufd	\$0b00011011,$ABCD,$ABCD	# flip word order
+	pshufd	\$0b00011011,$E,$E		# flip word order
+	jmp	.Loop_shaext
 
+.align	16
+.Loop_shaext:
+___
+	&$aesenc();
+$code.=<<___;
+	movdqu		($inp),@MSG[0]
+	movdqa		$E,$E_SAVE		# offload $E
+	pshufb		$BSWAP,@MSG[0]
+	movdqu		0x10($inp),@MSG[1]
+	movdqa		$ABCD,$ABCD_SAVE	# offload $ABCD
+___
+	&$aesenc();
+$code.=<<___;
+	pshufb		$BSWAP,@MSG[1]
+
+	paddd		@MSG[0],$E
+	movdqu		0x20($inp),@MSG[2]
+	lea		0x40($inp),$inp
+	pxor		$E_SAVE,@MSG[0]		# black magic
+___
+	&$aesenc();
+$code.=<<___;
+	pxor		$E_SAVE,@MSG[0]		# black magic
+	movdqa		$ABCD,$E_
+	pshufb		$BSWAP,@MSG[2]
+	sha1rnds4	\$0,$E,$ABCD		# 0-3
+	sha1nexte	@MSG[1],$E_
+___
+	&$aesenc();
+$code.=<<___;
+	sha1msg1	@MSG[1],@MSG[0]
+	movdqu		-0x10($inp),@MSG[3]
+	movdqa		$ABCD,$E
+	pshufb		$BSWAP,@MSG[3]
+___
+	&$aesenc();
+$code.=<<___;
+	sha1rnds4	\$0,$E_,$ABCD		# 4-7
+	sha1nexte	@MSG[2],$E
+	pxor		@MSG[2],@MSG[0]
+	sha1msg1	@MSG[2],@MSG[1]
+___
+	&$aesenc();
+
+for($i=2;$i<20-4;$i++) {
+$code.=<<___;
+	movdqa		$ABCD,$E_
+	sha1rnds4	\$`int($i/5)`,$E,$ABCD	# 8-11
+	sha1nexte	@MSG[3],$E_
+___
+	&$aesenc();
+$code.=<<___;
+	sha1msg2	@MSG[3],@MSG[0]
+	pxor		@MSG[3],@MSG[1]
+	sha1msg1	@MSG[3],@MSG[2]
+___
+	($E,$E_)=($E_,$E);
+	push(@MSG,shift(@MSG));
+
+	&$aesenc();
+}
+$code.=<<___;
+	movdqa		$ABCD,$E_
+	sha1rnds4	\$3,$E,$ABCD		# 64-67
+	sha1nexte	@MSG[3],$E_
+	sha1msg2	@MSG[3],@MSG[0]
+	pxor		@MSG[3],@MSG[1]
+___
+	&$aesenc();
+$code.=<<___;
+	movdqa		$ABCD,$E
+	sha1rnds4	\$3,$E_,$ABCD		# 68-71
+	sha1nexte	@MSG[0],$E
+	sha1msg2	@MSG[0],@MSG[1]
+___
+	&$aesenc();
+$code.=<<___;
+	movdqa		$E_SAVE,@MSG[0]
+	movdqa		$ABCD,$E_
+	sha1rnds4	\$3,$E,$ABCD		# 72-75
+	sha1nexte	@MSG[1],$E_
+___
+	&$aesenc();
+$code.=<<___;
+	movdqa		$ABCD,$E
+	sha1rnds4	\$3,$E_,$ABCD		# 76-79
+	sha1nexte	$MSG[0],$E
+___
+	while($r<40)	{ &$aesenc(); }		# remaining aesenc's
+$code.=<<___;
+	dec		$len
+
+	paddd		$ABCD_SAVE,$ABCD
+	movups		$iv,48($out,$in0)	# write output
+	lea		64($in0),$in0
+	jnz		.Loop_shaext
+
+	pshufd	\$0b00011011,$ABCD,$ABCD
+	pshufd	\$0b00011011,$E,$E
+	movups	$iv,($ivp)			# write IV
+	movdqu	$ABCD,($ctx)
+	movd	$E,16($ctx)
+___
+$code.=<<___ if ($win64);
+	movaps	-8-10*16(%rax),%xmm6
+	movaps	-8-9*16(%rax),%xmm7
+	movaps	-8-8*16(%rax),%xmm8
+	movaps	-8-7*16(%rax),%xmm9
+	movaps	-8-6*16(%rax),%xmm10
+	movaps	-8-5*16(%rax),%xmm11
+	movaps	-8-4*16(%rax),%xmm12
+	movaps	-8-3*16(%rax),%xmm13
+	movaps	-8-2*16(%rax),%xmm14
+	movaps	-8-1*16(%rax),%xmm15
+	mov	%rax,%rsp
+.Lepilogue_shaext:
+___
+$code.=<<___;
+	ret
+.size	aesni_cbc_sha1_enc_shaext,.-aesni_cbc_sha1_enc_shaext
+___
+						}}}
 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
 #		CONTEXT *context,DISPATCHER_CONTEXT *disp)
 if ($win64) {
@@ -1127,7 +1872,21 @@ ssse3_handler:
 	lea	(%rsi,%r10),%r10	# epilogue label
 	cmp	%r10,%rbx		# context->Rip>=epilogue label
 	jae	.Lcommon_seh_tail
+___
+$code.=<<___ if ($shaext);
+	lea	aesni_cbc_sha1_enc_shaext(%rip),%r10
+	cmp	%r10,%rbx
+	jb	.Lseh_no_shaext
 
+	lea	(%rax),%rsi
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	168(%rax),%rax		# adjust stack pointer
+	jmp	.Lcommon_seh_tail
+.Lseh_no_shaext:
+___
+$code.=<<___;
 	lea	96(%rax),%rsi
 	lea	512($context),%rdi	# &context.Xmm6
 	mov	\$20,%ecx
@@ -1199,6 +1958,11 @@ $code.=<<___ if ($avx);
 	.rva	.LSEH_end_aesni_cbc_sha1_enc_avx
 	.rva	.LSEH_info_aesni_cbc_sha1_enc_avx
 ___
+$code.=<<___ if ($shaext);
+	.rva	.LSEH_begin_aesni_cbc_sha1_enc_shaext
+	.rva	.LSEH_end_aesni_cbc_sha1_enc_shaext
+	.rva	.LSEH_info_aesni_cbc_sha1_enc_shaext
+___
 $code.=<<___;
 .section	.xdata
 .align	8
@@ -1213,6 +1977,12 @@ $code.=<<___ if ($avx);
 	.rva	ssse3_handler
 	.rva	.Lprologue_avx,.Lepilogue_avx		# HandlerData[]
 ___
+$code.=<<___ if ($shaext);
+.LSEH_info_aesni_cbc_sha1_enc_shaext:
+	.byte	9,0,0,0
+	.rva	ssse3_handler
+	.rva	.Lprologue_shaext,.Lepilogue_shaext	# HandlerData[]
+___
 }
 
 ####################################################################
@@ -1223,28 +1993,65 @@ sub rex {
 
     $rex|=0x04			if($dst>=8);
     $rex|=0x01			if($src>=8);
-    push @opcode,$rex|0x40	if($rex);
+    unshift @opcode,$rex|0x40	if($rex);
+}
+
+sub sha1rnds4 {
+    if (@_[0] =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x0f,0x3a,0xcc);
+	rex(\@opcode,$3,$2);
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
+	my $c=$1;
+	push @opcode,$c=~/^0/?oct($c):$c;
+	return ".byte\t".join(',',@opcode);
+    } else {
+	return "sha1rnds4\t".@_[0];
+    }
+}
+
+sub sha1op38 {
+    my $instr = shift;
+    my %opcodelet = (
+		"sha1nexte" => 0xc8,
+  		"sha1msg1"  => 0xc9,
+		"sha1msg2"  => 0xca	);
+
+    if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x0f,0x38);
+	rex(\@opcode,$2,$1);
+	push @opcode,$opcodelet{$instr};
+	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
+	return ".byte\t".join(',',@opcode);
+    } else {
+	return $instr."\t".@_[0];
+    }
 }
 
 sub aesni {
   my $line=shift;
-  my @opcode=(0x66);
+  my @opcode=(0x0f,0x38);
 
     if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
 	my %opcodelet = (
-		"aesenc" => 0xdc,	"aesenclast" => 0xdd
+		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
+		"aesdec" => 0xde,	"aesdeclast" => 0xdf
 	);
 	return undef if (!defined($opcodelet{$1}));
 	rex(\@opcode,$3,$2);
-	push @opcode,0x0f,0x38,$opcodelet{$1};
-	push @opcode,0xc0|($2&7)|(($3&7)<<3);	# ModR/M
+	push @opcode,$opcodelet{$1},0xc0|($2&7)|(($3&7)<<3);	# ModR/M
+	unshift @opcode,0x66;
 	return ".byte\t".join(',',@opcode);
     }
     return $line;
 }
 
-$code =~ s/\`([^\`]*)\`/eval($1)/gem;
-$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
+foreach (split("\n",$code)) {
+        s/\`([^\`]*)\`/eval $1/geo;
+
+	s/\b(sha1rnds4)\s+(.*)/sha1rnds4($2)/geo		or
+	s/\b(sha1[^\s]*)\s+(.*)/sha1op38($1,$2)/geo		or
+	s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/geo;
 
-print $code;
+	print $_,"\n";
+}
 close STDOUT;
diff --git a/openssl/crypto/aes/asm/aesni-sha256-x86_64.pl b/openssl/crypto/aes/asm/aesni-sha256-x86_64.pl
new file mode 100755
index 000000000..c1fce8983
--- /dev/null
+++ b/openssl/crypto/aes/asm/aesni-sha256-x86_64.pl
@@ -0,0 +1,1708 @@
+#!/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/.
+# ====================================================================
+#
+# January 2013
+#
+# This is AESNI-CBC+SHA256 stitch implementation. The idea, as spelled
+# in http://download.intel.com/design/intarch/papers/323686.pdf, is
+# that since AESNI-CBC encrypt exhibit *very* low instruction-level
+# parallelism, interleaving it with another algorithm would allow to
+# utilize processor resources better and achieve better performance.
+# SHA256 instruction sequences(*) are taken from sha512-x86_64.pl and
+# AESNI code is weaved into it. As SHA256 dominates execution time,
+# stitch performance does not depend on AES key length. Below are
+# performance numbers in cycles per processed byte, less is better,
+# for standalone AESNI-CBC encrypt, standalone SHA256, and stitched
+# subroutine:
+#
+#		 AES-128/-192/-256+SHA256	this(**)gain
+# Sandy Bridge	    5.05/6.05/7.05+11.6		13.0	+28%/36%/43%
+# Ivy Bridge	    5.05/6.05/7.05+10.3		11.6	+32%/41%/50%
+# Haswell	    4.43/5.29/6.19+7.80		8.79	+39%/49%/59%
+# Bulldozer	    5.77/6.89/8.00+13.7		13.7	+42%/50%/58%
+#
+# (*)	there are XOP, AVX1 and AVX2 code pathes, meaning that
+#	Westmere is omitted from loop, this is because gain was not
+#	estimated high enough to justify the effort;
+# (**)	these are EVP-free results, results obtained with 'speed
+#	-evp aes-256-cbc-hmac-sha256' will vary by percent or two;
+
+$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 $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.19) + ($1>=2.22);
+}
+
+if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
+	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.09) + ($1>=2.10);
+}
+
+if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
+	   `ml64 2>&1` =~ /Version ([0-9]+)\./) {
+	$avx = ($1>=10) + ($1>=12);
+}
+
+if (!$avx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/) {
+	$avx = ($2>=3.0) + ($2>3.0);
+}
+
+$shaext=$avx;	### set to zero if compiling for 1.0.1
+$avx=1		if (!$shaext && $avx);
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+$func="aesni_cbc_sha256_enc";
+$TABLE="K256";
+$SZ=4;
+@ROT=($A,$B,$C,$D,$E,$F,$G,$H)=("%eax","%ebx","%ecx","%edx",
+				"%r8d","%r9d","%r10d","%r11d");
+($T1,$a0,$a1,$a2,$a3)=("%r12d","%r13d","%r14d","%r15d","%esi");
+@Sigma0=( 2,13,22);
+@Sigma1=( 6,11,25);
+@sigma0=( 7,18, 3);
+@sigma1=(17,19,10);
+$rounds=64;
+
+########################################################################
+# void aesni_cbc_sha256_enc(const void *inp,
+#			void *out,
+#			size_t length,
+#			const AES_KEY *key,
+#			unsigned char *iv,
+#			SHA256_CTX *ctx,
+#			const void *in0);
+($inp,  $out,  $len,  $key,  $ivp, $ctx, $in0) =
+("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+
+$Tbl="%rbp";
+
+$_inp="16*$SZ+0*8(%rsp)";
+$_out="16*$SZ+1*8(%rsp)";
+$_end="16*$SZ+2*8(%rsp)";
+$_key="16*$SZ+3*8(%rsp)";
+$_ivp="16*$SZ+4*8(%rsp)";
+$_ctx="16*$SZ+5*8(%rsp)";
+$_in0="16*$SZ+6*8(%rsp)";
+$_rsp="16*$SZ+7*8(%rsp)";
+$framesz=16*$SZ+8*8;
+
+$code=<<___;
+.text
+
+.extern	OPENSSL_ia32cap_P
+.globl	$func
+.type	$func,\@abi-omnipotent
+.align	16
+$func:
+___
+						if ($avx) {
+$code.=<<___;
+	lea	OPENSSL_ia32cap_P(%rip),%r11
+	mov	\$1,%eax
+	cmp	\$0,`$win64?"%rcx":"%rdi"`
+	je	.Lprobe
+	mov	0(%r11),%eax
+	mov	4(%r11),%r10
+___
+$code.=<<___ if ($shaext);
+	bt	\$61,%r10			# check for SHA
+	jc	${func}_shaext
+___
+$code.=<<___;
+	mov	%r10,%r11
+	shr	\$32,%r11
+
+	test	\$`1<<11`,%r10d			# check for XOP
+	jnz	${func}_xop
+___
+$code.=<<___ if ($avx>1);
+	and	\$`1<<8|1<<5|1<<3`,%r11d	# check for BMI2+AVX2+BMI1
+	cmp	\$`1<<8|1<<5|1<<3`,%r11d
+	je	${func}_avx2
+___
+$code.=<<___;
+	and	\$`1<<30`,%eax			# mask "Intel CPU" bit
+	and	\$`1<<28|1<<9`,%r10d		# mask AVX+SSSE3 bits
+	or	%eax,%r10d
+	cmp	\$`1<<28|1<<9|1<<30`,%r10d
+	je	${func}_avx
+	ud2
+___
+						}
+$code.=<<___;
+	xor	%eax,%eax
+	cmp	\$0,`$win64?"%rcx":"%rdi"`
+	je	.Lprobe
+	ud2
+.Lprobe:
+	ret
+.size	$func,.-$func
+
+.align	64
+.type	$TABLE,\@object
+$TABLE:
+	.long	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+	.long	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+	.long	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+	.long	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+	.long	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+	.long	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+	.long	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+	.long	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+	.long	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+	.long	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+	.long	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+	.long	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+	.long	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+	.long	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+	.long	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+	.long	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+	.long	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+	.long	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+	.long	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+	.long	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+	.long	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+	.long	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+	.long	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+	.long	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+	.long	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+	.long	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+
+	.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f
+	.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f
+	.long	0,0,0,0,   0,0,0,0,   -1,-1,-1,-1
+	.long	0,0,0,0,   0,0,0,0
+	.asciz	"AESNI-CBC+SHA256 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align	64
+___
+
+######################################################################
+# SIMD code paths
+#
+{{{
+($iv,$inout,$roundkey,$temp,
+ $mask10,$mask12,$mask14,$offload)=map("%xmm$_",(8..15));
+
+$aesni_cbc_idx=0;
+@aesni_cbc_block = (
+##	&vmovdqu	($roundkey,"0x00-0x80($inp)");'
+##	&vmovdqu	($inout,($inp));
+##	&mov		($_inp,$inp);
+
+	'&vpxor		($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x10-0x80($inp)");',
+
+	'&vpxor		($inout,$inout,$iv);',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x20-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x30-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x40-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x50-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x60-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x70-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x80-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x90-0x80($inp)");',
+
+	'&vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0xa0-0x80($inp)");',
+
+	'&vaesenclast	($temp,$inout,$roundkey);'.
+	' &vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0xb0-0x80($inp)");',
+
+	'&vpand		($iv,$temp,$mask10);'.
+	' &vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0xc0-0x80($inp)");',
+
+	'&vaesenclast	($temp,$inout,$roundkey);'.
+	' &vaesenc	($inout,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0xd0-0x80($inp)");',
+
+	'&vpand		($temp,$temp,$mask12);'.
+	' &vaesenc	($inout,$inout,$roundkey);'.
+	 '&vmovdqu	($roundkey,"0xe0-0x80($inp)");',
+
+	'&vpor		($iv,$iv,$temp);'.
+	' &vaesenclast	($temp,$inout,$roundkey);'.
+	' &vmovdqu	($roundkey,"0x00-0x80($inp)");'
+
+##	&mov		($inp,$_inp);
+##	&mov		($out,$_out);
+##	&vpand		($temp,$temp,$mask14);
+##	&vpor		($iv,$iv,$temp);
+##	&vmovdqu	($iv,($out,$inp);
+##	&lea		(inp,16($inp));
+);
+
+my $a4=$T1;
+my ($a,$b,$c,$d,$e,$f,$g,$h);
+
+sub AUTOLOAD()		# thunk [simplified] 32-bit style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
+  my $arg = pop;
+    $arg = "\$$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
+}
+
+sub body_00_15 () {
+	(
+	'($a,$b,$c,$d,$e,$f,$g,$h)=@ROT;'.
+
+	'&ror	($a0,$Sigma1[2]-$Sigma1[1])',
+	'&mov	($a,$a1)',
+	'&mov	($a4,$f)',
+
+	'&xor	($a0,$e)',
+	'&ror	($a1,$Sigma0[2]-$Sigma0[1])',
+	'&xor	($a4,$g)',			# f^g
+
+	'&ror	($a0,$Sigma1[1]-$Sigma1[0])',
+	'&xor	($a1,$a)',
+	'&and	($a4,$e)',			# (f^g)&e
+
+	@aesni_cbc_block[$aesni_cbc_idx++].
+	'&xor	($a0,$e)',
+	'&add	($h,$SZ*($i&15)."(%rsp)")',	# h+=X[i]+K[i]
+	'&mov	($a2,$a)',
+
+	'&ror	($a1,$Sigma0[1]-$Sigma0[0])',
+	'&xor	($a4,$g)',			# Ch(e,f,g)=((f^g)&e)^g
+	'&xor	($a2,$b)',			# a^b, b^c in next round
+
+	'&ror	($a0,$Sigma1[0])',		# Sigma1(e)
+	'&add	($h,$a4)',			# h+=Ch(e,f,g)
+	'&and	($a3,$a2)',			# (b^c)&(a^b)
+
+	'&xor	($a1,$a)',
+	'&add	($h,$a0)',			# h+=Sigma1(e)
+	'&xor	($a3,$b)',			# Maj(a,b,c)=Ch(a^b,c,b)
+
+	'&add	($d,$h)',			# d+=h
+	'&ror	($a1,$Sigma0[0])',		# Sigma0(a)
+	'&add	($h,$a3)',			# h+=Maj(a,b,c)
+
+	'&mov	($a0,$d)',
+	'&add	($a1,$h);'.			# h+=Sigma0(a)
+	'($a2,$a3) = ($a3,$a2); unshift(@ROT,pop(@ROT)); $i++;'
+	);
+}
+
+if ($avx) {{
+######################################################################
+# XOP code path
+#
+$code.=<<___;
+.type	${func}_xop,\@function,6
+.align	64
+${func}_xop:
+.Lxop_shortcut:
+	mov	`($win64?56:8)`(%rsp),$in0	# load 7th parameter
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	sub	\$`$framesz+$win64*16*10`,%rsp
+	and	\$-64,%rsp		# align stack frame
+
+	shl	\$6,$len
+	sub	$inp,$out		# re-bias
+	sub	$inp,$in0
+	add	$inp,$len		# end of input
+
+	#mov	$inp,$_inp		# saved later
+	mov	$out,$_out
+	mov	$len,$_end
+	#mov	$key,$_key		# remains resident in $inp register
+	mov	$ivp,$_ivp
+	mov	$ctx,$_ctx
+	mov	$in0,$_in0
+	mov	%r11,$_rsp
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,`$framesz+16*0`(%rsp)
+	movaps	%xmm7,`$framesz+16*1`(%rsp)
+	movaps	%xmm8,`$framesz+16*2`(%rsp)
+	movaps	%xmm9,`$framesz+16*3`(%rsp)
+	movaps	%xmm10,`$framesz+16*4`(%rsp)
+	movaps	%xmm11,`$framesz+16*5`(%rsp)
+	movaps	%xmm12,`$framesz+16*6`(%rsp)
+	movaps	%xmm13,`$framesz+16*7`(%rsp)
+	movaps	%xmm14,`$framesz+16*8`(%rsp)
+	movaps	%xmm15,`$framesz+16*9`(%rsp)
+___
+$code.=<<___;
+.Lprologue_xop:
+	vzeroall
+
+	mov	$inp,%r12		# borrow $a4
+	lea	0x80($key),$inp		# size optimization, reassign
+	lea	$TABLE+`$SZ*2*$rounds+32`(%rip),%r13	# borrow $a0
+	mov	0xf0-0x80($inp),%r14d	# rounds, borrow $a1
+	mov	$ctx,%r15		# borrow $a2
+	mov	$in0,%rsi		# borrow $a3
+	vmovdqu	($ivp),$iv		# load IV
+	sub	\$9,%r14
+
+	mov	$SZ*0(%r15),$A
+	mov	$SZ*1(%r15),$B
+	mov	$SZ*2(%r15),$C
+	mov	$SZ*3(%r15),$D
+	mov	$SZ*4(%r15),$E
+	mov	$SZ*5(%r15),$F
+	mov	$SZ*6(%r15),$G
+	mov	$SZ*7(%r15),$H
+
+	vmovdqa	0x00(%r13,%r14,8),$mask14
+	vmovdqa	0x10(%r13,%r14,8),$mask12
+	vmovdqa	0x20(%r13,%r14,8),$mask10
+	vmovdqu	0x00-0x80($inp),$roundkey
+	jmp	.Lloop_xop
+___
+					if ($SZ==4) {	# SHA256
+    my @X = map("%xmm$_",(0..3));
+    my ($t0,$t1,$t2,$t3) = map("%xmm$_",(4..7));
+
+$code.=<<___;
+.align	16
+.Lloop_xop:
+	vmovdqa	$TABLE+`$SZ*2*$rounds`(%rip),$t3
+	vmovdqu	0x00(%rsi,%r12),@X[0]
+	vmovdqu	0x10(%rsi,%r12),@X[1]
+	vmovdqu	0x20(%rsi,%r12),@X[2]
+	vmovdqu	0x30(%rsi,%r12),@X[3]
+	vpshufb	$t3,@X[0],@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vpshufb	$t3,@X[1],@X[1]
+	vpshufb	$t3,@X[2],@X[2]
+	vpaddd	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[3],@X[3]
+	vpaddd	0x20($Tbl),@X[1],$t1
+	vpaddd	0x40($Tbl),@X[2],$t2
+	vpaddd	0x60($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x10(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x20(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x30(%rsp)
+	mov	$E,$a0
+	jmp	.Lxop_00_47
+
+.align	16
+.Lxop_00_47:
+	sub	\$-16*2*$SZ,$Tbl	# size optimization
+	vmovdqu	(%r12),$inout		# $a4
+	mov	%r12,$_inp		# $a4
+___
+sub XOP_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 104 instructions
+
+	&vpalignr	($t0,@X[1],@X[0],$SZ);	# X[1..4]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpalignr	($t3,@X[3],@X[2],$SZ);	# X[9..12]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotd		($t1,$t0,8*$SZ-$sigma0[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpsrld		($t0,$t0,$sigma0[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpaddd	(@X[0],@X[0],$t3);	# X[0..3] += X[9..12]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotd		($t2,$t1,$sigma0[1]-$sigma0[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t1);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t3,@X[3],8*$SZ-$sigma1[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t2);		# sigma0(X[1..4])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpsrld	($t2,@X[3],$sigma1[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t0);	# X[0..3] += sigma0(X[1..4])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t1,$t3,$sigma1[1]-$sigma1[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t1);		# sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpsrldq	($t3,$t3,8);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t3);	# X[0..1] += sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t3,@X[0],8*$SZ-$sigma1[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpsrld	($t2,@X[0],$sigma1[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t1,$t3,$sigma1[1]-$sigma1[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t1);		# sigma1(X[16..17])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpslldq	($t3,$t3,8);		# 22 instructions
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t3);	# X[2..3] += sigma1(X[16..17])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		($t2,@X[0],16*2*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    $aesni_cbc_idx=0;
+    for ($i=0,$j=0; $j<4; $j++) {
+	&XOP_256_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+    	&mov		("%r12",$_inp);		# borrow $a4
+	&vpand		($temp,$temp,$mask14);
+	&mov		("%r15",$_out);		# borrow $a2
+	&vpor		($iv,$iv,$temp);
+	&vmovdqu	("(%r15,%r12)",$iv);	# write output
+	&lea		("%r12","16(%r12)");	# inp++
+
+	&cmpb	($SZ-1+16*2*$SZ."($Tbl)",0);
+	&jne	(".Lxop_00_47");
+
+	&vmovdqu	($inout,"(%r12)");
+	&mov		($_inp,"%r12");
+
+    $aesni_cbc_idx=0;
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+					}
+$code.=<<___;
+	mov	$_inp,%r12		# borrow $a4
+	mov	$_out,%r13		# borrow $a0
+	mov	$_ctx,%r15		# borrow $a2
+	mov	$_in0,%rsi		# borrow $a3
+
+	vpand	$mask14,$temp,$temp
+	mov	$a1,$A
+	vpor	$temp,$iv,$iv
+	vmovdqu	$iv,(%r13,%r12)		# write output
+	lea	16(%r12),%r12		# inp++
+
+	add	$SZ*0(%r15),$A
+	add	$SZ*1(%r15),$B
+	add	$SZ*2(%r15),$C
+	add	$SZ*3(%r15),$D
+	add	$SZ*4(%r15),$E
+	add	$SZ*5(%r15),$F
+	add	$SZ*6(%r15),$G
+	add	$SZ*7(%r15),$H
+
+	cmp	$_end,%r12
+
+	mov	$A,$SZ*0(%r15)
+	mov	$B,$SZ*1(%r15)
+	mov	$C,$SZ*2(%r15)
+	mov	$D,$SZ*3(%r15)
+	mov	$E,$SZ*4(%r15)
+	mov	$F,$SZ*5(%r15)
+	mov	$G,$SZ*6(%r15)
+	mov	$H,$SZ*7(%r15)
+
+	jb	.Lloop_xop
+
+	mov	$_ivp,$ivp
+	mov	$_rsp,%rsi
+	vmovdqu	$iv,($ivp)		# output IV
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	`$framesz+16*0`(%rsp),%xmm6
+	movaps	`$framesz+16*1`(%rsp),%xmm7
+	movaps	`$framesz+16*2`(%rsp),%xmm8
+	movaps	`$framesz+16*3`(%rsp),%xmm9
+	movaps	`$framesz+16*4`(%rsp),%xmm10
+	movaps	`$framesz+16*5`(%rsp),%xmm11
+	movaps	`$framesz+16*6`(%rsp),%xmm12
+	movaps	`$framesz+16*7`(%rsp),%xmm13
+	movaps	`$framesz+16*8`(%rsp),%xmm14
+	movaps	`$framesz+16*9`(%rsp),%xmm15
+___
+$code.=<<___;
+	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
+.Lepilogue_xop:
+	ret
+.size	${func}_xop,.-${func}_xop
+___
+######################################################################
+# AVX+shrd code path
+#
+local *ror = sub { &shrd(@_[0],@_) };
+
+$code.=<<___;
+.type	${func}_avx,\@function,6
+.align	64
+${func}_avx:
+.Lavx_shortcut:
+	mov	`($win64?56:8)`(%rsp),$in0	# load 7th parameter
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	sub	\$`$framesz+$win64*16*10`,%rsp
+	and	\$-64,%rsp		# align stack frame
+
+	shl	\$6,$len
+	sub	$inp,$out		# re-bias
+	sub	$inp,$in0
+	add	$inp,$len		# end of input
+
+	#mov	$inp,$_inp		# saved later
+	mov	$out,$_out
+	mov	$len,$_end
+	#mov	$key,$_key		# remains resident in $inp register
+	mov	$ivp,$_ivp
+	mov	$ctx,$_ctx
+	mov	$in0,$_in0
+	mov	%r11,$_rsp
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,`$framesz+16*0`(%rsp)
+	movaps	%xmm7,`$framesz+16*1`(%rsp)
+	movaps	%xmm8,`$framesz+16*2`(%rsp)
+	movaps	%xmm9,`$framesz+16*3`(%rsp)
+	movaps	%xmm10,`$framesz+16*4`(%rsp)
+	movaps	%xmm11,`$framesz+16*5`(%rsp)
+	movaps	%xmm12,`$framesz+16*6`(%rsp)
+	movaps	%xmm13,`$framesz+16*7`(%rsp)
+	movaps	%xmm14,`$framesz+16*8`(%rsp)
+	movaps	%xmm15,`$framesz+16*9`(%rsp)
+___
+$code.=<<___;
+.Lprologue_avx:
+	vzeroall
+
+	mov	$inp,%r12		# borrow $a4
+	lea	0x80($key),$inp		# size optimization, reassign
+	lea	$TABLE+`$SZ*2*$rounds+32`(%rip),%r13	# borrow $a0
+	mov	0xf0-0x80($inp),%r14d	# rounds, borrow $a1
+	mov	$ctx,%r15		# borrow $a2
+	mov	$in0,%rsi		# borrow $a3
+	vmovdqu	($ivp),$iv		# load IV
+	sub	\$9,%r14
+
+	mov	$SZ*0(%r15),$A
+	mov	$SZ*1(%r15),$B
+	mov	$SZ*2(%r15),$C
+	mov	$SZ*3(%r15),$D
+	mov	$SZ*4(%r15),$E
+	mov	$SZ*5(%r15),$F
+	mov	$SZ*6(%r15),$G
+	mov	$SZ*7(%r15),$H
+
+	vmovdqa	0x00(%r13,%r14,8),$mask14
+	vmovdqa	0x10(%r13,%r14,8),$mask12
+	vmovdqa	0x20(%r13,%r14,8),$mask10
+	vmovdqu	0x00-0x80($inp),$roundkey
+___
+					if ($SZ==4) {	# SHA256
+    my @X = map("%xmm$_",(0..3));
+    my ($t0,$t1,$t2,$t3) = map("%xmm$_",(4..7));
+
+$code.=<<___;
+	jmp	.Lloop_avx
+.align	16
+.Lloop_avx:
+	vmovdqa	$TABLE+`$SZ*2*$rounds`(%rip),$t3
+	vmovdqu	0x00(%rsi,%r12),@X[0]
+	vmovdqu	0x10(%rsi,%r12),@X[1]
+	vmovdqu	0x20(%rsi,%r12),@X[2]
+	vmovdqu	0x30(%rsi,%r12),@X[3]
+	vpshufb	$t3,@X[0],@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vpshufb	$t3,@X[1],@X[1]
+	vpshufb	$t3,@X[2],@X[2]
+	vpaddd	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[3],@X[3]
+	vpaddd	0x20($Tbl),@X[1],$t1
+	vpaddd	0x40($Tbl),@X[2],$t2
+	vpaddd	0x60($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x10(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x20(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x30(%rsp)
+	mov	$E,$a0
+	jmp	.Lavx_00_47
+
+.align	16
+.Lavx_00_47:
+	sub	\$-16*2*$SZ,$Tbl	# size optimization
+	vmovdqu	(%r12),$inout		# $a4
+	mov	%r12,$_inp		# $a4
+___
+sub Xupdate_256_AVX () {
+	(
+	'&vpalignr	($t0,@X[1],@X[0],$SZ)',	# X[1..4]
+	 '&vpalignr	($t3,@X[3],@X[2],$SZ)',	# X[9..12]
+	'&vpsrld	($t2,$t0,$sigma0[0]);',
+	 '&vpaddd	(@X[0],@X[0],$t3)',	# X[0..3] += X[9..12]
+	'&vpsrld	($t3,$t0,$sigma0[2])',
+	'&vpslld	($t1,$t0,8*$SZ-$sigma0[1]);',
+	'&vpxor		($t0,$t3,$t2)',
+	 '&vpshufd	($t3,@X[3],0b11111010)',# X[14..15]
+	'&vpsrld	($t2,$t2,$sigma0[1]-$sigma0[0]);',
+	'&vpxor		($t0,$t0,$t1)',
+	'&vpslld	($t1,$t1,$sigma0[1]-$sigma0[0]);',
+	'&vpxor		($t0,$t0,$t2)',
+	 '&vpsrld	($t2,$t3,$sigma1[2]);',
+	'&vpxor		($t0,$t0,$t1)',		# sigma0(X[1..4])
+	 '&vpsrlq	($t3,$t3,$sigma1[0]);',
+	'&vpaddd	(@X[0],@X[0],$t0)',	# X[0..3] += sigma0(X[1..4])
+	 '&vpxor	($t2,$t2,$t3);',
+	 '&vpsrlq	($t3,$t3,$sigma1[1]-$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3)',		# sigma1(X[14..15])
+	 '&vpshufd	($t2,$t2,0b10000100)',
+	 '&vpsrldq	($t2,$t2,8)',
+	'&vpaddd	(@X[0],@X[0],$t2)',	# X[0..1] += sigma1(X[14..15])
+	 '&vpshufd	($t3,@X[0],0b01010000)',# X[16..17]
+	 '&vpsrld	($t2,$t3,$sigma1[2])',
+	 '&vpsrlq	($t3,$t3,$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3);',
+	 '&vpsrlq	($t3,$t3,$sigma1[1]-$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3)',
+	 '&vpshufd	($t2,$t2,0b11101000)',
+	 '&vpslldq	($t2,$t2,8)',
+	'&vpaddd	(@X[0],@X[0],$t2)'	# X[2..3] += sigma1(X[16..17])
+	);
+}
+
+sub AVX_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 104 instructions
+
+	foreach (Xupdate_256_AVX()) {		# 29 instructions
+	    eval;
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	}
+	&vpaddd		($t2,@X[0],16*2*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    $aesni_cbc_idx=0;
+    for ($i=0,$j=0; $j<4; $j++) {
+	&AVX_256_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+    	&mov		("%r12",$_inp);		# borrow $a4
+	&vpand		($temp,$temp,$mask14);
+	&mov		("%r15",$_out);		# borrow $a2
+	&vpor		($iv,$iv,$temp);
+	&vmovdqu	("(%r15,%r12)",$iv);	# write output
+	&lea		("%r12","16(%r12)");	# inp++
+
+	&cmpb	($SZ-1+16*2*$SZ."($Tbl)",0);
+	&jne	(".Lavx_00_47");
+
+	&vmovdqu	($inout,"(%r12)");
+	&mov		($_inp,"%r12");
+
+    $aesni_cbc_idx=0;
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+
+					}
+$code.=<<___;
+	mov	$_inp,%r12		# borrow $a4
+	mov	$_out,%r13		# borrow $a0
+	mov	$_ctx,%r15		# borrow $a2
+	mov	$_in0,%rsi		# borrow $a3
+
+	vpand	$mask14,$temp,$temp
+	mov	$a1,$A
+	vpor	$temp,$iv,$iv
+	vmovdqu	$iv,(%r13,%r12)		# write output
+	lea	16(%r12),%r12		# inp++
+
+	add	$SZ*0(%r15),$A
+	add	$SZ*1(%r15),$B
+	add	$SZ*2(%r15),$C
+	add	$SZ*3(%r15),$D
+	add	$SZ*4(%r15),$E
+	add	$SZ*5(%r15),$F
+	add	$SZ*6(%r15),$G
+	add	$SZ*7(%r15),$H
+
+	cmp	$_end,%r12
+
+	mov	$A,$SZ*0(%r15)
+	mov	$B,$SZ*1(%r15)
+	mov	$C,$SZ*2(%r15)
+	mov	$D,$SZ*3(%r15)
+	mov	$E,$SZ*4(%r15)
+	mov	$F,$SZ*5(%r15)
+	mov	$G,$SZ*6(%r15)
+	mov	$H,$SZ*7(%r15)
+	jb	.Lloop_avx
+
+	mov	$_ivp,$ivp
+	mov	$_rsp,%rsi
+	vmovdqu	$iv,($ivp)		# output IV
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	`$framesz+16*0`(%rsp),%xmm6
+	movaps	`$framesz+16*1`(%rsp),%xmm7
+	movaps	`$framesz+16*2`(%rsp),%xmm8
+	movaps	`$framesz+16*3`(%rsp),%xmm9
+	movaps	`$framesz+16*4`(%rsp),%xmm10
+	movaps	`$framesz+16*5`(%rsp),%xmm11
+	movaps	`$framesz+16*6`(%rsp),%xmm12
+	movaps	`$framesz+16*7`(%rsp),%xmm13
+	movaps	`$framesz+16*8`(%rsp),%xmm14
+	movaps	`$framesz+16*9`(%rsp),%xmm15
+___
+$code.=<<___;
+	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
+.Lepilogue_avx:
+	ret
+.size	${func}_avx,.-${func}_avx
+___
+
+if ($avx>1) {{
+######################################################################
+# AVX2+BMI code path
+#
+my $a5=$SZ==4?"%esi":"%rsi";	# zap $inp 
+my $PUSH8=8*2*$SZ;
+use integer;
+
+sub bodyx_00_15 () {
+	# at start $a1 should be zero, $a3 - $b^$c and $a4 copy of $f
+	(
+	'($a,$b,$c,$d,$e,$f,$g,$h)=@ROT;'.
+
+	'&add	($h,(32*($i/(16/$SZ))+$SZ*($i%(16/$SZ)))%$PUSH8.$base)',    # h+=X[i]+K[i]
+	'&and	($a4,$e)',		# f&e
+	'&rorx	($a0,$e,$Sigma1[2])',
+	'&rorx	($a2,$e,$Sigma1[1])',
+
+	'&lea	($a,"($a,$a1)")',	# h+=Sigma0(a) from the past
+	'&lea	($h,"($h,$a4)")',
+	'&andn	($a4,$e,$g)',		# ~e&g
+	'&xor	($a0,$a2)',
+
+	'&rorx	($a1,$e,$Sigma1[0])',
+	'&lea	($h,"($h,$a4)")',	# h+=Ch(e,f,g)=(e&f)+(~e&g)
+	'&xor	($a0,$a1)',		# Sigma1(e)
+	'&mov	($a2,$a)',
+
+	'&rorx	($a4,$a,$Sigma0[2])',
+	'&lea	($h,"($h,$a0)")',	# h+=Sigma1(e)
+	'&xor	($a2,$b)',		# a^b, b^c in next round
+	'&rorx	($a1,$a,$Sigma0[1])',
+
+	'&rorx	($a0,$a,$Sigma0[0])',
+	'&lea	($d,"($d,$h)")',	# d+=h
+	'&and	($a3,$a2)',		# (b^c)&(a^b)
+	@aesni_cbc_block[$aesni_cbc_idx++].
+	'&xor	($a1,$a4)',
+
+	'&xor	($a3,$b)',		# Maj(a,b,c)=Ch(a^b,c,b)
+	'&xor	($a1,$a0)',		# Sigma0(a)
+	'&lea	($h,"($h,$a3)");'.	# h+=Maj(a,b,c)
+	'&mov	($a4,$e)',		# copy of f in future
+
+	'($a2,$a3) = ($a3,$a2); unshift(@ROT,pop(@ROT)); $i++;'
+	);
+	# and at the finish one has to $a+=$a1
+}
+
+$code.=<<___;
+.type	${func}_avx2,\@function,6
+.align	64
+${func}_avx2:
+.Lavx2_shortcut:
+	mov	`($win64?56:8)`(%rsp),$in0	# load 7th parameter
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	sub	\$`2*$SZ*$rounds+8*8+$win64*16*10`,%rsp
+	and	\$-256*$SZ,%rsp		# align stack frame
+	add	\$`2*$SZ*($rounds-8)`,%rsp
+
+	shl	\$6,$len
+	sub	$inp,$out		# re-bias
+	sub	$inp,$in0
+	add	$inp,$len		# end of input
+
+	#mov	$inp,$_inp		# saved later
+	#mov	$out,$_out		# kept in $offload
+	mov	$len,$_end
+	#mov	$key,$_key		# remains resident in $inp register
+	mov	$ivp,$_ivp
+	mov	$ctx,$_ctx
+	mov	$in0,$_in0
+	mov	%r11,$_rsp
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,`$framesz+16*0`(%rsp)
+	movaps	%xmm7,`$framesz+16*1`(%rsp)
+	movaps	%xmm8,`$framesz+16*2`(%rsp)
+	movaps	%xmm9,`$framesz+16*3`(%rsp)
+	movaps	%xmm10,`$framesz+16*4`(%rsp)
+	movaps	%xmm11,`$framesz+16*5`(%rsp)
+	movaps	%xmm12,`$framesz+16*6`(%rsp)
+	movaps	%xmm13,`$framesz+16*7`(%rsp)
+	movaps	%xmm14,`$framesz+16*8`(%rsp)
+	movaps	%xmm15,`$framesz+16*9`(%rsp)
+___
+$code.=<<___;
+.Lprologue_avx2:
+	vzeroall
+
+	mov	$inp,%r13		# borrow $a0
+	vpinsrq	\$1,$out,$offload,$offload
+	lea	0x80($key),$inp		# size optimization, reassign
+	lea	$TABLE+`$SZ*2*$rounds+32`(%rip),%r12	# borrow $a4
+	mov	0xf0-0x80($inp),%r14d	# rounds, borrow $a1
+	mov	$ctx,%r15		# borrow $a2
+	mov	$in0,%rsi		# borrow $a3
+	vmovdqu	($ivp),$iv		# load IV
+	lea	-9(%r14),%r14
+
+	vmovdqa	0x00(%r12,%r14,8),$mask14
+	vmovdqa	0x10(%r12,%r14,8),$mask12
+	vmovdqa	0x20(%r12,%r14,8),$mask10
+
+	sub	\$-16*$SZ,%r13		# inp++, size optimization
+	mov	$SZ*0(%r15),$A
+	lea	(%rsi,%r13),%r12	# borrow $a0
+	mov	$SZ*1(%r15),$B
+	cmp	$len,%r13		# $_end
+	mov	$SZ*2(%r15),$C
+	cmove	%rsp,%r12		# next block or random data
+	mov	$SZ*3(%r15),$D
+	mov	$SZ*4(%r15),$E
+	mov	$SZ*5(%r15),$F
+	mov	$SZ*6(%r15),$G
+	mov	$SZ*7(%r15),$H
+	vmovdqu	0x00-0x80($inp),$roundkey
+___
+					if ($SZ==4) {	# SHA256
+    my @X = map("%ymm$_",(0..3));
+    my ($t0,$t1,$t2,$t3) = map("%ymm$_",(4..7));
+
+$code.=<<___;
+	jmp	.Loop_avx2
+.align	16
+.Loop_avx2:
+	vmovdqa	$TABLE+`$SZ*2*$rounds`(%rip),$t3
+	vmovdqu	-16*$SZ+0(%rsi,%r13),%xmm0
+	vmovdqu	-16*$SZ+16(%rsi,%r13),%xmm1
+	vmovdqu	-16*$SZ+32(%rsi,%r13),%xmm2
+	vmovdqu	-16*$SZ+48(%rsi,%r13),%xmm3
+
+	vinserti128	\$1,(%r12),@X[0],@X[0]
+	vinserti128	\$1,16(%r12),@X[1],@X[1]
+	 vpshufb	$t3,@X[0],@X[0]
+	vinserti128	\$1,32(%r12),@X[2],@X[2]
+	 vpshufb	$t3,@X[1],@X[1]
+	vinserti128	\$1,48(%r12),@X[3],@X[3]
+
+	lea	$TABLE(%rip),$Tbl
+	vpshufb	$t3,@X[2],@X[2]
+	lea	-16*$SZ(%r13),%r13
+	vpaddd	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[3],@X[3]
+	vpaddd	0x20($Tbl),@X[1],$t1
+	vpaddd	0x40($Tbl),@X[2],$t2
+	vpaddd	0x60($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	xor	$a1,$a1
+	vmovdqa	$t1,0x20(%rsp)
+	lea	-$PUSH8(%rsp),%rsp
+	mov	$B,$a3
+	vmovdqa	$t2,0x00(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x20(%rsp)
+	mov	$F,$a4
+	sub	\$-16*2*$SZ,$Tbl	# size optimization
+	jmp	.Lavx2_00_47
+
+.align	16
+.Lavx2_00_47:
+	vmovdqu	(%r13),$inout
+	vpinsrq	\$0,%r13,$offload,$offload
+___
+
+sub AVX2_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 96 instructions
+my $base = "+2*$PUSH8(%rsp)";
+
+	&lea	("%rsp","-$PUSH8(%rsp)")	if (($j%2)==0);
+	foreach (Xupdate_256_AVX()) {		# 29 instructions
+	    eval;
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	}
+	&vpaddd		($t2,@X[0],16*2*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	((32*$j)%$PUSH8."(%rsp)",$t2);
+}
+    $aesni_cbc_idx=0;
+    for ($i=0,$j=0; $j<4; $j++) {
+	&AVX2_256_00_47($j,\&bodyx_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&vmovq		("%r13",$offload);	# borrow $a0
+	&vpextrq	("%r15",$offload,1);	# borrow $a2
+	&vpand		($temp,$temp,$mask14);
+	&vpor		($iv,$iv,$temp);
+	&vmovdqu	("(%r15,%r13)",$iv);	# write output
+	&lea		("%r13","16(%r13)");	# inp++
+
+	&lea	($Tbl,16*2*$SZ."($Tbl)");
+	&cmpb	(($SZ-1)."($Tbl)",0);
+	&jne	(".Lavx2_00_47");
+
+	&vmovdqu	($inout,"(%r13)");
+	&vpinsrq	($offload,$offload,"%r13",0);
+
+    $aesni_cbc_idx=0;
+    for ($i=0; $i<16; ) {
+	my $base=$i<8?"+$PUSH8(%rsp)":"(%rsp)";
+	foreach(bodyx_00_15()) { eval; }
+    }
+					}
+$code.=<<___;
+	vpextrq	\$1,$offload,%r12		# $_out, borrow $a4
+	vmovq	$offload,%r13			# $_inp, borrow $a0
+	mov	`2*$SZ*$rounds+5*8`(%rsp),%r15	# $_ctx, borrow $a2
+	add	$a1,$A
+	lea	`2*$SZ*($rounds-8)`(%rsp),$Tbl
+
+	vpand	$mask14,$temp,$temp
+	vpor	$temp,$iv,$iv
+	vmovdqu	$iv,(%r12,%r13)			# write output
+	lea	16(%r13),%r13
+
+	add	$SZ*0(%r15),$A
+	add	$SZ*1(%r15),$B
+	add	$SZ*2(%r15),$C
+	add	$SZ*3(%r15),$D
+	add	$SZ*4(%r15),$E
+	add	$SZ*5(%r15),$F
+	add	$SZ*6(%r15),$G
+	add	$SZ*7(%r15),$H
+
+	mov	$A,$SZ*0(%r15)
+	mov	$B,$SZ*1(%r15)
+	mov	$C,$SZ*2(%r15)
+	mov	$D,$SZ*3(%r15)
+	mov	$E,$SZ*4(%r15)
+	mov	$F,$SZ*5(%r15)
+	mov	$G,$SZ*6(%r15)
+	mov	$H,$SZ*7(%r15)
+
+	cmp	`$PUSH8+2*8`($Tbl),%r13		# $_end
+	je	.Ldone_avx2
+
+	xor	$a1,$a1
+	mov	$B,$a3
+	mov	$F,$a4
+	xor	$C,$a3			# magic
+	jmp	.Lower_avx2
+.align	16
+.Lower_avx2:
+	vmovdqu	(%r13),$inout
+	vpinsrq	\$0,%r13,$offload,$offload
+___
+    $aesni_cbc_idx=0;
+    for ($i=0; $i<16; ) {
+	my $base="+16($Tbl)";
+	foreach(bodyx_00_15()) { eval; }
+	&lea	($Tbl,"-$PUSH8($Tbl)")	if ($i==8);
+    }
+$code.=<<___;
+	vmovq	$offload,%r13			# borrow $a0
+	vpextrq	\$1,$offload,%r15		# borrow $a2
+	vpand	$mask14,$temp,$temp
+	vpor	$temp,$iv,$iv
+	lea	-$PUSH8($Tbl),$Tbl
+	vmovdqu	$iv,(%r15,%r13)			# write output
+	lea	16(%r13),%r13			# inp++
+	cmp	%rsp,$Tbl
+	jae	.Lower_avx2
+
+	mov	`2*$SZ*$rounds+5*8`(%rsp),%r15	# $_ctx, borrow $a2
+	lea	16*$SZ(%r13),%r13
+	mov	`2*$SZ*$rounds+6*8`(%rsp),%rsi	# $_in0, borrow $a3
+	add	$a1,$A
+	lea	`2*$SZ*($rounds-8)`(%rsp),%rsp
+
+	add	$SZ*0(%r15),$A
+	add	$SZ*1(%r15),$B
+	add	$SZ*2(%r15),$C
+	add	$SZ*3(%r15),$D
+	add	$SZ*4(%r15),$E
+	add	$SZ*5(%r15),$F
+	add	$SZ*6(%r15),$G
+	lea	(%rsi,%r13),%r12
+	add	$SZ*7(%r15),$H
+
+	cmp	$_end,%r13
+
+	mov	$A,$SZ*0(%r15)
+	cmove	%rsp,%r12		# next block or stale data
+	mov	$B,$SZ*1(%r15)
+	mov	$C,$SZ*2(%r15)
+	mov	$D,$SZ*3(%r15)
+	mov	$E,$SZ*4(%r15)
+	mov	$F,$SZ*5(%r15)
+	mov	$G,$SZ*6(%r15)
+	mov	$H,$SZ*7(%r15)
+
+	jbe	.Loop_avx2
+	lea	(%rsp),$Tbl
+
+.Ldone_avx2:
+	lea	($Tbl),%rsp
+	mov	$_ivp,$ivp
+	mov	$_rsp,%rsi
+	vmovdqu	$iv,($ivp)		# output IV
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	`$framesz+16*0`(%rsp),%xmm6
+	movaps	`$framesz+16*1`(%rsp),%xmm7
+	movaps	`$framesz+16*2`(%rsp),%xmm8
+	movaps	`$framesz+16*3`(%rsp),%xmm9
+	movaps	`$framesz+16*4`(%rsp),%xmm10
+	movaps	`$framesz+16*5`(%rsp),%xmm11
+	movaps	`$framesz+16*6`(%rsp),%xmm12
+	movaps	`$framesz+16*7`(%rsp),%xmm13
+	movaps	`$framesz+16*8`(%rsp),%xmm14
+	movaps	`$framesz+16*9`(%rsp),%xmm15
+___
+$code.=<<___;
+	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
+.Lepilogue_avx2:
+	ret
+.size	${func}_avx2,.-${func}_avx2
+___
+}}
+}}
+{{
+my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+
+my ($rounds,$Tbl)=("%r11d","%rbx");
+
+my ($iv,$in,$rndkey0)=map("%xmm$_",(6,14,15));
+my @rndkey=("%xmm4","%xmm5");
+my $r=0;
+my $sn=0;
+
+my ($Wi,$ABEF,$CDGH,$TMP,$BSWAP,$ABEF_SAVE,$CDGH_SAVE)=map("%xmm$_",(0..3,7..9));
+my @MSG=map("%xmm$_",(10..13));
+
+my $aesenc=sub {
+  use integer;
+  my ($n,$k)=($r/10,$r%10);
+    if ($k==0) {
+      $code.=<<___;
+	movups		`16*$n`($in0),$in		# load input
+	xorps		$rndkey0,$in
+___
+      $code.=<<___ if ($n);
+	movups		$iv,`16*($n-1)`($out,$in0)	# write output
+___
+      $code.=<<___;
+	xorps		$in,$iv
+	movups		`32+16*$k-112`($key),$rndkey[1]
+	aesenc		$rndkey[0],$iv
+___
+    } elsif ($k==9) {
+      $sn++;
+      $code.=<<___;
+	cmp		\$11,$rounds
+	jb		.Laesenclast$sn
+	movups		`32+16*($k+0)-112`($key),$rndkey[1]
+	aesenc		$rndkey[0],$iv
+	movups		`32+16*($k+1)-112`($key),$rndkey[0]
+	aesenc		$rndkey[1],$iv
+	je		.Laesenclast$sn
+	movups		`32+16*($k+2)-112`($key),$rndkey[1]
+	aesenc		$rndkey[0],$iv
+	movups		`32+16*($k+3)-112`($key),$rndkey[0]
+	aesenc		$rndkey[1],$iv
+.Laesenclast$sn:
+	aesenclast	$rndkey[0],$iv
+	movups		16-112($key),$rndkey[1]		# forward reference
+	nop
+___
+    } else {
+      $code.=<<___;
+	movups		`32+16*$k-112`($key),$rndkey[1]
+	aesenc		$rndkey[0],$iv
+___
+    }
+    $r++;	unshift(@rndkey,pop(@rndkey));
+};
+
+if ($shaext) {
+my $Tbl="%rax";
+
+$code.=<<___;
+.type	${func}_shaext,\@function,6
+.align	32
+${func}_shaext:
+	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
+___
+$code.=<<___ if ($win64);
+	lea	`-8-10*16`(%rsp),%rsp
+	movaps	%xmm6,-8-10*16(%rax)
+	movaps	%xmm7,-8-9*16(%rax)
+	movaps	%xmm8,-8-8*16(%rax)
+	movaps	%xmm9,-8-7*16(%rax)
+	movaps	%xmm10,-8-6*16(%rax)
+	movaps	%xmm11,-8-5*16(%rax)
+	movaps	%xmm12,-8-4*16(%rax)
+	movaps	%xmm13,-8-3*16(%rax)
+	movaps	%xmm14,-8-2*16(%rax)
+	movaps	%xmm15,-8-1*16(%rax)
+.Lprologue_shaext:
+___
+$code.=<<___;
+	lea		K256+0x80(%rip),$Tbl
+	movdqu		($ctx),$ABEF		# DCBA
+	movdqu		16($ctx),$CDGH		# HGFE
+	movdqa		0x200-0x80($Tbl),$TMP	# byte swap mask
+
+	mov		240($key),$rounds
+	sub		$in0,$out
+	movups		($key),$rndkey0		# $key[0]
+	movups		16($key),$rndkey[0]	# forward reference
+	lea		112($key),$key		# size optimization
+
+	pshufd		\$0x1b,$ABEF,$Wi	# ABCD
+	pshufd		\$0xb1,$ABEF,$ABEF	# CDAB
+	pshufd		\$0x1b,$CDGH,$CDGH	# EFGH
+	movdqa		$TMP,$BSWAP		# offload
+	palignr		\$8,$CDGH,$ABEF		# ABEF
+	punpcklqdq	$Wi,$CDGH		# CDGH
+
+	jmp	.Loop_shaext
+
+.align	16
+.Loop_shaext:
+	movdqu		($inp),@MSG[0]
+	movdqu		0x10($inp),@MSG[1]
+	movdqu		0x20($inp),@MSG[2]
+	pshufb		$TMP,@MSG[0]
+	movdqu		0x30($inp),@MSG[3]
+
+	movdqa		0*32-0x80($Tbl),$Wi
+	paddd		@MSG[0],$Wi
+	pshufb		$TMP,@MSG[1]
+	movdqa		$CDGH,$CDGH_SAVE	# offload
+	movdqa		$ABEF,$ABEF_SAVE	# offload
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 0-3
+	pshufd		\$0x0e,$Wi,$Wi
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+
+	movdqa		1*32-0x80($Tbl),$Wi
+	paddd		@MSG[1],$Wi
+	pshufb		$TMP,@MSG[2]
+	lea		0x40($inp),$inp
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 4-7
+	pshufd		\$0x0e,$Wi,$Wi
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+
+	movdqa		2*32-0x80($Tbl),$Wi
+	paddd		@MSG[2],$Wi
+	pshufb		$TMP,@MSG[3]
+	sha256msg1	@MSG[1],@MSG[0]
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 8-11
+	pshufd		\$0x0e,$Wi,$Wi
+	movdqa		@MSG[3],$TMP
+	palignr		\$4,@MSG[2],$TMP
+	paddd		$TMP,@MSG[0]
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+
+	movdqa		3*32-0x80($Tbl),$Wi
+	paddd		@MSG[3],$Wi
+	sha256msg2	@MSG[3],@MSG[0]
+	sha256msg1	@MSG[2],@MSG[1]
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 12-15
+	pshufd		\$0x0e,$Wi,$Wi
+___
+	&$aesenc();
+$code.=<<___;
+	movdqa		@MSG[0],$TMP
+	palignr		\$4,@MSG[3],$TMP
+	paddd		$TMP,@MSG[1]
+	sha256rnds2	$CDGH,$ABEF
+___
+for($i=4;$i<16-3;$i++) {
+	&$aesenc()	if (($r%10)==0);
+$code.=<<___;
+	movdqa		$i*32-0x80($Tbl),$Wi
+	paddd		@MSG[0],$Wi
+	sha256msg2	@MSG[0],@MSG[1]
+	sha256msg1	@MSG[3],@MSG[2]
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 16-19...
+	pshufd		\$0x0e,$Wi,$Wi
+	movdqa		@MSG[1],$TMP
+	palignr		\$4,@MSG[0],$TMP
+	paddd		$TMP,@MSG[2]
+___
+	&$aesenc();
+	&$aesenc()	if ($r==19);
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+___
+	push(@MSG,shift(@MSG));
+}
+$code.=<<___;
+	movdqa		13*32-0x80($Tbl),$Wi
+	paddd		@MSG[0],$Wi
+	sha256msg2	@MSG[0],@MSG[1]
+	sha256msg1	@MSG[3],@MSG[2]
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 52-55
+	pshufd		\$0x0e,$Wi,$Wi
+	movdqa		@MSG[1],$TMP
+	palignr		\$4,@MSG[0],$TMP
+	paddd		$TMP,@MSG[2]
+___
+	&$aesenc();
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+
+	movdqa		14*32-0x80($Tbl),$Wi
+	paddd		@MSG[1],$Wi
+	sha256msg2	@MSG[1],@MSG[2]
+	movdqa		$BSWAP,$TMP
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 56-59
+	pshufd		\$0x0e,$Wi,$Wi
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+
+	movdqa		15*32-0x80($Tbl),$Wi
+	paddd		@MSG[2],$Wi
+___
+	&$aesenc();
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$ABEF,$CDGH		# 60-63
+	pshufd		\$0x0e,$Wi,$Wi
+___
+	&$aesenc();
+$code.=<<___;
+	sha256rnds2	$CDGH,$ABEF
+	#pxor		$CDGH,$rndkey0		# black magic
+___
+	while ($r<40)	{ &$aesenc(); }		# remaining aesenc's
+$code.=<<___;
+	#xorps		$CDGH,$rndkey0		# black magic
+	paddd		$CDGH_SAVE,$CDGH
+	paddd		$ABEF_SAVE,$ABEF
+
+	dec		$len
+	movups		$iv,48($out,$in0)	# write output
+	lea		64($in0),$in0
+	jnz		.Loop_shaext
+
+	pshufd		\$0xb1,$CDGH,$CDGH	# DCHG
+	pshufd		\$0x1b,$ABEF,$TMP	# FEBA
+	pshufd		\$0xb1,$ABEF,$ABEF	# BAFE
+	punpckhqdq	$CDGH,$ABEF		# DCBA
+	palignr		\$8,$TMP,$CDGH		# HGFE
+
+	movups		$iv,($ivp)		# write IV
+	movdqu		$ABEF,($ctx)
+	movdqu		$CDGH,16($ctx)
+___
+$code.=<<___ if ($win64);
+	movaps	0*16(%rsp),%xmm6
+	movaps	1*16(%rsp),%xmm7
+	movaps	2*16(%rsp),%xmm8
+	movaps	3*16(%rsp),%xmm9
+	movaps	4*16(%rsp),%xmm10
+	movaps	5*16(%rsp),%xmm11
+	movaps	6*16(%rsp),%xmm12
+	movaps	7*16(%rsp),%xmm13
+	movaps	8*16(%rsp),%xmm14
+	movaps	9*16(%rsp),%xmm15
+	lea	8+10*16(%rsp),%rsp
+.Lepilogue_shaext:
+___
+$code.=<<___;
+	ret
+.size	${func}_shaext,.-${func}_shaext
+___
+}
+}}}}}
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___ if ($avx);
+.extern	__imp_RtlVirtualUnwind
+.type	se_handler,\@abi-omnipotent
+.align	16
+se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HanderlData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_prologue
+
+	mov	152($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
+___
+$code.=<<___ if ($shaext);
+	lea	aesni_cbc_sha256_enc_shaext(%rip),%r10
+	cmp	%r10,%rbx
+	jb	.Lnot_in_shaext
+
+	lea	(%rax),%rsi
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	168(%rax),%rax		# adjust stack pointer
+	jmp	.Lin_prologue
+.Lnot_in_shaext:
+___
+$code.=<<___ if ($avx>1);
+	lea	.Lavx2_shortcut(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<avx2_shortcut
+	jb	.Lnot_in_avx2
+
+	and	\$-256*$SZ,%rax
+	add	\$`2*$SZ*($rounds-8)`,%rax
+.Lnot_in_avx2:
+___
+$code.=<<___;
+	mov	%rax,%rsi		# put aside Rsp
+	mov	16*$SZ+7*8(%rax),%rax	# pull $_rsp
+	lea	48(%rax),%rax
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($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($context)	# restore context->R15
+
+	lea	16*$SZ+8*8(%rsi),%rsi	# Xmm6- save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+
+.Lin_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($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	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	se_handler,.-se_handler
+
+.section	.pdata
+	.rva	.LSEH_begin_${func}_xop
+	.rva	.LSEH_end_${func}_xop
+	.rva	.LSEH_info_${func}_xop
+
+	.rva	.LSEH_begin_${func}_avx
+	.rva	.LSEH_end_${func}_avx
+	.rva	.LSEH_info_${func}_avx
+___
+$code.=<<___ if ($avx>1);
+	.rva	.LSEH_begin_${func}_avx2
+	.rva	.LSEH_end_${func}_avx2
+	.rva	.LSEH_info_${func}_avx2
+___
+$code.=<<___ if ($shaext);
+	.rva	.LSEH_begin_${func}_shaext
+	.rva	.LSEH_end_${func}_shaext
+	.rva	.LSEH_info_${func}_shaext
+___
+$code.=<<___ if ($avx);
+.section	.xdata
+.align	8
+.LSEH_info_${func}_xop:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_xop,.Lepilogue_xop		# HandlerData[]
+
+.LSEH_info_${func}_avx:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_avx,.Lepilogue_avx		# HandlerData[]
+___
+$code.=<<___ if ($avx>1);
+.LSEH_info_${func}_avx2:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_avx2,.Lepilogue_avx2		# HandlerData[]
+___
+$code.=<<___ if ($shaext);
+.LSEH_info_${func}_shaext:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_shaext,.Lepilogue_shaext	# HandlerData[]
+___
+}
+
+####################################################################
+sub rex {
+  local *opcode=shift;
+  my ($dst,$src)=@_;
+  my $rex=0;
+
+    $rex|=0x04			if($dst>=8);
+    $rex|=0x01			if($src>=8);
+    unshift @opcode,$rex|0x40	if($rex);
+}
+
+{
+  my %opcodelet = (
+		"sha256rnds2" => 0xcb,
+  		"sha256msg1"  => 0xcc,
+		"sha256msg2"  => 0xcd	);
+
+  sub sha256op38 {
+    my $instr = shift;
+
+    if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x0f,0x38);
+	rex(\@opcode,$2,$1);
+	push @opcode,$opcodelet{$instr};
+	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
+	return ".byte\t".join(',',@opcode);
+    } else {
+	return $instr."\t".@_[0];
+    }
+  }
+}
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/\b(sha256[^\s]*)\s+(.*)/sha256op38($1,$2)/gem;
+print $code;
+close STDOUT;
diff --git a/openssl/crypto/aes/asm/aesni-x86.pl b/openssl/crypto/aes/asm/aesni-x86.pl
index 3dc345b58..3deb86aed 100644
--- a/openssl/crypto/aes/asm/aesni-x86.pl
+++ b/openssl/crypto/aes/asm/aesni-x86.pl
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 
 # ====================================================================
-# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# 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/.
@@ -43,6 +43,17 @@
 # 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:-)
@@ -54,8 +65,8 @@ require "x86asm.pl";
 
 &asm_init($ARGV[0],$0);
 
-if ($PREFIX eq "aesni")	{ $movekey=*movups; }
-else			{ $movekey=*movups; }
+if ($PREFIX eq "aesni")	{ $movekey=\&movups; }
+else			{ $movekey=\&movups; }
 
 $len="eax";
 $rounds="ecx";
@@ -196,37 +207,71 @@ sub aesni_generate1	# fully unrolled loop
 # 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 makes no sense to implement 2x subroutine.
-# aes[enc|dec] latency in next processor generation is 8, but the
-# instructions can be scheduled every cycle. Optimal interleave for
-# new processor is therefore 8x, but it's unfeasible to accommodate it
-# in XMM registers addreassable in 32-bit mode and therefore 6x is
-# used instead...
+# 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));
-	&shr		($rounds,1);
+	&shl		($rounds,4);
 	&$movekey	($rndkey1,&QWP(16,$key));
-	&lea		($key,&DWP(32,$key));
 	&xorps		($inout0,$rndkey0);
 	&pxor		($inout1,$rndkey0);
 	&pxor		($inout2,$rndkey0);
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($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,$rndkey1)";
-	&dec		($rounds);
 	eval"&aes${p}	($inout2,$rndkey1)";
-	&$movekey	($rndkey1,&QWP(16,$key));
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
 	eval"&aes${p}	($inout0,$rndkey0)";
 	eval"&aes${p}	($inout1,$rndkey0)";
-	&lea		($key,&DWP(32,$key));
 	eval"&aes${p}	($inout2,$rndkey0)";
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
 	&jnz		(&label("${p}3_loop"));
     eval"&aes${p}	($inout0,$rndkey1)";
     eval"&aes${p}	($inout1,$rndkey1)";
@@ -248,27 +293,29 @@ sub aesni_generate4
     &function_begin_B("_aesni_${p}rypt4");
 	&$movekey	($rndkey0,&QWP(0,$key));
 	&$movekey	($rndkey1,&QWP(16,$key));
-	&shr		($rounds,1);
-	&lea		($key,&DWP(32,$key));
+	&shl		($rounds,4);
 	&xorps		($inout0,$rndkey0);
 	&pxor		($inout1,$rndkey0);
 	&pxor		($inout2,$rndkey0);
 	&pxor		($inout3,$rndkey0);
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($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,$rndkey1)";
-	&dec		($rounds);
 	eval"&aes${p}	($inout2,$rndkey1)";
 	eval"&aes${p}	($inout3,$rndkey1)";
-	&$movekey	($rndkey1,&QWP(16,$key));
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
 	eval"&aes${p}	($inout0,$rndkey0)";
 	eval"&aes${p}	($inout1,$rndkey0)";
-	&lea		($key,&DWP(32,$key));
 	eval"&aes${p}	($inout2,$rndkey0)";
 	eval"&aes${p}	($inout3,$rndkey0)";
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
     &jnz		(&label("${p}4_loop"));
 
     eval"&aes${p}	($inout0,$rndkey1)";
@@ -289,43 +336,43 @@ sub aesni_generate6
     &function_begin_B("_aesni_${p}rypt6");
     &static_label("_aesni_${p}rypt6_enter");
 	&$movekey	($rndkey0,&QWP(0,$key));
-	&shr		($rounds,1);
+	&shl		($rounds,4);
 	&$movekey	($rndkey1,&QWP(16,$key));
-	&lea		($key,&DWP(32,$key));
 	&xorps		($inout0,$rndkey0);
 	&pxor		($inout1,$rndkey0);	# pxor does better here
-	eval"&aes${p}	($inout0,$rndkey1)";
 	&pxor		($inout2,$rndkey0);
-	eval"&aes${p}	($inout1,$rndkey1)";
+	eval"&aes${p}	($inout0,$rndkey1)";
 	&pxor		($inout3,$rndkey0);
-	&dec		($rounds);
-	eval"&aes${p}	($inout2,$rndkey1)";
 	&pxor		($inout4,$rndkey0);
-	eval"&aes${p}	($inout3,$rndkey1)";
+	eval"&aes${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)";
-	&$movekey	($rndkey0,&QWP(0,$key));
 	eval"&aes${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,$rndkey1)";
-	&dec		($rounds);
 	eval"&aes${p}	($inout2,$rndkey1)";
 	eval"&aes${p}	($inout3,$rndkey1)";
 	eval"&aes${p}	($inout4,$rndkey1)";
 	eval"&aes${p}	($inout5,$rndkey1)";
-    &set_label("_aesni_${p}rypt6_enter",16);
-	&$movekey	($rndkey1,&QWP(16,$key));
+    &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)";
-	&lea		($key,&DWP(32,$key));
 	eval"&aes${p}	($inout2,$rndkey0)";
 	eval"&aes${p}	($inout3,$rndkey0)";
 	eval"&aes${p}	($inout4,$rndkey0)";
 	eval"&aes${p}	($inout5,$rndkey0)";
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
     &jnz		(&label("${p}6_loop"));
 
     eval"&aes${p}	($inout0,$rndkey1)";
@@ -343,6 +390,8 @@ sub aesni_generate6
     &ret();
     &function_end_B("_aesni_${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 ($PREFIX eq "aesni");
@@ -446,8 +495,7 @@ if ($PREFIX eq "aesni") {
 	&jmp	(&label("ecb_ret"));
 
 &set_label("ecb_enc_two",16);
-	&xorps	($inout2,$inout2);
-	&call	("_aesni_encrypt3");
+	&call	("_aesni_encrypt2");
 	&movups	(&QWP(0,$out),$inout0);
 	&movups	(&QWP(0x10,$out),$inout1);
 	&jmp	(&label("ecb_ret"));
@@ -547,8 +595,7 @@ if ($PREFIX eq "aesni") {
 	&jmp	(&label("ecb_ret"));
 
 &set_label("ecb_dec_two",16);
-	&xorps	($inout2,$inout2);
-	&call	("_aesni_decrypt3");
+	&call	("_aesni_decrypt2");
 	&movups	(&QWP(0,$out),$inout0);
 	&movups	(&QWP(0x10,$out),$inout1);
 	&jmp	(&label("ecb_ret"));
@@ -610,11 +657,13 @@ if ($PREFIX eq "aesni") {
 	&mov	(&DWP(24,"esp"),$key_);
 	&mov	(&DWP(28,"esp"),$key_);
 
-	&shr	($rounds,1);
+	&shl	($rounds,4);
+	&mov	($rounds_,16);
 	&lea	($key_,&DWP(0,$key));
 	&movdqa	($inout3,&QWP(0,"esp"));
 	&movdqa	($inout0,$ivec);
-	&mov	($rounds_,$rounds);
+	&lea	($key,&DWP(32,$key,$rounds));
+	&sub	($rounds_,$rounds);
 	&pshufb	($ivec,$inout3);
 
 &set_label("ccm64_enc_outer");
@@ -625,33 +674,31 @@ if ($PREFIX eq "aesni") {
 	&xorps		($inout0,$rndkey0);
 	&$movekey	($rndkey1,&QWP(16,$key_));
 	&xorps		($rndkey0,$in0);
-	&lea		($key,&DWP(32,$key_));
 	&xorps		($cmac,$rndkey0);		# cmac^=inp
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey0,&QWP(32,$key_));
 
 &set_label("ccm64_enc2_loop");
 	&aesenc		($inout0,$rndkey1);
-	&dec		($rounds);
 	&aesenc		($cmac,$rndkey1);
-	&$movekey	($rndkey1,&QWP(16,$key));
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
 	&aesenc		($inout0,$rndkey0);
-	&lea		($key,&DWP(32,$key));
 	&aesenc		($cmac,$rndkey0);
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($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);
 
-	&dec	($len);
 	&lea	($inp,&DWP(16,$inp));
 	&xorps	($in0,$inout0);			# inp^=E(ivec)
 	&movdqa	($inout0,$ivec);
 	&movups	(&QWP(0,$out),$in0);		# save output
-	&lea	($out,&DWP(16,$out));
 	&pshufb	($inout0,$inout3);
+	&lea	($out,&DWP(16,$out));
 	&jnz	(&label("ccm64_enc_outer"));
 
 	&mov	("esp",&DWP(48,"esp"));
@@ -700,15 +747,19 @@ if ($PREFIX eq "aesni") {
 	{   &aesni_inline_generate1("enc");	}
 	else
 	{   &call	("_aesni_encrypt1");	}
+	&shl	($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);
-	&mov	($rounds,$rounds_);
 	&movups	(&QWP(0,$out),$in0);		# save output
 	&lea	($out,&DWP(16,$out));
 	&pshufb	($inout0,$inout3);
@@ -717,34 +768,33 @@ if ($PREFIX eq "aesni") {
 	&jz	(&label("ccm64_dec_break"));
 
 	&$movekey	($rndkey0,&QWP(0,$key_));
-	&shr		($rounds,1);
+	&mov		($rounds,$rounds_);
 	&$movekey	($rndkey1,&QWP(16,$key_));
 	&xorps		($in0,$rndkey0);
-	&lea		($key,&DWP(32,$key_));
 	&xorps		($inout0,$rndkey0);
 	&xorps		($cmac,$in0);		# cmac^=out
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey0,&QWP(32,$key_));
 
 &set_label("ccm64_dec2_loop");
 	&aesenc		($inout0,$rndkey1);
-	&dec		($rounds);
 	&aesenc		($cmac,$rndkey1);
-	&$movekey	($rndkey1,&QWP(16,$key));
+	&$movekey	($rndkey1,&QWP(0,$key,$rounds));
+	&add		($rounds,32);
 	&aesenc		($inout0,$rndkey0);
-	&lea		($key,&DWP(32,$key));
 	&aesenc		($cmac,$rndkey0);
-	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($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,$rndkey1);
-	&lea		($inp,&QWP(16,$inp));
 	&aesenclast	($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,$in0);	}
@@ -763,7 +813,7 @@ if ($PREFIX eq "aesni") {
 #                         const char *ivec);
 #
 # Handles only complete blocks, operates on 32-bit counter and
-# does not update *ivec! (see engine/eng_aesni.c for details)
+# does not update *ivec! (see crypto/modes/ctr128.c for details)
 #
 # stack layout:
 #	0	pshufb mask
@@ -810,66 +860,61 @@ if ($PREFIX eq "aesni") {
 
 	# compose 2 vectors of 3x32-bit counters
 	&bswap	($rounds_);
-	&pxor	($rndkey1,$rndkey1);
 	&pxor	($rndkey0,$rndkey0);
+	&pxor	($rndkey1,$rndkey1);
 	&movdqa	($inout0,&QWP(0,"esp"));	# load byte-swap mask
-	&pinsrd	($rndkey1,$rounds_,0);
+	&pinsrd	($rndkey0,$rounds_,0);
 	&lea	($key_,&DWP(3,$rounds_));
-	&pinsrd	($rndkey0,$key_,0);
+	&pinsrd	($rndkey1,$key_,0);
 	&inc	($rounds_);
-	&pinsrd	($rndkey1,$rounds_,1);
+	&pinsrd	($rndkey0,$rounds_,1);
 	&inc	($key_);
-	&pinsrd	($rndkey0,$key_,1);
+	&pinsrd	($rndkey1,$key_,1);
 	&inc	($rounds_);
-	&pinsrd	($rndkey1,$rounds_,2);
+	&pinsrd	($rndkey0,$rounds_,2);
 	&inc	($key_);
-	&pinsrd	($rndkey0,$key_,2);
-	&movdqa	(&QWP(48,"esp"),$rndkey1);	# save 1st triplet
-	&pshufb	($rndkey1,$inout0);		# byte swap
-	&movdqa	(&QWP(64,"esp"),$rndkey0);	# save 2nd triplet
+	&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	($inout0,$rndkey1,3<<6);	# place counter to upper dword
-	&pshufd	($inout1,$rndkey1,2<<6);
+	&pshufd	($inout0,$rndkey0,3<<6);	# place counter to upper dword
+	&pshufd	($inout1,$rndkey0,2<<6);
 	&cmp	($len,6);
 	&jb	(&label("ctr32_tail"));
-	&movdqa	(&QWP(32,"esp"),$inout5);	# save counter-less ivec
-	&shr	($rounds,1);
+	&pxor	($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 $key
-	&mov	($rounds_,$rounds);		# backup $rounds
+	&sub	($rounds_,$rounds);		# backup twisted $rounds
+	&lea	($key,&DWP(32,$key,$rounds));
 	&sub	($len,6);
 	&jmp	(&label("ctr32_loop6"));
 
 &set_label("ctr32_loop6",16);
-	&pshufd	($inout2,$rndkey1,1<<6);
-	&movdqa	($rndkey1,&QWP(32,"esp"));	# pull counter-less ivec
-	&pshufd	($inout3,$rndkey0,3<<6);
-	&por	($inout0,$rndkey1);		# merge counter-less ivec
-	&pshufd	($inout4,$rndkey0,2<<6);
-	&por	($inout1,$rndkey1);
-	&pshufd	($inout5,$rndkey0,1<<6);
-	&por	($inout2,$rndkey1);
-	&por	($inout3,$rndkey1);
-	&por	($inout4,$rndkey1);
-	&por	($inout5,$rndkey1);
-
-	# inlining _aesni_encrypt6's prologue gives ~4% improvement...
-	&$movekey	($rndkey0,&QWP(0,$key_));
-	&$movekey	($rndkey1,&QWP(16,$key_));
-	&lea		($key,&DWP(32,$key_));
-	&dec		($rounds);
-	&pxor		($inout0,$rndkey0);
+	# 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,$rndkey0);
-	&aesenc		($inout0,$rndkey1);
+	&pshufd	($inout5,$rndkey1,1<<6);
+	&$movekey	($rndkey1,&QWP(16,$key_));
 	&pxor		($inout2,$rndkey0);
-	&aesenc		($inout1,$rndkey1);
 	&pxor		($inout3,$rndkey0);
-	&aesenc		($inout2,$rndkey1);
+	&aesenc		($inout0,$rndkey1);
 	&pxor		($inout4,$rndkey0);
-	&aesenc		($inout3,$rndkey1);
 	&pxor		($inout5,$rndkey0);
+	&aesenc		($inout1,$rndkey1);
+	&$movekey	($rndkey0,&QWP(32,$key_));
+	&mov		($rounds,$rounds_);
+	&aesenc		($inout2,$rndkey1);
+	&aesenc		($inout3,$rndkey1);
 	&aesenc		($inout4,$rndkey1);
-	&$movekey	($rndkey0,&QWP(0,$key));
 	&aesenc		($inout5,$rndkey1);
 
 	&call		(&label("_aesni_encrypt6_enter"));
@@ -882,12 +927,12 @@ if ($PREFIX eq "aesni") {
 	&movups	(&QWP(0,$out),$inout0);
 	&movdqa	($rndkey0,&QWP(16,"esp"));	# load increment
 	&xorps	($inout2,$rndkey1);
-	&movdqa	($rndkey1,&QWP(48,"esp"));	# load 1st triplet
+	&movdqa	($rndkey1,&QWP(64,"esp"));	# load 2nd triplet
 	&movups	(&QWP(0x10,$out),$inout1);
 	&movups	(&QWP(0x20,$out),$inout2);
 
-	&paddd	($rndkey1,$rndkey0);		# 1st triplet increment
-	&paddd	($rndkey0,&QWP(64,"esp"));	# 2nd triplet increment
+	&paddd	($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));
@@ -895,44 +940,44 @@ if ($PREFIX eq "aesni") {
 	&xorps	($inout3,$inout1);
 	&movups	($inout1,&QWP(0x50,$inp));
 	&lea	($inp,&DWP(0x60,$inp));
-	&movdqa	(&QWP(48,"esp"),$rndkey1);	# save 1st triplet
-	&pshufb	($rndkey1,$inout0);		# byte swap
+	&movdqa	(&QWP(48,"esp"),$rndkey0);	# save 1st triplet
+	&pshufb	($rndkey0,$inout0);		# byte swap
 	&xorps	($inout4,$inout2);
 	&movups	(&QWP(0x30,$out),$inout3);
 	&xorps	($inout5,$inout1);
-	&movdqa	(&QWP(64,"esp"),$rndkey0);	# save 2nd triplet
-	&pshufb	($rndkey0,$inout0);		# byte swap
+	&movdqa	(&QWP(64,"esp"),$rndkey1);	# save 2nd triplet
+	&pshufb	($rndkey1,$inout0);		# byte swap
 	&movups	(&QWP(0x40,$out),$inout4);
-	&pshufd	($inout0,$rndkey1,3<<6);
+	&pshufd	($inout0,$rndkey0,3<<6);
 	&movups	(&QWP(0x50,$out),$inout5);
 	&lea	($out,&DWP(0x60,$out));
 
-	&mov	($rounds,$rounds_);
-	&pshufd	($inout1,$rndkey1,2<<6);
+	&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,$key_);
-	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
-	&movdqa	($inout5,&QWP(32,"esp"));	# pull count-less ivec
+	&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	($inout2,$rndkey1,1<<6);
+	&pshufd	($inout2,$rndkey0,1<<6);
 	&por	($inout1,$inout5);
 	&je	(&label("ctr32_two"));
 
-	&pshufd	($inout3,$rndkey0,3<<6);
+	&pshufd	($inout3,$rndkey1,3<<6);
 	&por	($inout2,$inout5);
 	&cmp	($len,4);
 	&jb	(&label("ctr32_three"));
 
-	&pshufd	($inout4,$rndkey0,2<<6);
+	&pshufd	($inout4,$rndkey1,2<<6);
 	&por	($inout3,$inout5);
 	&je	(&label("ctr32_four"));
 
@@ -970,7 +1015,7 @@ if ($PREFIX eq "aesni") {
 	&jmp	(&label("ctr32_ret"));
 
 &set_label("ctr32_two",16);
-	&call	("_aesni_encrypt3");
+	&call	("_aesni_encrypt2");
 	&movups	($inout3,&QWP(0,$inp));
 	&movups	($inout4,&QWP(0x10,$inp));
 	&xorps	($inout0,$inout3);
@@ -1057,8 +1102,10 @@ if ($PREFIX eq "aesni") {
 	&sub	($len,16*6);
 	&jc	(&label("xts_enc_short"));
 
-	&shr	($rounds,1);
-	&mov	($rounds_,$rounds);
+	&shl	($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);
@@ -1080,6 +1127,7 @@ if ($PREFIX eq "aesni") {
 	&pxor	($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,$inp));
@@ -1096,19 +1144,17 @@ if ($PREFIX eq "aesni") {
 	&pxor	($inout5,$rndkey1);
 
 	 &$movekey	($rndkey1,&QWP(16,$key_));
-	 &lea		($key,&DWP(32,$key_));
 	&pxor	($inout1,&QWP(16*1,"esp"));
-	 &aesenc	($inout0,$rndkey1);
 	&pxor	($inout2,&QWP(16*2,"esp"));
-	 &aesenc	($inout1,$rndkey1);
+	 &aesenc	($inout0,$rndkey1);
 	&pxor	($inout3,&QWP(16*3,"esp"));
-	 &dec		($rounds);
-	 &aesenc	($inout2,$rndkey1);
 	&pxor	($inout4,&QWP(16*4,"esp"));
-	 &aesenc	($inout3,$rndkey1);
+	 &aesenc	($inout1,$rndkey1);
 	&pxor		($inout5,$rndkey0);
+	 &$movekey	($rndkey0,&QWP(32,$key_));
+	 &aesenc	($inout2,$rndkey1);
+	 &aesenc	($inout3,$rndkey1);
 	 &aesenc	($inout4,$rndkey1);
-	 &$movekey	($rndkey0,&QWP(0,$key));
 	 &aesenc	($inout5,$rndkey1);
 	&call		(&label("_aesni_encrypt6_enter"));
 
@@ -1135,13 +1181,12 @@ if ($PREFIX eq "aesni") {
 	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
 	&pand	($twres,$twmask);		# isolate carry and residue
 	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
-	&mov	($rounds,$rounds_);		# restore $rounds
 	&pxor	($tweak,$twres);
 
 	&sub	($len,16*6);
 	&jnc	(&label("xts_enc_loop6"));
 
-	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
+	&mov	($rounds,&DWP(240,$key_));	# restore $rounds
 	&mov	($key,$key_);			# restore $key
 	&mov	($rounds_,$rounds);
 
@@ -1241,9 +1286,8 @@ if ($PREFIX eq "aesni") {
 	&lea	($inp,&DWP(16*2,$inp));
 	&xorps	($inout0,$inout3);		# input^=tweak
 	&xorps	($inout1,$inout4);
-	&xorps	($inout2,$inout2);
 
-	&call	("_aesni_encrypt3");
+	&call	("_aesni_encrypt2");
 
 	&xorps	($inout0,$inout3);		# output^=tweak
 	&xorps	($inout1,$inout4);
@@ -1399,8 +1443,10 @@ if ($PREFIX eq "aesni") {
 	&sub	($len,16*6);
 	&jc	(&label("xts_dec_short"));
 
-	&shr	($rounds,1);
-	&mov	($rounds_,$rounds);
+	&shl	($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);
@@ -1422,6 +1468,7 @@ if ($PREFIX eq "aesni") {
 	&pxor	($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));
@@ -1438,19 +1485,17 @@ if ($PREFIX eq "aesni") {
 	&pxor	($inout5,$rndkey1);
 
 	 &$movekey	($rndkey1,&QWP(16,$key_));
-	 &lea		($key,&DWP(32,$key_));
 	&pxor	($inout1,&QWP(16*1,"esp"));
-	 &aesdec	($inout0,$rndkey1);
 	&pxor	($inout2,&QWP(16*2,"esp"));
-	 &aesdec	($inout1,$rndkey1);
+	 &aesdec	($inout0,$rndkey1);
 	&pxor	($inout3,&QWP(16*3,"esp"));
-	 &dec		($rounds);
-	 &aesdec	($inout2,$rndkey1);
 	&pxor	($inout4,&QWP(16*4,"esp"));
-	 &aesdec	($inout3,$rndkey1);
+	 &aesdec	($inout1,$rndkey1);
 	&pxor		($inout5,$rndkey0);
+	 &$movekey	($rndkey0,&QWP(32,$key_));
+	 &aesdec	($inout2,$rndkey1);
+	 &aesdec	($inout3,$rndkey1);
 	 &aesdec	($inout4,$rndkey1);
-	 &$movekey	($rndkey0,&QWP(0,$key));
 	 &aesdec	($inout5,$rndkey1);
 	&call		(&label("_aesni_decrypt6_enter"));
 
@@ -1477,13 +1522,12 @@ if ($PREFIX eq "aesni") {
 	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
 	&pand	($twres,$twmask);		# isolate carry and residue
 	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
-	&mov	($rounds,$rounds_);		# restore $rounds
 	&pxor	($tweak,$twres);
 
 	&sub	($len,16*6);
 	&jnc	(&label("xts_dec_loop6"));
 
-	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
+	&mov	($rounds,&DWP(240,$key_));	# restore $rounds
 	&mov	($key,$key_);			# restore $key
 	&mov	($rounds_,$rounds);
 
@@ -1584,7 +1628,7 @@ if ($PREFIX eq "aesni") {
 	&xorps	($inout0,$inout3);		# input^=tweak
 	&xorps	($inout1,$inout4);
 
-	&call	("_aesni_decrypt3");
+	&call	("_aesni_decrypt2");
 
 	&xorps	($inout0,$inout3);		# output^=tweak
 	&xorps	($inout1,$inout4);
@@ -1816,7 +1860,7 @@ if ($PREFIX eq "aesni") {
 	&movups	(&QWP(0x10,$out),$inout1);
 	&lea	($inp,&DWP(0x60,$inp));
 	&movups	(&QWP(0x20,$out),$inout2);
-	&mov	($rounds,$rounds_)		# restore $rounds
+	&mov	($rounds,$rounds_);		# restore $rounds
 	&movups	(&QWP(0x30,$out),$inout3);
 	&mov	($key,$key_);			# restore $key
 	&movups	(&QWP(0x40,$out),$inout4);
@@ -1884,8 +1928,7 @@ if ($PREFIX eq "aesni") {
 	&jmp	(&label("cbc_dec_tail_collected"));
 
 &set_label("cbc_dec_two",16);
-	&xorps	($inout2,$inout2);
-	&call	("_aesni_decrypt3");
+	&call	("_aesni_decrypt2");
 	&xorps	($inout0,$ivec);
 	&xorps	($inout1,$in0);
 	&movups	(&QWP(0,$out),$inout0);
@@ -2015,7 +2058,7 @@ if ($PREFIX eq "aesni") {
 &set_label("12rounds",16);
 	&movq		("xmm2",&QWP(16,"eax"));	# remaining 1/3 of *userKey
 	&mov		($rounds,11);
-	&$movekey	(&QWP(-16,$key),"xmm0")		# round 0
+	&$movekey	(&QWP(-16,$key),"xmm0");	# round 0
 	&aeskeygenassist("xmm1","xmm2",0x01);		# round 1,2
 	&call		(&label("key_192a_cold"));
 	&aeskeygenassist("xmm1","xmm2",0x02);		# round 2,3
@@ -2152,7 +2195,7 @@ if ($PREFIX eq "aesni") {
 	&mov	($key,&wparam(2));
 	&call	("_aesni_set_encrypt_key");
 	&mov	($key,&wparam(2));
-	&shl	($rounds,4)	# rounds-1 after _aesni_set_encrypt_key
+	&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
diff --git a/openssl/crypto/aes/asm/aesni-x86_64.pl b/openssl/crypto/aes/asm/aesni-x86_64.pl
index c9270dfdd..5f6174635 100644
--- a/openssl/crypto/aes/asm/aesni-x86_64.pl
+++ b/openssl/crypto/aes/asm/aesni-x86_64.pl
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 #
 # ====================================================================
-# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# 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/.
@@ -129,8 +129,8 @@
 #
 # Further data for other parallelizable modes:
 #
-# CBC decrypt				1.16	0.93	0.93
-# CTR					1.14	0.91	n/a
+# 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?
@@ -153,10 +153,25 @@
 
 # April 2011
 #
-# Add aesni_xts_[en|de]crypt. Westmere spends 1.33 cycles processing
-# one byte out of 8KB with 128-bit key, Sandy Bridge - 0.97. Just like
+# 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:-)
@@ -180,6 +195,7 @@ $movkey = $PREFIX eq "aesni" ? "movups" : "movups";
 		("%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 ...
@@ -272,10 +288,49 @@ ___
 # 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 makes no sense to implement 2x subroutine.
-# aes[enc|dec] latency in next processor generation is 8, but the
-# instructions can be scheduled every cycle. Optimal interleave for
-# new processor is therefore 8x...
+# 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*
@@ -285,25 +340,26 @@ $code.=<<___;
 .align	16
 _aesni_${dir}rypt3:
 	$movkey	($key),$rndkey0
-	shr	\$1,$rounds
+	shl	\$4,$rounds
 	$movkey	16($key),$rndkey1
-	lea	32($key),$key
 	xorps	$rndkey0,$inout0
 	xorps	$rndkey0,$inout1
 	xorps	$rndkey0,$inout2
-	$movkey		($key),$rndkey0
+	$movkey	32($key),$rndkey0
+	lea	32($key,$rounds),$key
+	neg	%rax				# $rounds
+	add	\$16,%rax
 
 .L${dir}_loop3:
 	aes${dir}	$rndkey1,$inout0
 	aes${dir}	$rndkey1,$inout1
-	dec		$rounds
 	aes${dir}	$rndkey1,$inout2
-	$movkey		16($key),$rndkey1
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
 	aes${dir}	$rndkey0,$inout0
 	aes${dir}	$rndkey0,$inout1
-	lea		32($key),$key
 	aes${dir}	$rndkey0,$inout2
-	$movkey		($key),$rndkey0
+	$movkey		-16($key,%rax),$rndkey0
 	jnz		.L${dir}_loop3
 
 	aes${dir}	$rndkey1,$inout0
@@ -329,28 +385,30 @@ $code.=<<___;
 .align	16
 _aesni_${dir}rypt4:
 	$movkey	($key),$rndkey0
-	shr	\$1,$rounds
+	shl	\$4,$rounds
 	$movkey	16($key),$rndkey1
-	lea	32($key),$key
 	xorps	$rndkey0,$inout0
 	xorps	$rndkey0,$inout1
 	xorps	$rndkey0,$inout2
 	xorps	$rndkey0,$inout3
-	$movkey	($key),$rndkey0
+	$movkey	32($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
-	dec		$rounds
 	aes${dir}	$rndkey1,$inout2
 	aes${dir}	$rndkey1,$inout3
-	$movkey		16($key),$rndkey1
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
 	aes${dir}	$rndkey0,$inout0
 	aes${dir}	$rndkey0,$inout1
-	lea		32($key),$key
 	aes${dir}	$rndkey0,$inout2
 	aes${dir}	$rndkey0,$inout3
-	$movkey		($key),$rndkey0
+	$movkey		-16($key,%rax),$rndkey0
 	jnz		.L${dir}_loop4
 
 	aes${dir}	$rndkey1,$inout0
@@ -374,43 +432,43 @@ $code.=<<___;
 .align	16
 _aesni_${dir}rypt6:
 	$movkey		($key),$rndkey0
-	shr		\$1,$rounds
+	shl		\$4,$rounds
 	$movkey		16($key),$rndkey1
-	lea		32($key),$key
 	xorps		$rndkey0,$inout0
 	pxor		$rndkey0,$inout1
-	aes${dir}	$rndkey1,$inout0
 	pxor		$rndkey0,$inout2
+	aes${dir}	$rndkey1,$inout0
+	lea		32($key,$rounds),$key
+	neg		%rax			# $rounds
 	aes${dir}	$rndkey1,$inout1
 	pxor		$rndkey0,$inout3
-	aes${dir}	$rndkey1,$inout2
 	pxor		$rndkey0,$inout4
-	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey1,$inout2
 	pxor		$rndkey0,$inout5
-	dec		$rounds
+	add		\$16,%rax
+	aes${dir}	$rndkey1,$inout3
 	aes${dir}	$rndkey1,$inout4
-	$movkey		($key),$rndkey0
 	aes${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
-	dec		$rounds
 	aes${dir}	$rndkey1,$inout2
 	aes${dir}	$rndkey1,$inout3
 	aes${dir}	$rndkey1,$inout4
 	aes${dir}	$rndkey1,$inout5
-.L${dir}_loop6_enter:				# happens to be 16-byte aligned
-	$movkey		16($key),$rndkey1
+.L${dir}_loop6_enter:
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
 	aes${dir}	$rndkey0,$inout0
 	aes${dir}	$rndkey0,$inout1
-	lea		32($key),$key
 	aes${dir}	$rndkey0,$inout2
 	aes${dir}	$rndkey0,$inout3
 	aes${dir}	$rndkey0,$inout4
 	aes${dir}	$rndkey0,$inout5
-	$movkey		($key),$rndkey0
+	$movkey		-16($key,%rax),$rndkey0
 	jnz		.L${dir}_loop6
 
 	aes${dir}	$rndkey1,$inout0
@@ -438,52 +496,51 @@ $code.=<<___;
 .align	16
 _aesni_${dir}rypt8:
 	$movkey		($key),$rndkey0
-	shr		\$1,$rounds
+	shl		\$4,$rounds
 	$movkey		16($key),$rndkey1
-	lea		32($key),$key
 	xorps		$rndkey0,$inout0
 	xorps		$rndkey0,$inout1
-	aes${dir}	$rndkey1,$inout0
 	pxor		$rndkey0,$inout2
-	aes${dir}	$rndkey1,$inout1
 	pxor		$rndkey0,$inout3
-	aes${dir}	$rndkey1,$inout2
 	pxor		$rndkey0,$inout4
-	aes${dir}	$rndkey1,$inout3
+	lea		32($key,$rounds),$key
+	neg		%rax			# $rounds
+	aes${dir}	$rndkey1,$inout0
+	add		\$16,%rax
 	pxor		$rndkey0,$inout5
-	dec		$rounds
-	aes${dir}	$rndkey1,$inout4
+	aes${dir}	$rndkey1,$inout1
 	pxor		$rndkey0,$inout6
-	aes${dir}	$rndkey1,$inout5
 	pxor		$rndkey0,$inout7
-	$movkey		($key),$rndkey0
+	aes${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),$rndkey1
+	$movkey		-16($key,%rax),$rndkey0
 	jmp		.L${dir}_loop8_enter
 .align	16
 .L${dir}_loop8:
 	aes${dir}	$rndkey1,$inout0
 	aes${dir}	$rndkey1,$inout1
-	dec		$rounds
 	aes${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),$rndkey1
-.L${dir}_loop8_enter:				# happens to be 16-byte aligned
+.L${dir}_loop8_enter:
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
 	aes${dir}	$rndkey0,$inout0
 	aes${dir}	$rndkey0,$inout1
-	lea		32($key),$key
 	aes${dir}	$rndkey0,$inout2
 	aes${dir}	$rndkey0,$inout3
 	aes${dir}	$rndkey0,$inout4
 	aes${dir}	$rndkey0,$inout5
 	aes${dir}	$rndkey0,$inout6
 	aes${dir}	$rndkey0,$inout7
-	$movkey		($key),$rndkey0
+	$movkey		-16($key,%rax),$rndkey0
 	jnz		.L${dir}_loop8
 
 	aes${dir}	$rndkey1,$inout0
@@ -506,6 +563,8 @@ _aesni_${dir}rypt8:
 .size	_aesni_${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");
@@ -637,8 +696,7 @@ $code.=<<___;
 	jmp	.Lecb_ret
 .align	16
 .Lecb_enc_two:
-	xorps	$inout2,$inout2
-	call	_aesni_encrypt3
+	call	_aesni_encrypt2
 	movups	$inout0,($out)
 	movups	$inout1,0x10($out)
 	jmp	.Lecb_ret
@@ -774,8 +832,7 @@ $code.=<<___;
 	jmp	.Lecb_ret
 .align	16
 .Lecb_dec_two:
-	xorps	$inout2,$inout2
-	call	_aesni_decrypt3
+	call	_aesni_decrypt2
 	movups	$inout0,($out)
 	movups	$inout1,0x10($out)
 	jmp	.Lecb_ret
@@ -842,7 +899,8 @@ ___
 {
 my $cmac="%r9";	# 6th argument
 
-my $increment="%xmm6";
+my $increment="%xmm9";
+my $iv="%xmm6";
 my $bswap_mask="%xmm7";
 
 $code.=<<___;
@@ -865,49 +923,49 @@ $code.=<<___;
 	movdqa	.Lincrement64(%rip),$increment
 	movdqa	.Lbswap_mask(%rip),$bswap_mask
 
-	shr	\$1,$rounds
+	shl	\$4,$rounds
+	mov	\$16,$rnds_
 	lea	0($key),$key_
 	movdqu	($cmac),$inout1
 	movdqa	$iv,$inout0
-	mov	$rounds,$rnds_
+	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	$rnds_,$rounds
+	mov	%r10,%rax
 	movups	($inp),$in0			# load inp
 
 	xorps	$rndkey0,$inout0		# counter
 	$movkey	16($key_),$rndkey1
 	xorps	$in0,$rndkey0
-	lea	32($key_),$key
 	xorps	$rndkey0,$inout1		# cmac^=inp
-	$movkey	($key),$rndkey0
+	$movkey	32($key_),$rndkey0
 
 .Lccm64_enc2_loop:
 	aesenc	$rndkey1,$inout0
-	dec	$rounds
 	aesenc	$rndkey1,$inout1
-	$movkey	16($key),$rndkey1
+	$movkey	($key,%rax),$rndkey1
+	add	\$32,%rax
 	aesenc	$rndkey0,$inout0
-	lea	32($key),$key
 	aesenc	$rndkey0,$inout1
-	$movkey	0($key),$rndkey0
+	$movkey	-16($key,%rax),$rndkey0
 	jnz	.Lccm64_enc2_loop
 	aesenc	$rndkey1,$inout0
 	aesenc	$rndkey1,$inout1
 	paddq	$increment,$iv
+	dec	$len
 	aesenclast	$rndkey0,$inout0
 	aesenclast	$rndkey0,$inout1
 
-	dec	$len
 	lea	16($inp),$inp
 	xorps	$inout0,$in0			# inp ^= E(iv)
 	movdqa	$iv,$inout0
 	movups	$in0,($out)			# save output
-	lea	16($out),$out
 	pshufb	$bswap_mask,$inout0
+	lea	16($out),$out
 	jnz	.Lccm64_enc_outer
 
 	movups	$inout1,($cmac)
@@ -953,15 +1011,19 @@ $code.=<<___;
 ___
 	&aesni_generate1("enc",$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
-	mov	$rnds_,$rounds
 	movups	$in0,($out)			# save output
 	lea	16($out),$out
 	pshufb	$bswap_mask,$inout0
@@ -970,36 +1032,36 @@ $code.=<<___;
 	jz	.Lccm64_dec_break
 
 	$movkey	($key_),$rndkey0
-	shr	\$1,$rounds
+	mov	%r10,%rax
 	$movkey	16($key_),$rndkey1
 	xorps	$rndkey0,$in0
-	lea	32($key_),$key
 	xorps	$rndkey0,$inout0
 	xorps	$in0,$inout1			# cmac^=out
-	$movkey	($key),$rndkey0
-
+	$movkey	32($key_),$rndkey0
+	jmp	.Lccm64_dec2_loop
+.align	16
 .Lccm64_dec2_loop:
 	aesenc	$rndkey1,$inout0
-	dec	$rounds
 	aesenc	$rndkey1,$inout1
-	$movkey	16($key),$rndkey1
+	$movkey	($key,%rax),$rndkey1
+	add	\$32,%rax
 	aesenc	$rndkey0,$inout0
-	lea	32($key),$key
 	aesenc	$rndkey0,$inout1
-	$movkey	0($key),$rndkey0
+	$movkey	-16($key,%rax),$rndkey0
 	jnz	.Lccm64_dec2_loop
 	movups	($inp),$in0			# load inp
 	paddq	$increment,$iv
 	aesenc	$rndkey1,$inout0
 	aesenc	$rndkey1,$inout1
-	lea	16($inp),$inp
 	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.=<<___;
@@ -1024,220 +1086,479 @@ ___
 #                         const char *ivec);
 #
 # Handles only complete blocks, operates on 32-bit counter and
-# does not update *ivec! (see engine/eng_aesni.c for details)
+# 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 $reserved = $win64?0:-0x28;
-my ($in0,$in1,$in2,$in3)=map("%xmm$_",(8..11));
-my ($iv0,$iv1,$ivec)=("%xmm12","%xmm13","%xmm14");
-my $bswap_mask="%xmm15";
+my ($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);
-	lea	-0xc8(%rsp),%rsp
-	movaps	%xmm6,0x20(%rsp)
-	movaps	%xmm7,0x30(%rsp)
-	movaps	%xmm8,0x40(%rsp)
-	movaps	%xmm9,0x50(%rsp)
-	movaps	%xmm10,0x60(%rsp)
-	movaps	%xmm11,0x70(%rsp)
-	movaps	%xmm12,0x80(%rsp)
-	movaps	%xmm13,0x90(%rsp)
-	movaps	%xmm14,0xa0(%rsp)
-	movaps	%xmm15,0xb0(%rsp)
+	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	($ivp),$ivec
-	movdqa	.Lbswap_mask(%rip),$bswap_mask
-	xor	$rounds,$rounds
-	pextrd	\$3,$ivec,$rnds_		# pull 32-bit counter
-	pinsrd	\$3,$rounds,$ivec		# wipe 32-bit counter
-
+	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
-	bswap	$rnds_
-	pxor	$iv0,$iv0			# vector of 3 32-bit counters
-	pxor	$iv1,$iv1			# vector of 3 32-bit counters
-	pinsrd	\$0,$rnds_,$iv0
-	lea	3($rnds_),$key_
-	pinsrd	\$0,$key_,$iv1
-	inc	$rnds_
-	pinsrd	\$1,$rnds_,$iv0
-	inc	$key_
-	pinsrd	\$1,$key_,$iv1
-	inc	$rnds_
-	pinsrd	\$2,$rnds_,$iv0
-	inc	$key_
-	pinsrd	\$2,$key_,$iv1
-	movdqa	$iv0,$reserved(%rsp)
-	pshufb	$bswap_mask,$iv0
-	movdqa	$iv1,`$reserved+0x10`(%rsp)
-	pshufb	$bswap_mask,$iv1
-
-	pshufd	\$`3<<6`,$iv0,$inout0		# place counter to upper dword
-	pshufd	\$`2<<6`,$iv0,$inout1
-	pshufd	\$`1<<6`,$iv0,$inout2
-	cmp	\$6,$len
+	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
-	shr	\$1,$rounds
-	mov	$key,$key_			# backup $key
-	mov	$rounds,$rnds_			# backup $rounds
+
 	sub	\$6,$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:
-	pshufd	\$`3<<6`,$iv1,$inout3
-	por	$ivec,$inout0			# merge counter-less ivec
-	 $movkey	($key_),$rndkey0
-	pshufd	\$`2<<6`,$iv1,$inout4
-	por	$ivec,$inout1
-	 $movkey	16($key_),$rndkey1
-	pshufd	\$`1<<6`,$iv1,$inout5
-	por	$ivec,$inout2
-	por	$ivec,$inout3
-	 xorps		$rndkey0,$inout0
-	por	$ivec,$inout4
-	por	$ivec,$inout5
-
-	# inline _aesni_encrypt6 and interleave last rounds
-	# with own code...
+	 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
 
-	pxor		$rndkey0,$inout1
+	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
-	lea		32($key_),$key
-	pxor		$rndkey0,$inout2
+	 mov		$ctr,%r9d
+	movdqa		0x70(%rsp),$inout7
 	aesenc		$rndkey1,$inout1
-	 movdqa		.Lincrement32(%rip),$iv1
-	pxor		$rndkey0,$inout3
+	 bswap		%r9d
+	$movkey		0x20-0x80($key),$rndkey0
 	aesenc		$rndkey1,$inout2
-	 movdqa		$reserved(%rsp),$iv0
-	pxor		$rndkey0,$inout4
+	 xor		$key0,%r9d
+	 nop
 	aesenc		$rndkey1,$inout3
-	pxor		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
-	dec		$rounds
+	 mov		%r9d,0x00+12(%rsp)
+	 lea		1($ctr),%r9
 	aesenc		$rndkey1,$inout4
 	aesenc		$rndkey1,$inout5
-	jmp		.Lctr32_enc_loop6_enter
-.align	16
-.Lctr32_enc_loop6:
+	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
-	dec		$rounds
 	aesenc		$rndkey1,$inout2
 	aesenc		$rndkey1,$inout3
 	aesenc		$rndkey1,$inout4
 	aesenc		$rndkey1,$inout5
-.Lctr32_enc_loop6_enter:
-	$movkey		16($key),$rndkey1
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	$movkey		0xb0-0x80($key),$rndkey1
+
 	aesenc		$rndkey0,$inout0
 	aesenc		$rndkey0,$inout1
-	lea		32($key),$key
 	aesenc		$rndkey0,$inout2
 	aesenc		$rndkey0,$inout3
 	aesenc		$rndkey0,$inout4
 	aesenc		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
-	jnz		.Lctr32_enc_loop6
+	aesenc		$rndkey0,$inout6
+	aesenc		$rndkey0,$inout7
+	$movkey		0xc0-0x80($key),$rndkey0
+	je		.Lctr32_enc_done
 
 	aesenc		$rndkey1,$inout0
-	 paddd		$iv1,$iv0		# increment counter vector
 	aesenc		$rndkey1,$inout1
-	 paddd		`$reserved+0x10`(%rsp),$iv1
 	aesenc		$rndkey1,$inout2
-	 movdqa		$iv0,$reserved(%rsp)	# save counter vector
 	aesenc		$rndkey1,$inout3
-	 movdqa		$iv1,`$reserved+0x10`(%rsp)
 	aesenc		$rndkey1,$inout4
-	 pshufb		$bswap_mask,$iv0	# byte swap
 	aesenc		$rndkey1,$inout5
-	 pshufb		$bswap_mask,$iv1
+	aesenc		$rndkey1,$inout6
+	aesenc		$rndkey1,$inout7
+	$movkey		0xd0-0x80($key),$rndkey1
 
-	aesenclast	$rndkey0,$inout0
-	 movups		($inp),$in0		# load input
-	aesenclast	$rndkey0,$inout1
+	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
-	aesenclast	$rndkey0,$inout2
 	 movups		0x20($inp),$in2
-	aesenclast	$rndkey0,$inout3
+	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
-	aesenclast	$rndkey0,$inout4
-	 movups		0x40($inp),$rndkey1
-	aesenclast	$rndkey0,$inout5
-	 movups		0x50($inp),$rndkey0
-	 lea	0x60($inp),$inp
-
-	xorps	$inout0,$in0			# xor
-	 pshufd	\$`3<<6`,$iv0,$inout0
-	xorps	$inout1,$in1
-	 pshufd	\$`2<<6`,$iv0,$inout1
-	movups	$in0,($out)			# store output
-	xorps	$inout2,$in2
-	 pshufd	\$`1<<6`,$iv0,$inout2
-	movups	$in1,0x10($out)
-	xorps	$inout3,$in3
-	movups	$in2,0x20($out)
-	xorps	$inout4,$rndkey1
-	movups	$in3,0x30($out)
-	xorps	$inout5,$rndkey0
-	movups	$rndkey1,0x40($out)
-	movups	$rndkey0,0x50($out)
-	lea	0x60($out),$out
-	mov	$rnds_,$rounds
-	sub	\$6,$len
-	jnc	.Lctr32_loop6
 
-	add	\$6,$len
-	jz	.Lctr32_done
-	mov	$key_,$key			# restore $key
-	lea	1($rounds,$rounds),$rounds	# restore original value
+	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
 
-.Lctr32_tail:
-	por	$ivec,$inout0
 	movups	($inp),$in0
+	xorps	$in0,$inout0
+	movups	$inout0,($out)
 	cmp	\$2,$len
-	jb	.Lctr32_one
+	jb	.Lctr32_done
 
-	por	$ivec,$inout1
 	movups	0x10($inp),$in1
-	je	.Lctr32_two
+	xorps	$in1,$inout1
+	movups	$inout1,0x10($out)
+	je	.Lctr32_done
 
-	pshufd	\$`3<<6`,$iv1,$inout3
-	por	$ivec,$inout2
 	movups	0x20($inp),$in2
-	cmp	\$4,$len
-	jb	.Lctr32_three
-
-	pshufd	\$`2<<6`,$iv1,$inout4
-	por	$ivec,$inout3
-	movups	0x30($inp),$in3
-	je	.Lctr32_four
-
-	por	$ivec,$inout4
-	xorps	$inout5,$inout5
-
-	call	_aesni_encrypt6
-
-	movups	0x40($inp),$rndkey1
-	xorps	$inout0,$in0
-	xorps	$inout1,$in1
-	movups	$in0,($out)
-	xorps	$inout2,$in2
-	movups	$in1,0x10($out)
-	xorps	$inout3,$in3
-	movups	$in2,0x20($out)
-	xorps	$inout4,$rndkey1
-	movups	$in3,0x30($out)
-	movups	$rndkey1,0x40($out)
+	xorps	$in2,$inout2
+	movups	$inout2,0x20($out)
 	jmp	.Lctr32_done
 
 .align	16
@@ -1245,64 +1566,32 @@ $code.=<<___;
 	movups	($ivp),$inout0
 	movups	($inp),$in0
 	mov	240($key),$rounds		# key->rounds
-.Lctr32_one:
 ___
 	&aesni_generate1("enc",$key,$rounds);
 $code.=<<___;
-	xorps	$inout0,$in0
-	movups	$in0,($out)
-	jmp	.Lctr32_done
-
-.align	16
-.Lctr32_two:
-	xorps	$inout2,$inout2
-	call	_aesni_encrypt3
-	xorps	$inout0,$in0
-	xorps	$inout1,$in1
-	movups	$in0,($out)
-	movups	$in1,0x10($out)
-	jmp	.Lctr32_done
-
-.align	16
-.Lctr32_three:
-	call	_aesni_encrypt3
-	xorps	$inout0,$in0
-	xorps	$inout1,$in1
-	movups	$in0,($out)
-	xorps	$inout2,$in2
-	movups	$in1,0x10($out)
-	movups	$in2,0x20($out)
+	xorps	$in0,$inout0
+	movups	$inout0,($out)
 	jmp	.Lctr32_done
 
 .align	16
-.Lctr32_four:
-	call	_aesni_encrypt4
-	xorps	$inout0,$in0
-	xorps	$inout1,$in1
-	movups	$in0,($out)
-	xorps	$inout2,$in2
-	movups	$in1,0x10($out)
-	xorps	$inout3,$in3
-	movups	$in2,0x20($out)
-	movups	$in3,0x30($out)
-
 .Lctr32_done:
 ___
 $code.=<<___ if ($win64);
-	movaps	0x20(%rsp),%xmm6
-	movaps	0x30(%rsp),%xmm7
-	movaps	0x40(%rsp),%xmm8
-	movaps	0x50(%rsp),%xmm9
-	movaps	0x60(%rsp),%xmm10
-	movaps	0x70(%rsp),%xmm11
-	movaps	0x80(%rsp),%xmm12
-	movaps	0x90(%rsp),%xmm13
-	movaps	0xa0(%rsp),%xmm14
-	movaps	0xb0(%rsp),%xmm15
-	lea	0xc8(%rsp),%rsp
-.Lctr32_ret:
+	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
 ___
 $code.=<<___;
+	lea	(%rbp),%rsp
+	pop	%rbp
+.Lctr32_epilogue:
 	ret
 .size	aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
 ___
@@ -1317,243 +1606,286 @@ ___
 my @tweak=map("%xmm$_",(10..15));
 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
 my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
-my $frame_size = 0x68 + ($win64?160:0);
+my $frame_size = 0x70 + ($win64?160:0);
 
 $code.=<<___;
 .globl	aesni_xts_encrypt
 .type	aesni_xts_encrypt,\@function,6
 .align	16
 aesni_xts_encrypt:
-	lea	-$frame_size(%rsp),%rsp
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
-	movaps	%xmm6,0x60(%rsp)
-	movaps	%xmm7,0x70(%rsp)
-	movaps	%xmm8,0x80(%rsp)
-	movaps	%xmm9,0x90(%rsp)
-	movaps	%xmm10,0xa0(%rsp)
-	movaps	%xmm11,0xb0(%rsp)
-	movaps	%xmm12,0xc0(%rsp)
-	movaps	%xmm13,0xd0(%rsp)
-	movaps	%xmm14,0xe0(%rsp)
-	movaps	%xmm15,0xf0(%rsp)
+	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:
 ___
 $code.=<<___;
-	movups	($ivp),@tweak[5]		# load clear-text tweak
+	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,$rounds,@tweak[5]);
+	&aesni_generate1("enc",$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
-	pxor	$twtmp,$twtmp
-	pcmpgtd	@tweak[5],$twtmp		# broadcast upper bits
+	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++) {
     $code.=<<___;
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
 	movdqa	@tweak[5],@tweak[$i]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	pand	$twmask,$twres			# isolate carry and residue
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
-	pxor	$twres,@tweak[5]
+	psrad	\$31,$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
 
-	shr	\$1,$rounds
-	sub	\$1,$rounds
-	mov	$rounds,$rnds_
+	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	16
+.align	32
 .Lxts_enc_grandloop:
-	pshufd	\$0x13,$twtmp,$twres
-	movdqa	@tweak[5],@tweak[4]
-	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
 	movdqu	`16*0`($inp),$inout0		# load input
-	pand	$twmask,$twres			# isolate carry and residue
+	movdqa	$rndkey0,$twmask
 	movdqu	`16*1`($inp),$inout1
-	pxor	$twres,@tweak[5]
-
+	pxor	@tweak[0],$inout0
 	movdqu	`16*2`($inp),$inout2
-	pxor	@tweak[0],$inout0		# input^=tweak
-	movdqu	`16*3`($inp),$inout3
 	pxor	@tweak[1],$inout1
-	movdqu	`16*4`($inp),$inout4
+	 aesenc		$rndkey1,$inout0
+	movdqu	`16*3`($inp),$inout3
 	pxor	@tweak[2],$inout2
-	movdqu	`16*5`($inp),$inout5
-	lea	`16*6`($inp),$inp
+	 aesenc		$rndkey1,$inout1
+	movdqu	`16*4`($inp),$inout4
 	pxor	@tweak[3],$inout3
-	$movkey		($key_),$rndkey0
+	 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],$inout4
-	pxor	@tweak[5],$inout5
+	 aesenc		$rndkey1,$inout3
+	$movkey	32($key_),$rndkey0
+	lea	`16*6`($inp),$inp
+	pxor	$twmask,$inout5
 
-	# inline _aesni_encrypt6 and interleave first and last rounds
-	# with own code...
-	$movkey		16($key_),$rndkey1
-	pxor		$rndkey0,$inout0
-	pxor		$rndkey0,$inout1
-	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks
-	aesenc		$rndkey1,$inout0
-	lea		32($key_),$key
-	pxor		$rndkey0,$inout2
-	 movdqa	@tweak[1],`16*1`(%rsp)
-	aesenc		$rndkey1,$inout1
-	pxor		$rndkey0,$inout3
-	 movdqa	@tweak[2],`16*2`(%rsp)
-	aesenc		$rndkey1,$inout2
-	pxor		$rndkey0,$inout4
-	 movdqa	@tweak[3],`16*3`(%rsp)
-	aesenc		$rndkey1,$inout3
-	pxor		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
-	dec		$rounds
-	 movdqa	@tweak[4],`16*4`(%rsp)
+	 pxor	$twres,@tweak[0]
 	aesenc		$rndkey1,$inout4
-	 movdqa	@tweak[5],`16*5`(%rsp)
+	 pxor	$twres,@tweak[1]
+	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks^last round key
 	aesenc		$rndkey1,$inout5
-	pxor	$twtmp,$twtmp
-	pcmpgtd	@tweak[5],$twtmp
-	jmp		.Lxts_enc_loop6_enter
+	$movkey		48($key_),$rndkey1
+	 pxor	$twres,@tweak[2]
 
-.align	16
+	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
-	dec		$rounds
 	aesenc		$rndkey1,$inout2
 	aesenc		$rndkey1,$inout3
 	aesenc		$rndkey1,$inout4
 	aesenc		$rndkey1,$inout5
-.Lxts_enc_loop6_enter:
-	$movkey		16($key),$rndkey1
+	$movkey		-64($key,%rax),$rndkey1
+	add		\$32,%rax
+
 	aesenc		$rndkey0,$inout0
 	aesenc		$rndkey0,$inout1
-	lea		32($key),$key
 	aesenc		$rndkey0,$inout2
 	aesenc		$rndkey0,$inout3
 	aesenc		$rndkey0,$inout4
 	aesenc		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
+	$movkey		-80($key,%rax),$rndkey0
 	jnz		.Lxts_enc_loop6
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	movdqa	(%r8),$twmask
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
 	 aesenc		$rndkey1,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	paddq	@tweak[5],@tweak[5]
+	psrad	\$31,$twtmp
 	 aesenc		$rndkey1,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcast upper bits
+	pand	$twmask,$twtmp
+	$movkey	($key_),@tweak[0]		# load round[0]
 	 aesenc		$rndkey1,$inout2
-	pxor	$twres,@tweak[5]
 	 aesenc		$rndkey1,$inout3
 	 aesenc		$rndkey1,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[0],@tweak[1]		# copy round[0]
 	 aesenc		$rndkey1,$inout5
-	 $movkey	16($key),$rndkey1
+	 $movkey	-64($key),$rndkey1
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[0]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	movdqa	$twres,$twtmp
 	 aesenc		$rndkey0,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	paddd	$twres,$twres
+	pxor	@tweak[5],@tweak[0]
 	 aesenc		$rndkey0,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	psrad	\$31,$twtmp
+	paddq	@tweak[5],@tweak[5]
 	 aesenc		$rndkey0,$inout2
-	pxor	$twres,@tweak[5]
 	 aesenc		$rndkey0,$inout3
+	pand	$twmask,$twtmp
+	movaps	@tweak[1],@tweak[2]
 	 aesenc		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movdqa	$twres,$twtmp
 	 aesenc		$rndkey0,$inout5
-	 $movkey	32($key),$rndkey0
+	 $movkey	-48($key),$rndkey0
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[1]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	paddd	$twres,$twres
 	 aesenc		$rndkey1,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	pxor	@tweak[5],@tweak[1]
+	psrad	\$31,$twtmp
 	 aesenc		$rndkey1,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
 	 aesenc		$rndkey1,$inout2
-	pxor	$twres,@tweak[5]
 	 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
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[2]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	 aesenclast	$rndkey0,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
-	 aesenclast	$rndkey0,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
-	 aesenclast	$rndkey0,$inout2
-	pxor	$twres,@tweak[5]
-	 aesenclast	$rndkey0,$inout3
-	 aesenclast	$rndkey0,$inout4
-	 aesenclast	$rndkey0,$inout5
-
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[3]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	 xorps	`16*0`(%rsp),$inout0		# output^=tweak
-	pand	$twmask,$twres			# isolate carry and residue
-	 xorps	`16*1`(%rsp),$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	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]
 
-	xorps	`16*2`(%rsp),$inout2
-	movups	$inout0,`16*0`($out)		# write output
-	xorps	`16*3`(%rsp),$inout3
-	movups	$inout1,`16*1`($out)
-	xorps	`16*4`(%rsp),$inout4
-	movups	$inout2,`16*2`($out)
-	xorps	`16*5`(%rsp),$inout5
-	movups	$inout3,`16*3`($out)
-	mov	$rnds_,$rounds			# restore $rounds
-	movups	$inout4,`16*4`($out)
-	movups	$inout5,`16*5`($out)
 	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
 
-	lea	3($rounds,$rounds),$rounds	# restore original value
+	mov	\$16+96,$rounds
+	sub	$rnds_,$rounds
 	mov	$key_,$key			# restore $key
-	mov	$rounds,$rnds_			# backup $rounds
+	shr	\$4,$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
 
-	pshufd	\$0x13,$twtmp,$twres
-	movdqa	@tweak[5],@tweak[4]
-	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
-	 movdqu	($inp),$inout0
-	pand	$twmask,$twres			# isolate carry and residue
-	 movdqu	16*1($inp),$inout1
-	pxor	$twres,@tweak[5]
-
+	movdqu	($inp),$inout0
+	movdqu	16*1($inp),$inout1
 	movdqu	16*2($inp),$inout2
 	pxor	@tweak[0],$inout0
 	movdqu	16*3($inp),$inout3
@@ -1602,7 +1934,7 @@ $code.=<<___;
 	xorps	@tweak[0],$inout0
 	xorps	@tweak[1],$inout1
 
-	call	_aesni_encrypt3
+	call	_aesni_encrypt2
 
 	xorps	@tweak[0],$inout0
 	movdqa	@tweak[2],@tweak[0]
@@ -1648,15 +1980,15 @@ $code.=<<___;
 
 	call	_aesni_encrypt4
 
-	xorps	@tweak[0],$inout0
-	movdqa	@tweak[5],@tweak[0]
-	xorps	@tweak[1],$inout1
-	xorps	@tweak[2],$inout2
-	movups	$inout0,($out)
-	xorps	@tweak[3],$inout3
-	movups	$inout1,16*1($out)
-	movups	$inout2,16*2($out)
-	movups	$inout3,16*3($out)
+	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
 
@@ -1691,19 +2023,20 @@ $code.=<<___;
 .Lxts_enc_ret:
 ___
 $code.=<<___ if ($win64);
-	movaps	0x60(%rsp),%xmm6
-	movaps	0x70(%rsp),%xmm7
-	movaps	0x80(%rsp),%xmm8
-	movaps	0x90(%rsp),%xmm9
-	movaps	0xa0(%rsp),%xmm10
-	movaps	0xb0(%rsp),%xmm11
-	movaps	0xc0(%rsp),%xmm12
-	movaps	0xd0(%rsp),%xmm13
-	movaps	0xe0(%rsp),%xmm14
-	movaps	0xf0(%rsp),%xmm15
+	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	$frame_size(%rsp),%rsp
+	lea	(%rbp),%rsp
+	pop	%rbp
 .Lxts_enc_epilogue:
 	ret
 .size	aesni_xts_encrypt,.-aesni_xts_encrypt
@@ -1714,28 +2047,32 @@ $code.=<<___;
 .type	aesni_xts_decrypt,\@function,6
 .align	16
 aesni_xts_decrypt:
-	lea	-$frame_size(%rsp),%rsp
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
-	movaps	%xmm6,0x60(%rsp)
-	movaps	%xmm7,0x70(%rsp)
-	movaps	%xmm8,0x80(%rsp)
-	movaps	%xmm9,0x90(%rsp)
-	movaps	%xmm10,0xa0(%rsp)
-	movaps	%xmm11,0xb0(%rsp)
-	movaps	%xmm12,0xc0(%rsp)
-	movaps	%xmm13,0xd0(%rsp)
-	movaps	%xmm14,0xe0(%rsp)
-	movaps	%xmm15,0xf0(%rsp)
+	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:
 ___
 $code.=<<___;
-	movups	($ivp),@tweak[5]		# load clear-text tweak
+	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,@tweak[5]);
+	&aesni_generate1("enc",$key2,$rounds,$inout0);
 $code.=<<___;
 	xor	%eax,%eax			# if ($len%16) len-=16;
 	test	\$15,$len
@@ -1743,213 +2080,249 @@ $code.=<<___;
 	shl	\$4,%rax
 	sub	%rax,$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
-	pxor	$twtmp,$twtmp
-	pcmpgtd	@tweak[5],$twtmp		# broadcast upper bits
+	movdqa	$inout0,@tweak[5]
+	pshufd	\$0x5f,$inout0,$twres
+	pxor	$rndkey0,$rndkey1
 ___
     for ($i=0;$i<4;$i++) {
     $code.=<<___;
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
 	movdqa	@tweak[5],@tweak[$i]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	pand	$twmask,$twres			# isolate carry and residue
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
-	pxor	$twres,@tweak[5]
+	psrad	\$31,$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
 
-	shr	\$1,$rounds
-	sub	\$1,$rounds
-	mov	$rounds,$rnds_
+	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	16
+.align	32
 .Lxts_dec_grandloop:
-	pshufd	\$0x13,$twtmp,$twres
-	movdqa	@tweak[5],@tweak[4]
-	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
 	movdqu	`16*0`($inp),$inout0		# load input
-	pand	$twmask,$twres			# isolate carry and residue
+	movdqa	$rndkey0,$twmask
 	movdqu	`16*1`($inp),$inout1
-	pxor	$twres,@tweak[5]
-
+	pxor	@tweak[0],$inout0
 	movdqu	`16*2`($inp),$inout2
-	pxor	@tweak[0],$inout0		# input^=tweak
-	movdqu	`16*3`($inp),$inout3
 	pxor	@tweak[1],$inout1
-	movdqu	`16*4`($inp),$inout4
+	 aesdec		$rndkey1,$inout0
+	movdqu	`16*3`($inp),$inout3
 	pxor	@tweak[2],$inout2
-	movdqu	`16*5`($inp),$inout5
-	lea	`16*6`($inp),$inp
+	 aesdec		$rndkey1,$inout1
+	movdqu	`16*4`($inp),$inout4
 	pxor	@tweak[3],$inout3
-	$movkey		($key_),$rndkey0
+	 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],$inout4
-	pxor	@tweak[5],$inout5
+	 aesdec		$rndkey1,$inout3
+	$movkey	32($key_),$rndkey0
+	lea	`16*6`($inp),$inp
+	pxor	$twmask,$inout5
 
-	# inline _aesni_decrypt6 and interleave first and last rounds
-	# with own code...
-	$movkey		16($key_),$rndkey1
-	pxor		$rndkey0,$inout0
-	pxor		$rndkey0,$inout1
-	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks
-	aesdec		$rndkey1,$inout0
-	lea		32($key_),$key
-	pxor		$rndkey0,$inout2
-	 movdqa	@tweak[1],`16*1`(%rsp)
-	aesdec		$rndkey1,$inout1
-	pxor		$rndkey0,$inout3
-	 movdqa	@tweak[2],`16*2`(%rsp)
-	aesdec		$rndkey1,$inout2
-	pxor		$rndkey0,$inout4
-	 movdqa	@tweak[3],`16*3`(%rsp)
-	aesdec		$rndkey1,$inout3
-	pxor		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
-	dec		$rounds
-	 movdqa	@tweak[4],`16*4`(%rsp)
+	 pxor	$twres,@tweak[0]
 	aesdec		$rndkey1,$inout4
-	 movdqa	@tweak[5],`16*5`(%rsp)
+	 pxor	$twres,@tweak[1]
+	 movdqa	@tweak[0],`16*0`(%rsp)		# put aside tweaks^last round key
 	aesdec		$rndkey1,$inout5
-	pxor	$twtmp,$twtmp
-	pcmpgtd	@tweak[5],$twtmp
-	jmp		.Lxts_dec_loop6_enter
+	$movkey		48($key_),$rndkey1
+	 pxor	$twres,@tweak[2]
 
-.align	16
+	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
-	dec		$rounds
 	aesdec		$rndkey1,$inout2
 	aesdec		$rndkey1,$inout3
 	aesdec		$rndkey1,$inout4
 	aesdec		$rndkey1,$inout5
-.Lxts_dec_loop6_enter:
-	$movkey		16($key),$rndkey1
+	$movkey		-64($key,%rax),$rndkey1
+	add		\$32,%rax
+
 	aesdec		$rndkey0,$inout0
 	aesdec		$rndkey0,$inout1
-	lea		32($key),$key
 	aesdec		$rndkey0,$inout2
 	aesdec		$rndkey0,$inout3
 	aesdec		$rndkey0,$inout4
 	aesdec		$rndkey0,$inout5
-	$movkey		($key),$rndkey0
+	$movkey		-80($key,%rax),$rndkey0
 	jnz		.Lxts_dec_loop6
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	movdqa	(%r8),$twmask
+	movdqa	$twres,$twtmp
+	paddd	$twres,$twres
 	 aesdec		$rndkey1,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	paddq	@tweak[5],@tweak[5]
+	psrad	\$31,$twtmp
 	 aesdec		$rndkey1,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcast upper bits
+	pand	$twmask,$twtmp
+	$movkey	($key_),@tweak[0]		# load round[0]
 	 aesdec		$rndkey1,$inout2
-	pxor	$twres,@tweak[5]
 	 aesdec		$rndkey1,$inout3
 	 aesdec		$rndkey1,$inout4
+	pxor	$twtmp,@tweak[5]
+	movaps	@tweak[0],@tweak[1]		# copy round[0]
 	 aesdec		$rndkey1,$inout5
-	 $movkey	16($key),$rndkey1
+	 $movkey	-64($key),$rndkey1
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[0]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	movdqa	$twres,$twtmp
 	 aesdec		$rndkey0,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	paddd	$twres,$twres
+	pxor	@tweak[5],@tweak[0]
 	 aesdec		$rndkey0,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	psrad	\$31,$twtmp
+	paddq	@tweak[5],@tweak[5]
 	 aesdec		$rndkey0,$inout2
-	pxor	$twres,@tweak[5]
 	 aesdec		$rndkey0,$inout3
+	pand	$twmask,$twtmp
+	movaps	@tweak[1],@tweak[2]
 	 aesdec		$rndkey0,$inout4
+	pxor	$twtmp,@tweak[5]
+	movdqa	$twres,$twtmp
 	 aesdec		$rndkey0,$inout5
-	 $movkey	32($key),$rndkey0
+	 $movkey	-48($key),$rndkey0
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[1]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
+	paddd	$twres,$twres
 	 aesdec		$rndkey1,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
+	pxor	@tweak[5],@tweak[1]
+	psrad	\$31,$twtmp
 	 aesdec		$rndkey1,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	paddq	@tweak[5],@tweak[5]
+	pand	$twmask,$twtmp
 	 aesdec		$rndkey1,$inout2
-	pxor	$twres,@tweak[5]
 	 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
 
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[2]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	 aesdeclast	$rndkey0,$inout0
-	pand	$twmask,$twres			# isolate carry and residue
-	 aesdeclast	$rndkey0,$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
-	 aesdeclast	$rndkey0,$inout2
-	pxor	$twres,@tweak[5]
-	 aesdeclast	$rndkey0,$inout3
-	 aesdeclast	$rndkey0,$inout4
-	 aesdeclast	$rndkey0,$inout5
-
-	pshufd	\$0x13,$twtmp,$twres
-	pxor	$twtmp,$twtmp
-	movdqa	@tweak[5],@tweak[3]
-	paddq	@tweak[5],@tweak[5]		# psllq	1,$tweak
-	 xorps	`16*0`(%rsp),$inout0		# output^=tweak
-	pand	$twmask,$twres			# isolate carry and residue
-	 xorps	`16*1`(%rsp),$inout1
-	pcmpgtd	@tweak[5],$twtmp		# broadcat upper bits
+	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]
 
-	xorps	`16*2`(%rsp),$inout2
-	movups	$inout0,`16*0`($out)		# write output
-	xorps	`16*3`(%rsp),$inout3
-	movups	$inout1,`16*1`($out)
-	xorps	`16*4`(%rsp),$inout4
-	movups	$inout2,`16*2`($out)
-	xorps	`16*5`(%rsp),$inout5
-	movups	$inout3,`16*3`($out)
-	mov	$rnds_,$rounds			# restore $rounds
-	movups	$inout4,`16*4`($out)
-	movups	$inout5,`16*5`($out)
 	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
 
-	lea	3($rounds,$rounds),$rounds	# restore original value
+	mov	\$16+96,$rounds
+	sub	$rnds_,$rounds
 	mov	$key_,$key			# restore $key
-	mov	$rounds,$rnds_			# backup $rounds
+	shr	\$4,$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
 
-	pshufd	\$0x13,$twtmp,$twres
-	movdqa	@tweak[5],@tweak[4]
-	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
-	 movdqu	($inp),$inout0
-	pand	$twmask,$twres			# isolate carry and residue
-	 movdqu	16*1($inp),$inout1
-	pxor	$twres,@tweak[5]
-
+	movdqu	($inp),$inout0
+	movdqu	16*1($inp),$inout1
 	movdqu	16*2($inp),$inout2
 	pxor	@tweak[0],$inout0
 	movdqu	16*3($inp),$inout3
@@ -2008,7 +2381,7 @@ $code.=<<___;
 	xorps	@tweak[0],$inout0
 	xorps	@tweak[1],$inout1
 
-	call	_aesni_decrypt3
+	call	_aesni_decrypt2
 
 	xorps	@tweak[0],$inout0
 	movdqa	@tweak[2],@tweak[0]
@@ -2034,7 +2407,7 @@ $code.=<<___;
 	xorps	@tweak[0],$inout0
 	movdqa	@tweak[3],@tweak[0]
 	xorps	@tweak[1],$inout1
-	movdqa	@tweak[5],@tweak[1]
+	movdqa	@tweak[4],@tweak[1]
 	xorps	@tweak[2],$inout2
 	movups	$inout0,($out)
 	movups	$inout1,16*1($out)
@@ -2044,14 +2417,8 @@ $code.=<<___;
 
 .align	16
 .Lxts_dec_four:
-	pshufd	\$0x13,$twtmp,$twres
-	movdqa	@tweak[5],@tweak[4]
-	paddq	@tweak[5],@tweak[5]		# psllq 1,$tweak
-	 movups	($inp),$inout0
-	pand	$twmask,$twres			# isolate carry and residue
-	 movups	16*1($inp),$inout1
-	pxor	$twres,@tweak[5]
-
+	movups	($inp),$inout0
+	movups	16*1($inp),$inout1
 	movups	16*2($inp),$inout2
 	xorps	@tweak[0],$inout0
 	movups	16*3($inp),$inout3
@@ -2062,16 +2429,16 @@ $code.=<<___;
 
 	call	_aesni_decrypt4
 
-	xorps	@tweak[0],$inout0
+	pxor	@tweak[0],$inout0
 	movdqa	@tweak[4],@tweak[0]
-	xorps	@tweak[1],$inout1
+	pxor	@tweak[1],$inout1
 	movdqa	@tweak[5],@tweak[1]
-	xorps	@tweak[2],$inout2
-	movups	$inout0,($out)
-	xorps	@tweak[3],$inout3
-	movups	$inout1,16*1($out)
-	movups	$inout2,16*2($out)
-	movups	$inout3,16*3($out)
+	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
 
@@ -2117,19 +2484,20 @@ $code.=<<___;
 .Lxts_dec_ret:
 ___
 $code.=<<___ if ($win64);
-	movaps	0x60(%rsp),%xmm6
-	movaps	0x70(%rsp),%xmm7
-	movaps	0x80(%rsp),%xmm8
-	movaps	0x90(%rsp),%xmm9
-	movaps	0xa0(%rsp),%xmm10
-	movaps	0xb0(%rsp),%xmm11
-	movaps	0xc0(%rsp),%xmm12
-	movaps	0xd0(%rsp),%xmm13
-	movaps	0xe0(%rsp),%xmm14
-	movaps	0xf0(%rsp),%xmm15
+	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	$frame_size(%rsp),%rsp
+	lea	(%rbp),%rsp
+	pop	%rbp
 .Lxts_dec_epilogue:
 	ret
 .size	aesni_xts_decrypt,.-aesni_xts_decrypt
@@ -2141,7 +2509,10 @@ ___
 #			    size_t length, const AES_KEY *key,
 #			    unsigned char *ivp,const int enc);
 {
-my $reserved = $win64?0x40:-0x18;	# used in decrypt
+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
@@ -2197,276 +2568,398 @@ $code.=<<___;
 #--------------------------- CBC DECRYPT ------------------------------#
 .align	16
 .Lcbc_decrypt:
+	lea	(%rsp),%rax
+	push	%rbp
+	sub	\$$frame_size,%rsp
+	and	\$-16,%rsp	# Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
-	lea	-0x58(%rsp),%rsp
-	movaps	%xmm6,(%rsp)
-	movaps	%xmm7,0x10(%rsp)
-	movaps	%xmm8,0x20(%rsp)
-	movaps	%xmm9,0x30(%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_decrypt_body:
 ___
 $code.=<<___;
+	lea	-8(%rax),%rbp
 	movups	($ivp),$iv
 	mov	$rnds_,$rounds
-	cmp	\$0x70,$len
+	cmp	\$0x50,$len
 	jbe	.Lcbc_dec_tail
-	shr	\$1,$rnds_
-	sub	\$0x70,$len
-	mov	$rnds_,$rounds
-	movaps	$iv,$reserved(%rsp)
+
+	$movkey	($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:
-	movaps	$rndkey0,$reserved(%rsp)	# save IV
 	movups	$inout7,($out)
 	lea	0x10($out),$out
 .Lcbc_dec_loop8_enter:
-	$movkey		($key),$rndkey0
-	movups	($inp),$inout0			# load input
-	movups	0x10($inp),$inout1
-	$movkey		16($key),$rndkey1
+	movdqu		0x60($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
 
-	lea		32($key),$key
-	movdqu	0x20($inp),$inout2
-	xorps		$rndkey0,$inout0
-	movdqu	0x30($inp),$inout3
-	xorps		$rndkey0,$inout1
-	movdqu	0x40($inp),$inout4
 	aesdec		$rndkey1,$inout0
-	pxor		$rndkey0,$inout2
-	movdqu	0x50($inp),$inout5
+	pxor		$rndkey0,$inout7
+	$movkey		0x20-0x70($key),$rndkey0
 	aesdec		$rndkey1,$inout1
-	pxor		$rndkey0,$inout3
-	movdqu	0x60($inp),$inout6
 	aesdec		$rndkey1,$inout2
-	pxor		$rndkey0,$inout4
-	movdqu	0x70($inp),$inout7
 	aesdec		$rndkey1,$inout3
-	pxor		$rndkey0,$inout5
-	dec		$rounds
 	aesdec		$rndkey1,$inout4
-	pxor		$rndkey0,$inout6
 	aesdec		$rndkey1,$inout5
-	pxor		$rndkey0,$inout7
-	$movkey		($key),$rndkey0
 	aesdec		$rndkey1,$inout6
+	setnc		${inp_}b
+	shl		\$7,$inp_
 	aesdec		$rndkey1,$inout7
-	$movkey		16($key),$rndkey1
-
-	call		.Ldec_loop8_enter
+	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
 
-	movups	($inp),$rndkey1		# re-load input
-	movups	0x10($inp),$rndkey0
-	xorps	$reserved(%rsp),$inout0	# ^= IV
-	xorps	$rndkey1,$inout1
-	movups	0x20($inp),$rndkey1
-	xorps	$rndkey0,$inout2
-	movups	0x30($inp),$rndkey0
-	xorps	$rndkey1,$inout3
-	movups	0x40($inp),$rndkey1
-	xorps	$rndkey0,$inout4
-	movups	0x50($inp),$rndkey0
-	xorps	$rndkey1,$inout5
-	movups	0x60($inp),$rndkey1
-	xorps	$rndkey0,$inout6
-	movups	0x70($inp),$rndkey0	# IV
-	xorps	$rndkey1,$inout7
-	movups	$inout0,($out)
-	movups	$inout1,0x10($out)
-	movups	$inout2,0x20($out)
-	movups	$inout3,0x30($out)
-	mov	$rnds_,$rounds		# restore $rounds
-	movups	$inout4,0x40($out)
-	mov	$key_,$key		# restore $key
-	movups	$inout5,0x50($out)
-	lea	0x80($inp),$inp
-	movups	$inout6,0x60($out)
-	lea	0x70($out),$out
 	sub	\$0x80,$len
 	ja	.Lcbc_dec_loop8
 
 	movaps	$inout7,$inout0
-	movaps	$rndkey0,$iv
+	lea	-0x70($key),$key
 	add	\$0x70,$len
 	jle	.Lcbc_dec_tail_collected
-	movups	$inout0,($out)
-	lea	1($rnds_,$rnds_),$rounds
+	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
-	movaps	$inout0,$in0
-	cmp	\$0x10,$len
+	sub	\$0x10,$len
 	jbe	.Lcbc_dec_one
 
 	movups	0x10($inp),$inout1
-	movaps	$inout1,$in1
-	cmp	\$0x20,$len
+	movaps	$inout0,$in0
+	sub	\$0x10,$len
 	jbe	.Lcbc_dec_two
 
 	movups	0x20($inp),$inout2
-	movaps	$inout2,$in2
-	cmp	\$0x30,$len
+	movaps	$inout1,$in1
+	sub	\$0x10,$len
 	jbe	.Lcbc_dec_three
 
 	movups	0x30($inp),$inout3
-	cmp	\$0x40,$len
+	movaps	$inout2,$in2
+	sub	\$0x10,$len
 	jbe	.Lcbc_dec_four
 
 	movups	0x40($inp),$inout4
-	cmp	\$0x50,$len
-	jbe	.Lcbc_dec_five
-
-	movups	0x50($inp),$inout5
-	cmp	\$0x60,$len
-	jbe	.Lcbc_dec_six
-
-	movups	0x60($inp),$inout6
-	movaps	$iv,$reserved(%rsp)	# save IV
-	call	_aesni_decrypt8
-	movups	($inp),$rndkey1
-	movups	0x10($inp),$rndkey0
-	xorps	$reserved(%rsp),$inout0	# ^= IV
-	xorps	$rndkey1,$inout1
-	movups	0x20($inp),$rndkey1
-	xorps	$rndkey0,$inout2
-	movups	0x30($inp),$rndkey0
-	xorps	$rndkey1,$inout3
-	movups	0x40($inp),$rndkey1
-	xorps	$rndkey0,$inout4
-	movups	0x50($inp),$rndkey0
-	xorps	$rndkey1,$inout5
-	movups	0x60($inp),$iv		# IV
-	xorps	$rndkey0,$inout6
-	movups	$inout0,($out)
-	movups	$inout1,0x10($out)
-	movups	$inout2,0x20($out)
-	movups	$inout3,0x30($out)
-	movups	$inout4,0x40($out)
-	movups	$inout5,0x50($out)
-	lea	0x60($out),$out
-	movaps	$inout6,$inout0
-	sub	\$0x70,$len
+	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
-	sub	\$0x10,$len
 	jmp	.Lcbc_dec_tail_collected
 .align	16
 .Lcbc_dec_two:
-	xorps	$inout2,$inout2
-	call	_aesni_decrypt3
-	xorps	$iv,$inout0
-	xorps	$in0,$inout1
-	movups	$inout0,($out)
+	movaps	$inout1,$in1
+	call	_aesni_decrypt2
+	pxor	$iv,$inout0
 	movaps	$in1,$iv
-	movaps	$inout1,$inout0
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	movdqa	$inout1,$inout0
 	lea	0x10($out),$out
-	sub	\$0x20,$len
 	jmp	.Lcbc_dec_tail_collected
 .align	16
 .Lcbc_dec_three:
+	movaps	$inout2,$in2
 	call	_aesni_decrypt3
-	xorps	$iv,$inout0
-	xorps	$in0,$inout1
-	movups	$inout0,($out)
-	xorps	$in1,$inout2
-	movups	$inout1,0x10($out)
+	pxor	$iv,$inout0
 	movaps	$in2,$iv
-	movaps	$inout2,$inout0
+	pxor	$in0,$inout1
+	movdqu	$inout0,($out)
+	pxor	$in1,$inout2
+	movdqu	$inout1,0x10($out)
+	movdqa	$inout2,$inout0
 	lea	0x20($out),$out
-	sub	\$0x30,$len
 	jmp	.Lcbc_dec_tail_collected
 .align	16
 .Lcbc_dec_four:
+	movaps	$inout3,$in3
 	call	_aesni_decrypt4
-	xorps	$iv,$inout0
-	movups	0x30($inp),$iv
-	xorps	$in0,$inout1
-	movups	$inout0,($out)
-	xorps	$in1,$inout2
-	movups	$inout1,0x10($out)
-	xorps	$in2,$inout3
-	movups	$inout2,0x20($out)
-	movaps	$inout3,$inout0
+	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
-	sub	\$0x40,$len
-	jmp	.Lcbc_dec_tail_collected
-.align	16
-.Lcbc_dec_five:
-	xorps	$inout5,$inout5
-	call	_aesni_decrypt6
-	movups	0x10($inp),$rndkey1
-	movups	0x20($inp),$rndkey0
-	xorps	$iv,$inout0
-	xorps	$in0,$inout1
-	xorps	$rndkey1,$inout2
-	movups	0x30($inp),$rndkey1
-	xorps	$rndkey0,$inout3
-	movups	0x40($inp),$iv
-	xorps	$rndkey1,$inout4
-	movups	$inout0,($out)
-	movups	$inout1,0x10($out)
-	movups	$inout2,0x20($out)
-	movups	$inout3,0x30($out)
-	lea	0x40($out),$out
-	movaps	$inout4,$inout0
-	sub	\$0x50,$len
-	jmp	.Lcbc_dec_tail_collected
-.align	16
-.Lcbc_dec_six:
-	call	_aesni_decrypt6
-	movups	0x10($inp),$rndkey1
-	movups	0x20($inp),$rndkey0
-	xorps	$iv,$inout0
-	xorps	$in0,$inout1
-	xorps	$rndkey1,$inout2
-	movups	0x30($inp),$rndkey1
-	xorps	$rndkey0,$inout3
-	movups	0x40($inp),$rndkey0
-	xorps	$rndkey1,$inout4
-	movups	0x50($inp),$iv
-	xorps	$rndkey0,$inout5
-	movups	$inout0,($out)
-	movups	$inout1,0x10($out)
-	movups	$inout2,0x20($out)
-	movups	$inout3,0x30($out)
-	movups	$inout4,0x40($out)
-	lea	0x50($out),$out
-	movaps	$inout5,$inout0
-	sub	\$0x60,$len
 	jmp	.Lcbc_dec_tail_collected
+
 .align	16
 .Lcbc_dec_tail_collected:
-	and	\$15,$len
 	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	$inout0,$reserved(%rsp)
+	movaps	$inout0,(%rsp)
 	mov	\$16,%rcx
 	mov	$out,%rdi
 	sub	$len,%rcx
-	lea	$reserved(%rsp),%rsi
+	lea	(%rsp),%rsi
 	.long	0x9066A4F3	# rep movsb
 
 .Lcbc_dec_ret:
 ___
 $code.=<<___ if ($win64);
-	movaps	(%rsp),%xmm6
-	movaps	0x10(%rsp),%xmm7
-	movaps	0x20(%rsp),%xmm8
-	movaps	0x30(%rsp),%xmm9
-	lea	0x58(%rsp),%rsp
+	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
@@ -2733,6 +3226,8 @@ $code.=<<___;
 	.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
@@ -2791,45 +3286,9 @@ ecb_ccm64_se_handler:
 	jmp	.Lcommon_seh_tail
 .size	ecb_ccm64_se_handler,.-ecb_ccm64_se_handler
 
-.type	ctr32_se_handler,\@abi-omnipotent
+.type	ctr_xts_se_handler,\@abi-omnipotent
 .align	16
-ctr32_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	.Lctr32_body(%rip),%r10
-	cmp	%r10,%rbx		# context->Rip<"prologue" label
-	jb	.Lcommon_seh_tail
-
-	mov	152($context),%rax	# pull context->Rsp
-
-	lea	.Lctr32_ret(%rip),%r10
-	cmp	%r10,%rbx
-	jae	.Lcommon_seh_tail
-
-	lea	0x20(%rax),%rsi		# %xmm save area
-	lea	512($context),%rdi	# &context.Xmm6
-	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
-	.long	0xa548f3fc		# cld; rep movsq
-	lea	0xc8(%rax),%rax		# adjust stack pointer
-
-	jmp	.Lcommon_seh_tail
-.size	ctr32_se_handler,.-ctr32_se_handler
-
-.type	xts_se_handler,\@abi-omnipotent
-.align	16
-xts_se_handler:
+ctr_xts_se_handler:
 	push	%rsi
 	push	%rdi
 	push	%rbx
@@ -2859,14 +3318,14 @@ xts_se_handler:
 	cmp	%r10,%rbx		# context->Rip>=epilogue label
 	jae	.Lcommon_seh_tail
 
-	lea	0x60(%rax),%rsi		# %xmm save area
+	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
-	lea	0x68+160(%rax),%rax	# adjust stack pointer
 
-	jmp	.Lcommon_seh_tail
-.size	xts_se_handler,.-xts_se_handler
+	jmp	.Lcommon_rbp_tail
+.size	ctr_xts_se_handler,.-ctr_xts_se_handler
 ___
 $code.=<<___;
 .type	cbc_se_handler,\@abi-omnipotent
@@ -2898,11 +3357,16 @@ cbc_se_handler:
 	cmp	%r10,%rbx		# context->Rip>="epilogue" label
 	jae	.Lcommon_seh_tail
 
-	lea	0(%rax),%rsi		# top of stack
+	lea	16(%rax),%rsi		# %xmm save area
 	lea	512($context),%rdi	# &context.Xmm6
-	mov	\$8,%ecx		# 4*sizeof(%xmm0)/sizeof(%rax)
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
 	.long	0xa548f3fc		# cld; rep movsq
-	lea	0x58(%rax),%rax		# adjust stack pointer
+
+.Lcommon_rbp_tail:
+	mov	160($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:
@@ -3006,14 +3470,15 @@ $code.=<<___ if ($PREFIX eq "aesni");
 	.rva	.Lccm64_dec_body,.Lccm64_dec_ret	# HandlerData[]
 .LSEH_info_ctr32:
 	.byte	9,0,0,0
-	.rva	ctr32_se_handler
+	.rva	ctr_xts_se_handler
+	.rva	.Lctr32_body,.Lctr32_epilogue		# HandlerData[]
 .LSEH_info_xts_enc:
 	.byte	9,0,0,0
-	.rva	xts_se_handler
+	.rva	ctr_xts_se_handler
 	.rva	.Lxts_enc_body,.Lxts_enc_epilogue	# HandlerData[]
 .LSEH_info_xts_dec:
 	.byte	9,0,0,0
-	.rva	xts_se_handler
+	.rva	ctr_xts_se_handler
 	.rva	.Lxts_dec_body,.Lxts_dec_epilogue	# HandlerData[]
 ___
 $code.=<<___;
@@ -3060,11 +3525,30 @@ sub aesni {
 	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;
 
diff --git a/openssl/crypto/aes/asm/aesp8-ppc.pl b/openssl/crypto/aes/asm/aesp8-ppc.pl
new file mode 100755
index 000000000..a1891cc03
--- /dev/null
+++ b/openssl/crypto/aes/asm/aesp8-ppc.pl
@@ -0,0 +1,1942 @@
+#!/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 AES instructions as per PowerISA
+# specification version 2.07, first implemented by POWER8 processor.
+# The module is endian-agnostic in sense that it supports both big-
+# and little-endian cases. Data alignment in parallelizable modes is
+# handled with VSX loads and stores, which implies MSR.VSX flag being
+# set. It should also be noted that ISA specification doesn't prohibit
+# alignment exceptions for these instructions on page boundaries.
+# Initially alignment was handled in pure AltiVec/VMX way [when data
+# is aligned programmatically, which in turn guarantees exception-
+# free execution], but it turned to hamper performance when vcipher
+# instructions are interleaved. It's reckoned that eventual
+# misalignment penalties at page boundaries are in average lower
+# than additional overhead in pure AltiVec approach.
+
+$flavour = shift;
+
+if ($flavour =~ /64/) {
+	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
+	$STU	="stdu";
+	$POP	="ld";
+	$PUSH	="std";
+	$UCMP	="cmpld";
+	$SHL	="sldi";
+} elsif ($flavour =~ /32/) {
+	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
+	$STU	="stwu";
+	$POP	="lwz";
+	$PUSH	="stw";
+	$UCMP	="cmplw";
+	$SHL	="slwi";
+} else { die "nonsense $flavour"; }
+
+$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
+
+$FRAME=8*$SIZE_T;
+$prefix="aes_p8";
+
+$sp="r1";
+$vrsave="r12";
+
+#########################################################################
+{{{	# Key setup procedures						#
+my ($inp,$bits,$out,$ptr,$cnt,$rounds)=map("r$_",(3..8));
+my ($zero,$in0,$in1,$key,$rcon,$mask,$tmp)=map("v$_",(0..6));
+my ($stage,$outperm,$outmask,$outhead,$outtail)=map("v$_",(7..11));
+
+$code.=<<___;
+.machine	"any"
+
+.text
+
+.align	7
+rcon:
+.long	0x01000000, 0x01000000, 0x01000000, 0x01000000	?rev
+.long	0x1b000000, 0x1b000000, 0x1b000000, 0x1b000000	?rev
+.long	0x0d0e0f0c, 0x0d0e0f0c, 0x0d0e0f0c, 0x0d0e0f0c	?rev
+.long	0,0,0,0						?asis
+Lconsts:
+	mflr	r0
+	bcl	20,31,\$+4
+	mflr	$ptr	 #vvvvv "distance between . and rcon
+	addi	$ptr,$ptr,-0x48
+	mtlr	r0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+.asciz	"AES for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>"
+
+.globl	.${prefix}_set_encrypt_key
+.align	5
+.${prefix}_set_encrypt_key:
+Lset_encrypt_key:
+	mflr		r11
+	$PUSH		r11,$LRSAVE($sp)
+
+	li		$ptr,-1
+	${UCMP}i	$inp,0
+	beq-		Lenc_key_abort		# if ($inp==0) return -1;
+	${UCMP}i	$out,0
+	beq-		Lenc_key_abort		# if ($out==0) return -1;
+	li		$ptr,-2
+	cmpwi		$bits,128
+	blt-		Lenc_key_abort
+	cmpwi		$bits,256
+	bgt-		Lenc_key_abort
+	andi.		r0,$bits,0x3f
+	bne-		Lenc_key_abort
+
+	lis		r0,0xfff0
+	mfspr		$vrsave,256
+	mtspr		256,r0
+
+	bl		Lconsts
+	mtlr		r11
+
+	neg		r9,$inp
+	lvx		$in0,0,$inp
+	addi		$inp,$inp,15		# 15 is not typo
+	lvsr		$key,0,r9		# borrow $key
+	li		r8,0x20
+	cmpwi		$bits,192
+	lvx		$in1,0,$inp
+	le?vspltisb	$mask,0x0f		# borrow $mask
+	lvx		$rcon,0,$ptr
+	le?vxor		$key,$key,$mask		# adjust for byte swap
+	lvx		$mask,r8,$ptr
+	addi		$ptr,$ptr,0x10
+	vperm		$in0,$in0,$in1,$key	# align [and byte swap in LE]
+	li		$cnt,8
+	vxor		$zero,$zero,$zero
+	mtctr		$cnt
+
+	?lvsr		$outperm,0,$out
+	vspltisb	$outmask,-1
+	lvx		$outhead,0,$out
+	?vperm		$outmask,$zero,$outmask,$outperm
+
+	blt		Loop128
+	addi		$inp,$inp,8
+	beq		L192
+	addi		$inp,$inp,8
+	b		L256
+
+.align	4
+Loop128:
+	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vcipherlast	$key,$key,$rcon
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	 vadduwm	$rcon,$rcon,$rcon
+	vxor		$in0,$in0,$key
+	bdnz		Loop128
+
+	lvx		$rcon,0,$ptr		# last two round keys
+
+	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vcipherlast	$key,$key,$rcon
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	 vadduwm	$rcon,$rcon,$rcon
+	vxor		$in0,$in0,$key
+
+	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vcipherlast	$key,$key,$rcon
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vxor		$in0,$in0,$key
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	 stvx		$stage,0,$out
+
+	addi		$inp,$out,15		# 15 is not typo
+	addi		$out,$out,0x50
+
+	li		$rounds,10
+	b		Ldone
+
+.align	4
+L192:
+	lvx		$tmp,0,$inp
+	li		$cnt,4
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+	vperm		$in1,$in1,$tmp,$key	# align [and byte swap in LE]
+	vspltisb	$key,8			# borrow $key
+	mtctr		$cnt
+	vsububm		$mask,$mask,$key	# adjust the mask
+
+Loop192:
+	vperm		$key,$in1,$in1,$mask	# roate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	vcipherlast	$key,$key,$rcon
+
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+
+	 vsldoi		$stage,$zero,$in1,8
+	vspltw		$tmp,$in0,3
+	vxor		$tmp,$tmp,$in1
+	vsldoi		$in1,$zero,$in1,12	# >>32
+	 vadduwm	$rcon,$rcon,$rcon
+	vxor		$in1,$in1,$tmp
+	vxor		$in0,$in0,$key
+	vxor		$in1,$in1,$key
+	 vsldoi		$stage,$stage,$in0,8
+
+	vperm		$key,$in1,$in1,$mask	# rotate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	 vperm		$outtail,$stage,$stage,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vcipherlast	$key,$key,$rcon
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	 vsldoi		$stage,$in0,$in1,8
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	 vperm		$outtail,$stage,$stage,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	vspltw		$tmp,$in0,3
+	vxor		$tmp,$tmp,$in1
+	vsldoi		$in1,$zero,$in1,12	# >>32
+	 vadduwm	$rcon,$rcon,$rcon
+	vxor		$in1,$in1,$tmp
+	vxor		$in0,$in0,$key
+	vxor		$in1,$in1,$key
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	 stvx		$stage,0,$out
+	 addi		$inp,$out,15		# 15 is not typo
+	 addi		$out,$out,16
+	bdnz		Loop192
+
+	li		$rounds,12
+	addi		$out,$out,0x20
+	b		Ldone
+
+.align	4
+L256:
+	lvx		$tmp,0,$inp
+	li		$cnt,7
+	li		$rounds,14
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+	vperm		$in1,$in1,$tmp,$key	# align [and byte swap in LE]
+	mtctr		$cnt
+
+Loop256:
+	vperm		$key,$in1,$in1,$mask	# rotate-n-splat
+	vsldoi		$tmp,$zero,$in0,12	# >>32
+	 vperm		$outtail,$in1,$in1,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	vcipherlast	$key,$key,$rcon
+	 stvx		$stage,0,$out
+	 addi		$out,$out,16
+
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in0,$in0,$tmp
+	 vadduwm	$rcon,$rcon,$rcon
+	vxor		$in0,$in0,$key
+	 vperm		$outtail,$in0,$in0,$outperm	# rotate
+	 vsel		$stage,$outhead,$outtail,$outmask
+	 vmr		$outhead,$outtail
+	 stvx		$stage,0,$out
+	 addi		$inp,$out,15		# 15 is not typo
+	 addi		$out,$out,16
+	bdz		Ldone
+
+	vspltw		$key,$in0,3		# just splat
+	vsldoi		$tmp,$zero,$in1,12	# >>32
+	vsbox		$key,$key
+
+	vxor		$in1,$in1,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in1,$in1,$tmp
+	vsldoi		$tmp,$zero,$tmp,12	# >>32
+	vxor		$in1,$in1,$tmp
+
+	vxor		$in1,$in1,$key
+	b		Loop256
+
+.align	4
+Ldone:
+	lvx		$in1,0,$inp		# redundant in aligned case
+	vsel		$in1,$outhead,$in1,$outmask
+	stvx		$in1,0,$inp
+	li		$ptr,0
+	mtspr		256,$vrsave
+	stw		$rounds,0($out)
+
+Lenc_key_abort:
+	mr		r3,$ptr
+	blr
+	.long		0
+	.byte		0,12,0x14,1,0,0,3,0
+	.long		0
+.size	.${prefix}_set_encrypt_key,.-.${prefix}_set_encrypt_key
+
+.globl	.${prefix}_set_decrypt_key
+.align	5
+.${prefix}_set_decrypt_key:
+	$STU		$sp,-$FRAME($sp)
+	mflr		r10
+	$PUSH		r10,$FRAME+$LRSAVE($sp)
+	bl		Lset_encrypt_key
+	mtlr		r10
+
+	cmpwi		r3,0
+	bne-		Ldec_key_abort
+
+	slwi		$cnt,$rounds,4
+	subi		$inp,$out,240		# first round key
+	srwi		$rounds,$rounds,1
+	add		$out,$inp,$cnt		# last round key
+	mtctr		$rounds
+
+Ldeckey:
+	lwz		r0, 0($inp)
+	lwz		r6, 4($inp)
+	lwz		r7, 8($inp)
+	lwz		r8, 12($inp)
+	addi		$inp,$inp,16
+	lwz		r9, 0($out)
+	lwz		r10,4($out)
+	lwz		r11,8($out)
+	lwz		r12,12($out)
+	stw		r0, 0($out)
+	stw		r6, 4($out)
+	stw		r7, 8($out)
+	stw		r8, 12($out)
+	subi		$out,$out,16
+	stw		r9, -16($inp)
+	stw		r10,-12($inp)
+	stw		r11,-8($inp)
+	stw		r12,-4($inp)
+	bdnz		Ldeckey
+
+	xor		r3,r3,r3		# return value
+Ldec_key_abort:
+	addi		$sp,$sp,$FRAME
+	blr
+	.long		0
+	.byte		0,12,4,1,0x80,0,3,0
+	.long		0
+.size	.${prefix}_set_decrypt_key,.-.${prefix}_set_decrypt_key
+___
+}}}
+#########################################################################
+{{{	# Single block en- and decrypt procedures			#
+sub gen_block () {
+my $dir = shift;
+my $n   = $dir eq "de" ? "n" : "";
+my ($inp,$out,$key,$rounds,$idx)=map("r$_",(3..7));
+
+$code.=<<___;
+.globl	.${prefix}_${dir}crypt
+.align	5
+.${prefix}_${dir}crypt:
+	lwz		$rounds,240($key)
+	lis		r0,0xfc00
+	mfspr		$vrsave,256
+	li		$idx,15			# 15 is not typo
+	mtspr		256,r0
+
+	lvx		v0,0,$inp
+	neg		r11,$out
+	lvx		v1,$idx,$inp
+	lvsl		v2,0,$inp		# inpperm
+	le?vspltisb	v4,0x0f
+	?lvsl		v3,0,r11		# outperm
+	le?vxor		v2,v2,v4
+	li		$idx,16
+	vperm		v0,v0,v1,v2		# align [and byte swap in LE]
+	lvx		v1,0,$key
+	?lvsl		v5,0,$key		# keyperm
+	srwi		$rounds,$rounds,1
+	lvx		v2,$idx,$key
+	addi		$idx,$idx,16
+	subi		$rounds,$rounds,1
+	?vperm		v1,v1,v2,v5		# align round key
+
+	vxor		v0,v0,v1
+	lvx		v1,$idx,$key
+	addi		$idx,$idx,16
+	mtctr		$rounds
+
+Loop_${dir}c:
+	?vperm		v2,v2,v1,v5
+	v${n}cipher	v0,v0,v2
+	lvx		v2,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		v1,v1,v2,v5
+	v${n}cipher	v0,v0,v1
+	lvx		v1,$idx,$key
+	addi		$idx,$idx,16
+	bdnz		Loop_${dir}c
+
+	?vperm		v2,v2,v1,v5
+	v${n}cipher	v0,v0,v2
+	lvx		v2,$idx,$key
+	?vperm		v1,v1,v2,v5
+	v${n}cipherlast	v0,v0,v1
+
+	vspltisb	v2,-1
+	vxor		v1,v1,v1
+	li		$idx,15			# 15 is not typo
+	?vperm		v2,v1,v2,v3		# outmask
+	le?vxor		v3,v3,v4
+	lvx		v1,0,$out		# outhead
+	vperm		v0,v0,v0,v3		# rotate [and byte swap in LE]
+	vsel		v1,v1,v0,v2
+	lvx		v4,$idx,$out
+	stvx		v1,0,$out
+	vsel		v0,v0,v4,v2
+	stvx		v0,$idx,$out
+
+	mtspr		256,$vrsave
+	blr
+	.long		0
+	.byte		0,12,0x14,0,0,0,3,0
+	.long		0
+.size	.${prefix}_${dir}crypt,.-.${prefix}_${dir}crypt
+___
+}
+&gen_block("en");
+&gen_block("de");
+}}}
+#########################################################################
+{{{	# CBC en- and decrypt procedures				#
+my ($inp,$out,$len,$key,$ivp,$enc,$rounds,$idx)=map("r$_",(3..10));
+my ($rndkey0,$rndkey1,$inout,$tmp)=		map("v$_",(0..3));
+my ($ivec,$inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm)=
+						map("v$_",(4..10));
+$code.=<<___;
+.globl	.${prefix}_cbc_encrypt
+.align	5
+.${prefix}_cbc_encrypt:
+	${UCMP}i	$len,16
+	bltlr-
+
+	cmpwi		$enc,0			# test direction
+	lis		r0,0xffe0
+	mfspr		$vrsave,256
+	mtspr		256,r0
+
+	li		$idx,15
+	vxor		$rndkey0,$rndkey0,$rndkey0
+	le?vspltisb	$tmp,0x0f
+
+	lvx		$ivec,0,$ivp		# load [unaligned] iv
+	lvsl		$inpperm,0,$ivp
+	lvx		$inptail,$idx,$ivp
+	le?vxor		$inpperm,$inpperm,$tmp
+	vperm		$ivec,$ivec,$inptail,$inpperm
+
+	neg		r11,$inp
+	?lvsl		$keyperm,0,$key		# prepare for unaligned key
+	lwz		$rounds,240($key)
+
+	lvsr		$inpperm,0,r11		# prepare for unaligned load
+	lvx		$inptail,0,$inp
+	addi		$inp,$inp,15		# 15 is not typo
+	le?vxor		$inpperm,$inpperm,$tmp
+
+	?lvsr		$outperm,0,$out		# prepare for unaligned store
+	vspltisb	$outmask,-1
+	lvx		$outhead,0,$out
+	?vperm		$outmask,$rndkey0,$outmask,$outperm
+	le?vxor		$outperm,$outperm,$tmp
+
+	srwi		$rounds,$rounds,1
+	li		$idx,16
+	subi		$rounds,$rounds,1
+	beq		Lcbc_dec
+
+Lcbc_enc:
+	vmr		$inout,$inptail
+	lvx		$inptail,0,$inp
+	addi		$inp,$inp,16
+	mtctr		$rounds
+	subi		$len,$len,16		# len-=16
+
+	lvx		$rndkey0,0,$key
+	 vperm		$inout,$inout,$inptail,$inpperm
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vxor		$inout,$inout,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+	vxor		$inout,$inout,$ivec
+
+Loop_cbc_enc:
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vcipher		$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vcipher		$inout,$inout,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+	bdnz		Loop_cbc_enc
+
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vcipher		$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	li		$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vcipherlast	$ivec,$inout,$rndkey0
+	${UCMP}i	$len,16
+
+	vperm		$tmp,$ivec,$ivec,$outperm
+	vsel		$inout,$outhead,$tmp,$outmask
+	vmr		$outhead,$tmp
+	stvx		$inout,0,$out
+	addi		$out,$out,16
+	bge		Lcbc_enc
+
+	b		Lcbc_done
+
+.align	4
+Lcbc_dec:
+	${UCMP}i	$len,128
+	bge		_aesp8_cbc_decrypt8x
+	vmr		$tmp,$inptail
+	lvx		$inptail,0,$inp
+	addi		$inp,$inp,16
+	mtctr		$rounds
+	subi		$len,$len,16		# len-=16
+
+	lvx		$rndkey0,0,$key
+	 vperm		$tmp,$tmp,$inptail,$inpperm
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vxor		$inout,$tmp,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+
+Loop_cbc_dec:
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vncipher	$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vncipher	$inout,$inout,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+	bdnz		Loop_cbc_dec
+
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vncipher	$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	li		$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vncipherlast	$inout,$inout,$rndkey0
+	${UCMP}i	$len,16
+
+	vxor		$inout,$inout,$ivec
+	vmr		$ivec,$tmp
+	vperm		$tmp,$inout,$inout,$outperm
+	vsel		$inout,$outhead,$tmp,$outmask
+	vmr		$outhead,$tmp
+	stvx		$inout,0,$out
+	addi		$out,$out,16
+	bge		Lcbc_dec
+
+Lcbc_done:
+	addi		$out,$out,-1
+	lvx		$inout,0,$out		# redundant in aligned case
+	vsel		$inout,$outhead,$inout,$outmask
+	stvx		$inout,0,$out
+
+	neg		$enc,$ivp		# write [unaligned] iv
+	li		$idx,15			# 15 is not typo
+	vxor		$rndkey0,$rndkey0,$rndkey0
+	vspltisb	$outmask,-1
+	le?vspltisb	$tmp,0x0f
+	?lvsl		$outperm,0,$enc
+	?vperm		$outmask,$rndkey0,$outmask,$outperm
+	le?vxor		$outperm,$outperm,$tmp
+	lvx		$outhead,0,$ivp
+	vperm		$ivec,$ivec,$ivec,$outperm
+	vsel		$inout,$outhead,$ivec,$outmask
+	lvx		$inptail,$idx,$ivp
+	stvx		$inout,0,$ivp
+	vsel		$inout,$ivec,$inptail,$outmask
+	stvx		$inout,$idx,$ivp
+
+	mtspr		256,$vrsave
+	blr
+	.long		0
+	.byte		0,12,0x14,0,0,0,6,0
+	.long		0
+___
+#########################################################################
+{{	# Optimized CBC decrypt procedure				#
+my $key_="r11";
+my ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70)=map("r$_",(0,8,26..31));
+    $x00=0 if ($flavour =~ /osx/);
+my ($in0, $in1, $in2, $in3, $in4, $in5, $in6, $in7 )=map("v$_",(0..3,10..13));
+my ($out0,$out1,$out2,$out3,$out4,$out5,$out6,$out7)=map("v$_",(14..21));
+my $rndkey0="v23";	# v24-v25 rotating buffer for first found keys
+			# v26-v31 last 6 round keys
+my ($tmp,$keyperm)=($in3,$in4);	# aliases with "caller", redundant assignment
+
+$code.=<<___;
+.align	5
+_aesp8_cbc_decrypt8x:
+	$STU		$sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
+	li		r10,`$FRAME+8*16+15`
+	li		r11,`$FRAME+8*16+31`
+	stvx		v20,r10,$sp		# ABI says so
+	addi		r10,r10,32
+	stvx		v21,r11,$sp
+	addi		r11,r11,32
+	stvx		v22,r10,$sp
+	addi		r10,r10,32
+	stvx		v23,r11,$sp
+	addi		r11,r11,32
+	stvx		v24,r10,$sp
+	addi		r10,r10,32
+	stvx		v25,r11,$sp
+	addi		r11,r11,32
+	stvx		v26,r10,$sp
+	addi		r10,r10,32
+	stvx		v27,r11,$sp
+	addi		r11,r11,32
+	stvx		v28,r10,$sp
+	addi		r10,r10,32
+	stvx		v29,r11,$sp
+	addi		r11,r11,32
+	stvx		v30,r10,$sp
+	stvx		v31,r11,$sp
+	li		r0,-1
+	stw		$vrsave,`$FRAME+21*16-4`($sp)	# save vrsave
+	li		$x10,0x10
+	$PUSH		r26,`$FRAME+21*16+0*$SIZE_T`($sp)
+	li		$x20,0x20
+	$PUSH		r27,`$FRAME+21*16+1*$SIZE_T`($sp)
+	li		$x30,0x30
+	$PUSH		r28,`$FRAME+21*16+2*$SIZE_T`($sp)
+	li		$x40,0x40
+	$PUSH		r29,`$FRAME+21*16+3*$SIZE_T`($sp)
+	li		$x50,0x50
+	$PUSH		r30,`$FRAME+21*16+4*$SIZE_T`($sp)
+	li		$x60,0x60
+	$PUSH		r31,`$FRAME+21*16+5*$SIZE_T`($sp)
+	li		$x70,0x70
+	mtspr		256,r0
+
+	subi		$rounds,$rounds,3	# -4 in total
+	subi		$len,$len,128		# bias
+
+	lvx		$rndkey0,$x00,$key	# load key schedule
+	lvx		v30,$x10,$key
+	addi		$key,$key,0x20
+	lvx		v31,$x00,$key
+	?vperm		$rndkey0,$rndkey0,v30,$keyperm
+	addi		$key_,$sp,$FRAME+15
+	mtctr		$rounds
+
+Load_cbc_dec_key:
+	?vperm		v24,v30,v31,$keyperm
+	lvx		v30,$x10,$key
+	addi		$key,$key,0x20
+	stvx		v24,$x00,$key_		# off-load round[1]
+	?vperm		v25,v31,v30,$keyperm
+	lvx		v31,$x00,$key
+	stvx		v25,$x10,$key_		# off-load round[2]
+	addi		$key_,$key_,0x20
+	bdnz		Load_cbc_dec_key
+
+	lvx		v26,$x10,$key
+	?vperm		v24,v30,v31,$keyperm
+	lvx		v27,$x20,$key
+	stvx		v24,$x00,$key_		# off-load round[3]
+	?vperm		v25,v31,v26,$keyperm
+	lvx		v28,$x30,$key
+	stvx		v25,$x10,$key_		# off-load round[4]
+	addi		$key_,$sp,$FRAME+15	# rewind $key_
+	?vperm		v26,v26,v27,$keyperm
+	lvx		v29,$x40,$key
+	?vperm		v27,v27,v28,$keyperm
+	lvx		v30,$x50,$key
+	?vperm		v28,v28,v29,$keyperm
+	lvx		v31,$x60,$key
+	?vperm		v29,v29,v30,$keyperm
+	lvx		$out0,$x70,$key		# borrow $out0
+	?vperm		v30,v30,v31,$keyperm
+	lvx		v24,$x00,$key_		# pre-load round[1]
+	?vperm		v31,v31,$out0,$keyperm
+	lvx		v25,$x10,$key_		# pre-load round[2]
+
+	#lvx		$inptail,0,$inp		# "caller" already did this
+	#addi		$inp,$inp,15		# 15 is not typo
+	subi		$inp,$inp,15		# undo "caller"
+
+	 le?li		$idx,8
+	lvx_u		$in0,$x00,$inp		# load first 8 "words"
+	 le?lvsl	$inpperm,0,$idx
+	 le?vspltisb	$tmp,0x0f
+	lvx_u		$in1,$x10,$inp
+	 le?vxor	$inpperm,$inpperm,$tmp	# transform for lvx_u/stvx_u
+	lvx_u		$in2,$x20,$inp
+	 le?vperm	$in0,$in0,$in0,$inpperm
+	lvx_u		$in3,$x30,$inp
+	 le?vperm	$in1,$in1,$in1,$inpperm
+	lvx_u		$in4,$x40,$inp
+	 le?vperm	$in2,$in2,$in2,$inpperm
+	vxor		$out0,$in0,$rndkey0
+	lvx_u		$in5,$x50,$inp
+	 le?vperm	$in3,$in3,$in3,$inpperm
+	vxor		$out1,$in1,$rndkey0
+	lvx_u		$in6,$x60,$inp
+	 le?vperm	$in4,$in4,$in4,$inpperm
+	vxor		$out2,$in2,$rndkey0
+	lvx_u		$in7,$x70,$inp
+	addi		$inp,$inp,0x80
+	 le?vperm	$in5,$in5,$in5,$inpperm
+	vxor		$out3,$in3,$rndkey0
+	 le?vperm	$in6,$in6,$in6,$inpperm
+	vxor		$out4,$in4,$rndkey0
+	 le?vperm	$in7,$in7,$in7,$inpperm
+	vxor		$out5,$in5,$rndkey0
+	vxor		$out6,$in6,$rndkey0
+	vxor		$out7,$in7,$rndkey0
+
+	mtctr		$rounds
+	b		Loop_cbc_dec8x
+.align	5
+Loop_cbc_dec8x:
+	vncipher	$out0,$out0,v24
+	vncipher	$out1,$out1,v24
+	vncipher	$out2,$out2,v24
+	vncipher	$out3,$out3,v24
+	vncipher	$out4,$out4,v24
+	vncipher	$out5,$out5,v24
+	vncipher	$out6,$out6,v24
+	vncipher	$out7,$out7,v24
+	lvx		v24,$x20,$key_		# round[3]
+	addi		$key_,$key_,0x20
+
+	vncipher	$out0,$out0,v25
+	vncipher	$out1,$out1,v25
+	vncipher	$out2,$out2,v25
+	vncipher	$out3,$out3,v25
+	vncipher	$out4,$out4,v25
+	vncipher	$out5,$out5,v25
+	vncipher	$out6,$out6,v25
+	vncipher	$out7,$out7,v25
+	lvx		v25,$x10,$key_		# round[4]
+	bdnz		Loop_cbc_dec8x
+
+	subic		$len,$len,128		# $len-=128
+	vncipher	$out0,$out0,v24
+	vncipher	$out1,$out1,v24
+	vncipher	$out2,$out2,v24
+	vncipher	$out3,$out3,v24
+	vncipher	$out4,$out4,v24
+	vncipher	$out5,$out5,v24
+	vncipher	$out6,$out6,v24
+	vncipher	$out7,$out7,v24
+
+	subfe.		r0,r0,r0		# borrow?-1:0
+	vncipher	$out0,$out0,v25
+	vncipher	$out1,$out1,v25
+	vncipher	$out2,$out2,v25
+	vncipher	$out3,$out3,v25
+	vncipher	$out4,$out4,v25
+	vncipher	$out5,$out5,v25
+	vncipher	$out6,$out6,v25
+	vncipher	$out7,$out7,v25
+
+	and		r0,r0,$len
+	vncipher	$out0,$out0,v26
+	vncipher	$out1,$out1,v26
+	vncipher	$out2,$out2,v26
+	vncipher	$out3,$out3,v26
+	vncipher	$out4,$out4,v26
+	vncipher	$out5,$out5,v26
+	vncipher	$out6,$out6,v26
+	vncipher	$out7,$out7,v26
+
+	add		$inp,$inp,r0		# $inp is adjusted in such
+						# way that at exit from the
+						# loop inX-in7 are loaded
+						# with last "words"
+	vncipher	$out0,$out0,v27
+	vncipher	$out1,$out1,v27
+	vncipher	$out2,$out2,v27
+	vncipher	$out3,$out3,v27
+	vncipher	$out4,$out4,v27
+	vncipher	$out5,$out5,v27
+	vncipher	$out6,$out6,v27
+	vncipher	$out7,$out7,v27
+
+	addi		$key_,$sp,$FRAME+15	# rewind $key_
+	vncipher	$out0,$out0,v28
+	vncipher	$out1,$out1,v28
+	vncipher	$out2,$out2,v28
+	vncipher	$out3,$out3,v28
+	vncipher	$out4,$out4,v28
+	vncipher	$out5,$out5,v28
+	vncipher	$out6,$out6,v28
+	vncipher	$out7,$out7,v28
+	lvx		v24,$x00,$key_		# re-pre-load round[1]
+
+	vncipher	$out0,$out0,v29
+	vncipher	$out1,$out1,v29
+	vncipher	$out2,$out2,v29
+	vncipher	$out3,$out3,v29
+	vncipher	$out4,$out4,v29
+	vncipher	$out5,$out5,v29
+	vncipher	$out6,$out6,v29
+	vncipher	$out7,$out7,v29
+	lvx		v25,$x10,$key_		# re-pre-load round[2]
+
+	vncipher	$out0,$out0,v30
+	 vxor		$ivec,$ivec,v31		# xor with last round key
+	vncipher	$out1,$out1,v30
+	 vxor		$in0,$in0,v31
+	vncipher	$out2,$out2,v30
+	 vxor		$in1,$in1,v31
+	vncipher	$out3,$out3,v30
+	 vxor		$in2,$in2,v31
+	vncipher	$out4,$out4,v30
+	 vxor		$in3,$in3,v31
+	vncipher	$out5,$out5,v30
+	 vxor		$in4,$in4,v31
+	vncipher	$out6,$out6,v30
+	 vxor		$in5,$in5,v31
+	vncipher	$out7,$out7,v30
+	 vxor		$in6,$in6,v31
+
+	vncipherlast	$out0,$out0,$ivec
+	vncipherlast	$out1,$out1,$in0
+	 lvx_u		$in0,$x00,$inp		# load next input block
+	vncipherlast	$out2,$out2,$in1
+	 lvx_u		$in1,$x10,$inp
+	vncipherlast	$out3,$out3,$in2
+	 le?vperm	$in0,$in0,$in0,$inpperm
+	 lvx_u		$in2,$x20,$inp
+	vncipherlast	$out4,$out4,$in3
+	 le?vperm	$in1,$in1,$in1,$inpperm
+	 lvx_u		$in3,$x30,$inp
+	vncipherlast	$out5,$out5,$in4
+	 le?vperm	$in2,$in2,$in2,$inpperm
+	 lvx_u		$in4,$x40,$inp
+	vncipherlast	$out6,$out6,$in5
+	 le?vperm	$in3,$in3,$in3,$inpperm
+	 lvx_u		$in5,$x50,$inp
+	vncipherlast	$out7,$out7,$in6
+	 le?vperm	$in4,$in4,$in4,$inpperm
+	 lvx_u		$in6,$x60,$inp
+	vmr		$ivec,$in7
+	 le?vperm	$in5,$in5,$in5,$inpperm
+	 lvx_u		$in7,$x70,$inp
+	 addi		$inp,$inp,0x80
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	 le?vperm	$in6,$in6,$in6,$inpperm
+	 vxor		$out0,$in0,$rndkey0
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	 le?vperm	$in7,$in7,$in7,$inpperm
+	 vxor		$out1,$in1,$rndkey0
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	 vxor		$out2,$in2,$rndkey0
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x30,$out
+	 vxor		$out3,$in3,$rndkey0
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x40,$out
+	 vxor		$out4,$in4,$rndkey0
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x50,$out
+	 vxor		$out5,$in5,$rndkey0
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x60,$out
+	 vxor		$out6,$in6,$rndkey0
+	stvx_u		$out7,$x70,$out
+	addi		$out,$out,0x80
+	 vxor		$out7,$in7,$rndkey0
+
+	mtctr		$rounds
+	beq		Loop_cbc_dec8x		# did $len-=128 borrow?
+
+	addic.		$len,$len,128
+	beq		Lcbc_dec8x_done
+	nop
+	nop
+
+Loop_cbc_dec8x_tail:				# up to 7 "words" tail...
+	vncipher	$out1,$out1,v24
+	vncipher	$out2,$out2,v24
+	vncipher	$out3,$out3,v24
+	vncipher	$out4,$out4,v24
+	vncipher	$out5,$out5,v24
+	vncipher	$out6,$out6,v24
+	vncipher	$out7,$out7,v24
+	lvx		v24,$x20,$key_		# round[3]
+	addi		$key_,$key_,0x20
+
+	vncipher	$out1,$out1,v25
+	vncipher	$out2,$out2,v25
+	vncipher	$out3,$out3,v25
+	vncipher	$out4,$out4,v25
+	vncipher	$out5,$out5,v25
+	vncipher	$out6,$out6,v25
+	vncipher	$out7,$out7,v25
+	lvx		v25,$x10,$key_		# round[4]
+	bdnz		Loop_cbc_dec8x_tail
+
+	vncipher	$out1,$out1,v24
+	vncipher	$out2,$out2,v24
+	vncipher	$out3,$out3,v24
+	vncipher	$out4,$out4,v24
+	vncipher	$out5,$out5,v24
+	vncipher	$out6,$out6,v24
+	vncipher	$out7,$out7,v24
+
+	vncipher	$out1,$out1,v25
+	vncipher	$out2,$out2,v25
+	vncipher	$out3,$out3,v25
+	vncipher	$out4,$out4,v25
+	vncipher	$out5,$out5,v25
+	vncipher	$out6,$out6,v25
+	vncipher	$out7,$out7,v25
+
+	vncipher	$out1,$out1,v26
+	vncipher	$out2,$out2,v26
+	vncipher	$out3,$out3,v26
+	vncipher	$out4,$out4,v26
+	vncipher	$out5,$out5,v26
+	vncipher	$out6,$out6,v26
+	vncipher	$out7,$out7,v26
+
+	vncipher	$out1,$out1,v27
+	vncipher	$out2,$out2,v27
+	vncipher	$out3,$out3,v27
+	vncipher	$out4,$out4,v27
+	vncipher	$out5,$out5,v27
+	vncipher	$out6,$out6,v27
+	vncipher	$out7,$out7,v27
+
+	vncipher	$out1,$out1,v28
+	vncipher	$out2,$out2,v28
+	vncipher	$out3,$out3,v28
+	vncipher	$out4,$out4,v28
+	vncipher	$out5,$out5,v28
+	vncipher	$out6,$out6,v28
+	vncipher	$out7,$out7,v28
+
+	vncipher	$out1,$out1,v29
+	vncipher	$out2,$out2,v29
+	vncipher	$out3,$out3,v29
+	vncipher	$out4,$out4,v29
+	vncipher	$out5,$out5,v29
+	vncipher	$out6,$out6,v29
+	vncipher	$out7,$out7,v29
+
+	vncipher	$out1,$out1,v30
+	 vxor		$ivec,$ivec,v31		# last round key
+	vncipher	$out2,$out2,v30
+	 vxor		$in1,$in1,v31
+	vncipher	$out3,$out3,v30
+	 vxor		$in2,$in2,v31
+	vncipher	$out4,$out4,v30
+	 vxor		$in3,$in3,v31
+	vncipher	$out5,$out5,v30
+	 vxor		$in4,$in4,v31
+	vncipher	$out6,$out6,v30
+	 vxor		$in5,$in5,v31
+	vncipher	$out7,$out7,v30
+	 vxor		$in6,$in6,v31
+
+	cmplwi		$len,32			# switch($len)
+	blt		Lcbc_dec8x_one
+	nop
+	beq		Lcbc_dec8x_two
+	cmplwi		$len,64
+	blt		Lcbc_dec8x_three
+	nop
+	beq		Lcbc_dec8x_four
+	cmplwi		$len,96
+	blt		Lcbc_dec8x_five
+	nop
+	beq		Lcbc_dec8x_six
+
+Lcbc_dec8x_seven:
+	vncipherlast	$out1,$out1,$ivec
+	vncipherlast	$out2,$out2,$in1
+	vncipherlast	$out3,$out3,$in2
+	vncipherlast	$out4,$out4,$in3
+	vncipherlast	$out5,$out5,$in4
+	vncipherlast	$out6,$out6,$in5
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out1,$out1,$out1,$inpperm
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x00,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x10,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x20,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x30,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x40,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x50,$out
+	stvx_u		$out7,$x60,$out
+	addi		$out,$out,0x70
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_six:
+	vncipherlast	$out2,$out2,$ivec
+	vncipherlast	$out3,$out3,$in2
+	vncipherlast	$out4,$out4,$in3
+	vncipherlast	$out5,$out5,$in4
+	vncipherlast	$out6,$out6,$in5
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out2,$out2,$out2,$inpperm
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x00,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x10,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x20,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x30,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x40,$out
+	stvx_u		$out7,$x50,$out
+	addi		$out,$out,0x60
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_five:
+	vncipherlast	$out3,$out3,$ivec
+	vncipherlast	$out4,$out4,$in3
+	vncipherlast	$out5,$out5,$in4
+	vncipherlast	$out6,$out6,$in5
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out3,$out3,$out3,$inpperm
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x00,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x10,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x20,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x30,$out
+	stvx_u		$out7,$x40,$out
+	addi		$out,$out,0x50
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_four:
+	vncipherlast	$out4,$out4,$ivec
+	vncipherlast	$out5,$out5,$in4
+	vncipherlast	$out6,$out6,$in5
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out4,$out4,$out4,$inpperm
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x00,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x10,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x20,$out
+	stvx_u		$out7,$x30,$out
+	addi		$out,$out,0x40
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_three:
+	vncipherlast	$out5,$out5,$ivec
+	vncipherlast	$out6,$out6,$in5
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out5,$out5,$out5,$inpperm
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x00,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x10,$out
+	stvx_u		$out7,$x20,$out
+	addi		$out,$out,0x30
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_two:
+	vncipherlast	$out6,$out6,$ivec
+	vncipherlast	$out7,$out7,$in6
+	vmr		$ivec,$in7
+
+	le?vperm	$out6,$out6,$out6,$inpperm
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x00,$out
+	stvx_u		$out7,$x10,$out
+	addi		$out,$out,0x20
+	b		Lcbc_dec8x_done
+
+.align	5
+Lcbc_dec8x_one:
+	vncipherlast	$out7,$out7,$ivec
+	vmr		$ivec,$in7
+
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out7,0,$out
+	addi		$out,$out,0x10
+
+Lcbc_dec8x_done:
+	le?vperm	$ivec,$ivec,$ivec,$inpperm
+	stvx_u		$ivec,0,$ivp		# write [unaligned] iv
+
+	li		r10,`$FRAME+15`
+	li		r11,`$FRAME+31`
+	stvx		$inpperm,r10,$sp	# wipe copies of round keys
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+
+	mtspr		256,$vrsave
+	lvx		v20,r10,$sp		# ABI says so
+	addi		r10,r10,32
+	lvx		v21,r11,$sp
+	addi		r11,r11,32
+	lvx		v22,r10,$sp
+	addi		r10,r10,32
+	lvx		v23,r11,$sp
+	addi		r11,r11,32
+	lvx		v24,r10,$sp
+	addi		r10,r10,32
+	lvx		v25,r11,$sp
+	addi		r11,r11,32
+	lvx		v26,r10,$sp
+	addi		r10,r10,32
+	lvx		v27,r11,$sp
+	addi		r11,r11,32
+	lvx		v28,r10,$sp
+	addi		r10,r10,32
+	lvx		v29,r11,$sp
+	addi		r11,r11,32
+	lvx		v30,r10,$sp
+	lvx		v31,r11,$sp
+	$POP		r26,`$FRAME+21*16+0*$SIZE_T`($sp)
+	$POP		r27,`$FRAME+21*16+1*$SIZE_T`($sp)
+	$POP		r28,`$FRAME+21*16+2*$SIZE_T`($sp)
+	$POP		r29,`$FRAME+21*16+3*$SIZE_T`($sp)
+	$POP		r30,`$FRAME+21*16+4*$SIZE_T`($sp)
+	$POP		r31,`$FRAME+21*16+5*$SIZE_T`($sp)
+	addi		$sp,$sp,`$FRAME+21*16+6*$SIZE_T`
+	blr
+	.long		0
+	.byte		0,12,0x04,0,0x80,6,6,0
+	.long		0
+.size	.${prefix}_cbc_encrypt,.-.${prefix}_cbc_encrypt
+___
+}}	}}}
+
+#########################################################################
+{{{	# CTR procedure[s]						#
+my ($inp,$out,$len,$key,$ivp,$x10,$rounds,$idx)=map("r$_",(3..10));
+my ($rndkey0,$rndkey1,$inout,$tmp)=		map("v$_",(0..3));
+my ($ivec,$inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm,$one)=
+						map("v$_",(4..11));
+my $dat=$tmp;
+
+$code.=<<___;
+.globl	.${prefix}_ctr32_encrypt_blocks
+.align	5
+.${prefix}_ctr32_encrypt_blocks:
+	${UCMP}i	$len,1
+	bltlr-
+
+	lis		r0,0xfff0
+	mfspr		$vrsave,256
+	mtspr		256,r0
+
+	li		$idx,15
+	vxor		$rndkey0,$rndkey0,$rndkey0
+	le?vspltisb	$tmp,0x0f
+
+	lvx		$ivec,0,$ivp		# load [unaligned] iv
+	lvsl		$inpperm,0,$ivp
+	lvx		$inptail,$idx,$ivp
+	 vspltisb	$one,1
+	le?vxor		$inpperm,$inpperm,$tmp
+	vperm		$ivec,$ivec,$inptail,$inpperm
+	 vsldoi		$one,$rndkey0,$one,1
+
+	neg		r11,$inp
+	?lvsl		$keyperm,0,$key		# prepare for unaligned key
+	lwz		$rounds,240($key)
+
+	lvsr		$inpperm,0,r11		# prepare for unaligned load
+	lvx		$inptail,0,$inp
+	addi		$inp,$inp,15		# 15 is not typo
+	le?vxor		$inpperm,$inpperm,$tmp
+
+	srwi		$rounds,$rounds,1
+	li		$idx,16
+	subi		$rounds,$rounds,1
+
+	${UCMP}i	$len,8
+	bge		_aesp8_ctr32_encrypt8x
+
+	?lvsr		$outperm,0,$out		# prepare for unaligned store
+	vspltisb	$outmask,-1
+	lvx		$outhead,0,$out
+	?vperm		$outmask,$rndkey0,$outmask,$outperm
+	le?vxor		$outperm,$outperm,$tmp
+
+	lvx		$rndkey0,0,$key
+	mtctr		$rounds
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vxor		$inout,$ivec,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+	b		Loop_ctr32_enc
+
+.align	5
+Loop_ctr32_enc:
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vcipher		$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	addi		$idx,$idx,16
+	?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vcipher		$inout,$inout,$rndkey0
+	lvx		$rndkey0,$idx,$key
+	addi		$idx,$idx,16
+	bdnz		Loop_ctr32_enc
+
+	vadduwm		$ivec,$ivec,$one
+	 vmr		$dat,$inptail
+	 lvx		$inptail,0,$inp
+	 addi		$inp,$inp,16
+	 subic.		$len,$len,1		# blocks--
+
+	?vperm		$rndkey1,$rndkey1,$rndkey0,$keyperm
+	vcipher		$inout,$inout,$rndkey1
+	lvx		$rndkey1,$idx,$key
+	 vperm		$dat,$dat,$inptail,$inpperm
+	 li		$idx,16
+	?vperm		$rndkey1,$rndkey0,$rndkey1,$keyperm
+	 lvx		$rndkey0,0,$key
+	vxor		$dat,$dat,$rndkey1	# last round key
+	vcipherlast	$inout,$inout,$dat
+
+	 lvx		$rndkey1,$idx,$key
+	 addi		$idx,$idx,16
+	vperm		$inout,$inout,$inout,$outperm
+	vsel		$dat,$outhead,$inout,$outmask
+	 mtctr		$rounds
+	 ?vperm		$rndkey0,$rndkey0,$rndkey1,$keyperm
+	vmr		$outhead,$inout
+	 vxor		$inout,$ivec,$rndkey0
+	 lvx		$rndkey0,$idx,$key
+	 addi		$idx,$idx,16
+	stvx		$dat,0,$out
+	addi		$out,$out,16
+	bne		Loop_ctr32_enc
+
+	addi		$out,$out,-1
+	lvx		$inout,0,$out		# redundant in aligned case
+	vsel		$inout,$outhead,$inout,$outmask
+	stvx		$inout,0,$out
+
+	mtspr		256,$vrsave
+	blr
+	.long		0
+	.byte		0,12,0x14,0,0,0,6,0
+	.long		0
+___
+#########################################################################
+{{	# Optimized CTR procedure					#
+my $key_="r11";
+my ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70)=map("r$_",(0,8,26..31));
+    $x00=0 if ($flavour =~ /osx/);
+my ($in0, $in1, $in2, $in3, $in4, $in5, $in6, $in7 )=map("v$_",(0..3,10,12..14));
+my ($out0,$out1,$out2,$out3,$out4,$out5,$out6,$out7)=map("v$_",(15..22));
+my $rndkey0="v23";	# v24-v25 rotating buffer for first found keys
+			# v26-v31 last 6 round keys
+my ($tmp,$keyperm)=($in3,$in4);	# aliases with "caller", redundant assignment
+my ($two,$three,$four)=($outhead,$outperm,$outmask);
+
+$code.=<<___;
+.align	5
+_aesp8_ctr32_encrypt8x:
+	$STU		$sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
+	li		r10,`$FRAME+8*16+15`
+	li		r11,`$FRAME+8*16+31`
+	stvx		v20,r10,$sp		# ABI says so
+	addi		r10,r10,32
+	stvx		v21,r11,$sp
+	addi		r11,r11,32
+	stvx		v22,r10,$sp
+	addi		r10,r10,32
+	stvx		v23,r11,$sp
+	addi		r11,r11,32
+	stvx		v24,r10,$sp
+	addi		r10,r10,32
+	stvx		v25,r11,$sp
+	addi		r11,r11,32
+	stvx		v26,r10,$sp
+	addi		r10,r10,32
+	stvx		v27,r11,$sp
+	addi		r11,r11,32
+	stvx		v28,r10,$sp
+	addi		r10,r10,32
+	stvx		v29,r11,$sp
+	addi		r11,r11,32
+	stvx		v30,r10,$sp
+	stvx		v31,r11,$sp
+	li		r0,-1
+	stw		$vrsave,`$FRAME+21*16-4`($sp)	# save vrsave
+	li		$x10,0x10
+	$PUSH		r26,`$FRAME+21*16+0*$SIZE_T`($sp)
+	li		$x20,0x20
+	$PUSH		r27,`$FRAME+21*16+1*$SIZE_T`($sp)
+	li		$x30,0x30
+	$PUSH		r28,`$FRAME+21*16+2*$SIZE_T`($sp)
+	li		$x40,0x40
+	$PUSH		r29,`$FRAME+21*16+3*$SIZE_T`($sp)
+	li		$x50,0x50
+	$PUSH		r30,`$FRAME+21*16+4*$SIZE_T`($sp)
+	li		$x60,0x60
+	$PUSH		r31,`$FRAME+21*16+5*$SIZE_T`($sp)
+	li		$x70,0x70
+	mtspr		256,r0
+
+	subi		$rounds,$rounds,3	# -4 in total
+
+	lvx		$rndkey0,$x00,$key	# load key schedule
+	lvx		v30,$x10,$key
+	addi		$key,$key,0x20
+	lvx		v31,$x00,$key
+	?vperm		$rndkey0,$rndkey0,v30,$keyperm
+	addi		$key_,$sp,$FRAME+15
+	mtctr		$rounds
+
+Load_ctr32_enc_key:
+	?vperm		v24,v30,v31,$keyperm
+	lvx		v30,$x10,$key
+	addi		$key,$key,0x20
+	stvx		v24,$x00,$key_		# off-load round[1]
+	?vperm		v25,v31,v30,$keyperm
+	lvx		v31,$x00,$key
+	stvx		v25,$x10,$key_		# off-load round[2]
+	addi		$key_,$key_,0x20
+	bdnz		Load_ctr32_enc_key
+
+	lvx		v26,$x10,$key
+	?vperm		v24,v30,v31,$keyperm
+	lvx		v27,$x20,$key
+	stvx		v24,$x00,$key_		# off-load round[3]
+	?vperm		v25,v31,v26,$keyperm
+	lvx		v28,$x30,$key
+	stvx		v25,$x10,$key_		# off-load round[4]
+	addi		$key_,$sp,$FRAME+15	# rewind $key_
+	?vperm		v26,v26,v27,$keyperm
+	lvx		v29,$x40,$key
+	?vperm		v27,v27,v28,$keyperm
+	lvx		v30,$x50,$key
+	?vperm		v28,v28,v29,$keyperm
+	lvx		v31,$x60,$key
+	?vperm		v29,v29,v30,$keyperm
+	lvx		$out0,$x70,$key		# borrow $out0
+	?vperm		v30,v30,v31,$keyperm
+	lvx		v24,$x00,$key_		# pre-load round[1]
+	?vperm		v31,v31,$out0,$keyperm
+	lvx		v25,$x10,$key_		# pre-load round[2]
+
+	vadduwm		$two,$one,$one
+	subi		$inp,$inp,15		# undo "caller"
+	$SHL		$len,$len,4
+
+	vadduwm		$out1,$ivec,$one	# counter values ...
+	vadduwm		$out2,$ivec,$two
+	vxor		$out0,$ivec,$rndkey0	# ... xored with rndkey[0]
+	 le?li		$idx,8
+	vadduwm		$out3,$out1,$two
+	vxor		$out1,$out1,$rndkey0
+	 le?lvsl	$inpperm,0,$idx
+	vadduwm		$out4,$out2,$two
+	vxor		$out2,$out2,$rndkey0
+	 le?vspltisb	$tmp,0x0f
+	vadduwm		$out5,$out3,$two
+	vxor		$out3,$out3,$rndkey0
+	 le?vxor	$inpperm,$inpperm,$tmp	# transform for lvx_u/stvx_u
+	vadduwm		$out6,$out4,$two
+	vxor		$out4,$out4,$rndkey0
+	vadduwm		$out7,$out5,$two
+	vxor		$out5,$out5,$rndkey0
+	vadduwm		$ivec,$out6,$two	# next counter value
+	vxor		$out6,$out6,$rndkey0
+	vxor		$out7,$out7,$rndkey0
+
+	mtctr		$rounds
+	b		Loop_ctr32_enc8x
+.align	5
+Loop_ctr32_enc8x:
+	vcipher 	$out0,$out0,v24
+	vcipher 	$out1,$out1,v24
+	vcipher 	$out2,$out2,v24
+	vcipher 	$out3,$out3,v24
+	vcipher 	$out4,$out4,v24
+	vcipher 	$out5,$out5,v24
+	vcipher 	$out6,$out6,v24
+	vcipher 	$out7,$out7,v24
+Loop_ctr32_enc8x_middle:
+	lvx		v24,$x20,$key_		# round[3]
+	addi		$key_,$key_,0x20
+
+	vcipher 	$out0,$out0,v25
+	vcipher 	$out1,$out1,v25
+	vcipher 	$out2,$out2,v25
+	vcipher 	$out3,$out3,v25
+	vcipher 	$out4,$out4,v25
+	vcipher 	$out5,$out5,v25
+	vcipher 	$out6,$out6,v25
+	vcipher 	$out7,$out7,v25
+	lvx		v25,$x10,$key_		# round[4]
+	bdnz		Loop_ctr32_enc8x
+
+	subic		r11,$len,256		# $len-256, borrow $key_
+	vcipher 	$out0,$out0,v24
+	vcipher 	$out1,$out1,v24
+	vcipher 	$out2,$out2,v24
+	vcipher 	$out3,$out3,v24
+	vcipher 	$out4,$out4,v24
+	vcipher 	$out5,$out5,v24
+	vcipher 	$out6,$out6,v24
+	vcipher 	$out7,$out7,v24
+
+	subfe		r0,r0,r0		# borrow?-1:0
+	vcipher 	$out0,$out0,v25
+	vcipher 	$out1,$out1,v25
+	vcipher 	$out2,$out2,v25
+	vcipher 	$out3,$out3,v25
+	vcipher 	$out4,$out4,v25
+	vcipher		$out5,$out5,v25
+	vcipher		$out6,$out6,v25
+	vcipher		$out7,$out7,v25
+
+	and		r0,r0,r11
+	addi		$key_,$sp,$FRAME+15	# rewind $key_
+	vcipher		$out0,$out0,v26
+	vcipher		$out1,$out1,v26
+	vcipher		$out2,$out2,v26
+	vcipher		$out3,$out3,v26
+	vcipher		$out4,$out4,v26
+	vcipher		$out5,$out5,v26
+	vcipher		$out6,$out6,v26
+	vcipher		$out7,$out7,v26
+	lvx		v24,$x00,$key_		# re-pre-load round[1]
+
+	subic		$len,$len,129		# $len-=129
+	vcipher		$out0,$out0,v27
+	addi		$len,$len,1		# $len-=128 really
+	vcipher		$out1,$out1,v27
+	vcipher		$out2,$out2,v27
+	vcipher		$out3,$out3,v27
+	vcipher		$out4,$out4,v27
+	vcipher		$out5,$out5,v27
+	vcipher		$out6,$out6,v27
+	vcipher		$out7,$out7,v27
+	lvx		v25,$x10,$key_		# re-pre-load round[2]
+
+	vcipher		$out0,$out0,v28
+	 lvx_u		$in0,$x00,$inp		# load input
+	vcipher		$out1,$out1,v28
+	 lvx_u		$in1,$x10,$inp
+	vcipher		$out2,$out2,v28
+	 lvx_u		$in2,$x20,$inp
+	vcipher		$out3,$out3,v28
+	 lvx_u		$in3,$x30,$inp
+	vcipher		$out4,$out4,v28
+	 lvx_u		$in4,$x40,$inp
+	vcipher		$out5,$out5,v28
+	 lvx_u		$in5,$x50,$inp
+	vcipher		$out6,$out6,v28
+	 lvx_u		$in6,$x60,$inp
+	vcipher		$out7,$out7,v28
+	 lvx_u		$in7,$x70,$inp
+	 addi		$inp,$inp,0x80
+
+	vcipher		$out0,$out0,v29
+	 le?vperm	$in0,$in0,$in0,$inpperm
+	vcipher		$out1,$out1,v29
+	 le?vperm	$in1,$in1,$in1,$inpperm
+	vcipher		$out2,$out2,v29
+	 le?vperm	$in2,$in2,$in2,$inpperm
+	vcipher		$out3,$out3,v29
+	 le?vperm	$in3,$in3,$in3,$inpperm
+	vcipher		$out4,$out4,v29
+	 le?vperm	$in4,$in4,$in4,$inpperm
+	vcipher		$out5,$out5,v29
+	 le?vperm	$in5,$in5,$in5,$inpperm
+	vcipher		$out6,$out6,v29
+	 le?vperm	$in6,$in6,$in6,$inpperm
+	vcipher		$out7,$out7,v29
+	 le?vperm	$in7,$in7,$in7,$inpperm
+
+	add		$inp,$inp,r0		# $inp is adjusted in such
+						# way that at exit from the
+						# loop inX-in7 are loaded
+						# with last "words"
+	subfe.		r0,r0,r0		# borrow?-1:0
+	vcipher		$out0,$out0,v30
+	 vxor		$in0,$in0,v31		# xor with last round key
+	vcipher		$out1,$out1,v30
+	 vxor		$in1,$in1,v31
+	vcipher		$out2,$out2,v30
+	 vxor		$in2,$in2,v31
+	vcipher		$out3,$out3,v30
+	 vxor		$in3,$in3,v31
+	vcipher		$out4,$out4,v30
+	 vxor		$in4,$in4,v31
+	vcipher		$out5,$out5,v30
+	 vxor		$in5,$in5,v31
+	vcipher		$out6,$out6,v30
+	 vxor		$in6,$in6,v31
+	vcipher		$out7,$out7,v30
+	 vxor		$in7,$in7,v31
+
+	bne		Lctr32_enc8x_break	# did $len-129 borrow?
+
+	vcipherlast	$in0,$out0,$in0
+	vcipherlast	$in1,$out1,$in1
+	 vadduwm	$out1,$ivec,$one	# counter values ...
+	vcipherlast	$in2,$out2,$in2
+	 vadduwm	$out2,$ivec,$two
+	 vxor		$out0,$ivec,$rndkey0	# ... xored with rndkey[0]
+	vcipherlast	$in3,$out3,$in3
+	 vadduwm	$out3,$out1,$two
+	 vxor		$out1,$out1,$rndkey0
+	vcipherlast	$in4,$out4,$in4
+	 vadduwm	$out4,$out2,$two
+	 vxor		$out2,$out2,$rndkey0
+	vcipherlast	$in5,$out5,$in5
+	 vadduwm	$out5,$out3,$two
+	 vxor		$out3,$out3,$rndkey0
+	vcipherlast	$in6,$out6,$in6
+	 vadduwm	$out6,$out4,$two
+	 vxor		$out4,$out4,$rndkey0
+	vcipherlast	$in7,$out7,$in7
+	 vadduwm	$out7,$out5,$two
+	 vxor		$out5,$out5,$rndkey0
+	le?vperm	$in0,$in0,$in0,$inpperm
+	 vadduwm	$ivec,$out6,$two	# next counter value
+	 vxor		$out6,$out6,$rndkey0
+	le?vperm	$in1,$in1,$in1,$inpperm
+	 vxor		$out7,$out7,$rndkey0
+	mtctr		$rounds
+
+	 vcipher	$out0,$out0,v24
+	stvx_u		$in0,$x00,$out
+	le?vperm	$in2,$in2,$in2,$inpperm
+	 vcipher	$out1,$out1,v24
+	stvx_u		$in1,$x10,$out
+	le?vperm	$in3,$in3,$in3,$inpperm
+	 vcipher	$out2,$out2,v24
+	stvx_u		$in2,$x20,$out
+	le?vperm	$in4,$in4,$in4,$inpperm
+	 vcipher	$out3,$out3,v24
+	stvx_u		$in3,$x30,$out
+	le?vperm	$in5,$in5,$in5,$inpperm
+	 vcipher	$out4,$out4,v24
+	stvx_u		$in4,$x40,$out
+	le?vperm	$in6,$in6,$in6,$inpperm
+	 vcipher	$out5,$out5,v24
+	stvx_u		$in5,$x50,$out
+	le?vperm	$in7,$in7,$in7,$inpperm
+	 vcipher	$out6,$out6,v24
+	stvx_u		$in6,$x60,$out
+	 vcipher	$out7,$out7,v24
+	stvx_u		$in7,$x70,$out
+	addi		$out,$out,0x80
+
+	b		Loop_ctr32_enc8x_middle
+
+.align	5
+Lctr32_enc8x_break:
+	cmpwi		$len,-0x60
+	blt		Lctr32_enc8x_one
+	nop
+	beq		Lctr32_enc8x_two
+	cmpwi		$len,-0x40
+	blt		Lctr32_enc8x_three
+	nop
+	beq		Lctr32_enc8x_four
+	cmpwi		$len,-0x20
+	blt		Lctr32_enc8x_five
+	nop
+	beq		Lctr32_enc8x_six
+	cmpwi		$len,0x00
+	blt		Lctr32_enc8x_seven
+
+Lctr32_enc8x_eight:
+	vcipherlast	$out0,$out0,$in0
+	vcipherlast	$out1,$out1,$in1
+	vcipherlast	$out2,$out2,$in2
+	vcipherlast	$out3,$out3,$in3
+	vcipherlast	$out4,$out4,$in4
+	vcipherlast	$out5,$out5,$in5
+	vcipherlast	$out6,$out6,$in6
+	vcipherlast	$out7,$out7,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x30,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x40,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x50,$out
+	le?vperm	$out7,$out7,$out7,$inpperm
+	stvx_u		$out6,$x60,$out
+	stvx_u		$out7,$x70,$out
+	addi		$out,$out,0x80
+	b		Lctr32_enc8x_done
+
+.align	5
+Lctr32_enc8x_seven:
+	vcipherlast	$out0,$out0,$in1
+	vcipherlast	$out1,$out1,$in2
+	vcipherlast	$out2,$out2,$in3
+	vcipherlast	$out3,$out3,$in4
+	vcipherlast	$out4,$out4,$in5
+	vcipherlast	$out5,$out5,$in6
+	vcipherlast	$out6,$out6,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x30,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x40,$out
+	le?vperm	$out6,$out6,$out6,$inpperm
+	stvx_u		$out5,$x50,$out
+	stvx_u		$out6,$x60,$out
+	addi		$out,$out,0x70
+	b		Lctr32_enc8x_done
+
+.align	5
+Lctr32_enc8x_six:
+	vcipherlast	$out0,$out0,$in2
+	vcipherlast	$out1,$out1,$in3
+	vcipherlast	$out2,$out2,$in4
+	vcipherlast	$out3,$out3,$in5
+	vcipherlast	$out4,$out4,$in6
+	vcipherlast	$out5,$out5,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x30,$out
+	le?vperm	$out5,$out5,$out5,$inpperm
+	stvx_u		$out4,$x40,$out
+	stvx_u		$out5,$x50,$out
+	addi		$out,$out,0x60
+	b		Lctr32_enc8x_done
+
+.align	5
+Lctr32_enc8x_five:
+	vcipherlast	$out0,$out0,$in3
+	vcipherlast	$out1,$out1,$in4
+	vcipherlast	$out2,$out2,$in5
+	vcipherlast	$out3,$out3,$in6
+	vcipherlast	$out4,$out4,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	le?vperm	$out4,$out4,$out4,$inpperm
+	stvx_u		$out3,$x30,$out
+	stvx_u		$out4,$x40,$out
+	addi		$out,$out,0x50
+	b		Lctr32_enc8x_done
+
+.align	5
+Lctr32_enc8x_four:
+	vcipherlast	$out0,$out0,$in4
+	vcipherlast	$out1,$out1,$in5
+	vcipherlast	$out2,$out2,$in6
+	vcipherlast	$out3,$out3,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	le?vperm	$out3,$out3,$out3,$inpperm
+	stvx_u		$out2,$x20,$out
+	stvx_u		$out3,$x30,$out
+	addi		$out,$out,0x40
+	b		Lctr32_enc8x_done
+
+.align	5
+Lctr32_enc8x_three:
+	vcipherlast	$out0,$out0,$in5
+	vcipherlast	$out1,$out1,$in6
+	vcipherlast	$out2,$out2,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	le?vperm	$out2,$out2,$out2,$inpperm
+	stvx_u		$out1,$x10,$out
+	stvx_u		$out2,$x20,$out
+	addi		$out,$out,0x30
+	b		Lcbc_dec8x_done
+
+.align	5
+Lctr32_enc8x_two:
+	vcipherlast	$out0,$out0,$in6
+	vcipherlast	$out1,$out1,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	le?vperm	$out1,$out1,$out1,$inpperm
+	stvx_u		$out0,$x00,$out
+	stvx_u		$out1,$x10,$out
+	addi		$out,$out,0x20
+	b		Lcbc_dec8x_done
+
+.align	5
+Lctr32_enc8x_one:
+	vcipherlast	$out0,$out0,$in7
+
+	le?vperm	$out0,$out0,$out0,$inpperm
+	stvx_u		$out0,0,$out
+	addi		$out,$out,0x10
+
+Lctr32_enc8x_done:
+	li		r10,`$FRAME+15`
+	li		r11,`$FRAME+31`
+	stvx		$inpperm,r10,$sp	# wipe copies of round keys
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+	stvx		$inpperm,r10,$sp
+	addi		r10,r10,32
+	stvx		$inpperm,r11,$sp
+	addi		r11,r11,32
+
+	mtspr		256,$vrsave
+	lvx		v20,r10,$sp		# ABI says so
+	addi		r10,r10,32
+	lvx		v21,r11,$sp
+	addi		r11,r11,32
+	lvx		v22,r10,$sp
+	addi		r10,r10,32
+	lvx		v23,r11,$sp
+	addi		r11,r11,32
+	lvx		v24,r10,$sp
+	addi		r10,r10,32
+	lvx		v25,r11,$sp
+	addi		r11,r11,32
+	lvx		v26,r10,$sp
+	addi		r10,r10,32
+	lvx		v27,r11,$sp
+	addi		r11,r11,32
+	lvx		v28,r10,$sp
+	addi		r10,r10,32
+	lvx		v29,r11,$sp
+	addi		r11,r11,32
+	lvx		v30,r10,$sp
+	lvx		v31,r11,$sp
+	$POP		r26,`$FRAME+21*16+0*$SIZE_T`($sp)
+	$POP		r27,`$FRAME+21*16+1*$SIZE_T`($sp)
+	$POP		r28,`$FRAME+21*16+2*$SIZE_T`($sp)
+	$POP		r29,`$FRAME+21*16+3*$SIZE_T`($sp)
+	$POP		r30,`$FRAME+21*16+4*$SIZE_T`($sp)
+	$POP		r31,`$FRAME+21*16+5*$SIZE_T`($sp)
+	addi		$sp,$sp,`$FRAME+21*16+6*$SIZE_T`
+	blr
+	.long		0
+	.byte		0,12,0x04,0,0x80,6,6,0
+	.long		0
+.size	.${prefix}_ctr32_encrypt_blocks,.-.${prefix}_ctr32_encrypt_blocks
+___
+}}	}}}
+
+my $consts=1;
+foreach(split("\n",$code)) {
+        s/\`([^\`]*)\`/eval($1)/geo;
+
+	# constants table endian-specific conversion
+	if ($consts && m/\.(long|byte)\s+(.+)\s+(\?[a-z]*)$/o) {
+	    my $conv=$3;
+	    my @bytes=();
+
+	    # convert to endian-agnostic format
+	    if ($1 eq "long") {
+	      foreach (split(/,\s*/,$2)) {
+		my $l = /^0/?oct:int;
+		push @bytes,($l>>24)&0xff,($l>>16)&0xff,($l>>8)&0xff,$l&0xff;
+	      }
+	    } else {
+		@bytes = map(/^0/?oct:int,split(/,\s*/,$2));
+	    }
+
+	    # little-endian conversion
+	    if ($flavour =~ /le$/o) {
+		SWITCH: for($conv)  {
+		    /\?inv/ && do   { @bytes=map($_^0xf,@bytes); last; };
+		    /\?rev/ && do   { @bytes=reverse(@bytes);    last; }; 
+		}
+	    }
+
+	    #emit
+	    print ".byte\t",join(',',map (sprintf("0x%02x",$_),@bytes)),"\n";
+	    next;
+	}
+	$consts=0 if (m/Lconsts:/o);	# end of table
+
+	# instructions prefixed with '?' are endian-specific and need
+	# to be adjusted accordingly...
+	if ($flavour =~ /le$/o) {	# little-endian
+	    s/le\?//o		or
+	    s/be\?/#be#/o	or
+	    s/\?lvsr/lvsl/o	or
+	    s/\?lvsl/lvsr/o	or
+	    s/\?(vperm\s+v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+)/$1$3$2$4/o or
+	    s/\?(vsldoi\s+v[0-9]+,\s*)(v[0-9]+,)\s*(v[0-9]+,\s*)([0-9]+)/$1$3$2 16-$4/o or
+	    s/\?(vspltw\s+v[0-9]+,\s*)(v[0-9]+,)\s*([0-9])/$1$2 3-$3/o;
+	} else {			# big-endian
+	    s/le\?/#le#/o	or
+	    s/be\?//o		or
+	    s/\?([a-z]+)/$1/o;
+	}
+
+        print $_,"\n";
+}
+
+close STDOUT;
diff --git a/openssl/crypto/aes/asm/aest4-sparcv9.pl b/openssl/crypto/aes/asm/aest4-sparcv9.pl
new file mode 100755
index 000000000..536f23b47
--- /dev/null
+++ b/openssl/crypto/aes/asm/aest4-sparcv9.pl
@@ -0,0 +1,919 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by David S. Miller <davem@devemloft.net> and Andy Polyakov
+# <appro@openssl.org>. The module is licensed under 2-clause BSD
+# license. October 2012. All rights reserved.
+# ====================================================================
+
+######################################################################
+# AES for SPARC T4.
+#
+# AES round instructions complete in 3 cycles and can be issued every
+# cycle. It means that round calculations should take 4*rounds cycles,
+# because any given round instruction depends on result of *both*
+# previous instructions:
+#
+#	|0 |1 |2 |3 |4
+#	|01|01|01|
+#	   |23|23|23|
+#	            |01|01|...
+#	               |23|...
+#
+# Provided that fxor [with IV] takes 3 cycles to complete, critical
+# path length for CBC encrypt would be 3+4*rounds, or in other words
+# it should process one byte in at least (3+4*rounds)/16 cycles. This
+# estimate doesn't account for "collateral" instructions, such as
+# fetching input from memory, xor-ing it with zero-round key and
+# storing the result. Yet, *measured* performance [for data aligned
+# at 64-bit boundary!] deviates from this equation by less than 0.5%:
+#
+#		128-bit key	192-		256-
+# CBC encrypt	2.70/2.90(*)	3.20/3.40	3.70/3.90
+#			 (*) numbers after slash are for
+#			     misaligned data.
+#
+# Out-of-order execution logic managed to fully overlap "collateral"
+# instructions with those on critical path. Amazing!
+#
+# As with Intel AES-NI, question is if it's possible to improve
+# performance of parallelizeable modes by interleaving round
+# instructions. Provided round instruction latency and throughput
+# optimal interleave factor is 2. But can we expect 2x performance
+# improvement? Well, as round instructions can be issued one per
+# cycle, they don't saturate the 2-way issue pipeline and therefore
+# there is room for "collateral" calculations... Yet, 2x speed-up
+# over CBC encrypt remains unattaintable:
+#
+#		128-bit key	192-		256-
+# CBC decrypt	1.64/2.11	1.89/2.37	2.23/2.61
+# CTR		1.64/2.08(*)	1.89/2.33	2.23/2.61
+#			 (*) numbers after slash are for
+#			     misaligned data.
+#
+# Estimates based on amount of instructions under assumption that
+# round instructions are not pairable with any other instruction
+# suggest that latter is the actual case and pipeline runs
+# underutilized. It should be noted that T4 out-of-order execution
+# logic is so capable that performance gain from 2x interleave is
+# not even impressive, ~7-13% over non-interleaved code, largest
+# for 256-bit keys.
+
+# To anchor to something else, software implementation processes
+# one byte in 29 cycles with 128-bit key on same processor. Intel
+# Sandy Bridge encrypts byte in 5.07 cycles in CBC mode and decrypts
+# in 0.93, naturally with AES-NI.
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "sparcv9_modes.pl";
+
+&asm_init(@ARGV);
+
+$::evp=1;	# if $evp is set to 0, script generates module with
+# AES_[en|de]crypt, AES_set_[en|de]crypt_key and AES_cbc_encrypt entry
+# points. These however are not fully compatible with openssl/aes.h,
+# because they expect AES_KEY to be aligned at 64-bit boundary. When
+# used through EVP, alignment is arranged at EVP layer. Second thing
+# that is arranged by EVP is at least 32-bit alignment of IV.
+
+######################################################################
+# single-round subroutines
+#
+{
+my ($inp,$out,$key,$rounds,$tmp,$mask)=map("%o$_",(0..5));
+
+$code.=<<___ if ($::abibits==64);
+.register	%g2,#scratch
+.register	%g3,#scratch
+
+___
+$code.=<<___;
+.text
+
+.globl	aes_t4_encrypt
+.align	32
+aes_t4_encrypt:
+	andcc		$inp, 7, %g1		! is input aligned?
+	andn		$inp, 7, $inp
+
+	ldx		[$key + 0], %g4
+	ldx		[$key + 8], %g5
+
+	ldx		[$inp + 0], %o4
+	bz,pt		%icc, 1f
+	ldx		[$inp + 8], %o5
+	ldx		[$inp + 16], $inp
+	sll		%g1, 3, %g1
+	sub		%g0, %g1, %o3
+	sllx		%o4, %g1, %o4
+	sllx		%o5, %g1, %g1
+	srlx		%o5, %o3, %o5
+	srlx		$inp, %o3, %o3
+	or		%o5, %o4, %o4
+	or		%o3, %g1, %o5
+1:
+	ld		[$key + 240], $rounds
+	ldd		[$key + 16], %f12
+	ldd		[$key + 24], %f14
+	xor		%g4, %o4, %o4
+	xor		%g5, %o5, %o5
+	movxtod		%o4, %f0
+	movxtod		%o5, %f2
+	srl		$rounds, 1, $rounds
+	ldd		[$key + 32], %f16
+	sub		$rounds, 1, $rounds
+	ldd		[$key + 40], %f18
+	add		$key, 48, $key
+
+.Lenc:
+	aes_eround01	%f12, %f0, %f2, %f4
+	aes_eround23	%f14, %f0, %f2, %f2
+	ldd		[$key + 0], %f12
+	ldd		[$key + 8], %f14
+	sub		$rounds,1,$rounds
+	aes_eround01	%f16, %f4, %f2, %f0
+	aes_eround23	%f18, %f4, %f2, %f2
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	brnz,pt		$rounds, .Lenc
+	add		$key, 32, $key
+
+	andcc		$out, 7, $tmp		! is output aligned?
+	aes_eround01	%f12, %f0, %f2, %f4
+	aes_eround23	%f14, %f0, %f2, %f2
+	aes_eround01_l	%f16, %f4, %f2, %f0
+	aes_eround23_l	%f18, %f4, %f2, %f2
+
+	bnz,pn		%icc, 2f
+	nop
+
+	std		%f0, [$out + 0]
+	retl
+	std		%f2, [$out + 8]
+
+2:	alignaddrl	$out, %g0, $out
+	mov		0xff, $mask
+	srl		$mask, $tmp, $mask
+
+	faligndata	%f0, %f0, %f4
+	faligndata	%f0, %f2, %f6
+	faligndata	%f2, %f2, %f8
+
+	stda		%f4, [$out + $mask]0xc0	! partial store
+	std		%f6, [$out + 8]
+	add		$out, 16, $out
+	orn		%g0, $mask, $mask
+	retl
+	stda		%f8, [$out + $mask]0xc0	! partial store
+.type	aes_t4_encrypt,#function
+.size	aes_t4_encrypt,.-aes_t4_encrypt
+
+.globl	aes_t4_decrypt
+.align	32
+aes_t4_decrypt:
+	andcc		$inp, 7, %g1		! is input aligned?
+	andn		$inp, 7, $inp
+
+	ldx		[$key + 0], %g4
+	ldx		[$key + 8], %g5
+
+	ldx		[$inp + 0], %o4
+	bz,pt		%icc, 1f
+	ldx		[$inp + 8], %o5
+	ldx		[$inp + 16], $inp
+	sll		%g1, 3, %g1
+	sub		%g0, %g1, %o3
+	sllx		%o4, %g1, %o4
+	sllx		%o5, %g1, %g1
+	srlx		%o5, %o3, %o5
+	srlx		$inp, %o3, %o3
+	or		%o5, %o4, %o4
+	or		%o3, %g1, %o5
+1:
+	ld		[$key + 240], $rounds
+	ldd		[$key + 16], %f12
+	ldd		[$key + 24], %f14
+	xor		%g4, %o4, %o4
+	xor		%g5, %o5, %o5
+	movxtod		%o4, %f0
+	movxtod		%o5, %f2
+	srl		$rounds, 1, $rounds
+	ldd		[$key + 32], %f16
+	sub		$rounds, 1, $rounds
+	ldd		[$key + 40], %f18
+	add		$key, 48, $key
+
+.Ldec:
+	aes_dround01	%f12, %f0, %f2, %f4
+	aes_dround23	%f14, %f0, %f2, %f2
+	ldd		[$key + 0], %f12
+	ldd		[$key + 8], %f14
+	sub		$rounds,1,$rounds
+	aes_dround01	%f16, %f4, %f2, %f0
+	aes_dround23	%f18, %f4, %f2, %f2
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	brnz,pt		$rounds, .Ldec
+	add		$key, 32, $key
+
+	andcc		$out, 7, $tmp		! is output aligned?
+	aes_dround01	%f12, %f0, %f2, %f4
+	aes_dround23	%f14, %f0, %f2, %f2
+	aes_dround01_l	%f16, %f4, %f2, %f0
+	aes_dround23_l	%f18, %f4, %f2, %f2
+
+	bnz,pn		%icc, 2f
+	nop
+
+	std		%f0, [$out + 0]
+	retl
+	std		%f2, [$out + 8]
+
+2:	alignaddrl	$out, %g0, $out
+	mov		0xff, $mask
+	srl		$mask, $tmp, $mask
+
+	faligndata	%f0, %f0, %f4
+	faligndata	%f0, %f2, %f6
+	faligndata	%f2, %f2, %f8
+
+	stda		%f4, [$out + $mask]0xc0	! partial store
+	std		%f6, [$out + 8]
+	add		$out, 16, $out
+	orn		%g0, $mask, $mask
+	retl
+	stda		%f8, [$out + $mask]0xc0	! partial store
+.type	aes_t4_decrypt,#function
+.size	aes_t4_decrypt,.-aes_t4_decrypt
+___
+}
+
+######################################################################
+# key setup subroutines
+#
+{
+my ($inp,$bits,$out,$tmp)=map("%o$_",(0..5));
+$code.=<<___;
+.globl	aes_t4_set_encrypt_key
+.align	32
+aes_t4_set_encrypt_key:
+.Lset_encrypt_key:
+	and		$inp, 7, $tmp
+	alignaddr	$inp, %g0, $inp
+	cmp		$bits, 192
+	ldd		[$inp + 0], %f0
+	bl,pt		%icc,.L128
+	ldd		[$inp + 8], %f2
+
+	be,pt		%icc,.L192
+	ldd		[$inp + 16], %f4
+	brz,pt		$tmp, .L256aligned
+	ldd		[$inp + 24], %f6
+
+	ldd		[$inp + 32], %f8
+	faligndata	%f0, %f2, %f0
+	faligndata	%f2, %f4, %f2
+	faligndata	%f4, %f6, %f4
+	faligndata	%f6, %f8, %f6
+.L256aligned:
+___
+for ($i=0; $i<6; $i++) {
+    $code.=<<___;
+	std		%f0, [$out + `32*$i+0`]
+	aes_kexpand1	%f0, %f6, $i, %f0
+	std		%f2, [$out + `32*$i+8`]
+	aes_kexpand2	%f2, %f0, %f2
+	std		%f4, [$out + `32*$i+16`]
+	aes_kexpand0	%f4, %f2, %f4
+	std		%f6, [$out + `32*$i+24`]
+	aes_kexpand2	%f6, %f4, %f6
+___
+}
+$code.=<<___;
+	std		%f0, [$out + `32*$i+0`]
+	aes_kexpand1	%f0, %f6, $i, %f0
+	std		%f2, [$out + `32*$i+8`]
+	aes_kexpand2	%f2, %f0, %f2
+	std		%f4, [$out + `32*$i+16`]
+	std		%f6, [$out + `32*$i+24`]
+	std		%f0, [$out + `32*$i+32`]
+	std		%f2, [$out + `32*$i+40`]
+
+	mov		14, $tmp
+	st		$tmp, [$out + 240]
+	retl
+	xor		%o0, %o0, %o0
+
+.align	16
+.L192:
+	brz,pt		$tmp, .L192aligned
+	nop
+
+	ldd		[$inp + 24], %f6
+	faligndata	%f0, %f2, %f0
+	faligndata	%f2, %f4, %f2
+	faligndata	%f4, %f6, %f4
+.L192aligned:
+___
+for ($i=0; $i<7; $i++) {
+    $code.=<<___;
+	std		%f0, [$out + `24*$i+0`]
+	aes_kexpand1	%f0, %f4, $i, %f0
+	std		%f2, [$out + `24*$i+8`]
+	aes_kexpand2	%f2, %f0, %f2
+	std		%f4, [$out + `24*$i+16`]
+	aes_kexpand2	%f4, %f2, %f4
+___
+}
+$code.=<<___;
+	std		%f0, [$out + `24*$i+0`]
+	aes_kexpand1	%f0, %f4, $i, %f0
+	std		%f2, [$out + `24*$i+8`]
+	aes_kexpand2	%f2, %f0, %f2
+	std		%f4, [$out + `24*$i+16`]
+	std		%f0, [$out + `24*$i+24`]
+	std		%f2, [$out + `24*$i+32`]
+
+	mov		12, $tmp
+	st		$tmp, [$out + 240]
+	retl
+	xor		%o0, %o0, %o0
+
+.align	16
+.L128:
+	brz,pt		$tmp, .L128aligned
+	nop
+
+	ldd		[$inp + 16], %f4
+	faligndata	%f0, %f2, %f0
+	faligndata	%f2, %f4, %f2
+.L128aligned:
+___
+for ($i=0; $i<10; $i++) {
+    $code.=<<___;
+	std		%f0, [$out + `16*$i+0`]
+	aes_kexpand1	%f0, %f2, $i, %f0
+	std		%f2, [$out + `16*$i+8`]
+	aes_kexpand2	%f2, %f0, %f2
+___
+}
+$code.=<<___;
+	std		%f0, [$out + `16*$i+0`]
+	std		%f2, [$out + `16*$i+8`]
+
+	mov		10, $tmp
+	st		$tmp, [$out + 240]
+	retl
+	xor		%o0, %o0, %o0
+.type	aes_t4_set_encrypt_key,#function
+.size	aes_t4_set_encrypt_key,.-aes_t4_set_encrypt_key
+
+.globl	aes_t4_set_decrypt_key
+.align	32
+aes_t4_set_decrypt_key:
+	mov		%o7, %o5
+	call		.Lset_encrypt_key
+	nop
+
+	mov		%o5, %o7
+	sll		$tmp, 4, $inp		! $tmp is number of rounds
+	add		$tmp, 2, $tmp
+	add		$out, $inp, $inp	! $inp=$out+16*rounds
+	srl		$tmp, 2, $tmp		! $tmp=(rounds+2)/4
+
+.Lkey_flip:
+	ldd		[$out + 0],  %f0
+	ldd		[$out + 8],  %f2
+	ldd		[$out + 16], %f4
+	ldd		[$out + 24], %f6
+	ldd		[$inp + 0],  %f8
+	ldd		[$inp + 8],  %f10
+	ldd		[$inp - 16], %f12
+	ldd		[$inp - 8],  %f14
+	sub		$tmp, 1, $tmp
+	std		%f0, [$inp + 0]
+	std		%f2, [$inp + 8]
+	std		%f4, [$inp - 16]
+	std		%f6, [$inp - 8]
+	std		%f8, [$out + 0]
+	std		%f10, [$out + 8]
+	std		%f12, [$out + 16]
+	std		%f14, [$out + 24]
+	add		$out, 32, $out
+	brnz		$tmp, .Lkey_flip
+	sub		$inp, 32, $inp
+
+	retl
+	xor		%o0, %o0, %o0
+.type	aes_t4_set_decrypt_key,#function
+.size	aes_t4_set_decrypt_key,.-aes_t4_set_decrypt_key
+___
+}
+
+{{{
+my ($inp,$out,$len,$key,$ivec,$enc)=map("%i$_",(0..5));
+my ($ileft,$iright,$ooff,$omask,$ivoff)=map("%l$_",(1..7));
+
+$code.=<<___;
+.align	32
+_aes128_encrypt_1x:
+___
+for ($i=0; $i<4; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_eround01	%f48, %f0, %f2, %f4
+	aes_eround23	%f50, %f0, %f2, %f2
+	aes_eround01_l	%f52, %f4, %f2, %f0
+	retl
+	aes_eround23_l	%f54, %f4, %f2, %f2
+.type	_aes128_encrypt_1x,#function
+.size	_aes128_encrypt_1x,.-_aes128_encrypt_1x
+
+.align	32
+_aes128_encrypt_2x:
+___
+for ($i=0; $i<4; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_eround01	%f48, %f0, %f2, %f8
+	aes_eround23	%f50, %f0, %f2, %f2
+	aes_eround01	%f48, %f4, %f6, %f10
+	aes_eround23	%f50, %f4, %f6, %f6
+	aes_eround01_l	%f52, %f8, %f2, %f0
+	aes_eround23_l	%f54, %f8, %f2, %f2
+	aes_eround01_l	%f52, %f10, %f6, %f4
+	retl
+	aes_eround23_l	%f54, %f10, %f6, %f6
+.type	_aes128_encrypt_2x,#function
+.size	_aes128_encrypt_2x,.-_aes128_encrypt_2x
+
+.align	32
+_aes128_loadkey:
+	ldx		[$key + 0], %g4
+	ldx		[$key + 8], %g5
+___
+for ($i=2; $i<22;$i++) {			# load key schedule
+    $code.=<<___;
+	ldd		[$key + `8*$i`], %f`12+2*$i`
+___
+}
+$code.=<<___;
+	retl
+	nop
+.type	_aes128_loadkey,#function
+.size	_aes128_loadkey,.-_aes128_loadkey
+_aes128_load_enckey=_aes128_loadkey
+_aes128_load_deckey=_aes128_loadkey
+
+___
+
+&alg_cbc_encrypt_implement("aes",128);
+if ($::evp) {
+    &alg_ctr32_implement("aes",128);
+    &alg_xts_implement("aes",128,"en");
+    &alg_xts_implement("aes",128,"de");
+}
+&alg_cbc_decrypt_implement("aes",128);
+
+$code.=<<___;
+.align	32
+_aes128_decrypt_1x:
+___
+for ($i=0; $i<4; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_dround01	%f48, %f0, %f2, %f4
+	aes_dround23	%f50, %f0, %f2, %f2
+	aes_dround01_l	%f52, %f4, %f2, %f0
+	retl
+	aes_dround23_l	%f54, %f4, %f2, %f2
+.type	_aes128_decrypt_1x,#function
+.size	_aes128_decrypt_1x,.-_aes128_decrypt_1x
+
+.align	32
+_aes128_decrypt_2x:
+___
+for ($i=0; $i<4; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_dround01	%f48, %f0, %f2, %f8
+	aes_dround23	%f50, %f0, %f2, %f2
+	aes_dround01	%f48, %f4, %f6, %f10
+	aes_dround23	%f50, %f4, %f6, %f6
+	aes_dround01_l	%f52, %f8, %f2, %f0
+	aes_dround23_l	%f54, %f8, %f2, %f2
+	aes_dround01_l	%f52, %f10, %f6, %f4
+	retl
+	aes_dround23_l	%f54, %f10, %f6, %f6
+.type	_aes128_decrypt_2x,#function
+.size	_aes128_decrypt_2x,.-_aes128_decrypt_2x
+___
+
+$code.=<<___;
+.align	32
+_aes192_encrypt_1x:
+___
+for ($i=0; $i<5; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_eround01	%f56, %f0, %f2, %f4
+	aes_eround23	%f58, %f0, %f2, %f2
+	aes_eround01_l	%f60, %f4, %f2, %f0
+	retl
+	aes_eround23_l	%f62, %f4, %f2, %f2
+.type	_aes192_encrypt_1x,#function
+.size	_aes192_encrypt_1x,.-_aes192_encrypt_1x
+
+.align	32
+_aes192_encrypt_2x:
+___
+for ($i=0; $i<5; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_eround01	%f56, %f0, %f2, %f8
+	aes_eround23	%f58, %f0, %f2, %f2
+	aes_eround01	%f56, %f4, %f6, %f10
+	aes_eround23	%f58, %f4, %f6, %f6
+	aes_eround01_l	%f60, %f8, %f2, %f0
+	aes_eround23_l	%f62, %f8, %f2, %f2
+	aes_eround01_l	%f60, %f10, %f6, %f4
+	retl
+	aes_eround23_l	%f62, %f10, %f6, %f6
+.type	_aes192_encrypt_2x,#function
+.size	_aes192_encrypt_2x,.-_aes192_encrypt_2x
+
+.align	32
+_aes256_encrypt_1x:
+	aes_eround01	%f16, %f0, %f2, %f4
+	aes_eround23	%f18, %f0, %f2, %f2
+	ldd		[$key + 208], %f16
+	ldd		[$key + 216], %f18
+	aes_eround01	%f20, %f4, %f2, %f0
+	aes_eround23	%f22, %f4, %f2, %f2
+	ldd		[$key + 224], %f20
+	ldd		[$key + 232], %f22
+___
+for ($i=1; $i<6; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_eround01	%f16, %f0, %f2, %f4
+	aes_eround23	%f18, %f0, %f2, %f2
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	aes_eround01_l	%f20, %f4, %f2, %f0
+	aes_eround23_l	%f22, %f4, %f2, %f2
+	ldd		[$key + 32], %f20
+	retl
+	ldd		[$key + 40], %f22
+.type	_aes256_encrypt_1x,#function
+.size	_aes256_encrypt_1x,.-_aes256_encrypt_1x
+
+.align	32
+_aes256_encrypt_2x:
+	aes_eround01	%f16, %f0, %f2, %f8
+	aes_eround23	%f18, %f0, %f2, %f2
+	aes_eround01	%f16, %f4, %f6, %f10
+	aes_eround23	%f18, %f4, %f6, %f6
+	ldd		[$key + 208], %f16
+	ldd		[$key + 216], %f18
+	aes_eround01	%f20, %f8, %f2, %f0
+	aes_eround23	%f22, %f8, %f2, %f2
+	aes_eround01	%f20, %f10, %f6, %f4
+	aes_eround23	%f22, %f10, %f6, %f6
+	ldd		[$key + 224], %f20
+	ldd		[$key + 232], %f22
+___
+for ($i=1; $i<6; $i++) {
+    $code.=<<___;
+	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_eround01	%f16, %f0, %f2, %f8
+	aes_eround23	%f18, %f0, %f2, %f2
+	aes_eround01	%f16, %f4, %f6, %f10
+	aes_eround23	%f18, %f4, %f6, %f6
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	aes_eround01_l	%f20, %f8, %f2, %f0
+	aes_eround23_l	%f22, %f8, %f2, %f2
+	aes_eround01_l	%f20, %f10, %f6, %f4
+	aes_eround23_l	%f22, %f10, %f6, %f6
+	ldd		[$key + 32], %f20
+	retl
+	ldd		[$key + 40], %f22
+.type	_aes256_encrypt_2x,#function
+.size	_aes256_encrypt_2x,.-_aes256_encrypt_2x
+
+.align	32
+_aes192_loadkey:
+	ldx		[$key + 0], %g4
+	ldx		[$key + 8], %g5
+___
+for ($i=2; $i<26;$i++) {			# load key schedule
+    $code.=<<___;
+	ldd		[$key + `8*$i`], %f`12+2*$i`
+___
+}
+$code.=<<___;
+	retl
+	nop
+.type	_aes192_loadkey,#function
+.size	_aes192_loadkey,.-_aes192_loadkey
+_aes256_loadkey=_aes192_loadkey
+_aes192_load_enckey=_aes192_loadkey
+_aes192_load_deckey=_aes192_loadkey
+_aes256_load_enckey=_aes192_loadkey
+_aes256_load_deckey=_aes192_loadkey
+___
+
+&alg_cbc_encrypt_implement("aes",256);
+&alg_cbc_encrypt_implement("aes",192);
+if ($::evp) {
+    &alg_ctr32_implement("aes",256);
+    &alg_xts_implement("aes",256,"en");
+    &alg_xts_implement("aes",256,"de");
+    &alg_ctr32_implement("aes",192);
+}
+&alg_cbc_decrypt_implement("aes",192);
+&alg_cbc_decrypt_implement("aes",256);
+
+$code.=<<___;
+.align	32
+_aes256_decrypt_1x:
+	aes_dround01	%f16, %f0, %f2, %f4
+	aes_dround23	%f18, %f0, %f2, %f2
+	ldd		[$key + 208], %f16
+	ldd		[$key + 216], %f18
+	aes_dround01	%f20, %f4, %f2, %f0
+	aes_dround23	%f22, %f4, %f2, %f2
+	ldd		[$key + 224], %f20
+	ldd		[$key + 232], %f22
+___
+for ($i=1; $i<6; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_dround01	%f16, %f0, %f2, %f4
+	aes_dround23	%f18, %f0, %f2, %f2
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	aes_dround01_l	%f20, %f4, %f2, %f0
+	aes_dround23_l	%f22, %f4, %f2, %f2
+	ldd		[$key + 32], %f20
+	retl
+	ldd		[$key + 40], %f22
+.type	_aes256_decrypt_1x,#function
+.size	_aes256_decrypt_1x,.-_aes256_decrypt_1x
+
+.align	32
+_aes256_decrypt_2x:
+	aes_dround01	%f16, %f0, %f2, %f8
+	aes_dround23	%f18, %f0, %f2, %f2
+	aes_dround01	%f16, %f4, %f6, %f10
+	aes_dround23	%f18, %f4, %f6, %f6
+	ldd		[$key + 208], %f16
+	ldd		[$key + 216], %f18
+	aes_dround01	%f20, %f8, %f2, %f0
+	aes_dround23	%f22, %f8, %f2, %f2
+	aes_dround01	%f20, %f10, %f6, %f4
+	aes_dround23	%f22, %f10, %f6, %f6
+	ldd		[$key + 224], %f20
+	ldd		[$key + 232], %f22
+___
+for ($i=1; $i<6; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_dround01	%f16, %f0, %f2, %f8
+	aes_dround23	%f18, %f0, %f2, %f2
+	aes_dround01	%f16, %f4, %f6, %f10
+	aes_dround23	%f18, %f4, %f6, %f6
+	ldd		[$key + 16], %f16
+	ldd		[$key + 24], %f18
+	aes_dround01_l	%f20, %f8, %f2, %f0
+	aes_dround23_l	%f22, %f8, %f2, %f2
+	aes_dround01_l	%f20, %f10, %f6, %f4
+	aes_dround23_l	%f22, %f10, %f6, %f6
+	ldd		[$key + 32], %f20
+	retl
+	ldd		[$key + 40], %f22
+.type	_aes256_decrypt_2x,#function
+.size	_aes256_decrypt_2x,.-_aes256_decrypt_2x
+
+.align	32
+_aes192_decrypt_1x:
+___
+for ($i=0; $i<5; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
+___
+}
+$code.=<<___;
+	aes_dround01	%f56, %f0, %f2, %f4
+	aes_dround23	%f58, %f0, %f2, %f2
+	aes_dround01_l	%f60, %f4, %f2, %f0
+	retl
+	aes_dround23_l	%f62, %f4, %f2, %f2
+.type	_aes192_decrypt_1x,#function
+.size	_aes192_decrypt_1x,.-_aes192_decrypt_1x
+
+.align	32
+_aes192_decrypt_2x:
+___
+for ($i=0; $i<5; $i++) {
+    $code.=<<___;
+	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
+	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
+	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
+	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
+	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
+	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
+	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
+	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
+___
+}
+$code.=<<___;
+	aes_dround01	%f56, %f0, %f2, %f8
+	aes_dround23	%f58, %f0, %f2, %f2
+	aes_dround01	%f56, %f4, %f6, %f10
+	aes_dround23	%f58, %f4, %f6, %f6
+	aes_dround01_l	%f60, %f8, %f2, %f0
+	aes_dround23_l	%f62, %f8, %f2, %f2
+	aes_dround01_l	%f60, %f10, %f6, %f4
+	retl
+	aes_dround23_l	%f62, %f10, %f6, %f6
+.type	_aes192_decrypt_2x,#function
+.size	_aes192_decrypt_2x,.-_aes192_decrypt_2x
+___
+}}}
+
+if (!$::evp) {
+$code.=<<___;
+.global	AES_encrypt
+AES_encrypt=aes_t4_encrypt
+.global	AES_decrypt
+AES_decrypt=aes_t4_decrypt
+.global	AES_set_encrypt_key
+.align	32
+AES_set_encrypt_key:
+	andcc		%o2, 7, %g0		! check alignment
+	bnz,a,pn	%icc, 1f
+	mov		-1, %o0
+	brz,a,pn	%o0, 1f
+	mov		-1, %o0
+	brz,a,pn	%o2, 1f
+	mov		-1, %o0
+	andncc		%o1, 0x1c0, %g0
+	bnz,a,pn	%icc, 1f
+	mov		-2, %o0
+	cmp		%o1, 128
+	bl,a,pn		%icc, 1f
+	mov		-2, %o0
+	b		aes_t4_set_encrypt_key
+	nop
+1:	retl
+	nop
+.type	AES_set_encrypt_key,#function
+.size	AES_set_encrypt_key,.-AES_set_encrypt_key
+
+.global	AES_set_decrypt_key
+.align	32
+AES_set_decrypt_key:
+	andcc		%o2, 7, %g0		! check alignment
+	bnz,a,pn	%icc, 1f
+	mov		-1, %o0
+	brz,a,pn	%o0, 1f
+	mov		-1, %o0
+	brz,a,pn	%o2, 1f
+	mov		-1, %o0
+	andncc		%o1, 0x1c0, %g0
+	bnz,a,pn	%icc, 1f
+	mov		-2, %o0
+	cmp		%o1, 128
+	bl,a,pn		%icc, 1f
+	mov		-2, %o0
+	b		aes_t4_set_decrypt_key
+	nop
+1:	retl
+	nop
+.type	AES_set_decrypt_key,#function
+.size	AES_set_decrypt_key,.-AES_set_decrypt_key
+___
+
+my ($inp,$out,$len,$key,$ivec,$enc)=map("%o$_",(0..5));
+
+$code.=<<___;
+.globl	AES_cbc_encrypt
+.align	32
+AES_cbc_encrypt:
+	ld		[$key + 240], %g1
+	nop
+	brz		$enc, .Lcbc_decrypt
+	cmp		%g1, 12
+
+	bl,pt		%icc, aes128_t4_cbc_encrypt
+	nop
+	be,pn		%icc, aes192_t4_cbc_encrypt
+	nop
+	ba		aes256_t4_cbc_encrypt
+	nop
+
+.Lcbc_decrypt:
+	bl,pt		%icc, aes128_t4_cbc_decrypt
+	nop
+	be,pn		%icc, aes192_t4_cbc_decrypt
+	nop
+	ba		aes256_t4_cbc_decrypt
+	nop
+.type	AES_cbc_encrypt,#function
+.size	AES_cbc_encrypt,.-AES_cbc_encrypt
+___
+}
+$code.=<<___;
+.asciz	"AES for SPARC T4, David S. Miller, Andy Polyakov"
+.align	4
+___
+
+&emit_assembler();
+
+close STDOUT;
diff --git a/openssl/crypto/aes/asm/aesv8-armx.pl b/openssl/crypto/aes/asm/aesv8-armx.pl
new file mode 100755
index 000000000..1e93f8685
--- /dev/null
+++ b/openssl/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/openssl/crypto/aes/asm/bsaes-armv7.pl b/openssl/crypto/aes/asm/bsaes-armv7.pl
new file mode 100755
index 000000000..fcc81d1a4
--- /dev/null
+++ b/openssl/crypto/aes/asm/bsaes-armv7.pl
@@ -0,0 +1,2469 @@
+#!/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.=<<___;
+#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__
+# define __ARM_MAX_ARCH__ __LINUX_ARM_ARCH__
+#endif
+
+#ifdef __thumb__
+# define adrl adr
+#endif
+
+#if __ARM_MAX_ARCH__>=7
+.arch	armv7-a
+.fpu	neon
+
+.text
+.syntax	unified 	@ ARMv7-capable assembler is expected to handle this
+#ifdef __thumb2__
+.thumb
+#else
+.code   32
+#endif
+
+.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
+.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,$out			@ pass key schedule
+	bl	_bsaes_key_convert
+	veor	@XMM[7],@XMM[7],@XMM[15]	@ fix up last round key
+	vstmia	r12, {@XMM[7]}			@ save last round key
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_enc_key_convert,.-bsaes_enc_key_convert
+
+.globl	bsaes_encrypt_128
+.type	bsaes_encrypt_128,%function
+.align	4
+bsaes_encrypt_128:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+.Lenc128_loop:
+	vld1.8	{@XMM[0]-@XMM[1]}, [$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]}, [$out]!
+	bhi	.Lenc128_loop
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_encrypt_128,.-bsaes_encrypt_128
+
+.globl	bsaes_dec_key_convert
+.type	bsaes_dec_key_convert,%function
+.align	4
+bsaes_dec_key_convert:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+
+	ldr	r5,[$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	$out, {@XMM[7]}
+
+	vldmia	sp!,{d8-d15}
+	ldmia	sp!,{r4-r6,pc}
+.size	bsaes_dec_key_convert,.-bsaes_dec_key_convert
+
+.globl	bsaes_decrypt_128
+.type	bsaes_decrypt_128,%function
+.align	4
+bsaes_decrypt_128:
+	stmdb	sp!,{r4-r6,lr}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+.Ldec128_loop:
+	vld1.8	{@XMM[0]-@XMM[1]}, [$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
+.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
+.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
+.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
+.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
+___
+
+$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/openssl/crypto/aes/asm/bsaes-x86_64.pl b/openssl/crypto/aes/asm/bsaes-x86_64.pl
index 41b90f084..3f7d33c45 100644
--- a/openssl/crypto/aes/asm/bsaes-x86_64.pl
+++ b/openssl/crypto/aes/asm/bsaes-x86_64.pl
@@ -38,8 +38,9 @@
 #		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
 #
 # (*)	Comparison is not completely fair, because "this" is ECB,
 #	i.e. no extra processing such as counter values calculation
@@ -50,14 +51,6 @@
 # (**)	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
@@ -67,7 +60,7 @@
 # 		conversion	conversion/8x block
 # Core 2	240		0.22
 # Nehalem	180		0.20
-# Atom		430		0.19
+# 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%,
@@ -83,9 +76,10 @@
 # Add decryption procedure. Performance in CPU cycles spent to decrypt
 # one byte out of 4096-byte buffer with 128-bit key is:
 #
-# Core 2	9.83
-# Nehalem	7.74
-# Atom		19.0
+# Core 2	9.98
+# Nehalem	7.80
+# Atom		17.9
+# Silvermont	14.0
 #
 # November 2011.
 #
@@ -434,21 +428,21 @@ 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
 ___
 }
 
@@ -820,18 +814,18 @@ _bsaes_encrypt8:
 	movdqa	0x50($const), @XMM[8]	# .LM0SR
 	pxor	@XMM[9], @XMM[0]	# xor with round0 key
 	pxor	@XMM[9], @XMM[1]
-	 pshufb	@XMM[8], @XMM[0]
 	pxor	@XMM[9], @XMM[2]
-	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[3]
-	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[0]
+	 pshufb	@XMM[8], @XMM[1]
 	pxor	@XMM[9], @XMM[4]
-	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[5]
-	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[2]
+	 pshufb	@XMM[8], @XMM[3]
 	pxor	@XMM[9], @XMM[6]
-	 pshufb	@XMM[8], @XMM[5]
 	pxor	@XMM[9], @XMM[7]
+	 pshufb	@XMM[8], @XMM[4]
+	 pshufb	@XMM[8], @XMM[5]
 	 pshufb	@XMM[8], @XMM[6]
 	 pshufb	@XMM[8], @XMM[7]
 _bsaes_encrypt8_bitslice:
@@ -884,18 +878,18 @@ _bsaes_decrypt8:
 	movdqa	-0x30($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]
 ___
@@ -1937,21 +1931,21 @@ $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
diff --git a/openssl/crypto/aes/asm/vpaes-ppc.pl b/openssl/crypto/aes/asm/vpaes-ppc.pl
new file mode 100755
index 000000000..7fda60ed9
--- /dev/null
+++ b/openssl/crypto/aes/asm/vpaes-ppc.pl
@@ -0,0 +1,1512 @@
+#!/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/.
+
+# CBC encrypt/decrypt performance in cycles per byte processed with
+# 128-bit key.
+#
+#		aes-ppc.pl		this
+# G4e		35.5/52.1/(23.8)	11.9(*)/15.4
+# POWER6	42.7/54.3/(28.2)	63.0/92.8(**)
+# POWER7	32.3/42.9/(18.4)	18.5/23.3
+#
+# (*)	This is ~10% worse than reported in paper. The reason is
+#	twofold. This module doesn't make any assumption about
+#	key schedule (or data for that matter) alignment and handles
+#	it in-line. Secondly it, being transliterated from
+#	vpaes-x86_64.pl, relies on "nested inversion" better suited
+#	for Intel CPUs.
+# (**)	Inadequate POWER6 performance is due to astronomic AltiVec
+#	latency, 9 cycles per simple logical operation.
+
+$flavour = shift;
+
+if ($flavour =~ /64/) {
+	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
+	$STU	="stdu";
+	$POP	="ld";
+	$PUSH	="std";
+	$UCMP	="cmpld";
+} elsif ($flavour =~ /32/) {
+	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
+	$STU	="stwu";
+	$POP	="lwz";
+	$PUSH	="stw";
+	$UCMP	="cmplw";
+} else { die "nonsense $flavour"; }
+
+$sp="r1";
+$FRAME=6*$SIZE_T+13*16;	# 13*16 is for v20-v31 offload
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
+
+$code.=<<___;
+.machine	"any"
+
+.text
+
+.align	7	# totally strategic alignment
+_vpaes_consts:
+Lk_mc_forward:	# mc_forward
+	.long	0x01020300, 0x05060704, 0x090a0b08, 0x0d0e0f0c	?inv
+	.long	0x05060704, 0x090a0b08, 0x0d0e0f0c, 0x01020300	?inv
+	.long	0x090a0b08, 0x0d0e0f0c, 0x01020300, 0x05060704	?inv
+	.long	0x0d0e0f0c, 0x01020300, 0x05060704, 0x090a0b08	?inv
+Lk_mc_backward:	# mc_backward
+	.long	0x03000102, 0x07040506, 0x0b08090a, 0x0f0c0d0e	?inv
+	.long	0x0f0c0d0e, 0x03000102, 0x07040506, 0x0b08090a	?inv
+	.long	0x0b08090a, 0x0f0c0d0e, 0x03000102, 0x07040506	?inv
+	.long	0x07040506, 0x0b08090a, 0x0f0c0d0e, 0x03000102	?inv
+Lk_sr:		# sr
+	.long	0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f	?inv
+	.long	0x00050a0f, 0x04090e03, 0x080d0207, 0x0c01060b	?inv
+	.long	0x0009020b, 0x040d060f, 0x08010a03, 0x0c050e07	?inv
+	.long	0x000d0a07, 0x04010e0b, 0x0805020f, 0x0c090603	?inv
+
+##
+## "Hot" constants
+##
+Lk_inv:		# inv, inva
+	.long	0xf001080d, 0x0f06050e, 0x020c0b0a, 0x09030704	?rev
+	.long	0xf0070b0f, 0x060a0401, 0x09080502, 0x0c0e0d03	?rev
+Lk_ipt:		# input transform (lo, hi)
+	.long	0x00702a5a, 0x98e8b2c2, 0x08782252, 0x90e0baca	?rev
+	.long	0x004d7c31, 0x7d30014c, 0x81ccfdb0, 0xfcb180cd	?rev
+Lk_sbo:		# sbou, sbot
+	.long	0x00c7bd6f, 0x176dd2d0, 0x78a802c5, 0x7abfaa15	?rev
+	.long	0x006abb5f, 0xa574e4cf, 0xfa352b41, 0xd1901e8e	?rev
+Lk_sb1:		# sb1u, sb1t
+	.long	0x0023e2fa, 0x15d41836, 0xefd92e0d, 0xc1ccf73b	?rev
+	.long	0x003e50cb, 0x8fe19bb1, 0x44f52a14, 0x6e7adfa5	?rev
+Lk_sb2:		# sb2u, sb2t
+	.long	0x0029e10a, 0x4088eb69, 0x4a2382ab, 0xc863a1c2	?rev
+	.long	0x0024710b, 0xc6937ae2, 0xcd2f98bc, 0x55e9b75e	?rev
+
+##
+##  Decryption stuff
+##
+Lk_dipt:	# decryption input transform
+	.long	0x005f540b, 0x045b500f, 0x1a454e11, 0x1e414a15	?rev
+	.long	0x00650560, 0xe683e386, 0x94f191f4, 0x72177712	?rev
+Lk_dsbo:	# decryption sbox final output
+	.long	0x0040f97e, 0x53ea8713, 0x2d3e94d4, 0xb96daac7	?rev
+	.long	0x001d4493, 0x0f56d712, 0x9c8ec5d8, 0x59814bca	?rev
+Lk_dsb9:	# decryption sbox output *9*u, *9*t
+	.long	0x00d6869a, 0x53031c85, 0xc94c994f, 0x501fd5ca	?rev
+	.long	0x0049d7ec, 0x89173bc0, 0x65a5fbb2, 0x9e2c5e72	?rev
+Lk_dsbd:	# decryption sbox output *D*u, *D*t
+	.long	0x00a2b1e6, 0xdfcc577d, 0x39442a88, 0x139b6ef5	?rev
+	.long	0x00cbc624, 0xf7fae23c, 0xd3efde15, 0x0d183129	?rev
+Lk_dsbb:	# decryption sbox output *B*u, *B*t
+	.long	0x0042b496, 0x926422d0, 0x04d4f2b0, 0xf6462660	?rev
+	.long	0x006759cd, 0xa69894c1, 0x6baa5532, 0x3e0cfff3	?rev
+Lk_dsbe:	# decryption sbox output *E*u, *E*t
+	.long	0x00d0d426, 0x9692f246, 0xb0f6b464, 0x04604222	?rev
+	.long	0x00c1aaff, 0xcda6550c, 0x323e5998, 0x6bf36794	?rev
+
+##
+##  Key schedule constants
+##
+Lk_dksd:	# decryption key schedule: invskew x*D
+	.long	0x0047e4a3, 0x5d1ab9fe, 0xf9be1d5a, 0xa4e34007	?rev
+	.long	0x008336b5, 0xf477c241, 0x1e9d28ab, 0xea69dc5f	?rev
+Lk_dksb:	# decryption key schedule: invskew x*B
+	.long	0x00d55085, 0x1fca4f9a, 0x994cc91c, 0x8653d603	?rev
+	.long	0x004afcb6, 0xa7ed5b11, 0xc882347e, 0x6f2593d9	?rev
+Lk_dkse:	# decryption key schedule: invskew x*E + 0x63
+	.long	0x00d6c91f, 0xca1c03d5, 0x86504f99, 0x4c9a8553	?rev
+	.long	0xe87bdc4f, 0x059631a2, 0x8714b320, 0x6af95ecd	?rev
+Lk_dks9:	# decryption key schedule: invskew x*9
+	.long	0x00a7d97e, 0xc86f11b6, 0xfc5b2582, 0x3493ed4a	?rev
+	.long	0x00331427, 0x62517645, 0xcefddae9, 0xac9fb88b	?rev
+
+Lk_rcon:	# rcon
+	.long	0xb6ee9daf, 0xb991831f, 0x817d7c4d, 0x08982a70	?asis
+Lk_s63:
+	.long	0x5b5b5b5b, 0x5b5b5b5b, 0x5b5b5b5b, 0x5b5b5b5b	?asis
+
+Lk_opt:		# output transform
+	.long	0x0060b6d6, 0x29499fff, 0x0868bede, 0x214197f7	?rev
+	.long	0x00ecbc50, 0x51bded01, 0xe00c5cb0, 0xb15d0de1	?rev
+Lk_deskew:	# deskew tables: inverts the sbox's "skew"
+	.long	0x00e3a447, 0x40a3e407, 0x1af9be5d, 0x5ab9fe1d	?rev
+	.long	0x0069ea83, 0xdcb5365f, 0x771e9df4, 0xabc24128	?rev
+.align	5
+Lconsts:
+	mflr	r0
+	bcl	20,31,\$+4
+	mflr	r12	#vvvvv "distance between . and _vpaes_consts
+	addi	r12,r12,-0x308
+	mtlr	r0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+.asciz  "Vector Permutation AES for AltiVec, Mike Hamburg (Stanford University)"
+.align	6
+___
+
+my ($inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm) = map("v$_",(26..31));
+{
+my ($inp,$out,$key) = map("r$_",(3..5));
+
+my ($invlo,$invhi,$iptlo,$ipthi,$sbou,$sbot) = map("v$_",(10..15));
+my ($sb1u,$sb1t,$sb2u,$sb2t) = map("v$_",(16..19));
+my ($sb9u,$sb9t,$sbdu,$sbdt,$sbbu,$sbbt,$sbeu,$sbet)=map("v$_",(16..23));
+
+$code.=<<___;
+##
+##  _aes_preheat
+##
+##  Fills register %r10 -> .aes_consts (so you can -fPIC)
+##  and %xmm9-%xmm15 as specified below.
+##
+.align	4
+_vpaes_encrypt_preheat:
+	mflr	r8
+	bl	Lconsts
+	mtlr	r8
+	li	r11, 0xc0		# Lk_inv
+	li	r10, 0xd0
+	li	r9,  0xe0		# Lk_ipt
+	li	r8,  0xf0
+	vxor	v7, v7, v7		# 0x00..00
+	vspltisb	v8,4		# 0x04..04
+	vspltisb	v9,0x0f		# 0x0f..0f
+	lvx	$invlo, r12, r11
+	li	r11, 0x100
+	lvx	$invhi, r12, r10
+	li	r10, 0x110
+	lvx	$iptlo, r12, r9
+	li	r9,  0x120
+	lvx	$ipthi, r12, r8
+	li	r8,  0x130
+	lvx	$sbou, r12, r11
+	li	r11, 0x140
+	lvx	$sbot, r12, r10
+	li	r10, 0x150
+	lvx	$sb1u, r12, r9
+	lvx	$sb1t, r12, r8
+	lvx	$sb2u, r12, r11
+	lvx	$sb2t, r12, r10
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  _aes_encrypt_core
+##
+##  AES-encrypt %xmm0.
+##
+##  Inputs:
+##     %xmm0 = input
+##     %xmm9-%xmm15 as in _vpaes_preheat
+##    (%rdx) = scheduled keys
+##
+##  Output in %xmm0
+##  Clobbers  %xmm1-%xmm6, %r9, %r10, %r11, %rax
+##
+##
+.align 5
+_vpaes_encrypt_core:
+	lwz	r8, 240($key)		# pull rounds
+	li	r9, 16
+	lvx	v5, 0, $key		# vmovdqu	(%r9),	%xmm5		# round0 key
+	li	r11, 0x10
+	lvx	v6, r9, $key
+	addi	r9, r9, 16
+	?vperm	v5, v5, v6, $keyperm	# align round key
+	addi	r10, r11, 0x40
+	vsrb	v1, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0
+	vperm	v0, $iptlo, $iptlo, v0	# vpshufb	%xmm1,	%xmm2,	%xmm1
+	vperm	v1, $ipthi, $ipthi, v1	# vpshufb	%xmm0,	%xmm3,	%xmm2
+	vxor	v0, v0, v5		# vpxor	%xmm5,	%xmm1,	%xmm0
+	vxor	v0, v0, v1		# vpxor	%xmm2,	%xmm0,	%xmm0
+	mtctr	r8
+	b	Lenc_entry
+
+.align 4
+Lenc_loop:
+	# middle of middle round
+	vperm	v4, $sb1t, v7, v2	# vpshufb	%xmm2,	%xmm13,	%xmm4	# 4 = sb1u
+	lvx	v1, r12, r11		# vmovdqa	-0x40(%r11,%r10), %xmm1	# .Lk_mc_forward[]
+	addi	r11, r11, 16
+	vperm	v0, $sb1u, v7, v3	# vpshufb	%xmm3,	%xmm12,	%xmm0	# 0 = sb1t
+	vxor	v4, v4, v5		# vpxor		%xmm5,	%xmm4,	%xmm4	# 4 = sb1u + k
+	andi.	r11, r11, 0x30		# and		\$0x30, %r11	# ... mod 4
+	vperm	v5, $sb2t, v7, v2	# vpshufb	%xmm2,	%xmm15,	%xmm5	# 4 = sb2u
+	vxor	v0, v0, v4		# vpxor		%xmm4,	%xmm0,	%xmm0	# 0 = A
+	vperm	v2, $sb2u, v7, v3	# vpshufb	%xmm3,	%xmm14,	%xmm2	# 2 = sb2t
+	lvx	v4, r12, r10		# vmovdqa	(%r11,%r10), %xmm4	# .Lk_mc_backward[]
+	addi	r10, r11, 0x40
+	vperm	v3, v0, v7, v1		# vpshufb	%xmm1,	%xmm0,	%xmm3	# 0 = B
+	vxor	v2, v2, v5		# vpxor		%xmm5,	%xmm2,	%xmm2	# 2 = 2A
+	vperm	v0, v0, v7, v4		# vpshufb	%xmm4,	%xmm0,	%xmm0	# 3 = D
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3	# 0 = 2A+B
+	vperm	v4, v3, v7, v1		# vpshufb	%xmm1,	%xmm3,	%xmm4	# 0 = 2B+C
+	vxor	v0, v0, v3		# vpxor		%xmm3,	%xmm0,	%xmm0	# 3 = 2A+B+D
+	vxor	v0, v0, v4		# vpxor		%xmm4,	%xmm0, %xmm0	# 0 = 2A+3B+C+D
+
+Lenc_entry:
+	# top of round
+	vsrb	v1, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0	# 1 = i
+	vperm	v5, $invhi, $invhi, v0	# vpshufb	%xmm1,	%xmm11,	%xmm5	# 2 = a/k
+	vxor	v0, v0, v1		# vpxor		%xmm0,	%xmm1,	%xmm1	# 0 = j
+	vperm	v3, $invlo, $invlo, v1	# vpshufb	%xmm0, 	%xmm10,	%xmm3  	# 3 = 1/i
+	vperm	v4, $invlo, $invlo, v0	# vpshufb	%xmm1, 	%xmm10,	%xmm4  	# 4 = 1/j
+	vand	v0, v0, v9
+	vxor	v3, v3, v5		# vpxor		%xmm5,	%xmm3,	%xmm3	# 3 = iak = 1/i + a/k
+	vxor	v4, v4, v5		# vpxor		%xmm5,	%xmm4,	%xmm4  	# 4 = jak = 1/j + a/k
+	vperm	v2, $invlo, v7, v3	# vpshufb	%xmm3,	%xmm10,	%xmm2  	# 2 = 1/iak
+	vmr	v5, v6
+	lvx	v6, r9, $key		# vmovdqu	(%r9), %xmm5
+	vperm	v3, $invlo, v7, v4	# vpshufb	%xmm4,	%xmm10,	%xmm3	# 3 = 1/jak
+	addi	r9, r9, 16
+	vxor	v2, v2, v0		# vpxor		%xmm1,	%xmm2,	%xmm2  	# 2 = io
+	?vperm	v5, v5, v6, $keyperm	# align round key
+	vxor	v3, v3, v1		# vpxor		%xmm0,	%xmm3,	%xmm3	# 3 = jo
+	bdnz	Lenc_loop
+
+	# middle of last round
+	addi	r10, r11, 0x80
+					# vmovdqa	-0x60(%r10), %xmm4	# 3 : sbou	.Lk_sbo
+					# vmovdqa	-0x50(%r10), %xmm0	# 0 : sbot	.Lk_sbo+16
+	vperm	v4, $sbou, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4	# 4 = sbou
+	lvx	v1, r12, r10		# vmovdqa	0x40(%r11,%r10), %xmm1	# .Lk_sr[]
+	vperm	v0, $sbot, v7, v3	# vpshufb	%xmm3,	%xmm0,	%xmm0	# 0 = sb1t
+	vxor	v4, v4, v5		# vpxor		%xmm5,	%xmm4,	%xmm4	# 4 = sb1u + k
+	vxor	v0, v0, v4		# vpxor		%xmm4,	%xmm0,	%xmm0	# 0 = A
+	vperm	v0, v0, v7, v1		# vpshufb	%xmm1,	%xmm0,	%xmm0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+.globl	.vpaes_encrypt
+.align	5
+.vpaes_encrypt:
+	$STU	$sp,-$FRAME($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mflr	r6
+	mfspr	r7, 256			# save vrsave
+	stvx	v20,r10,$sp
+	addi	r10,r10,32
+	stvx	v21,r11,$sp
+	addi	r11,r11,32
+	stvx	v22,r10,$sp
+	addi	r10,r10,32
+	stvx	v23,r11,$sp
+	addi	r11,r11,32
+	stvx	v24,r10,$sp
+	addi	r10,r10,32
+	stvx	v25,r11,$sp
+	addi	r11,r11,32
+	stvx	v26,r10,$sp
+	addi	r10,r10,32
+	stvx	v27,r11,$sp
+	addi	r11,r11,32
+	stvx	v28,r10,$sp
+	addi	r10,r10,32
+	stvx	v29,r11,$sp
+	addi	r11,r11,32
+	stvx	v30,r10,$sp
+	stvx	v31,r11,$sp
+	stw	r7,`$FRAME-4`($sp)	# save vrsave
+	li	r0, -1
+	$PUSH	r6,`$FRAME+$LRSAVE`($sp)
+	mtspr	256, r0			# preserve all AltiVec registers
+
+	bl	_vpaes_encrypt_preheat
+
+	?lvsl	$inpperm, 0, $inp	# prepare for unaligned access
+	lvx	v0, 0, $inp
+	addi	$inp, $inp, 15		# 15 is not a typo
+	 ?lvsr	$outperm, 0, $out
+	?lvsl	$keyperm, 0, $key	# prepare for unaligned access
+	 vnor	$outmask, v7, v7	# 0xff..ff
+	lvx	$inptail, 0, $inp	# redundant in aligned case
+	 ?vperm	$outmask, v7, $outmask, $outperm
+	 lvx	$outhead, 0, $out
+	?vperm	v0, v0, $inptail, $inpperm
+
+	bl	_vpaes_encrypt_core
+
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v1, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v1, 0, $out
+	addi	$out, $out, 15		# 15 is not a typo
+	########
+
+	lvx	v1, 0, $out		# redundant in aligned case
+	vsel	v1, $outhead, v1, $outmask
+	stvx	v1, 0, $out
+
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mtlr	r6
+	mtspr	256, r7			# restore vrsave
+	lvx	v20,r10,$sp
+	addi	r10,r10,32
+	lvx	v21,r11,$sp
+	addi	r11,r11,32
+	lvx	v22,r10,$sp
+	addi	r10,r10,32
+	lvx	v23,r11,$sp
+	addi	r11,r11,32
+	lvx	v24,r10,$sp
+	addi	r10,r10,32
+	lvx	v25,r11,$sp
+	addi	r11,r11,32
+	lvx	v26,r10,$sp
+	addi	r10,r10,32
+	lvx	v27,r11,$sp
+	addi	r11,r11,32
+	lvx	v28,r10,$sp
+	addi	r10,r10,32
+	lvx	v29,r11,$sp
+	addi	r11,r11,32
+	lvx	v30,r10,$sp
+	lvx	v31,r11,$sp
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,0x04,1,0x80,0,3,0
+	.long	0
+.size	.vpaes_encrypt,.-.vpaes_encrypt
+
+.align	4
+_vpaes_decrypt_preheat:
+	mflr	r8
+	bl	Lconsts
+	mtlr	r8
+	li	r11, 0xc0		# Lk_inv
+	li	r10, 0xd0
+	li	r9,  0x160		# Ldipt
+	li	r8,  0x170
+	vxor	v7, v7, v7		# 0x00..00
+	vspltisb	v8,4		# 0x04..04
+	vspltisb	v9,0x0f		# 0x0f..0f
+	lvx	$invlo, r12, r11
+	li	r11, 0x180
+	lvx	$invhi, r12, r10
+	li	r10, 0x190
+	lvx	$iptlo, r12, r9
+	li	r9,  0x1a0
+	lvx	$ipthi, r12, r8
+	li	r8,  0x1b0
+	lvx	$sbou, r12, r11
+	li	r11, 0x1c0
+	lvx	$sbot, r12, r10
+	li	r10, 0x1d0
+	lvx	$sb9u, r12, r9
+	li	r9,  0x1e0
+	lvx	$sb9t, r12, r8
+	li	r8,  0x1f0
+	lvx	$sbdu, r12, r11
+	li	r11, 0x200
+	lvx	$sbdt, r12, r10
+	li	r10, 0x210
+	lvx	$sbbu, r12, r9
+	lvx	$sbbt, r12, r8
+	lvx	$sbeu, r12, r11
+	lvx	$sbet, r12, r10
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  Decryption core
+##
+##  Same API as encryption core.
+##
+.align	4
+_vpaes_decrypt_core:
+	lwz	r8, 240($key)		# pull rounds
+	li	r9, 16
+	lvx	v5, 0, $key		# vmovdqu	(%r9),	%xmm4		# round0 key
+	li	r11, 0x30
+	lvx	v6, r9, $key
+	addi	r9, r9, 16
+	?vperm	v5, v5, v6, $keyperm	# align round key
+	vsrb	v1, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0
+	vperm	v0, $iptlo, $iptlo, v0	# vpshufb	%xmm1,	%xmm2,	%xmm2
+	vperm	v1, $ipthi, $ipthi, v1	# vpshufb	%xmm0,	%xmm1,	%xmm0
+	vxor	v0, v0, v5		# vpxor	%xmm4,	%xmm2,	%xmm2
+	vxor	v0, v0, v1		# vpxor	%xmm2,	%xmm0,	%xmm0
+	mtctr	r8
+	b	Ldec_entry
+
+.align 4
+Ldec_loop:
+#
+#  Inverse mix columns
+#
+	lvx	v0, r12, r11		# v5 and v0 are flipped
+					# vmovdqa	-0x20(%r10),%xmm4		# 4 : sb9u
+					# vmovdqa	-0x10(%r10),%xmm1		# 0 : sb9t
+	vperm	v4, $sb9u, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4		# 4 = sb9u
+	subi	r11, r11, 16
+	vperm	v1, $sb9t, v7, v3	# vpshufb	%xmm3,	%xmm1,	%xmm1		# 0 = sb9t
+	andi.	r11, r11, 0x30
+	vxor	v5, v5, v4		# vpxor		%xmm4,	%xmm0,	%xmm0
+					# vmovdqa	0x00(%r10),%xmm4		# 4 : sbdu
+	vxor	v5, v5, v1		# vpxor		%xmm1,	%xmm0,	%xmm0		# 0 = ch
+					# vmovdqa	0x10(%r10),%xmm1		# 0 : sbdt
+
+	vperm	v4, $sbdu, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4		# 4 = sbdu
+	vperm 	v5, v5, v7, v0		# vpshufb	%xmm5,	%xmm0,	%xmm0		# MC ch
+	vperm	v1, $sbdt, v7, v3	# vpshufb	%xmm3,	%xmm1,	%xmm1		# 0 = sbdt
+	vxor	v5, v5, v4		# vpxor		%xmm4,	%xmm0,	%xmm0		# 4 = ch
+					# vmovdqa	0x20(%r10),	%xmm4		# 4 : sbbu
+	vxor	v5, v5, v1		# vpxor		%xmm1,	%xmm0,	%xmm0		# 0 = ch
+					# vmovdqa	0x30(%r10),	%xmm1		# 0 : sbbt
+
+	vperm	v4, $sbbu, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4		# 4 = sbbu
+	vperm	v5, v5, v7, v0		# vpshufb	%xmm5,	%xmm0,	%xmm0		# MC ch
+	vperm	v1, $sbbt, v7, v3	# vpshufb	%xmm3,	%xmm1,	%xmm1		# 0 = sbbt
+	vxor	v5, v5, v4		# vpxor		%xmm4,	%xmm0,	%xmm0		# 4 = ch
+					# vmovdqa	0x40(%r10),	%xmm4		# 4 : sbeu
+	vxor	v5, v5, v1		# vpxor		%xmm1,	%xmm0,	%xmm0		# 0 = ch
+					# vmovdqa	0x50(%r10),	%xmm1		# 0 : sbet
+
+	vperm	v4, $sbeu, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4		# 4 = sbeu
+	vperm	v5, v5, v7, v0		# vpshufb	%xmm5,	%xmm0,	%xmm0		# MC ch
+	vperm	v1, $sbet, v7, v3	# vpshufb	%xmm3,	%xmm1,	%xmm1		# 0 = sbet
+	vxor	v0, v5, v4		# vpxor		%xmm4,	%xmm0,	%xmm0		# 4 = ch
+	vxor	v0, v0, v1		# vpxor		%xmm1,	%xmm0,	%xmm0		# 0 = ch
+
+Ldec_entry:
+	# top of round
+	vsrb	v1, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0	# 1 = i
+	vperm	v2, $invhi, $invhi, v0	# vpshufb	%xmm1,	%xmm11,	%xmm2	# 2 = a/k
+	vxor	v0, v0, v1		# vpxor		%xmm0,	%xmm1,	%xmm1	# 0 = j
+	vperm	v3, $invlo, $invlo, v1	# vpshufb	%xmm0, 	%xmm10,	%xmm3	# 3 = 1/i
+	vperm	v4, $invlo, $invlo, v0	# vpshufb	%xmm1,	%xmm10,	%xmm4	# 4 = 1/j
+	vand	v0, v0, v9
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3	# 3 = iak = 1/i + a/k
+	vxor	v4, v4, v2		# vpxor		%xmm2, 	%xmm4,	%xmm4	# 4 = jak = 1/j + a/k
+	vperm	v2, $invlo, v7, v3	# vpshufb	%xmm3,	%xmm10,	%xmm2	# 2 = 1/iak
+	vmr	v5, v6
+	lvx	v6, r9, $key		# vmovdqu	(%r9),	%xmm0
+	vperm	v3, $invlo, v7, v4	# vpshufb	%xmm4,  %xmm10,	%xmm3	# 3 = 1/jak
+	addi	r9, r9, 16
+	vxor	v2, v2, v0		# vpxor		%xmm1,	%xmm2,	%xmm2	# 2 = io
+	?vperm	v5, v5, v6, $keyperm	# align round key
+	vxor	v3, v3, v1		# vpxor		%xmm0,  %xmm3,	%xmm3	# 3 = jo
+	bdnz	Ldec_loop
+
+	# middle of last round
+	addi	r10, r11, 0x80
+					# vmovdqa	0x60(%r10),	%xmm4	# 3 : sbou
+	vperm	v4, $sbou, v7, v2	# vpshufb	%xmm2,	%xmm4,	%xmm4	# 4 = sbou
+					# vmovdqa	0x70(%r10),	%xmm1	# 0 : sbot
+	lvx	v2, r12, r10		# vmovdqa	-0x160(%r11),	%xmm2	# .Lk_sr-.Lk_dsbd=-0x160
+	vperm	v1, $sbot, v7, v3	# vpshufb	%xmm3,	%xmm1,	%xmm1	# 0 = sb1t
+	vxor	v4, v4, v5		# vpxor		%xmm0,	%xmm4,	%xmm4	# 4 = sb1u + k
+	vxor	v0, v1, v4		# vpxor		%xmm4,	%xmm1,	%xmm0	# 0 = A
+	vperm	v0, v0, v7, v2		# vpshufb	%xmm2,	%xmm0,	%xmm0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+.globl	.vpaes_decrypt
+.align	5
+.vpaes_decrypt:
+	$STU	$sp,-$FRAME($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mflr	r6
+	mfspr	r7, 256			# save vrsave
+	stvx	v20,r10,$sp
+	addi	r10,r10,32
+	stvx	v21,r11,$sp
+	addi	r11,r11,32
+	stvx	v22,r10,$sp
+	addi	r10,r10,32
+	stvx	v23,r11,$sp
+	addi	r11,r11,32
+	stvx	v24,r10,$sp
+	addi	r10,r10,32
+	stvx	v25,r11,$sp
+	addi	r11,r11,32
+	stvx	v26,r10,$sp
+	addi	r10,r10,32
+	stvx	v27,r11,$sp
+	addi	r11,r11,32
+	stvx	v28,r10,$sp
+	addi	r10,r10,32
+	stvx	v29,r11,$sp
+	addi	r11,r11,32
+	stvx	v30,r10,$sp
+	stvx	v31,r11,$sp
+	stw	r7,`$FRAME-4`($sp)	# save vrsave
+	li	r0, -1
+	$PUSH	r6,`$FRAME+$LRSAVE`($sp)
+	mtspr	256, r0			# preserve all AltiVec registers
+
+	bl	_vpaes_decrypt_preheat
+
+	?lvsl	$inpperm, 0, $inp	# prepare for unaligned access
+	lvx	v0, 0, $inp
+	addi	$inp, $inp, 15		# 15 is not a typo
+	 ?lvsr	$outperm, 0, $out
+	?lvsl	$keyperm, 0, $key
+	 vnor	$outmask, v7, v7	# 0xff..ff
+	lvx	$inptail, 0, $inp	# redundant in aligned case
+	 ?vperm	$outmask, v7, $outmask, $outperm
+	 lvx	$outhead, 0, $out
+	?vperm	v0, v0, $inptail, $inpperm
+
+	bl	_vpaes_decrypt_core
+
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v1, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v1, 0, $out
+	addi	$out, $out, 15		# 15 is not a typo
+	########
+
+	lvx	v1, 0, $out		# redundant in aligned case
+	vsel	v1, $outhead, v1, $outmask
+	stvx	v1, 0, $out
+
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mtlr	r6
+	mtspr	256, r7			# restore vrsave
+	lvx	v20,r10,$sp
+	addi	r10,r10,32
+	lvx	v21,r11,$sp
+	addi	r11,r11,32
+	lvx	v22,r10,$sp
+	addi	r10,r10,32
+	lvx	v23,r11,$sp
+	addi	r11,r11,32
+	lvx	v24,r10,$sp
+	addi	r10,r10,32
+	lvx	v25,r11,$sp
+	addi	r11,r11,32
+	lvx	v26,r10,$sp
+	addi	r10,r10,32
+	lvx	v27,r11,$sp
+	addi	r11,r11,32
+	lvx	v28,r10,$sp
+	addi	r10,r10,32
+	lvx	v29,r11,$sp
+	addi	r11,r11,32
+	lvx	v30,r10,$sp
+	lvx	v31,r11,$sp
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,0x04,1,0x80,0,3,0
+	.long	0
+.size	.vpaes_decrypt,.-.vpaes_decrypt
+
+.globl	.vpaes_cbc_encrypt
+.align	5
+.vpaes_cbc_encrypt:
+	${UCMP}i r5,16
+	bltlr-
+
+	$STU	$sp,-`($FRAME+2*$SIZE_T)`($sp)
+	mflr	r0
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mfspr	r12, 256
+	stvx	v20,r10,$sp
+	addi	r10,r10,32
+	stvx	v21,r11,$sp
+	addi	r11,r11,32
+	stvx	v22,r10,$sp
+	addi	r10,r10,32
+	stvx	v23,r11,$sp
+	addi	r11,r11,32
+	stvx	v24,r10,$sp
+	addi	r10,r10,32
+	stvx	v25,r11,$sp
+	addi	r11,r11,32
+	stvx	v26,r10,$sp
+	addi	r10,r10,32
+	stvx	v27,r11,$sp
+	addi	r11,r11,32
+	stvx	v28,r10,$sp
+	addi	r10,r10,32
+	stvx	v29,r11,$sp
+	addi	r11,r11,32
+	stvx	v30,r10,$sp
+	stvx	v31,r11,$sp
+	stw	r12,`$FRAME-4`($sp)	# save vrsave
+	$PUSH	r30,`$FRAME+$SIZE_T*0`($sp)
+	$PUSH	r31,`$FRAME+$SIZE_T*1`($sp)
+	li	r9, -16
+	$PUSH	r0, `$FRAME+$SIZE_T*2+$LRSAVE`($sp)
+
+	and	r30, r5, r9		# copy length&-16
+	mr	r5, r6			# copy pointer to key
+	mr	r31, r7			# copy pointer to iv
+	blt	Lcbc_abort
+	cmpwi	r8, 0			# test direction
+	li	r6, -1
+	mr	r7, r12			# copy vrsave
+	mtspr	256, r6			# preserve all AltiVec registers
+
+	lvx	v24, 0, r31		# load [potentially unaligned] iv
+	li	r9, 15
+	?lvsl	$inpperm, 0, r31
+	lvx	v25, r9, r31
+	?vperm	v24, v24, v25, $inpperm
+
+	neg	r8, $inp		# prepare for unaligned access
+	 vxor	v7, v7, v7
+	?lvsl	$keyperm, 0, $key
+	 ?lvsr	$outperm, 0, $out
+	?lvsr	$inpperm, 0, r8		# -$inp
+	 vnor	$outmask, v7, v7	# 0xff..ff
+	lvx	$inptail, 0, $inp
+	 ?vperm	$outmask, v7, $outmask, $outperm
+	addi	$inp, $inp, 15		# 15 is not a typo
+	 lvx	$outhead, 0, $out
+
+	beq	Lcbc_decrypt
+
+	bl	_vpaes_encrypt_preheat
+	li	r0, 16
+
+Lcbc_enc_loop:
+	vmr	v0, $inptail
+	lvx	$inptail, 0, $inp
+	addi	$inp, $inp, 16
+	?vperm	v0, v0, $inptail, $inpperm
+	vxor	v0, v0, v24		# ^= iv
+
+	bl	_vpaes_encrypt_core
+
+	vmr	v24, v0			# put aside iv
+	sub.	r30, r30, r0		# len -= 16
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v1, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v1, 0, $out
+	addi	$out, $out, 16
+	bne	Lcbc_enc_loop
+
+	b	Lcbc_done
+
+.align	5
+Lcbc_decrypt:
+	bl	_vpaes_decrypt_preheat
+	li	r0, 16
+
+Lcbc_dec_loop:
+	vmr	v0, $inptail
+	lvx	$inptail, 0, $inp
+	addi	$inp, $inp, 16
+	?vperm	v0, v0, $inptail, $inpperm
+	vmr	v25, v0			# put aside input
+
+	bl	_vpaes_decrypt_core
+
+	vxor	v0, v0, v24		# ^= iv
+	vmr	v24, v25
+	sub.	r30, r30, r0		# len -= 16
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v1, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v1, 0, $out
+	addi	$out, $out, 16
+	bne	Lcbc_dec_loop
+
+Lcbc_done:
+	addi	$out, $out, -1
+	lvx	v1, 0, $out		# redundant in aligned case
+	vsel	v1, $outhead, v1, $outmask
+	stvx	v1, 0, $out
+
+	neg	r8, r31			# write [potentially unaligned] iv
+	?lvsl	$outperm, 0, r8
+	li	r6, 15
+	vnor	$outmask, v7, v7	# 0xff..ff
+	?vperm	$outmask, v7, $outmask, $outperm
+	lvx	$outhead, 0, r31
+	vperm	v24, v24, v24, $outperm	# rotate right/left
+	vsel	v0, $outhead, v24, $outmask
+	lvx	v1, r6, r31
+	stvx	v0, 0, r31
+	vsel	v1, v24, v1, $outmask
+	stvx	v1, r6, r31
+
+	mtspr	256, r7			# restore vrsave
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	lvx	v20,r10,$sp
+	addi	r10,r10,32
+	lvx	v21,r11,$sp
+	addi	r11,r11,32
+	lvx	v22,r10,$sp
+	addi	r10,r10,32
+	lvx	v23,r11,$sp
+	addi	r11,r11,32
+	lvx	v24,r10,$sp
+	addi	r10,r10,32
+	lvx	v25,r11,$sp
+	addi	r11,r11,32
+	lvx	v26,r10,$sp
+	addi	r10,r10,32
+	lvx	v27,r11,$sp
+	addi	r11,r11,32
+	lvx	v28,r10,$sp
+	addi	r10,r10,32
+	lvx	v29,r11,$sp
+	addi	r11,r11,32
+	lvx	v30,r10,$sp
+	lvx	v31,r11,$sp
+Lcbc_abort:
+	$POP	r0, `$FRAME+$SIZE_T*2+$LRSAVE`($sp)
+	$POP	r30,`$FRAME+$SIZE_T*0`($sp)
+	$POP	r31,`$FRAME+$SIZE_T*1`($sp)
+	mtlr	r0
+	addi	$sp,$sp,`$FRAME+$SIZE_T*2`
+	blr
+	.long	0
+	.byte	0,12,0x04,1,0x80,2,6,0
+	.long	0
+.size	.vpaes_cbc_encrypt,.-.vpaes_cbc_encrypt
+___
+}
+{
+my ($inp,$bits,$out)=map("r$_",(3..5));
+my $dir="cr1";
+my ($invlo,$invhi,$iptlo,$ipthi,$rcon) = map("v$_",(10..13,24));
+
+$code.=<<___;
+########################################################
+##                                                    ##
+##                  AES key schedule                  ##
+##                                                    ##
+########################################################
+.align	4
+_vpaes_key_preheat:
+	mflr	r8
+	bl	Lconsts
+	mtlr	r8
+	li	r11, 0xc0		# Lk_inv
+	li	r10, 0xd0
+	li	r9,  0xe0		# L_ipt
+	li	r8,  0xf0
+
+	vspltisb	v8,4		# 0x04..04
+	vxor	v9,v9,v9		# 0x00..00
+	lvx	$invlo, r12, r11	# Lk_inv
+	li	r11, 0x120
+	lvx	$invhi, r12, r10
+	li	r10, 0x130
+	lvx	$iptlo, r12, r9		# Lk_ipt
+	li	r9, 0x220
+	lvx	$ipthi, r12, r8
+	li	r8, 0x230
+
+	lvx	v14, r12, r11		# Lk_sb1
+	li	r11, 0x240
+	lvx	v15, r12, r10
+	li	r10, 0x250
+
+	lvx	v16, r12, r9		# Lk_dksd
+	li	r9, 0x260
+	lvx	v17, r12, r8
+	li	r8, 0x270
+	lvx	v18, r12, r11		# Lk_dksb
+	li	r11, 0x280
+	lvx	v19, r12, r10
+	li	r10, 0x290
+	lvx	v20, r12, r9		# Lk_dkse
+	li	r9, 0x2a0
+	lvx	v21, r12, r8
+	li	r8, 0x2b0
+	lvx	v22, r12, r11		# Lk_dks9
+	lvx	v23, r12, r10
+
+	lvx	v24, r12, r9		# Lk_rcon
+	lvx	v25, 0, r12		# Lk_mc_forward[0]
+	lvx	v26, r12, r8		# Lks63
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+.align	4
+_vpaes_schedule_core:
+	mflr	r7
+
+	bl	_vpaes_key_preheat	# load the tables
+
+	#lvx	v0, 0, $inp		# vmovdqu	(%rdi),	%xmm0		# load key (unaligned)
+	neg	r8, $inp		# prepare for unaligned access
+	lvx	v0, 0, $inp
+	addi	$inp, $inp, 15		# 15 is not typo
+	?lvsr	$inpperm, 0, r8		# -$inp
+	lvx	v6, 0, $inp		# v6 serves as inptail
+	addi	$inp, $inp, 8
+	?vperm	v0, v0, v6, $inpperm
+
+	# input transform
+	vmr	v3, v0			# vmovdqa	%xmm0,	%xmm3
+	bl	_vpaes_schedule_transform
+	vmr	v7, v0			# vmovdqa	%xmm0,	%xmm7
+
+	bne	$dir, Lschedule_am_decrypting
+
+	# encrypting, output zeroth round key after transform
+	li	r8, 0x30		# mov	\$0x30,%r8d
+	addi	r10, r12, 0x80		# lea	.Lk_sr(%rip),%r10
+
+	?lvsr	$outperm, 0, $out	# prepare for unaligned access
+	vnor	$outmask, v9, v9	# 0xff..ff
+	lvx	$outhead, 0, $out
+	?vperm	$outmask, v9, $outmask, $outperm
+
+	#stvx	v0, 0, $out		# vmovdqu	%xmm0,	(%rdx)
+	vperm	v1, v0, v0, $outperm	# rotate right/left
+	vsel	v2, $outhead, v1, $outmask
+	vmr	$outhead, v1
+	stvx	v2, 0, $out
+	b	Lschedule_go
+
+Lschedule_am_decrypting:
+	srwi	r8, $bits, 1		# shr	\$1,%r8d
+	andi.	r8, r8, 32		# and	\$32,%r8d
+	xori	r8, r8, 32		# xor	\$32,%r8d	# nbits==192?0:32
+	addi	r10, r12, 0x80		# lea	.Lk_sr(%rip),%r10
+	# decrypting, output zeroth round key after shiftrows
+	lvx	v1, r8, r10		# vmovdqa	(%r8,%r10),	%xmm1
+	vperm	v4, v3, v3, v1		# vpshufb	%xmm1,	%xmm3,	%xmm3
+
+	neg	r0, $out		# prepare for unaligned access
+	?lvsl	$outperm, 0, r0
+	addi	$out, $out, 15		# 15 is not typo
+	vnor	$outmask, v9, v9	# 0xff..ff
+	lvx	$outhead, 0, $out
+	?vperm	$outmask, $outmask, v9, $outperm
+
+	#stvx	v4, 0, $out		# vmovdqu	%xmm3,	(%rdx)
+	vperm	v4, v4, v4, $outperm	# rotate right/left
+	vsel	v2, $outhead, v4, $outmask
+	vmr	$outhead, v4
+	stvx	v2, 0, $out
+	xori	r8, r8, 0x30		# xor	\$0x30, %r8
+
+Lschedule_go:
+	cmplwi	$bits, 192		# cmp	\$192,	%esi
+	bgt	Lschedule_256
+	beq	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:
+	li	r0, 10			# mov	\$10, %esi
+	mtctr	r0
+
+Loop_schedule_128:
+	bl 	_vpaes_schedule_round
+	bdz 	Lschedule_mangle_last	# dec	%esi
+	bl	_vpaes_schedule_mangle	# write output
+	b 	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	4
+Lschedule_192:
+	li	r0, 4			# mov	\$4,	%esi
+	lvx	v0, 0, $inp
+	?vperm	v0, v6, v0, $inpperm
+	?vsldoi	v0, v3, v0, 8		# vmovdqu	8(%rdi),%xmm0		# load key part 2 (very unaligned)
+	bl	_vpaes_schedule_transform	# input transform
+	?vsldoi	v6, v0, v9, 8
+	?vsldoi	v6, v9, v6, 8		# clobber "low" side with zeros
+	mtctr	r0
+
+Loop_schedule_192:
+	bl	_vpaes_schedule_round
+	?vsldoi	v0, v6, v0, 8		# vpalignr	\$8,%xmm6,%xmm0,%xmm0
+	bl	_vpaes_schedule_mangle	# save key n
+	bl	_vpaes_schedule_192_smear
+	bl	_vpaes_schedule_mangle	# save key n+1
+	bl	_vpaes_schedule_round
+	bdz 	Lschedule_mangle_last	# dec	%esi
+	bl	_vpaes_schedule_mangle	# save key n+2
+	bl	_vpaes_schedule_192_smear
+	b	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	4
+Lschedule_256:
+	li	r0, 7			# mov	\$7, %esi
+	addi	$inp, $inp, 8
+	lvx	v0, 0, $inp		# vmovdqu	16(%rdi),%xmm0		# load key part 2 (unaligned)
+	?vperm	v0, v6, v0, $inpperm
+	bl	_vpaes_schedule_transform	# input transform
+	mtctr	r0
+
+Loop_schedule_256:
+	bl	_vpaes_schedule_mangle	# output low result
+	vmr	v6, v0			# vmovdqa	%xmm0,	%xmm6		# save cur_lo in xmm6
+
+	# high round
+	bl	_vpaes_schedule_round
+	bdz 	Lschedule_mangle_last	# dec	%esi
+	bl	_vpaes_schedule_mangle	
+
+	# low round. swap xmm7 and xmm6
+	?vspltw	v0, v0, 3		# vpshufd	\$0xFF,	%xmm0,	%xmm0
+	vmr	v5, v7			# vmovdqa	%xmm7,	%xmm5
+	vmr	v7, v6			# vmovdqa	%xmm6,	%xmm7
+	bl	_vpaes_schedule_low_round
+	vmr	v7, v5			# vmovdqa	%xmm5,	%xmm7
+	
+	b	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	4
+Lschedule_mangle_last:
+	# schedule last round key from xmm0
+	li	r11, 0x2e0		# lea	.Lk_deskew(%rip),%r11
+	li	r9,  0x2f0
+	bne	$dir, Lschedule_mangle_last_dec
+
+	# encrypting
+	lvx	v1, r8, r10		# vmovdqa	(%r8,%r10),%xmm1
+	li	r11, 0x2c0		# lea		.Lk_opt(%rip),	%r11	# prepare to output transform
+	li	r9,  0x2d0		# prepare to output transform
+	vperm	v0, v0, v0, v1		# vpshufb	%xmm1,	%xmm0,	%xmm0	# output permute
+
+	lvx	$iptlo, r11, r12	# reload $ipt
+	lvx	$ipthi, r9, r12
+	addi	$out, $out, 16		# add	\$16,	%rdx
+	vxor	v0, v0, v26		# vpxor		.Lk_s63(%rip),	%xmm0,	%xmm0
+	bl	_vpaes_schedule_transform	# output transform
+
+	#stvx	v0, r0, $out		# vmovdqu	%xmm0,	(%rdx)		# save last key
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v2, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v2, 0, $out
+
+	addi	$out, $out, 15		# 15 is not typo
+	lvx	v1, 0, $out		# redundant in aligned case
+	vsel	v1, $outhead, v1, $outmask
+	stvx	v1, 0, $out
+	b	Lschedule_mangle_done
+
+.align	4
+Lschedule_mangle_last_dec:
+	lvx	$iptlo, r11, r12	# reload $ipt
+	lvx	$ipthi, r9,  r12
+	addi	$out, $out, -16		# add	\$-16,	%rdx 
+	vxor	v0, v0, v26		# vpxor	.Lk_s63(%rip),	%xmm0,	%xmm0
+	bl	_vpaes_schedule_transform	# output transform
+
+	#stvx	v0, r0, $out		# vmovdqu	%xmm0,	(%rdx)		# save last key
+	vperm	v0, v0, v0, $outperm	# rotate right/left
+	vsel	v2, $outhead, v0, $outmask
+	vmr	$outhead, v0
+	stvx	v2, 0, $out
+
+	addi	$out, $out, -15		# -15 is not typo
+	lvx	v1, 0, $out		# redundant in aligned case
+	vsel	v1, $outhead, v1, $outmask
+	stvx	v1, 0, $out
+
+Lschedule_mangle_done:
+	mtlr	r7
+	# cleanup
+	vxor	v0, v0, v0		# vpxor		%xmm0,	%xmm0,	%xmm0
+	vxor	v1, v1, v1		# vpxor		%xmm1,	%xmm1,	%xmm1
+	vxor	v2, v2, v2		# vpxor		%xmm2,	%xmm2,	%xmm2
+	vxor	v3, v3, v3		# vpxor		%xmm3,	%xmm3,	%xmm3
+	vxor	v4, v4, v4		# vpxor		%xmm4,	%xmm4,	%xmm4
+	vxor	v5, v5, v5		# vpxor		%xmm5,	%xmm5,	%xmm5
+	vxor	v6, v6, v6		# vpxor		%xmm6,	%xmm6,	%xmm6
+	vxor	v7, v7, v7		# vpxor		%xmm7,	%xmm7,	%xmm7
+
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  .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
+##
+.align	4
+_vpaes_schedule_192_smear:
+	?vspltw	v0, v7, 3
+	?vsldoi	v1, v9, v6, 12		# vpshufd	\$0x80,	%xmm6,	%xmm1	# d c 0 0 -> c 0 0 0
+	?vsldoi	v0, v7, v0, 8		# vpshufd	\$0xFE,	%xmm7,	%xmm0	# b a _ _ -> b b b a
+	vxor	v6, v6, v1		# vpxor		%xmm1,	%xmm6,	%xmm6	# -> c+d c 0 0
+	vxor	v6, v6, v0		# vpxor		%xmm0,	%xmm6,	%xmm6	# -> b+c+d b+c b a
+	vmr	v0, v6
+	?vsldoi	v6, v6, v9, 8
+	?vsldoi	v6, v9, v6, 8		# clobber low side with zeros
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  .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.
+##
+.align	4
+_vpaes_schedule_round:
+	# extract rcon from xmm8
+	#vxor	v4, v4, v4		# vpxor		%xmm4,	%xmm4,	%xmm4
+	?vsldoi	v1, $rcon, v9, 15	# vpalignr	\$15,	%xmm8,	%xmm4,	%xmm1
+	?vsldoi	$rcon, $rcon, $rcon, 15	# vpalignr	\$15,	%xmm8,	%xmm8,	%xmm8
+	vxor	v7, v7, v1		# vpxor		%xmm1,	%xmm7,	%xmm7
+
+	# rotate
+	?vspltw	v0, v0, 3		# vpshufd	\$0xFF,	%xmm0,	%xmm0
+	?vsldoi	v0, v0, v0, 1		# vpalignr	\$1,	%xmm0,	%xmm0,	%xmm0
+
+	# fall through...
+
+	# low round: same as high round, but no rotation and no rcon.
+_vpaes_schedule_low_round:
+	# smear xmm7
+	?vsldoi	v1, v9, v7, 12		# vpslldq	\$4,	%xmm7,	%xmm1
+	vxor	v7, v7, v1		# vpxor		%xmm1,	%xmm7,	%xmm7
+	vspltisb	v1, 0x0f	# 0x0f..0f
+	?vsldoi	v4, v9, v7, 8		# vpslldq	\$8,	%xmm7,	%xmm4
+
+	# subbytes
+	vand	v1, v1, v0		# vpand		%xmm9,	%xmm0,	%xmm1		# 0 = k
+	vsrb	v0, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0		# 1 = i
+	 vxor	v7, v7, v4		# vpxor		%xmm4,	%xmm7,	%xmm7
+	vperm	v2, $invhi, v9, v1	# vpshufb	%xmm1,	%xmm11,	%xmm2		# 2 = a/k
+	vxor	v1, v1, v0		# vpxor		%xmm0,	%xmm1,	%xmm1		# 0 = j
+	vperm	v3, $invlo, v9, v0	# vpshufb	%xmm0, 	%xmm10,	%xmm3		# 3 = 1/i
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3		# 3 = iak = 1/i + a/k
+	vperm	v4, $invlo, v9, v1	# vpshufb	%xmm1,	%xmm10,	%xmm4		# 4 = 1/j
+	 vxor	v7, v7, v26		# vpxor		.Lk_s63(%rip),	%xmm7,	%xmm7
+	vperm	v3, $invlo, v9, v3	# vpshufb	%xmm3,	%xmm10,	%xmm3		# 2 = 1/iak
+	vxor	v4, v4, v2		# vpxor		%xmm2,	%xmm4,	%xmm4		# 4 = jak = 1/j + a/k
+	vperm	v2, $invlo, v9, v4	# vpshufb	%xmm4,	%xmm10,	%xmm2		# 3 = 1/jak
+	vxor	v3, v3, v1		# vpxor		%xmm1,	%xmm3,	%xmm3		# 2 = io
+	vxor	v2, v2, v0		# vpxor		%xmm0,	%xmm2,	%xmm2		# 3 = jo
+	vperm	v4, v15, v9, v3		# vpshufb	%xmm3,	%xmm13,	%xmm4		# 4 = sbou
+	vperm	v1, v14, v9, v2		# vpshufb	%xmm2,	%xmm12,	%xmm1		# 0 = sb1t
+	vxor	v1, v1, v4		# vpxor		%xmm4,	%xmm1,	%xmm1		# 0 = sbox output
+
+	# add in smeared stuff
+	vxor	v0, v1, v7		# vpxor		%xmm7,	%xmm1,	%xmm0
+	vxor	v7, v1, v7		# vmovdqa	%xmm0,	%xmm7
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  .aes_schedule_transform
+##
+##  Linear-transform %xmm0 according to tables at (%r11)
+##
+##  Requires that %xmm9 = 0x0F0F... as in preheat
+##  Output in %xmm0
+##  Clobbers %xmm2
+##
+.align	4
+_vpaes_schedule_transform:
+	#vand	v1, v0, v9		# vpand		%xmm9,	%xmm0,	%xmm1
+	vsrb	v2, v0, v8		# vpsrlb	\$4,	%xmm0,	%xmm0
+					# vmovdqa	(%r11),	%xmm2 	# lo
+	vperm	v0, $iptlo, $iptlo, v0	# vpshufb	%xmm1,	%xmm2,	%xmm2
+					# vmovdqa	16(%r11),	%xmm1 # hi
+	vperm	v2, $ipthi, $ipthi, v2	# vpshufb	%xmm0,	%xmm1,	%xmm0
+	vxor	v0, v0, v2		# vpxor		%xmm2,	%xmm0,	%xmm0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+##
+##  .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
+##
+.align	4
+_vpaes_schedule_mangle:
+	#vmr	v4, v0			# vmovdqa	%xmm0,	%xmm4	# save xmm0 for later
+					# vmovdqa	.Lk_mc_forward(%rip),%xmm5
+	bne	$dir, Lschedule_mangle_dec
+
+	# encrypting
+	vxor	v4, v0, v26		# vpxor	.Lk_s63(%rip),	%xmm0,	%xmm4
+	addi	$out, $out, 16		# add	\$16,	%rdx
+	vperm	v4, v4, v4, v25		# vpshufb	%xmm5,	%xmm4,	%xmm4
+	vperm	v1, v4, v4, v25		# vpshufb	%xmm5,	%xmm4,	%xmm1
+	vperm	v3, v1, v1, v25		# vpshufb	%xmm5,	%xmm1,	%xmm3
+	vxor	v4, v4, v1		# vpxor		%xmm1,	%xmm4,	%xmm4
+	lvx	v1, r8, r10		# vmovdqa	(%r8,%r10),	%xmm1
+	vxor	v3, v3, v4		# vpxor		%xmm4,	%xmm3,	%xmm3
+
+	vperm	v3, v3, v3, v1		# vpshufb	%xmm1,	%xmm3,	%xmm3
+	addi	r8, r8, -16		# add	\$-16,	%r8
+	andi.	r8, r8, 0x30		# and	\$0x30,	%r8
+
+	#stvx	v3, 0, $out		# vmovdqu	%xmm3,	(%rdx)
+	vperm	v1, v3, v3, $outperm	# rotate right/left
+	vsel	v2, $outhead, v1, $outmask
+	vmr	$outhead, v1
+	stvx	v2, 0, $out
+	blr
+
+.align	4
+Lschedule_mangle_dec:
+	# inverse mix columns
+					# lea	.Lk_dksd(%rip),%r11
+	vsrb	v1, v0, v8		# vpsrlb	\$4,	%xmm4,	%xmm1	# 1 = hi
+	#and	v4, v0, v9		# vpand		%xmm9,	%xmm4,	%xmm4	# 4 = lo
+
+					# vmovdqa	0x00(%r11),	%xmm2
+	vperm	v2, v16, v16, v0	# vpshufb	%xmm4,	%xmm2,	%xmm2
+					# vmovdqa	0x10(%r11),	%xmm3
+	vperm	v3, v17, v17, v1	# vpshufb	%xmm1,	%xmm3,	%xmm3
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3
+	vperm	v3, v3, v9, v25		# vpshufb	%xmm5,	%xmm3,	%xmm3
+
+					# vmovdqa	0x20(%r11),	%xmm2
+	vperm	v2, v18, v18, v0	# vpshufb	%xmm4,	%xmm2,	%xmm2
+	vxor	v2, v2, v3		# vpxor		%xmm3,	%xmm2,	%xmm2
+					# vmovdqa	0x30(%r11),	%xmm3
+	vperm	v3, v19, v19, v1	# vpshufb	%xmm1,	%xmm3,	%xmm3
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3
+	vperm	v3, v3, v9, v25		# vpshufb	%xmm5,	%xmm3,	%xmm3
+
+					# vmovdqa	0x40(%r11),	%xmm2
+	vperm	v2, v20, v20, v0	# vpshufb	%xmm4,	%xmm2,	%xmm2
+	vxor	v2, v2, v3		# vpxor		%xmm3,	%xmm2,	%xmm2
+					# vmovdqa	0x50(%r11),	%xmm3
+	vperm	v3, v21, v21, v1	# vpshufb	%xmm1,	%xmm3,	%xmm3
+	vxor	v3, v3, v2		# vpxor		%xmm2,	%xmm3,	%xmm3
+
+					# vmovdqa	0x60(%r11),	%xmm2
+	vperm	v2, v22, v22, v0	# vpshufb	%xmm4,	%xmm2,	%xmm2
+	vperm	v3, v3, v9, v25		# vpshufb	%xmm5,	%xmm3,	%xmm3
+					# vmovdqa	0x70(%r11),	%xmm4
+	vperm	v4, v23, v23, v1	# vpshufb	%xmm1,	%xmm4,	%xmm4
+	lvx	v1, r8, r10		# vmovdqa	(%r8,%r10),	%xmm1
+	vxor	v2, v2, v3		# vpxor		%xmm3,	%xmm2,	%xmm2
+	vxor	v3, v4, v2		# vpxor		%xmm2,	%xmm4,	%xmm3
+
+	addi	$out, $out, -16		# add	\$-16,	%rdx
+
+	vperm	v3, v3, v3, v1		# vpshufb	%xmm1,	%xmm3,	%xmm3
+	addi	r8, r8, -16		# add	\$-16,	%r8
+	andi.	r8, r8, 0x30		# and	\$0x30,	%r8
+
+	#stvx	v3, 0, $out		# vmovdqu	%xmm3,	(%rdx)
+	vperm	v1, v3, v3, $outperm	# rotate right/left
+	vsel	v2, $outhead, v1, $outmask
+	vmr	$outhead, v1
+	stvx	v2, 0, $out
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
+.globl	.vpaes_set_encrypt_key
+.align	5
+.vpaes_set_encrypt_key:
+	$STU	$sp,-$FRAME($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mflr	r0
+	mfspr	r6, 256			# save vrsave
+	stvx	v20,r10,$sp
+	addi	r10,r10,32
+	stvx	v21,r11,$sp
+	addi	r11,r11,32
+	stvx	v22,r10,$sp
+	addi	r10,r10,32
+	stvx	v23,r11,$sp
+	addi	r11,r11,32
+	stvx	v24,r10,$sp
+	addi	r10,r10,32
+	stvx	v25,r11,$sp
+	addi	r11,r11,32
+	stvx	v26,r10,$sp
+	addi	r10,r10,32
+	stvx	v27,r11,$sp
+	addi	r11,r11,32
+	stvx	v28,r10,$sp
+	addi	r10,r10,32
+	stvx	v29,r11,$sp
+	addi	r11,r11,32
+	stvx	v30,r10,$sp
+	stvx	v31,r11,$sp
+	stw	r6,`$FRAME-4`($sp)	# save vrsave
+	li	r7, -1
+	$PUSH	r0, `$FRAME+$LRSAVE`($sp)
+	mtspr	256, r7			# preserve all AltiVec registers
+
+	srwi	r9, $bits, 5		# shr	\$5,%eax
+	addi	r9, r9, 6		# add	\$5,%eax
+	stw	r9, 240($out)		# mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
+
+	cmplw	$dir, $bits, $bits	# set encrypt direction
+	li	r8, 0x30		# mov	\$0x30,%r8d
+	bl	_vpaes_schedule_core
+
+	$POP	r0, `$FRAME+$LRSAVE`($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mtspr	256, r6			# restore vrsave
+	mtlr	r0
+	xor	r3, r3, r3
+	lvx	v20,r10,$sp
+	addi	r10,r10,32
+	lvx	v21,r11,$sp
+	addi	r11,r11,32
+	lvx	v22,r10,$sp
+	addi	r10,r10,32
+	lvx	v23,r11,$sp
+	addi	r11,r11,32
+	lvx	v24,r10,$sp
+	addi	r10,r10,32
+	lvx	v25,r11,$sp
+	addi	r11,r11,32
+	lvx	v26,r10,$sp
+	addi	r10,r10,32
+	lvx	v27,r11,$sp
+	addi	r11,r11,32
+	lvx	v28,r10,$sp
+	addi	r10,r10,32
+	lvx	v29,r11,$sp
+	addi	r11,r11,32
+	lvx	v30,r10,$sp
+	lvx	v31,r11,$sp
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,0x04,1,0x80,0,3,0
+	.long	0
+.size	.vpaes_set_encrypt_key,.-.vpaes_set_encrypt_key
+
+.globl	.vpaes_set_decrypt_key
+.align	4
+.vpaes_set_decrypt_key:
+	$STU	$sp,-$FRAME($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mflr	r0
+	mfspr	r6, 256			# save vrsave
+	stvx	v20,r10,$sp
+	addi	r10,r10,32
+	stvx	v21,r11,$sp
+	addi	r11,r11,32
+	stvx	v22,r10,$sp
+	addi	r10,r10,32
+	stvx	v23,r11,$sp
+	addi	r11,r11,32
+	stvx	v24,r10,$sp
+	addi	r10,r10,32
+	stvx	v25,r11,$sp
+	addi	r11,r11,32
+	stvx	v26,r10,$sp
+	addi	r10,r10,32
+	stvx	v27,r11,$sp
+	addi	r11,r11,32
+	stvx	v28,r10,$sp
+	addi	r10,r10,32
+	stvx	v29,r11,$sp
+	addi	r11,r11,32
+	stvx	v30,r10,$sp
+	stvx	v31,r11,$sp
+	stw	r6,`$FRAME-4`($sp)	# save vrsave
+	li	r7, -1
+	$PUSH	r0, `$FRAME+$LRSAVE`($sp)
+	mtspr	256, r7			# preserve all AltiVec registers
+
+	srwi	r9, $bits, 5		# shr	\$5,%eax
+	addi	r9, r9, 6		# add	\$5,%eax
+	stw	r9, 240($out)		# mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
+
+	slwi	r9, r9, 4		# shl	\$4,%eax
+	add	$out, $out, r9		# lea	(%rdx,%rax),%rdx
+
+	cmplwi	$dir, $bits, 0		# set decrypt direction
+	srwi	r8, $bits, 1		# shr	\$1,%r8d
+	andi.	r8, r8, 32		# and	\$32,%r8d
+	xori	r8, r8, 32		# xor	\$32,%r8d	# nbits==192?0:32
+	bl	_vpaes_schedule_core
+
+	$POP	r0,  `$FRAME+$LRSAVE`($sp)
+	li	r10,`15+6*$SIZE_T`
+	li	r11,`31+6*$SIZE_T`
+	mtspr	256, r6			# restore vrsave
+	mtlr	r0
+	xor	r3, r3, r3
+	lvx	v20,r10,$sp
+	addi	r10,r10,32
+	lvx	v21,r11,$sp
+	addi	r11,r11,32
+	lvx	v22,r10,$sp
+	addi	r10,r10,32
+	lvx	v23,r11,$sp
+	addi	r11,r11,32
+	lvx	v24,r10,$sp
+	addi	r10,r10,32
+	lvx	v25,r11,$sp
+	addi	r11,r11,32
+	lvx	v26,r10,$sp
+	addi	r10,r10,32
+	lvx	v27,r11,$sp
+	addi	r11,r11,32
+	lvx	v28,r10,$sp
+	addi	r10,r10,32
+	lvx	v29,r11,$sp
+	addi	r11,r11,32
+	lvx	v30,r10,$sp
+	lvx	v31,r11,$sp
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,0x04,1,0x80,0,3,0
+	.long	0
+.size	.vpaes_set_decrypt_key,.-.vpaes_set_decrypt_key
+___
+}
+
+my $consts=1;
+foreach  (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	# constants table endian-specific conversion
+	if ($consts && m/\.long\s+(.+)\s+(\?[a-z]*)$/o) {
+	    my $conv=$2;
+	    my @bytes=();
+
+	    # convert to endian-agnostic format
+	    foreach (split(/,\s+/,$1)) {
+		my $l = /^0/?oct:int;
+		push @bytes,($l>>24)&0xff,($l>>16)&0xff,($l>>8)&0xff,$l&0xff;
+	    }
+
+	    # little-endian conversion
+	    if ($flavour =~ /le$/o) {
+		SWITCH: for($conv)  {
+		    /\?inv/ && do   { @bytes=map($_^0xf,@bytes); last; };
+		    /\?rev/ && do   { @bytes=reverse(@bytes);    last; }; 
+		}
+	    }
+
+	    #emit
+	    print ".byte\t",join(',',map (sprintf("0x%02x",$_),@bytes)),"\n";
+	    next;
+	}
+	$consts=0 if (m/Lconsts:/o);	# end of table
+
+	# instructions prefixed with '?' are endian-specific and need
+	# to be adjusted accordingly...
+	if ($flavour =~ /le$/o) {	# little-endian
+	    s/\?lvsr/lvsl/o or
+	    s/\?lvsl/lvsr/o or
+	    s/\?(vperm\s+v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+)/$1$3$2$4/o or
+	    s/\?(vsldoi\s+v[0-9]+,\s*)(v[0-9]+,)\s*(v[0-9]+,\s*)([0-9]+)/$1$3$2 16-$4/o or
+	    s/\?(vspltw\s+v[0-9]+,\s*)(v[0-9]+,)\s*([0-9])/$1$2 3-$3/o;
+	} else {			# big-endian
+	    s/\?([a-z]+)/$1/o;
+	}
+
+	print $_,"\n";
+}
+
+close STDOUT;
diff --git a/openssl/crypto/aes/asm/vpaes-x86.pl b/openssl/crypto/aes/asm/vpaes-x86.pl
index 1533e2c30..2ba149c3f 100644
--- a/openssl/crypto/aes/asm/vpaes-x86.pl
+++ b/openssl/crypto/aes/asm/vpaes-x86.pl
@@ -27,9 +27,10 @@
 #
 #		aes-586.pl		vpaes-x86.pl
 #
-# Core 2(**)	29.1/42.3/18.3		22.0/25.6(***)
-# Nehalem	27.9/40.4/18.1		10.3/12.0
-# Atom		102./119./60.1		64.5/85.3(***)
+# 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
@@ -40,8 +41,8 @@
 # (**)	"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 +32%/65%  improvement on Core 2
-#	and +58%/40% on Atom (as implied, over "hyper-threading-safe"
+#	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>
@@ -183,35 +184,35 @@ $k_dsbo=0x2c0;		# decryption sbox final output
 	&movdqa	("xmm1","xmm6")
 	&movdqa	("xmm2",&QWP($k_ipt,$const));
 	&pandn	("xmm1","xmm0");
-	&movdqu	("xmm5",&QWP(0,$key));
-	&psrld	("xmm1",4);
 	&pand	("xmm0","xmm6");
+	&movdqu	("xmm5",&QWP(0,$key));
 	&pshufb	("xmm2","xmm0");
 	&movdqa	("xmm0",&QWP($k_ipt+16,$const));
-	&pshufb	("xmm0","xmm1");
 	&pxor	("xmm2","xmm5");
-	&pxor	("xmm0","xmm2");
+	&psrld	("xmm1",4);
 	&add	($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
-	&pshufb	("xmm4","xmm2");		# 4 = sb1u
-	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
 	&movdqa	("xmm0",&QWP($k_sb1+16,$const));# 0 : sb1t
+	&pshufb	("xmm4","xmm2");		# 4 = sb1u
 	&pshufb	("xmm0","xmm3");		# 0 = sb1t
-	&pxor	("xmm0","xmm4");		# 0 = A
+	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
 	&movdqa	("xmm5",&QWP($k_sb2,$const));	# 4 : sb2u
-	&pshufb	("xmm5","xmm2");		# 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
-	&pshufb	("xmm2","xmm3");		# 2 = sb2t
-	&pxor	("xmm2","xmm5");		# 2 = 2A
 	&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
@@ -220,30 +221,30 @@ $k_dsbo=0x2c0;		# decryption sbox final output
 	&pxor	("xmm3","xmm0");		# 3 = 2A+B+D
 	&pshufb	("xmm0","xmm1");		# 0 = 2B+C
 	&and	($magic,0x30);			# ... mod 4
-	&pxor	("xmm0","xmm3");		# 0 = 2A+3B+C+D
 	&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
-	&movdqa	("xmm5",&QWP($k_inv+16,$const));# 2 : a/k
 	&pshufb	("xmm5","xmm0");		# 2 = a/k
-	&pxor	("xmm0","xmm1");		# 0 = j
 	&movdqa	("xmm3","xmm7");		# 3 : 1/i
+	&pxor	("xmm0","xmm1");		# 0 = j
 	&pshufb	("xmm3","xmm1");		# 3 = 1/i
-	&pxor	("xmm3","xmm5");		# 3 = iak = 1/i + a/k
 	&movdqa	("xmm4","xmm7");		# 4 : 1/j
+	&pxor	("xmm3","xmm5");		# 3 = iak = 1/i + a/k
 	&pshufb	("xmm4","xmm0");		# 4 = 1/j
-	&pxor	("xmm4","xmm5");		# 4 = jak = 1/j + a/k
 	&movdqa	("xmm2","xmm7");		# 2 : 1/iak
+	&pxor	("xmm4","xmm5");		# 4 = jak = 1/j + a/k
 	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
-	&pxor	("xmm2","xmm0");		# 2 = io
 	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
-	&movdqu	("xmm5",&QWP(0,$key));
+	&pxor	("xmm2","xmm0");		# 2 = io
 	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
+	&movdqu	("xmm5",&QWP(0,$key));
 	&pxor	("xmm3","xmm1");		# 3 = jo
 	&jnz	(&label("enc_loop"));
 
@@ -265,8 +266,8 @@ $k_dsbo=0x2c0;		# decryption sbox final output
 ##  Same API as encryption core.
 ##
 &function_begin_B("_vpaes_decrypt_core");
-	&mov	($round,&DWP(240,$key));
 	&lea	($base,&DWP($k_dsbd,$const));
+	&mov	($round,&DWP(240,$key));
 	&movdqa	("xmm1","xmm6");
 	&movdqa	("xmm2",&QWP($k_dipt-$k_dsbd,$base));
 	&pandn	("xmm1","xmm0");
@@ -292,62 +293,61 @@ $k_dsbo=0x2c0;		# decryption sbox final output
 ##  Inverse mix columns
 ##
 	&movdqa	("xmm4",&QWP(-0x20,$base));	# 4 : sb9u
+	&movdqa	("xmm1",&QWP(-0x10,$base));	# 0 : sb9t
 	&pshufb	("xmm4","xmm2");		# 4 = sb9u
-	&pxor	("xmm4","xmm0");
-	&movdqa	("xmm0",&QWP(-0x10,$base));	# 0 : sb9t
-	&pshufb	("xmm0","xmm3");		# 0 = sb9t
-	&pxor	("xmm0","xmm4");		# 0 = ch
-	&add	($key,16);			# next round key
-
-	&pshufb	("xmm0","xmm5");		# MC ch
+	&pshufb	("xmm1","xmm3");		# 0 = sb9t
+	&pxor	("xmm0","xmm4");
 	&movdqa	("xmm4",&QWP(0,$base));		# 4 : sbdu
-	&pshufb	("xmm4","xmm2");		# 4 = sbdu
-	&pxor	("xmm4","xmm0");		# 4 = ch
-	&movdqa	("xmm0",&QWP(0x10,$base));	# 0 : sbdt
-	&pshufb	("xmm0","xmm3");		# 0 = sbdt
-	&pxor	("xmm0","xmm4");		# 0 = ch
-	&sub	($round,1);			# nr--
+	&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,$base));	# 4 : sbbu
-	&pshufb	("xmm4","xmm2");		# 4 = sbbu
-	&pxor	("xmm4","xmm0");		# 4 = ch
-	&movdqa	("xmm0",&QWP(0x30,$base));	# 0 : sbbt
-	&pshufb	("xmm0","xmm3");		# 0 = sbbt
-	&pxor	("xmm0","xmm4");		# 0 = ch
+	&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,$base));	# 4 : sbeu
-	&pshufb	("xmm4","xmm2");		# 4 = sbeu
-	&pxor	("xmm4","xmm0");		# 4 = ch
-	&movdqa	("xmm0",&QWP(0x50,$base));	# 0 : sbet
-	&pshufb	("xmm0","xmm3");		# 0 = sbet
-	&pxor	("xmm0","xmm4");		# 0 = ch
+	&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
-	&psrld	("xmm1",4);			# 1 = i
 	&pand	("xmm0","xmm6");		# 0 = k
-	&movdqa	("xmm2",&QWP($k_inv+16,$const));# 2 : a/k
+	&psrld	("xmm1",4);			# 1 = i
 	&pshufb	("xmm2","xmm0");		# 2 = a/k
-	&pxor	("xmm0","xmm1");		# 0 = j
 	&movdqa	("xmm3","xmm7");		# 3 : 1/i
+	&pxor	("xmm0","xmm1");		# 0 = j
 	&pshufb	("xmm3","xmm1");		# 3 = 1/i
-	&pxor	("xmm3","xmm2");		# 3 = iak = 1/i + a/k
 	&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
-	&pxor	("xmm2","xmm0");		# 2 = io
 	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
+	&pxor	("xmm2","xmm0");		# 2 = io
 	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
-	&pxor	("xmm3","xmm1");		# 3 = jo
 	&movdqu	("xmm0",&QWP(0,$key));
+	&pxor	("xmm3","xmm1");		# 3 = jo
 	&jnz	(&label("dec_loop"));
 
 	# middle of last round
@@ -542,12 +542,12 @@ $k_dsbo=0x2c0;		# decryption sbox final output
 ##    %xmm0: b+c+d  b+c  b  a
 ##
 &function_begin_B("_vpaes_schedule_192_smear");
-	&pshufd	("xmm0","xmm6",0x80);		# d c 0 0 -> c 0 0 0
-	&pxor	("xmm6","xmm0");		# -> c+d c 0 0
+	&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");
-	&pxor	("xmm1","xmm1");
 	&movhlps("xmm6","xmm1");		# clobber low side with zeros
 	&ret	();
 &function_end_B("_vpaes_schedule_192_smear");
diff --git a/openssl/crypto/aes/asm/vpaes-x86_64.pl b/openssl/crypto/aes/asm/vpaes-x86_64.pl
index bd7f45b85..f2ef318fa 100644
--- a/openssl/crypto/aes/asm/vpaes-x86_64.pl
+++ b/openssl/crypto/aes/asm/vpaes-x86_64.pl
@@ -27,9 +27,10 @@
 #
 #		aes-x86_64.pl		vpaes-x86_64.pl
 #
-# Core 2(**)	30.5/43.7/14.3		21.8/25.7(***)
-# Nehalem	30.5/42.2/14.6		 9.8/11.8
-# Atom		63.9/79.0/32.1		64.0/84.8(***)
+# 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
@@ -40,7 +41,7 @@
 # (**)	"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 +40%/78% improvement on Core 2
+#	pshufb,	yet it's respectable +36%/62% improvement on Core 2
 #	(as implied, over "hyper-threading-safe" code path).
 #
 #						<appro@openssl.org>
@@ -95,8 +96,8 @@ _vpaes_encrypt_core:
 	movdqa	.Lk_ipt+16(%rip), %xmm0	# ipthi
 	pshufb	%xmm1,	%xmm0
 	pxor	%xmm5,	%xmm2
-	pxor	%xmm2,	%xmm0
 	add	\$16,	%r9
+	pxor	%xmm2,	%xmm0
 	lea	.Lk_mc_backward(%rip),%r10
 	jmp	.Lenc_entry
 
@@ -104,19 +105,19 @@ _vpaes_encrypt_core:
 .Lenc_loop:
 	# middle of middle round
 	movdqa  %xmm13,	%xmm4	# 4 : sb1u
-	pshufb  %xmm2,	%xmm4	# 4 = sb1u
-	pxor	%xmm5,	%xmm4	# 4 = sb1u + k
 	movdqa  %xmm12,	%xmm0	# 0 : sb1t
+	pshufb  %xmm2,	%xmm4	# 4 = sb1u
 	pshufb  %xmm3,	%xmm0	# 0 = sb1t
-	pxor	%xmm4,	%xmm0	# 0 = A
+	pxor	%xmm5,	%xmm4	# 4 = sb1u + k
 	movdqa  %xmm15,	%xmm5	# 4 : sb2u
-	pshufb	%xmm2,	%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
-	pxor	%xmm5,	%xmm2	# 2 = 2A
-	movdqa	(%r11,%r10), %xmm4		# .Lk_mc_backward[]
 	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
@@ -125,30 +126,30 @@ _vpaes_encrypt_core:
 	pxor	%xmm0,	%xmm3	# 3 = 2A+B+D
 	pshufb  %xmm1,	%xmm0	# 0 = 2B+C
 	and	\$0x30,	%r11	# ... mod 4
-	pxor	%xmm3,	%xmm0	# 0 = 2A+3B+C+D
 	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
-	movdqa	%xmm11, %xmm5	# 2 : a/k
 	pshufb  %xmm0,  %xmm5	# 2 = a/k
-	pxor	%xmm1,	%xmm0	# 0 = j
 	movdqa	%xmm10,	%xmm3  	# 3 : 1/i
+	pxor	%xmm1,	%xmm0	# 0 = j
 	pshufb  %xmm1, 	%xmm3  	# 3 = 1/i
-	pxor	%xmm5, 	%xmm3  	# 3 = iak = 1/i + a/k
 	movdqa	%xmm10,	%xmm4  	# 4 : 1/j
+	pxor	%xmm5, 	%xmm3  	# 3 = iak = 1/i + a/k
 	pshufb	%xmm0, 	%xmm4  	# 4 = 1/j
-	pxor	%xmm5, 	%xmm4  	# 4 = jak = 1/j + a/k
 	movdqa	%xmm10,	%xmm2  	# 2 : 1/iak
+	pxor	%xmm5, 	%xmm4  	# 4 = jak = 1/j + a/k
 	pshufb  %xmm3,	%xmm2  	# 2 = 1/iak
-	pxor	%xmm0, 	%xmm2  	# 2 = io
 	movdqa	%xmm10, %xmm3   # 3 : 1/jak
-	movdqu	(%r9),	%xmm5
+	pxor	%xmm0, 	%xmm2  	# 2 = io
 	pshufb  %xmm4,  %xmm3   # 3 = 1/jak
+	movdqu	(%r9),	%xmm5
 	pxor	%xmm1,  %xmm3   # 3 = jo
 	jnz	.Lenc_loop
 
@@ -201,62 +202,61 @@ _vpaes_decrypt_core:
 ##  Inverse mix columns
 ##
 	movdqa  -0x20(%r10),%xmm4	# 4 : sb9u
+	movdqa  -0x10(%r10),%xmm1	# 0 : sb9t
 	pshufb	%xmm2,	%xmm4		# 4 = sb9u
-	pxor	%xmm0,	%xmm4
-	movdqa  -0x10(%r10),%xmm0	# 0 : sb9t
-	pshufb	%xmm3,	%xmm0		# 0 = sb9t
-	pxor	%xmm4,	%xmm0		# 0 = ch
-	add	\$16, %r9		# next round key
-
-	pshufb	%xmm5,	%xmm0		# MC ch
+	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
-	pxor	%xmm0,	%xmm4		# 4 = ch
-	movdqa  0x10(%r10),%xmm0	# 0 : sbdt
-	pshufb	%xmm3,	%xmm0		# 0 = sbdt
-	pxor	%xmm4,	%xmm0		# 0 = ch
-	sub	\$1,%rax		# nr--
-	
 	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
-	pxor	%xmm0,	%xmm4		# 4 = ch
-	movdqa  0x30(%r10),%xmm0	# 0 : sbbt
-	pshufb	%xmm3,	%xmm0		# 0 = sbbt
-	pxor	%xmm4,	%xmm0		# 0 = ch
-	
 	pshufb	%xmm5,	%xmm0		# MC ch
+	pshufb	%xmm3,	%xmm1		# 0 = sbbt
+	pxor	%xmm4,	%xmm0		# 4 = ch
 	movdqa  0x40(%r10),%xmm4	# 4 : sbeu
-	pshufb	%xmm2,	%xmm4		# 4 = sbeu
-	pxor	%xmm0,	%xmm4		# 4 = ch
-	movdqa  0x50(%r10),%xmm0	# 0 : sbet
-	pshufb	%xmm3,	%xmm0		# 0 = sbet
-	pxor	%xmm4,	%xmm0		# 0 = ch
+	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
-	movdqa	%xmm11, %xmm2	# 2 : a/k
 	pshufb  %xmm0,  %xmm2	# 2 = a/k
-	pxor	%xmm1,	%xmm0	# 0 = j
 	movdqa	%xmm10,	%xmm3	# 3 : 1/i
+	pxor	%xmm1,	%xmm0	# 0 = j
 	pshufb  %xmm1, 	%xmm3	# 3 = 1/i
-	pxor	%xmm2, 	%xmm3	# 3 = iak = 1/i + a/k
 	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
-	pxor	%xmm0, 	%xmm2	# 2 = io
 	movdqa	%xmm10, %xmm3	# 3 : 1/jak
+	pxor	%xmm0, 	%xmm2	# 2 = io
 	pshufb  %xmm4,  %xmm3	# 3 = 1/jak
-	pxor	%xmm1,  %xmm3	# 3 = jo
 	movdqu	(%r9),	%xmm0
+	pxor	%xmm1,  %xmm3	# 3 = jo
 	jnz	.Ldec_loop
 
 	# middle of last round
@@ -464,12 +464,12 @@ _vpaes_schedule_core:
 .type	_vpaes_schedule_192_smear,\@abi-omnipotent
 .align	16
 _vpaes_schedule_192_smear:
-	pshufd	\$0x80,	%xmm6,	%xmm0	# d c 0 0 -> c 0 0 0
-	pxor	%xmm0,	%xmm6		# -> c+d c 0 0
+	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
-	pxor	%xmm1,	%xmm1
 	movhlps	%xmm1,	%xmm6		# clobber low side with zeros
 	ret
 .size	_vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear