x86[_64] assembly pack: add optimized AES-NI OCB subroutines.

Reviewed-by: Richard Levitte <levitte@openssl.org>

5 files changed, 2007 insertions, 90 deletions

diff --git a/crypto/aes/asm/aesni-x86.pl b/crypto/aes/asm/aesni-x86.pl
index 9b2e37a..536f035 100644
--- a/crypto/aes/asm/aesni-x86.pl
+++ b/crypto/aes/asm/aesni-x86.pl
@@ -43,16 +43,20 @@
 # 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.
 
+# November 2015
+#
+# Add aesni_ocb_[en|de]crypt.
+
 ######################################################################
 # Current large-block performance in cycles per byte processed with
 # 128-bit key (less is better).
 #
-#		CBC en-/decrypt	CTR	XTS	ECB
+#		CBC en-/decrypt	CTR	XTS	ECB	OCB
 # 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
-# Silvermont	5.77/3.56	3.67	4.03	3.46
-# Bulldozer	5.80/0.98	1.05	1.24	0.93
+# * Bridge	5.07/0.98	0.99	1.09	0.91	1.10
+# Haswell	4.44/0.80	0.97	1.03	0.72	0.76
+# Silvermont	5.77/3.56	3.67	4.03	3.46	4.03
+# Bulldozer	5.80/0.98	1.05	1.24	0.93	1.23
 
 $PREFIX="aesni";	# if $PREFIX is set to "AES", the script
 			# generates drop-in replacement for
@@ -1831,6 +1835,877 @@ if ($PREFIX eq "aesni") {
 	&mov	("esp",&DWP(16*7+4,"esp"));	# restore %esp
 &function_end("aesni_xts_decrypt");
 }
+
+######################################################################
+# void aesni_ocb_[en|de]crypt(const char *inp, char *out, size_t blocks,
+#	const AES_KEY *key, unsigned int start_block_num,
+#	unsigned char offset_i[16], const unsigned char L_[][16],
+#	unsigned char checksum[16]);
+#
+{
+# offsets within stack frame
+my $checksum = 16*6;
+my ($key_off,$rounds_off,$out_off,$end_off,$esp_off)=map(16*7+4*$_,(0..4));
+
+# reassigned registers
+my ($l_,$block,$i1,$i3,$i5) = ($rounds_,$key_,$rounds,$len,$out);
+# $l_, $blocks, $inp, $key are permanently allocated in registers;
+# remaining non-volatile ones are offloaded to stack, which even
+# stay invariant after written to stack.
+
+&function_begin("aesni_ocb_encrypt");
+	&mov	($rounds,&wparam(5));		# &offset_i
+	&mov	($rounds_,&wparam(7));		# &checksum
+
+	&mov	($inp,&wparam(0));
+	&mov	($out,&wparam(1));
+	&mov	($len,&wparam(2));
+	&mov	($key,&wparam(3));
+	&movdqu	($rndkey0,&QWP(0,$rounds));	# load offset_i
+	&mov	($block,&wparam(4));		# start_block_num
+	&movdqu	($rndkey1,&QWP(0,$rounds_));	# load checksum
+	&mov	($l_,&wparam(6));		# L_
+
+	&mov	($rounds,"esp");
+	&sub	("esp",$esp_off+4);		# alloca
+	&and	("esp",-16);			# align stack
+
+	&sub	($out,$inp);
+	&shl	($len,4);
+	&lea	($len,&DWP(-16*6,$inp,$len));	# end of input - 16*6
+	&mov	(&DWP($out_off,"esp"),$out);
+	&mov	(&DWP($end_off,"esp"),$len);
+	&mov	(&DWP($esp_off,"esp"),$rounds);
+
+	&mov	($rounds,&DWP(240,$key));
+
+	&test	($block,1);
+	&jnz	(&label("odd"));
+
+	&bsf		($i3,$block);
+	&add		($block,1);
+	&shl		($i3,4);
+	&movdqu		($inout5,&QWP(0,$l_,$i3));
+	&mov		($i3,$key);			# put aside key
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&lea		($inp,&DWP(16,$inp));
+
+	&pxor		($inout5,$rndkey0);		# ^ last offset_i
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$inout5);		# ^ offset_i
+
+	&movdqa		($inout4,$rndkey1);
+	if ($inline)
+	{   &aesni_inline_generate1("enc");	}
+	else
+	{   &call	("_aesni_encrypt1");	}
+
+	&xorps		($inout0,$inout5);		# ^ offset_i
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movdqa		($rndkey1,$inout4);		# pass the checksum
+
+	&movups		(&QWP(-16,$out,$inp),$inout0);	# store output
+
+	&mov		($rounds,&DWP(240,$i3));
+	&mov		($key,$i3);			# restore key
+	&mov		($len,&DWP($end_off,"esp"));
+
+&set_label("odd");
+	&shl		($rounds,4);
+	&mov		($out,16);
+	&sub		($out,$rounds);			# twisted rounds
+	&mov		(&DWP($key_off,"esp"),$key);
+	&lea		($key,&DWP(32,$key,$rounds));	# end of key schedule
+	&mov		(&DWP($rounds_off,"esp"),$out);
+
+	&cmp		($inp,$len);
+	&ja		(&label("short"));
+	&jmp		(&label("grandloop"));
+
+&set_label("grandloop",32);
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&lea		($i5,&DWP(5,$block));
+	&add		($block,6);
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&bsf		($i5,$i5);
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&shl		($i5,4);
+	&movdqu		($inout0,&QWP(0,$l_));
+	&movdqu		($inout1,&QWP(0,$l_,$i1));
+	&mov		($rounds,&DWP($rounds_off,"esp"));
+	&movdqa		($inout2,$inout0);
+	&movdqu		($inout3,&QWP(0,$l_,$i3));
+	&movdqa		($inout4,$inout0);
+	&movdqu		($inout5,&QWP(0,$l_,$i5));
+
+	&pxor		($inout0,$rndkey0);		# ^ last offset_i
+	&pxor		($inout1,$inout0);
+	&movdqa		(&QWP(16*0,"esp"),$inout0);
+	&pxor		($inout2,$inout1);
+	&movdqa		(&QWP(16*1,"esp"),$inout1);
+	&pxor		($inout3,$inout2);
+	&movdqa		(&QWP(16*2,"esp"),$inout2);
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*3,"esp"),$inout3);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*4,"esp"),$inout4);
+	&movdqa		(&QWP(16*5,"esp"),$inout5);
+
+	&$movekey	($rndkey0,&QWP(-48,$key,$rounds));
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&movdqu		($inout4,&QWP(16*4,$inp));
+	&movdqu		($inout5,&QWP(16*5,$inp));
+	&lea		($inp,&DWP(16*6,$inp));
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$rndkey0);		# ^ roundkey[0]
+	&pxor		($rndkey1,$inout1);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($rndkey1,$inout2);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($rndkey1,$inout3);
+	&pxor		($inout3,$rndkey0);
+	&pxor		($rndkey1,$inout4);
+	&pxor		($inout4,$rndkey0);
+	&pxor		($rndkey1,$inout5);
+	&pxor		($inout5,$rndkey0);
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+
+	&$movekey	($rndkey1,&QWP(-32,$key,$rounds));
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+	&pxor		($inout5,&QWP(16*5,"esp"));
+
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($inout1,$rndkey1);
+	&aesenc		($inout2,$rndkey1);
+	&aesenc		($inout3,$rndkey1);
+	&aesenc		($inout4,$rndkey1);
+	&aesenc		($inout5,$rndkey1);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&mov		($len,&DWP($end_off,"esp"));
+	&call		("_aesni_encrypt6_enter");
+
+	&movdqa		($rndkey0,&QWP(16*5,"esp"));	# pass last offset_i
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+	&pxor		($inout5,$rndkey0);
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+
+	&movdqu		(&QWP(-16*6,$out,$inp),$inout0);# store output
+	&movdqu		(&QWP(-16*5,$out,$inp),$inout1);
+	&movdqu		(&QWP(-16*4,$out,$inp),$inout2);
+	&movdqu		(&QWP(-16*3,$out,$inp),$inout3);
+	&movdqu		(&QWP(-16*2,$out,$inp),$inout4);
+	&movdqu		(&QWP(-16*1,$out,$inp),$inout5);
+	&cmp		($inp,$len);			# done yet?
+	&jb		(&label("grandloop"));
+
+&set_label("short");
+	&add		($len,16*6);
+	&sub		($len,$inp);
+	&jz		(&label("done"));
+
+	&cmp		($len,16*2);
+	&jb		(&label("one"));
+	&je		(&label("two"));
+
+	&cmp		($len,16*4);
+	&jb		(&label("three"));
+	&je		(&label("four"));
+
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&movdqu		($inout0,&QWP(0,$l_));
+	&movdqu		($inout1,&QWP(0,$l_,$i1));
+	&mov		($rounds,&DWP($rounds_off,"esp"));
+	&movdqa		($inout2,$inout0);
+	&movdqu		($inout3,&QWP(0,$l_,$i3));
+	&movdqa		($inout4,$inout0);
+
+	&pxor		($inout0,$rndkey0);		# ^ last offset_i
+	&pxor		($inout1,$inout0);
+	&movdqa		(&QWP(16*0,"esp"),$inout0);
+	&pxor		($inout2,$inout1);
+	&movdqa		(&QWP(16*1,"esp"),$inout1);
+	&pxor		($inout3,$inout2);
+	&movdqa		(&QWP(16*2,"esp"),$inout2);
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*3,"esp"),$inout3);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*4,"esp"),$inout4);
+
+	&$movekey	($rndkey0,&QWP(-48,$key,$rounds));
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&movdqu		($inout4,&QWP(16*4,$inp));
+	&pxor		($inout5,$inout5);
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$rndkey0);		# ^ roundkey[0]
+	&pxor		($rndkey1,$inout1);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($rndkey1,$inout2);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($rndkey1,$inout3);
+	&pxor		($inout3,$rndkey0);
+	&pxor		($rndkey1,$inout4);
+	&pxor		($inout4,$rndkey0);
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+
+	&$movekey	($rndkey1,&QWP(-32,$key,$rounds));
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&aesenc		($inout0,$rndkey1);
+	&aesenc		($inout1,$rndkey1);
+	&aesenc		($inout2,$rndkey1);
+	&aesenc		($inout3,$rndkey1);
+	&aesenc		($inout4,$rndkey1);
+	&aesenc		($inout5,$rndkey1);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_encrypt6_enter");
+
+	&movdqa		($rndkey0,&QWP(16*4,"esp"));	# pass last offset_i
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,$rndkey0);
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+
+	&movdqu		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&movdqu		(&QWP(16*1,$out,$inp),$inout1);
+	&movdqu		(&QWP(16*2,$out,$inp),$inout2);
+	&movdqu		(&QWP(16*3,$out,$inp),$inout3);
+	&movdqu		(&QWP(16*4,$out,$inp),$inout4);
+
+	&jmp		(&label("done"));
+
+&set_label("one",16);
+	&movdqu		($inout5,&QWP(0,$l_));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&mov		($rounds,&DWP(240,$key));
+
+	&pxor		($inout5,$rndkey0);		# ^ last offset_i
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$inout5);		# ^ offset_i
+
+	&movdqa		($inout4,$rndkey1);
+	&mov		($out,&DWP($out_off,"esp"));
+	if ($inline)
+	{   &aesni_inline_generate1("enc");	}
+	else
+	{   &call	("_aesni_encrypt1");	}
+
+	&xorps		($inout0,$inout5);		# ^ offset_i
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movdqa		($rndkey1,$inout4);		# pass the checksum
+	&movups		(&QWP(0,$out,$inp),$inout0);
+
+	&jmp		(&label("done"));
+
+&set_label("two",16);
+	&lea		($i1,&DWP(1,$block));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&bsf		($i1,$i1);
+	&shl		($i1,4);
+	&movdqu		($inout4,&QWP(0,$l_));
+	&movdqu		($inout5,&QWP(0,$l_,$i1));
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&pxor		($inout4,$rndkey0);		# ^ last offset_i
+	&pxor		($inout5,$inout4);
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$inout4);		# ^ offset_i
+	&pxor		($rndkey1,$inout1);
+	&pxor		($inout1,$inout5);
+
+	&movdqa		($inout3,$rndkey1)
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_encrypt2");
+
+	&xorps		($inout0,$inout4);		# ^ offset_i
+	&xorps		($inout1,$inout5);
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movdqa		($rndkey1,$inout3);		# pass the checksum
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+
+	&jmp		(&label("done"));
+
+&set_label("three",16);
+	&lea		($i1,&DWP(1,$block));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&bsf		($i1,$i1);
+	&shl		($i1,4);
+	&movdqu		($inout3,&QWP(0,$l_));
+	&movdqu		($inout4,&QWP(0,$l_,$i1));
+	&movdqa		($inout5,$inout3);
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&pxor		($inout3,$rndkey0);		# ^ last offset_i
+	&pxor		($inout4,$inout3);
+	&pxor		($inout5,$inout4);
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,$inout3);		# ^ offset_i
+	&pxor		($rndkey1,$inout1);
+	&pxor		($inout1,$inout4);
+	&pxor		($rndkey1,$inout2);
+	&pxor		($inout2,$inout5);
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_encrypt3");
+
+	&xorps		($inout0,$inout3);		# ^ offset_i
+	&xorps		($inout1,$inout4);
+	&xorps		($inout2,$inout5);
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+	&movups		(&QWP(16*2,$out,$inp),$inout2);
+
+	&jmp		(&label("done"));
+
+&set_label("four",16);
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&movdqu		($inout2,&QWP(0,$l_));
+	&movdqu		($inout3,&QWP(0,$l_,$i1));
+	&movdqa		($inout4,$inout2);
+	&movdqu		($inout5,&QWP(0,$l_,$i3));
+
+	&pxor		($inout2,$rndkey0);		# ^ last offset_i
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&pxor		($inout3,$inout2);
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*0,"esp"),$inout2);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*1,"esp"),$inout3);
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($rndkey1,$inout1);
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($rndkey1,$inout2);
+	&pxor		($inout2,$inout4);
+	&pxor		($rndkey1,$inout3);
+	&pxor		($inout3,$inout5);
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1)
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_encrypt4");
+
+	&xorps		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&xorps		($inout1,&QWP(16*1,"esp"));
+	&xorps		($inout2,$inout4);
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&xorps		($inout3,$inout5);
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movups		(&QWP(16*2,$out,$inp),$inout2);
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+	&movups		(&QWP(16*3,$out,$inp),$inout3);
+
+&set_label("done");
+	&mov	($key,&DWP($esp_off,"esp"));
+	&pxor	($inout0,$inout0);		# clear register bank
+	&pxor	($inout1,$inout1);
+	&movdqa	(&QWP(16*0,"esp"),$inout0);	# clear stack
+	&pxor	($inout2,$inout2);
+	&movdqa	(&QWP(16*1,"esp"),$inout0);
+	&pxor	($inout3,$inout3);
+	&movdqa	(&QWP(16*2,"esp"),$inout0);
+	&pxor	($inout4,$inout4);
+	&movdqa	(&QWP(16*3,"esp"),$inout0);
+	&pxor	($inout5,$inout5);
+	&movdqa	(&QWP(16*4,"esp"),$inout0);
+	&movdqa	(&QWP(16*5,"esp"),$inout0);
+	&movdqa	(&QWP(16*6,"esp"),$inout0);
+
+	&lea	("esp",&DWP(0,$key));
+	&mov	($rounds,&wparam(5));		# &offset_i
+	&mov	($rounds_,&wparam(7));		# &checksum
+	&movdqu	(&QWP(0,$rounds),$rndkey0);
+	&pxor	($rndkey0,$rndkey0);
+	&movdqu	(&QWP(0,$rounds_),$rndkey1);
+	&pxor	($rndkey1,$rndkey1);
+&function_end("aesni_ocb_encrypt");
+
+&function_begin("aesni_ocb_decrypt");
+	&mov	($rounds,&wparam(5));		# &offset_i
+	&mov	($rounds_,&wparam(7));		# &checksum
+
+	&mov	($inp,&wparam(0));
+	&mov	($out,&wparam(1));
+	&mov	($len,&wparam(2));
+	&mov	($key,&wparam(3));
+	&movdqu	($rndkey0,&QWP(0,$rounds));	# load offset_i
+	&mov	($block,&wparam(4));		# start_block_num
+	&movdqu	($rndkey1,&QWP(0,$rounds_));	# load checksum
+	&mov	($l_,&wparam(6));		# L_
+
+	&mov	($rounds,"esp");
+	&sub	("esp",$esp_off+4);		# alloca
+	&and	("esp",-16);			# align stack
+
+	&sub	($out,$inp);
+	&shl	($len,4);
+	&lea	($len,&DWP(-16*6,$inp,$len));	# end of input - 16*6
+	&mov	(&DWP($out_off,"esp"),$out);
+	&mov	(&DWP($end_off,"esp"),$len);
+	&mov	(&DWP($esp_off,"esp"),$rounds);
+
+	&mov	($rounds,&DWP(240,$key));
+
+	&test	($block,1);
+	&jnz	(&label("odd"));
+
+	&bsf		($i3,$block);
+	&add		($block,1);
+	&shl		($i3,4);
+	&movdqu		($inout5,&QWP(0,$l_,$i3));
+	&mov		($i3,$key);			# put aside key
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&lea		($inp,&DWP(16,$inp));
+
+	&pxor		($inout5,$rndkey0);		# ^ last offset_i
+	&pxor		($inout0,$inout5);		# ^ offset_i
+
+	&movdqa		($inout4,$rndkey1);
+	if ($inline)
+	{   &aesni_inline_generate1("dec");	}
+	else
+	{   &call	("_aesni_decrypt1");	}
+
+	&xorps		($inout0,$inout5);		# ^ offset_i
+	&movaps		($rndkey1,$inout4);		# pass the checksum
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&xorps		($rndkey1,$inout0);		# checksum
+	&movups		(&QWP(-16,$out,$inp),$inout0);	# store output
+
+	&mov		($rounds,&DWP(240,$i3));
+	&mov		($key,$i3);			# restore key
+	&mov		($len,&DWP($end_off,"esp"));
+
+&set_label("odd");
+	&shl		($rounds,4);
+	&mov		($out,16);
+	&sub		($out,$rounds);			# twisted rounds
+	&mov		(&DWP($key_off,"esp"),$key);
+	&lea		($key,&DWP(32,$key,$rounds));	# end of key schedule
+	&mov		(&DWP($rounds_off,"esp"),$out);
+
+	&cmp		($inp,$len);
+	&ja		(&label("short"));
+	&jmp		(&label("grandloop"));
+
+&set_label("grandloop",32);
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&lea		($i5,&DWP(5,$block));
+	&add		($block,6);
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&bsf		($i5,$i5);
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&shl		($i5,4);
+	&movdqu		($inout0,&QWP(0,$l_));
+	&movdqu		($inout1,&QWP(0,$l_,$i1));
+	&mov		($rounds,&DWP($rounds_off,"esp"));
+	&movdqa		($inout2,$inout0);
+	&movdqu		($inout3,&QWP(0,$l_,$i3));
+	&movdqa		($inout4,$inout0);
+	&movdqu		($inout5,&QWP(0,$l_,$i5));
+
+	&pxor		($inout0,$rndkey0);		# ^ last offset_i
+	&pxor		($inout1,$inout0);
+	&movdqa		(&QWP(16*0,"esp"),$inout0);
+	&pxor		($inout2,$inout1);
+	&movdqa		(&QWP(16*1,"esp"),$inout1);
+	&pxor		($inout3,$inout2);
+	&movdqa		(&QWP(16*2,"esp"),$inout2);
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*3,"esp"),$inout3);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*4,"esp"),$inout4);
+	&movdqa		(&QWP(16*5,"esp"),$inout5);
+
+	&$movekey	($rndkey0,&QWP(-48,$key,$rounds));
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&movdqu		($inout4,&QWP(16*4,$inp));
+	&movdqu		($inout5,&QWP(16*5,$inp));
+	&lea		($inp,&DWP(16*6,$inp));
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+	&pxor		($inout0,$rndkey0);		# ^ roundkey[0]
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($inout3,$rndkey0);
+	&pxor		($inout4,$rndkey0);
+	&pxor		($inout5,$rndkey0);
+
+	&$movekey	($rndkey1,&QWP(-32,$key,$rounds));
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+	&pxor		($inout5,&QWP(16*5,"esp"));
+
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&aesdec		($inout0,$rndkey1);
+	&aesdec		($inout1,$rndkey1);
+	&aesdec		($inout2,$rndkey1);
+	&aesdec		($inout3,$rndkey1);
+	&aesdec		($inout4,$rndkey1);
+	&aesdec		($inout5,$rndkey1);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&mov		($len,&DWP($end_off,"esp"));
+	&call		("_aesni_decrypt6_enter");
+
+	&movdqa		($rndkey0,&QWP(16*5,"esp"));	# pass last offset_i
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+	&pxor		($inout5,$rndkey0);
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&movdqu		(&QWP(-16*6,$out,$inp),$inout0);# store output
+	&pxor		($rndkey1,$inout1);
+	&movdqu		(&QWP(-16*5,$out,$inp),$inout1);
+	&pxor		($rndkey1,$inout2);
+	&movdqu		(&QWP(-16*4,$out,$inp),$inout2);
+	&pxor		($rndkey1,$inout3);
+	&movdqu		(&QWP(-16*3,$out,$inp),$inout3);
+	&pxor		($rndkey1,$inout4);
+	&movdqu		(&QWP(-16*2,$out,$inp),$inout4);
+	&pxor		($rndkey1,$inout5);
+	&movdqu		(&QWP(-16*1,$out,$inp),$inout5);
+	&cmp		($inp,$len);			# done yet?
+	&jb		(&label("grandloop"));
+
+&set_label("short");
+	&add		($len,16*6);
+	&sub		($len,$inp);
+	&jz		(&label("done"));
+
+	&cmp		($len,16*2);
+	&jb		(&label("one"));
+	&je		(&label("two"));
+
+	&cmp		($len,16*4);
+	&jb		(&label("three"));
+	&je		(&label("four"));
+
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&movdqu		($inout0,&QWP(0,$l_));
+	&movdqu		($inout1,&QWP(0,$l_,$i1));
+	&mov		($rounds,&DWP($rounds_off,"esp"));
+	&movdqa		($inout2,$inout0);
+	&movdqu		($inout3,&QWP(0,$l_,$i3));
+	&movdqa		($inout4,$inout0);
+
+	&pxor		($inout0,$rndkey0);		# ^ last offset_i
+	&pxor		($inout1,$inout0);
+	&movdqa		(&QWP(16*0,"esp"),$inout0);
+	&pxor		($inout2,$inout1);
+	&movdqa		(&QWP(16*1,"esp"),$inout1);
+	&pxor		($inout3,$inout2);
+	&movdqa		(&QWP(16*2,"esp"),$inout2);
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*3,"esp"),$inout3);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*4,"esp"),$inout4);
+
+	&$movekey	($rndkey0,&QWP(-48,$key,$rounds));
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&movdqu		($inout4,&QWP(16*4,$inp));
+	&pxor		($inout5,$inout5);
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+	&pxor		($inout0,$rndkey0);		# ^ roundkey[0]
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($inout3,$rndkey0);
+	&pxor		($inout4,$rndkey0);
+
+	&$movekey	($rndkey1,&QWP(-32,$key,$rounds));
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,&QWP(16*4,"esp"));
+
+	&$movekey	($rndkey0,&QWP(-16,$key,$rounds));
+	&aesdec		($inout0,$rndkey1);
+	&aesdec		($inout1,$rndkey1);
+	&aesdec		($inout2,$rndkey1);
+	&aesdec		($inout3,$rndkey1);
+	&aesdec		($inout4,$rndkey1);
+	&aesdec		($inout5,$rndkey1);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_decrypt6_enter");
+
+	&movdqa		($rndkey0,&QWP(16*4,"esp"));	# pass last offset_i
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,&QWP(16*2,"esp"));
+	&pxor		($inout3,&QWP(16*3,"esp"));
+	&pxor		($inout4,$rndkey0);
+
+	&pxor		($rndkey1,$inout0);		# checksum
+	&movdqu		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&pxor		($rndkey1,$inout1);
+	&movdqu		(&QWP(16*1,$out,$inp),$inout1);
+	&pxor		($rndkey1,$inout2);
+	&movdqu		(&QWP(16*2,$out,$inp),$inout2);
+	&pxor		($rndkey1,$inout3);
+	&movdqu		(&QWP(16*3,$out,$inp),$inout3);
+	&pxor		($rndkey1,$inout4);
+	&movdqu		(&QWP(16*4,$out,$inp),$inout4);
+
+	&jmp		(&label("done"));
+
+&set_label("one",16);
+	&movdqu		($inout5,&QWP(0,$l_));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&mov		($rounds,&DWP(240,$key));
+
+	&pxor		($inout5,$rndkey0);		# ^ last offset_i
+	&pxor		($inout0,$inout5);		# ^ offset_i
+
+	&movdqa		($inout4,$rndkey1);
+	&mov		($out,&DWP($out_off,"esp"));
+	if ($inline)
+	{   &aesni_inline_generate1("dec");	}
+	else
+	{   &call	("_aesni_decrypt1");	}
+
+	&xorps		($inout0,$inout5);		# ^ offset_i
+	&movaps		($rndkey1,$inout4);		# pass the checksum
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&xorps		($rndkey1,$inout0);		# checksum
+	&movups		(&QWP(0,$out,$inp),$inout0);
+
+	&jmp		(&label("done"));
+
+&set_label("two",16);
+	&lea		($i1,&DWP(1,$block));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&bsf		($i1,$i1);
+	&shl		($i1,4);
+	&movdqu		($inout4,&QWP(0,$l_));
+	&movdqu		($inout5,&QWP(0,$l_,$i1));
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&movdqa		($inout3,$rndkey1);
+	&pxor		($inout4,$rndkey0);		# ^ last offset_i
+	&pxor		($inout5,$inout4);
+
+	&pxor		($inout0,$inout4);		# ^ offset_i
+	&pxor		($inout1,$inout5);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_decrypt2");
+
+	&xorps		($inout0,$inout4);		# ^ offset_i
+	&xorps		($inout1,$inout5);
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&xorps		($inout3,$inout0);		# checksum
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&xorps		($inout3,$inout1);
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+	&movaps		($rndkey1,$inout3);		# pass the checksum
+
+	&jmp		(&label("done"));
+
+&set_label("three",16);
+	&lea		($i1,&DWP(1,$block));
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&bsf		($i1,$i1);
+	&shl		($i1,4);
+	&movdqu		($inout3,&QWP(0,$l_));
+	&movdqu		($inout4,&QWP(0,$l_,$i1));
+	&movdqa		($inout5,$inout3);
+
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+	&pxor		($inout3,$rndkey0);		# ^ last offset_i
+	&pxor		($inout4,$inout3);
+	&pxor		($inout5,$inout4);
+
+	&pxor		($inout0,$inout3);		# ^ offset_i
+	&pxor		($inout1,$inout4);
+	&pxor		($inout2,$inout5);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_decrypt3");
+
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+	&xorps		($inout0,$inout3);		# ^ offset_i
+	&xorps		($inout1,$inout4);
+	&xorps		($inout2,$inout5);
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&pxor		($rndkey1,$inout0);		# checksum
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+	&pxor		($rndkey1,$inout1);
+	&movups		(&QWP(16*2,$out,$inp),$inout2);
+	&pxor		($rndkey1,$inout2);
+
+	&jmp		(&label("done"));
+
+&set_label("four",16);
+	&lea		($i1,&DWP(1,$block));
+	&lea		($i3,&DWP(3,$block));
+	&bsf		($i1,$i1);
+	&bsf		($i3,$i3);
+	&mov		($key,&DWP($key_off,"esp"));	# restore key
+	&shl		($i1,4);
+	&shl		($i3,4);
+	&movdqu		($inout2,&QWP(0,$l_));
+	&movdqu		($inout3,&QWP(0,$l_,$i1));
+	&movdqa		($inout4,$inout2);
+	&movdqu		($inout5,&QWP(0,$l_,$i3));
+
+	&pxor		($inout2,$rndkey0);		# ^ last offset_i
+	&movdqu		($inout0,&QWP(16*0,$inp));	# load input
+	&pxor		($inout3,$inout2);
+	&movdqu		($inout1,&QWP(16*1,$inp));
+	&pxor		($inout4,$inout3);
+	&movdqa		(&QWP(16*0,"esp"),$inout2);
+	&pxor		($inout5,$inout4);
+	&movdqa		(&QWP(16*1,"esp"),$inout3);
+	&movdqu		($inout2,&QWP(16*2,$inp));
+	&movdqu		($inout3,&QWP(16*3,$inp));
+	&mov		($rounds,&DWP(240,$key));
+
+	&movdqa		(&QWP($checksum,"esp"),$rndkey1);
+	&pxor		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&pxor		($inout1,&QWP(16*1,"esp"));
+	&pxor		($inout2,$inout4);
+	&pxor		($inout3,$inout5);
+
+	&mov		($out,&DWP($out_off,"esp"));
+	&call		("_aesni_decrypt4");
+
+	&movdqa		($rndkey1,&QWP($checksum,"esp"));# pass the checksum
+	&xorps		($inout0,&QWP(16*0,"esp"));	# ^ offset_i
+	&xorps		($inout1,&QWP(16*1,"esp"));
+	&xorps		($inout2,$inout4);
+	&movups		(&QWP(16*0,$out,$inp),$inout0);	# store output
+	&pxor		($rndkey1,$inout0);		# checksum
+	&xorps		($inout3,$inout5);
+	&movups		(&QWP(16*1,$out,$inp),$inout1);
+	&pxor		($rndkey1,$inout1);
+	&movdqa		($rndkey0,$inout5);		# pass last offset_i
+	&movups		(&QWP(16*2,$out,$inp),$inout2);
+	&pxor		($rndkey1,$inout2);
+	&movups		(&QWP(16*3,$out,$inp),$inout3);
+	&pxor		($rndkey1,$inout3);
+
+&set_label("done");
+	&mov	($key,&DWP($esp_off,"esp"));
+	&pxor	($inout0,$inout0);		# clear register bank
+	&pxor	($inout1,$inout1);
+	&movdqa	(&QWP(16*0,"esp"),$inout0);	# clear stack
+	&pxor	($inout2,$inout2);
+	&movdqa	(&QWP(16*1,"esp"),$inout0);
+	&pxor	($inout3,$inout3);
+	&movdqa	(&QWP(16*2,"esp"),$inout0);
+	&pxor	($inout4,$inout4);
+	&movdqa	(&QWP(16*3,"esp"),$inout0);
+	&pxor	($inout5,$inout5);
+	&movdqa	(&QWP(16*4,"esp"),$inout0);
+	&movdqa	(&QWP(16*5,"esp"),$inout0);
+	&movdqa	(&QWP(16*6,"esp"),$inout0);
+
+	&lea	("esp",&DWP(0,$key));
+	&mov	($rounds,&wparam(5));		# &offset_i
+	&mov	($rounds_,&wparam(7));		# &checksum
+	&movdqu	(&QWP(0,$rounds),$rndkey0);
+	&pxor	($rndkey0,$rndkey0);
+	&movdqu	(&QWP(0,$rounds_),$rndkey1);
+	&pxor	($rndkey1,$rndkey1);
+&function_end("aesni_ocb_decrypt");
+}
 }
 
 ######################################################################
@@ -2419,7 +3294,7 @@ if ($PREFIX eq "aesni") {
 	&pxor		("xmm3","xmm3");
 	&aesenclast	("xmm2","xmm3");
 
-	&movdqa		("xmm3","xmm1")
+	&movdqa		("xmm3","xmm1");
 	&pslldq		("xmm1",4);
 	&pxor		("xmm3","xmm1");
 	&pslldq		("xmm1",4);
diff --git a/crypto/aes/asm/aesni-x86_64.pl b/crypto/aes/asm/aesni-x86_64.pl
index 6037e9e..6e41a1ae 100644
--- a/crypto/aes/asm/aesni-x86_64.pl
+++ b/crypto/aes/asm/aesni-x86_64.pl
@@ -157,17 +157,22 @@
 # 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.
 
+# November 2015
+#
+# Add aesni_ocb_[en|de]crypt. 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
+#		CBC en-/decrypt	CTR	XTS	ECB	OCB
 # 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
+# * Bridge	5.07/0.74	0.75	0.90	0.85	0.98
+# Haswell	4.44/0.63	0.63	0.73	0.63	0.70
 # Skylake	2.62/0.63	0.63	0.63	0.63
-# Silvermont	5.75/3.54	3.56	4.12	3.87(*)
-# Bulldozer	5.77/0.70	0.72	0.90	0.70
+# Silvermont	5.75/3.54	3.56	4.12	3.87(*)	4.11
+# Bulldozer	5.77/0.70	0.72	0.90	0.70	0.95
 #
 # (*)	Atom Silvermont ECB result is suboptimal because of penalties
 #	incurred by operations on %xmm8-15. As ECB is not considered
@@ -2709,6 +2714,925 @@ $code.=<<___;
 	ret
 .size	aesni_xts_decrypt,.-aesni_xts_decrypt
 ___
+}
+
+######################################################################
+# void aesni_ocb_[en|de]crypt(const char *inp, char *out, size_t blocks,
+#	const AES_KEY *key, unsigned int start_block_num,
+#	unsigned char offset_i[16], const unsigned char L_[][16],
+#	unsigned char checksum[16]);
+#
+{
+my @offset=map("%xmm$_",(10..15));
+my ($checksum,$rndkey0l)=("%xmm8","%xmm9");
+my ($block_num,$offset_p)=("%r8","%r9");		# 5th and 6th arguments
+my ($L_p,$checksum_p) = ("%rbx","%rbp");
+my ($i1,$i3,$i5) = ("%r12","%r13","%r14");
+my $seventh_arg = $win64 ? 56 : 8;
+my $blocks = $len;
+
+$code.=<<___;
+.globl	aesni_ocb_encrypt
+.type	aesni_ocb_encrypt,\@function,6
+.align	32
+aesni_ocb_encrypt:
+	lea	(%rsp),%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+___
+$code.=<<___ if ($win64);
+	lea	-0xa0(%rsp),%rsp
+	movaps	%xmm6,0x00(%rsp)		# offload everything
+	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	%xmm13,0x70(%rsp)
+	movaps	%xmm14,0x80(%rsp)
+	movaps	%xmm15,0x90(%rsp)
+.Locb_enc_body:
+___
+$code.=<<___;
+	mov	$seventh_arg(%rax),$L_p		# 7th argument
+	mov	$seventh_arg+8(%rax),$checksum_p# 8th argument
+
+	mov	240($key),$rnds_
+	mov	$key,$key_
+	shl	\$4,$rnds_
+	$movkey	($key),$rndkey0l		# round[0]
+	$movkey	16($key,$rnds_),$rndkey1	# round[last]
+
+	movdqu	($offset_p),@offset[5]		# load last offset_i
+	pxor	$rndkey1,$rndkey0l		# round[0] ^ round[last]
+	pxor	$rndkey1,@offset[5]		# offset_i ^ round[last]
+
+	mov	\$16+32,$rounds
+	lea	32($key_,$rnds_),$key
+	$movkey	16($key_),$rndkey1		# round[1]
+	sub	%r10,%rax			# twisted $rounds
+	mov	%rax,%r10			# backup twisted $rounds
+
+	movdqu	($L_p),@offset[0]		# L_0 for all odd-numbered blocks
+	movdqu	($checksum_p),$checksum		# load checksum
+
+	test	\$1,$block_num			# is first block number odd?
+	jnz	.Locb_enc_odd
+
+	bsf	$block_num,$i1
+	add	\$1,$block_num
+	shl	\$4,$i1
+	movdqu	($L_p,$i1),$inout5		# borrow
+	movdqu	($inp),$inout0
+	lea	16($inp),$inp
+
+	call	__ocb_encrypt1
+
+	movdqa	$inout5,@offset[5]
+	movups	$inout0,($out)
+	lea	16($out),$out
+	sub	\$1,$blocks
+	jz	.Locb_enc_done
+
+.Locb_enc_odd:
+	lea	1($block_num),$i1		# even-numbered blocks
+	lea	3($block_num),$i3
+	lea	5($block_num),$i5
+	lea	6($block_num),$block_num
+	bsf	$i1,$i1				# ntz(block)
+	bsf	$i3,$i3
+	bsf	$i5,$i5
+	shl	\$4,$i1				# ntz(block) -> table offset
+	shl	\$4,$i3
+	shl	\$4,$i5
+
+	sub	\$6,$blocks
+	jc	.Locb_enc_short
+	jmp	.Locb_enc_grandloop
+
+.align	32
+.Locb_enc_grandloop:
+	movdqu	`16*0`($inp),$inout0		# load input
+	movdqu	`16*1`($inp),$inout1
+	movdqu	`16*2`($inp),$inout2
+	movdqu	`16*3`($inp),$inout3
+	movdqu	`16*4`($inp),$inout4
+	movdqu	`16*5`($inp),$inout5
+	lea	`16*6`($inp),$inp
+
+	call	__ocb_encrypt6
+
+	movups	$inout0,`16*0`($out)		# store output
+	movups	$inout1,`16*1`($out)
+	movups	$inout2,`16*2`($out)
+	movups	$inout3,`16*3`($out)
+	movups	$inout4,`16*4`($out)
+	movups	$inout5,`16*5`($out)
+	lea	`16*6`($out),$out
+	sub	\$6,$blocks
+	jnc	.Locb_enc_grandloop
+
+.Locb_enc_short:
+	add	\$6,$blocks
+	jz	.Locb_enc_done
+
+	movdqu	`16*0`($inp),$inout0
+	cmp	\$2,$blocks
+	jb	.Locb_enc_one
+	movdqu	`16*1`($inp),$inout1
+	je	.Locb_enc_two
+
+	movdqu	`16*2`($inp),$inout2
+	cmp	\$4,$blocks
+	jb	.Locb_enc_three
+	movdqu	`16*3`($inp),$inout3
+	je	.Locb_enc_four
+
+	movdqu	`16*4`($inp),$inout4
+	pxor	$inout5,$inout5
+
+	call	__ocb_encrypt6
+
+	movdqa	@offset[4],@offset[5]
+	movups	$inout0,`16*0`($out)
+	movups	$inout1,`16*1`($out)
+	movups	$inout2,`16*2`($out)
+	movups	$inout3,`16*3`($out)
+	movups	$inout4,`16*4`($out)
+
+	jmp	.Locb_enc_done
+
+.align	16
+.Locb_enc_one:
+	movdqa	@offset[0],$inout5		# borrow
+
+	call	__ocb_encrypt1
+
+	movdqa	$inout5,@offset[5]
+	movups	$inout0,`16*0`($out)
+	jmp	.Locb_enc_done
+
+.align	16
+.Locb_enc_two:
+	pxor	$inout2,$inout2
+	pxor	$inout3,$inout3
+
+	call	__ocb_encrypt4
+
+	movdqa	@offset[1],@offset[5]
+	movups	$inout0,`16*0`($out)
+	movups	$inout1,`16*1`($out)
+
+	jmp	.Locb_enc_done
+
+.align	16
+.Locb_enc_three:
+	pxor	$inout3,$inout3
+
+	call	__ocb_encrypt4
+
+	movdqa	@offset[2],@offset[5]
+	movups	$inout0,`16*0`($out)
+	movups	$inout1,`16*1`($out)
+	movups	$inout2,`16*2`($out)
+
+	jmp	.Locb_enc_done
+
+.align	16
+.Locb_enc_four:
+	call	__ocb_encrypt4
+
+	movdqa	@offset[3],@offset[5]
+	movups	$inout0,`16*0`($out)
+	movups	$inout1,`16*1`($out)
+	movups	$inout2,`16*2`($out)
+	movups	$inout3,`16*3`($out)
+
+.Locb_enc_done:
+	pxor	$rndkey0,@offset[5]		# "remove" round[last]
+	movdqu	$checksum,($checksum_p)		# store checksum
+	movdqu	@offset[5],($offset_p)		# store last offset_i
+
+	xorps	%xmm0,%xmm0			# clear register bank
+	pxor	%xmm1,%xmm1
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+___
+$code.=<<___ if (!$win64);
+	pxor	%xmm6,%xmm6
+	pxor	%xmm7,%xmm7
+	pxor	%xmm8,%xmm8
+	pxor	%xmm9,%xmm9
+	pxor	%xmm10,%xmm10
+	pxor	%xmm11,%xmm11
+	pxor	%xmm12,%xmm12
+	pxor	%xmm13,%xmm13
+	pxor	%xmm14,%xmm14
+	pxor	%xmm15,%xmm15
+___
+$code.=<<___ if ($win64);
+	movaps	0x00(%rsp),%xmm6
+	movaps	%xmm0,0x00(%rsp)		# clear stack
+	movaps	0x10(%rsp),%xmm7
+	movaps	%xmm0,0x10(%rsp)
+	movaps	0x20(%rsp),%xmm8
+	movaps	%xmm0,0x20(%rsp)
+	movaps	0x30(%rsp),%xmm9
+	movaps	%xmm0,0x30(%rsp)
+	movaps	0x40(%rsp),%xmm10
+	movaps	%xmm0,0x40(%rsp)
+	movaps	0x50(%rsp),%xmm11
+	movaps	%xmm0,0x50(%rsp)
+	movaps	0x60(%rsp),%xmm12
+	movaps	%xmm0,0x60(%rsp)
+	movaps	0x70(%rsp),%xmm13
+	movaps	%xmm0,0x70(%rsp)
+	movaps	0x80(%rsp),%xmm14
+	movaps	%xmm0,0x80(%rsp)
+	movaps	0x90(%rsp),%xmm15
+	movaps	%xmm0,0x90(%rsp)
+	lea	0xa0+0x28(%rsp),%rax
+.Locb_enc_pop:
+	lea	0xa0(%rsp),%rsp
+___
+$code.=<<___;
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+.Locb_enc_epilogue:
+	ret
+.size	aesni_ocb_encrypt,.-aesni_ocb_encrypt
+
+.type	__ocb_encrypt6,\@abi-omnipotent
+.align	32
+__ocb_encrypt6:
+	 pxor		$rndkey0l,@offset[5]	# offset_i ^ round[0]
+	 movdqu		($L_p,$i1),@offset[1]
+	 movdqa		@offset[0],@offset[2]
+	 movdqu		($L_p,$i3),@offset[3]
+	 movdqa		@offset[0],@offset[4]
+	 pxor		@offset[5],@offset[0]
+	 movdqu		($L_p,$i5),@offset[5]
+	 pxor		@offset[0],@offset[1]
+	pxor		$inout0,$checksum	# accumulate checksum
+	pxor		@offset[0],$inout0	# input ^ round[0] ^ offset_i
+	 pxor		@offset[1],@offset[2]
+	pxor		$inout1,$checksum
+	pxor		@offset[1],$inout1
+	 pxor		@offset[2],@offset[3]
+	pxor		$inout2,$checksum
+	pxor		@offset[2],$inout2
+	 pxor		@offset[3],@offset[4]
+	pxor		$inout3,$checksum
+	pxor		@offset[3],$inout3
+	 pxor		@offset[4],@offset[5]
+	pxor		$inout4,$checksum
+	pxor		@offset[4],$inout4
+	pxor		$inout5,$checksum
+	pxor		@offset[5],$inout5
+	$movkey		32($key_),$rndkey0
+
+	lea		1($block_num),$i1	# even-numbered blocks
+	lea		3($block_num),$i3
+	lea		5($block_num),$i5
+	add		\$6,$block_num
+	 pxor		$rndkey0l,@offset[0]	# offset_i ^ round[last]
+	bsf		$i1,$i1			# ntz(block)
+	bsf		$i3,$i3
+	bsf		$i5,$i5
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	 pxor		$rndkey0l,@offset[1]
+	 pxor		$rndkey0l,@offset[2]
+	aesenc		$rndkey1,$inout4
+	 pxor		$rndkey0l,@offset[3]
+	 pxor		$rndkey0l,@offset[4]
+	aesenc		$rndkey1,$inout5
+	$movkey		48($key_),$rndkey1
+	 pxor		$rndkey0l,@offset[5]
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	$movkey		64($key_),$rndkey0
+	shl		\$4,$i1			# ntz(block) -> table offset
+	shl		\$4,$i3
+	jmp		.Locb_enc_loop6
+
+.align	32
+.Locb_enc_loop6:
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	aesenc		$rndkey0,$inout4
+	aesenc		$rndkey0,$inout5
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_enc_loop6
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	aesenc		$rndkey1,$inout4
+	aesenc		$rndkey1,$inout5
+	$movkey		16($key_),$rndkey1
+	shl		\$4,$i5
+
+	aesenclast	@offset[0],$inout0
+	movdqu		($L_p),@offset[0]	# L_0 for all odd-numbered blocks
+	mov		%r10,%rax		# restore twisted rounds
+	aesenclast	@offset[1],$inout1
+	aesenclast	@offset[2],$inout2
+	aesenclast	@offset[3],$inout3
+	aesenclast	@offset[4],$inout4
+	aesenclast	@offset[5],$inout5
+	ret
+.size	__ocb_encrypt6,.-__ocb_encrypt6
+
+.type	__ocb_encrypt4,\@abi-omnipotent
+.align	32
+__ocb_encrypt4:
+	 pxor		$rndkey0l,@offset[5]	# offset_i ^ round[0]
+	 movdqu		($L_p,$i1),@offset[1]
+	 movdqa		@offset[0],@offset[2]
+	 movdqu		($L_p,$i3),@offset[3]
+	 pxor		@offset[5],@offset[0]
+	 pxor		@offset[0],@offset[1]
+	pxor		$inout0,$checksum	# accumulate checksum
+	pxor		@offset[0],$inout0	# input ^ round[0] ^ offset_i
+	 pxor		@offset[1],@offset[2]
+	pxor		$inout1,$checksum
+	pxor		@offset[1],$inout1
+	 pxor		@offset[2],@offset[3]
+	pxor		$inout2,$checksum
+	pxor		@offset[2],$inout2
+	pxor		$inout3,$checksum
+	pxor		@offset[3],$inout3
+	$movkey		32($key_),$rndkey0
+
+	 pxor		$rndkey0l,@offset[0]	# offset_i ^ round[last]
+	 pxor		$rndkey0l,@offset[1]
+	 pxor		$rndkey0l,@offset[2]
+	 pxor		$rndkey0l,@offset[3]
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	$movkey		48($key_),$rndkey1
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	$movkey		64($key_),$rndkey0
+	jmp		.Locb_enc_loop4
+
+.align	32
+.Locb_enc_loop4:
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesenc		$rndkey0,$inout0
+	aesenc		$rndkey0,$inout1
+	aesenc		$rndkey0,$inout2
+	aesenc		$rndkey0,$inout3
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_enc_loop4
+
+	aesenc		$rndkey1,$inout0
+	aesenc		$rndkey1,$inout1
+	aesenc		$rndkey1,$inout2
+	aesenc		$rndkey1,$inout3
+	$movkey		16($key_),$rndkey1
+	mov		%r10,%rax		# restore twisted rounds
+
+	aesenclast	@offset[0],$inout0
+	aesenclast	@offset[1],$inout1
+	aesenclast	@offset[2],$inout2
+	aesenclast	@offset[3],$inout3
+	ret
+.size	__ocb_encrypt4,.-__ocb_encrypt4
+
+.type	__ocb_encrypt1,\@abi-omnipotent
+.align	32
+__ocb_encrypt1:
+	 pxor		@offset[5],$inout5	# offset_i
+	 pxor		$rndkey0l,$inout5	# offset_i ^ round[0]
+	pxor		$inout0,$checksum	# accumulate checksum
+	pxor		$inout5,$inout0		# input ^ round[0] ^ offset_i
+	$movkey		32($key_),$rndkey0
+
+	aesenc		$rndkey1,$inout0
+	$movkey		48($key_),$rndkey1
+	pxor		$rndkey0l,$inout5	# offset_i ^ round[last]
+
+	aesenc		$rndkey0,$inout0
+	$movkey		64($key_),$rndkey0
+	jmp		.Locb_enc_loop1
+
+.align	32
+.Locb_enc_loop1:
+	aesenc		$rndkey1,$inout0
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesenc		$rndkey0,$inout0
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_enc_loop1
+
+	aesenc		$rndkey1,$inout0
+	$movkey		16($key_),$rndkey1	# redundant in tail
+	mov		%r10,%rax		# restore twisted rounds
+
+	aesenclast	$inout5,$inout0
+	ret
+.size	__ocb_encrypt1,.-__ocb_encrypt1
+
+.globl	aesni_ocb_decrypt
+.type	aesni_ocb_decrypt,\@function,6
+.align	32
+aesni_ocb_decrypt:
+	lea	(%rsp),%rax
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+___
+$code.=<<___ if ($win64);
+	lea	-0xa0(%rsp),%rsp
+	movaps	%xmm6,0x00(%rsp)		# offload everything
+	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	%xmm13,0x70(%rsp)
+	movaps	%xmm14,0x80(%rsp)
+	movaps	%xmm15,0x90(%rsp)
+.Locb_dec_body:
+___
+$code.=<<___;
+	mov	$seventh_arg(%rax),$L_p		# 7th argument
+	mov	$seventh_arg+8(%rax),$checksum_p# 8th argument
+
+	mov	240($key),$rnds_
+	mov	$key,$key_
+	shl	\$4,$rnds_
+	$movkey	($key),$rndkey0l		# round[0]
+	$movkey	16($key,$rnds_),$rndkey1	# round[last]
+
+	movdqu	($offset_p),@offset[5]		# load last offset_i
+	pxor	$rndkey1,$rndkey0l		# round[0] ^ round[last]
+	pxor	$rndkey1,@offset[5]		# offset_i ^ round[last]
+
+	mov	\$16+32,$rounds
+	lea	32($key_,$rnds_),$key
+	$movkey	16($key_),$rndkey1		# round[1]
+	sub	%r10,%rax			# twisted $rounds
+	mov	%rax,%r10			# backup twisted $rounds
+
+	movdqu	($L_p),@offset[0]		# L_0 for all odd-numbered blocks
+	movdqu	($checksum_p),$checksum		# load checksum
+
+	test	\$1,$block_num			# is first block number odd?
+	jnz	.Locb_dec_odd
+
+	bsf	$block_num,$i1
+	add	\$1,$block_num
+	shl	\$4,$i1
+	movdqu	($L_p,$i1),$inout5		# borrow
+	movdqu	($inp),$inout0
+	lea	16($inp),$inp
+
+	call	__ocb_decrypt1
+
+	movdqa	$inout5,@offset[5]
+	movups	$inout0,($out)
+	xorps	$inout0,$checksum		# accumulate checksum
+	lea	16($out),$out
+	sub	\$1,$blocks
+	jz	.Locb_dec_done
+
+.Locb_dec_odd:
+	lea	1($block_num),$i1		# even-numbered blocks
+	lea	3($block_num),$i3
+	lea	5($block_num),$i5
+	lea	6($block_num),$block_num
+	bsf	$i1,$i1				# ntz(block)
+	bsf	$i3,$i3
+	bsf	$i5,$i5
+	shl	\$4,$i1				# ntz(block) -> table offset
+	shl	\$4,$i3
+	shl	\$4,$i5
+
+	sub	\$6,$blocks
+	jc	.Locb_dec_short
+	jmp	.Locb_dec_grandloop
+
+.align	32
+.Locb_dec_grandloop:
+	movdqu	`16*0`($inp),$inout0		# load input
+	movdqu	`16*1`($inp),$inout1
+	movdqu	`16*2`($inp),$inout2
+	movdqu	`16*3`($inp),$inout3
+	movdqu	`16*4`($inp),$inout4
+	movdqu	`16*5`($inp),$inout5
+	lea	`16*6`($inp),$inp
+
+	call	__ocb_decrypt6
+
+	movups	$inout0,`16*0`($out)		# store output
+	pxor	$inout0,$checksum		# accumulate checksum
+	movups	$inout1,`16*1`($out)
+	pxor	$inout1,$checksum
+	movups	$inout2,`16*2`($out)
+	pxor	$inout2,$checksum
+	movups	$inout3,`16*3`($out)
+	pxor	$inout3,$checksum
+	movups	$inout4,`16*4`($out)
+	pxor	$inout4,$checksum
+	movups	$inout5,`16*5`($out)
+	pxor	$inout5,$checksum
+	lea	`16*6`($out),$out
+	sub	\$6,$blocks
+	jnc	.Locb_dec_grandloop
+
+.Locb_dec_short:
+	add	\$6,$blocks
+	jz	.Locb_dec_done
+
+	movdqu	`16*0`($inp),$inout0
+	cmp	\$2,$blocks
+	jb	.Locb_dec_one
+	movdqu	`16*1`($inp),$inout1
+	je	.Locb_dec_two
+
+	movdqu	`16*2`($inp),$inout2
+	cmp	\$4,$blocks
+	jb	.Locb_dec_three
+	movdqu	`16*3`($inp),$inout3
+	je	.Locb_dec_four
+
+	movdqu	`16*4`($inp),$inout4
+	pxor	$inout5,$inout5
+
+	call	__ocb_decrypt6
+
+	movdqa	@offset[4],@offset[5]
+	movups	$inout0,`16*0`($out)		# store output
+	pxor	$inout0,$checksum		# accumulate checksum
+	movups	$inout1,`16*1`($out)
+	pxor	$inout1,$checksum
+	movups	$inout2,`16*2`($out)
+	pxor	$inout2,$checksum
+	movups	$inout3,`16*3`($out)
+	pxor	$inout3,$checksum
+	movups	$inout4,`16*4`($out)
+	pxor	$inout4,$checksum
+
+	jmp	.Locb_dec_done
+
+.align	16
+.Locb_dec_one:
+	movdqa	@offset[0],$inout5		# borrow
+
+	call	__ocb_decrypt1
+
+	movdqa	$inout5,@offset[5]
+	movups	$inout0,`16*0`($out)		# store output
+	xorps	$inout0,$checksum		# accumulate checksum
+	jmp	.Locb_dec_done
+
+.align	16
+.Locb_dec_two:
+	pxor	$inout2,$inout2
+	pxor	$inout3,$inout3
+
+	call	__ocb_decrypt4
+
+	movdqa	@offset[1],@offset[5]
+	movups	$inout0,`16*0`($out)		# store output
+	xorps	$inout0,$checksum		# accumulate checksum
+	movups	$inout1,`16*1`($out)
+	xorps	$inout1,$checksum
+
+	jmp	.Locb_dec_done
+
+.align	16
+.Locb_dec_three:
+	pxor	$inout3,$inout3
+
+	call	__ocb_decrypt4
+
+	movdqa	@offset[2],@offset[5]
+	movups	$inout0,`16*0`($out)		# store output
+	xorps	$inout0,$checksum		# accumulate checksum
+	movups	$inout1,`16*1`($out)
+	xorps	$inout1,$checksum
+	movups	$inout2,`16*2`($out)
+	xorps	$inout2,$checksum
+
+	jmp	.Locb_dec_done
+
+.align	16
+.Locb_dec_four:
+	call	__ocb_decrypt4
+
+	movdqa	@offset[3],@offset[5]
+	movups	$inout0,`16*0`($out)		# store output
+	pxor	$inout0,$checksum		# accumulate checksum
+	movups	$inout1,`16*1`($out)
+	pxor	$inout1,$checksum
+	movups	$inout2,`16*2`($out)
+	pxor	$inout2,$checksum
+	movups	$inout3,`16*3`($out)
+	pxor	$inout3,$checksum
+
+.Locb_dec_done:
+	pxor	$rndkey0,@offset[5]		# "remove" round[last]
+	movdqu	$checksum,($checksum_p)		# store checksum
+	movdqu	@offset[5],($offset_p)		# store last offset_i
+
+	xorps	%xmm0,%xmm0			# clear register bank
+	pxor	%xmm1,%xmm1
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+___
+$code.=<<___ if (!$win64);
+	pxor	%xmm6,%xmm6
+	pxor	%xmm7,%xmm7
+	pxor	%xmm8,%xmm8
+	pxor	%xmm9,%xmm9
+	pxor	%xmm10,%xmm10
+	pxor	%xmm11,%xmm11
+	pxor	%xmm12,%xmm12
+	pxor	%xmm13,%xmm13
+	pxor	%xmm14,%xmm14
+	pxor	%xmm15,%xmm15
+___
+$code.=<<___ if ($win64);
+	movaps	0x00(%rsp),%xmm6
+	movaps	%xmm0,0x00(%rsp)		# clear stack
+	movaps	0x10(%rsp),%xmm7
+	movaps	%xmm0,0x10(%rsp)
+	movaps	0x20(%rsp),%xmm8
+	movaps	%xmm0,0x20(%rsp)
+	movaps	0x30(%rsp),%xmm9
+	movaps	%xmm0,0x30(%rsp)
+	movaps	0x40(%rsp),%xmm10
+	movaps	%xmm0,0x40(%rsp)
+	movaps	0x50(%rsp),%xmm11
+	movaps	%xmm0,0x50(%rsp)
+	movaps	0x60(%rsp),%xmm12
+	movaps	%xmm0,0x60(%rsp)
+	movaps	0x70(%rsp),%xmm13
+	movaps	%xmm0,0x70(%rsp)
+	movaps	0x80(%rsp),%xmm14
+	movaps	%xmm0,0x80(%rsp)
+	movaps	0x90(%rsp),%xmm15
+	movaps	%xmm0,0x90(%rsp)
+	lea	0xa0+0x28(%rsp),%rax
+.Locb_dec_pop:
+	lea	0xa0(%rsp),%rsp
+___
+$code.=<<___;
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+.Locb_dec_epilogue:
+	ret
+.size	aesni_ocb_decrypt,.-aesni_ocb_decrypt
+
+.type	__ocb_decrypt6,\@abi-omnipotent
+.align	32
+__ocb_decrypt6:
+	 pxor		$rndkey0l,@offset[5]	# offset_i ^ round[0]
+	 movdqu		($L_p,$i1),@offset[1]
+	 movdqa		@offset[0],@offset[2]
+	 movdqu		($L_p,$i3),@offset[3]
+	 movdqa		@offset[0],@offset[4]
+	 pxor		@offset[5],@offset[0]
+	 movdqu		($L_p,$i5),@offset[5]
+	 pxor		@offset[0],@offset[1]
+	pxor		@offset[0],$inout0	# input ^ round[0] ^ offset_i
+	 pxor		@offset[1],@offset[2]
+	pxor		@offset[1],$inout1
+	 pxor		@offset[2],@offset[3]
+	pxor		@offset[2],$inout2
+	 pxor		@offset[3],@offset[4]
+	pxor		@offset[3],$inout3
+	 pxor		@offset[4],@offset[5]
+	pxor		@offset[4],$inout4
+	pxor		@offset[5],$inout5
+	$movkey		32($key_),$rndkey0
+
+	lea		1($block_num),$i1	# even-numbered blocks
+	lea		3($block_num),$i3
+	lea		5($block_num),$i5
+	add		\$6,$block_num
+	 pxor		$rndkey0l,@offset[0]	# offset_i ^ round[last]
+	bsf		$i1,$i1			# ntz(block)
+	bsf		$i3,$i3
+	bsf		$i5,$i5
+
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	 pxor		$rndkey0l,@offset[1]
+	 pxor		$rndkey0l,@offset[2]
+	aesdec		$rndkey1,$inout4
+	 pxor		$rndkey0l,@offset[3]
+	 pxor		$rndkey0l,@offset[4]
+	aesdec		$rndkey1,$inout5
+	$movkey		48($key_),$rndkey1
+	 pxor		$rndkey0l,@offset[5]
+
+	aesdec		$rndkey0,$inout0
+	aesdec		$rndkey0,$inout1
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	aesdec		$rndkey0,$inout4
+	aesdec		$rndkey0,$inout5
+	$movkey		64($key_),$rndkey0
+	shl		\$4,$i1			# ntz(block) -> table offset
+	shl		\$4,$i3
+	jmp		.Locb_dec_loop6
+
+.align	32
+.Locb_dec_loop6:
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	aesdec		$rndkey1,$inout4
+	aesdec		$rndkey1,$inout5
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesdec		$rndkey0,$inout0
+	aesdec		$rndkey0,$inout1
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	aesdec		$rndkey0,$inout4
+	aesdec		$rndkey0,$inout5
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_dec_loop6
+
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	aesdec		$rndkey1,$inout4
+	aesdec		$rndkey1,$inout5
+	$movkey		16($key_),$rndkey1
+	shl		\$4,$i5
+
+	aesdeclast	@offset[0],$inout0
+	movdqu		($L_p),@offset[0]	# L_0 for all odd-numbered blocks
+	mov		%r10,%rax		# restore twisted rounds
+	aesdeclast	@offset[1],$inout1
+	aesdeclast	@offset[2],$inout2
+	aesdeclast	@offset[3],$inout3
+	aesdeclast	@offset[4],$inout4
+	aesdeclast	@offset[5],$inout5
+	ret
+.size	__ocb_decrypt6,.-__ocb_decrypt6
+
+.type	__ocb_decrypt4,\@abi-omnipotent
+.align	32
+__ocb_decrypt4:
+	 pxor		$rndkey0l,@offset[5]	# offset_i ^ round[0]
+	 movdqu		($L_p,$i1),@offset[1]
+	 movdqa		@offset[0],@offset[2]
+	 movdqu		($L_p,$i3),@offset[3]
+	 pxor		@offset[5],@offset[0]
+	 pxor		@offset[0],@offset[1]
+	pxor		@offset[0],$inout0	# input ^ round[0] ^ offset_i
+	 pxor		@offset[1],@offset[2]
+	pxor		@offset[1],$inout1
+	 pxor		@offset[2],@offset[3]
+	pxor		@offset[2],$inout2
+	pxor		@offset[3],$inout3
+	$movkey		32($key_),$rndkey0
+
+	 pxor		$rndkey0l,@offset[0]	# offset_i ^ round[last]
+	 pxor		$rndkey0l,@offset[1]
+	 pxor		$rndkey0l,@offset[2]
+	 pxor		$rndkey0l,@offset[3]
+
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	$movkey		48($key_),$rndkey1
+
+	aesdec		$rndkey0,$inout0
+	aesdec		$rndkey0,$inout1
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	$movkey		64($key_),$rndkey0
+	jmp		.Locb_dec_loop4
+
+.align	32
+.Locb_dec_loop4:
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesdec		$rndkey0,$inout0
+	aesdec		$rndkey0,$inout1
+	aesdec		$rndkey0,$inout2
+	aesdec		$rndkey0,$inout3
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_dec_loop4
+
+	aesdec		$rndkey1,$inout0
+	aesdec		$rndkey1,$inout1
+	aesdec		$rndkey1,$inout2
+	aesdec		$rndkey1,$inout3
+	$movkey		16($key_),$rndkey1
+	mov		%r10,%rax		# restore twisted rounds
+
+	aesdeclast	@offset[0],$inout0
+	aesdeclast	@offset[1],$inout1
+	aesdeclast	@offset[2],$inout2
+	aesdeclast	@offset[3],$inout3
+	ret
+.size	__ocb_decrypt4,.-__ocb_decrypt4
+
+.type	__ocb_decrypt1,\@abi-omnipotent
+.align	32
+__ocb_decrypt1:
+	 pxor		@offset[5],$inout5	# offset_i
+	 pxor		$rndkey0l,$inout5	# offset_i ^ round[0]
+	pxor		$inout5,$inout0		# input ^ round[0] ^ offset_i
+	$movkey		32($key_),$rndkey0
+
+	aesdec		$rndkey1,$inout0
+	$movkey		48($key_),$rndkey1
+	pxor		$rndkey0l,$inout5	# offset_i ^ round[last]
+
+	aesdec		$rndkey0,$inout0
+	$movkey		64($key_),$rndkey0
+	jmp		.Locb_dec_loop1
+
+.align	32
+.Locb_dec_loop1:
+	aesdec		$rndkey1,$inout0
+	$movkey		($key,%rax),$rndkey1
+	add		\$32,%rax
+
+	aesdec		$rndkey0,$inout0
+	$movkey		-16($key,%rax),$rndkey0
+	jnz		.Locb_dec_loop1
+
+	aesdec		$rndkey1,$inout0
+	$movkey		16($key_),$rndkey1	# redundant in tail
+	mov		%r10,%rax		# restore twisted rounds
+
+	aesdeclast	$inout5,$inout0
+	ret
+.size	__ocb_decrypt1,.-__ocb_decrypt1
+___
 } }}
 
 ########################################################################
@@ -3820,6 +4744,65 @@ ctr_xts_se_handler:
 
 	jmp	.Lcommon_rbp_tail
 .size	ctr_xts_se_handler,.-ctr_xts_se_handler
+
+.type	ocb_se_handler,\@abi-omnipotent
+.align	16
+ocb_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue lable
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lcommon_seh_tail
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lcommon_seh_tail
+
+	mov	8(%r11),%r10d		# HandlerData[2]
+	lea	(%rsi,%r10),%r10
+	cmp	%r10,%rbx		# context->Rip>=pop label
+	jae	.Locb_no_xmm
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	lea	(%rax),%rsi		# %xmm save area
+	lea	512($context),%rdi	# & context.Xmm6
+	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	0xa0+0x28(%rax),%rax
+
+.Locb_no_xmm:
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+
+	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
+
+	jmp	.Lcommon_seh_tail
+.size	ocb_se_handler,.-ocb_se_handler
 ___
 $code.=<<___;
 .type	cbc_se_handler,\@abi-omnipotent
@@ -3933,6 +4916,14 @@ $code.=<<___ if ($PREFIX eq "aesni");
 	.rva	.LSEH_begin_aesni_xts_decrypt
 	.rva	.LSEH_end_aesni_xts_decrypt
 	.rva	.LSEH_info_xts_dec
+
+	.rva	.LSEH_begin_aesni_ocb_encrypt
+	.rva	.LSEH_end_aesni_ocb_encrypt
+	.rva	.LSEH_info_ocb_enc
+
+	.rva	.LSEH_begin_aesni_ocb_decrypt
+	.rva	.LSEH_end_aesni_ocb_decrypt
+	.rva	.LSEH_info_ocb_dec
 ___
 $code.=<<___;
 	.rva	.LSEH_begin_${PREFIX}_cbc_encrypt
@@ -3974,6 +4965,18 @@ $code.=<<___ if ($PREFIX eq "aesni");
 	.byte	9,0,0,0
 	.rva	ctr_xts_se_handler
 	.rva	.Lxts_dec_body,.Lxts_dec_epilogue	# HandlerData[]
+.LSEH_info_ocb_enc:
+	.byte	9,0,0,0
+	.rva	ocb_se_handler
+	.rva	.Locb_enc_body,.Locb_enc_epilogue	# HandlerData[]
+	.rva	.Locb_enc_pop
+	.long	0
+.LSEH_info_ocb_dec:
+	.byte	9,0,0,0
+	.rva	ocb_se_handler
+	.rva	.Locb_dec_body,.Locb_dec_epilogue	# HandlerData[]
+	.rva	.Locb_dec_pop
+	.long	0
 ___
 $code.=<<___;
 .LSEH_info_cbc:
diff --git a/crypto/evp/e_aes.c b/crypto/evp/e_aes.c
index efa724a..c356c9a 100644
--- a/crypto/evp/e_aes.c
+++ b/crypto/evp/e_aes.c
@@ -461,6 +461,19 @@ static int aesni_ccm_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
                             const unsigned char *in, size_t len);
 
 #  ifndef OPENSSL_NO_OCB
+void aesni_ocb_encrypt(const unsigned char *in, unsigned char *out,
+                       size_t blocks, const void *key,
+                       size_t start_block_num,
+                       unsigned char offset_i[16],
+                       const unsigned char L_[][16],
+                       unsigned char checksum[16]);
+void aesni_ocb_decrypt(const unsigned char *in, unsigned char *out,
+                       size_t blocks, const void *key,
+                       size_t start_block_num,
+                       unsigned char offset_i[16],
+                       const unsigned char L_[][16],
+                       unsigned char checksum[16]);
+
 static int aesni_ocb_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
                               const unsigned char *iv, int enc)
 {
@@ -479,7 +492,9 @@ static int aesni_ocb_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
             if (!CRYPTO_ocb128_init(&octx->ocb,
                                     &octx->ksenc.ks, &octx->ksdec.ks,
                                     (block128_f) aesni_encrypt,
-                                    (block128_f) aesni_decrypt))
+                                    (block128_f) aesni_decrypt,
+                                    enc ? aesni_ocb_encrypt
+                                        : aesni_ocb_decrypt))
                 return 0;
         }
         while (0);
@@ -2348,7 +2363,8 @@ static int aes_ocb_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
                 if (!CRYPTO_ocb128_init(&octx->ocb,
                                         &octx->ksenc.ks, &octx->ksdec.ks,
                                         (block128_f) vpaes_encrypt,
-                                        (block128_f) vpaes_decrypt))
+                                        (block128_f) vpaes_decrypt,
+                                        NULL))
                     return 0;
                 break;
             }
@@ -2358,7 +2374,8 @@ static int aes_ocb_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
             if (!CRYPTO_ocb128_init(&octx->ocb,
                                     &octx->ksenc.ks, &octx->ksdec.ks,
                                     (block128_f) AES_encrypt,
-                                    (block128_f) AES_decrypt))
+                                    (block128_f) AES_decrypt,
+                                    NULL))
                 return 0;
         }
         while (0);
diff --git a/crypto/modes/modes_lcl.h b/crypto/modes/modes_lcl.h
index 2f61afe..071b014 100644
--- a/crypto/modes/modes_lcl.h
+++ b/crypto/modes/modes_lcl.h
@@ -164,6 +164,7 @@ struct ocb128_context {
     block128_f decrypt;
     void *keyenc;
     void *keydec;
+    ocb128_f stream;    /* direction dependent */
     /* Key dependent variables. Can be reused if key remains the same */
     size_t l_index;
     size_t max_l_index;
diff --git a/crypto/modes/ocb128.c b/crypto/modes/ocb128.c
index 3a3f7a8..c3daf7c 100644
--- a/crypto/modes/ocb128.c
+++ b/crypto/modes/ocb128.c
@@ -159,7 +159,7 @@ static OCB_BLOCK *ocb_lookup_l(OCB128_CONTEXT *ctx, size_t idx)
         ctx->max_l_index += (idx - ctx->max_l_index + 4) & ~3;
         ctx->l =
             OPENSSL_realloc(ctx->l, ctx->max_l_index * sizeof(OCB_BLOCK));
-        if (!ctx->l)
+        if (ctx->l == NULL)
             return NULL;
     }
     while (l_index < idx) {
@@ -172,34 +172,18 @@ static OCB_BLOCK *ocb_lookup_l(OCB128_CONTEXT *ctx, size_t idx)
 }
 
 /*
- * Encrypt a block from |in| and store the result in |out|
- */
-static void ocb_encrypt(OCB128_CONTEXT *ctx, OCB_BLOCK *in, OCB_BLOCK *out,
-                        void *keyenc)
-{
-    ctx->encrypt(in->c, out->c, keyenc);
-}
-
-/*
- * Decrypt a block from |in| and store the result in |out|
- */
-static void ocb_decrypt(OCB128_CONTEXT *ctx, OCB_BLOCK *in, OCB_BLOCK *out,
-                        void *keydec)
-{
-    ctx->decrypt(in->c, out->c, keydec);
-}
-
-/*
  * Create a new OCB128_CONTEXT
  */
 OCB128_CONTEXT *CRYPTO_ocb128_new(void *keyenc, void *keydec,
-                                  block128_f encrypt, block128_f decrypt)
+                                  block128_f encrypt, block128_f decrypt,
+                                  ocb128_f stream)
 {
     OCB128_CONTEXT *octx;
     int ret;
 
     if ((octx = OPENSSL_malloc(sizeof(*octx))) != NULL) {
-        ret = CRYPTO_ocb128_init(octx, keyenc, keydec, encrypt, decrypt);
+        ret = CRYPTO_ocb128_init(octx, keyenc, keydec, encrypt, decrypt,
+                                 stream);
         if (ret)
             return octx;
         OPENSSL_free(octx);
@@ -212,7 +196,8 @@ OCB128_CONTEXT *CRYPTO_ocb128_new(void *keyenc, void *keydec,
  * Initialise an existing OCB128_CONTEXT
  */
 int CRYPTO_ocb128_init(OCB128_CONTEXT *ctx, void *keyenc, void *keydec,
-                       block128_f encrypt, block128_f decrypt)
+                       block128_f encrypt, block128_f decrypt,
+                       ocb128_f stream)
 {
     memset(ctx, 0, sizeof(*ctx));
     ctx->l_index = 0;
@@ -228,11 +213,12 @@ int CRYPTO_ocb128_init(OCB128_CONTEXT *ctx, void *keyenc, void *keydec,
      */
     ctx->encrypt = encrypt;
     ctx->decrypt = decrypt;
+    ctx->stream = stream;
     ctx->keyenc = keyenc;
     ctx->keydec = keydec;
 
     /* L_* = ENCIPHER(K, zeros(128)) */
-    ocb_encrypt(ctx, &ctx->l_star, &ctx->l_star, ctx->keyenc);
+    ctx->encrypt(ctx->l_star.c, ctx->l_star.c, ctx->keyenc);
 
     /* L_$ = double(L_*) */
     ocb_double(&ctx->l_star, &ctx->l_dollar);
@@ -324,11 +310,10 @@ int CRYPTO_ocb128_setiv(OCB128_CONTEXT *ctx, const unsigned char *iv,
 int CRYPTO_ocb128_aad(OCB128_CONTEXT *ctx, const unsigned char *aad,
                       size_t len)
 {
-    u64 all_num_blocks, num_blocks;
-    u64 i;
+    u64 i, all_num_blocks;
+    size_t num_blocks, last_len;
     OCB_BLOCK tmp1;
     OCB_BLOCK tmp2;
-    int last_len;
 
     /* Calculate the number of blocks of AAD provided now, and so far */
     num_blocks = len / 16;
@@ -341,14 +326,14 @@ int CRYPTO_ocb128_aad(OCB128_CONTEXT *ctx, const unsigned char *aad,
 
         /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
         lookup = ocb_lookup_l(ctx, ocb_ntz(i));
-        if (!lookup)
+        if (lookup == NULL)
             return 0;
         ocb_block16_xor(&ctx->offset_aad, lookup, &ctx->offset_aad);
 
         /* Sum_i = Sum_{i-1} xor ENCIPHER(K, A_i xor Offset_i) */
         aad_block = (OCB_BLOCK *)(aad + ((i - ctx->blocks_hashed - 1) * 16));
         ocb_block16_xor(&ctx->offset_aad, aad_block, &tmp1);
-        ocb_encrypt(ctx, &tmp1, &tmp2, ctx->keyenc);
+        ctx->encrypt(tmp1.c, tmp2.c, ctx->keyenc);
         ocb_block16_xor(&ctx->sum, &tmp2, &ctx->sum);
     }
 
@@ -369,7 +354,7 @@ int CRYPTO_ocb128_aad(OCB128_CONTEXT *ctx, const unsigned char *aad,
         ocb_block16_xor(&ctx->offset_aad, &tmp1, &tmp2);
 
         /* Sum = Sum_m xor ENCIPHER(K, CipherInput) */
-        ocb_encrypt(ctx, &tmp2, &tmp1, ctx->keyenc);
+        ctx->encrypt(tmp2.c, tmp1.c, ctx->keyenc);
         ocb_block16_xor(&ctx->sum, &tmp1, &ctx->sum);
     }
 
@@ -386,12 +371,11 @@ int CRYPTO_ocb128_encrypt(OCB128_CONTEXT *ctx,
                           const unsigned char *in, unsigned char *out,
                           size_t len)
 {
-    u64 i;
-    u64 all_num_blocks, num_blocks;
+    u64 i, all_num_blocks;
+    size_t num_blocks, last_len;
     OCB_BLOCK tmp1;
     OCB_BLOCK tmp2;
     OCB_BLOCK pad;
-    int last_len;
 
     /*
      * Calculate the number of blocks of data to be encrypted provided now, and
@@ -400,28 +384,46 @@ int CRYPTO_ocb128_encrypt(OCB128_CONTEXT *ctx,
     num_blocks = len / 16;
     all_num_blocks = num_blocks + ctx->blocks_processed;
 
-    /* Loop through all full blocks to be encrypted */
-    for (i = ctx->blocks_processed + 1; i <= all_num_blocks; i++) {
-        OCB_BLOCK *lookup;
-        OCB_BLOCK *inblock;
-        OCB_BLOCK *outblock;
+    if (num_blocks && all_num_blocks == (size_t)all_num_blocks
+        && ctx->stream != NULL) {
+        size_t max_idx = 0, top = (size_t)all_num_blocks;
 
-        /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
-        lookup = ocb_lookup_l(ctx, ocb_ntz(i));
-        if (!lookup)
+        /*
+         * See how many L_{i} entries we need to process data at hand
+         * and pre-compute missing entries in the table [if any]...
+         */
+        while (top >>= 1)
+            max_idx++;
+        if (ocb_lookup_l(ctx, max_idx) == NULL)
             return 0;
-        ocb_block16_xor(&ctx->offset, lookup, &ctx->offset);
-
-        /* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i) */
-        inblock = (OCB_BLOCK *)(in + ((i - ctx->blocks_processed - 1) * 16));
-        ocb_block16_xor_misaligned(&ctx->offset, inblock, &tmp1);
-        /* Checksum_i = Checksum_{i-1} xor P_i */
-        ocb_block16_xor_misaligned(&ctx->checksum, inblock, &ctx->checksum);
-        ocb_encrypt(ctx, &tmp1, &tmp2, ctx->keyenc);
-        outblock =
-            (OCB_BLOCK *)(out + ((i - ctx->blocks_processed - 1) * 16));
-        ocb_block16_xor_misaligned(&ctx->offset, &tmp2, outblock);
 
+        ctx->stream(in, out, num_blocks, ctx->keyenc,
+                    (size_t)ctx->blocks_processed + 1, ctx->offset.c,
+                    (const unsigned char (*)[16])ctx->l, ctx->checksum.c);
+    } else {
+        /* Loop through all full blocks to be encrypted */
+        for (i = ctx->blocks_processed + 1; i <= all_num_blocks; i++) {
+            OCB_BLOCK *lookup;
+            OCB_BLOCK *inblock;
+            OCB_BLOCK *outblock;
+
+            /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
+            lookup = ocb_lookup_l(ctx, ocb_ntz(i));
+            if (lookup == NULL)
+                return 0;
+            ocb_block16_xor(&ctx->offset, lookup, &ctx->offset);
+
+            /* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i) */
+            inblock =
+	        (OCB_BLOCK *)(in + ((i - ctx->blocks_processed - 1) * 16));
+            ocb_block16_xor_misaligned(&ctx->offset, inblock, &tmp1);
+            /* Checksum_i = Checksum_{i-1} xor P_i */
+            ocb_block16_xor_misaligned(&ctx->checksum, inblock, &ctx->checksum);
+            ctx->encrypt(tmp1.c, tmp2.c, ctx->keyenc);
+            outblock =
+                (OCB_BLOCK *)(out + ((i - ctx->blocks_processed - 1) * 16));
+            ocb_block16_xor_misaligned(&ctx->offset, &tmp2, outblock);
+        }
     }
 
     /*
@@ -435,7 +437,7 @@ int CRYPTO_ocb128_encrypt(OCB128_CONTEXT *ctx,
         ocb_block16_xor(&ctx->offset, &ctx->l_star, &ctx->offset);
 
         /* Pad = ENCIPHER(K, Offset_*) */
-        ocb_encrypt(ctx, &ctx->offset, &pad, ctx->keyenc);
+        ctx->encrypt(ctx->offset.c, pad.c, ctx->keyenc);
 
         /* C_* = P_* xor Pad[1..bitlen(P_*)] */
         ocb_block_xor(in + (len / 16) * 16, (unsigned char *)&pad, last_len,
@@ -461,12 +463,12 @@ int CRYPTO_ocb128_decrypt(OCB128_CONTEXT *ctx,
                           const unsigned char *in, unsigned char *out,
                           size_t len)
 {
-    u64 i;
-    u64 all_num_blocks, num_blocks;
+    u64 i, all_num_blocks;
+    size_t num_blocks, last_len;
     OCB_BLOCK tmp1;
     OCB_BLOCK tmp2;
     OCB_BLOCK pad;
-    int last_len;
+
     /*
      * Calculate the number of blocks of data to be decrypted provided now, and
      * so far
@@ -474,27 +476,46 @@ int CRYPTO_ocb128_decrypt(OCB128_CONTEXT *ctx,
     num_blocks = len / 16;
     all_num_blocks = num_blocks + ctx->blocks_processed;
 
-    /* Loop through all full blocks to be decrypted */
-    for (i = ctx->blocks_processed + 1; i <= all_num_blocks; i++) {
-        OCB_BLOCK *inblock;
-        OCB_BLOCK *outblock;
+    if (num_blocks && all_num_blocks == (size_t)all_num_blocks
+        && ctx->stream != NULL) {
+        size_t max_idx = 0, top = (size_t)all_num_blocks;
 
-        /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
-        OCB_BLOCK *lookup = ocb_lookup_l(ctx, ocb_ntz(i));
-        if (!lookup)
+        /*
+         * See how many L_{i} entries we need to process data at hand
+         * and pre-compute missing entries in the table [if any]...
+         */
+        while (top >>= 1)
+            max_idx++;
+        if (ocb_lookup_l(ctx, max_idx) == NULL)
             return 0;
-        ocb_block16_xor(&ctx->offset, lookup, &ctx->offset);
-
-        /* P_i = Offset_i xor DECIPHER(K, C_i xor Offset_i) */
-        inblock = (OCB_BLOCK *)(in + ((i - ctx->blocks_processed - 1) * 16));
-        ocb_block16_xor_misaligned(&ctx->offset, inblock, &tmp1);
-        ocb_decrypt(ctx, &tmp1, &tmp2, ctx->keydec);
-        outblock =
-            (OCB_BLOCK *)(out + ((i - ctx->blocks_processed - 1) * 16));
-        ocb_block16_xor_misaligned(&ctx->offset, &tmp2, outblock);
-
-        /* Checksum_i = Checksum_{i-1} xor P_i */
-        ocb_block16_xor_misaligned(&ctx->checksum, outblock, &ctx->checksum);
+
+        ctx->stream(in, out, num_blocks, ctx->keydec,
+                    (size_t)ctx->blocks_processed + 1, ctx->offset.c,
+                    (const unsigned char (*)[16])ctx->l, ctx->checksum.c);
+    } else {
+        /* Loop through all full blocks to be decrypted */
+        for (i = ctx->blocks_processed + 1; i <= all_num_blocks; i++) {
+            OCB_BLOCK *inblock;
+            OCB_BLOCK *outblock;
+
+            /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */
+            OCB_BLOCK *lookup = ocb_lookup_l(ctx, ocb_ntz(i));
+            if (lookup == NULL)
+                return 0;
+            ocb_block16_xor(&ctx->offset, lookup, &ctx->offset);
+
+            /* P_i = Offset_i xor DECIPHER(K, C_i xor Offset_i) */
+            inblock =
+                (OCB_BLOCK *)(in + ((i - ctx->blocks_processed - 1) * 16));
+            ocb_block16_xor_misaligned(&ctx->offset, inblock, &tmp1);
+            ctx->decrypt(tmp1.c, tmp2.c, ctx->keydec);
+            outblock =
+                (OCB_BLOCK *)(out + ((i - ctx->blocks_processed - 1) * 16));
+            ocb_block16_xor_misaligned(&ctx->offset, &tmp2, outblock);
+
+            /* Checksum_i = Checksum_{i-1} xor P_i */
+            ocb_block16_xor_misaligned(&ctx->checksum, outblock, &ctx->checksum);
+        }
     }
 
     /*
@@ -508,7 +529,7 @@ int CRYPTO_ocb128_decrypt(OCB128_CONTEXT *ctx,
         ocb_block16_xor(&ctx->offset, &ctx->l_star, &ctx->offset);
 
         /* Pad = ENCIPHER(K, Offset_*) */
-        ocb_encrypt(ctx, &ctx->offset, &pad, ctx->keyenc);
+        ctx->encrypt(ctx->offset.c, pad.c, ctx->keyenc);
 
         /* P_* = C_* xor Pad[1..bitlen(C_*)] */
         ocb_block_xor(in + (len / 16) * 16, (unsigned char *)&pad, last_len,
@@ -539,7 +560,7 @@ int CRYPTO_ocb128_finish(OCB128_CONTEXT *ctx, const unsigned char *tag,
      */
     ocb_block16_xor(&ctx->checksum, &ctx->offset, &tmp1);
     ocb_block16_xor(&tmp1, &ctx->l_dollar, &tmp2);
-    ocb_encrypt(ctx, &tmp2, &tmp1, ctx->keyenc);
+    ctx->encrypt(tmp2.c, tmp1.c, ctx->keyenc);
     ocb_block16_xor(&tmp1, &ctx->sum, &ctx->tag);
 
     if (len > 16 || len < 1) {