Blob Blame Raw
diff -up openssl-1.0.0-beta4/Configure.aesni openssl-1.0.0-beta4/Configure
--- openssl-1.0.0-beta4/Configure.aesni	2010-01-07 23:38:31.000000000 +0100
+++ openssl-1.0.0-beta4/Configure	2010-01-12 22:18:06.000000000 +0100
@@ -123,11 +123,11 @@ my $tlib="-lnsl -lsocket";
 my $bits1="THIRTY_TWO_BIT ";
 my $bits2="SIXTY_FOUR_BIT ";
 
-my $x86_asm="x86cpuid.o:bn-586.o co-586.o x86-mont.o:des-586.o crypt586.o:aes-586.o:bf-586.o:md5-586.o:sha1-586.o sha256-586.o sha512-586.o:cast-586.o:rc4-586.o:rmd-586.o:rc5-586.o:wp_block.o wp-mmx.o:cmll-x86.o";
+my $x86_asm="x86cpuid.o:bn-586.o co-586.o x86-mont.o:des-586.o crypt586.o:aes-586.o aesni-x86.o:bf-586.o:md5-586.o:sha1-586.o sha256-586.o sha512-586.o:cast-586.o:rc4-586.o:rmd-586.o:rc5-586.o:wp_block.o wp-mmx.o:cmll-x86.o";
 
 my $x86_elf_asm="$x86_asm:elf";
 
-my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o";
+my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o aesni-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o";
 my $ia64_asm="ia64cpuid.o:bn-ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o rc4_skey.o:::::void";
 my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o:des_enc-sparc.o fcrypt_b.o:aes_core.o aes_cbc.o aes-sparcv9.o:::sha1-sparcv9.o sha256-sparcv9.o sha512-sparcv9.o:::::::void";
 my $sparcv8_asm=":sparcv8.o:des_enc-sparc.o fcrypt_b.o:::::::::::void";
@@ -491,7 +491,7 @@ my %table=(
 #
 # Win64 targets, WIN64I denotes IA-64 and WIN64A - AMD64
 "VC-WIN64I","cl:-W3 -Gs0 -Gy -nologo -DOPENSSL_SYSNAME_WIN32 -DWIN32_LEAN_AND_MEAN -DL_ENDIAN -DUNICODE -D_UNICODE -D_CRT_SECURE_NO_DEPRECATE:::WIN64I::SIXTY_FOUR_BIT RC4_CHUNK_LL DES_INT EXPORT_VAR_AS_FN:ia64cpuid.o:ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o:::::::ias:win32",
-"VC-WIN64A","cl:-W3 -Gs0 -Gy -nologo -DOPENSSL_SYSNAME_WIN32 -DWIN32_LEAN_AND_MEAN -DL_ENDIAN -DUNICODE -D_UNICODE -D_CRT_SECURE_NO_DEPRECATE:::WIN64A::SIXTY_FOUR_BIT RC4_CHUNK_LL DES_INT EXPORT_VAR_AS_FN:x86_64cpuid.o:bn_asm.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:auto:win32",
+"VC-WIN64A","cl:-W3 -Gs0 -Gy -nologo -DOPENSSL_SYSNAME_WIN32 -DWIN32_LEAN_AND_MEAN -DL_ENDIAN -DUNICODE -D_UNICODE -D_CRT_SECURE_NO_DEPRECATE:::WIN64A::SIXTY_FOUR_BIT RC4_CHUNK_LL DES_INT EXPORT_VAR_AS_FN:x86_64cpuid.o:bn_asm.o x86_64-mont.o::aes-x86_64.o aesni-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:auto:win32",
 # x86 Win32 target defaults to ANSI API, if you want UNICODE, complement
 # 'perl Configure VC-WIN32' with '-DUNICODE -D_UNICODE'
 "VC-WIN32","cl:-W3 -WX -Gs0 -GF -Gy -nologo -DOPENSSL_SYSNAME_WIN32 -DWIN32_LEAN_AND_MEAN -DL_ENDIAN -D_CRT_SECURE_NO_DEPRECATE:::WIN32::BN_LLONG RC4_INDEX EXPORT_VAR_AS_FN ${x86_gcc_opts}:${x86_asm}:win32n:win32",
@@ -1410,6 +1410,7 @@ if ($rmd160_obj =~ /\.o$/)
 if ($aes_obj =~ /\.o$/)
 	{
 	$cflags.=" -DAES_ASM";
+	$aes_obj =~ s/\s*aesni\-x86\.o// if ($no_sse2);
 	}
 else	{
 	$aes_obj=$aes_enc;
diff -up openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86.pl.aesni openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86.pl
--- openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86.pl.aesni	2010-01-12 22:18:06.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86.pl	2010-01-12 22:18:06.000000000 +0100
@@ -0,0 +1,765 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements support for Intel AES-NI extension. In
+# OpenSSL context it's used with Intel engine, but can also be used as
+# drop-in replacement for crypto/aes/asm/aes-586.pl [see below for
+# details].
+
+$PREFIX="aesni";	# if $PREFIX is set to "AES", the script
+			# generates drop-in replacement for
+			# crypto/aes/asm/aes-586.pl:-)
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],$0);
+
+$movekey = eval($RREFIX eq "aseni" ? "*movaps" : "*movups");
+
+$len="eax";
+$rounds="ecx";
+$key="edx";
+$inp="esi";
+$out="edi";
+$rounds_="ebx";	# backup copy for $rounds
+$key_="ebp";	# backup copy for $key
+
+$inout0="xmm0";
+$inout1="xmm1";
+$inout2="xmm2";
+$rndkey0="xmm3";
+$rndkey1="xmm4";
+$ivec="xmm5";
+$in0="xmm6";
+$in1="xmm7";	$inout3="xmm7";
+
+# Inline version of internal aesni_[en|de]crypt1
+sub aesni_inline_generate1
+{ my $p=shift;
+
+    &$movekey		($rndkey0,&QWP(0,$key));
+    &$movekey		($rndkey1,&QWP(16,$key));
+    &lea		($key,&DWP(32,$key));
+    &pxor		($inout0,$rndkey0);
+    &set_label("${p}1_loop");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&dec		($rounds);
+	&$movekey	($rndkey1,&QWP(0,$key));
+	&lea		($key,&DWP(16,$key));
+    &jnz		(&label("${p}1_loop"));
+    eval"&aes${p}last	($inout0,$rndkey1)";
+}
+
+sub aesni_generate1	# fully unrolled loop
+{ my $p=shift;
+
+    &function_begin_B("_aesni_${p}rypt1");
+	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey1,&QWP(0x10,$key));
+	&cmp		($rounds,11);
+	&pxor		($inout0,$rndkey0);
+	&$movekey	($rndkey0,&QWP(0x20,$key));
+	&lea		($key,&DWP(0x30,$key));
+	&jb		(&label("${p}128"));
+	&lea		($key,&DWP(0x20,$key));
+	&je		(&label("${p}192"));
+	&lea		($key,&DWP(0x20,$key));
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(-0x40,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-0x30,$key));
+    &set_label("${p}192");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(-0x20,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(-0x10,$key));
+    &set_label("${p}128");
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(0x10,$key));
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0x20,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(0x30,$key));
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0x40,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(0x50,$key));
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(0x60,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&$movekey	($rndkey0,&QWP(0x70,$key));
+	eval"&aes${p}	($inout0,$rndkey1)";
+    eval"&aes${p}last	($inout0,$rndkey0)";
+    &ret();
+    &function_end_B("_aesni_${p}rypt1");
+}
+
+# void $PREFIX_encrypt (const void *inp,void *out,const AES_KEY *key);
+# &aesni_generate1("dec");
+&function_begin_B("${PREFIX}_encrypt");
+	&mov	("eax",&wparam(0));
+	&mov	($key,&wparam(2));
+	&movups	($inout0,&QWP(0,"eax"));
+	&mov	($rounds,&DWP(240,$key));
+	&mov	("eax",&wparam(1));
+	&aesni_inline_generate1("enc");	# &call	("_aesni_encrypt1");
+	&movups	(&QWP(0,"eax"),$inout0);
+	&ret	();
+&function_end_B("${PREFIX}_encrypt");
+
+# void $PREFIX_decrypt (const void *inp,void *out,const AES_KEY *key);
+# &aesni_generate1("dec");
+&function_begin_B("${PREFIX}_decrypt");
+	&mov	("eax",&wparam(0));
+	&mov	($key,&wparam(2));
+	&movups	($inout0,&QWP(0,"eax"));
+	&mov	($rounds,&DWP(240,$key));
+	&mov	("eax",&wparam(1));
+	&aesni_inline_generate1("dec");	# &call	("_aesni_decrypt1");
+	&movups	(&QWP(0,"eax"),$inout0);
+	&ret	();
+&function_end_B("${PREFIX}_decrypt");
+
+# _aesni_[en|de]crypt[34] are private interfaces, N denotes interleave
+# factor. Why 3x subroutine is used in loops? Even though aes[enc|dec]
+# latency is 6, it turned out that it can be scheduled only every
+# *second* cycle. Thus 3x interleave is 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. As soon
+# as/if Intel improves throughput by making it possible to schedule
+# the instructions in question *every* cycles I would have to
+# implement 6x interleave and use it in loop...
+sub aesni_generate3
+{ my $p=shift;
+
+    &function_begin_B("_aesni_${p}rypt3");
+	&$movekey	($rndkey0,&QWP(0,$key));
+	&shr		($rounds,1);
+	&$movekey	($rndkey1,&QWP(16,$key));
+	&lea		($key,&DWP(32,$key));
+	&pxor		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&jmp		(&label("${p}3_loop"));
+    &set_label("${p}3_loop",16);
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey0,&QWP(0,$key));
+	eval"&aes${p}	($inout1,$rndkey1)";
+	&dec		($rounds);
+	eval"&aes${p}	($inout2,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(16,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&lea		($key,&DWP(32,$key));
+	eval"&aes${p}	($inout1,$rndkey0)";
+	eval"&aes${p}	($inout2,$rndkey0)";
+	&jnz		(&label("${p}3_loop"));
+    eval"&aes${p}	($inout0,$rndkey1)";
+    &$movekey		($rndkey0,&QWP(0,$key));
+    eval"&aes${p}	($inout1,$rndkey1)";
+    eval"&aes${p}	($inout2,$rndkey1)";
+    eval"&aes${p}last	($inout0,$rndkey0)";
+    eval"&aes${p}last	($inout1,$rndkey0)";
+    eval"&aes${p}last	($inout2,$rndkey0)";
+    &ret();
+    &function_end_B("_aesni_${p}rypt3");
+}
+
+# 4x interleave is implemented to improve small block performance,
+# most notably [and naturally] 4 block by ~30%. One can argue that one
+# should have implemented 5x as well, but improvement  would be <20%,
+# so it's not worth it...
+sub aesni_generate4
+{ my $p=shift;
+
+    &function_begin_B("_aesni_${p}rypt4");
+	&$movekey	($rndkey0,&QWP(0,$key));
+	&$movekey	($rndkey1,&QWP(16,$key));
+	&shr		($rounds,1);
+	&lea		($key,&DWP(32,$key));
+	&pxor		($inout0,$rndkey0);
+	&pxor		($inout1,$rndkey0);
+	&pxor		($inout2,$rndkey0);
+	&pxor		($inout3,$rndkey0);
+	&jmp		(&label("${p}3_loop"));
+    &set_label("${p}3_loop",16);
+	eval"&aes${p}	($inout0,$rndkey1)";
+	&$movekey	($rndkey0,&QWP(0,$key));
+	eval"&aes${p}	($inout1,$rndkey1)";
+	&dec		($rounds);
+	eval"&aes${p}	($inout2,$rndkey1)";
+	eval"&aes${p}	($inout3,$rndkey1)";
+	&$movekey	($rndkey1,&QWP(16,$key));
+	eval"&aes${p}	($inout0,$rndkey0)";
+	&lea		($key,&DWP(32,$key));
+	eval"&aes${p}	($inout1,$rndkey0)";
+	eval"&aes${p}	($inout2,$rndkey0)";
+	eval"&aes${p}	($inout3,$rndkey0)";
+	&jnz		(&label("${p}3_loop"));
+    eval"&aes${p}	($inout0,$rndkey1)";
+    &$movekey		($rndkey0,&QWP(0,$key));
+    eval"&aes${p}	($inout1,$rndkey1)";
+    eval"&aes${p}	($inout2,$rndkey1)";
+    eval"&aes${p}	($inout3,$rndkey1)";
+    eval"&aes${p}last	($inout0,$rndkey0)";
+    eval"&aes${p}last	($inout1,$rndkey0)";
+    eval"&aes${p}last	($inout2,$rndkey0)";
+    eval"&aes${p}last	($inout3,$rndkey0)";
+    &ret();
+    &function_end_B("_aesni_${p}rypt4");
+}
+&aesni_generate3("enc") if ($PREFIX eq "aesni");
+&aesni_generate3("dec");
+&aesni_generate4("enc") if ($PREFIX eq "aesni");
+&aesni_generate4("dec");
+
+if ($PREFIX eq "aesni") {
+# void aesni_ecb_encrypt (const void *in, void *out,
+#                         size_t length, const AES_KEY *key,
+#                         int enc);
+&function_begin("aesni_ecb_encrypt");
+	&mov	($inp,&wparam(0));
+	&mov	($out,&wparam(1));
+	&mov	($len,&wparam(2));
+	&mov	($key,&wparam(3));
+	&mov	($rounds,&wparam(4));
+	&cmp	($len,16);
+	&jb	(&label("ecb_ret"));
+	&and	($len,-16);
+	&test	($rounds,$rounds)
+	&mov	($rounds,&DWP(240,$key));
+	&mov	($key_,$key);		# backup $key
+	&mov	($rounds_,$rounds);	# backup $rounds
+	&jz	(&label("ecb_decrypt"));
+
+	&sub	($len,0x40);
+	&jbe	(&label("ecb_enc_tail"));
+	&jmp	(&label("ecb_enc_loop3"));
+
+&set_label("ecb_enc_loop3",16);
+	&movups	($inout0,&QWP(0,$inp));
+	&movups	($inout1,&QWP(0x10,$inp));
+	&movups	($inout2,&QWP(0x20,$inp));
+	&call	("_aesni_encrypt3");
+	&sub	($len,0x30);
+	&lea	($inp,&DWP(0x30,$inp));
+	&lea	($out,&DWP(0x30,$out));
+	&movups	(&QWP(-0x30,$out),$inout0);
+	&mov	($key,$key_);		# restore $key
+	&movups	(&QWP(-0x20,$out),$inout1);
+	&mov	($rounds,$rounds_);	# restore $rounds
+	&movups	(&QWP(-0x10,$out),$inout2);
+	&ja	(&label("ecb_enc_loop3"));
+
+&set_label("ecb_enc_tail");
+	&add	($len,0x40);
+	&jz	(&label("ecb_ret"));
+
+	&cmp	($len,0x10);
+	&movups	($inout0,&QWP(0,$inp));
+	&je	(&label("ecb_enc_one"));
+	&cmp	($len,0x20);
+	&movups	($inout1,&QWP(0x10,$inp));
+	&je	(&label("ecb_enc_two"));
+	&cmp	($len,0x30);
+	&movups	($inout2,&QWP(0x20,$inp));
+	&je	(&label("ecb_enc_three"));
+	&movups	($inout3,&QWP(0x30,$inp));
+	&call	("_aesni_encrypt4");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+	&movups	(&QWP(0x30,$out),$inout3);
+	jmp	(&label("ecb_ret"));
+
+&set_label("ecb_enc_one",16);
+	&aesni_inline_generate1("enc");	# &call	("_aesni_encrypt1");
+	&movups	(&QWP(0,$out),$inout0);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_enc_two",16);
+	&call	("_aesni_encrypt3");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_enc_three",16);
+	&call	("_aesni_encrypt3");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_decrypt",16);
+	&sub	($len,0x40);
+	&jbe	(&label("ecb_dec_tail"));
+	&jmp	(&label("ecb_dec_loop3"));
+
+&set_label("ecb_dec_loop3",16);
+	&movups	($inout0,&QWP(0,$inp));
+	&movups	($inout1,&QWP(0x10,$inp));
+	&movups	($inout2,&QWP(0x20,$inp));
+	&call	("_aesni_decrypt3");
+	&sub	($len,0x30);
+	&lea	($inp,&DWP(0x30,$inp));
+	&lea	($out,&DWP(0x30,$out));
+	&movups	(&QWP(-0x30,$out),$inout0);
+	&mov	($key,$key_);		# restore $key
+	&movups	(&QWP(-0x20,$out),$inout1);
+	&mov	($rounds,$rounds_);	# restore $rounds
+	&movups	(&QWP(-0x10,$out),$inout2);
+	&ja	(&label("ecb_dec_loop3"));
+
+&set_label("ecb_dec_tail");
+	&add	($len,0x40);
+	&jz	(&label("ecb_ret"));
+
+	&cmp	($len,0x10);
+	&movups	($inout0,&QWP(0,$inp));
+	&je	(&label("ecb_dec_one"));
+	&cmp	($len,0x20);
+	&movups	($inout1,&QWP(0x10,$inp));
+	&je	(&label("ecb_dec_two"));
+	&cmp	($len,0x30);
+	&movups	($inout2,&QWP(0x20,$inp));
+	&je	(&label("ecb_dec_three"));
+	&movups	($inout3,&QWP(0x30,$inp));
+	&call	("_aesni_decrypt4");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+	&movups	(&QWP(0x30,$out),$inout3);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_dec_one",16);
+	&aesni_inline_generate1("dec");	# &call	("_aesni_decrypt3");
+	&movups	(&QWP(0,$out),$inout0);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_dec_two",16);
+	&call	("_aesni_decrypt3");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&jmp	(&label("ecb_ret"));
+
+&set_label("ecb_dec_three",16);
+	&call	("_aesni_decrypt3");
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+
+&set_label("ecb_ret");
+&function_end("aesni_ecb_encrypt");
+}
+
+# void $PREFIX_cbc_encrypt (const void *inp, void *out,
+#                           size_t length, const AES_KEY *key,
+#                           unsigned char *ivp,const int enc);
+&function_begin("${PREFIX}_cbc_encrypt");
+	&mov	($inp,&wparam(0));
+	&mov	($out,&wparam(1));
+	&mov	($len,&wparam(2));
+	&mov	($key,&wparam(3));
+	&test	($len,$len);
+	&mov	($key_,&wparam(4));
+	&jz	(&label("cbc_ret"));
+
+	&cmp	(&wparam(5),0);
+	&movups	($ivec,&QWP(0,$key_));	# load IV
+	&mov	($rounds,&DWP(240,$key));
+	&mov	($key_,$key);		# backup $key
+	&mov	($rounds_,$rounds);	# backup $rounds
+	&je	(&label("cbc_decrypt"));
+
+	&movaps	($inout0,$ivec);
+	&cmp	($len,16);
+	&jb	(&label("cbc_enc_tail"));
+	&sub	($len,16);
+	&jmp	(&label("cbc_enc_loop"));
+
+&set_label("cbc_enc_loop",16);
+	&movups	($ivec,&QWP(0,$inp));
+	&lea	($inp,&DWP(16,$inp));
+	&pxor	($inout0,$ivec);
+	&aesni_inline_generate1("enc");	# &call	("_aesni_encrypt3");
+	&sub	($len,16);
+	&lea	($out,&DWP(16,$out));
+	&mov	($rounds,$rounds_);	# restore $rounds
+	&mov	($key,$key_);		# restore $key
+	&movups	(&QWP(-16,$out),$inout0);
+	&jnc	(&label("cbc_enc_loop"));
+	&add	($len,16);
+	&jnz	(&label("cbc_enc_tail"));
+	&movaps	($ivec,$inout0);
+	&jmp	(&label("cbc_ret"));
+
+&set_label("cbc_enc_tail");
+	&mov	("ecx",$len);		# zaps $rounds
+	&data_word(0xA4F3F689);		# rep movsb
+	&mov	("ecx",16);		# zero tail
+	&sub	("ecx",$len);
+	&xor	("eax","eax");		# zaps $len
+	&data_word(0xAAF3F689);		# rep stosb
+	&lea	($out,&DWP(-16,$out));	# rewind $out by 1 block
+	&mov	($rounds,$rounds_);	# restore $rounds
+	&mov	($inp,$out);		# $inp and $out are the same
+	&mov	($key,$key_);		# restore $key
+	&jmp	(&label("cbc_enc_loop"));
+
+&set_label("cbc_decrypt",16);
+	&sub	($len,0x40);
+	&jbe	(&label("cbc_dec_tail"));
+	&jmp	(&label("cbc_dec_loop3"));
+
+&set_label("cbc_dec_loop3",16);
+	&movups	($inout0,&QWP(0,$inp));
+	&movups	($inout1,&QWP(0x10,$inp));
+	&movups	($inout2,&QWP(0x20,$inp));
+	&movaps	($in0,$inout0);
+	&movaps	($in1,$inout1);
+	&call	("_aesni_decrypt3");
+	&sub	($len,0x30);
+	&lea	($inp,&DWP(0x30,$inp));
+	&lea	($out,&DWP(0x30,$out));
+	&pxor	($inout0,$ivec);
+	&pxor	($inout1,$in0);
+	&movups	($ivec,&QWP(-0x10,$inp));
+	&pxor	($inout2,$in1);
+	&movups	(&QWP(-0x30,$out),$inout0);
+	&mov	($rounds,$rounds_)	# restore $rounds
+	&movups	(&QWP(-0x20,$out),$inout1);
+	&mov	($key,$key_);		# restore $key
+	&movups	(&QWP(-0x10,$out),$inout2);
+	&ja	(&label("cbc_dec_loop3"));
+
+&set_label("cbc_dec_tail");
+	&add	($len,0x40);
+	&jz	(&label("cbc_ret"));
+
+	&movups	($inout0,&QWP(0,$inp));
+	&cmp	($len,0x10);
+	&movaps	($in0,$inout0);
+	&jbe	(&label("cbc_dec_one"));
+	&movups	($inout1,&QWP(0x10,$inp));
+	&cmp	($len,0x20);
+	&movaps	($in1,$inout1);
+	&jbe	(&label("cbc_dec_two"));
+	&movups	($inout2,&QWP(0x20,$inp));
+	&cmp	($len,0x30);
+	&jbe	(&label("cbc_dec_three"));
+	&movups	($inout3,&QWP(0x30,$inp));
+	&call	("_aesni_decrypt4");
+	&movups	($rndkey0,&QWP(0x10,$inp));
+	&movups	($rndkey1,&QWP(0x20,$inp));
+	&pxor	($inout0,$ivec);
+	&pxor	($inout1,$in0);
+	&movups	($ivec,&QWP(0x30,$inp));
+	&movups	(&QWP(0,$out),$inout0);
+	&pxor	($inout2,$rndkey0);
+	&pxor	($inout3,$rndkey1);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movups	(&QWP(0x20,$out),$inout2);
+	&movaps	($inout0,$inout3);
+	&lea	($out,&DWP(0x30,$out));
+	&jmp	(&label("cbc_dec_tail_collected"));
+
+&set_label("cbc_dec_one");
+	&aesni_inline_generate1("dec");	# &call	("_aesni_decrypt3");
+	&pxor	($inout0,$ivec);
+	&movaps	($ivec,$in0);
+	&jmp	(&label("cbc_dec_tail_collected"));
+
+&set_label("cbc_dec_two");
+	&call	("_aesni_decrypt3");
+	&pxor	($inout0,$ivec);
+	&pxor	($inout1,$in0);
+	&movups	(&QWP(0,$out),$inout0);
+	&movaps	($inout0,$inout1);
+	&movaps	($ivec,$in1);
+	&lea	($out,&DWP(0x10,$out));
+	&jmp	(&label("cbc_dec_tail_collected"));
+
+&set_label("cbc_dec_three");
+	&call	("_aesni_decrypt3");
+	&pxor	($inout0,$ivec);
+	&pxor	($inout1,$in0);
+	&pxor	($inout2,$in1);
+	&movups	(&QWP(0,$out),$inout0);
+	&movups	(&QWP(0x10,$out),$inout1);
+	&movaps	($inout0,$inout2);
+	&movups	($ivec,&QWP(0x20,$inp));
+	&lea	($out,&DWP(0x20,$out));
+
+&set_label("cbc_dec_tail_collected");
+	&and	($len,15);
+	&jnz	(&label("cbc_dec_tail_partial"));
+	&movups	(&QWP(0,$out),$inout0);
+	&jmp	(&label("cbc_ret"));
+
+&set_label("cbc_dec_tail_partial");
+	&mov	($key_,"esp");
+	&sub	("esp",16);
+	&and	("esp",-16);
+	&movaps	(&QWP(0,"esp"),$inout0);
+	&mov	($inp,"esp");
+	&mov	("ecx",$len);
+	&data_word(0xA4F3F689);		# rep movsb
+	&mov	("esp",$key_);
+
+&set_label("cbc_ret");
+	&mov	($key_,&wparam(4));
+	&movups	(&QWP(0,$key_),$ivec);	# output IV
+&function_end("${PREFIX}_cbc_encrypt");
+
+# Mechanical port from aesni-x86_64.pl.
+#
+# _aesni_set_encrypt_key is private interface,
+# input:
+#	"eax"	const unsigned char *userKey
+#	$rounds	int bits
+#	$key	AES_KEY *key
+# output:
+#	"eax"	return code
+#	$round	rounds
+
+&function_begin_B("_aesni_set_encrypt_key");
+	&test	("eax","eax");
+	&jz	(&label("bad_pointer"));
+	&test	($key,$key);
+	&jz	(&label("bad_pointer"));
+
+	&movups	("xmm0",&QWP(0,"eax"));	# pull first 128 bits of *userKey
+	&pxor	("xmm4","xmm4");	# low dword of xmm4 is assumed 0
+	&lea	($key,&DWP(16,$key));
+	&cmp	($rounds,256);
+	&je	(&label("14rounds"));
+	&cmp	($rounds,192);
+	&je	(&label("12rounds"));
+	&cmp	($rounds,128);
+	&jne	(&label("bad_keybits"));
+
+&set_label("10rounds",16);
+	&mov		($rounds,9);
+	&$movekey	(&QWP(-16,$key),"xmm0");	# round 0
+	&aeskeygenassist("xmm1","xmm0",0x01);		# round 1
+	&call		(&label("key_128_cold"));
+	&aeskeygenassist("xmm1","xmm0",0x2);		# round 2
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x04);		# round 3
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x08);		# round 4
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x10);		# round 5
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x20);		# round 6
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x40);		# round 7
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x80);		# round 8
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x1b);		# round 9
+	&call		(&label("key_128"));
+	&aeskeygenassist("xmm1","xmm0",0x36);		# round 10
+	&call		(&label("key_128"));
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&mov		(&DWP(80,$key),$rounds);
+	&xor		("eax","eax");
+	&ret();
+
+&set_label("key_128",16);
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&lea		($key,&DWP(16,$key));
+&set_label("key_128_cold");
+	&shufps		("xmm4","xmm0",0b00010000);
+	&pxor		("xmm0","xmm4");
+	&shufps		("xmm4","xmm0",0b10001100,);
+	&pxor		("xmm0","xmm4");
+	&pshufd		("xmm1","xmm1",0b11111111);	# critical path
+	&pxor		("xmm0","xmm1");
+	&ret();
+
+&set_label("12rounds",16);
+	&movq		("xmm2",&QWP(16,"eax"));	# remaining 1/3 of *userKey
+	&mov		($rounds,11);
+	&$movekey	(&QWP(-16,$key),"xmm0")		# round 0
+	&aeskeygenassist("xmm1","xmm2",0x01);		# round 1,2
+	&call		(&label("key_192a_cold"));
+	&aeskeygenassist("xmm1","xmm2",0x02);		# round 2,3
+	&call		(&label("key_192b"));
+	&aeskeygenassist("xmm1","xmm2",0x04);		# round 4,5
+	&call		(&label("key_192a"));
+	&aeskeygenassist("xmm1","xmm2",0x08);		# round 5,6
+	&call		(&label("key_192b"));
+	&aeskeygenassist("xmm1","xmm2",0x10);		# round 7,8
+	&call		(&label("key_192a"));
+	&aeskeygenassist("xmm1","xmm2",0x20);		# round 8,9
+	&call		(&label("key_192b"));
+	&aeskeygenassist("xmm1","xmm2",0x40);		# round 10,11
+	&call		(&label("key_192a"));
+	&aeskeygenassist("xmm1","xmm2",0x80);		# round 11,12
+	&call		(&label("key_192b"));
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&mov		(&DWP(48,$key),$rounds);
+	&xor		("eax","eax");
+	&ret();
+
+&set_label("key_192a",16);
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&lea		($key,&DWP(16,$key));
+&set_label("key_192a_cold",16);
+	&movaps		("xmm5","xmm2");
+&set_label("key_192b_warm");
+	&shufps		("xmm4","xmm0",0b00010000);
+	&movaps		("xmm3","xmm2");
+	&pxor		("xmm0","xmm4");
+	&shufps		("xmm4","xmm0",0b10001100);
+	&pslldq		("xmm3",4);
+	&pxor		("xmm0","xmm4");
+	&pshufd		("xmm1","xmm1",0b01010101);	# critical path
+	&pxor		("xmm2","xmm3");
+	&pxor		("xmm0","xmm1");
+	&pshufd		("xmm3","xmm0",0b11111111);
+	&pxor		("xmm2","xmm3");
+	&ret();
+
+&set_label("key_192b",16);
+	&movaps		("xmm3","xmm0");
+	&shufps		("xmm5","xmm0",0b01000100);
+	&$movekey	(&QWP(0,$key),"xmm5");
+	&shufps		("xmm3","xmm2",0b01001110);
+	&$movekey	(&QWP(16,$key),"xmm3");
+	&lea		($key,&DWP(32,$key));
+	&jmp		(&label("key_192b_warm"));
+
+&set_label("14rounds",16);
+	&movups		("xmm2",&QWP(16,"eax"));	# remaining half of *userKey
+	&mov		($rounds,13);
+	&lea		($key,&DWP(16,$key));
+	&$movekey	(&QWP(-32,$key),"xmm0");	# round 0
+	&$movekey	(&QWP(-16,$key),"xmm2");	# round 1
+	&aeskeygenassist("xmm1","xmm2",0x01);		# round 2
+	&call		(&label("key_256a_cold"));
+	&aeskeygenassist("xmm1","xmm0",0x01);		# round 3
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x02);		# round 4
+	&call		(&label("key_256a"));
+	&aeskeygenassist("xmm1","xmm0",0x02);		# round 5
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x04);		# round 6
+	&call		(&label("key_256a"));
+	&aeskeygenassist("xmm1","xmm0",0x04);		# round 7
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x08);		# round 8
+	&call		(&label("key_256a"));
+	&aeskeygenassist("xmm1","xmm0",0x08);		# round 9
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x10);		# round 10
+	&call		(&label("key_256a"));
+	&aeskeygenassist("xmm1","xmm0",0x10);		# round 11
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x20);		# round 12
+	&call		(&label("key_256a"));
+	&aeskeygenassist("xmm1","xmm0",0x20);		# round 13
+	&call		(&label("key_256b"));
+	&aeskeygenassist("xmm1","xmm2",0x40);		# round 14
+	&call		(&label("key_256a"));
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&mov		(&DWP(16,$key),$rounds);
+	&xor		("eax","eax");
+	&ret();
+
+&set_label("key_256a",16);
+	&$movekey	(&QWP(0,$key),"xmm2");
+	&lea		($key,&DWP(16,$key));
+&set_label("key_256a_cold");
+	&shufps		("xmm4","xmm0",0b00010000);
+	&pxor		("xmm0","xmm4");
+	&shufps		("xmm4","xmm0",0b10001100);
+	&pxor		("xmm0","xmm4");
+	&pshufd		("xmm1","xmm1",0b11111111);	# critical path
+	&pxor		("xmm0","xmm1");
+	&ret();
+
+&set_label("key_256b",16);
+	&$movekey	(&QWP(0,$key),"xmm0");
+	&lea		($key,&DWP(16,$key));
+
+	&shufps		("xmm4","xmm2",0b00010000);
+	&pxor		("xmm2","xmm4");
+	&shufps		("xmm4","xmm2",0b10001100);
+	&pxor		("xmm2","xmm4");
+	&pshufd		("xmm1","xmm1",0b10101010);	# critical path
+	&pxor		("xmm2","xmm1");
+	&ret();
+
+&set_label("bad_pointer",4);
+	&mov	("eax",-1);
+	&ret	();
+&set_label("bad_keybits",4);
+	&mov	("eax",-2);
+	&ret	();
+&function_end_B("_aesni_set_encrypt_key");
+
+# int $PREFIX_set_encrypt_key (const unsigned char *userKey, int bits,
+#                              AES_KEY *key)
+&function_begin_B("${PREFIX}_set_encrypt_key");
+	&mov	("eax",&wparam(0));
+	&mov	($rounds,&wparam(1));
+	&mov	($key,&wparam(2));
+	&call	("_aesni_set_encrypt_key");
+	&ret	();
+&function_end_B("${PREFIX}_set_encrypt_key");
+
+# int $PREFIX_set_decrypt_key (const unsigned char *userKey, int bits,
+#                              AES_KEY *key)
+&function_begin_B("${PREFIX}_set_decrypt_key");
+	&mov	("eax",&wparam(0));
+	&mov	($rounds,&wparam(1));
+	&mov	($key,&wparam(2));
+	&call	("_aesni_set_encrypt_key");
+	&mov	($key,&wparam(2));
+	&shl	($rounds,4)	# rounds-1 after _aesni_set_encrypt_key
+	&test	("eax","eax");
+	&jnz	(&label("dec_key_ret"));
+	&lea	("eax",&DWP(16,$key,$rounds));	# end of key schedule
+
+	&$movekey	("xmm0",&QWP(0,$key));	# just swap
+	&$movekey	("xmm1",&QWP(0,"eax"));
+	&$movekey	(&QWP(0,"eax"),"xmm0");
+	&$movekey	(&QWP(0,$key),"xmm1");
+	&lea		($key,&DWP(16,$key));
+	&lea		("eax",&DWP(-16,"eax"));
+
+&set_label("dec_key_inverse");
+	&$movekey	("xmm0",&QWP(0,$key));	# swap and inverse
+	&$movekey	("xmm1",&QWP(0,"eax"));
+	&aesimc		("xmm0","xmm0");
+	&aesimc		("xmm1","xmm1");
+	&lea		($key,&DWP(16,$key));
+	&lea		("eax",&DWP(-16,"eax"));
+	&cmp		("eax",$key);
+	&$movekey	(&QWP(16,"eax"),"xmm0");
+	&$movekey	(&QWP(-16,$key),"xmm1");
+	&ja		(&label("dec_key_inverse"));
+
+	&$movekey	("xmm0",&QWP(0,$key));	# inverse middle
+	&aesimc		("xmm0","xmm0");
+	&$movekey	(&QWP(0,$key),"xmm0");
+
+	&xor		("eax","eax");		# return success
+&set_label("dec_key_ret");
+	&ret	();
+&function_end_B("${PREFIX}_set_decrypt_key");
+&asciz("AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>");
+
+&asm_finish();
diff -up openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86_64.pl.aesni openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86_64.pl
--- openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86_64.pl.aesni	2010-01-12 22:18:06.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/aes/asm/aesni-x86_64.pl	2010-01-12 22:18:06.000000000 +0100
@@ -0,0 +1,991 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements support for Intel AES-NI extension. In
+# OpenSSL context it's used with Intel engine, but can also be used as
+# drop-in replacement for crypto/aes/asm/aes-x86_64.pl [see below for
+# details].
+
+$PREFIX="aesni";	# if $PREFIX is set to "AES", the script
+			# generates drop-in replacement for
+			# crypto/aes/asm/aes-x86_64.pl:-)
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour $output";
+
+$movkey = $PREFIX eq "aesni" ? "movaps" : "movups";
+@_4args=$win64?	("%rcx","%rdx","%r8", "%r9") :	# Win64 order
+		("%rdi","%rsi","%rdx","%rcx");	# Unix order
+
+$code=".text\n";
+
+$rounds="%eax";	# input to and changed by aesni_[en|de]cryptN !!!
+# this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ...
+$inp="%rdi";
+$out="%rsi";
+$len="%rdx";
+$key="%rcx";	# input to and changed by aesni_[en|de]cryptN !!!
+$ivp="%r8";	# cbc
+
+$rnds_="%r10d";	# backup copy for $rounds
+$key_="%r11";	# backup copy for $key
+
+# %xmm register layout
+$inout0="%xmm0";	$inout1="%xmm1";
+$inout2="%xmm2";	$inout3="%xmm3";
+$rndkey0="%xmm4";	$rndkey1="%xmm5";
+
+$iv="%xmm6";		$in0="%xmm7";	# used in CBC decrypt
+$in1="%xmm8";		$in2="%xmm9";
+
+# Inline version of internal aesni_[en|de]crypt1.
+#
+# Why folded loop? Because aes[enc|dec] is slow enough to accommodate
+# cycles which take care of loop variables...
+{ my $sn;
+sub aesni_generate1 {
+my ($p,$key,$rounds)=@_;
+++$sn;
+$code.=<<___;
+	$movkey	($key),$rndkey0
+	$movkey	16($key),$rndkey1
+	lea	32($key),$key
+	pxor	$rndkey0,$inout0
+.Loop_${p}1_$sn:
+	aes${p}	$rndkey1,$inout0
+	dec	$rounds
+	$movkey	($key),$rndkey1
+	lea	16($key),$key
+	jnz	.Loop_${p}1_$sn	# loop body is 16 bytes
+	aes${p}last	$rndkey1,$inout0
+___
+}}
+# void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key);
+#
+{ my ($inp,$out,$key) = @_4args;
+
+$code.=<<___;
+.globl	${PREFIX}_encrypt
+.type	${PREFIX}_encrypt,\@abi-omnipotent
+.align	16
+${PREFIX}_encrypt:
+	movups	($inp),$inout0		# load input
+	mov	240($key),$rounds	# pull $rounds
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)		# output
+	ret
+.size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
+
+.globl	${PREFIX}_decrypt
+.type	${PREFIX}_decrypt,\@abi-omnipotent
+.align	16
+${PREFIX}_decrypt:
+	movups	($inp),$inout0		# load input
+	mov	240($key),$rounds	# pull $rounds
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)		# output
+	ret
+.size	${PREFIX}_decrypt, .-${PREFIX}_decrypt
+___
+}
+
+# _aesni_[en|de]crypt[34] are private interfaces, N denotes interleave
+# factor. Why 3x subroutine is used in loops? Even though aes[enc|dec]
+# latency is 6, it turned out that it can be scheduled only every
+# *second* cycle. Thus 3x interleave is 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. As soon
+# as/if Intel improves throughput by making it possible to schedule
+# the instructions in question *every* cycles I would have to
+# implement 6x interleave and use it in loop...
+sub aesni_generate3 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-2] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt3,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt3:
+	$movkey	($key),$rndkey0
+	shr	\$1,$rounds
+	$movkey	16($key),$rndkey1
+	lea	32($key),$key
+	pxor	$rndkey0,$inout0
+	pxor	$rndkey0,$inout1
+	pxor	$rndkey0,$inout2
+
+.L${dir}_loop3:
+	aes${dir}	$rndkey1,$inout0
+	$movkey		($key),$rndkey0
+	aes${dir}	$rndkey1,$inout1
+	dec		$rounds
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey0,$inout0
+	$movkey		16($key),$rndkey1
+	aes${dir}	$rndkey0,$inout1
+	lea		32($key),$key
+	aes${dir}	$rndkey0,$inout2
+	jnz		.L${dir}_loop3
+
+	aes${dir}	$rndkey1,$inout0
+	$movkey		($key),$rndkey0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	ret
+.size	_aesni_${dir}rypt3,.-_aesni_${dir}rypt3
+___
+}
+# 4x interleave is implemented to improve small block performance,
+# most notably [and naturally] 4 block by ~30%. One can argue that one
+# should have implemented 5x as well, but improvement would be <20%,
+# so it's not worth it...
+sub aesni_generate4 {
+my $dir=shift;
+# As already mentioned it takes in $key and $rounds, which are *not*
+# preserved. $inout[0-3] is cipher/clear text...
+$code.=<<___;
+.type	_aesni_${dir}rypt4,\@abi-omnipotent
+.align	16
+_aesni_${dir}rypt4:
+	$movkey	($key),$rndkey0
+	shr	\$1,$rounds
+	$movkey	16($key),$rndkey1
+	lea	32($key),$key
+	pxor	$rndkey0,$inout0
+	pxor	$rndkey0,$inout1
+	pxor	$rndkey0,$inout2
+	pxor	$rndkey0,$inout3
+
+.L${dir}_loop4:
+	aes${dir}	$rndkey1,$inout0
+	$movkey		($key),$rndkey0
+	aes${dir}	$rndkey1,$inout1
+	dec		$rounds
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}	$rndkey0,$inout0
+	$movkey		16($key),$rndkey1
+	aes${dir}	$rndkey0,$inout1
+	lea		32($key),$key
+	aes${dir}	$rndkey0,$inout2
+	aes${dir}	$rndkey0,$inout3
+	jnz		.L${dir}_loop4
+
+	aes${dir}	$rndkey1,$inout0
+	$movkey		($key),$rndkey0
+	aes${dir}	$rndkey1,$inout1
+	aes${dir}	$rndkey1,$inout2
+	aes${dir}	$rndkey1,$inout3
+	aes${dir}last	$rndkey0,$inout0
+	aes${dir}last	$rndkey0,$inout1
+	aes${dir}last	$rndkey0,$inout2
+	aes${dir}last	$rndkey0,$inout3
+	ret
+.size	_aesni_${dir}rypt4,.-_aesni_${dir}rypt4
+___
+}
+&aesni_generate3("enc") if ($PREFIX eq "aesni");
+&aesni_generate3("dec");
+&aesni_generate4("enc") if ($PREFIX eq "aesni");
+&aesni_generate4("dec");
+
+if ($PREFIX eq "aesni") {
+# void aesni_ecb_encrypt (const void *in, void *out,
+#			  size_t length, const AES_KEY *key,
+#			  int enc);
+$code.=<<___;
+.globl	aesni_ecb_encrypt
+.type	aesni_ecb_encrypt,\@function,5
+.align	16
+aesni_ecb_encrypt:
+	cmp	\$16,$len		# check length
+	jb	.Lecb_ret
+
+	mov	240($key),$rounds	# pull $rounds
+	and	\$-16,$len
+	mov	$key,$key_		# backup $key
+	test	%r8d,%r8d		# 5th argument
+	mov	$rounds,$rnds_		# backup $rounds
+	jz	.Lecb_decrypt
+#--------------------------- ECB ENCRYPT ------------------------------#
+	sub	\$0x40,$len
+	jbe	.Lecb_enc_tail
+	jmp	.Lecb_enc_loop3
+.align 16
+.Lecb_enc_loop3:
+	movups	($inp),$inout0
+	movups	0x10($inp),$inout1
+	movups	0x20($inp),$inout2
+	call	_aesni_encrypt3
+	sub	\$0x30,$len
+	lea	0x30($inp),$inp
+	lea	0x30($out),$out
+	movups	$inout0,-0x30($out)
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout1,-0x20($out)
+	mov	$key_,$key		# restore $key
+	movups	$inout2,-0x10($out)
+	ja	.Lecb_enc_loop3
+
+.Lecb_enc_tail:
+	add	\$0x40,$len
+	jz	.Lecb_ret
+
+	cmp	\$0x10,$len
+	movups	($inp),$inout0
+	je	.Lecb_enc_one
+	cmp	\$0x20,$len
+	movups	0x10($inp),$inout1
+	je	.Lecb_enc_two
+	cmp	\$0x30,$len
+	movups	0x20($inp),$inout2
+	je	.Lecb_enc_three
+	movups	0x30($inp),$inout3
+	call	_aesni_encrypt4
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_one:
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_two:
+	call	_aesni_encrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_enc_three:
+	call	_aesni_encrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	jmp	.Lecb_ret
+#--------------------------- ECB DECRYPT ------------------------------#
+.align	16
+.Lecb_decrypt:
+	sub	\$0x40,$len
+	jbe	.Lecb_dec_tail
+	jmp	.Lecb_dec_loop3
+.align 16
+.Lecb_dec_loop3:
+	movups	($inp),$inout0
+	movups	0x10($inp),$inout1
+	movups	0x20($inp),$inout2
+	call	_aesni_decrypt3
+	sub	\$0x30,$len
+	lea	0x30($inp),$inp
+	lea	0x30($out),$out
+	movups	$inout0,-0x30($out)
+	mov	$rnds_,$rounds		# restore $rounds
+	movups	$inout1,-0x20($out)
+	mov	$key_,$key		# restore $key
+	movups	$inout2,-0x10($out)
+	ja	.Lecb_dec_loop3
+
+.Lecb_dec_tail:
+	add	\$0x40,$len
+	jz	.Lecb_ret
+
+	cmp	\$0x10,$len
+	movups	($inp),$inout0
+	je	.Lecb_dec_one
+	cmp	\$0x20,$len
+	movups	0x10($inp),$inout1
+	je	.Lecb_dec_two
+	cmp	\$0x30,$len
+	movups	0x20($inp),$inout2
+	je	.Lecb_dec_three
+	movups	0x30($inp),$inout3
+	call	_aesni_decrypt4
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+	movups	$inout3,0x30($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_one:
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	movups	$inout0,($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_two:
+	call	_aesni_decrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	jmp	.Lecb_ret
+.align	16
+.Lecb_dec_three:
+	call	_aesni_decrypt3
+	movups	$inout0,($out)
+	movups	$inout1,0x10($out)
+	movups	$inout2,0x20($out)
+
+.Lecb_ret:
+	ret
+.size	aesni_ecb_encrypt,.-aesni_ecb_encrypt
+___
+}
+
+# void $PREFIX_cbc_encrypt (const void *inp, void *out,
+#			    size_t length, const AES_KEY *key,
+#			    unsigned char *ivp,const int enc);
+$reserved = $win64?0x40:-0x18;	# used in decrypt
+$code.=<<___;
+.globl	${PREFIX}_cbc_encrypt
+.type	${PREFIX}_cbc_encrypt,\@function,6
+.align	16
+${PREFIX}_cbc_encrypt:
+	test	$len,$len		# check length
+	jz	.Lcbc_ret
+
+	mov	240($key),$rnds_	# pull $rounds
+	mov	$key,$key_		# backup $key
+	test	%r9d,%r9d		# 6th argument
+	jz	.Lcbc_decrypt
+#--------------------------- CBC ENCRYPT ------------------------------#
+	movups	($ivp),$inout0		# load iv as initial state
+	cmp	\$16,$len
+	mov	$rnds_,$rounds
+	jb	.Lcbc_enc_tail
+	sub	\$16,$len
+	jmp	.Lcbc_enc_loop
+.align 16
+.Lcbc_enc_loop:
+	movups	($inp),$inout1		# load input
+	lea	16($inp),$inp
+	pxor	$inout1,$inout0
+___
+	&aesni_generate1("enc",$key,$rounds);
+$code.=<<___;
+	sub	\$16,$len
+	lea	16($out),$out
+	mov	$rnds_,$rounds		# restore $rounds
+	mov	$key_,$key		# restore $key
+	movups	$inout0,-16($out)	# store output
+	jnc	.Lcbc_enc_loop
+	add	\$16,$len
+	jnz	.Lcbc_enc_tail
+	movups	$inout0,($ivp)
+	jmp	.Lcbc_ret
+
+.Lcbc_enc_tail:
+	mov	$len,%rcx	# zaps $key
+	xchg	$inp,$out	# $inp is %rsi and $out is %rdi now
+	.long	0x9066A4F3	# rep movsb
+	mov	\$16,%ecx	# zero tail
+	sub	$len,%rcx
+	xor	%eax,%eax
+	.long	0x9066AAF3	# rep stosb
+	lea	-16(%rdi),%rdi	# rewind $out by 1 block
+	mov	$rnds_,$rounds	# restore $rounds
+	mov	%rdi,%rsi	# $inp and $out are the same
+	mov	$key_,$key	# restore $key
+	xor	$len,$len	# len=16
+	jmp	.Lcbc_enc_loop	# one more spin
+#--------------------------- CBC DECRYPT ------------------------------#
+.align	16
+.Lcbc_decrypt:
+___
+$code.=<<___ if ($win64);
+	lea	-0x58(%rsp),%rsp
+	movaps	%xmm6,(%rsp)
+	movaps	%xmm7,0x10(%rsp)
+	movaps	%xmm8,0x20(%rsp)
+	movaps	%xmm9,0x30(%rsp)
+.Lcbc_decrypt_body:
+___
+$code.=<<___;
+	movups	($ivp),$iv
+	sub	\$0x40,$len
+	mov	$rnds_,$rounds
+	jbe	.Lcbc_dec_tail
+	jmp	.Lcbc_dec_loop3
+.align 16
+.Lcbc_dec_loop3:
+	movups	($inp),$inout0
+	movups	0x10($inp),$inout1
+	movups	0x20($inp),$inout2
+	movaps	$inout0,$in0
+	movaps	$inout1,$in1
+	movaps	$inout2,$in2
+	call	_aesni_decrypt3
+	sub	\$0x30,$len
+	lea	0x30($inp),$inp
+	lea	0x30($out),$out
+	pxor	$iv,$inout0
+	pxor	$in0,$inout1
+	movaps	$in2,$iv
+	pxor	$in1,$inout2
+	movups	$inout0,-0x30($out)
+	mov	$rnds_,$rounds	# restore $rounds
+	movups	$inout1,-0x20($out)
+	mov	$key_,$key	# restore $key
+	movups	$inout2,-0x10($out)
+	ja	.Lcbc_dec_loop3
+
+.Lcbc_dec_tail:
+	add	\$0x40,$len
+	movups	$iv,($ivp)
+	jz	.Lcbc_dec_ret
+
+	movups	($inp),$inout0
+	cmp	\$0x10,$len
+	movaps	$inout0,$in0
+	jbe	.Lcbc_dec_one
+	movups	0x10($inp),$inout1
+	cmp	\$0x20,$len
+	movaps	$inout1,$in1
+	jbe	.Lcbc_dec_two
+	movups	0x20($inp),$inout2
+	cmp	\$0x30,$len
+	movaps	$inout2,$in2
+	jbe	.Lcbc_dec_three
+	movups	0x30($inp),$inout3
+	call	_aesni_decrypt4
+	pxor	$iv,$inout0
+	movups	0x30($inp),$iv
+	pxor	$in0,$inout1
+	movups	$inout0,($out)
+	pxor	$in1,$inout2
+	movups	$inout1,0x10($out)
+	pxor	$in2,$inout3
+	movups	$inout2,0x20($out)
+	movaps	$inout3,$inout0
+	lea	0x30($out),$out
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_one:
+___
+	&aesni_generate1("dec",$key,$rounds);
+$code.=<<___;
+	pxor	$iv,$inout0
+	movaps	$in0,$iv
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_two:
+	call	_aesni_decrypt3
+	pxor	$iv,$inout0
+	pxor	$in0,$inout1
+	movups	$inout0,($out)
+	movaps	$in1,$iv
+	movaps	$inout1,$inout0
+	lea	0x10($out),$out
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_three:
+	call	_aesni_decrypt3
+	pxor	$iv,$inout0
+	pxor	$in0,$inout1
+	movups	$inout0,($out)
+	pxor	$in1,$inout2
+	movups	$inout1,0x10($out)
+	movaps	$in2,$iv
+	movaps	$inout2,$inout0
+	lea	0x20($out),$out
+	jmp	.Lcbc_dec_tail_collected
+.align	16
+.Lcbc_dec_tail_collected:
+	and	\$15,$len
+	movups	$iv,($ivp)
+	jnz	.Lcbc_dec_tail_partial
+	movups	$inout0,($out)
+	jmp	.Lcbc_dec_ret
+.Lcbc_dec_tail_partial:
+	movaps	$inout0,$reserved(%rsp)
+	mov	$out,%rdi
+	mov	$len,%rcx
+	lea	$reserved(%rsp),%rsi
+	.long	0x9066A4F3	# rep movsb
+
+.Lcbc_dec_ret:
+___
+$code.=<<___ if ($win64);
+	movaps	(%rsp),%xmm6
+	movaps	0x10(%rsp),%xmm7
+	movaps	0x20(%rsp),%xmm8
+	movaps	0x30(%rsp),%xmm9
+	lea	0x58(%rsp),%rsp
+___
+$code.=<<___;
+.Lcbc_ret:
+	ret
+.size	${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
+___
+
+# int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
+#				int bits, AES_KEY *key)
+{ my ($inp,$bits,$key) = @_4args;
+  $bits =~ s/%r/%e/;
+
+$code.=<<___;
+.globl	${PREFIX}_set_decrypt_key
+.type	${PREFIX}_set_decrypt_key,\@abi-omnipotent
+.align	16
+${PREFIX}_set_decrypt_key:
+	.byte	0x48,0x83,0xEC,0x08	# sub rsp,8
+	call	_aesni_set_encrypt_key
+	shl	\$4,$bits		# rounds-1 after _aesni_set_encrypt_key
+	test	%eax,%eax
+	jnz	.Ldec_key_ret
+	lea	16($key,$bits),$inp	# points at the end of key schedule
+
+	$movkey	($key),%xmm0		# just swap
+	$movkey	($inp),%xmm1
+	$movkey	%xmm0,($inp)
+	$movkey	%xmm1,($key)
+	lea	16($key),$key
+	lea	-16($inp),$inp
+
+.Ldec_key_inverse:
+	$movkey	($key),%xmm0		# swap and inverse
+	$movkey	($inp),%xmm1
+	aesimc	%xmm0,%xmm0
+	aesimc	%xmm1,%xmm1
+	lea	16($key),$key
+	lea	-16($inp),$inp
+	cmp	$key,$inp
+	$movkey	%xmm0,16($inp)
+	$movkey	%xmm1,-16($key)
+	ja	.Ldec_key_inverse
+
+	$movkey	($key),%xmm0		# inverse middle
+	aesimc	%xmm0,%xmm0
+	$movkey	%xmm0,($inp)
+.Ldec_key_ret:
+	add	\$8,%rsp
+	ret
+.LSEH_end_set_decrypt_key:
+.size	${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
+___
+
+# This is based on submission by
+#
+#	Huang Ying <ying.huang@intel.com>
+#	Vinodh Gopal <vinodh.gopal@intel.com>
+#	Kahraman Akdemir
+#
+# Agressively optimized in respect to aeskeygenassist's critical path
+# and is contained in %xmm0-5 to meet Win64 ABI requirement.
+#
+$code.=<<___;
+.globl	${PREFIX}_set_encrypt_key
+.type	${PREFIX}_set_encrypt_key,\@abi-omnipotent
+.align	16
+${PREFIX}_set_encrypt_key:
+_aesni_set_encrypt_key:
+	.byte	0x48,0x83,0xEC,0x08	# sub rsp,8
+	test	$inp,$inp
+	mov	\$-1,%rax
+	jz	.Lenc_key_ret
+	test	$key,$key
+	jz	.Lenc_key_ret
+
+	movups	($inp),%xmm0		# pull first 128 bits of *userKey
+	pxor	%xmm4,%xmm4		# low dword of xmm4 is assumed 0
+	lea	16($key),%rax
+	cmp	\$256,$bits
+	je	.L14rounds
+	cmp	\$192,$bits
+	je	.L12rounds
+	cmp	\$128,$bits
+	jne	.Lbad_keybits
+
+.L10rounds:
+	mov	\$9,$bits			# 10 rounds for 128-bit key
+	$movkey	%xmm0,($key)			# round 0
+	aeskeygenassist	\$0x1,%xmm0,%xmm1	# round 1
+	call		.Lkey_expansion_128_cold
+	aeskeygenassist	\$0x2,%xmm0,%xmm1	# round 2
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x4,%xmm0,%xmm1	# round 3
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x8,%xmm0,%xmm1	# round 4
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x10,%xmm0,%xmm1	# round 5
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x20,%xmm0,%xmm1	# round 6
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x40,%xmm0,%xmm1	# round 7
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x80,%xmm0,%xmm1	# round 8
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x1b,%xmm0,%xmm1	# round 9
+	call		.Lkey_expansion_128
+	aeskeygenassist	\$0x36,%xmm0,%xmm1	# round 10
+	call		.Lkey_expansion_128
+	$movkey	%xmm0,(%rax)
+	mov	$bits,80(%rax)	# 240(%rdx)
+	xor	%eax,%eax
+	jmp	.Lenc_key_ret
+
+.align	16
+.L12rounds:
+	movq	16($inp),%xmm2			# remaining 1/3 of *userKey
+	mov	\$11,$bits			# 12 rounds for 192
+	$movkey	%xmm0,($key)			# round 0
+	aeskeygenassist	\$0x1,%xmm2,%xmm1	# round 1,2
+	call		.Lkey_expansion_192a_cold
+	aeskeygenassist	\$0x2,%xmm2,%xmm1	# round 2,3
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x4,%xmm2,%xmm1	# round 4,5
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x8,%xmm2,%xmm1	# round 5,6
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x10,%xmm2,%xmm1	# round 7,8
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x20,%xmm2,%xmm1	# round 8,9
+	call		.Lkey_expansion_192b
+	aeskeygenassist	\$0x40,%xmm2,%xmm1	# round 10,11
+	call		.Lkey_expansion_192a
+	aeskeygenassist	\$0x80,%xmm2,%xmm1	# round 11,12
+	call		.Lkey_expansion_192b
+	$movkey	%xmm0,(%rax)
+	mov	$bits,48(%rax)	# 240(%rdx)
+	xor	%rax, %rax
+	jmp	.Lenc_key_ret
+
+.align	16
+.L14rounds:
+	movups	16($inp),%xmm2			# remaning half of *userKey
+	mov	\$13,$bits			# 14 rounds for 256
+	lea	16(%rax),%rax
+	$movkey	%xmm0,($key)			# round 0
+	$movkey	%xmm2,16($key)			# round 1
+	aeskeygenassist	\$0x1,%xmm2,%xmm1	# round 2
+	call		.Lkey_expansion_256a_cold
+	aeskeygenassist	\$0x1,%xmm0,%xmm1	# round 3
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x2,%xmm2,%xmm1	# round 4
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x2,%xmm0,%xmm1	# round 5
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x4,%xmm2,%xmm1	# round 6
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x4,%xmm0,%xmm1	# round 7
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x8,%xmm2,%xmm1	# round 8
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x8,%xmm0,%xmm1	# round 9
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x10,%xmm2,%xmm1	# round 10
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x10,%xmm0,%xmm1	# round 11
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x20,%xmm2,%xmm1	# round 12
+	call		.Lkey_expansion_256a
+	aeskeygenassist	\$0x20,%xmm0,%xmm1	# round 13
+	call		.Lkey_expansion_256b
+	aeskeygenassist	\$0x40,%xmm2,%xmm1	# round 14
+	call		.Lkey_expansion_256a
+	$movkey	%xmm0,(%rax)
+	mov	$bits,16(%rax)	# 240(%rdx)
+	xor	%rax,%rax
+	jmp	.Lenc_key_ret
+
+.align	16
+.Lbad_keybits:
+	mov	\$-2,%rax
+.Lenc_key_ret:
+	add	\$8,%rsp
+	ret
+.LSEH_end_set_encrypt_key:
+
+.align	16
+.Lkey_expansion_128:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_128_cold:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	pxor	%xmm4, %xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	pxor	%xmm4, %xmm0
+	pshufd	\$0b11111111,%xmm1,%xmm1	# critical path
+	pxor	%xmm1,%xmm0
+	ret
+
+.align 16
+.Lkey_expansion_192a:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_192a_cold:
+	movaps	%xmm2, %xmm5
+.Lkey_expansion_192b_warm:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	movaps	%xmm2,%xmm3
+	pxor	%xmm4,%xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	pslldq	\$4,%xmm3
+	pxor	%xmm4,%xmm0
+	pshufd	\$0b01010101,%xmm1,%xmm1	# critical path
+	pxor	%xmm3,%xmm2
+	pxor	%xmm1,%xmm0
+	pshufd	\$0b11111111,%xmm0,%xmm3
+	pxor	%xmm3,%xmm2
+	ret
+
+.align 16
+.Lkey_expansion_192b:
+	movaps	%xmm0,%xmm3
+	shufps	\$0b01000100,%xmm0,%xmm5
+	$movkey	%xmm5,(%rax)
+	shufps	\$0b01001110,%xmm2,%xmm3
+	$movkey	%xmm3,16(%rax)
+	lea	32(%rax),%rax
+	jmp	.Lkey_expansion_192b_warm
+
+.align	16
+.Lkey_expansion_256a:
+	$movkey	%xmm2,(%rax)
+	lea	16(%rax),%rax
+.Lkey_expansion_256a_cold:
+	shufps	\$0b00010000,%xmm0,%xmm4
+	pxor	%xmm4,%xmm0
+	shufps	\$0b10001100,%xmm0,%xmm4
+	pxor	%xmm4,%xmm0
+	pshufd	\$0b11111111,%xmm1,%xmm1	# critical path
+	pxor	%xmm1,%xmm0
+	ret
+
+.align 16
+.Lkey_expansion_256b:
+	$movkey	%xmm0,(%rax)
+	lea	16(%rax),%rax
+
+	shufps	\$0b00010000,%xmm2,%xmm4
+	pxor	%xmm4,%xmm2
+	shufps	\$0b10001100,%xmm2,%xmm4
+	pxor	%xmm4,%xmm2
+	pshufd	\$0b10101010,%xmm1,%xmm1	# critical path
+	pxor	%xmm1,%xmm2
+	ret
+.size	${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
+___
+}
+
+$code.=<<___;
+.asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
+.align	64
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	cbc_se_handler,\@abi-omnipotent
+.align	16
+cbc_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	152($context),%rax	# pull context->Rsp
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lcbc_decrypt(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<"prologue" label
+	jb	.Lin_prologue
+
+	lea	.Lcbc_decrypt_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<cbc_decrypt_body
+	jb	.Lrestore_rax
+
+	lea	.Lcbc_ret(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>="epilogue" label
+	jae	.Lin_prologue
+
+	lea	0(%rax),%rsi		# top of stack
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$8,%ecx		# 4*sizeof(%xmm0)/sizeof(%rax)
+	.long	0xa548f3fc		# cld; rep movsq
+	lea	0x58(%rax),%rax		# adjust stack pointer
+	jmp	.Lin_prologue
+
+.Lrestore_rax:
+	mov	120($context),%rax
+.Lin_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	jmp	.Lcommon_seh_exit
+.size	cbc_se_handler,.-cbc_se_handler
+
+.type	ecb_se_handler,\@abi-omnipotent
+.align	16
+ecb_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	152($context),%rax	# pull context->Rsp
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+.Lcommon_seh_exit:
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	cbc_se_handler,.-cbc_se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_${PREFIX}_ecb_encrypt
+	.rva	.LSEH_end_${PREFIX}_ecb_encrypt
+	.rva	.LSEH_info_ecb
+
+	.rva	.LSEH_begin_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_end_${PREFIX}_cbc_encrypt
+	.rva	.LSEH_info_cbc
+
+	.rva	${PREFIX}_set_decrypt_key
+	.rva	.LSEH_end_set_decrypt_key
+	.rva	.LSEH_info_key
+
+	.rva	${PREFIX}_set_encrypt_key
+	.rva	.LSEH_end_set_encrypt_key
+	.rva	.LSEH_info_key
+.section	.xdata
+.align	8
+.LSEH_info_ecb:
+	.byte	9,0,0,0
+	.rva	ecb_se_handler
+.LSEH_info_cbc:
+	.byte	9,0,0,0
+	.rva	cbc_se_handler
+.LSEH_info_key:
+	.byte	0x01,0x04,0x01,0x00
+	.byte	0x04,0x02,0x00,0x00
+___
+}
+
+sub rex {
+ local *opcode=shift;
+ my ($dst,$src)=@_;
+
+   if ($dst>=8 || $src>=8) {
+	$rex=0x40;
+	$rex|=0x04 if($dst>=8);
+	$rex|=0x01 if($src>=8);
+	push @opcode,$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);
+    }
+    return $line;
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
+
+print $code;
+
+close STDOUT;
diff -up openssl-1.0.0-beta4/crypto/aes/Makefile.aesni openssl-1.0.0-beta4/crypto/aes/Makefile
--- openssl-1.0.0-beta4/crypto/aes/Makefile.aesni	2008-12-23 12:33:00.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/aes/Makefile	2010-01-12 22:18:06.000000000 +0100
@@ -50,9 +50,13 @@ aes-ia64.s: asm/aes-ia64.S
 
 aes-586.s:	asm/aes-586.pl ../perlasm/x86asm.pl
 	$(PERL) asm/aes-586.pl $(PERLASM_SCHEME) $(CFLAGS) $(PROCESSOR) > $@
+aesni-x86.s:	asm/aesni-x86.pl ../perlasm/x86asm.pl
+	$(PERL) asm/aesni-x86.pl $(PERLASM_SCHEME) $(CFLAGS) $(PROCESSOR) > $@
 
 aes-x86_64.s: asm/aes-x86_64.pl
 	$(PERL) asm/aes-x86_64.pl $(PERLASM_SCHEME) > $@
+aesni-x86_64.s: asm/aesni-x86_64.pl
+	$(PERL) asm/aesni-x86_64.pl $(PERLASM_SCHEME) > $@
 
 aes-sparcv9.s: asm/aes-sparcv9.pl
 	$(PERL) asm/aes-sparcv9.pl $(CFLAGS) > $@
diff -up openssl-1.0.0-beta4/crypto/engine/eng_aesni.c.aesni openssl-1.0.0-beta4/crypto/engine/eng_aesni.c
--- openssl-1.0.0-beta4/crypto/engine/eng_aesni.c.aesni	2010-01-12 22:18:06.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/engine/eng_aesni.c	2010-01-12 22:18:06.000000000 +0100
@@ -0,0 +1,413 @@
+/*
+ * Support for Intel AES-NI intruction set
+ *   Author: Huang Ying <ying.huang@intel.com>
+ *
+ * Intel AES-NI is a new set of Single Instruction Multiple Data
+ * (SIMD) instructions that are going to be introduced in the next
+ * generation of Intel processor, as of 2009. These instructions
+ * enable fast and secure data encryption and decryption, using the
+ * Advanced Encryption Standard (AES), defined by FIPS Publication
+ * number 197.  The architecture introduces six instructions that
+ * offer full hardware support for AES. Four of them support high
+ * performance data encryption and decryption, and the other two
+ * instructions support the AES key expansion procedure.
+ *
+ * The white paper can be downloaded from:
+ *   http://softwarecommunity.intel.com/isn/downloads/intelavx/AES-Instructions-Set_WP.pdf
+ *
+ * This file is based on engines/e_padlock.c
+ */
+
+/* ====================================================================
+ * Copyright (c) 1999-2001 The OpenSSL Project.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * 3. All advertising materials mentioning features or use of this
+ *    software must display the following acknowledgment:
+ *    "This product includes software developed by the OpenSSL Project
+ *    for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
+ *
+ * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
+ *    endorse or promote products derived from this software without
+ *    prior written permission. For written permission, please contact
+ *    licensing@OpenSSL.org.
+ *
+ * 5. Products derived from this software may not be called "OpenSSL"
+ *    nor may "OpenSSL" appear in their names without prior written
+ *    permission of the OpenSSL Project.
+ *
+ * 6. Redistributions of any form whatsoever must retain the following
+ *    acknowledgment:
+ *    "This product includes software developed by the OpenSSL Project
+ *    for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
+ * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
+ * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ * ====================================================================
+ *
+ * This product includes cryptographic software written by Eric Young
+ * (eay@cryptsoft.com).  This product includes software written by Tim
+ * Hudson (tjh@cryptsoft.com).
+ *
+ */
+
+
+#include <openssl/opensslconf.h>
+
+#if !defined(OPENSSL_NO_HW) && !defined(OPENSSL_NO_HW_AES_NI) && !defined(OPENSSL_NO_AES)
+
+#include <stdio.h>
+#include "cryptlib.h"
+#include <openssl/dso.h>
+#include <openssl/engine.h>
+#include <openssl/evp.h>
+#include <openssl/aes.h>
+#include <openssl/err.h>
+#include <openssl/modes.h>
+
+/* AES-NI is available *ONLY* on some x86 CPUs.  Not only that it
+   doesn't exist elsewhere, but it even can't be compiled on other
+   platforms! */
+#undef COMPILE_HW_AESNI
+#if (defined(__x86_64) || defined(__x86_64__) || \
+     defined(_M_AMD64) || defined(_M_X64) || \
+     defined(OPENSSL_IA32_SSE2)) && !defined(OPENSSL_NO_ASM)
+#define COMPILE_HW_AESNI
+static ENGINE *ENGINE_aesni (void);
+#endif
+
+void ENGINE_load_aesni (void)
+{
+/* On non-x86 CPUs it just returns. */
+#ifdef COMPILE_HW_AESNI
+	ENGINE *toadd = ENGINE_aesni();
+	if (!toadd)
+		return;
+	ENGINE_add (toadd);
+	ENGINE_register_complete (toadd);
+	ENGINE_free (toadd);
+	ERR_clear_error ();
+#endif
+}
+
+#ifdef COMPILE_HW_AESNI
+int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
+			      AES_KEY *key);
+int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
+			      AES_KEY *key);
+
+void aesni_encrypt(const unsigned char *in, unsigned char *out,
+		       const AES_KEY *key);
+void aesni_decrypt(const unsigned char *in, unsigned char *out,
+		       const AES_KEY *key);
+
+void aesni_ecb_encrypt(const unsigned char *in,
+			   unsigned char *out,
+			   size_t length,
+			   const AES_KEY *key,
+			   int enc);
+void aesni_cbc_encrypt(const unsigned char *in,
+			   unsigned char *out,
+			   size_t length,
+			   const AES_KEY *key,
+			   unsigned char *ivec, int enc);
+
+/* Function for ENGINE detection and control */
+static int aesni_init(ENGINE *e);
+
+/* Cipher Stuff */
+static int aesni_ciphers(ENGINE *e, const EVP_CIPHER **cipher,
+				const int **nids, int nid);
+
+#define AESNI_MIN_ALIGN	16
+#define AESNI_ALIGN(x) \
+	((void *)(((unsigned long)(x)+AESNI_MIN_ALIGN-1)&~(AESNI_MIN_ALIGN-1)))
+
+/* Engine names */
+static const char   aesni_id[] = "aesni",
+		    aesni_name[] = "Intel AES-NI engine",
+		    no_aesni_name[] = "Intel AES-NI engine (no-aesni)";
+
+/* ===== Engine "management" functions ===== */
+
+#if defined(_WIN32)
+typedef unsigned __int64 IA32CAP;
+#else
+typedef unsigned long long IA32CAP;
+#endif
+
+/* Prepare the ENGINE structure for registration */
+static int
+aesni_bind_helper(ENGINE *e)
+{
+	int engage;
+	if (sizeof(OPENSSL_ia32cap_P) > 4) {
+		engage = (OPENSSL_ia32cap_P >> 57) & 1;
+	} else {
+		IA32CAP OPENSSL_ia32_cpuid(void);
+		engage = (OPENSSL_ia32_cpuid() >> 57) & 1;
+	}
+
+	/* Register everything or return with an error */
+	if (!ENGINE_set_id(e, aesni_id) ||
+	    !ENGINE_set_name(e, engage ? aesni_name : no_aesni_name) ||
+
+	    !ENGINE_set_init_function(e, aesni_init) ||
+	    (engage && !ENGINE_set_ciphers (e, aesni_ciphers))
+	    )
+		return 0;
+
+	/* Everything looks good */
+	return 1;
+}
+
+/* Constructor */
+static ENGINE *
+ENGINE_aesni(void)
+{
+	ENGINE *eng = ENGINE_new();
+
+	if (!eng) {
+		return NULL;
+	}
+
+	if (!aesni_bind_helper(eng)) {
+		ENGINE_free(eng);
+		return NULL;
+	}
+
+	return eng;
+}
+
+/* Check availability of the engine */
+static int
+aesni_init(ENGINE *e)
+{
+	return 1;
+}
+
+#if defined(NID_aes_128_cfb128) && ! defined (NID_aes_128_cfb)
+#define NID_aes_128_cfb	NID_aes_128_cfb128
+#endif
+
+#if defined(NID_aes_128_ofb128) && ! defined (NID_aes_128_ofb)
+#define NID_aes_128_ofb	NID_aes_128_ofb128
+#endif
+
+#if defined(NID_aes_192_cfb128) && ! defined (NID_aes_192_cfb)
+#define NID_aes_192_cfb	NID_aes_192_cfb128
+#endif
+
+#if defined(NID_aes_192_ofb128) && ! defined (NID_aes_192_ofb)
+#define NID_aes_192_ofb	NID_aes_192_ofb128
+#endif
+
+#if defined(NID_aes_256_cfb128) && ! defined (NID_aes_256_cfb)
+#define NID_aes_256_cfb	NID_aes_256_cfb128
+#endif
+
+#if defined(NID_aes_256_ofb128) && ! defined (NID_aes_256_ofb)
+#define NID_aes_256_ofb	NID_aes_256_ofb128
+#endif
+
+/* List of supported ciphers. */
+static int aesni_cipher_nids[] = {
+	NID_aes_128_ecb,
+	NID_aes_128_cbc,
+	NID_aes_128_cfb,
+	NID_aes_128_ofb,
+
+	NID_aes_192_ecb,
+	NID_aes_192_cbc,
+	NID_aes_192_cfb,
+	NID_aes_192_ofb,
+
+	NID_aes_256_ecb,
+	NID_aes_256_cbc,
+	NID_aes_256_cfb,
+	NID_aes_256_ofb,
+};
+static int aesni_cipher_nids_num =
+	(sizeof(aesni_cipher_nids)/sizeof(aesni_cipher_nids[0]));
+
+typedef struct
+{
+	AES_KEY ks;
+	unsigned int _pad1[3];
+} AESNI_KEY;
+
+static int
+aesni_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *user_key,
+		    const unsigned char *iv, int enc)
+{
+	int ret;
+	AES_KEY *key = AESNI_ALIGN(ctx->cipher_data);
+
+	if ((ctx->cipher->flags & EVP_CIPH_MODE) == EVP_CIPH_CFB_MODE
+	    || (ctx->cipher->flags & EVP_CIPH_MODE) == EVP_CIPH_OFB_MODE
+	    || enc)
+		ret=aesni_set_encrypt_key(user_key, ctx->key_len * 8, key);
+	else
+		ret=aesni_set_decrypt_key(user_key, ctx->key_len * 8, key);
+
+	if(ret < 0) {
+		EVPerr(EVP_F_AESNI_INIT_KEY,EVP_R_AES_KEY_SETUP_FAILED);
+		return 0;
+	}
+
+	return 1;
+}
+
+static int aesni_cipher_ecb(EVP_CIPHER_CTX *ctx, unsigned char *out,
+		 const unsigned char *in, size_t inl)
+{	AES_KEY *key = AESNI_ALIGN(ctx->cipher_data);
+	aesni_ecb_encrypt(in, out, inl, key, ctx->encrypt);
+	return 1;
+}
+static int aesni_cipher_cbc(EVP_CIPHER_CTX *ctx, unsigned char *out,
+		 const unsigned char *in, size_t inl)
+{	AES_KEY *key = AESNI_ALIGN(ctx->cipher_data);
+	aesni_cbc_encrypt(in, out, inl, key,
+			      ctx->iv, ctx->encrypt);
+	return 1;
+}
+static int aesni_cipher_cfb(EVP_CIPHER_CTX *ctx, unsigned char *out,
+		 const unsigned char *in, size_t inl)
+{	AES_KEY *key = AESNI_ALIGN(ctx->cipher_data);
+	CRYPTO_cfb128_encrypt(in, out, inl, key, ctx->iv,
+				&ctx->num, ctx->encrypt,
+				(block128_f)aesni_encrypt);
+	return 1;
+}
+static int aesni_cipher_ofb(EVP_CIPHER_CTX *ctx, unsigned char *out,
+		 const unsigned char *in, size_t inl)
+{	AES_KEY *key = AESNI_ALIGN(ctx->cipher_data);
+	CRYPTO_ofb128_encrypt(in, out, inl, key, ctx->iv,
+				&ctx->num, (block128_f)aesni_encrypt);
+	return 1;
+}
+
+#define AES_BLOCK_SIZE		16
+
+#define EVP_CIPHER_block_size_ECB	AES_BLOCK_SIZE
+#define EVP_CIPHER_block_size_CBC	AES_BLOCK_SIZE
+#define EVP_CIPHER_block_size_OFB	1
+#define EVP_CIPHER_block_size_CFB	1
+
+/* Declaring so many ciphers by hand would be a pain.
+   Instead introduce a bit of preprocessor magic :-) */
+#define	DECLARE_AES_EVP(ksize,lmode,umode)	\
+static const EVP_CIPHER aesni_##ksize##_##lmode = {	\
+	NID_aes_##ksize##_##lmode,			\
+	EVP_CIPHER_block_size_##umode,			\
+	ksize / 8,					\
+	AES_BLOCK_SIZE,					\
+	0 | EVP_CIPH_##umode##_MODE,			\
+	aesni_init_key,				\
+	aesni_cipher_##lmode,				\
+	NULL,						\
+	sizeof(AESNI_KEY),				\
+	EVP_CIPHER_set_asn1_iv,				\
+	EVP_CIPHER_get_asn1_iv,				\
+	NULL,						\
+	NULL						\
+}
+
+DECLARE_AES_EVP(128,ecb,ECB);
+DECLARE_AES_EVP(128,cbc,CBC);
+DECLARE_AES_EVP(128,cfb,CFB);
+DECLARE_AES_EVP(128,ofb,OFB);
+
+DECLARE_AES_EVP(192,ecb,ECB);
+DECLARE_AES_EVP(192,cbc,CBC);
+DECLARE_AES_EVP(192,cfb,CFB);
+DECLARE_AES_EVP(192,ofb,OFB);
+
+DECLARE_AES_EVP(256,ecb,ECB);
+DECLARE_AES_EVP(256,cbc,CBC);
+DECLARE_AES_EVP(256,cfb,CFB);
+DECLARE_AES_EVP(256,ofb,OFB);
+
+static int
+aesni_ciphers (ENGINE *e, const EVP_CIPHER **cipher,
+		      const int **nids, int nid)
+{
+	/* No specific cipher => return a list of supported nids ... */
+	if (!cipher) {
+		*nids = aesni_cipher_nids;
+		return aesni_cipher_nids_num;
+	}
+
+	/* ... or the requested "cipher" otherwise */
+	switch (nid) {
+	case NID_aes_128_ecb:
+		*cipher = &aesni_128_ecb;
+		break;
+	case NID_aes_128_cbc:
+		*cipher = &aesni_128_cbc;
+		break;
+	case NID_aes_128_cfb:
+		*cipher = &aesni_128_cfb;
+		break;
+	case NID_aes_128_ofb:
+		*cipher = &aesni_128_ofb;
+		break;
+
+	case NID_aes_192_ecb:
+		*cipher = &aesni_192_ecb;
+		break;
+	case NID_aes_192_cbc:
+		*cipher = &aesni_192_cbc;
+		break;
+	case NID_aes_192_cfb:
+		*cipher = &aesni_192_cfb;
+		break;
+	case NID_aes_192_ofb:
+		*cipher = &aesni_192_ofb;
+		break;
+
+	case NID_aes_256_ecb:
+		*cipher = &aesni_256_ecb;
+		break;
+	case NID_aes_256_cbc:
+		*cipher = &aesni_256_cbc;
+		break;
+	case NID_aes_256_cfb:
+		*cipher = &aesni_256_cfb;
+		break;
+	case NID_aes_256_ofb:
+		*cipher = &aesni_256_ofb;
+		break;
+
+	default:
+		/* Sorry, we don't support this NID */
+		*cipher = NULL;
+		return 0;
+	}
+
+	return 1;
+}
+
+#endif /* COMPILE_HW_AESNI */
+#endif /* !defined(OPENSSL_NO_HW) && !defined(OPENSSL_NO_HW_AESNI) && !defined(OPENSSL_NO_AES) */
diff -up openssl-1.0.0-beta4/crypto/engine/eng_all.c.aesni openssl-1.0.0-beta4/crypto/engine/eng_all.c
--- openssl-1.0.0-beta4/crypto/engine/eng_all.c.aesni	2010-01-07 23:38:31.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/engine/eng_all.c	2010-01-12 22:18:06.000000000 +0100
@@ -85,6 +85,9 @@ void ENGINE_load_builtin_engines(void)
 #if !defined(OPENSSL_NO_HW) && (defined(__OpenBSD__) || defined(__FreeBSD__) || defined(HAVE_CRYPTODEV))
 	ENGINE_load_cryptodev();
 #endif
+#if !defined(OPENSSL_NO_HW) && !defined(OPENSSL_NO_HW_AESNI)
+	ENGINE_load_aesni();
+#endif
 	ENGINE_load_dynamic();
 #ifndef OPENSSL_NO_STATIC_ENGINE
 #ifndef OPENSSL_NO_HW
diff -up openssl-1.0.0-beta4/crypto/engine/engine.h.aesni openssl-1.0.0-beta4/crypto/engine/engine.h
--- openssl-1.0.0-beta4/crypto/engine/engine.h.aesni	2010-01-07 23:38:30.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/engine/engine.h	2010-01-12 22:18:06.000000000 +0100
@@ -342,6 +342,7 @@ void ENGINE_load_gost(void);
 #endif
 #endif
 void ENGINE_load_cryptodev(void);
+void ENGINE_load_aesni(void);
 void ENGINE_load_builtin_engines(void);
 
 /* Get and set global flags (ENGINE_TABLE_FLAG_***) for the implementation
diff -up openssl-1.0.0-beta4/crypto/engine/Makefile.aesni openssl-1.0.0-beta4/crypto/engine/Makefile
--- openssl-1.0.0-beta4/crypto/engine/Makefile.aesni	2008-06-04 13:01:29.000000000 +0200
+++ openssl-1.0.0-beta4/crypto/engine/Makefile	2010-01-12 22:18:06.000000000 +0100
@@ -21,12 +21,14 @@ LIBSRC= eng_err.c eng_lib.c eng_list.c e
 	eng_table.c eng_pkey.c eng_fat.c eng_all.c \
 	tb_rsa.c tb_dsa.c tb_ecdsa.c tb_dh.c tb_ecdh.c tb_rand.c tb_store.c \
 	tb_cipher.c tb_digest.c tb_pkmeth.c tb_asnmth.c \
-	eng_openssl.c eng_cnf.c eng_dyn.c eng_cryptodev.c
+	eng_openssl.c eng_cnf.c eng_dyn.c eng_cryptodev.c \
+	eng_aesni.c
 LIBOBJ= eng_err.o eng_lib.o eng_list.o eng_init.o eng_ctrl.o \
 	eng_table.o eng_pkey.o eng_fat.o eng_all.o \
 	tb_rsa.o tb_dsa.o tb_ecdsa.o tb_dh.o tb_ecdh.o tb_rand.o tb_store.o \
 	tb_cipher.o tb_digest.o tb_pkmeth.o tb_asnmth.o \
-	eng_openssl.o eng_cnf.o eng_dyn.o eng_cryptodev.o
+	eng_openssl.o eng_cnf.o eng_dyn.o eng_cryptodev.o \
+	eng_aesni.o
 
 SRC= $(LIBSRC)
 
diff -up openssl-1.0.0-beta4/crypto/evp/evp_err.c.aesni openssl-1.0.0-beta4/crypto/evp/evp_err.c
--- openssl-1.0.0-beta4/crypto/evp/evp_err.c.aesni	2010-01-07 23:38:31.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/evp/evp_err.c	2010-01-12 22:18:06.000000000 +0100
@@ -1,6 +1,6 @@
 /* crypto/evp/evp_err.c */
 /* ====================================================================
- * Copyright (c) 1999-2008 The OpenSSL Project.  All rights reserved.
+ * Copyright (c) 1999-2009 The OpenSSL Project.  All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
@@ -70,6 +70,7 @@
 
 static ERR_STRING_DATA EVP_str_functs[]=
 	{
+{ERR_FUNC(EVP_F_AESNI_INIT_KEY),	"AESNI_INIT_KEY"},
 {ERR_FUNC(EVP_F_AES_INIT_KEY),	"AES_INIT_KEY"},
 {ERR_FUNC(EVP_F_CAMELLIA_INIT_KEY),	"CAMELLIA_INIT_KEY"},
 {ERR_FUNC(EVP_F_D2I_PKEY),	"D2I_PKEY"},
@@ -85,7 +86,7 @@ static ERR_STRING_DATA EVP_str_functs[]=
 {ERR_FUNC(EVP_F_EVP_DIGESTINIT_EX),	"EVP_DigestInit_ex"},
 {ERR_FUNC(EVP_F_EVP_ENCRYPTFINAL_EX),	"EVP_EncryptFinal_ex"},
 {ERR_FUNC(EVP_F_EVP_MD_CTX_COPY_EX),	"EVP_MD_CTX_copy_ex"},
-{ERR_FUNC(EVP_F_EVP_MD_SIZE),	"EVP_MD_SIZE"},
+{ERR_FUNC(EVP_F_EVP_MD_SIZE),	"EVP_MD_size"},
 {ERR_FUNC(EVP_F_EVP_OPENINIT),	"EVP_OpenInit"},
 {ERR_FUNC(EVP_F_EVP_PBE_ALG_ADD),	"EVP_PBE_alg_add"},
 {ERR_FUNC(EVP_F_EVP_PBE_ALG_ADD_TYPE),	"EVP_PBE_alg_add_type"},
diff -up openssl-1.0.0-beta4/crypto/evp/evp.h.aesni openssl-1.0.0-beta4/crypto/evp/evp.h
--- openssl-1.0.0-beta4/crypto/evp/evp.h.aesni	2010-01-07 23:38:31.000000000 +0100
+++ openssl-1.0.0-beta4/crypto/evp/evp.h	2010-01-12 22:18:06.000000000 +0100
@@ -1162,6 +1162,7 @@ void ERR_load_EVP_strings(void);
 /* Error codes for the EVP functions. */
 
 /* Function codes. */
+#define EVP_F_AESNI_INIT_KEY				 163
 #define EVP_F_AES_INIT_KEY				 133
 #define EVP_F_CAMELLIA_INIT_KEY				 159
 #define EVP_F_D2I_PKEY					 100
diff -up openssl-1.0.0-beta4/test/test_aesni.aesni openssl-1.0.0-beta4/test/test_aesni
--- openssl-1.0.0-beta4/test/test_aesni.aesni	2010-01-12 22:18:06.000000000 +0100
+++ openssl-1.0.0-beta4/test/test_aesni	2010-01-12 22:18:06.000000000 +0100
@@ -0,0 +1,69 @@
+#!/bin/sh
+
+PROG=$1
+
+if [ -x $PROG ]; then
+    if expr "x`$PROG version`" : "xOpenSSL" > /dev/null; then
+	:
+    else
+	echo "$PROG is not OpenSSL executable"
+	exit 1
+    fi
+else
+    echo "$PROG is not executable"
+    exit 1;
+fi
+
+if $PROG engine aesni | grep -v no-aesni; then
+
+    HASH=`cat $PROG | $PROG dgst -hex`
+
+    AES_ALGS="	aes-128-ecb aes-192-ecb aes-256-ecb \
+		aes-128-cbc aes-192-cbc aes-256-cbc \
+		aes-128-cfb aes-192-cfb aes-256-cfb \
+		aes-128-ofb aes-192-ofb aes-256-ofb"
+    BUFSIZE="16 32 48 64 80 96 128 144 999"
+
+    nerr=0
+
+    for alg in $AES_ALGS; do
+	echo $alg
+	for bufsize in $BUFSIZE; do
+	    TEST=`(	cat $PROG | \
+		$PROG enc -e -k "$HASH" -$alg -bufsize $bufsize -engine aesni | \
+		$PROG enc -d -k "$HASH" -$alg | \
+		$PROG dgst -hex ) 2>/dev/null`
+	    if [ "$TEST" != "$HASH" ]; then
+		echo "-$alg/$bufsize encrypt test failed"
+		nerr=`expr $nerr + 1`
+	    fi
+	done
+	for bufsize in $BUFSIZE; do 
+	    TEST=`(	cat $PROG | \
+		$PROG enc -e -k "$HASH" -$alg | \
+		$PROG enc -d -k "$HASH" -$alg -bufsize $bufsize -engine aesni | \
+		$PROG dgst -hex ) 2>/dev/null`
+	    if [ "$TEST" != "$HASH" ]; then
+		echo "-$alg/$bufsize decrypt test failed"
+		nerr=`expr $nerr + 1`
+	    fi
+	done
+	TEST=`(	cat $PROG | \
+		$PROG enc -e -k "$HASH" -$alg -engine aesni | \
+		$PROG enc -d -k "$HASH" -$alg -engine aesni | \
+		$PROG dgst -hex ) 2>/dev/null`
+	if [ "$TEST" != "$HASH" ]; then
+		echo "-$alg en/decrypt test failed"
+		nerr=`expr $nerr + 1`
+	fi
+    done
+
+    if [ $nerr -gt 0 ]; then
+	echo "AESNI engine test failed."
+	exit 1;
+    fi
+else
+    echo "AESNI engine is not available"
+fi
+
+exit 0