Revert "Remove engine_rsax and its asm file."

This reverts commit 5226c62b7632dfaf38480919d406307318a7d145.

Reviewed-by: Andy Polyakov <appro@openssl.org>

6 files changed, 2220 insertions, 3 deletions

diff --git a/Configure b/Configure
index 02161d8636..541be9ec57 100755
--- a/Configure
+++ b/Configure
@@ -128,7 +128,7 @@ my $x86_asm="x86cpuid.o:bn-586.o co-586.o x86-mont.o x86-gf2m.o:des-586.o crypt5
 
 my $x86_elf_asm="$x86_asm:elf";
 
-my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o x86_64-mont5.o x86_64-gf2m.o::aes-x86_64.o vpaes-x86_64.o bsaes-x86_64.o aesni-x86_64.o aesni-sha1-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o rc4-md5-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:ghash-x86_64.o:";
+my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o x86_64-mont5.o x86_64-gf2m.o modexp512-x86_64.o::aes-x86_64.o vpaes-x86_64.o bsaes-x86_64.o aesni-x86_64.o aesni-sha1-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o rc4-md5-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:ghash-x86_64.o:";
 my $ia64_asm="ia64cpuid.o:bn-ia64.o ia64-mont.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:::::ghash-ia64.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:::::::ghash-sparcv9.o::void";
 my $sparcv8_asm=":sparcv8.o:des_enc-sparc.o fcrypt_b.o:::::::::::::void";
diff --git a/crypto/bn/Makefile b/crypto/bn/Makefile
index e1452b3060..6dd136be5d 100644
--- a/crypto/bn/Makefile
+++ b/crypto/bn/Makefile
@@ -102,6 +102,8 @@ x86_64-mont5.s:	asm/x86_64-mont5.pl
 	$(PERL) asm/x86_64-mont5.pl $(PERLASM_SCHEME) > $@
 x86_64-gf2m.s:	asm/x86_64-gf2m.pl
 	$(PERL) asm/x86_64-gf2m.pl $(PERLASM_SCHEME) > $@
+modexp512-x86_64.s:	asm/modexp512-x86_64.pl
+	$(PERL) asm/modexp512-x86_64.pl $(PERLASM_SCHEME) > $@
 
 bn-ia64.s:	asm/ia64.S
 	$(CC) $(CFLAGS) -E asm/ia64.S > $@
diff --git a/crypto/bn/asm/modexp512-x86_64.pl b/crypto/bn/asm/modexp512-x86_64.pl
new file mode 100644
index 0000000000..bfd6e97541
--- /dev/null
+++ b/crypto/bn/asm/modexp512-x86_64.pl
@@ -0,0 +1,1497 @@
+#!/usr/bin/env perl
+#
+# Copyright (c) 2010-2011 Intel Corp.
+#   Author: Vinodh.Gopal@intel.com
+#           Jim Guilford
+#           Erdinc.Ozturk@intel.com
+#           Maxim.Perminov@intel.com
+#
+# More information about algorithm used can be found at:
+#   http://www.cse.buffalo.edu/srds2009/escs2009_submission_Gopal.pdf
+#
+# ====================================================================
+# Copyright (c) 2011 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.
+# ====================================================================
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+my $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 $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+use strict;
+my $code=".text\n\n";
+my $m=0;
+
+#
+# Define x512 macros
+#
+
+#MULSTEP_512_ADD	MACRO	x7, x6, x5, x4, x3, x2, x1, x0, dst, src1, src2, add_src, tmp1, tmp2
+#
+# uses rax, rdx, and args
+sub MULSTEP_512_ADD
+{
+ my ($x, $DST, $SRC2, $ASRC, $OP, $TMP)=@_;
+ my @X=@$x;	# make a copy
+$code.=<<___;
+	 mov	(+8*0)($SRC2), %rax
+	 mul	$OP			# rdx:rax = %OP * [0]
+	 mov	($ASRC), $X[0]
+	 add	%rax, $X[0]
+	 adc	\$0, %rdx
+	 mov	$X[0], $DST
+___
+for(my $i=1;$i<8;$i++) {
+$code.=<<___;
+	 mov	%rdx, $TMP
+
+	 mov	(+8*$i)($SRC2), %rax
+	 mul	$OP			# rdx:rax = %OP * [$i]
+	 mov	(+8*$i)($ASRC), $X[$i]
+	 add	%rax, $X[$i]
+	 adc	\$0, %rdx
+	 add	$TMP, $X[$i]
+	 adc	\$0, %rdx
+___
+}
+$code.=<<___;
+	 mov	%rdx, $X[0]
+___
+}
+
+#MULSTEP_512	MACRO	x7, x6, x5, x4, x3, x2, x1, x0, dst, src2, src1_val, tmp
+#
+# uses rax, rdx, and args
+sub MULSTEP_512
+{
+ my ($x, $DST, $SRC2, $OP, $TMP)=@_;
+ my @X=@$x;	# make a copy
+$code.=<<___;
+	 mov	(+8*0)($SRC2), %rax
+	 mul	$OP			# rdx:rax = %OP * [0]
+	 add	%rax, $X[0]
+	 adc	\$0, %rdx
+	 mov	$X[0], $DST
+___
+for(my $i=1;$i<8;$i++) {
+$code.=<<___;
+	 mov	%rdx, $TMP
+
+	 mov	(+8*$i)($SRC2), %rax
+	 mul	$OP			# rdx:rax = %OP * [$i]
+	 add	%rax, $X[$i]
+	 adc	\$0, %rdx
+	 add	$TMP, $X[$i]
+	 adc	\$0, %rdx
+___
+}
+$code.=<<___;
+	 mov	%rdx, $X[0]
+___
+}
+
+#
+# Swizzle Macros
+#
+
+# macro to copy data from flat space to swizzled table
+#MACRO swizzle	pDst, pSrc, tmp1, tmp2
+# pDst and pSrc are modified
+sub swizzle
+{
+ my ($pDst, $pSrc, $cnt, $d0)=@_;
+$code.=<<___;
+	 mov	\$8, $cnt
+loop_$m:
+	 mov	($pSrc), $d0
+	 mov	$d0#w, ($pDst)
+	 shr	\$16, $d0
+	 mov	$d0#w, (+64*1)($pDst)
+	 shr	\$16, $d0
+	 mov	$d0#w, (+64*2)($pDst)
+	 shr	\$16, $d0
+	 mov	$d0#w, (+64*3)($pDst)
+	 lea	8($pSrc), $pSrc
+	 lea	64*4($pDst), $pDst
+	 dec	$cnt
+	 jnz	loop_$m
+___
+
+ $m++;
+}
+
+# macro to copy data from swizzled table to  flat space
+#MACRO unswizzle	pDst, pSrc, tmp*3
+sub unswizzle
+{
+ my ($pDst, $pSrc, $cnt, $d0, $d1)=@_;
+$code.=<<___;
+	 mov	\$4, $cnt
+loop_$m:
+	 movzxw	(+64*3+256*0)($pSrc), $d0
+	 movzxw	(+64*3+256*1)($pSrc), $d1
+	 shl	\$16, $d0
+	 shl	\$16, $d1
+	 mov	(+64*2+256*0)($pSrc), $d0#w
+	 mov	(+64*2+256*1)($pSrc), $d1#w
+	 shl	\$16, $d0
+	 shl	\$16, $d1
+	 mov	(+64*1+256*0)($pSrc), $d0#w
+	 mov	(+64*1+256*1)($pSrc), $d1#w
+	 shl	\$16, $d0
+	 shl	\$16, $d1
+	 mov	(+64*0+256*0)($pSrc), $d0#w
+	 mov	(+64*0+256*1)($pSrc), $d1#w
+	 mov	$d0, (+8*0)($pDst)
+	 mov	$d1, (+8*1)($pDst)
+	 lea	256*2($pSrc), $pSrc
+	 lea	8*2($pDst), $pDst
+	 sub	\$1, $cnt
+	 jnz	loop_$m
+___
+
+ $m++;
+}
+
+#
+# Data Structures
+#
+
+# Reduce Data
+#
+#
+# Offset  Value
+# 0C0     Carries
+# 0B8     X2[10]
+# 0B0     X2[9]
+# 0A8     X2[8]
+# 0A0     X2[7]
+# 098     X2[6]
+# 090     X2[5]
+# 088     X2[4]
+# 080     X2[3]
+# 078     X2[2]
+# 070     X2[1]
+# 068     X2[0]
+# 060     X1[12]  P[10]
+# 058     X1[11]  P[9]  Z[8]
+# 050     X1[10]  P[8]  Z[7]
+# 048     X1[9]   P[7]  Z[6]
+# 040     X1[8]   P[6]  Z[5]
+# 038     X1[7]   P[5]  Z[4]
+# 030     X1[6]   P[4]  Z[3]
+# 028     X1[5]   P[3]  Z[2]
+# 020     X1[4]   P[2]  Z[1]
+# 018     X1[3]   P[1]  Z[0]
+# 010     X1[2]   P[0]  Y[2]
+# 008     X1[1]   Q[1]  Y[1]
+# 000     X1[0]   Q[0]  Y[0]
+
+my $X1_offset           =  0;			# 13 qwords
+my $X2_offset           =  $X1_offset + 13*8;			# 11 qwords
+my $Carries_offset      =  $X2_offset + 11*8;			# 1 qword
+my $Q_offset            =  0;			# 2 qwords
+my $P_offset            =  $Q_offset + 2*8;			# 11 qwords
+my $Y_offset            =  0;			# 3 qwords
+my $Z_offset            =  $Y_offset + 3*8;			# 9 qwords
+
+my $Red_Data_Size       =  $Carries_offset + 1*8;			# (25 qwords)
+
+#
+# Stack Frame
+#
+#
+# offset	value
+# ...		<old stack contents>
+# ...
+# 280		Garray
+
+# 278		tmp16[15]
+# ...		...
+# 200		tmp16[0]
+
+# 1F8		tmp[7]
+# ...		...
+# 1C0		tmp[0]
+
+# 1B8		GT[7]
+# ...		...
+# 180		GT[0]
+
+# 178		Reduce Data
+# ...		...
+# 0B8		Reduce Data
+# 0B0		reserved
+# 0A8		reserved
+# 0A0		reserved
+# 098		reserved
+# 090		reserved
+# 088		reduce result addr
+# 080		exp[8]
+
+# ...
+# 048		exp[1]
+# 040		exp[0]
+
+# 038		reserved
+# 030		loop_idx
+# 028		pg
+# 020		i
+# 018		pData	; arg 4
+# 010		pG	; arg 2
+# 008		pResult	; arg 1
+# 000		rsp	; stack pointer before subtract
+
+my $rsp_offset          =  0;
+my $pResult_offset      =  8*1 + $rsp_offset;
+my $pG_offset           =  8*1 + $pResult_offset;
+my $pData_offset        =  8*1 + $pG_offset;
+my $i_offset            =  8*1 + $pData_offset;
+my $pg_offset           =  8*1 + $i_offset;
+my $loop_idx_offset     =  8*1 + $pg_offset;
+my $reserved1_offset    =  8*1 + $loop_idx_offset;
+my $exp_offset          =  8*1 + $reserved1_offset;
+my $red_result_addr_offset=  8*9 + $exp_offset;
+my $reserved2_offset    =  8*1 + $red_result_addr_offset;
+my $Reduce_Data_offset  =  8*5 + $reserved2_offset;
+my $GT_offset           =  $Red_Data_Size + $Reduce_Data_offset;
+my $tmp_offset          =  8*8 + $GT_offset;
+my $tmp16_offset        =  8*8 + $tmp_offset;
+my $garray_offset       =  8*16 + $tmp16_offset;
+my $mem_size            =  8*8*32 + $garray_offset;
+
+#
+# Offsets within Reduce Data
+#
+#
+#	struct MODF_2FOLD_MONT_512_C1_DATA {
+#	UINT64 t[8][8];
+#	UINT64 m[8];
+#	UINT64 m1[8]; /* 2^768 % m */
+#	UINT64 m2[8]; /* 2^640 % m */
+#	UINT64 k1[2]; /* (- 1/m) % 2^128 */
+#	};
+
+my $T                   =  0;
+my $M                   =  512;			# = 8 * 8 * 8
+my $M1                  =  576;			# = 8 * 8 * 9 /* += 8 * 8 */
+my $M2                  =  640;			# = 8 * 8 * 10 /* += 8 * 8 */
+my $K1                  =  704;			# = 8 * 8 * 11 /* += 8 * 8 */
+
+#
+#   FUNCTIONS
+#
+
+{{{
+#
+# MULADD_128x512 : Function to multiply 128-bits (2 qwords) by 512-bits (8 qwords)
+#                       and add 512-bits (8 qwords)
+#                       to get 640 bits (10 qwords)
+# Input: 128-bit mul source: [rdi+8*1], rbp
+#        512-bit mul source: [rsi+8*n]
+#        512-bit add source: r15, r14, ..., r9, r8
+# Output: r9, r8, r15, r14, r13, r12, r11, r10, [rcx+8*1], [rcx+8*0]
+# Clobbers all regs except: rcx, rsi, rdi
+$code.=<<___;
+.type	MULADD_128x512,\@abi-omnipotent
+.align	16
+MULADD_128x512:
+___
+	&MULSTEP_512([map("%r$_",(8..15))], "(+8*0)(%rcx)", "%rsi", "%rbp", "%rbx");
+$code.=<<___;
+	 mov	(+8*1)(%rdi), %rbp
+___
+	&MULSTEP_512([map("%r$_",(9..15,8))], "(+8*1)(%rcx)", "%rsi", "%rbp", "%rbx");
+$code.=<<___;
+	 ret
+.size	MULADD_128x512,.-MULADD_128x512
+___
+}}}
+
+{{{
+#MULADD_256x512	MACRO	pDst, pA, pB, OP, TMP, X7, X6, X5, X4, X3, X2, X1, X0
+#
+# Inputs: pDst: Destination  (768 bits, 12 qwords)
+#         pA:   Multiplicand (1024 bits, 16 qwords)
+#         pB:   Multiplicand (512 bits, 8 qwords)
+# Dst = Ah * B + Al
+# where Ah is (in qwords) A[15:12] (256 bits) and Al is A[7:0] (512 bits)
+# Results in X3 X2 X1 X0 X7 X6 X5 X4 Dst[3:0]
+# Uses registers: arguments, RAX, RDX
+sub MULADD_256x512
+{
+ my ($pDst, $pA, $pB, $OP, $TMP, $X)=@_;
+$code.=<<___;
+	mov	(+8*12)($pA), $OP
+___
+	&MULSTEP_512_ADD($X, "(+8*0)($pDst)", $pB, $pA, $OP, $TMP);
+	push(@$X,shift(@$X));
+
+$code.=<<___;
+	 mov	(+8*13)($pA), $OP
+___
+	&MULSTEP_512($X, "(+8*1)($pDst)", $pB, $OP, $TMP);
+	push(@$X,shift(@$X));
+
+$code.=<<___;
+	 mov	(+8*14)($pA), $OP
+___
+	&MULSTEP_512($X, "(+8*2)($pDst)", $pB, $OP, $TMP);
+	push(@$X,shift(@$X));
+
+$code.=<<___;
+	 mov	(+8*15)($pA), $OP
+___
+	&MULSTEP_512($X, "(+8*3)($pDst)", $pB, $OP, $TMP);
+	push(@$X,shift(@$X));
+}
+
+#
+# mont_reduce(UINT64 *x,  /* 1024 bits, 16 qwords */
+#	       UINT64 *m,  /*  512 bits,  8 qwords */
+#	       MODF_2FOLD_MONT_512_C1_DATA *data,
+#             UINT64 *r)  /*  512 bits,  8 qwords */
+# Input:  x (number to be reduced): tmp16 (Implicit)
+#         m (modulus):              [pM]  (Implicit)
+#         data (reduce data):       [pData] (Implicit)
+# Output: r (result):		     Address in [red_res_addr]
+#         result also in: r9, r8, r15, r14, r13, r12, r11, r10
+
+my @X=map("%r$_",(8..15));
+
+$code.=<<___;
+.type	mont_reduce,\@abi-omnipotent
+.align	16
+mont_reduce:
+___
+
+my $STACK_DEPTH         =  8;
+	#
+	# X1 = Xh * M1 + Xl
+$code.=<<___;
+	 lea	(+$Reduce_Data_offset+$X1_offset+$STACK_DEPTH)(%rsp), %rdi			# pX1 (Dst) 769 bits, 13 qwords
+	 mov	(+$pData_offset+$STACK_DEPTH)(%rsp), %rsi			# pM1 (Bsrc) 512 bits, 8 qwords
+	 add	\$$M1, %rsi
+	 lea	(+$tmp16_offset+$STACK_DEPTH)(%rsp), %rcx			# X (Asrc) 1024 bits, 16 qwords
+
+___
+
+	&MULADD_256x512("%rdi", "%rcx", "%rsi", "%rbp", "%rbx", \@X);	# rotates @X 4 times
+	# results in r11, r10, r9, r8, r15, r14, r13, r12, X1[3:0]
+
+$code.=<<___;
+	 xor	%rax, %rax
+	# X1 += xl
+	 add	(+8*8)(%rcx), $X[4]
+	 adc	(+8*9)(%rcx), $X[5]
+	 adc	(+8*10)(%rcx), $X[6]
+	 adc	(+8*11)(%rcx), $X[7]
+	 adc	\$0, %rax
+	# X1 is now rax, r11-r8, r15-r12, tmp16[3:0]
+
+	#
+	# check for carry ;; carry stored in rax
+	 mov	$X[4], (+8*8)(%rdi)			# rdi points to X1
+	 mov	$X[5], (+8*9)(%rdi)
+	 mov	$X[6], %rbp
+	 mov	$X[7], (+8*11)(%rdi)
+
+	 mov	%rax, (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp)
+
+	 mov	(+8*0)(%rdi), $X[4]
+	 mov	(+8*1)(%rdi), $X[5]
+	 mov	(+8*2)(%rdi), $X[6]
+	 mov	(+8*3)(%rdi), $X[7]
+
+	# X1 is now stored in: X1[11], rbp, X1[9:8], r15-r8
+	# rdi -> X1
+	# rsi -> M1
+
+	#
+	# X2 = Xh * M2 + Xl
+	# do first part (X2 = Xh * M2)
+	 add	\$8*10, %rdi			# rdi -> pXh ; 128 bits, 2 qwords
+				#        Xh is actually { [rdi+8*1], rbp }
+	 add	\$`$M2-$M1`, %rsi			# rsi -> M2
+	 lea	(+$Reduce_Data_offset+$X2_offset+$STACK_DEPTH)(%rsp), %rcx			# rcx -> pX2 ; 641 bits, 11 qwords
+___
+	unshift(@X,pop(@X));	unshift(@X,pop(@X));
+$code.=<<___;
+
+	 call	MULADD_128x512			# args in rcx, rdi / rbp, rsi, r15-r8
+	# result in r9, r8, r15, r14, r13, r12, r11, r10, X2[1:0]
+	 mov	(+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp), %rax
+
+	# X2 += Xl
+	 add	(+8*8-8*10)(%rdi), $X[6]		# (-8*10) is to adjust rdi -> Xh to Xl
+	 adc	(+8*9-8*10)(%rdi), $X[7]
+	 mov	$X[6], (+8*8)(%rcx)
+	 mov	$X[7], (+8*9)(%rcx)
+
+	 adc	%rax, %rax
+	 mov	%rax, (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp)
+
+	 lea	(+$Reduce_Data_offset+$Q_offset+$STACK_DEPTH)(%rsp), %rdi			# rdi -> pQ ; 128 bits, 2 qwords
+	 add	\$`$K1-$M2`, %rsi			# rsi -> pK1 ; 128 bits, 2 qwords
+
+	# MUL_128x128t128	rdi, rcx, rsi	; Q = X2 * K1 (bottom half)
+	# B1:B0 = rsi[1:0] = K1[1:0]
+	# A1:A0 = rcx[1:0] = X2[1:0]
+	# Result = rdi[1],rbp = Q[1],rbp
+	 mov	(%rsi), %r8			# B0
+	 mov	(+8*1)(%rsi), %rbx			# B1
+
+	 mov	(%rcx), %rax			# A0
+	 mul	%r8			# B0
+	 mov	%rax, %rbp
+	 mov	%rdx, %r9
+
+	 mov	(+8*1)(%rcx), %rax			# A1
+	 mul	%r8			# B0
+	 add	%rax, %r9
+
+	 mov	(%rcx), %rax			# A0
+	 mul	%rbx			# B1
+	 add	%rax, %r9
+
+	 mov	%r9, (+8*1)(%rdi)
+	# end MUL_128x128t128
+
+	 sub	\$`$K1-$M`, %rsi
+
+	 mov	(%rcx), $X[6]
+	 mov	(+8*1)(%rcx), $X[7]			# r9:r8 = X2[1:0]
+
+	 call	MULADD_128x512			# args in rcx, rdi / rbp, rsi, r15-r8
+	# result in r9, r8, r15, r14, r13, r12, r11, r10, X2[1:0]
+
+	# load first half of m to rdx, rdi, rbx, rax
+	# moved this here for efficiency
+	 mov	(+8*0)(%rsi), %rax
+	 mov	(+8*1)(%rsi), %rbx
+	 mov	(+8*2)(%rsi), %rdi
+	 mov	(+8*3)(%rsi), %rdx
+
+	# continue with reduction
+	 mov	(+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp), %rbp
+
+	 add	(+8*8)(%rcx), $X[6]
+	 adc	(+8*9)(%rcx), $X[7]
+
+	#accumulate the final carry to rbp
+	 adc	%rbp, %rbp
+
+	# Add in overflow corrections: R = (X2>>128) += T[overflow]
+	# R = {r9, r8, r15, r14, ..., r10}
+	 shl	\$3, %rbp
+	 mov	(+$pData_offset+$STACK_DEPTH)(%rsp), %rcx			# rsi -> Data (and points to T)
+	 add	%rcx, %rbp			# pT ; 512 bits, 8 qwords, spread out
+
+	# rsi will be used to generate a mask after the addition
+	 xor	%rsi, %rsi
+
+	 add	(+8*8*0)(%rbp), $X[0]
+	 adc	(+8*8*1)(%rbp), $X[1]
+	 adc	(+8*8*2)(%rbp), $X[2]
+	 adc	(+8*8*3)(%rbp), $X[3]
+	 adc	(+8*8*4)(%rbp), $X[4]
+	 adc	(+8*8*5)(%rbp), $X[5]
+	 adc	(+8*8*6)(%rbp), $X[6]
+	 adc	(+8*8*7)(%rbp), $X[7]
+
+	# if there is a carry:	rsi = 0xFFFFFFFFFFFFFFFF
+	# if carry is clear:	rsi = 0x0000000000000000
+	 sbb	\$0, %rsi
+
+	# if carry is clear, subtract 0. Otherwise, subtract 256 bits of m
+	 and	%rsi, %rax
+	 and	%rsi, %rbx
+	 and	%rsi, %rdi
+	 and	%rsi, %rdx
+
+	 mov	\$1, %rbp
+	 sub	%rax, $X[0]
+	 sbb	%rbx, $X[1]
+	 sbb	%rdi, $X[2]
+	 sbb	%rdx, $X[3]
+
+	# if there is a borrow:		rbp = 0
+	# if there is no borrow:	rbp = 1
+	# this is used to save the borrows in between the first half and the 2nd half of the subtraction of m
+	 sbb	\$0, %rbp
+
+	#load second half of m to rdx, rdi, rbx, rax
+
+	 add	\$$M, %rcx
+	 mov	(+8*4)(%rcx), %rax
+	 mov	(+8*5)(%rcx), %rbx
+	 mov	(+8*6)(%rcx), %rdi
+	 mov	(+8*7)(%rcx), %rdx
+
+	# use the rsi mask as before
+	# if carry is clear, subtract 0. Otherwise, subtract 256 bits of m
+	 and	%rsi, %rax
+	 and	%rsi, %rbx
+	 and	%rsi, %rdi
+	 and	%rsi, %rdx
+
+	# if rbp = 0, there was a borrow before, it is moved to the carry flag
+	# if rbp = 1, there was not a borrow before, carry flag is cleared
+	 sub	\$1, %rbp
+
+	 sbb	%rax, $X[4]
+	 sbb	%rbx, $X[5]
+	 sbb	%rdi, $X[6]
+	 sbb	%rdx, $X[7]
+
+	# write R back to memory
+
+	 mov	(+$red_result_addr_offset+$STACK_DEPTH)(%rsp), %rsi
+	 mov	$X[0], (+8*0)(%rsi)
+	 mov	$X[1], (+8*1)(%rsi)
+	 mov	$X[2], (+8*2)(%rsi)
+	 mov	$X[3], (+8*3)(%rsi)
+	 mov	$X[4], (+8*4)(%rsi)
+	 mov	$X[5], (+8*5)(%rsi)
+	 mov	$X[6], (+8*6)(%rsi)
+	 mov	$X[7], (+8*7)(%rsi)
+
+	 ret
+.size	mont_reduce,.-mont_reduce
+___
+}}}
+
+{{{
+#MUL_512x512	MACRO	pDst, pA, pB, x7, x6, x5, x4, x3, x2, x1, x0, tmp*2
+#
+# Inputs: pDst: Destination  (1024 bits, 16 qwords)
+#         pA:   Multiplicand (512 bits, 8 qwords)
+#         pB:   Multiplicand (512 bits, 8 qwords)
+# Uses registers rax, rdx, args
+#   B operand in [pB] and also in x7...x0
+sub MUL_512x512
+{
+ my ($pDst, $pA, $pB, $x, $OP, $TMP, $pDst_o)=@_;
+ my ($pDst,  $pDst_o) = ($pDst =~ m/([^+]*)\+?(.*)?/);
+ my @X=@$x;	# make a copy
+
+$code.=<<___;
+	 mov	(+8*0)($pA), $OP
+
+	 mov	$X[0], %rax
+	 mul	$OP			# rdx:rax = %OP * [0]
+	 mov	%rax, (+$pDst_o+8*0)($pDst)
+	 mov	%rdx, $X[0]
+___
+for(my $i=1;$i<8;$i++) {
+$code.=<<___;
+	 mov	$X[$i], %rax
+	 mul	$OP			# rdx:rax = %OP * [$i]
+	 add	%rax, $X[$i-1]
+	 adc	\$0, %rdx
+	 mov	%rdx, $X[$i]
+___
+}
+
+for(my $i=1;$i<8;$i++) {
+$code.=<<___;
+	 mov	(+8*$i)($pA), $OP
+___
+
+	&MULSTEP_512(\@X, "(+$pDst_o+8*$i)($pDst)", $pB, $OP, $TMP);
+	push(@X,shift(@X));
+}
+
+$code.=<<___;
+	 mov	$X[0], (+$pDst_o+8*8)($pDst)
+	 mov	$X[1], (+$pDst_o+8*9)($pDst)
+	 mov	$X[2], (+$pDst_o+8*10)($pDst)
+	 mov	$X[3], (+$pDst_o+8*11)($pDst)
+	 mov	$X[4], (+$pDst_o+8*12)($pDst)
+	 mov	$X[5], (+$pDst_o+8*13)($pDst)
+	 mov	$X[6], (+$pDst_o+8*14)($pDst)
+	 mov	$X[7], (+$pDst_o+8*15)($pDst)
+___
+}
+
+#
+# mont_mul_a3b : subroutine to compute (Src1 * Src2) % M (all 512-bits)
+# Input:  src1: Address of source 1: rdi
+#         src2: Address of source 2: rsi
+# Output: dst:  Address of destination: [red_res_addr]
+#    src2 and result also in: r9, r8, r15, r14, r13, r12, r11, r10
+# Temp:   Clobbers [tmp16], all registers
+$code.=<<___;
+.type	mont_mul_a3b,\@abi-omnipotent
+.align	16
+mont_mul_a3b:
+	#
+	# multiply tmp = src1 * src2
+	# For multiply: dst = rcx, src1 = rdi, src2 = rsi
+	# stack depth is extra 8 from call
+___
+	&MUL_512x512("%rsp+$tmp16_offset+8", "%rdi", "%rsi", [map("%r$_",(10..15,8..9))], "%rbp", "%rbx");
+$code.=<<___;
+	#
+	# Dst = tmp % m
+	# Call reduce(tmp, m, data, dst)
+
+	# tail recursion optimization: jmp to mont_reduce and return from there
+	 jmp	mont_reduce
+	# call	mont_reduce
+	# ret
+.size	mont_mul_a3b,.-mont_mul_a3b
+___
+}}}
+
+{{{
+#SQR_512 MACRO pDest, pA, x7, x6, x5, x4, x3, x2, x1, x0, tmp*4
+#
+# Input in memory [pA] and also in x7...x0
+# Uses all argument registers plus rax and rdx
+#
+# This version computes all of the off-diagonal terms into memory,
+# and then it adds in the diagonal terms
+
+sub SQR_512
+{
+ my ($pDst, $pA, $x, $A, $tmp, $x7, $x6, $pDst_o)=@_;
+ my ($pDst,  $pDst_o) = ($pDst =~ m/([^+]*)\+?(.*)?/);
+ my @X=@$x;	# make a copy
+$code.=<<___;
+	# ------------------
+	# first pass 01...07
+	# ------------------
+	 mov	$X[0], $A
+
+	 mov	$X[1],%rax
+	 mul	$A
+	 mov	%rax, (+$pDst_o+8*1)($pDst)
+___
+for(my $i=2;$i<8;$i++) {
+$code.=<<___;
+	 mov	%rdx, $X[$i-2]
+	 mov	$X[$i],%rax
+	 mul	$A
+	 add	%rax, $X[$i-2]
+	 adc	\$0, %rdx
+___
+}
+$code.=<<___;
+	 mov	%rdx, $x7
+
+	 mov	$X[0], (+$pDst_o+8*2)($pDst)
+
+	# ------------------
+	# second pass 12...17
+	# ------------------
+
+	 mov	(+8*1)($pA), $A
+
+	 mov	(+8*2)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[1]
+	 adc	\$0, %rdx
+	 mov	$X[1], (+$pDst_o+8*3)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*3)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[2]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[2]
+	 adc	\$0, %rdx
+	 mov	$X[2], (+$pDst_o+8*4)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*4)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[3]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[3]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*5)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[4]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[4]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	$X[6],%rax
+	 mul	$A
+	 add	%rax, $X[5]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[5]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $x7
+	 adc	\$0, %rdx
+	 add	$X[0], $x7
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[1]
+
+	# ------------------
+	# third pass 23...27
+	# ------------------
+	 mov	(+8*2)($pA), $A
+
+	 mov	(+8*3)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[3]
+	 adc	\$0, %rdx
+	 mov	$X[3], (+$pDst_o+8*5)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*4)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[4]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[4]
+	 adc	\$0, %rdx
+	 mov	$X[4], (+$pDst_o+8*6)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*5)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[5]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[5]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	$X[6],%rax
+	 mul	$A
+	 add	%rax, $x7
+	 adc	\$0, %rdx
+	 add	$X[0], $x7
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $X[1]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[1]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[2]
+
+	# ------------------
+	# fourth pass 34...37
+	# ------------------
+
+	 mov	(+8*3)($pA), $A
+
+	 mov	(+8*4)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[5]
+	 adc	\$0, %rdx
+	 mov	$X[5], (+$pDst_o+8*7)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	(+8*5)($pA),%rax
+	 mul	$A
+	 add	%rax, $x7
+	 adc	\$0, %rdx
+	 add	$X[0], $x7
+	 adc	\$0, %rdx
+	 mov	$x7, (+$pDst_o+8*8)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	$X[6],%rax
+	 mul	$A
+	 add	%rax, $X[1]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[1]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[0]
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $X[2]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[2]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[5]
+
+	# ------------------
+	# fifth pass 45...47
+	# ------------------
+	 mov	(+8*4)($pA), $A
+
+	 mov	(+8*5)($pA),%rax
+	 mul	$A
+	 add	%rax, $X[1]
+	 adc	\$0, %rdx
+	 mov	$X[1], (+$pDst_o+8*9)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	$X[6],%rax
+	 mul	$A
+	 add	%rax, $X[2]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[2]
+	 adc	\$0, %rdx
+	 mov	$X[2], (+$pDst_o+8*10)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $X[5]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[5]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $X[1]
+
+	# ------------------
+	# sixth pass 56...57
+	# ------------------
+	 mov	(+8*5)($pA), $A
+
+	 mov	$X[6],%rax
+	 mul	$A
+	 add	%rax, $X[5]
+	 adc	\$0, %rdx
+	 mov	$X[5], (+$pDst_o+8*11)($pDst)
+
+	 mov	%rdx, $X[0]
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $X[1]
+	 adc	\$0, %rdx
+	 add	$X[0], $X[1]
+	 adc	\$0, %rdx
+	 mov	$X[1], (+$pDst_o+8*12)($pDst)
+
+	 mov	%rdx, $X[2]
+
+	# ------------------
+	# seventh pass 67
+	# ------------------
+	 mov	$X[6], $A
+
+	 mov	$X[7],%rax
+	 mul	$A
+	 add	%rax, $X[2]
+	 adc	\$0, %rdx
+	 mov	$X[2], (+$pDst_o+8*13)($pDst)
+
+	 mov	%rdx, (+$pDst_o+8*14)($pDst)
+
+	# start finalize (add	in squares, and double off-terms)
+	 mov	(+$pDst_o+8*1)($pDst), $X[0]
+	 mov	(+$pDst_o+8*2)($pDst), $X[1]
+	 mov	(+$pDst_o+8*3)($pDst), $X[2]
+	 mov	(+$pDst_o+8*4)($pDst), $X[3]
+	 mov	(+$pDst_o+8*5)($pDst), $X[4]
+	 mov	(+$pDst_o+8*6)($pDst), $X[5]
+
+	 mov	(+8*3)($pA), %rax
+	 mul	%rax
+	 mov	%rax, $x6
+	 mov	%rdx, $X[6]
+
+	 add	$X[0], $X[0]
+	 adc	$X[1], $X[1]
+	 adc	$X[2], $X[2]
+	 adc	$X[3], $X[3]
+	 adc	$X[4], $X[4]
+	 adc	$X[5], $X[5]
+	 adc	\$0, $X[6]
+
+	 mov	(+8*0)($pA), %rax
+	 mul	%rax
+	 mov	%rax, (+$pDst_o+8*0)($pDst)
+	 mov	%rdx, $A
+
+	 mov	(+8*1)($pA), %rax
+	 mul	%rax
+
+	 add	$A, $X[0]
+	 adc	%rax, $X[1]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $A
+	 mov	$X[0], (+$pDst_o+8*1)($pDst)
+	 mov	$X[1], (+$pDst_o+8*2)($pDst)
+
+	 mov	(+8*2)($pA), %rax
+	 mul	%rax
+
+	 add	$A, $X[2]
+	 adc	%rax, $X[3]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $A
+
+	 mov	$X[2], (+$pDst_o+8*3)($pDst)
+	 mov	$X[3], (+$pDst_o+8*4)($pDst)
+
+	 xor	$tmp, $tmp
+	 add	$A, $X[4]
+	 adc	$x6, $X[5]
+	 adc	\$0, $tmp
+
+	 mov	$X[4], (+$pDst_o+8*5)($pDst)
+	 mov	$X[5], (+$pDst_o+8*6)($pDst)
+
+	# %%tmp has 0/1 in column 7
+	# %%A6 has a full value in column 7
+
+	 mov	(+$pDst_o+8*7)($pDst), $X[0]
+	 mov	(+$pDst_o+8*8)($pDst), $X[1]
+	 mov	(+$pDst_o+8*9)($pDst), $X[2]
+	 mov	(+$pDst_o+8*10)($pDst), $X[3]
+	 mov	(+$pDst_o+8*11)($pDst), $X[4]
+	 mov	(+$pDst_o+8*12)($pDst), $X[5]
+	 mov	(+$pDst_o+8*13)($pDst), $x6
+	 mov	(+$pDst_o+8*14)($pDst), $x7
+
+	 mov	$X[7], %rax
+	 mul	%rax
+	 mov	%rax, $X[7]
+	 mov	%rdx, $A
+
+	 add	$X[0], $X[0]
+	 adc	$X[1], $X[1]
+	 adc	$X[2], $X[2]
+	 adc	$X[3], $X[3]
+	 adc	$X[4], $X[4]
+	 adc	$X[5], $X[5]
+	 adc	$x6, $x6
+	 adc	$x7, $x7
+	 adc	\$0, $A
+
+	 add	$tmp, $X[0]
+
+	 mov	(+8*4)($pA), %rax
+	 mul	%rax
+
+	 add	$X[6], $X[0]
+	 adc	%rax, $X[1]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $tmp
+
+	 mov	$X[0], (+$pDst_o+8*7)($pDst)
+	 mov	$X[1], (+$pDst_o+8*8)($pDst)
+
+	 mov	(+8*5)($pA), %rax
+	 mul	%rax
+
+	 add	$tmp, $X[2]
+	 adc	%rax, $X[3]
+	 adc	\$0, %rdx
+
+	 mov	%rdx, $tmp
+
+	 mov	$X[2], (+$pDst_o+8*9)($pDst)
+	 mov	$X[3], (+$pDst_o+8*10)($pDst)
+
+	 mov	(+8*6)($pA), %rax
+	 mul	%rax
+
+	 add	$tmp, $X[4]
+	 adc	%rax, $X[5]
+	 adc	\$0, %rdx
+
+	 mov	$X[4], (+$pDst_o+8*11)($pDst)
+	 mov	$X[5], (+$pDst_o+8*12)($pDst)
+
+	 add	%rdx, $x6
+	 adc	$X[7], $x7
+	 adc	\$0, $A
+
+	 mov	$x6, (+$pDst_o+8*13)($pDst)
+	 mov	$x7, (+$pDst_o+8*14)($pDst)
+	 mov	$A, (+$pDst_o+8*15)($pDst)
+___
+}
+
+#
+# sqr_reduce: subroutine to compute Result = reduce(Result * Result)
+#
+# input and result also in: r9, r8, r15, r14, r13, r12, r11, r10
+#
+$code.=<<___;
+.type	sqr_reduce,\@abi-omnipotent
+.align	16
+sqr_reduce:
+	 mov	(+$pResult_offset+8)(%rsp), %rcx
+___
+	&SQR_512("%rsp+$tmp16_offset+8", "%rcx", [map("%r$_",(10..15,8..9))], "%rbx", "%rbp", "%rsi", "%rdi");
+$code.=<<___;
+	# tail recursion optimization: jmp to mont_reduce and return from there
+	 jmp	mont_reduce
+	# call	mont_reduce
+	# ret
+.size	sqr_reduce,.-sqr_reduce
+___
+}}}
+
+#
+# MAIN FUNCTION
+#
+
+#mod_exp_512(UINT64 *result, /* 512 bits, 8 qwords */
+#           UINT64 *g,   /* 512 bits, 8 qwords */
+#           UINT64 *exp, /* 512 bits, 8 qwords */
+#           struct mod_ctx_512 *data)
+
+# window size = 5
+# table size = 2^5 = 32
+#table_entries	equ	32
+#table_size	equ	table_entries * 8
+$code.=<<___;
+.globl	mod_exp_512
+.type	mod_exp_512,\@function,4
+mod_exp_512:
+	 push	%rbp
+	 push	%rbx
+	 push	%r12
+	 push	%r13
+	 push	%r14
+	 push	%r15
+
+	# adjust stack down and then align it with cache boundary
+	 mov	%rsp, %r8
+	 sub	\$$mem_size, %rsp
+	 and	\$-64, %rsp
+
+	# store previous stack pointer and arguments
+	 mov	%r8, (+$rsp_offset)(%rsp)
+	 mov	%rdi, (+$pResult_offset)(%rsp)
+	 mov	%rsi, (+$pG_offset)(%rsp)
+	 mov	%rcx, (+$pData_offset)(%rsp)
+.Lbody:
+	# transform g into montgomery space
+	# GT = reduce(g * C2) = reduce(g * (2^256))
+	# reduce expects to have the input in [tmp16]
+	 pxor	%xmm4, %xmm4
+	 movdqu	(+16*0)(%rsi), %xmm0
+	 movdqu	(+16*1)(%rsi), %xmm1
+	 movdqu	(+16*2)(%rsi), %xmm2
+	 movdqu	(+16*3)(%rsi), %xmm3
+	 movdqa	%xmm4, (+$tmp16_offset+16*0)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*1)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*6)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*7)(%rsp)
+	 movdqa	%xmm0, (+$tmp16_offset+16*2)(%rsp)
+	 movdqa	%xmm1, (+$tmp16_offset+16*3)(%rsp)
+	 movdqa	%xmm2, (+$tmp16_offset+16*4)(%rsp)
+	 movdqa	%xmm3, (+$tmp16_offset+16*5)(%rsp)
+
+	# load pExp before rdx gets blown away
+	 movdqu	(+16*0)(%rdx), %xmm0
+	 movdqu	(+16*1)(%rdx), %xmm1
+	 movdqu	(+16*2)(%rdx), %xmm2
+	 movdqu	(+16*3)(%rdx), %xmm3
+
+	 lea	(+$GT_offset)(%rsp), %rbx
+	 mov	%rbx, (+$red_result_addr_offset)(%rsp)
+	 call	mont_reduce
+
+	# Initialize tmp = C
+	 lea	(+$tmp_offset)(%rsp), %rcx
+	 xor	%rax, %rax
+	 mov	%rax, (+8*0)(%rcx)
+	 mov	%rax, (+8*1)(%rcx)
+	 mov	%rax, (+8*3)(%rcx)
+	 mov	%rax, (+8*4)(%rcx)
+	 mov	%rax, (+8*5)(%rcx)
+	 mov	%rax, (+8*6)(%rcx)
+	 mov	%rax, (+8*7)(%rcx)
+	 mov	%rax, (+$exp_offset+8*8)(%rsp)
+	 movq	\$1, (+8*2)(%rcx)
+
+	 lea	(+$garray_offset)(%rsp), %rbp
+	 mov	%rcx, %rsi			# pTmp
+	 mov	%rbp, %rdi			# Garray[][0]
+___
+
+	&swizzle("%rdi", "%rcx", "%rax", "%rbx");
+
+	# for (rax = 31; rax != 0; rax--) {
+	#     tmp = reduce(tmp * G)
+	#     swizzle(pg, tmp);
+	#     pg += 2; }
+$code.=<<___;
+	 mov	\$31, %rax
+	 mov	%rax, (+$i_offset)(%rsp)
+	 mov	%rbp, (+$pg_offset)(%rsp)
+	# rsi -> pTmp
+	 mov	%rsi, (+$red_result_addr_offset)(%rsp)
+	 mov	(+8*0)(%rsi), %r10
+	 mov	(+8*1)(%rsi), %r11
+	 mov	(+8*2)(%rsi), %r12
+	 mov	(+8*3)(%rsi), %r13
+	 mov	(+8*4)(%rsi), %r14
+	 mov	(+8*5)(%rsi), %r15
+	 mov	(+8*6)(%rsi), %r8
+	 mov	(+8*7)(%rsi), %r9
+init_loop:
+	 lea	(+$GT_offset)(%rsp), %rdi
+	 call	mont_mul_a3b
+	 lea	(+$tmp_offset)(%rsp), %rsi
+	 mov	(+$pg_offset)(%rsp), %rbp
+	 add	\$2, %rbp
+	 mov	%rbp, (+$pg_offset)(%rsp)
+	 mov	%rsi, %rcx			# rcx = rsi = addr of tmp
+___
+
+	&swizzle("%rbp", "%rcx", "%rax", "%rbx");
+$code.=<<___;
+	 mov	(+$i_offset)(%rsp), %rax
+	 sub	\$1, %rax
+	 mov	%rax, (+$i_offset)(%rsp)
+	 jne	init_loop
+
+	#
+	# Copy exponent onto stack
+	 movdqa	%xmm0, (+$exp_offset+16*0)(%rsp)
+	 movdqa	%xmm1, (+$exp_offset+16*1)(%rsp)
+	 movdqa	%xmm2, (+$exp_offset+16*2)(%rsp)
+	 movdqa	%xmm3, (+$exp_offset+16*3)(%rsp)
+
+
+	#
+	# Do exponentiation
+	# Initialize result to G[exp{511:507}]
+	 mov	(+$exp_offset+62)(%rsp), %eax
+	 mov	%rax, %rdx
+	 shr	\$11, %rax
+	 and	\$0x07FF, %edx
+	 mov	%edx, (+$exp_offset+62)(%rsp)
+	 lea	(+$garray_offset)(%rsp,%rax,2), %rsi
+	 mov	(+$pResult_offset)(%rsp), %rdx
+___
+
+	&unswizzle("%rdx", "%rsi", "%rbp", "%rbx", "%rax");
+
+	#
+	# Loop variables
+	# rcx = [loop_idx] = index: 510-5 to 0 by 5
+$code.=<<___;
+	 movq	\$505, (+$loop_idx_offset)(%rsp)
+
+	 mov	(+$pResult_offset)(%rsp), %rcx
+	 mov	%rcx, (+$red_result_addr_offset)(%rsp)
+	 mov	(+8*0)(%rcx), %r10
+	 mov	(+8*1)(%rcx), %r11
+	 mov	(+8*2)(%rcx), %r12
+	 mov	(+8*3)(%rcx), %r13
+	 mov	(+8*4)(%rcx), %r14
+	 mov	(+8*5)(%rcx), %r15
+	 mov	(+8*6)(%rcx), %r8
+	 mov	(+8*7)(%rcx), %r9
+	 jmp	sqr_2
+
+main_loop_a3b:
+	 call	sqr_reduce
+	 call	sqr_reduce
+	 call	sqr_reduce
+sqr_2:
+	 call	sqr_reduce
+	 call	sqr_reduce
+
+	#
+	# Do multiply, first look up proper value in Garray
+	 mov	(+$loop_idx_offset)(%rsp), %rcx			# bit index
+	 mov	%rcx, %rax
+	 shr	\$4, %rax			# rax is word pointer
+	 mov	(+$exp_offset)(%rsp,%rax,2), %edx
+	 and	\$15, %rcx
+	 shrq	%cl, %rdx
+	 and	\$0x1F, %rdx
+
+	 lea	(+$garray_offset)(%rsp,%rdx,2), %rsi
+	 lea	(+$tmp_offset)(%rsp), %rdx
+	 mov	%rdx, %rdi
+___
+
+	&unswizzle("%rdx", "%rsi", "%rbp", "%rbx", "%rax");
+	# rdi = tmp = pG
+
+	#
+	# Call mod_mul_a1(pDst,  pSrc1, pSrc2, pM, pData)
+	#                 result result pG     M   Data
+$code.=<<___;
+	 mov	(+$pResult_offset)(%rsp), %rsi
+	 call	mont_mul_a3b
+
+	#
+	# finish loop
+	 mov	(+$loop_idx_offset)(%rsp), %rcx
+	 sub	\$5, %rcx
+	 mov	%rcx, (+$loop_idx_offset)(%rsp)
+	 jge	main_loop_a3b
+
+	#
+
+end_main_loop_a3b:
+	# transform result out of Montgomery space
+	# result = reduce(result)
+	 mov	(+$pResult_offset)(%rsp), %rdx
+	 pxor	%xmm4, %xmm4
+	 movdqu	(+16*0)(%rdx), %xmm0
+	 movdqu	(+16*1)(%rdx), %xmm1
+	 movdqu	(+16*2)(%rdx), %xmm2
+	 movdqu	(+16*3)(%rdx), %xmm3
+	 movdqa	%xmm4, (+$tmp16_offset+16*4)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*5)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*6)(%rsp)
+	 movdqa	%xmm4, (+$tmp16_offset+16*7)(%rsp)
+	 movdqa	%xmm0, (+$tmp16_offset+16*0)(%rsp)
+	 movdqa	%xmm1, (+$tmp16_offset+16*1)(%rsp)
+	 movdqa	%xmm2, (+$tmp16_offset+16*2)(%rsp)
+	 movdqa	%xmm3, (+$tmp16_offset+16*3)(%rsp)
+	 call	mont_reduce
+
+	# If result > m, subract m
+	# load result into r15:r8
+	 mov	(+$pResult_offset)(%rsp), %rax
+	 mov	(+8*0)(%rax), %r8
+	 mov	(+8*1)(%rax), %r9
+	 mov	(+8*2)(%rax), %r10
+	 mov	(+8*3)(%rax), %r11
+	 mov	(+8*4)(%rax), %r12
+	 mov	(+8*5)(%rax), %r13
+	 mov	(+8*6)(%rax), %r14
+	 mov	(+8*7)(%rax), %r15
+
+	# subtract m
+	 mov	(+$pData_offset)(%rsp), %rbx
+	 add	\$$M, %rbx
+
+	 sub	(+8*0)(%rbx), %r8
+	 sbb	(+8*1)(%rbx), %r9
+	 sbb	(+8*2)(%rbx), %r10
+	 sbb	(+8*3)(%rbx), %r11
+	 sbb	(+8*4)(%rbx), %r12
+	 sbb	(+8*5)(%rbx), %r13
+	 sbb	(+8*6)(%rbx), %r14
+	 sbb	(+8*7)(%rbx), %r15
+
+	# if Carry is clear, replace result with difference
+	 mov	(+8*0)(%rax), %rsi
+	 mov	(+8*1)(%rax), %rdi
+	 mov	(+8*2)(%rax), %rcx
+	 mov	(+8*3)(%rax), %rdx
+	 cmovnc	%r8, %rsi
+	 cmovnc	%r9, %rdi
+	 cmovnc	%r10, %rcx
+	 cmovnc	%r11, %rdx
+	 mov	%rsi, (+8*0)(%rax)
+	 mov	%rdi, (+8*1)(%rax)
+	 mov	%rcx, (+8*2)(%rax)
+	 mov	%rdx, (+8*3)(%rax)
+
+	 mov	(+8*4)(%rax), %rsi
+	 mov	(+8*5)(%rax), %rdi
+	 mov	(+8*6)(%rax), %rcx
+	 mov	(+8*7)(%rax), %rdx
+	 cmovnc	%r12, %rsi
+	 cmovnc	%r13, %rdi
+	 cmovnc	%r14, %rcx
+	 cmovnc	%r15, %rdx
+	 mov	%rsi, (+8*4)(%rax)
+	 mov	%rdi, (+8*5)(%rax)
+	 mov	%rcx, (+8*6)(%rax)
+	 mov	%rdx, (+8*7)(%rax)
+
+	 mov	(+$rsp_offset)(%rsp), %rsi
+	 mov	0(%rsi),%r15
+	 mov	8(%rsi),%r14
+	 mov	16(%rsi),%r13
+	 mov	24(%rsi),%r12
+	 mov	32(%rsi),%rbx
+	 mov	40(%rsi),%rbp
+	 lea	48(%rsi),%rsp
+.Lepilogue:
+	 ret
+.size mod_exp_512, . - mod_exp_512
+___
+
+if ($win64) {
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+my $rec="%rcx";
+my $frame="%rdx";
+my $context="%r8";
+my $disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	mod_exp_512_se_handler,\@abi-omnipotent
+.align	16
+mod_exp_512_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lbody(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	lea	.Lepilogue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_prologue
+
+	mov	$rsp_offset(%rax),%rax	# pull saved Rsp
+
+	mov	32(%rax),%rbx
+	mov	40(%rax),%rbp
+	mov	24(%rax),%r12
+	mov	16(%rax),%r13
+	mov	8(%rax),%r14
+	mov	0(%rax),%r15
+	lea	48(%rax),%rax
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	mod_exp_512_se_handler,.-mod_exp_512_se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_mod_exp_512
+	.rva	.LSEH_end_mod_exp_512
+	.rva	.LSEH_info_mod_exp_512
+
+.section	.xdata
+.align	8
+.LSEH_info_mod_exp_512:
+	.byte	9,0,0,0
+	.rva	mod_exp_512_se_handler
+___
+}
+
+sub reg_part {
+my ($reg,$conv)=@_;
+    if ($reg =~ /%r[0-9]+/)	{ $reg .= $conv; }
+    elsif ($conv eq "b")	{ $reg =~ s/%[er]([^x]+)x?/%$1l/;	}
+    elsif ($conv eq "w")	{ $reg =~ s/%[er](.+)/%$1/;		}
+    elsif ($conv eq "d")	{ $reg =~ s/%[er](.+)/%e$1/;		}
+    return $reg;
+}
+
+$code =~ s/(%[a-z0-9]+)#([bwd])/reg_part($1,$2)/gem;
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/(\(\+[^)]+\))/eval $1/gem;
+print $code;
+close STDOUT;
diff --git a/crypto/engine/Makefile b/crypto/engine/Makefile
index 2ee6c72362..d29bdd09a0 100644
--- a/crypto/engine/Makefile
+++ b/crypto/engine/Makefile
@@ -22,13 +22,13 @@ LIBSRC= eng_err.c eng_lib.c eng_list.c eng_init.c eng_ctrl.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_rdrand.c
+	eng_rsax.c eng_rdrand.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_rdrand.o
+	eng_rsax.o eng_rdrand.o
 
 SRC= $(LIBSRC)
 
@@ -265,6 +265,20 @@ eng_rdrand.o: ../../include/openssl/safestack.h ../../include/openssl/sha.h
 eng_rdrand.o: ../../include/openssl/stack.h ../../include/openssl/symhacks.h
 eng_rdrand.o: ../../include/openssl/x509.h ../../include/openssl/x509_vfy.h
 eng_rdrand.o: eng_rdrand.c
+eng_rsax.o: ../../include/openssl/asn1.h ../../include/openssl/bio.h
+eng_rsax.o: ../../include/openssl/bn.h ../../include/openssl/buffer.h
+eng_rsax.o: ../../include/openssl/crypto.h ../../include/openssl/e_os2.h
+eng_rsax.o: ../../include/openssl/ec.h ../../include/openssl/ecdh.h
+eng_rsax.o: ../../include/openssl/ecdsa.h ../../include/openssl/engine.h
+eng_rsax.o: ../../include/openssl/err.h ../../include/openssl/evp.h
+eng_rsax.o: ../../include/openssl/lhash.h ../../include/openssl/obj_mac.h
+eng_rsax.o: ../../include/openssl/objects.h ../../include/openssl/opensslconf.h
+eng_rsax.o: ../../include/openssl/opensslv.h ../../include/openssl/ossl_typ.h
+eng_rsax.o: ../../include/openssl/pkcs7.h ../../include/openssl/rsa.h
+eng_rsax.o: ../../include/openssl/safestack.h ../../include/openssl/sha.h
+eng_rsax.o: ../../include/openssl/stack.h ../../include/openssl/symhacks.h
+eng_rsax.o: ../../include/openssl/x509.h ../../include/openssl/x509_vfy.h
+eng_rsax.o: eng_rsax.c
 eng_table.o: ../../e_os.h ../../include/openssl/asn1.h
 eng_table.o: ../../include/openssl/bio.h ../../include/openssl/buffer.h
 eng_table.o: ../../include/openssl/crypto.h ../../include/openssl/e_os2.h
diff --git a/crypto/engine/eng_all.c b/crypto/engine/eng_all.c
index 195a3a9554..7edf12e7e1 100644
--- a/crypto/engine/eng_all.c
+++ b/crypto/engine/eng_all.c
@@ -76,6 +76,9 @@ void ENGINE_load_builtin_engines(void)
 #if !defined(OPENSSL_NO_HW) && (defined(__OpenBSD__) || defined(__FreeBSD__) || defined(HAVE_CRYPTODEV))
     ENGINE_load_cryptodev();
 #endif
+#ifndef OPENSSL_NO_RSAX
+    ENGINE_load_rsax();
+#endif
 #ifndef OPENSSL_NO_RDRAND
     ENGINE_load_rdrand();
 #endif
diff --git a/crypto/engine/eng_rsax.c b/crypto/engine/eng_rsax.c
new file mode 100644
index 0000000000..8362754c7a
--- /dev/null
+++ b/crypto/engine/eng_rsax.c
@@ -0,0 +1,701 @@
+/* crypto/engine/eng_rsax.c */
+/* Copyright (c) 2010-2010 Intel Corp.
+ *   Author: Vinodh.Gopal@intel.com
+ *           Jim Guilford
+ *           Erdinc.Ozturk@intel.com
+ *           Maxim.Perminov@intel.com
+ *           Ying.Huang@intel.com
+ *
+ * More information about algorithm used can be found at:
+ *   http://www.cse.buffalo.edu/srds2009/escs2009_submission_Gopal.pdf
+ */
+/* ====================================================================
+ * 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>
+
+#include <stdio.h>
+#include <string.h>
+#include <openssl/crypto.h>
+#include <openssl/buffer.h>
+#include <openssl/engine.h>
+#ifndef OPENSSL_NO_RSA
+# include <openssl/rsa.h>
+#endif
+#include <openssl/bn.h>
+#include <openssl/err.h>
+
+/* RSAX is available **ONLY* on x86_64 CPUs */
+#undef COMPILE_RSAX
+
+#if (defined(__x86_64) || defined(__x86_64__) || \
+     defined(_M_AMD64) || defined (_M_X64)) && !defined(OPENSSL_NO_ASM)
+# define COMPILE_RSAX
+static ENGINE *ENGINE_rsax(void);
+#endif
+
+void ENGINE_load_rsax(void)
+{
+/* On non-x86 CPUs it just returns. */
+#ifdef COMPILE_RSAX
+    ENGINE *toadd = ENGINE_rsax();
+    if (!toadd)
+        return;
+    ENGINE_add(toadd);
+    ENGINE_free(toadd);
+    ERR_clear_error();
+#endif
+}
+
+#ifdef COMPILE_RSAX
+# define E_RSAX_LIB_NAME "rsax engine"
+
+static int e_rsax_destroy(ENGINE *e);
+static int e_rsax_init(ENGINE *e);
+static int e_rsax_finish(ENGINE *e);
+static int e_rsax_ctrl(ENGINE *e, int cmd, long i, void *p, void (*f) (void));
+
+# ifndef OPENSSL_NO_RSA
+/* RSA stuff */
+static int e_rsax_rsa_mod_exp(BIGNUM *r, const BIGNUM *I, RSA *rsa,
+                              BN_CTX *ctx);
+static int e_rsax_rsa_finish(RSA *r);
+# endif
+
+static const ENGINE_CMD_DEFN e_rsax_cmd_defns[] = {
+    {0, NULL, NULL, 0}
+};
+
+# ifndef OPENSSL_NO_RSA
+/* Our internal RSA_METHOD that we provide pointers to */
+static RSA_METHOD e_rsax_rsa = {
+    "Intel RSA-X method",
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    e_rsax_rsa_mod_exp,
+    NULL,
+    NULL,
+    e_rsax_rsa_finish,
+    RSA_FLAG_CACHE_PUBLIC | RSA_FLAG_CACHE_PRIVATE,
+    NULL,
+    NULL,
+    NULL
+};
+# endif
+
+/* Constants used when creating the ENGINE */
+static const char *engine_e_rsax_id = "rsax";
+static const char *engine_e_rsax_name = "RSAX engine support";
+
+/* This internal function is used by ENGINE_rsax() */
+static int bind_helper(ENGINE *e)
+{
+# ifndef OPENSSL_NO_RSA
+    const RSA_METHOD *meth1;
+# endif
+    if (!ENGINE_set_id(e, engine_e_rsax_id) ||
+        !ENGINE_set_name(e, engine_e_rsax_name) ||
+# ifndef OPENSSL_NO_RSA
+        !ENGINE_set_RSA(e, &e_rsax_rsa) ||
+# endif
+        !ENGINE_set_destroy_function(e, e_rsax_destroy) ||
+        !ENGINE_set_init_function(e, e_rsax_init) ||
+        !ENGINE_set_finish_function(e, e_rsax_finish) ||
+        !ENGINE_set_ctrl_function(e, e_rsax_ctrl) ||
+        !ENGINE_set_cmd_defns(e, e_rsax_cmd_defns))
+        return 0;
+
+# ifndef OPENSSL_NO_RSA
+    meth1 = RSA_PKCS1_SSLeay();
+    e_rsax_rsa.rsa_pub_enc = meth1->rsa_pub_enc;
+    e_rsax_rsa.rsa_pub_dec = meth1->rsa_pub_dec;
+    e_rsax_rsa.rsa_priv_enc = meth1->rsa_priv_enc;
+    e_rsax_rsa.rsa_priv_dec = meth1->rsa_priv_dec;
+    e_rsax_rsa.bn_mod_exp = meth1->bn_mod_exp;
+# endif
+    return 1;
+}
+
+static ENGINE *ENGINE_rsax(void)
+{
+    ENGINE *ret = ENGINE_new();
+    if (!ret)
+        return NULL;
+    if (!bind_helper(ret)) {
+        ENGINE_free(ret);
+        return NULL;
+    }
+    return ret;
+}
+
+# ifndef OPENSSL_NO_RSA
+/* Used to attach our own key-data to an RSA structure */
+static int rsax_ex_data_idx = -1;
+# endif
+
+static int e_rsax_destroy(ENGINE *e)
+{
+    return 1;
+}
+
+/* (de)initialisation functions. */
+static int e_rsax_init(ENGINE *e)
+{
+# ifndef OPENSSL_NO_RSA
+    if (rsax_ex_data_idx == -1)
+        rsax_ex_data_idx = RSA_get_ex_new_index(0, NULL, NULL, NULL, NULL);
+# endif
+    if (rsax_ex_data_idx == -1)
+        return 0;
+    return 1;
+}
+
+static int e_rsax_finish(ENGINE *e)
+{
+    return 1;
+}
+
+static int e_rsax_ctrl(ENGINE *e, int cmd, long i, void *p, void (*f) (void))
+{
+    int to_return = 1;
+
+    switch (cmd) {
+        /* The command isn't understood by this engine */
+    default:
+        to_return = 0;
+        break;
+    }
+
+    return to_return;
+}
+
+# ifndef OPENSSL_NO_RSA
+
+#  ifdef _WIN32
+typedef unsigned __int64 UINT64;
+#  else
+typedef unsigned long long UINT64;
+#  endif
+typedef unsigned short UINT16;
+
+/*
+ * Table t is interleaved in the following manner: The order in memory is
+ * t[0][0], t[0][1], ..., t[0][7], t[1][0], ... A particular 512-bit value is
+ * stored in t[][index] rather than the more normal t[index][]; i.e. the
+ * qwords of a particular entry in t are not adjacent in memory
+ */
+
+/* Init BIGNUM b from the interleaved UINT64 array */
+static int interleaved_array_to_bn_512(BIGNUM *b, UINT64 *array);
+
+/*
+ * Extract array elements from BIGNUM b To set the whole array from b, call
+ * with n=8
+ */
+static int bn_extract_to_array_512(const BIGNUM *b, unsigned int n,
+                                   UINT64 *array);
+
+struct mod_ctx_512 {
+    UINT64 t[8][8];
+    UINT64 m[8];
+    UINT64 m1[8];               /* 2^278 % m */
+    UINT64 m2[8];               /* 2^640 % m */
+    UINT64 k1[2];               /* (- 1/m) % 2^128 */
+};
+
+static int mod_exp_pre_compute_data_512(UINT64 *m, struct mod_ctx_512 *data);
+
+void mod_exp_512(UINT64 *result, /* 512 bits, 8 qwords */
+                 UINT64 *g,     /* 512 bits, 8 qwords */
+                 UINT64 *exp,   /* 512 bits, 8 qwords */
+                 struct mod_ctx_512 *data);
+
+typedef struct st_e_rsax_mod_ctx {
+    UINT64 type;
+    union {
+        struct mod_ctx_512 b512;
+    } ctx;
+
+} E_RSAX_MOD_CTX;
+
+static E_RSAX_MOD_CTX *e_rsax_get_ctx(RSA *rsa, int idx, BIGNUM *m)
+{
+    E_RSAX_MOD_CTX *hptr;
+
+    if (idx < 0 || idx > 2)
+        return NULL;
+
+    hptr = RSA_get_ex_data(rsa, rsax_ex_data_idx);
+    if (!hptr) {
+        hptr = OPENSSL_malloc(3 * sizeof(E_RSAX_MOD_CTX));
+        if (!hptr)
+            return NULL;
+        hptr[2].type = hptr[1].type = hptr[0].type = 0;
+        RSA_set_ex_data(rsa, rsax_ex_data_idx, hptr);
+    }
+
+    if (hptr[idx].type == (UINT64)BN_num_bits(m))
+        return hptr + idx;
+
+    if (BN_num_bits(m) == 512) {
+        UINT64 _m[8];
+        bn_extract_to_array_512(m, 8, _m);
+        memset(&hptr[idx].ctx.b512, 0, sizeof(struct mod_ctx_512));
+        mod_exp_pre_compute_data_512(_m, &hptr[idx].ctx.b512);
+    }
+
+    hptr[idx].type = BN_num_bits(m);
+    return hptr + idx;
+}
+
+static int e_rsax_rsa_finish(RSA *rsa)
+{
+    E_RSAX_MOD_CTX *hptr = RSA_get_ex_data(rsa, rsax_ex_data_idx);
+    if (hptr) {
+        OPENSSL_free(hptr);
+        RSA_set_ex_data(rsa, rsax_ex_data_idx, NULL);
+    }
+    if (rsa->_method_mod_n)
+        BN_MONT_CTX_free(rsa->_method_mod_n);
+    if (rsa->_method_mod_p)
+        BN_MONT_CTX_free(rsa->_method_mod_p);
+    if (rsa->_method_mod_q)
+        BN_MONT_CTX_free(rsa->_method_mod_q);
+    return 1;
+}
+
+static int e_rsax_bn_mod_exp(BIGNUM *r, const BIGNUM *g, const BIGNUM *e,
+                             const BIGNUM *m, BN_CTX *ctx,
+                             BN_MONT_CTX *in_mont,
+                             E_RSAX_MOD_CTX *rsax_mod_ctx)
+{
+    if (rsax_mod_ctx && BN_get_flags(e, BN_FLG_CONSTTIME) != 0) {
+        if (BN_num_bits(m) == 512) {
+            UINT64 _r[8];
+            UINT64 _g[8];
+            UINT64 _e[8];
+
+            /* Init the arrays from the BIGNUMs */
+            bn_extract_to_array_512(g, 8, _g);
+            bn_extract_to_array_512(e, 8, _e);
+
+            mod_exp_512(_r, _g, _e, &rsax_mod_ctx->ctx.b512);
+            /* Return the result in the BIGNUM */
+            interleaved_array_to_bn_512(r, _r);
+            return 1;
+        }
+    }
+
+    return BN_mod_exp_mont(r, g, e, m, ctx, in_mont);
+}
+
+/*
+ * Declares for the Intel CIAP 512-bit / CRT / 1024 bit RSA modular
+ * exponentiation routine precalculations and a structure to hold the
+ * necessary values.  These files are meant to live in crypto/rsa/ in the
+ * target openssl.
+ */
+
+/*
+ * Local method: extracts a piece from a BIGNUM, to fit it into
+ * an array. Call with n=8 to extract an entire 512-bit BIGNUM
+ */
+static int bn_extract_to_array_512(const BIGNUM *b, unsigned int n,
+                                   UINT64 *array)
+{
+    int i;
+    UINT64 tmp;
+    unsigned char bn_buff[64];
+    memset(bn_buff, 0, 64);
+    if (BN_num_bytes(b) > 64) {
+        printf("Can't support this byte size\n");
+        return 0;
+    }
+    if (BN_num_bytes(b) != 0) {
+        if (!BN_bn2bin(b, bn_buff + (64 - BN_num_bytes(b)))) {
+            printf("Error's in bn2bin\n");
+            /* We have to error, here */
+            return 0;
+        }
+    }
+    while (n-- > 0) {
+        array[n] = 0;
+        for (i = 7; i >= 0; i--) {
+            tmp = bn_buff[63 - (n * 8 + i)];
+            array[n] |= tmp << (8 * i);
+        }
+    }
+    return 1;
+}
+
+/* Init a 512-bit BIGNUM from the UINT64*_ (8 * 64) interleaved array */
+static int interleaved_array_to_bn_512(BIGNUM *b, UINT64 *array)
+{
+    unsigned char tmp[64];
+    int n = 8;
+    int i;
+    while (n-- > 0) {
+        for (i = 7; i >= 0; i--) {
+            tmp[63 - (n * 8 + i)] = (unsigned char)(array[n] >> (8 * i));
+    }}
+    BN_bin2bn(tmp, 64, b);
+    return 0;
+}
+
+/* The main 512bit precompute call */
+static int mod_exp_pre_compute_data_512(UINT64 *m, struct mod_ctx_512 *data)
+{
+    BIGNUM two_768, two_640, two_128, two_512, tmp, _m, tmp2;
+
+    /* We need a BN_CTX for the modulo functions */
+    BN_CTX *ctx;
+    /* Some tmps */
+    UINT64 _t[8];
+    int i, j, ret = 0;
+
+    /* Init _m with m */
+    BN_init(&_m);
+    interleaved_array_to_bn_512(&_m, m);
+    memset(_t, 0, 64);
+
+    /* Inits */
+    BN_init(&two_768);
+    BN_init(&two_640);
+    BN_init(&two_128);
+    BN_init(&two_512);
+    BN_init(&tmp);
+    BN_init(&tmp2);
+
+    /* Create our context */
+    if ((ctx = BN_CTX_new()) == NULL) {
+        goto err;
+    }
+    BN_CTX_start(ctx);
+
+    /*
+     * For production, if you care, these only need to be set once,
+     * and may be made constants.
+     */
+    BN_lshift(&two_768, BN_value_one(), 768);
+    BN_lshift(&two_640, BN_value_one(), 640);
+    BN_lshift(&two_128, BN_value_one(), 128);
+    BN_lshift(&two_512, BN_value_one(), 512);
+
+    if (0 == (m[7] & 0x8000000000000000)) {
+        exit(1);
+    }
+    if (0 == (m[0] & 0x1)) {    /* Odd modulus required for Mont */
+        exit(1);
+    }
+
+    /* Precompute m1 */
+    BN_mod(&tmp, &two_768, &_m, ctx);
+    if (!bn_extract_to_array_512(&tmp, 8, &data->m1[0])) {
+        goto err;
+    }
+
+    /* Precompute m2 */
+    BN_mod(&tmp, &two_640, &_m, ctx);
+    if (!bn_extract_to_array_512(&tmp, 8, &data->m2[0])) {
+        goto err;
+    }
+
+    /*
+     * Precompute k1, a 128b number = ((-1)* m-1 ) mod 2128; k1 should
+     * be non-negative.
+     */
+    BN_mod_inverse(&tmp, &_m, &two_128, ctx);
+    if (!BN_is_zero(&tmp)) {
+        BN_sub(&tmp, &two_128, &tmp);
+    }
+    if (!bn_extract_to_array_512(&tmp, 2, &data->k1[0])) {
+        goto err;
+    }
+
+    /* Precompute t */
+    for (i = 0; i < 8; i++) {
+        BN_zero(&tmp);
+        if (i & 1) {
+            BN_add(&tmp, &two_512, &tmp);
+        }
+        if (i & 2) {
+            BN_add(&tmp, &two_512, &tmp);
+        }
+        if (i & 4) {
+            BN_add(&tmp, &two_640, &tmp);
+        }
+
+        BN_nnmod(&tmp2, &tmp, &_m, ctx);
+        if (!bn_extract_to_array_512(&tmp2, 8, _t)) {
+            goto err;
+        }
+        for (j = 0; j < 8; j++)
+            data->t[j][i] = _t[j];
+    }
+
+    /* Precompute m */
+    for (i = 0; i < 8; i++) {
+        data->m[i] = m[i];
+    }
+
+    ret = 1;
+
+ err:
+    /* Cleanup */
+    if (ctx != NULL) {
+        BN_CTX_end(ctx);
+        BN_CTX_free(ctx);
+    }
+    BN_free(&two_768);
+    BN_free(&two_640);
+    BN_free(&two_128);
+    BN_free(&two_512);
+    BN_free(&tmp);
+    BN_free(&tmp2);
+    BN_free(&_m);
+
+    return ret;
+}
+
+static int e_rsax_rsa_mod_exp(BIGNUM *r0, const BIGNUM *I, RSA *rsa,
+                              BN_CTX *ctx)
+{
+    BIGNUM *r1, *m1, *vrfy;
+    BIGNUM local_dmp1, local_dmq1, local_c, local_r1;
+    BIGNUM *dmp1, *dmq1, *c, *pr1;
+    int ret = 0;
+
+    BN_CTX_start(ctx);
+    r1 = BN_CTX_get(ctx);
+    m1 = BN_CTX_get(ctx);
+    vrfy = BN_CTX_get(ctx);
+
+    {
+        BIGNUM local_p, local_q;
+        BIGNUM *p = NULL, *q = NULL;
+        int error = 0;
+
+        /*
+         * Make sure BN_mod_inverse in Montgomery intialization uses the
+         * BN_FLG_CONSTTIME flag (unless RSA_FLAG_NO_CONSTTIME is set)
+         */
+        if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+            BN_init(&local_p);
+            p = &local_p;
+            BN_with_flags(p, rsa->p, BN_FLG_CONSTTIME);
+
+            BN_init(&local_q);
+            q = &local_q;
+            BN_with_flags(q, rsa->q, BN_FLG_CONSTTIME);
+        } else {
+            p = rsa->p;
+            q = rsa->q;
+        }
+
+        if (rsa->flags & RSA_FLAG_CACHE_PRIVATE) {
+            if (!BN_MONT_CTX_set_locked
+                (&rsa->_method_mod_p, CRYPTO_LOCK_RSA, p, ctx))
+                error = 1;
+            if (!BN_MONT_CTX_set_locked
+                (&rsa->_method_mod_q, CRYPTO_LOCK_RSA, q, ctx))
+                error = 1;
+        }
+
+        /* clean up */
+        if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+            BN_free(&local_p);
+            BN_free(&local_q);
+        }
+        if (error)
+            goto err;
+    }
+
+    if (rsa->flags & RSA_FLAG_CACHE_PUBLIC)
+        if (!BN_MONT_CTX_set_locked
+            (&rsa->_method_mod_n, CRYPTO_LOCK_RSA, rsa->n, ctx))
+            goto err;
+
+    /* compute I mod q */
+    if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+        c = &local_c;
+        BN_with_flags(c, I, BN_FLG_CONSTTIME);
+        if (!BN_mod(r1, c, rsa->q, ctx))
+            goto err;
+    } else {
+        if (!BN_mod(r1, I, rsa->q, ctx))
+            goto err;
+    }
+
+    /* compute r1^dmq1 mod q */
+    if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+        dmq1 = &local_dmq1;
+        BN_with_flags(dmq1, rsa->dmq1, BN_FLG_CONSTTIME);
+    } else
+        dmq1 = rsa->dmq1;
+
+    if (!e_rsax_bn_mod_exp(m1, r1, dmq1, rsa->q, ctx,
+                           rsa->_method_mod_q, e_rsax_get_ctx(rsa, 0,
+                                                              rsa->q)))
+        goto err;
+
+    /* compute I mod p */
+    if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+        c = &local_c;
+        BN_with_flags(c, I, BN_FLG_CONSTTIME);
+        if (!BN_mod(r1, c, rsa->p, ctx))
+            goto err;
+    } else {
+        if (!BN_mod(r1, I, rsa->p, ctx))
+            goto err;
+    }
+
+    /* compute r1^dmp1 mod p */
+    if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+        dmp1 = &local_dmp1;
+        BN_with_flags(dmp1, rsa->dmp1, BN_FLG_CONSTTIME);
+    } else
+        dmp1 = rsa->dmp1;
+
+    if (!e_rsax_bn_mod_exp(r0, r1, dmp1, rsa->p, ctx,
+                           rsa->_method_mod_p, e_rsax_get_ctx(rsa, 1,
+                                                              rsa->p)))
+        goto err;
+
+    if (!BN_sub(r0, r0, m1))
+        goto err;
+    /*
+     * This will help stop the size of r0 increasing, which does affect the
+     * multiply if it optimised for a power of 2 size
+     */
+    if (BN_is_negative(r0))
+        if (!BN_add(r0, r0, rsa->p))
+            goto err;
+
+    if (!BN_mul(r1, r0, rsa->iqmp, ctx))
+        goto err;
+
+    /* Turn BN_FLG_CONSTTIME flag on before division operation */
+    if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+        pr1 = &local_r1;
+        BN_with_flags(pr1, r1, BN_FLG_CONSTTIME);
+    } else
+        pr1 = r1;
+    if (!BN_mod(r0, pr1, rsa->p, ctx))
+        goto err;
+
+    /*
+     * If p < q it is occasionally possible for the correction of adding 'p'
+     * if r0 is negative above to leave the result still negative. This can
+     * break the private key operations: the following second correction
+     * should *always* correct this rare occurrence. This will *never* happen
+     * with OpenSSL generated keys because they ensure p > q [steve]
+     */
+    if (BN_is_negative(r0))
+        if (!BN_add(r0, r0, rsa->p))
+            goto err;
+    if (!BN_mul(r1, r0, rsa->q, ctx))
+        goto err;
+    if (!BN_add(r0, r1, m1))
+        goto err;
+
+    if (rsa->e && rsa->n) {
+        if (!e_rsax_bn_mod_exp
+            (vrfy, r0, rsa->e, rsa->n, ctx, rsa->_method_mod_n,
+             e_rsax_get_ctx(rsa, 2, rsa->n)))
+            goto err;
+
+        /*
+         * If 'I' was greater than (or equal to) rsa->n, the operation will
+         * be equivalent to using 'I mod n'. However, the result of the
+         * verify will *always* be less than 'n' so we don't check for
+         * absolute equality, just congruency.
+         */
+        if (!BN_sub(vrfy, vrfy, I))
+            goto err;
+        if (!BN_mod(vrfy, vrfy, rsa->n, ctx))
+            goto err;
+        if (BN_is_negative(vrfy))
+            if (!BN_add(vrfy, vrfy, rsa->n))
+                goto err;
+        if (!BN_is_zero(vrfy)) {
+            /*
+             * 'I' and 'vrfy' aren't congruent mod n. Don't leak
+             * miscalculated CRT output, just do a raw (slower) mod_exp and
+             * return that instead.
+             */
+
+            BIGNUM local_d;
+            BIGNUM *d = NULL;
+
+            if (!(rsa->flags & RSA_FLAG_NO_CONSTTIME)) {
+                d = &local_d;
+                BN_with_flags(d, rsa->d, BN_FLG_CONSTTIME);
+            } else
+                d = rsa->d;
+            if (!e_rsax_bn_mod_exp(r0, I, d, rsa->n, ctx,
+                                   rsa->_method_mod_n, e_rsax_get_ctx(rsa, 2,
+                                                                      rsa->n)))
+                goto err;
+        }
+    }
+    ret = 1;
+
+ err:
+    BN_CTX_end(ctx);
+
+    return ret;
+}
+# endif                         /* !OPENSSL_NO_RSA */
+#endif                          /* !COMPILE_RSAX */