[thirdparty/openssl.git] / crypto / bn / asm / x86_64-gf2m.pl

#! /usr/bin/env perl
# Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
#
# Licensed under the OpenSSL license (the "License").  You may not use
# this file except in compliance with the License.  You can obtain a copy
# in the file LICENSE in the source distribution or at
# https://www.openssl.org/source/license.html

#
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
#
# May 2011
#
# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
# the time being... Except that it has two code paths: code suitable
# for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
# later. Improvement varies from one benchmark and µ-arch to another.
# Vanilla code path is at most 20% faster than compiler-generated code
# [not very impressive], while PCLMULQDQ - whole 85%-160% better on
# 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
# all CPU time is burnt in it...

$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 OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
*STDOUT=*OUT;

($lo,$hi)=("%rax","%rdx");	$a=$lo;
($i0,$i1)=("%rsi","%rdi");
($t0,$t1)=("%rbx","%rcx");
($b,$mask)=("%rbp","%r8");
($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
($R,$Tx)=("%xmm0","%xmm1");

$code.=<<___;
.text

.type	_mul_1x1,\@abi-omnipotent
.align	16
_mul_1x1:
.cfi_startproc
	sub	\$128+8,%rsp
.cfi_adjust_cfa_offset	128+8
	mov	\$-1,$a1
	lea	($a,$a),$i0
	shr	\$3,$a1
	lea	(,$a,4),$i1
	and	$a,$a1			# a1=a&0x1fffffffffffffff
	lea	(,$a,8),$a8
	sar	\$63,$a			# broadcast 63rd bit
	lea	($a1,$a1),$a2
	sar	\$63,$i0		# broadcast 62nd bit
	lea	(,$a1,4),$a4
	and	$b,$a
	sar	\$63,$i1		# broadcast 61st bit
	mov	$a,$hi			# $a is $lo
	shl	\$63,$lo
	and	$b,$i0
	shr	\$1,$hi
	mov	$i0,$t1
	shl	\$62,$i0
	and	$b,$i1
	shr	\$2,$t1
	xor	$i0,$lo
	mov	$i1,$t0
	shl	\$61,$i1
	xor	$t1,$hi
	shr	\$3,$t0
	xor	$i1,$lo
	xor	$t0,$hi

	mov	$a1,$a12
	movq	\$0,0(%rsp)		# tab[0]=0
	xor	$a2,$a12		# a1^a2
	mov	$a1,8(%rsp)		# tab[1]=a1
	 mov	$a4,$a48
	mov	$a2,16(%rsp)		# tab[2]=a2
	 xor	$a8,$a48		# a4^a8
	mov	$a12,24(%rsp)		# tab[3]=a1^a2

	xor	$a4,$a1
	mov	$a4,32(%rsp)		# tab[4]=a4
	xor	$a4,$a2
	mov	$a1,40(%rsp)		# tab[5]=a1^a4
	xor	$a4,$a12
	mov	$a2,48(%rsp)		# tab[6]=a2^a4
	 xor	$a48,$a1		# a1^a4^a4^a8=a1^a8
	mov	$a12,56(%rsp)		# tab[7]=a1^a2^a4
	 xor	$a48,$a2		# a2^a4^a4^a8=a1^a8

	mov	$a8,64(%rsp)		# tab[8]=a8
	xor	$a48,$a12		# a1^a2^a4^a4^a8=a1^a2^a8
	mov	$a1,72(%rsp)		# tab[9]=a1^a8
	 xor	$a4,$a1			# a1^a8^a4
	mov	$a2,80(%rsp)		# tab[10]=a2^a8
	 xor	$a4,$a2			# a2^a8^a4
	mov	$a12,88(%rsp)		# tab[11]=a1^a2^a8

	xor	$a4,$a12		# a1^a2^a8^a4
	mov	$a48,96(%rsp)		# tab[12]=a4^a8
	 mov	$mask,$i0
	mov	$a1,104(%rsp)		# tab[13]=a1^a4^a8
	 and	$b,$i0
	mov	$a2,112(%rsp)		# tab[14]=a2^a4^a8
	 shr	\$4,$b
	mov	$a12,120(%rsp)		# tab[15]=a1^a2^a4^a8
	 mov	$mask,$i1
	 and	$b,$i1
	 shr	\$4,$b

	movq	(%rsp,$i0,8),$R		# half of calculations is done in SSE2
	mov	$mask,$i0
	and	$b,$i0
	shr	\$4,$b
___
    for ($n=1;$n<8;$n++) {
	$code.=<<___;
	mov	(%rsp,$i1,8),$t1
	mov	$mask,$i1
	mov	$t1,$t0
	shl	\$`8*$n-4`,$t1
	and	$b,$i1
	 movq	(%rsp,$i0,8),$Tx
	shr	\$`64-(8*$n-4)`,$t0
	xor	$t1,$lo
	 pslldq	\$$n,$Tx
	 mov	$mask,$i0
	shr	\$4,$b
	xor	$t0,$hi
	 and	$b,$i0
	 shr	\$4,$b
	 pxor	$Tx,$R
___
    }
$code.=<<___;
	mov	(%rsp,$i1,8),$t1
	mov	$t1,$t0
	shl	\$`8*$n-4`,$t1
	movq	$R,$i0
	shr	\$`64-(8*$n-4)`,$t0
	xor	$t1,$lo
	psrldq	\$8,$R
	xor	$t0,$hi
	movq	$R,$i1
	xor	$i0,$lo
	xor	$i1,$hi

	add	\$128+8,%rsp
.cfi_adjust_cfa_offset	-128-8
	ret
.Lend_mul_1x1:
.cfi_endproc
.size	_mul_1x1,.-_mul_1x1
___

($rp,$a1,$a0,$b1,$b0) = $win64?	("%rcx","%rdx","%r8", "%r9","%r10") :	# Win64 order
				("%rdi","%rsi","%rdx","%rcx","%r8");	# Unix order

$code.=<<___;
.extern	OPENSSL_ia32cap_P
.globl	bn_GF2m_mul_2x2
.type	bn_GF2m_mul_2x2,\@abi-omnipotent
.align	16
bn_GF2m_mul_2x2:
.cfi_startproc
	mov	%rsp,%rax
	mov	OPENSSL_ia32cap_P(%rip),%r10
	bt	\$33,%r10
	jnc	.Lvanilla_mul_2x2

	movq		$a1,%xmm0
	movq		$b1,%xmm1
	movq		$a0,%xmm2
___
$code.=<<___ if ($win64);
	movq		40(%rsp),%xmm3
___
$code.=<<___ if (!$win64);
	movq		$b0,%xmm3
___
$code.=<<___;
	movdqa		%xmm0,%xmm4
	movdqa		%xmm1,%xmm5
	pclmulqdq	\$0,%xmm1,%xmm0	# a1·b1
	pxor		%xmm2,%xmm4
	pxor		%xmm3,%xmm5
	pclmulqdq	\$0,%xmm3,%xmm2	# a0·b0
	pclmulqdq	\$0,%xmm5,%xmm4	# (a0+a1)·(b0+b1)
	xorps		%xmm0,%xmm4
	xorps		%xmm2,%xmm4	# (a0+a1)·(b0+b1)-a0·b0-a1·b1
	movdqa		%xmm4,%xmm5
	pslldq		\$8,%xmm4
	psrldq		\$8,%xmm5
	pxor		%xmm4,%xmm2
	pxor		%xmm5,%xmm0
	movdqu		%xmm2,0($rp)
	movdqu		%xmm0,16($rp)
	ret

.align	16
.Lvanilla_mul_2x2:
	lea	-8*17(%rsp),%rsp
.cfi_adjust_cfa_offset	8*17
___
$code.=<<___ if ($win64);
	mov	`8*17+40`(%rsp),$b0
	mov	%rdi,8*15(%rsp)
	mov	%rsi,8*16(%rsp)
___
$code.=<<___;
	mov	%r14,8*10(%rsp)
.cfi_rel_offset	%r14,8*10
	mov	%r13,8*11(%rsp)
.cfi_rel_offset	%r13,8*11
	mov	%r12,8*12(%rsp)
.cfi_rel_offset	%r12,8*12
	mov	%rbp,8*13(%rsp)
.cfi_rel_offset	%rbp,8*13
	mov	%rbx,8*14(%rsp)
.cfi_rel_offset	%rbx,8*14
.Lbody_mul_2x2:
	mov	$rp,32(%rsp)		# save the arguments
	mov	$a1,40(%rsp)
	mov	$a0,48(%rsp)
	mov	$b1,56(%rsp)
	mov	$b0,64(%rsp)

	mov	\$0xf,$mask
	mov	$a1,$a
	mov	$b1,$b
	call	_mul_1x1		# a1·b1
	mov	$lo,16(%rsp)
	mov	$hi,24(%rsp)

	mov	48(%rsp),$a
	mov	64(%rsp),$b
	call	_mul_1x1		# a0·b0
	mov	$lo,0(%rsp)
	mov	$hi,8(%rsp)

	mov	40(%rsp),$a
	mov	56(%rsp),$b
	xor	48(%rsp),$a
	xor	64(%rsp),$b
	call	_mul_1x1		# (a0+a1)·(b0+b1)
___
	@r=("%rbx","%rcx","%rdi","%rsi");
$code.=<<___;
	mov	0(%rsp),@r[0]
	mov	8(%rsp),@r[1]
	mov	16(%rsp),@r[2]
	mov	24(%rsp),@r[3]
	mov	32(%rsp),%rbp

	xor	$hi,$lo
	xor	@r[1],$hi
	xor	@r[0],$lo
	mov	@r[0],0(%rbp)
	xor	@r[2],$hi
	mov	@r[3],24(%rbp)
	xor	@r[3],$lo
	xor	@r[3],$hi
	xor	$hi,$lo
	mov	$hi,16(%rbp)
	mov	$lo,8(%rbp)

	mov	8*10(%rsp),%r14
.cfi_restore	%r14
	mov	8*11(%rsp),%r13
.cfi_restore	%r13
	mov	8*12(%rsp),%r12
.cfi_restore	%r12
	mov	8*13(%rsp),%rbp
.cfi_restore	%rbp
	mov	8*14(%rsp),%rbx
.cfi_restore	%rbx
___
$code.=<<___ if ($win64);
	mov	8*15(%rsp),%rdi
	mov	8*16(%rsp),%rsi
___
$code.=<<___;
	lea	8*17(%rsp),%rsp
.cfi_adjust_cfa_offset	-8*17
.Lepilogue_mul_2x2:
	ret
.Lend_mul_2x2:
.cfi_endproc
.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
.asciz	"GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
.align	16
___

# 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	se_handler,\@abi-omnipotent
.align	16
se_handler:
	push	%rsi
	push	%rdi
	push	%rbx
	push	%rbp
	push	%r12
	push	%r13
	push	%r14
	push	%r15
	pushfq
	sub	\$64,%rsp

	mov	120($context),%rax	# pull context->Rax
	mov	248($context),%rbx	# pull context->Rip

	lea	.Lbody_mul_2x2(%rip),%r10
	cmp	%r10,%rbx		# context->Rip<"prologue" label
	jb	.Lin_prologue

	mov	152($context),%rax	# pull context->Rsp

	lea	.Lepilogue_mul_2x2(%rip),%r10
	cmp	%r10,%rbx		# context->Rip>="epilogue" label
	jae	.Lin_prologue

	mov	8*10(%rax),%r14		# mimic epilogue
	mov	8*11(%rax),%r13
	mov	8*12(%rax),%r12
	mov	8*13(%rax),%rbp
	mov	8*14(%rax),%rbx
	mov	8*15(%rax),%rdi
	mov	8*16(%rax),%rsi

	mov	%rbx,144($context)	# restore context->Rbx
	mov	%rbp,160($context)	# restore context->Rbp
	mov	%rsi,168($context)	# restore context->Rsi
	mov	%rdi,176($context)	# restore context->Rdi
	mov	%r12,216($context)	# restore context->R12
	mov	%r13,224($context)	# restore context->R13
	mov	%r14,232($context)	# restore context->R14

	lea	8*17(%rax),%rax

.Lin_prologue:
	mov	%rax,152($context)	# restore context->Rsp

	mov	40($disp),%rdi		# disp->ContextRecord
	mov	$context,%rsi		# context
	mov	\$154,%ecx		# sizeof(CONTEXT)
	.long	0xa548f3fc		# cld; rep movsq

	mov	$disp,%rsi
	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
	mov	0(%rsi),%r8		# arg3, disp->ControlPc
	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
	mov	40(%rsi),%r10		# disp->ContextRecord
	lea	56(%rsi),%r11		# &disp->HandlerData
	lea	24(%rsi),%r12		# &disp->EstablisherFrame
	mov	%r10,32(%rsp)		# arg5
	mov	%r11,40(%rsp)		# arg6
	mov	%r12,48(%rsp)		# arg7
	mov	%rcx,56(%rsp)		# arg8, (NULL)
	call	*__imp_RtlVirtualUnwind(%rip)

	mov	\$1,%eax		# ExceptionContinueSearch
	add	\$64,%rsp
	popfq
	pop	%r15
	pop	%r14
	pop	%r13
	pop	%r12
	pop	%rbp
	pop	%rbx
	pop	%rdi
	pop	%rsi
	ret
.size	se_handler,.-se_handler

.section	.pdata
.align	4
	.rva	_mul_1x1
	.rva	.Lend_mul_1x1
	.rva	.LSEH_info_1x1

	.rva	.Lvanilla_mul_2x2
	.rva	.Lend_mul_2x2
	.rva	.LSEH_info_2x2
.section	.xdata
.align	8
.LSEH_info_1x1:
	.byte	0x01,0x07,0x02,0x00
	.byte	0x07,0x01,0x11,0x00	# sub rsp,128+8
.LSEH_info_2x2:
	.byte	9,0,0,0
	.rva	se_handler
___
}

$code =~ s/\`([^\`]*)\`/eval($1)/gem;
print $code;
close STDOUT;
Commit	Line	Data
6aa36e8e RS	1	#! /usr/bin/env perl
	2	# Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
	3	#
	4	# Licensed under the OpenSSL license (the "License"). You may not use
	5	# this file except in compliance with the License. You can obtain a copy
	6	# in the file LICENSE in the source distribution or at
	7	# https://www.openssl.org/source/license.html
	8
afebe623 AP	9	#
	10	# ====================================================================
	11	# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
	12	# project. The module is, however, dual licensed under OpenSSL and
	13	# CRYPTOGAMS licenses depending on where you obtain it. For further
	14	# details see http://www.openssl.org/~appro/cryptogams/.
	15	# ====================================================================
	16	#
	17	# May 2011
	18	#
	19	# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
	20	# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
	21	# the time being... Except that it has two code paths: code suitable
	22	# for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
053fa39a	23	# later. Improvement varies from one benchmark and µ-arch to another.
afebe623 AP	24	# Vanilla code path is at most 20% faster than compiler-generated code
	25	# [not very impressive], while PCLMULQDQ - whole 85%-160% better on
	26	# 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
	27	# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
	28	# all CPU time is burnt in it...
	29
	30	$flavour = shift;
	31	$output = shift;
	32	if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
	33
	34	$win64=0; $win64=1 if ($flavour =~ /[nm]asm\|mingw64/ \|\| $output =~ /\.asm$/);
	35
	36	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
	37	( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
	38	( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
	39	die "can't locate x86_64-xlate.pl";
	40
cfe1d992	41	open OUT,"\| \"$^X\" \"$xlate\" $flavour \"$output\"";
342dbbbe	42	STDOUT=OUT;
afebe623 AP	43
	44	($lo,$hi)=("%rax","%rdx"); $a=$lo;
	45	($i0,$i1)=("%rsi","%rdi");
	46	($t0,$t1)=("%rbx","%rcx");
	47	($b,$mask)=("%rbp","%r8");
	48	($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
	49	($R,$Tx)=("%xmm0","%xmm1");
	50
	51	$code.=<<___;
	52	.text
	53
	54	.type _mul_1x1,\@abi-omnipotent
	55	.align 16
	56	_mul_1x1:
76e624a0	57	.cfi_startproc
afebe623	58	sub \$128+8,%rsp
76e624a0	59	.cfi_adjust_cfa_offset 128+8
afebe623 AP	60	mov \$-1,$a1
	61	lea ($a,$a),$i0
	62	shr \$3,$a1
	63	lea (,$a,4),$i1
	64	and $a,$a1 # a1=a&0x1fffffffffffffff
	65	lea (,$a,8),$a8
	66	sar \$63,$a # broadcast 63rd bit
	67	lea ($a1,$a1),$a2
	68	sar \$63,$i0 # broadcast 62nd bit
	69	lea (,$a1,4),$a4
	70	and $b,$a
46f4e1be	71	sar \$63,$i1 # broadcast 61st bit
afebe623 AP	72	mov $a,$hi # $a is $lo
	73	shl \$63,$lo
	74	and $b,$i0
	75	shr \$1,$hi
	76	mov $i0,$t1
	77	shl \$62,$i0
	78	and $b,$i1
	79	shr \$2,$t1
	80	xor $i0,$lo
	81	mov $i1,$t0
	82	shl \$61,$i1
	83	xor $t1,$hi
	84	shr \$3,$t0
	85	xor $i1,$lo
	86	xor $t0,$hi
	87
	88	mov $a1,$a12
	89	movq \$0,0(%rsp) # tab[0]=0
	90	xor $a2,$a12 # a1^a2
	91	mov $a1,8(%rsp) # tab[1]=a1
	92	mov $a4,$a48
	93	mov $a2,16(%rsp) # tab[2]=a2
	94	xor $a8,$a48 # a4^a8
	95	mov $a12,24(%rsp) # tab[3]=a1^a2
	96
	97	xor $a4,$a1
	98	mov $a4,32(%rsp) # tab[4]=a4
	99	xor $a4,$a2
	100	mov $a1,40(%rsp) # tab[5]=a1^a4
	101	xor $a4,$a12
	102	mov $a2,48(%rsp) # tab[6]=a2^a4
	103	xor $a48,$a1 # a1^a4^a4^a8=a1^a8
	104	mov $a12,56(%rsp) # tab[7]=a1^a2^a4
	105	xor $a48,$a2 # a2^a4^a4^a8=a1^a8
	106
	107	mov $a8,64(%rsp) # tab[8]=a8
	108	xor $a48,$a12 # a1^a2^a4^a4^a8=a1^a2^a8
	109	mov $a1,72(%rsp) # tab[9]=a1^a8
	110	xor $a4,$a1 # a1^a8^a4
	111	mov $a2,80(%rsp) # tab[10]=a2^a8
	112	xor $a4,$a2 # a2^a8^a4
	113	mov $a12,88(%rsp) # tab[11]=a1^a2^a8
	114
	115	xor $a4,$a12 # a1^a2^a8^a4
	116	mov $a48,96(%rsp) # tab[12]=a4^a8
	117	mov $mask,$i0
	118	mov $a1,104(%rsp) # tab[13]=a1^a4^a8
	119	and $b,$i0
	120	mov $a2,112(%rsp) # tab[14]=a2^a4^a8
	121	shr \$4,$b
	122	mov $a12,120(%rsp) # tab[15]=a1^a2^a4^a8
	123	mov $mask,$i1
	124	and $b,$i1
	125	shr \$4,$b
	126
	127	movq (%rsp,$i0,8),$R # half of calculations is done in SSE2
	128	mov $mask,$i0
	129	and $b,$i0
	130	shr \$4,$b
	131	___
	132	for ($n=1;$n<8;$n++) {
	133	$code.=<<___;
	134	mov (%rsp,$i1,8),$t1
	135	mov $mask,$i1
136	mov $t1,$t0
137	shl \$`8*$n-4`,$t1
138	and $b,$i1
139	movq (%rsp,$i0,8),$Tx
140	shr \$`64-(8*$n-4)`,$t0
141	xor $t1,$lo
142	pslldq \$$n,$Tx
143	mov $mask,$i0
144	shr \$4,$b
145	xor $t0,$hi
146	and $b,$i0
147	shr \$4,$b
148	pxor $Tx,$R
149	___
150	}
151	$code.=<<___;
152	mov (%rsp,$i1,8),$t1
153	mov $t1,$t0
154	shl \$`8*$n-4`,$t1
155	movq $R,$i0
156	shr \$`64-(8*$n-4)`,$t0
157	xor $t1,$lo
158	psrldq \$8,$R
159	xor $t0,$hi
160	movq $R,$i1
161	xor $i0,$lo
162	xor $i1,$hi
163
164	add \$128+8,%rsp
76e624a0	165	.cfi_adjust_cfa_offset -128-8
afebe623	166	ret
96abea33	167	.Lend_mul_1x1:
76e624a0	168	.cfi_endproc
afebe623 AP	169	.size _mul_1x1,.-_mul_1x1
	170	___
	171
	172	($rp,$a1,$a0,$b1,$b0) = $win64? ("%rcx","%rdx","%r8", "%r9","%r10") : # Win64 order
	173	("%rdi","%rsi","%rdx","%rcx","%r8"); # Unix order
	174
	175	$code.=<<___;
	176	.extern OPENSSL_ia32cap_P
	177	.globl bn_GF2m_mul_2x2
	178	.type bn_GF2m_mul_2x2,\@abi-omnipotent
	179	.align 16
	180	bn_GF2m_mul_2x2:
76e624a0	181	.cfi_startproc
384e6de4 AP	182	mov %rsp,%rax
	183	mov OPENSSL_ia32cap_P(%rip),%r10
	184	bt \$33,%r10
96abea33	185	jnc .Lvanilla_mul_2x2
afebe623 AP	186
	187	movq $a1,%xmm0
	188	movq $b1,%xmm1
	189	movq $a0,%xmm2
	190	___
	191	$code.=<<___ if ($win64);
	192	movq 40(%rsp),%xmm3
	193	___
	194	$code.=<<___ if (!$win64);
	195	movq $b0,%xmm3
	196	___
	197	$code.=<<___;
	198	movdqa %xmm0,%xmm4
	199	movdqa %xmm1,%xmm5
053fa39a	200	pclmulqdq \$0,%xmm1,%xmm0 # a1·b1
afebe623 AP	201	pxor %xmm2,%xmm4
afebe623 AP	202	pxor %xmm3,%xmm5
053fa39a RL	203	pclmulqdq \$0,%xmm3,%xmm2 # a0·b0
053fa39a RL	204	pclmulqdq \$0,%xmm5,%xmm4 # (a0+a1)·(b0+b1)
afebe623	205	xorps %xmm0,%xmm4
053fa39a	206	xorps %xmm2,%xmm4 # (a0+a1)·(b0+b1)-a0·b0-a1·b1
afebe623 AP	207	movdqa %xmm4,%xmm5
	208	pslldq \$8,%xmm4
	209	psrldq \$8,%xmm5
	210	pxor %xmm4,%xmm2
	211	pxor %xmm5,%xmm0
	212	movdqu %xmm2,0($rp)
	213	movdqu %xmm0,16($rp)
	214	ret
	215
	216	.align 16
96abea33	217	.Lvanilla_mul_2x2:
afebe623	218	lea -8*17(%rsp),%rsp
76e624a0	219	.cfi_adjust_cfa_offset 8*17
afebe623 AP	220	___
	221	$code.=<<___ if ($win64);
	222	mov `8*17+40`(%rsp),$b0
	223	mov %rdi,8*15(%rsp)
	224	mov %rsi,8*16(%rsp)
	225	___
	226	$code.=<<___;
	227	mov %r14,8*10(%rsp)
76e624a0	228	.cfi_rel_offset %r14,8*10
afebe623	229	mov %r13,8*11(%rsp)
76e624a0	230	.cfi_rel_offset %r13,8*11
afebe623	231	mov %r12,8*12(%rsp)
76e624a0	232	.cfi_rel_offset %r12,8*12
afebe623	233	mov %rbp,8*13(%rsp)
76e624a0	234	.cfi_rel_offset %rbp,8*13
afebe623	235	mov %rbx,8*14(%rsp)
76e624a0	236	.cfi_rel_offset %rbx,8*14
96abea33	237	.Lbody_mul_2x2:
afebe623 AP	238	mov $rp,32(%rsp) # save the arguments
	239	mov $a1,40(%rsp)
	240	mov $a0,48(%rsp)
	241	mov $b1,56(%rsp)
	242	mov $b0,64(%rsp)
	243
	244	mov \$0xf,$mask
	245	mov $a1,$a
	246	mov $b1,$b
053fa39a	247	call _mul_1x1 # a1·b1
afebe623 AP	248	mov $lo,16(%rsp)
	249	mov $hi,24(%rsp)
	250
	251	mov 48(%rsp),$a
	252	mov 64(%rsp),$b
053fa39a	253	call _mul_1x1 # a0·b0
afebe623 AP	254	mov $lo,0(%rsp)
	255	mov $hi,8(%rsp)
	256
	257	mov 40(%rsp),$a
	258	mov 56(%rsp),$b
	259	xor 48(%rsp),$a
	260	xor 64(%rsp),$b
053fa39a	261	call _mul_1x1 # (a0+a1)·(b0+b1)
afebe623 AP	262	___
	263	@r=("%rbx","%rcx","%rdi","%rsi");
	264	$code.=<<___;
	265	mov 0(%rsp),@r[0]
	266	mov 8(%rsp),@r[1]
	267	mov 16(%rsp),@r[2]
	268	mov 24(%rsp),@r[3]
	269	mov 32(%rsp),%rbp
	270
	271	xor $hi,$lo
	272	xor @r[1],$hi
	273	xor @r[0],$lo
	274	mov @r[0],0(%rbp)
	275	xor @r[2],$hi
	276	mov @r[3],24(%rbp)
	277	xor @r[3],$lo
	278	xor @r[3],$hi
	279	xor $hi,$lo
	280	mov $hi,16(%rbp)
	281	mov $lo,8(%rbp)
	282
	283	mov 8*10(%rsp),%r14
76e624a0	284	.cfi_restore %r14
afebe623	285	mov 8*11(%rsp),%r13
76e624a0	286	.cfi_restore %r13
afebe623	287	mov 8*12(%rsp),%r12
76e624a0	288	.cfi_restore %r12
afebe623	289	mov 8*13(%rsp),%rbp
76e624a0	290	.cfi_restore %rbp
afebe623	291	mov 8*14(%rsp),%rbx
76e624a0	292	.cfi_restore %rbx
afebe623 AP	293	___
	294	$code.=<<___ if ($win64);
	295	mov 8*15(%rsp),%rdi
	296	mov 8*16(%rsp),%rsi
	297	___
	298	$code.=<<___;
	299	lea 8*17(%rsp),%rsp
76e624a0	300	.cfi_adjust_cfa_offset -8*17
384e6de4	301	.Lepilogue_mul_2x2:
afebe623	302	ret
96abea33	303	.Lend_mul_2x2:
76e624a0	304	.cfi_endproc
afebe623 AP	305	.size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
afebe623 AP	306	.asciz "GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
96abea33 AP	307	.align 16
	308	___
	309
	310	# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
	311	# CONTEXT context,DISPATCHER_CONTEXT disp)
	312	if ($win64) {
	313	$rec="%rcx";
	314	$frame="%rdx";
	315	$context="%r8";
	316	$disp="%r9";
	317
	318	$code.=<<___;
	319	.extern __imp_RtlVirtualUnwind
	320
	321	.type se_handler,\@abi-omnipotent
	322	.align 16
	323	se_handler:
	324	push %rsi
	325	push %rdi
	326	push %rbx
	327	push %rbp
	328	push %r12
	329	push %r13
	330	push %r14
	331	push %r15
	332	pushfq
	333	sub \$64,%rsp
	334
384e6de4	335	mov 120($context),%rax # pull context->Rax
96abea33 AP	336	mov 248($context),%rbx # pull context->Rip
	337
	338	lea .Lbody_mul_2x2(%rip),%r10
	339	cmp %r10,%rbx # context->Rip<"prologue" label
	340	jb .Lin_prologue
	341
384e6de4 AP	342	mov 152($context),%rax # pull context->Rsp
	343
	344	lea .Lepilogue_mul_2x2(%rip),%r10
	345	cmp %r10,%rbx # context->Rip>="epilogue" label
	346	jae .Lin_prologue
	347
96abea33 AP	348	mov 8*10(%rax),%r14 # mimic epilogue
	349	mov 8*11(%rax),%r13
	350	mov 8*12(%rax),%r12
	351	mov 8*13(%rax),%rbp
	352	mov 8*14(%rax),%rbx
	353	mov 8*15(%rax),%rdi
	354	mov 8*16(%rax),%rsi
	355
	356	mov %rbx,144($context) # restore context->Rbx
	357	mov %rbp,160($context) # restore context->Rbp
	358	mov %rsi,168($context) # restore context->Rsi
	359	mov %rdi,176($context) # restore context->Rdi
	360	mov %r12,216($context) # restore context->R12
	361	mov %r13,224($context) # restore context->R13
	362	mov %r14,232($context) # restore context->R14
	363
96abea33	364	lea 8*17(%rax),%rax
384e6de4 AP	365
384e6de4 AP	366	.Lin_prologue:
96abea33 AP	367	mov %rax,152($context) # restore context->Rsp
	368
	369	mov 40($disp),%rdi # disp->ContextRecord
	370	mov $context,%rsi # context
	371	mov \$154,%ecx # sizeof(CONTEXT)
	372	.long 0xa548f3fc # cld; rep movsq
	373
	374	mov $disp,%rsi
	375	xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
	376	mov 8(%rsi),%rdx # arg2, disp->ImageBase
	377	mov 0(%rsi),%r8 # arg3, disp->ControlPc
	378	mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
	379	mov 40(%rsi),%r10 # disp->ContextRecord
	380	lea 56(%rsi),%r11 # &disp->HandlerData
	381	lea 24(%rsi),%r12 # &disp->EstablisherFrame
	382	mov %r10,32(%rsp) # arg5
	383	mov %r11,40(%rsp) # arg6
	384	mov %r12,48(%rsp) # arg7
	385	mov %rcx,56(%rsp) # arg8, (NULL)
	386	call *__imp_RtlVirtualUnwind(%rip)
	387
	388	mov \$1,%eax # ExceptionContinueSearch
	389	add \$64,%rsp
	390	popfq
	391	pop %r15
	392	pop %r14
	393	pop %r13
	394	pop %r12
	395	pop %rbp
	396	pop %rbx
	397	pop %rdi
	398	pop %rsi
	399	ret
	400	.size se_handler,.-se_handler
	401
	402	.section .pdata
	403	.align 4
	404	.rva _mul_1x1
	405	.rva .Lend_mul_1x1
	406	.rva .LSEH_info_1x1
	407
	408	.rva .Lvanilla_mul_2x2
	409	.rva .Lend_mul_2x2
	410	.rva .LSEH_info_2x2
	411	.section .xdata
	412	.align 8
	413	.LSEH_info_1x1:
	414	.byte 0x01,0x07,0x02,0x00
	415	.byte 0x07,0x01,0x11,0x00 # sub rsp,128+8
	416	.LSEH_info_2x2:
	417	.byte 9,0,0,0
	418	.rva se_handler
afebe623	419	___
96abea33	420	}
afebe623 AP	421
	422	$code =~ s/\`([^\`]*)\`/eval($1)/gem;
	423	print $code;
	424	close STDOUT;