[thirdparty/openssl.git] / crypto / md5 / asm / md5-x86_64.pl

#!/usr/bin/perl -w
#
# MD5 optimized for AMD64.
#
# Author: Marc Bevand <bevand_m (at) epita.fr>
# Licence: I hereby disclaim the copyright on this code and place it
# in the public domain.
#

use strict;

my $code;

# round1_step() does:
#   dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
#   %r10d = X[k_next]
#   %r11d = z' (copy of z for the next step)
# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
sub round1_step
{
    my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
    $code .= " mov	0*4(%rsi),	%r10d		/* (NEXT STEP) X[0] */\n" if ($pos == -1);
    $code .= " mov	%edx,		%r11d		/* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
    $code .= <<EOF;
	xor	$y,		%r11d		/* y ^ ... */
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	and	$x,		%r11d		/* x & ... */
	xor	$z,		%r11d		/* z ^ ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	add	%r11d,		$dst		/* dst += ... */
	rol	\$$s,		$dst		/* dst <<< s */
	mov	$y,		%r11d		/* (NEXT STEP) z' = $y */
	add	$x,		$dst		/* dst += x */
EOF
}

# round2_step() does:
#   dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
#   %r10d = X[k_next]
#   %r11d = z' (copy of z for the next step)
#   %r12d = z' (copy of z for the next step)
# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
sub round2_step
{
    my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
    $code .= " mov	1*4(%rsi),	%r10d		/* (NEXT STEP) X[1] */\n" if ($pos == -1);
    $code .= " mov	%edx,		%r11d		/* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
    $code .= " mov	%edx,		%r12d		/* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
    $code .= <<EOF;
	not	%r11d				/* not z */
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	and	$x,		%r12d		/* x & z */
	and	$y,		%r11d		/* y & (not z) */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	or	%r11d,		%r12d		/* (y & (not z)) | (x & z) */
	mov	$y,		%r11d		/* (NEXT STEP) z' = $y */
	add	%r12d,		$dst		/* dst += ... */
	mov	$y,		%r12d		/* (NEXT STEP) z' = $y */
	rol	\$$s,		$dst		/* dst <<< s */
	add	$x,		$dst		/* dst += x */
EOF
}

# round3_step() does:
#   dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
#   %r10d = X[k_next]
#   %r11d = y' (copy of y for the next step)
# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
sub round3_step
{
    my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
    $code .= " mov	5*4(%rsi),	%r10d		/* (NEXT STEP) X[5] */\n" if ($pos == -1);
    $code .= " mov	%ecx,		%r11d		/* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
    $code .= <<EOF;
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	xor	$z,		%r11d		/* z ^ ... */
	xor	$x,		%r11d		/* x ^ ... */
	add	%r11d,		$dst		/* dst += ... */
	rol	\$$s,		$dst		/* dst <<< s */
	mov	$x,		%r11d		/* (NEXT STEP) y' = $x */
	add	$x,		$dst		/* dst += x */
EOF
}

# round4_step() does:
#   dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
#   %r10d = X[k_next]
#   %r11d = not z' (copy of not z for the next step)
# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
sub round4_step
{
    my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
    $code .= " mov	0*4(%rsi),	%r10d		/* (NEXT STEP) X[0] */\n" if ($pos == -1);
    $code .= " mov	\$0xffffffff,	%r11d\n" if ($pos == -1);
    $code .= " xor	%edx,		%r11d		/* (NEXT STEP) not z' = not %edx*/\n"
    if ($pos == -1);
    $code .= <<EOF;
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	or	$x,		%r11d		/* x | ... */
	xor	$y,		%r11d		/* y ^ ... */
	add	%r11d,		$dst		/* dst += ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	mov	\$0xffffffff,	%r11d
	rol	\$$s,		$dst		/* dst <<< s */
	xor	$y,		%r11d		/* (NEXT STEP) not z' = not $y */
	add	$x,		$dst		/* dst += x */
EOF
}

my $flavour = shift;
my $output  = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }

my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);

$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
( $xlate="${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";

no warnings qw(uninitialized);
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;

$code .= <<EOF;
.text
.align 16

.globl md5_block_asm_data_order
.type md5_block_asm_data_order,\@function,3
md5_block_asm_data_order:
	push	%rbp
	push	%rbx
	push	%r12
	push	%r14
	push	%r15
.Lprologue:

	# rdi = arg #1 (ctx, MD5_CTX pointer)
	# rsi = arg #2 (ptr, data pointer)
	# rdx = arg #3 (nbr, number of 16-word blocks to process)
	mov	%rdi,		%rbp	# rbp = ctx
	shl	\$6,		%rdx	# rdx = nbr in bytes
	lea	(%rsi,%rdx),	%rdi	# rdi = end
	mov	0*4(%rbp),	%eax	# eax = ctx->A
	mov	1*4(%rbp),	%ebx	# ebx = ctx->B
	mov	2*4(%rbp),	%ecx	# ecx = ctx->C
	mov	3*4(%rbp),	%edx	# edx = ctx->D
	# end is 'rdi'
	# ptr is 'rsi'
	# A is 'eax'
	# B is 'ebx'
	# C is 'ecx'
	# D is 'edx'

	cmp	%rdi,		%rsi		# cmp end with ptr
	je	.Lend				# jmp if ptr == end

	# BEGIN of loop over 16-word blocks
.Lloop:	# save old values of A, B, C, D
	mov	%eax,		%r8d
	mov	%ebx,		%r9d
	mov	%ecx,		%r14d
	mov	%edx,		%r15d
EOF
round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');

round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');

round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');

round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
$code .= <<EOF;
	# add old values of A, B, C, D
	add	%r8d,	%eax
	add	%r9d,	%ebx
	add	%r14d,	%ecx
	add	%r15d,	%edx

	# loop control
	add	\$64,		%rsi		# ptr += 64
	cmp	%rdi,		%rsi		# cmp end with ptr
	jb	.Lloop				# jmp if ptr < end
	# END of loop over 16-word blocks

.Lend:
	mov	%eax,		0*4(%rbp)	# ctx->A = A
	mov	%ebx,		1*4(%rbp)	# ctx->B = B
	mov	%ecx,		2*4(%rbp)	# ctx->C = C
	mov	%edx,		3*4(%rbp)	# ctx->D = D

	mov	(%rsp),%r15
	mov	8(%rsp),%r14
	mov	16(%rsp),%r12
	mov	24(%rsp),%rbx
	mov	32(%rsp),%rbp
	add	\$40,%rsp
.Lepilogue:
	ret
.size md5_block_asm_data_order,.-md5_block_asm_data_order
EOF

# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
if ($win64) {
my $rec="%rcx";
my $frame="%rdx";
my $context="%r8";
my $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	.Lprologue(%rip),%r10
	cmp	%r10,%rbx		# context->Rip<.Lprologue
	jb	.Lin_prologue

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

	lea	.Lepilogue(%rip),%r10
	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
	jae	.Lin_prologue

	lea	40(%rax),%rax

	mov	-8(%rax),%rbp
	mov	-16(%rax),%rbx
	mov	-24(%rax),%r12
	mov	-32(%rax),%r14
	mov	-40(%rax),%r15
	mov	%rbx,144($context)	# restore context->Rbx
	mov	%rbp,160($context)	# restore context->Rbp
	mov	%r12,216($context)	# restore context->R12
	mov	%r14,232($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	se_handler,.-se_handler

.section	.pdata
.align	4
	.rva	.LSEH_begin_md5_block_asm_data_order
	.rva	.LSEH_end_md5_block_asm_data_order
	.rva	.LSEH_info_md5_block_asm_data_order

.section	.xdata
.align	8
.LSEH_info_md5_block_asm_data_order:
	.byte	9,0,0,0
	.rva	se_handler
___
}

print $code;

close STDOUT;
Commit	Line	Data
d37a65bc AP	1	#!/usr/bin/perl -w
	2	#
	3	# MD5 optimized for AMD64.
	4	#
	5	# Author: Marc Bevand <bevand_m (at) epita.fr>
	6	# Licence: I hereby disclaim the copyright on this code and place it
	7	# in the public domain.
	8	#
	9
	10	use strict;
	11
	12	my $code;
	13
	14	# round1_step() does:
	15	# dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
	16	# %r10d = X[k_next]
	17	# %r11d = z' (copy of z for the next step)
9c4fe782	18	# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
d37a65bc AP	19	sub round1_step
	20	{
	21	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
	22	$code .= " mov 04(%rsi), %r10d / (NEXT STEP) X[0] */\n" if ($pos == -1);
	23	$code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
	24	$code .= <<EOF;
	25	xor $y, %r11d /* y ^ ... */
	26	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
	27	and $x, %r11d /* x & ... */
	28	xor $z, %r11d /* z ^ ... */
	29	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
	30	add %r11d, $dst /* dst += ... */
	31	rol \$$s, $dst /* dst <<< s */
	32	mov $y, %r11d /* (NEXT STEP) z' = $y */
	33	add $x, $dst /* dst += x */
	34	EOF
	35	}
	36
	37	# round2_step() does:
	38	# dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
	39	# %r10d = X[k_next]
9c4fe782 AP	40	# %r11d = z' (copy of z for the next step)
	41	# %r12d = z' (copy of z for the next step)
	42	# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
d37a65bc AP	43	sub round2_step
	44	{
	45	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
	46	$code .= " mov 14(%rsi), %r10d / (NEXT STEP) X[1] */\n" if ($pos == -1);
9c4fe782 AP	47	$code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
9c4fe782 AP	48	$code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
d37a65bc	49	$code .= <<EOF;
9c4fe782	50	not %r11d /* not z */
d37a65bc	51	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
9c4fe782 AP	52	and $x, %r12d /* x & z */
9c4fe782 AP	53	and $y, %r11d /* y & (not z) */
d37a65bc	54	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
9c4fe782 AP	55	or %r11d, %r12d /* (y & (not z)) \| (x & z) */
	56	mov $y, %r11d /* (NEXT STEP) z' = $y */
	57	add %r12d, $dst /* dst += ... */
	58	mov $y, %r12d /* (NEXT STEP) z' = $y */
d37a65bc	59	rol \$$s, $dst /* dst <<< s */
d37a65bc AP	60	add $x, $dst /* dst += x */
	61	EOF
	62	}
	63
	64	# round3_step() does:
	65	# dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
	66	# %r10d = X[k_next]
	67	# %r11d = y' (copy of y for the next step)
9c4fe782	68	# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
d37a65bc AP	69	sub round3_step
	70	{
	71	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
	72	$code .= " mov 54(%rsi), %r10d / (NEXT STEP) X[5] */\n" if ($pos == -1);
	73	$code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
	74	$code .= <<EOF;
	75	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
	76	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
	77	xor $z, %r11d /* z ^ ... */
	78	xor $x, %r11d /* x ^ ... */
	79	add %r11d, $dst /* dst += ... */
	80	rol \$$s, $dst /* dst <<< s */
	81	mov $x, %r11d /* (NEXT STEP) y' = $x */
	82	add $x, $dst /* dst += x */
	83	EOF
	84	}
	85
	86	# round4_step() does:
	87	# dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
	88	# %r10d = X[k_next]
	89	# %r11d = not z' (copy of not z for the next step)
9c4fe782	90	# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
d37a65bc AP	91	sub round4_step
	92	{
	93	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
	94	$code .= " mov 04(%rsi), %r10d / (NEXT STEP) X[0] */\n" if ($pos == -1);
	95	$code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1);
	96	$code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n"
	97	if ($pos == -1);
	98	$code .= <<EOF;
	99	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
	100	or $x, %r11d /* x \| ... */
	101	xor $y, %r11d /* y ^ ... */
	102	add %r11d, $dst /* dst += ... */
	103	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
	104	mov \$0xffffffff, %r11d
	105	rol \$$s, $dst /* dst <<< s */
	106	xor $y, %r11d /* (NEXT STEP) not z' = not $y */
	107	add $x, $dst /* dst += x */
	108	EOF
	109	}
	110
be01f79d AP	111	my $flavour = shift;
	112	my $output = shift;
	113	if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
	114
	115	my $win64=0; $win64=1 if ($flavour =~ /[nm]asm\|mingw64/ \|\| $output =~ /\.asm$/);
abe7f8b4 AP	116
	117	$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
	118	( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
	119	( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
	120	die "can't locate x86_64-xlate.pl";
	121
be01f79d	122	no warnings qw(uninitialized);
b286a968 AP	123	open OUT,"\| \"$^X\" $xlate $flavour $output";
b286a968 AP	124	STDOUT=OUT;
d37a65bc AP	125
	126	$code .= <<EOF;
	127	.text
	128	.align 16
	129
c69ed6ea AP	130	.globl md5_block_asm_data_order
	131	.type md5_block_asm_data_order,\@function,3
	132	md5_block_asm_data_order:
d37a65bc AP	133	push %rbp
d37a65bc AP	134	push %rbx
9c4fe782	135	push %r12
d37a65bc AP	136	push %r14
d37a65bc AP	137	push %r15
be01f79d	138	.Lprologue:
d37a65bc AP	139
	140	# rdi = arg #1 (ctx, MD5_CTX pointer)
	141	# rsi = arg #2 (ptr, data pointer)
	142	# rdx = arg #3 (nbr, number of 16-word blocks to process)
	143	mov %rdi, %rbp # rbp = ctx
	144	shl \$6, %rdx # rdx = nbr in bytes
	145	lea (%rsi,%rdx), %rdi # rdi = end
	146	mov 0*4(%rbp), %eax # eax = ctx->A
	147	mov 1*4(%rbp), %ebx # ebx = ctx->B
	148	mov 2*4(%rbp), %ecx # ecx = ctx->C
	149	mov 3*4(%rbp), %edx # edx = ctx->D
d37a65bc AP	150	# end is 'rdi'
	151	# ptr is 'rsi'
	152	# A is 'eax'
	153	# B is 'ebx'
	154	# C is 'ecx'
	155	# D is 'edx'
	156
	157	cmp %rdi, %rsi # cmp end with ptr
8b5bf52a	158	je .Lend # jmp if ptr == end
d37a65bc AP	159
d37a65bc AP	160	# BEGIN of loop over 16-word blocks
8b5bf52a	161	.Lloop: # save old values of A, B, C, D
d37a65bc AP	162	mov %eax, %r8d
	163	mov %ebx, %r9d
	164	mov %ecx, %r14d
	165	mov %edx, %r15d
	166	EOF
	167	round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
	168	round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
	169	round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
	170	round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
	171	round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
	172	round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
	173	round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
	174	round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
	175	round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
	176	round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
	177	round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
	178	round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
	179	round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
	180	round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
	181	round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
	182	round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');
	183
	184	round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
	185	round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
	186	round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
	187	round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
	188	round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
	189	round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
	190	round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
	191	round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
	192	round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
	193	round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
	194	round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
	195	round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
	196	round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
	197	round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
	198	round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
	199	round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');
	200
	201	round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
	202	round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
	203	round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
	204	round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
	205	round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
	206	round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
	207	round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
	208	round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
	209	round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
	210	round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
	211	round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
	212	round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
	213	round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
	214	round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
	215	round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
	216	round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
	217
	218	round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
	219	round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
	220	round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
	221	round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
	222	round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
	223	round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
	224	round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
	225	round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
226	round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
227	round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
228	round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
229	round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
230	round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
231	round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
232	round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
233	round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
234	$code .= <<EOF;
235	# add old values of A, B, C, D
236	add %r8d, %eax
237	add %r9d, %ebx
238	add %r14d, %ecx
239	add %r15d, %edx
240
241	# loop control
242	add \$64, %rsi # ptr += 64
243	cmp %rdi, %rsi # cmp end with ptr
8b5bf52a	244	jb .Lloop # jmp if ptr < end
d37a65bc AP	245	# END of loop over 16-word blocks
d37a65bc AP	246
8b5bf52a	247	.Lend:
d37a65bc AP	248	mov %eax, 0*4(%rbp) # ctx->A = A
	249	mov %ebx, 1*4(%rbp) # ctx->B = B
	250	mov %ecx, 2*4(%rbp) # ctx->C = C
	251	mov %edx, 3*4(%rbp) # ctx->D = D
	252
be01f79d AP	253	mov (%rsp),%r15
	254	mov 8(%rsp),%r14
	255	mov 16(%rsp),%r12
	256	mov 24(%rsp),%rbx
	257	mov 32(%rsp),%rbp
	258	add \$40,%rsp
	259	.Lepilogue:
	260	ret
	261	.size md5_block_asm_data_order,.-md5_block_asm_data_order
	262	EOF
	263
	264	# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
	265	# CONTEXT context,DISPATCHER_CONTEXT disp)
	266	if ($win64) {
	267	my $rec="%rcx";
	268	my $frame="%rdx";
	269	my $context="%r8";
	270	my $disp="%r9";
	271
	272	$code.=<<___;
	273	.extern __imp_RtlVirtualUnwind
	274	.type se_handler,\@abi-omnipotent
	275	.align 16
	276	se_handler:
	277	push %rsi
	278	push %rdi
	279	push %rbx
	280	push %rbp
	281	push %r12
	282	push %r13
	283	push %r14
	284	push %r15
	285	pushfq
	286	sub \$64,%rsp
	287
	288	mov 120($context),%rax # pull context->Rax
	289	mov 248($context),%rbx # pull context->Rip
	290
	291	lea .Lprologue(%rip),%r10
	292	cmp %r10,%rbx # context->Rip<.Lprologue
	293	jb .Lin_prologue
	294
	295	mov 152($context),%rax # pull context->Rsp
	296
	297	lea .Lepilogue(%rip),%r10
	298	cmp %r10,%rbx # context->Rip>=.Lepilogue
	299	jae .Lin_prologue
	300
	301	lea 40(%rax),%rax
	302
	303	mov -8(%rax),%rbp
	304	mov -16(%rax),%rbx
	305	mov -24(%rax),%r12
	306	mov -32(%rax),%r14
	307	mov -40(%rax),%r15
	308	mov %rbx,144($context) # restore context->Rbx
	309	mov %rbp,160($context) # restore context->Rbp
	310	mov %r12,216($context) # restore context->R12
	311	mov %r14,232($context) # restore context->R14
	312	mov %r15,240($context) # restore context->R15
	313
	314	.Lin_prologue:
	315	mov 8(%rax),%rdi
	316	mov 16(%rax),%rsi
317	mov %rax,152($context) # restore context->Rsp
318	mov %rsi,168($context) # restore context->Rsi
319	mov %rdi,176($context) # restore context->Rdi
320
321	mov 40($disp),%rdi # disp->ContextRecord
322	mov $context,%rsi # context
323	mov \$154,%ecx # sizeof(CONTEXT)
324	.long 0xa548f3fc # cld; rep movsq
325
326	mov $disp,%rsi
327	xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
328	mov 8(%rsi),%rdx # arg2, disp->ImageBase
329	mov 0(%rsi),%r8 # arg3, disp->ControlPc
330	mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
331	mov 40(%rsi),%r10 # disp->ContextRecord
332	lea 56(%rsi),%r11 # &disp->HandlerData
333	lea 24(%rsi),%r12 # &disp->EstablisherFrame
334	mov %r10,32(%rsp) # arg5
335	mov %r11,40(%rsp) # arg6
336	mov %r12,48(%rsp) # arg7
337	mov %rcx,56(%rsp) # arg8, (NULL)
338	call *__imp_RtlVirtualUnwind(%rip)
339
340	mov \$1,%eax # ExceptionContinueSearch
341	add \$64,%rsp
342	popfq
d37a65bc AP	343	pop %r15
d37a65bc AP	344	pop %r14
be01f79d	345	pop %r13
9c4fe782	346	pop %r12
d37a65bc	347	pop %rbp
be01f79d AP	348	pop %rbx
	349	pop %rdi
	350	pop %rsi
d37a65bc	351	ret
be01f79d AP	352	.size se_handler,.-se_handler
	353
	354	.section .pdata
	355	.align 4
	356	.rva .LSEH_begin_md5_block_asm_data_order
	357	.rva .LSEH_end_md5_block_asm_data_order
	358	.rva .LSEH_info_md5_block_asm_data_order
	359
	360	.section .xdata
	361	.align 8
	362	.LSEH_info_md5_block_asm_data_order:
	363	.byte 9,0,0,0
	364	.rva se_handler
	365	___
	366	}
d37a65bc AP	367
d37a65bc AP	368	print $code;
8b5bf52a AP	369
8b5bf52a AP	370	close STDOUT;