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

#! /usr/bin/env perl
# Author: Marc Bevand <bevand_m (at) epita.fr>
# Copyright 2005-2016 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

# MD5 optimized for AMD64.

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 & ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	xor	$z,		%r11d		/* z ^ ... */
	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) = @_;
    $code .= " mov	%edx,		%r11d		/* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
    $code .= " mov	%edx,		%r12d		/* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
    $code .= <<EOF;
	not	%r11d				/* not z */
	and	$x,		%r12d		/* x & z */
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	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)
{ my $round3_alter=0;
sub round3_step
{
    my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
    $code .= " mov	%ecx,		%r11d		/* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
    $code .= <<EOF;
	lea	$T_i($dst,%r10d),$dst		/* Const + dst + ... */
	xor	$z,		%r11d		/* z ^ ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	xor	$x,		%r11d		/* x ^ ... */
	add	%r11d,		$dst		/* dst += ... */
EOF
    $code .= <<EOF if ($round3_alter);
	rol	\$$s,		$dst		/* dst <<< s */
	mov	$x,		%r11d		/* (NEXT STEP) y' = $x */
EOF
    $code .= <<EOF if (!$round3_alter);
	mov	$x,		%r11d		/* (NEXT STEP) y' = $x */
	rol	\$$s,		$dst		/* dst <<< s */
EOF
    $code .= <<EOF;
	add	$x,		$dst		/* dst += x */
EOF
    $round3_alter^=1;
}
}

# 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) = @_;
    $code .= " mov	\$0xffffffff,	%r11d\n" if ($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 | ... */
	mov	$k_next*4(%rsi),%r10d		/* (NEXT STEP) X[$k_next] */
	xor	$y,		%r11d		/* y ^ ... */
	add	%r11d,		$dst		/* dst += ... */
	mov	\$0xffffffff,	%r11d
	rol	\$$s,		$dst		/* dst <<< s */
	xor	$y,		%r11d		/* (NEXT STEP) not z' = not $y */
	add	$x,		$dst		/* dst += x */
EOF
}

no warnings qw(uninitialized);
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";

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', '1','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', '5','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
e0a65194	1	#! /usr/bin/env perl
d37a65bc	2	# Author: Marc Bevand <bevand_m (at) epita.fr>
e0a65194	3	# Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
d37a65bc	4	#
e0a65194 RS	5	# Licensed under the OpenSSL license (the "License"). You may not use
	6	# this file except in compliance with the License. You can obtain a copy
	7	# in the file LICENSE in the source distribution or at
	8	# https://www.openssl.org/source/license.html
	9
	10	# MD5 optimized for AMD64.
d37a65bc AP	11
	12	use strict;
	13
	14	my $code;
	15
	16	# round1_step() does:
	17	# dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
	18	# %r10d = X[k_next]
	19	# %r11d = z' (copy of z for the next step)
9c4fe782	20	# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
d37a65bc AP	21	sub round1_step
	22	{
	23	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
	24	$code .= " mov 04(%rsi), %r10d / (NEXT STEP) X[0] */\n" if ($pos == -1);
	25	$code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
	26	$code .= <<EOF;
	27	xor $y, %r11d /* y ^ ... */
	28	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
	29	and $x, %r11d /* x & ... */
d37a65bc	30	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
2036c9a5	31	xor $z, %r11d /* z ^ ... */
d37a65bc AP	32	add %r11d, $dst /* dst += ... */
	33	rol \$$s, $dst /* dst <<< s */
	34	mov $y, %r11d /* (NEXT STEP) z' = $y */
	35	add $x, $dst /* dst += x */
	36	EOF
	37	}
	38
	39	# round2_step() does:
	40	# dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
	41	# %r10d = X[k_next]
9c4fe782 AP	42	# %r11d = z' (copy of z for the next step)
	43	# %r12d = z' (copy of z for the next step)
	44	# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
d37a65bc AP	45	sub round2_step
	46	{
	47	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
9c4fe782 AP	48	$code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
9c4fe782 AP	49	$code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
d37a65bc	50	$code .= <<EOF;
9c4fe782	51	not %r11d /* not z */
9c4fe782	52	and $x, %r12d /* x & z */
b943b7d2	53	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
9c4fe782	54	and $y, %r11d /* y & (not z) */
d37a65bc	55	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
9c4fe782 AP	56	or %r11d, %r12d /* (y & (not z)) \| (x & z) */
	57	mov $y, %r11d /* (NEXT STEP) z' = $y */
	58	add %r12d, $dst /* dst += ... */
	59	mov $y, %r12d /* (NEXT STEP) z' = $y */
d37a65bc	60	rol \$$s, $dst /* dst <<< s */
d37a65bc AP	61	add $x, $dst /* dst += x */
	62	EOF
	63	}
	64
	65	# round3_step() does:
	66	# dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
	67	# %r10d = X[k_next]
	68	# %r11d = y' (copy of y for the next step)
9c4fe782	69	# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
b943b7d2	70	{ my $round3_alter=0;
d37a65bc AP	71	sub round3_step
	72	{
	73	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
d37a65bc AP	74	$code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
	75	$code .= <<EOF;
	76	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
d37a65bc	77	xor $z, %r11d /* z ^ ... */
2036c9a5	78	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
d37a65bc AP	79	xor $x, %r11d /* x ^ ... */
d37a65bc AP	80	add %r11d, $dst /* dst += ... */
b943b7d2 AP	81	EOF
b943b7d2 AP	82	$code .= <<EOF if ($round3_alter);
d37a65bc AP	83	rol \$$s, $dst /* dst <<< s */
d37a65bc AP	84	mov $x, %r11d /* (NEXT STEP) y' = $x */
b943b7d2 AP	85	EOF
	86	$code .= <<EOF if (!$round3_alter);
	87	mov $x, %r11d /* (NEXT STEP) y' = $x */
	88	rol \$$s, $dst /* dst <<< s */
	89	EOF
	90	$code .= <<EOF;
d37a65bc AP	91	add $x, $dst /* dst += x */
d37a65bc AP	92	EOF
b943b7d2 AP	93	$round3_alter^=1;
b943b7d2 AP	94	}
d37a65bc AP	95	}
	96
	97	# round4_step() does:
	98	# dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
	99	# %r10d = X[k_next]
	100	# %r11d = not z' (copy of not z for the next step)
9c4fe782	101	# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
d37a65bc AP	102	sub round4_step
	103	{
	104	my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
d37a65bc AP	105	$code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1);
	106	$code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n"
	107	if ($pos == -1);
	108	$code .= <<EOF;
	109	lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
	110	or $x, %r11d /* x \| ... */
2036c9a5	111	mov $k_next4(%rsi),%r10d / (NEXT STEP) X[$k_next] */
d37a65bc AP	112	xor $y, %r11d /* y ^ ... */
d37a65bc AP	113	add %r11d, $dst /* dst += ... */
d37a65bc AP	114	mov \$0xffffffff, %r11d
	115	rol \$$s, $dst /* dst <<< s */
	116	xor $y, %r11d /* (NEXT STEP) not z' = not $y */
	117	add $x, $dst /* dst += x */
	118	EOF
	119	}
	120
4d86e8df	121	no warnings qw(uninitialized);
be01f79d AP	122	my $flavour = shift;
	123	my $output = shift;
	124	if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
	125
	126	my $win64=0; $win64=1 if ($flavour =~ /[nm]asm\|mingw64/ \|\| $output =~ /\.asm$/);
abe7f8b4 AP	127
	128	$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
	129	( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
	130	( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
	131	die "can't locate x86_64-xlate.pl";
	132
cfe1d992	133	open OUT,"\| \"$^X\" \"$xlate\" $flavour \"$output\"";
46bf83f0	134	STDOUT=OUT;
d37a65bc AP	135
	136	$code .= <<EOF;
	137	.text
	138	.align 16
	139
c69ed6ea AP	140	.globl md5_block_asm_data_order
	141	.type md5_block_asm_data_order,\@function,3
	142	md5_block_asm_data_order:
d37a65bc AP	143	push %rbp
d37a65bc AP	144	push %rbx
9c4fe782	145	push %r12
d37a65bc AP	146	push %r14
d37a65bc AP	147	push %r15
be01f79d	148	.Lprologue:
d37a65bc AP	149
	150	# rdi = arg #1 (ctx, MD5_CTX pointer)
	151	# rsi = arg #2 (ptr, data pointer)
	152	# rdx = arg #3 (nbr, number of 16-word blocks to process)
	153	mov %rdi, %rbp # rbp = ctx
	154	shl \$6, %rdx # rdx = nbr in bytes
	155	lea (%rsi,%rdx), %rdi # rdi = end
	156	mov 0*4(%rbp), %eax # eax = ctx->A
	157	mov 1*4(%rbp), %ebx # ebx = ctx->B
	158	mov 2*4(%rbp), %ecx # ecx = ctx->C
	159	mov 3*4(%rbp), %edx # edx = ctx->D
d37a65bc AP	160	# end is 'rdi'
	161	# ptr is 'rsi'
	162	# A is 'eax'
	163	# B is 'ebx'
	164	# C is 'ecx'
	165	# D is 'edx'
	166
	167	cmp %rdi, %rsi # cmp end with ptr
8b5bf52a	168	je .Lend # jmp if ptr == end
d37a65bc AP	169
d37a65bc AP	170	# BEGIN of loop over 16-word blocks
8b5bf52a	171	.Lloop: # save old values of A, B, C, D
d37a65bc AP	172	mov %eax, %r8d
	173	mov %ebx, %r9d
	174	mov %ecx, %r14d
	175	mov %edx, %r15d
	176	EOF
	177	round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
	178	round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
	179	round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
	180	round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
	181	round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
	182	round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
	183	round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
	184	round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
	185	round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
	186	round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
	187	round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
	188	round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
	189	round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
	190	round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
	191	round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
2036c9a5	192	round1_step( 1,'%ebx','%ecx','%edx','%eax', '1','0x49b40821','22');
d37a65bc AP	193
	194	round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
	195	round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
	196	round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
	197	round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
	198	round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
	199	round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
	200	round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
	201	round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
	202	round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
	203	round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
	204	round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
	205	round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
	206	round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
	207	round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
	208	round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
2036c9a5	209	round2_step( 1,'%ebx','%ecx','%edx','%eax', '5','0x8d2a4c8a','20');
d37a65bc AP	210
	211	round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
	212	round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
	213	round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
	214	round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
	215	round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
	216	round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
	217	round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
	218	round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
	219	round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
	220	round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
	221	round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
	222	round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
	223	round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
	224	round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
	225	round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
	226	round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
	227
	228	round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
	229	round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
	230	round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
	231	round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
	232	round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
	233	round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
	234	round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
	235	round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
	236	round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
	237	round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
	238	round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
	239	round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
	240	round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
	241	round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
	242	round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
	243	round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
	244	$code .= <<EOF;
	245	# add old values of A, B, C, D
	246	add %r8d, %eax
	247	add %r9d, %ebx
	248	add %r14d, %ecx
	249	add %r15d, %edx
	250
	251	# loop control
	252	add \$64, %rsi # ptr += 64
	253	cmp %rdi, %rsi # cmp end with ptr
8b5bf52a	254	jb .Lloop # jmp if ptr < end
d37a65bc AP	255	# END of loop over 16-word blocks
d37a65bc AP	256
8b5bf52a	257	.Lend:
d37a65bc AP	258	mov %eax, 0*4(%rbp) # ctx->A = A
	259	mov %ebx, 1*4(%rbp) # ctx->B = B
	260	mov %ecx, 2*4(%rbp) # ctx->C = C
	261	mov %edx, 3*4(%rbp) # ctx->D = D
	262
be01f79d AP	263	mov (%rsp),%r15
	264	mov 8(%rsp),%r14
	265	mov 16(%rsp),%r12
	266	mov 24(%rsp),%rbx
	267	mov 32(%rsp),%rbp
	268	add \$40,%rsp
	269	.Lepilogue:
	270	ret
	271	.size md5_block_asm_data_order,.-md5_block_asm_data_order
	272	EOF
	273
	274	# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
	275	# CONTEXT context,DISPATCHER_CONTEXT disp)
	276	if ($win64) {
	277	my $rec="%rcx";
	278	my $frame="%rdx";
	279	my $context="%r8";
	280	my $disp="%r9";
	281
	282	$code.=<<___;
	283	.extern __imp_RtlVirtualUnwind
	284	.type se_handler,\@abi-omnipotent
	285	.align 16
	286	se_handler:
	287	push %rsi
	288	push %rdi
	289	push %rbx
	290	push %rbp
	291	push %r12
	292	push %r13
	293	push %r14
	294	push %r15
	295	pushfq
	296	sub \$64,%rsp
	297
	298	mov 120($context),%rax # pull context->Rax
	299	mov 248($context),%rbx # pull context->Rip
	300
	301	lea .Lprologue(%rip),%r10
	302	cmp %r10,%rbx # context->Rip<.Lprologue
	303	jb .Lin_prologue
	304
	305	mov 152($context),%rax # pull context->Rsp
	306
	307	lea .Lepilogue(%rip),%r10
	308	cmp %r10,%rbx # context->Rip>=.Lepilogue
	309	jae .Lin_prologue
	310
	311	lea 40(%rax),%rax
	312
	313	mov -8(%rax),%rbp
	314	mov -16(%rax),%rbx
	315	mov -24(%rax),%r12
	316	mov -32(%rax),%r14
	317	mov -40(%rax),%r15
	318	mov %rbx,144($context) # restore context->Rbx
	319	mov %rbp,160($context) # restore context->Rbp
	320	mov %r12,216($context) # restore context->R12
	321	mov %r14,232($context) # restore context->R14
	322	mov %r15,240($context) # restore context->R15
	323
	324	.Lin_prologue:
	325	mov 8(%rax),%rdi
	326	mov 16(%rax),%rsi
327	mov %rax,152($context) # restore context->Rsp
328	mov %rsi,168($context) # restore context->Rsi
329	mov %rdi,176($context) # restore context->Rdi
330
331	mov 40($disp),%rdi # disp->ContextRecord
332	mov $context,%rsi # context
333	mov \$154,%ecx # sizeof(CONTEXT)
334	.long 0xa548f3fc # cld; rep movsq
335
336	mov $disp,%rsi
337	xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
338	mov 8(%rsi),%rdx # arg2, disp->ImageBase
339	mov 0(%rsi),%r8 # arg3, disp->ControlPc
340	mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
341	mov 40(%rsi),%r10 # disp->ContextRecord
342	lea 56(%rsi),%r11 # &disp->HandlerData
343	lea 24(%rsi),%r12 # &disp->EstablisherFrame
344	mov %r10,32(%rsp) # arg5
345	mov %r11,40(%rsp) # arg6
346	mov %r12,48(%rsp) # arg7
347	mov %rcx,56(%rsp) # arg8, (NULL)
348	call *__imp_RtlVirtualUnwind(%rip)
349
350	mov \$1,%eax # ExceptionContinueSearch
351	add \$64,%rsp
352	popfq
d37a65bc AP	353	pop %r15
d37a65bc AP	354	pop %r14
be01f79d	355	pop %r13
9c4fe782	356	pop %r12
d37a65bc	357	pop %rbp
be01f79d AP	358	pop %rbx
	359	pop %rdi
	360	pop %rsi
d37a65bc	361	ret
be01f79d AP	362	.size se_handler,.-se_handler
	363
	364	.section .pdata
	365	.align 4
	366	.rva .LSEH_begin_md5_block_asm_data_order
	367	.rva .LSEH_end_md5_block_asm_data_order
	368	.rva .LSEH_info_md5_block_asm_data_order
	369
	370	.section .xdata
	371	.align 8
	372	.LSEH_info_md5_block_asm_data_order:
	373	.byte 9,0,0,0
	374	.rva se_handler
	375	___
	376	}
d37a65bc AP	377
d37a65bc AP	378	print $code;
8b5bf52a AP	379
8b5bf52a AP	380	close STDOUT;