Throw in x86_64 AT&T to MASM assembler converter to facilitate development

of dual-ABI Unix/Win64 modules.

crypto/perlasm/x86_64-xlate.pl

+#!/usr/bin/env perl
+
+# Ascetic x86_64 AT&T to MASM assembler translator by <appro>.
+#
+# Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
+# format is way easier to parse. Because it's simpler to "gear" from
+# Unix ABI to Windows one [see cross-reference "card" at the end of
+# file]. Because Linux targets were available first...
+#
+# In addition the script also "distills" code suitable for GNU
+# assembler, so that it can be compiled with more rigid assemblers,
+# such as Solaris /usr/ccs/bin/as.
+#
+# This translator is not designed to convert *arbitrary* assembler
+# code from AT&T format to MASM one. It's designed to convert just
+# enough to provide for dual-ABI OpenSSL modules development...
+# There *are* limitations and you might have to modify your assembler
+# code or this script to achieve the desired result...
+#
+# Currently recognized limitations:
+#
+# - can't use multiple ops per line;
+# - indirect calls and jumps are not supported;
+#
+# Dual-ABI styling rules.
+#
+# 1. Adhere to Unix register and stack layout [see the end for
+#    explanation].
+# 2. Forget about "red zone," stick to more traditional blended
+#    stack frame allocation. If volatile storage is actually required
+#    that is. If not, just leave the stack as is.
+# 3. Functions tagged with ".type name,@function" get crafted with
+#    unified Windows prologue and epilogue automatically. If you want
+#    to take care of ABI differences yourself, tag functions as
+#    ".type name,@abi-omnipotent."
+# 4. As minor optimization you can specify number of input arguments
+#    as ".type name,@function,N." Keep in mind that if N is larger
+#    than 6, then you *have to* write "abi-omnipotent" code, because
+#    it can't be resolved with unified prologue.
+# 5. Name local labels as .L*.
+# 6. Don't use repret, it's generated automatically.
+
+my $output = shift;
+open STDOUT,">$output" || die "can't open $output: $!";
+
+my $masm=1 if ($output =~ /\.asm/);
+
+my $current_segment;
+my $current_function;
+
+{ package opcode;	# pick up opcodes
+    sub re {
+	my	$self = shift;	# single instance in enough...
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /^([a-z]+)/i) {
+	    $self->{op} = $1;
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+
+	    undef $self->{sz};
+	    if ($self->{op} =~ /(movz)b.*/) {	# movz is pain...
+		$self->{op} = $1;
+		$self->{sz} = "b";
+	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])/) {
+		$self->{op} = $1;
+		$self->{sz} = $2;
+	    }
+	}
+	$ret;
+    }
+    sub size {
+	my $self = shift;
+	my $sz   = shift;
+	$self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
+	$self->{sz};
+    }
+    sub out {
+	my $self = shift;
+	if (!$masm) {
+	    if ($self->{op} eq "movz") {	# movz in pain...
+		sprintf "%s%s%s",$self->{op},$self->{sz},shift;
+	    } elsif ($self->{op} eq "ret") {
+	    	".byte	0xf3,0xc3";
+	    } else {
+		"$self->{op}$self->{sz}";
+	    }
+	} else {
+	    $self->{op} =~ s/movz/movzx/;
+	    if ($self->{op} eq "ret") {
+		$self->{op} = "";
+		if ($current_function->{abi} eq "svr4") {
+		    $self->{op} = "mov	rdi,QWORD PTR 8[rsp]\t;WIN64 epilogue\n\t".
+				  "mov	rsi,QWORD PTR 16[rsp]\n\t";
+	    	}
+		$self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
+	    }
+	    $self->{op};
+	}
+    }
+}
+{ package const;	# pick up constants, which start with $
+    sub re {
+	my	$self = shift;	# single instance in enough...
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /^\$([^,]+)/) {
+	    $self->{value} = $1;
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+	}
+	$ret;
+    }
+    sub out {
+    	my $self = shift;
+	sprintf $masm?"%s":"\$%s",$self->{value};
+    }
+}
+{ package ea;		# pick up effective addresses: expr(%reg,%reg,scale)
+    sub re {
+	my	$self = shift;	# single instance in enough...
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /^([^\(,]*)\(([%\w,]+)\)/) {
+	    $self->{label} = $1;
+	    ($self->{base},$self->{index},$self->{scale})=split(/,/,$2);
+	    $self->{scale} = 1 if (!defined($self->{scale}));
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+
+	    $self->{label} =~ s/\.L/\$L/g;
+	    $self->{base}  =~ s/^%//;
+	    $self->{index} =~ s/^%// if (defined($self->{index}));
+	}
+	$ret;
+    }
+    sub size {}
+    sub out {
+    	my $self = shift;
+	my $sz = shift;
+
+	if (!$masm) {
+	    if (defined($self->{index})) {
+		sprintf "%s(%%%s,%%%s,%d)",	$self->{label},$self->{base},
+					$self->{index},$self->{scale};
+	    }
+	    else {
+		sprintf "%s(%%%s)",	$self->{label},$self->{base};
+	    }
+	} else {
+	    %szmap = ( b=>"BYTE", w=>"WORD", l=>"DWORD", q=>"QWORD" );
+
+	    if (defined($self->{index})) {
+		sprintf "%s PTR %s[%s*%d+%s]",$szmap{$sz},
+					$self->{label},
+					$self->{index},$self->{scale},
+					$self->{base};
+	    }
+	    else {
+		sprintf "%s PTR %s[%s]",$szmap{$sz},
+					$self->{label},$self->{base};
+	    }
+	}
+    }
+}
+{ package register;	# pick up registers, which start with %.
+    sub re {
+	my	$class = shift;	# muliple instances...
+	my	$self = {};
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /^%(\w+)/) {
+	    bless $self,$class;
+	    $self->{value} = $1;
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+	}
+	$ret;
+    }
+    sub size {
+	my	$self = shift;
+	undef	$ret;
+
+	if    ($self->{value} =~ /^r[\d]+b$/i)	{ $ret="b"; }
+	elsif ($self->{value} =~ /^r[\d]+w$/i)	{ $ret="w"; }
+	elsif ($self->{value} =~ /^r[\d]+d$/i)	{ $ret="l"; }
+	elsif ($self->{value} =~ /^r[\w]+$/i)	{ $ret="q"; }
+	elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
+	elsif ($self->{value} =~ /^[\w]{2}l$/i)	{ $ret="b"; }
+	elsif ($self->{value} =~ /^[\w]{2}$/i)	{ $ret="w"; }
+	elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
+
+	$ret;
+    }
+    sub out {
+    	my $self = shift;
+	sprintf $masm?"%s":"%%%s",$self->{value};
+    }
+}
+{ package label;	# pick up labels, which end with :
+    sub re {
+	my	$self = shift;	# single instance is enough...
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /(^[\.\w]+\:)/) {
+	    $self->{value} = $1;
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+
+	    $self->{value} =~ s/\.L/\$L/ if ($masm);
+	}
+	$ret;
+    }
+    sub out {
+	my $self = shift;
+
+	if (!$masm) {
+	    $self->{value};
+	} elsif ($self->{value} ne "$current_function->{name}:") {
+	    $self->{value};
+	} elsif ($current_function->{abi} eq "svr4") {
+	    my $func =	"$current_function->{name}	PROC\n".
+			"	mov	QWORD PTR 8[rsp],rdi\t;WIN64 prologue\n".
+			"	mov	QWORD PTR 16[rsp],rsi\n";
+	    my $narg = $current_function->{narg};
+	    $narg=6 if (!defined($narg));
+	    $func .= "	mov	rdi,rcx\n" if ($narg>0);
+	    $func .= "	mov	rsi,rdx\n" if ($narg>1);
+	    $func .= "	mov	rdx,r8\n"  if ($narg>2);
+	    $func .= "	mov	rcx,r9\n"  if ($narg>3);
+	    $func .= "	mov	r8,QWORD PTR 40[rsp]\n" if ($narg>4);
+	    $func .= "	mov	r9,QWORD PTR 48[rsp]\n" if ($narg>5);
+	    $func .= "\n";
+	} else {
+	   "$current_function->{name}	PROC";
+	}
+    }
+}
+{ package expr;		# pick up expressioins
+    sub re {
+	my	$self = shift;	# single instance is enough...
+	local	*line = shift;
+	undef	$ret;
+
+	if ($line =~ /(^[^,]+)/) {
+	    $self->{value} = $1;
+	    $ret = $self;
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+
+	    $self->{value} =~ s/\.L/\$L/g if ($masm);
+	}
+	$ret;
+    }
+    sub out {
+	my $self = shift;
+	$self->{value};
+    }
+}
+{ package directive;	# pick up directives, which start with .
+    sub re {
+	my	$self = shift;	# single instance is enough...
+	local	*line = shift;
+	undef	$ret;
+	my	$dir;
+
+	if ($line =~ /^\s*(\.\w+)/) {
+	    if (!$masm) {
+		$self->{value} = $1;
+		$line =~ s/\@abi\-omnipotent/\@function/;
+		$line =~ s/\@function.*/\@function/;
+		$self->{value} = $line;
+		$line = "";
+		return $self;
+	    }
+
+	    $dir = $1;
+	    $ret = $self;
+	    undef $self->{value};
+	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
+	    SWITCH: for ($dir) {
+		/\.(text|data)/
+			    && do { my $v=undef;
+				    $v="$current_segment\tENDS\n" if ($current_segment);
+				    $current_segment = "_$1";
+				    $current_segment =~ tr/[a-z]/[A-Z]/;
+				    $v.="$current_segment\tSEGMENT PARA";
+				    $self->{value} = $v;
+				    last;
+				  };
+		/\.globl/   && do { $self->{value} = "PUBLIC\t".$line; last; };
+		/\.type/    && do { ($sym,$type,$narg) = split(',',$line);
+				    if ($type eq "\@function")
+				    {	undef $current_function;
+					$current_function->{name} = $sym;
+					$current_function->{abi}  = "svr4";
+					$current_function->{narg} = $narg;
+				    }
+				    elsif ($type eq "\@abi-omnipotent")
+				    {	undef $current_function;
+					$current_function->{name} = $sym;
+				    }
+				    last;
+				  };
+		/\.size/    && do { if (defined($current_function))
+				    {	$self->{value}="$current_function->{name}\tENDP";
+					undef $current_function;
+				    }
+				    last;
+				  };
+		/\.align/   && do { $self->{value} = "ALIGN\t".$line; last; };
+		/\.(byte|value|long|quad)/
+			    && do { my @arr = split(',',$line);
+				    my $sz  = substr($1,0,1);
+				    my $last = pop(@arr);
+
+				    $sz =~ tr/bvlq/BWDQ/;
+				    $self->{value} = "\tD$sz\t";
+				    for (@arr) { $self->{value} .= sprintf"0%Xh,",oct; }
+				    $self->{value} .= sprintf"0%Xh",oct($last);
+				    last;
+				  };
+	    }
+	    $line = "";
+	}
+
+	$ret;
+    }
+    sub out {
+	my $self = shift;
+	$self->{value};
+    }
+}
+
+while($line=<>) {
+
+    chomp($line);
+
+    $line =~ s/\[#!].*$//;	# get rid of comments...
+    $line =~ s/^\s+//;		# ... and skip white spaces
+
+    undef $label;
+    undef $opcode;
+    undef $dst;
+    undef $src;
+    undef $sz;
+
+    if ($label=label->re(\$line))	{ print $label->out(); }
+
+    if (directive->re(\$line)) {
+	printf "%s",directive->out();
+    } elsif ($opcode=opcode->re(\$line)) { ARGUMENT: {
+
+	if ($src=register->re(\$line))	{ opcode->size($src->size()); }
+	elsif ($src=const->re(\$line))	{ }
+	elsif ($src=ea->re(\$line))	{ }
+	elsif ($src=expr->re(\$line))	{ }
+
+	last ARGUMENT if ($line !~ /^,/);
+
+	$line = substr($line,1); $line =~ s/^\s+//;
+
+	if ($dst=register->re(\$line))	{ opcode->size($dst->size()); }
+	elsif ($dst=const->re(\$line))	{ }
+	elsif ($dst=ea->re(\$line))	{ }
+
+	} # ARGUMENT:
+
+	$sz=opcode->size();
+
+	if (defined($dst)) {
+	    if (!$masm) {
+		printf "\t%s\t%s,%s",	$opcode->out($dst->size()),
+					$src->out($sz),$dst->out($sz);
+	    }
+	    else {
+		printf "\t%s\t%s,%s",	$opcode->out(),
+					$dst->out($sz),$src->out($sz);
+	    }
+	}
+	elsif (defined($src)) {
+	    printf "\t%s\t%s",$opcode->out(),$src->out($sz);
+	} else {
+	    printf "\t%s",$opcode->out();
+	}
+    }
+
+    print $line,"\n";
+}
+
+print "\n$current_segment\tENDS\nEND\n" if ($masm);
+
+close STDOUT;
+
+#################################################
+# Cross-reference x86_64 ABI "card"
+#
+# 		Unix		Win64
+# %rax		*		*
+# %rbx		-		-
+# %rcx		#4		#1
+# %rdx		#3		#2
+# %rsi		#2		-
+# %rdi		#1		-
+# %rbp		-		-
+# %rsp		-		-
+# %r8		#5		#3
+# %r9		#6		#4
+# %r10		*		*
+# %r11		*		*
+# %r12		-		-
+# %r13		-		-
+# %r14		-		-
+# %r15		-		-
+# 
+# (*)	volatile register
+# (-)	preserved by callee
+# (#)	Nth argument, volatile
+#
+# In Unix terms top of stack is argument transfer area for arguments
+# which could not be accomodated in registers. Or in other words 7th
+# [integer] argument resides at 8(%rsp) upon function entry point.
+# 128 bytes above %rsp constitute a "red zone" which is not touched
+# by signal handlers and can be used as temporal storage without
+# allocating a frame.
+#
+# In Win64 terms N*8 bytes on top of stack is argument transfer area,
+# which belongs to/can be overwritten by callee. N is the number of
+# arguments passed to callee, *but* not less than 4! This means that
+# upon function entry point 5th argument resides at 40(%rsp), as well
+# as that 32 bytes from 8(%rsp) can always be used as temporal
+# storage [without allocating a frame].
+#
+# All the above means that if assembler programmer adheres to Unix
+# register and stack layout, but disregards the "red zone" existense,
+# it's possible to use following prologue and epilogue to "gear" from
+# Unix to Win64 ABI in leaf functions with not more than 6 arguments.
+#
+# omnipotent_function:
+# ifdef WIN64
+#	movq	%rdi,8(%rsp)
+#	movq	%rsi,16(%rsp)
+#	movq	%rcx,%rdi	; if 1st argument is actually present
+#	movq	%rdx,%rsi	; if 2nd argument is actually ...
+#	movq	%r8,%rdx	; if 3rd argument is ...
+#	movq	%r9,%rcx	; if 4th argument ...
+#	movq	40(%rsp),%r8	; if 5th ...
+#	movq	48(%rsp),%r9	; if 6th ...
+# endif
+#	...
+# ifdef WIN64
+#	movq	8(%rsp),%rdi
+#	movq	16(%rsp),%rsi
+# endif
+#	ret

crypto/rc4/asm/rc4-amd64.pl

@@ -35,145 +35,99 @@
 # of code remain redundant.
 
 $output=shift;
+open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
 
-$win64a=1 if ($output =~ /win64a.[s|asm]/);
-
-open STDOUT,">$output" || die "can't open $output: $!";
-
-if (defined($win64a)) {
-    $dat="%rcx";	# arg1
-    $len="%rdx";	# arg2
-    $inp="%rsi";	# r8, arg3 moves here
-    $out="%rdi";	# r9, arg4 moves here
-} else {
-    $dat="%rdi";	# arg1
-    $len="%rsi";	# arg2
-    $inp="%rdx";	# arg3
-    $out="%rcx";	# arg4
-}
+$dat="%rdi";	    # arg1
+$len="%rsi";	    # arg2
+$inp="%rdx";	    # arg3
+$out="%rcx";	    # arg4
 
 $XX="%r10";
 $TX="%r8";
 $YY="%r11";
 $TY="%r9";
 
-sub PTR() {
-    my $ret=shift;
-    if (defined($win64a)) {
-	$ret =~ s/\[([\S]+)\+([\S]+)\]/[$2+$1]/g;   # [%rN+%rM*4]->[%rM*4+%rN]
-	$ret =~ s/:([^\[]+)\[([^\]]+)\]/:[$2+$1]/g; # :off[ea]->:[ea+off]
-    } else {
-	$ret =~ s/[\+\*]/,/g;		# [%rN+%rM*4]->[%rN,%rM,4]
-	$ret =~ s/\[([^\]]+)\]/($1)/g;	# [%rN]->(%rN)
-    }
-    $ret;
-}
-
-$code=<<___ if (!defined($win64a));
+$code=<<___;
 .text
 
 .globl	RC4
-.type	RC4,\@function
+.type	RC4,\@function,4
 .align	16
 RC4:	or	$len,$len
 	jne	.Lentry
-	repret
+	ret
 .Lentry:
-___
-$code=<<___ if (defined($win64a));
-_TEXT	SEGMENT
-PUBLIC	RC4
-ALIGN	16
-RC4	PROC
-	or	$len,$len
-	jne	.Lentry
-	repret
-.Lentry:
-	push	%rdi
-	push	%rsi
-	sub	\$40,%rsp
-	mov	%r8,$inp
-	mov	%r9,$out
-___
-$code.=<<___;
 	add	\$8,$dat
-	movl	`&PTR("DWORD:-8[$dat]")`,$XX#d
-	movl	`&PTR("DWORD:-4[$dat]")`,$YY#d
-	cmpl	\$-1,`&PTR("DWORD:256[$dat]")`
+	movl	-8($dat),$XX#d
+	movl	-4($dat),$YY#d
+	cmpl	\$-1,256($dat)
 	je	.LRC4_CHAR
 	test	\$-8,$len
 	jz	.Lloop1
 .align	16
 .Lloop8:
 	inc	$XX#b
-	movl	`&PTR("DWORD:[$dat+$XX*4]")`,$TX#d
+	movl	($dat,$XX,4),$TX#d
 	add	$TX#b,$YY#b
-	movl	`&PTR("DWORD:[$dat+$YY*4]")`,$TY#d
-	movl	$TX#d,`&PTR("DWORD:[$dat+$YY*4]")`
-	movl	$TY#d,`&PTR("DWORD:[$dat+$XX*4]")`
+	movl	($dat,$YY,4),$TY#d
+	movl	$TX#d,($dat,$YY,4)
+	movl	$TY#d,($dat,$XX,4)
 	add	$TX#b,$TY#b
 	inc	$XX#b
-	movl	`&PTR("DWORD:[$dat+$XX*4]")`,$TX#d
-	movb	`&PTR("BYTE:[$dat+$TY*4]")`,%al
+	movl	($dat,$XX,4),$TX#d
+	movb	($dat,$TY,4),%al
 ___
 for ($i=1;$i<=6;$i++) {
 $code.=<<___;
 	add	$TX#b,$YY#b
 	ror	\$8,%rax
-	movl	`&PTR("DWORD:[$dat+$YY*4]")`,$TY#d
-	movl	$TX#d,`&PTR("DWORD:[$dat+$YY*4]")`
-	movl	$TY#d,`&PTR("DWORD:[$dat+$XX*4]")`
+	movl	($dat,$YY,4),$TY#d
+	movl	$TX#d,($dat,$YY,4)
+	movl	$TY#d,($dat,$XX,4)
 	add	$TX#b,$TY#b
 	inc	$XX#b
-	movl	`&PTR("DWORD:[$dat+$XX*4]")`,$TX#d
-	movb	`&PTR("BYTE:[$dat+$TY*4]")`,%al
+	movl	($dat,$XX,4),$TX#d
+	movb	($dat,$TY,4),%al
 ___
 }
 $code.=<<___;
 	add	$TX#b,$YY#b
 	ror	\$8,%rax
-	movl	`&PTR("DWORD:[$dat+$YY*4]")`,$TY#d
-	movl	$TX#d,`&PTR("DWORD:[$dat+$YY*4]")`
-	movl	$TY#d,`&PTR("DWORD:[$dat+$XX*4]")`
+	movl	($dat,$YY,4),$TY#d
+	movl	$TX#d,($dat,$YY,4)
+	movl	$TY#d,($dat,$XX,4)
 	sub	\$8,$len
 	add	$TY#b,$TX#b
-	movb	`&PTR("BYTE:[$dat+$TX*4]")`,%al
+	movb	($dat,$TX,4),%al
 	ror	\$8,%rax
 	add	\$8,$inp
 	add	\$8,$out
 
-	xor	`&PTR("QWORD:-8[$inp]")`,%rax
-	mov	%rax,`&PTR("QWORD:-8[$out]")`
+	xor	-8($inp),%rax
+	mov	%rax,-8($out)
 
 	test	\$-8,$len
 	jnz	.Lloop8
 	cmp	\$0,$len
 	jne	.Lloop1
 .Lexit:
-	movl	$XX#d,`&PTR("DWORD:-8[$dat]")`
-	movl	$YY#d,`&PTR("DWORD:-4[$dat]")`
-___
-$code.=<<___ if (defined($win64a));
-	add	\$40,%rsp
-	pop	%rsi
-	pop	%rdi
-___
-$code.=<<___;
-	repret
+	movl	$XX#d,-8($dat)
+	movl	$YY#d,-4($dat)
+	ret
 .align	16
 .Lloop1:
-	movzb	`&PTR("BYTE:[$inp]")`,%eax
+	movzb	($inp),%eax
 	inc	$XX#b
-	movl	`&PTR("DWORD:[$dat+$XX*4]")`,$TX#d
+	movl	($dat,$XX,4),$TX#d
 	add	$TX#b,$YY#b
-	movl	`&PTR("DWORD:[$dat+$YY*4]")`,$TY#d
-	movl	$TX#d,`&PTR("DWORD:[$dat+$YY*4]")`
-	movl	$TY#d,`&PTR("DWORD:[$dat+$XX*4]")`
+	movl	($dat,$YY,4),$TY#d
+	movl	$TX#d,($dat,$YY,4)
+	movl	$TY#d,($dat,$XX,4)
 	add	$TY#b,$TX#b
-	movl	`&PTR("DWORD:[$dat+$TX*4]")`,$TY#d
+	movl	($dat,$TX,4),$TY#d
 	xor	$TY,%rax
 	inc	$inp
-	movb	%al,`&PTR("BYTE:[$out]")`
+	movb	%al,($out)
 	inc	$out
 	dec	$len
 	jnz	.Lloop1
@@ -182,46 +136,25 @@ $code.=<<___;
 .align	16
 .LRC4_CHAR:
 	add	\$1,$XX#b
-	movzb	`&PTR("BYTE:[$dat+$XX]")`,$TX#d
+	movzb	($dat,$XX),$TX#d
 	add	$TX#b,$YY#b
-	movzb	`&PTR("BYTE:[$dat+$YY]")`,$TY#d
-	movb	$TX#b,`&PTR("BYTE:[$dat+$YY]")`
-	movb	$TY#b,`&PTR("BYTE:[$dat+$XX]")`
+	movzb	($dat,$YY),$TY#d
+	movb	$TX#b,($dat,$YY)
+	movb	$TY#b,($dat,$XX)
 	add	$TX#b,$TY#b
-	movzb	`&PTR("BYTE:[$dat+$TY]")`,$TY#d
-	xorb	`&PTR("BYTE:[$inp]")`,$TY#b
-	movb	$TY#b,`&PTR("BYTE:[$out]")`
+	movzb	($dat,$TY),$TY#d
+	xorb	($inp),$TY#b
+	movb	$TY#b,($out)
 	lea	1($inp),$inp
 	lea	1($out),$out
 	sub	\$1,$len
 	jnz	.LRC4_CHAR
 	jmp	.Lexit
-___
-$code.=<<___ if (defined($win64a));
-RC4	ENDP
-_TEXT	ENDS
-END
-___
-$code.=<<___ if (!defined($win64a));
 .size	RC4,.-RC4
 ___
 
 $code =~ s/#([bwd])/$1/gm;
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
 
-if (defined($win64a)) {
-    $code =~ s/\.align/ALIGN/gm;
-    $code =~ s/[\$%]//gm;
-    $code =~ s/\.L/\$L/gm;
-    $code =~ s/([\w]+)([\s]+)([\S]+),([\S]+)/$1$2$4,$3/gm;
-    $code =~ s/([QD]*WORD|BYTE):/$1 PTR/gm;
-    $code =~ s/mov[bwlq]/mov/gm;
-    $code =~ s/movzb/movzx/gm;
-    $code =~ s/repret/DB\t0F3h,0C3h/gm;
-    $code =~ s/cmpl/cmp/gm;
-    $code =~ s/xorb/xor/gm;
-} else {
-    $code =~ s/([QD]*WORD|BYTE)://gm;
-    $code =~ s/repret/.byte\t0xF3,0xC3/gm;
-}
 print $code;
+
+close STDOUT;