X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fperlasm%2Fx86_64-xlate.pl;h=ed74b85ac96a40db973f952c52e6adbafe23b982;hp=1f138508249151530cb18e8888fb5e43525df45c;hb=41965a84c4bc0f6694fd17203920a6672e3fac7b;hpb=852537726574b0b568a3bd8d12e5c797c5e56c63

diff --git a/crypto/perlasm/x86_64-xlate.pl b/crypto/perlasm/x86_64-xlate.pl
index 1f13850824..ed74b85ac9 100755
--- a/crypto/perlasm/x86_64-xlate.pl
+++ b/crypto/perlasm/x86_64-xlate.pl
@@ -55,22 +55,21 @@
 #    Win64 prologue copies %rsp value to %rax. For further details
 #    see SEH paragraph at the end.
 # 9. .init segment is allowed to contain calls to functions only.
+# a. If function accepts more than 4 arguments *and* >4th argument
+#    is declared as non 64-bit value, do clear its upper part.
 
 my $flavour = shift;
 my $output  = shift;
 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 
-{ my ($stddev,$stdino,@junk)=stat(STDOUT);
-  my ($outdev,$outino,@junk)=stat($output);
-
-    open STDOUT,">$output" || die "can't open $output: $!"
-	if ($stddev!=$outdev || $stdino!=$outino);
-}
+open STDOUT,">$output" || die "can't open $output: $!"
+	if (defined($output));
 
 my $gas=1;	$gas=0 if ($output =~ /\.asm$/);
 my $elf=1;	$elf=0 if (!$gas);
 my $win64=0;
 my $prefix="";
+my $decor=".L";
 
 my $masmref=8 + 50727*2**-32;	# 8.00.50727 shipped with VS2005
 my $masm=0;
@@ -79,10 +78,13 @@ my $PTR=" PTR";
 my $nasmref=2.03;
 my $nasm=0;
 
-if    ($flavour eq "mingw64")	{ $gas=1; $elf=0; $win64=1; $prefix="_"; }
-elsif ($flavour eq "macosx")	{ $gas=1; $elf=0; $prefix="_"; }
-elsif ($flavour eq "masm")	{ $gas=0; $elf=0; $masm=$masmref; $win64=1; }
-elsif ($flavour eq "nasm")	{ $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $PTR=""; }
+if    ($flavour eq "mingw64")	{ $gas=1; $elf=0; $win64=1;
+				  $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
+				  chomp($prefix);
+				}
+elsif ($flavour eq "macosx")	{ $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
+elsif ($flavour eq "masm")	{ $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
+elsif ($flavour eq "nasm")	{ $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
 elsif (!$gas)
 {   if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
     {	$nasm = $1 + $2*0.01; $PTR="";  }
@@ -91,6 +93,7 @@ elsif (!$gas)
     die "no assembler found on %PATH" if (!($nasm || $masm));
     $win64=1;
     $elf=0;
+    $decor="\$L\$";
 }
 
 my $current_segment;
@@ -109,11 +112,17 @@ my %globals;
 	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
 
 	    undef $self->{sz};
-	    if ($self->{op} =~ /^(movz)b.*/) {	# movz is pain...
+	    if ($self->{op} =~ /^(movz)x?([bw]).*/) {	# movz is pain...
 		$self->{op} = $1;
-		$self->{sz} = "b";
+		$self->{sz} = $2;
 	    } elsif ($self->{op} =~ /call|jmp/) {
-		$self->{sz} = ""
+		$self->{sz} = "";
+	    } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
+		$self->{sz} = "";
+	    } elsif ($self->{op} =~ /^v/) { # VEX
+		$self->{sz} = "";
+	    } elsif ($self->{op} =~ /movq/ && $line =~ /%xmm/) {
+		$self->{sz} = "";
 	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
 		$self->{op} = $1;
 		$self->{sz} = $2;
@@ -158,7 +167,7 @@ my %globals;
 	    } elsif ($self->{op} =~ /^(pop|push)f/) {
 		$self->{op} .= $self->{sz};
 	    } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
-		$self->{op} = "ALIGN\t8\n\tDQ";
+		$self->{op} = "\tDQ";
 	    } 
 	    $self->{op};
 	}
@@ -189,11 +198,12 @@ my %globals;
 	if ($gas) {
 	    # Solaris /usr/ccs/bin/as can't handle multiplications
 	    # in $self->{value}
-	    $self->{value} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
+	    $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
 	    $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
 	    sprintf "\$%s",$self->{value};
 	} else {
-	    $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig;
+	    $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig;
+	    $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
 	    sprintf "%s",$self->{value};
 	}
     }
@@ -228,6 +238,7 @@ my %globals;
 	my $sz = shift;
 
 	$self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
+	$self->{label} =~ s/\.L/$decor/g;
 
 	# Silently convert all EAs to 64-bit. This is required for
 	# elder GNU assembler and results in more compact code,
@@ -235,34 +246,45 @@ my %globals;
 	$self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
 	$self->{base}  =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
 
+	# Solaris /usr/ccs/bin/as can't handle multiplications
+	# in $self->{label}, new gas requires sign extension...
+	use integer;
+	$self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
+	$self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
+	$self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
+
+	if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
+	    $self->{base} =~ /(rbp|r13)/) {
+		$self->{base} = $self->{index}; $self->{index} = $1;
+	}
+
 	if ($gas) {
-	    # Solaris /usr/ccs/bin/as can't handle multiplications
-	    # in $self->{label}
-	    $self->{label} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
-	    $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
-	    $self->{label} =~ s/\.L/L\$/g	if (!$elf);
+	    $self->{label} =~ s/^___imp_/__imp__/   if ($flavour eq "mingw64");
 
 	    if (defined($self->{index})) {
-		sprintf "%s%s(%%%s,%%%s,%d)",$self->{asterisk},
-					$self->{label},$self->{base},
+		sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk},
+					$self->{label},
+					$self->{base}?"%$self->{base}":"",
 					$self->{index},$self->{scale};
 	    } else {
 		sprintf "%s%s(%%%s)",	$self->{asterisk},$self->{label},$self->{base};
 	    }
 	} else {
-	    %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", q=>"QWORD$PTR" );
+	    %szmap = (	b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR",
+	    		q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR",
+			y=>"" );
 
-	    $self->{label} =~ s/\.L/\$L\$/g;
 	    $self->{label} =~ s/\./\$/g;
-	    $self->{label} =~ s/0x([0-9a-f]+)/0$1h/ig;
+	    $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
 	    $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
-	    $sz="q" if ($self->{asterisk});
+	    $sz="q" if ($self->{asterisk} || opcode->mnemonic() =~ /^v?movq$/);
+	    $sz="l" if (opcode->mnemonic() =~ /^v?movd$/);
 
 	    if (defined($self->{index})) {
-		sprintf "%s[%s%s*%d+%s]",$szmap{$sz},
+		sprintf "%s[%s%s*%d%s]",$szmap{$sz},
 					$self->{label}?"$self->{label}+":"",
 					$self->{index},$self->{scale},
-					$self->{base};
+					$self->{base}?"+$self->{base}":"";
 	    } elsif ($self->{base} eq "rip") {
 		sprintf "%s[%s]",$szmap{$sz},$self->{label};
 	    } else {
@@ -322,8 +344,7 @@ my %globals;
 	    $ret = $self;
 	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
 
-	    $self->{value} =~ s/\.L/\$L\$/ if (!$gas);
-	    $self->{value} =~ s/\.L/L\$/   if (!$elf);
+	    $self->{value} =~ s/^\.L/$decor/;
 	}
 	$ret;
     }
@@ -339,7 +360,7 @@ my %globals;
 		$func .= "	movq	%rdi,8(%rsp)\n";
 		$func .= "	movq	%rsi,16(%rsp)\n";
 		$func .= "	movq	%rsp,%rax\n";
-		$func .= ".LSEH_begin_$current_function->{name}:\n";
+		$func .= "${decor}SEH_begin_$current_function->{name}:\n";
 		my $narg = $current_function->{narg};
 		$narg=6 if (!defined($narg));
 		$func .= "	movq	%rcx,%rdi\n" if ($narg>0);
@@ -360,7 +381,7 @@ my %globals;
 	    $func .= "	mov	QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n";
 	    $func .= "	mov	QWORD${PTR}[16+rsp],rsi\n";
 	    $func .= "	mov	rax,rsp\n";
-	    $func .= "\$L\$SEH_begin_$current_function->{name}:";
+	    $func .= "${decor}SEH_begin_$current_function->{name}:";
 	    $func .= ":" if ($masm);
 	    $func .= "\n";
 	    my $narg = $current_function->{narg};
@@ -391,14 +412,13 @@ my %globals;
 
 	    $self->{value} =~ s/\@PLT// if (!$elf);
 	    $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
-	    $self->{value} =~ s/\.L/\$L\$/g if (!$gas);
-	    $self->{value} =~ s/\.L/L\$/g   if (!$elf);
+	    $self->{value} =~ s/\.L/$decor/g;
 	}
 	$ret;
     }
     sub out {
 	my $self = shift;
-	if ($nasm && opcode->mnemonic()=~m/^j/) {
+	if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) {
 	    "NEAR ".$self->{value};
 	} else {
 	    $self->{value};
@@ -461,6 +481,11 @@ my %globals;
 				    }
 				    last;
 				  };
+		/\.rva|\.long|\.quad/
+			    && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
+				    $line =~ s/\.L/$decor/g;
+				    last;
+				  };
 	    }
 
 	    if ($gas) {
@@ -477,7 +502,7 @@ my %globals;
 		} elsif (!$elf && $dir =~ /\.size/) {
 		    $self->{value} = "";
 		    if (defined($current_function)) {
-			$self->{value} .= ".LSEH_end_$current_function->{name}:"
+			$self->{value} .= "${decor}SEH_end_$current_function->{name}:"
 				if ($win64 && $current_function->{abi} eq "svr4");
 			undef $current_function;
 		    }
@@ -491,6 +516,12 @@ my %globals;
 		    }
 		} elsif ($dir =~ /\.(text|data)/) {
 		    $current_segment=".$1";
+		} elsif ($dir =~ /\.hidden/) {
+		    if    ($flavour eq "macosx")  { $self->{value} = ".private_extern\t$prefix$line"; }
+		    elsif ($flavour eq "mingw64") { $self->{value} = ""; }
+		} elsif ($dir =~ /\.comm/) {
+		    $self->{value} = "$dir\t$prefix$line";
+		    $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
 		}
 		$line = "";
 		return $self;
@@ -530,6 +561,8 @@ my %globals;
 					if ($line=~/\.([px])data/) {
 					    $v.=" rdata align=";
 					    $v.=$1 eq "p"? 4 : 8;
+					} elsif ($line=~/\.CRT\$/i) {
+					    $v.=" rdata align=8";
 					}
 				    } else {
 					$v="$current_segment\tENDS\n" if ($current_segment);
@@ -537,6 +570,9 @@ my %globals;
 					if ($line=~/\.([px])data/) {
 					    $v.=" READONLY";
 					    $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
+					} elsif ($line=~/\.CRT\$/i) {
+					    $v.=" READONLY ";
+					    $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
 					}
 				    }
 				    $current_segment = $line;
@@ -555,10 +591,10 @@ my %globals;
 		/\.size/    && do { if (defined($current_function)) {
 					undef $self->{value};
 					if ($current_function->{abi} eq "svr4") {
-					    $self->{value}="\$L\$SEH_end_$current_function->{name}:";
+					    $self->{value}="${decor}SEH_end_$current_function->{name}:";
 					    $self->{value}.=":\n" if($masm);
 					}
-					$self->{value}.="$current_function->{name}\tENDP" if($masm);
+					$self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
 					undef $current_function;
 				    }
 				    last;
@@ -566,11 +602,11 @@ my %globals;
 		/\.align/   && do { $self->{value} = "ALIGN\t".$line; last; };
 		/\.(value|long|rva|quad)/
 			    && do { my $sz  = substr($1,0,1);
-				    my @arr = split(',',$line);
+				    my @arr = split(/,\s*/,$line);
 				    my $last = pop(@arr);
 				    my $conv = sub  {	my $var=shift;
-							$var=~s/0x([0-9a-f]+)/0$1h/ig;
-							$var=~s/\.L/\$L\$/g;
+							$var=~s/^(0b[0-1]+)/oct($1)/eig;
+							$var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
 							if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
 							{ $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
 							$var;
@@ -582,7 +618,9 @@ my %globals;
 				    $self->{value} .= &$conv($last);
 				    last;
 				  };
-		/\.byte/    && do { my @str=split(",",$line);
+		/\.byte/    && do { my @str=split(/,\s*/,$line);
+				    map(s/(0b[0-1]+)/oct($1)/eig,@str);
+				    map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);	
 				    while ($#str>15) {
 					$self->{value}.="DB\t"
 						.join(",",@str[0..15])."\n";
@@ -592,6 +630,19 @@ my %globals;
 						.join(",",@str) if (@str);
 				    last;
 				  };
+		/\.comm/    && do { my @str=split(/,\s*/,$line);
+				    my $v=undef;
+				    if ($nasm) {
+					$v.="common	$prefix@str[0] @str[1]";
+				    } else {
+					$v="$current_segment\tENDS\n" if ($current_segment);
+					$current_segment = "_DATA";
+					$v.="$current_segment\tSEGMENT\n";
+					$v.="COMM	@str[0]:DWORD:".@str[1]/4;
+				    }
+				    $self->{value} = $v;
+				    last;
+				  };
 	    }
 	    $line = "";
 	}
@@ -604,9 +655,172 @@ my %globals;
     }
 }
 
+sub rex {
+ local *opcode=shift;
+ my ($dst,$src,$rex)=@_;
+
+   $rex|=0x04 if($dst>=8);
+   $rex|=0x01 if($src>=8);
+   push @opcode,($rex|0x40) if ($rex);
+}
+
+# older gas and ml64 don't handle SSE>2 instructions
+my %regrm = (	"%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
+		"%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7	);
+
+my $movq = sub {	# elderly gas can't handle inter-register movq
+  my $arg = shift;
+  my @opcode=(0x66);
+    if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
+	my ($src,$dst)=($1,$2);
+	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }
+	rex(\@opcode,$src,$dst,0x8);
+	push @opcode,0x0f,0x7e;
+	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
+	@opcode;
+    } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
+	my ($src,$dst)=($2,$1);
+	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }
+	rex(\@opcode,$src,$dst,0x8);
+	push @opcode,0x0f,0x6e;
+	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $pextrd = sub {
+    if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
+      my @opcode=(0x66);
+	$imm=$1;
+	$src=$2;
+	$dst=$3;
+	if ($dst =~ /%r([0-9]+)d/)	{ $dst = $1; }
+	elsif ($dst =~ /%e/)		{ $dst = $regrm{$dst}; }
+	rex(\@opcode,$src,$dst);
+	push @opcode,0x0f,0x3a,0x16;
+	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
+	push @opcode,$imm;
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $pinsrd = sub {
+    if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x66);
+	$imm=$1;
+	$src=$2;
+	$dst=$3;
+	if ($src =~ /%r([0-9]+)/)	{ $src = $1; }
+	elsif ($src =~ /%e/)		{ $src = $regrm{$src}; }
+	rex(\@opcode,$dst,$src);
+	push @opcode,0x0f,0x3a,0x22;
+	push @opcode,0xc0|(($dst&7)<<3)|($src&7);	# ModR/M
+	push @opcode,$imm;
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $pshufb = sub {
+    if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x66);
+	rex(\@opcode,$2,$1);
+	push @opcode,0x0f,0x38,0x00;
+	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $palignr = sub {
+    if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x66);
+	rex(\@opcode,$3,$2);
+	push @opcode,0x0f,0x3a,0x0f;
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
+	push @opcode,$1;
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $pclmulqdq = sub {
+    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x66);
+	rex(\@opcode,$3,$2);
+	push @opcode,0x0f,0x3a,0x44;
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
+	my $c=$1;
+	push @opcode,$c=~/^0/?oct($c):$c;
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $rdrand = sub {
+    if (shift =~ /%[er](\w+)/) {
+      my @opcode=();
+      my $dst=$1;
+	if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
+	rex(\@opcode,0,$1,8);
+	push @opcode,0x0f,0xc7,0xf0|($dst&7);
+	@opcode;
+    } else {
+	();
+    }
+};
+
+sub rxb {
+ local *opcode=shift;
+ my ($dst,$src1,$src2,$rxb)=@_;
+
+   $rxb|=0x7<<5;
+   $rxb&=~(0x04<<5) if($dst>=8);
+   $rxb&=~(0x01<<5) if($src1>=8);
+   $rxb&=~(0x02<<5) if($src2>=8);
+   push @opcode,$rxb;
+}
+
+my $vprotd = sub {
+    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x8f);
+	rxb(\@opcode,$3,$2,-1,0x08);
+	push @opcode,0x78,0xc2;
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
+	my $c=$1;
+	push @opcode,$c=~/^0/?oct($c):$c;
+	@opcode;
+    } else {
+	();
+    }
+};
+
+my $vprotq = sub {
+    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+      my @opcode=(0x8f);
+	rxb(\@opcode,$3,$2,-1,0x08);
+	push @opcode,0x78,0xc3;
+	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
+	my $c=$1;
+	push @opcode,$c=~/^0/?oct($c):$c;
+	@opcode;
+    } else {
+	();
+    }
+};
+
 if ($nasm) {
     print <<___;
 default	rel
+%define XMMWORD
 ___
 } elsif ($masm) {
     print <<___;
@@ -620,47 +834,62 @@ while($line=<>) {
     $line =~ s|[#!].*$||;	# get rid of asm-style comments...
     $line =~ s|/\*.*\*/||;	# ... and C-style comments...
     $line =~ s|^\s+||;		# ... and skip white spaces in beginning
+    $line =~ s|\s+$||;		# ... and at the end
 
     undef $label;
     undef $opcode;
-    undef $dst;
-    undef $src;
-    undef $sz;
+    undef @args;
 
     if ($label=label->re(\$line))	{ print $label->out(); }
 
     if (directive->re(\$line)) {
 	printf "%s",directive->out();
-    } elsif ($opcode=opcode->re(\$line)) { ARGUMENT: {
+    } elsif ($opcode=opcode->re(\$line)) {
+	my $asm = eval("\$".$opcode->mnemonic());
+	undef @bytes;
+	
+	if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) {
+	    print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
+	    next;
+	}
 
-	if ($src=register->re(\$line))	{ opcode->size($src->size()); }
-	elsif ($src=const->re(\$line))	{ }
-	elsif ($src=ea->re(\$line))	{ }
-	elsif ($src=expr->re(\$line))	{ }
+	ARGUMENT: while (1) {
+	my $arg;
 
-	last ARGUMENT if ($line !~ /^,/);
+	if ($arg=register->re(\$line))	{ opcode->size($arg->size()); }
+	elsif ($arg=const->re(\$line))	{ }
+	elsif ($arg=ea->re(\$line))	{ }
+	elsif ($arg=expr->re(\$line))	{ }
+	else				{ last ARGUMENT; }
 
-	$line = substr($line,1); $line =~ s/^\s+//;
+	push @args,$arg;
 
-	if ($dst=register->re(\$line))	{ opcode->size($dst->size()); }
-	elsif ($dst=const->re(\$line))	{ }
-	elsif ($dst=ea->re(\$line))	{ }
+	last ARGUMENT if ($line !~ /^,/);
 
+	$line =~ s/^,\s*//;
 	} # ARGUMENT:
 
-	$sz=opcode->size();
+	if ($#args>=0) {
+	    my $insn;
+	    my $sz=opcode->size();
 
-	if (defined($dst)) {
 	    if ($gas) {
-		printf "\t%s\t%s,%s",	$opcode->out($dst->size()),
-					$src->out($sz),$dst->out($sz);
+		$insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
+		@args = map($_->out($sz),@args);
+		printf "\t%s\t%s",$insn,join(",",@args);
 	    } else {
+		$insn = $opcode->out();
+		foreach (@args) {
+		    my $arg = $_->out();
+		    # $insn.=$sz compensates for movq, pinsrw, ...
+		    if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
+		    if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
+		    if ($arg =~ /^mm[0-9]+$/)  { $insn.=$sz; $sz="q" if(!$sz); last; }
+		}
+		@args = reverse(@args);
 		undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
-		printf "\t%s\t%s,%s",	$opcode->out(),
-					$dst->out($sz),$src->out($sz);
+		printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
 	    }
-	} elsif (defined($src)) {
-	    printf "\t%s\t%s",$opcode->out(),$src->out($sz);
 	} else {
 	    printf "\t%s",$opcode->out();
 	}
@@ -826,6 +1055,7 @@ close STDOUT;
 #	CONTEXT.R14				232
 #	CONTEXT.R15				240
 #	CONTEXT.Rip				248
+#	CONTEXT.Xmm6				512
 #	sizeof(CONTEXT)				1232
 #	DISPATCHER_CONTEXT.ControlPc		0
 #	DISPATCHER_CONTEXT.ImageBase		8
@@ -863,7 +1093,7 @@ close STDOUT;
 #	.rva	.LSEH_end_function
 #	.rva	function_unwind_info
 #
-# Reference to functon_unwind_info from .xdata segment is the anchor.
+# Reference to function_unwind_info from .xdata segment is the anchor.
 # In case you wonder why references are 32-bit .rvas and not 64-bit
 # .quads. References put into these two segments are required to be
 # *relative* to the base address of the current binary module, a.k.a.