nasm/insns.pl

#!/usr/bin/perl
#
# insns.pl   produce insnsa.c, insnsd.c, insnsi.h, insnsn.c from insns.dat
#
# The Netwide Assembler is copyright (C) 1996 Simon Tatham and
# Julian Hall. All rights reserved. The software is
# redistributable under the license given in the file "LICENSE"
# distributed in the NASM archive.

# Opcode prefixes which need their own opcode tables
# LONGER PREFIXES FIRST!
@disasm_prefixes = qw(0F24 0F25 0F38 0F3A 0F7A 0FA6 0FA7 0F);

# This should match MAX_OPERANDS from nasm.h
$MAX_OPERANDS = 5;

print STDERR "Reading insns.dat...\n";

@args   = ();
undef $output;
foreach $arg ( @ARGV ) {
    if ( $arg =~ /^\-/ ) {
	if  ( $arg =~ /^\-([abdin])$/ ) {
	    $output = $1;
	} else {
	    die "$0: Unknown option: ${arg}\n";
	}
    } else {
	push (@args, $arg);
    }
}

$fname = "insns.dat" unless $fname = $args[0];
open (F, $fname) || die "unable to open $fname";

%dinstables = ();
@bytecode_list = ();

$line = 0;
$insns = 0;
while (<F>) {
  $line++;
  chomp;
  next if ( /^\s*(\;.*|)$/ );   # comments or blank lines

  unless (/^\s*(\S+)\s+(\S+)\s+(\S+|\[.*\])\s+(\S+)\s*$/) {
      warn "line $line does not contain four fields\n";
      next;
  }
  @fields = ($1, $2, $3, $4);
  ($formatted, $nd) = format(@fields);
  if ($formatted) {
    $insns++;
    $aname = "aa_$fields[0]";
    push @$aname, $formatted;
  }
  if ( $fields[0] =~ /cc$/ ) {
      # Conditional instruction
      $k_opcodes_cc{$fields[0]}++;
  } else {
      # Unconditional instruction
      $k_opcodes{$fields[0]}++;
  }
  if ($formatted && !$nd) {
    push @big, $formatted;
    my @sseq = startseq($fields[2]);
    foreach $i (@sseq) {
	if (!defined($dinstables{$i})) {
	    $dinstables{$i} = [];
	}
	push(@{$dinstables{$i}}, $#big);
    }
  }
}

close F;

#
# Generate the bytecode array.  At this point, @bytecode_list contains
# the full set of bytecodes.
#

# Sort by descending length
@bytecode_list = sort { scalar(@$b) <=> scalar(@$a) } @bytecode_list;
@bytecode_array = ();
%bytecode_pos = ();
$bytecode_next = 0;
foreach $bl (@bytecode_list) {
    my $h = hexstr(@$bl);
    next if (defined($bytecode_pos{$h}));

    push(@bytecode_array, $bl);
    while ($h ne '') {
	$bytecode_pos{$h} = $bytecode_next;
	$h = substr($h, 2);
	$bytecode_next++;
    }
}
undef @bytecode_list;

@opcodes    = sort keys(%k_opcodes);
@opcodes_cc = sort keys(%k_opcodes_cc);

if ( !defined($output) || $output eq 'b') {
    print STDERR "Writing insnsb.c...\n";

    open B, ">insnsb.c";
    
    print B "/* This file auto-generated from insns.dat by insns.pl" .
        " - don't edit it */\n\n";

    print B "#include \"nasm.h\"\n";
    print B "#include \"insns.h\"\n\n";

    print B "const uint8_t nasm_bytecodes[$bytecode_next] = {\n";

    $p = 0;
    foreach $bl (@bytecode_array) {
	printf B "    /* %5d */ ", $p;
	foreach $d (@$bl) {
	    printf B "%#o,", $d;
	    $p++;
	}
	printf B "\n";
    }
    print B "};\n";

    close B;
}
    
if ( !defined($output) || $output eq 'a' ) {
    print STDERR "Writing insnsa.c...\n";

    open A, ">insnsa.c";

    print A "/* This file auto-generated from insns.dat by insns.pl" .
        " - don't edit it */\n\n";

    print A "#include \"nasm.h\"\n";
    print A "#include \"insns.h\"\n\n";

    foreach $i (@opcodes, @opcodes_cc) {
	print A "static const struct itemplate instrux_${i}[] = {\n";
	$aname = "aa_$i";
	foreach $j (@$aname) {
	    print A "    ", codesubst($j), "\n";
	}
	print A "    ITEMPLATE_END\n};\n\n";
    }
    print A "const struct itemplate * const nasm_instructions[] = {\n";
    foreach $i (@opcodes, @opcodes_cc) {
	print A "    instrux_${i},\n";
    }
    print A "};\n";

    close A;
}

if ( !defined($output) || $output eq 'd' ) {
    print STDERR "Writing insnsd.c...\n";

    open D, ">insnsd.c";

    print D "/* This file auto-generated from insns.dat by insns.pl" .
        " - don't edit it */\n\n";

    print D "#include \"nasm.h\"\n";
    print D "#include \"insns.h\"\n\n";

    print D "static const struct itemplate instrux[] = {\n";
    $n = 0;
    foreach $j (@big) {
	printf D "    /* %4d */ %s\n", $n++, codesubst($j);
    }
    print D "};\n";

    foreach $h (sort(keys(%dinstables))) {
	print D "\nstatic const struct itemplate * const itable_${h}[] = {\n";
	foreach $j (@{$dinstables{$h}}) {
	    print D "    instrux + $j,\n";
	}
	print D "};\n";
    }

    foreach $h (@disasm_prefixes, '') {
	$is_prefix{$h} = 1;
	print D "\n";
	print D "static " unless ($h eq '');
	print D "const struct disasm_index ";
	print D ($h eq '') ? 'itable' : "itable_$h";
	print D "[256] = {\n";
	for ($c = 0; $c < 256; $c++) {
	    $nn = sprintf("%s%02X", $h, $c);
	    if ($is_prefix{$nn}) {
		die "$0: ambiguous decoding of $nn\n"
		    if (defined($dinstables{$nn}));
		printf D "    { itable_%s, -1 },\n", $nn;
	    } elsif (defined($dinstables{$nn})) {
		printf D "    { itable_%s, %u },\n",
		$nn, scalar(@{$dinstables{$nn}});
	    } else {
		printf D "    { NULL, 0 },\n";
	    }
	}
    print D "};\n";
    }

    close D;
}

if ( !defined($output) || $output eq 'i' ) {
    print STDERR "Writing insnsi.h...\n";

    open I, ">insnsi.h";

    print I "/* This file is auto-generated from insns.dat by insns.pl" .
        " - don't edit it */\n\n";
    print I "/* This file in included by nasm.h */\n\n";

    print I "/* Instruction names */\n\n";
    print I "#ifndef NASM_INSNSI_H\n";
    print I "#define NASM_INSNSI_H 1\n\n";
    print I "enum opcode {\n";
    $maxlen = 0;
    foreach $i (@opcodes, @opcodes_cc) {
	print I "\tI_${i},\n";
	$len = length($i);
	$len++ if ( $i =~ /cc$/ );	# Condition codes can be 3 characters long
	$maxlen = $len if ( $len > $maxlen );
    }
    print I "\tI_none = -1\n";
    print I "\n};\n\n";
    print I "#define MAX_INSLEN ", $maxlen, "\n\n";
    print I "#endif /* NASM_INSNSI_H */\n";

    close I;
}

if ( !defined($output) || $output eq 'n' ) {
    print STDERR "Writing insnsn.c...\n";

    open N, ">insnsn.c";

    print N "/* This file is auto-generated from insns.dat by insns.pl" .
        " - don't edit it */\n\n";
    print N "/* This file in included by names.c */\n\n";

    print N "static const char * const insn_names[] = {";
    $first = 1;
    foreach $i (@opcodes) {
	print N "," if ( !$first );
	$first = 0;
	$ilower = $i;
	$ilower =~ tr/A-Z/a-z/;	# Change to lower case (Perl 4 compatible)
	print N "\n\t\"${ilower}\"";
    }
    print N "\n};\n\n";
    print N "/* Conditional instructions */\n";
    print N "static const char *icn[] = {";
    $first = 1;
    foreach $i (@opcodes_cc) {
	print N "," if ( !$first );
	$first = 0;
	$ilower = $i;
	$ilower =~ s/cc$//;		# Skip cc suffix
	$ilower =~ tr/A-Z/a-z/;	# Change to lower case (Perl 4 compatible)
	print N "\n\t\"${ilower}\"";
    }

    print N "\n};\n\n";
    print N "/* and the corresponding opcodes */\n";
    print N "static const enum opcode ico[] = {";
    $first = 1;
    foreach $i (@opcodes_cc) {
	print N "," if ( !$first );
	$first = 0;
	print N "\n\tI_$i";
    }

    print N "\n};\n";

    close N;
}

printf STDERR "Done: %d instructions\n", $insns;

sub format(@) {
    my ($opcode, $operands, $codes, $flags) = @_;
    my $num, $nd = 0;
    my @bytecode;

    return (undef, undef) if $operands eq "ignore";

    # format the operands
    $operands =~ s/:/|colon,/g;
    $operands =~ s/mem(\d+)/mem|bits$1/g;
    $operands =~ s/mem/memory/g;
    $operands =~ s/memory_offs/mem_offs/g;
    $operands =~ s/imm(\d+)/imm|bits$1/g;
    $operands =~ s/imm/immediate/g;
    $operands =~ s/rm(\d+)/rm_gpr|bits$1/g;
    $operands =~ s/(mmx|xmm|ymm)rm/rm_$1/g;
    $operands =~ s/\=([0-9]+)/same_as|$1/g;
    if ($operands eq 'void') {
	@ops = ();
    } else {
	@ops = split(/\,/, $operands);
    }
    $num = scalar(@ops);
    while (scalar(@ops) < $MAX_OPERANDS) {
	push(@ops, '0');
    }
    $operands = join(',', @ops);
    $operands =~ tr/a-z/A-Z/;

    # format the flags
    $flags =~ s/,/|IF_/g;
    $flags =~ s/(\|IF_ND|IF_ND\|)//, $nd = 1 if $flags =~ /IF_ND/;
    $flags = "IF_" . $flags;

    @bytecode = (decodify($codes), 0);
    push(@bytecode_list, [@bytecode]);
    $codes = hexstr(@bytecode);

    ("{I_$opcode, $num, {$operands}, \@\@CODES-$codes\@\@, $flags},", $nd);
}

#
# Look for @@CODES-xxx@@ sequences and replace them with the appropriate
# offset into nasm_bytecodes
#
sub codesubst($) {
    my($s) = @_;
    my $n;

    while ($s =~ /\@\@CODES-([0-9A-F]+)\@\@/) {
	my $pos = $bytecode_pos{$1};
	if (!defined($pos)) {
	    die "$0: no position assigned to byte code $1\n";
	}
	$s = $` . "nasm_bytecodes+${pos}" . "$'";
    }
    return $s;
}

sub addprefix ($@) {
    my ($prefix, @list) = @_;
    my $x;
    my @l = ();

    foreach $x (@list) {
	push(@l, sprintf("%s%02X", $prefix, $x));
    }

    return @l;
}

#
# Turn a code string into a sequence of bytes
#
sub decodify($) {
  # Although these are C-syntax strings, by convention they should have
  # only octal escapes (for directives) and hexadecimal escapes
  # (for verbatim bytes)
    my($codestr) = @_;
    my $c = $codestr;
    my @codes = ();

    if ($codestr =~ /^\s*\[([^\]]*)\]\s*$/) {
	return byte_code_compile($1);
    }

    while ($c ne '') {
	if ($c =~ /^\\x([0-9a-f]+)(.*)$/i) {
	    push(@codes, hex $1);
	    $c = $2;
	    next;
	} elsif ($c =~ /^\\([0-7]{1,3})(.*)$/) {
	    push(@codes, oct $1);
	    $c = $2;
	    next;
	} else {
	    die "$0: unknown code format in \"$codestr\"\n";
	}
    }

    return @codes;
}

# Turn a numeric list into a hex string
sub hexstr(@) {
    my $s = '';
    my $c;

    foreach $c (@_) {
	$s .= sprintf("%02X", $c);
    }
    return $s;
}

# Here we determine the range of possible starting bytes for a given
# instruction. We need only consider the codes:
# \1 \2 \3     mean literal bytes, of course
# \4 \5 \6 \7  mean PUSH/POP of segment registers: special case
# \1[0123]     mean byte plus register value
# \330         means byte plus condition code
# \0 or \340   mean give up and return empty set
sub startseq($) {
  my ($codestr) = @_;
  my $word, @range;
  my @codes = ();
  my $c = $codestr;
  my $c0, $c1, $i;
  my $prefix = '';

  @codes = decodify($codestr);

  while ($c0 = shift(@codes)) {
      $c1 = $codes[0];
      if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
	  # Fixed byte string
	  my $fbs = $prefix;
	  while (1) {
	      if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
		  while ($c0--) {
		      $fbs .= sprintf("%02X", shift(@codes));
		  }
	      } else {
		  last;
	      }
	      $c0 = shift(@codes);
	  }

	  foreach $pfx (@disasm_prefixes) {
	      if (substr($fbs, 0, length($pfx)) eq $pfx) {
		  $prefix = $pfx;
		  $fbs = substr($fbs, length($pfx));
		  last;
	      }
	  }

	  if ($fbs ne '') {
	      return ($prefix.substr($fbs,0,2));
	  }

	  unshift(@codes, $c0);
      } elsif ($c0 == 04) {
	  return addprefix($prefix, 0x07, 0x17, 0x1F);
      } elsif ($c0 == 05) {
	  return addprefix($prefix, 0xA1, 0xA9);
      } elsif ($c0 == 06) {
	  return addprefix($prefix, 0x06, 0x0E, 0x16, 0x1E);
      } elsif ($c0 == 07) {
	  return addprefix($prefix, 0xA0, 0xA8);
      } elsif ($c0 >= 010 && $c0 <= 013) {
	  return addprefix($prefix, $c1..($c1+7));
      } elsif (($c0 & ~013) == 0144) {
	  return addprefix($prefix, $c1, $c1|2);
      } elsif ($c0 == 0330) {
	  return addprefix($prefix, $c1..($c1+15));
      } elsif ($c0 == 0 || $c0 == 0340) {
	  return $prefix;
      } elsif (($c0 & ~3) == 0260 || $c0 == 270) {
	  shift(@codes);
	  shift(@codes);
      } elsif ($c0 == 0172) {
	  shift(@codes);
      } else {
	  # We really need to be able to distinguish "forbidden"
	  # and "ignorable" codes here
      }
  }
  return $prefix;
}

#
# This function takes a series of byte codes in a format which is more
# typical of the Intel documentation, and encode it.
#
# The format looks like:
#
# [operands: opcodes]
#
# The operands word lists the order of the operands:
#
# r = register field in the modr/m
# m = modr/m
# v = VEX "v" field or DREX "src" field
# i = immediate
# z = register field of is4 or imz2 field
# 
sub byte_code_compile($) {
    my($str) = @_;
    my $opr;
    my $opc;	
    my @codes = ();
    my $litix = undef;
    my %oppos = ();
    my $i;
    my $op, $oq;
    
    if ($str =~ /^(\S*)\:\s*(.*\S)\s*$/) {
	$opr = "\L$1";
	$opc = "\L$2";
    } else {
	$opr = '';
	$opc = $str;
    }
    
    for ($i = 0; $i < length($opr); $i++) {
	$oppos{substr($opr,$i,1)} = $i;
    }

    foreach $op (split($opc)) {
	if ($op =~ /^[0-9a-f]{2}$/) {
	    if (defined($litix) && $litix+$codes[$litix]+1 == scalar @codes) {
		$codes[$litix]++;
		push(@codes, hex $op);
	    } else {
		$litix = scalar(@codes);
		push(@codes, 01, hex $op);
	    }
	} elsif ($op eq '/r') {
	    if (!defined($oppos{'r'}) || !defined($oppos{'m'})) {
		die "$0: $line: $op requires r and m operands\n";
	    }
	    push(@codes, 0100 + ($oppos{'m'} << 3) + $oppos{'r'});
	} elsif ($op =~ m:^/([0-7])$:) {
	    if (!defined($oppos{'m'})) {
		die "$0: $line: $op requires m operand\n";
	    }
	    push(@codes, 0200 + ($oppos{'m'} << 3) + $1;
	} elsif ($op =~ /^vex\./) {
	    my ($m,$w,$l,$p) = (undef,2,undef,0);
	    foreach $oq (split(/\./, $op)) {
		if ($oq eq 'vex') {
		    # prefix
		} elsif ($oq eq '128' || $oq eq 'l0') {
		    $l = 0;
		} elsif ($oq eq '256' || $oq eq 'l1') {
		    $l = 1;
		} elsif ($oq eq 'w0') {
		    $w = 0;
		} elsif ($oq eq 'w1') {
		    $w = 1;
		} elsif ($oq eq '66') {
		    $p = 1;
		} elsif ($oq eq 'f3') {
		    $p = 2;
		} elsif ($oq eq 'f2') {
		    $p = 3;
		} elsif ($oq eq '0f') {
		    $m = 1;
		} elsif ($oq eq '0f38') {
		    $m = 2;
		} elsif ($oq eq '0f3a') {
		    $m = 3;
		} elsif ($oq =~ /^m([0-9]+)$/) {
		    $m = $1+0;
		} elsif ($oq eq 'nds' || $oq eq 'ndd') {
		    return undef if (!defined($oppos{'v'}));
		} else {
		    die "$0: $line: undefined VEX subcode: $oq\n";
		}
	    }
	    if (!defined($m) || !defined($w) || !defined($l) || !defined($p)) {
		die "$0: $line: missing fields in VEX specification\n";
	    }
	    push(@codes, defined($oppos{'v'}) ? 0260+$oppos{'v'} : 0270,
		 $m, ($w << 3)+($l << 2)+$p);
	}