insns.pl

   1 #!/usr/bin/perl
   2 #
   3 # insns.pl   produce insnsa.c, insnsd.c, insnsi.h, insnsn.c from insns.dat
   4 #
   5 # The Netwide Assembler is copyright (C) 1996 Simon Tatham and
   6 # Julian Hall. All rights reserved. The software is
   7 # redistributable under the license given in the file "LICENSE"
   8 # distributed in the NASM archive.
   9
  10 # Opcode prefixes which need their own opcode tables
  11 # LONGER PREFIXES FIRST!
  12 @disasm_prefixes = qw(0F24 0F25 0F38 0F3A 0F7A 0FA6 0FA7 0F);
  13
  14 # This should match MAX_OPERANDS from nasm.h
  15 $MAX_OPERANDS = 5;
  16
  17 # Add VEX/XOP prefixes
  18 @vex_class = ( 'vex', 'xop' );
  19 $vex_classes = scalar(@vex_class);
  20 @vexlist = ();
  21 %vexmap = ();
  22 for ($c = 0; $c < $vex_classes; $c++) {
  23     $vexmap{$vex_class[$c]} = $c;
  24     for ($m = 0; $m < 32; $m++) {
  25         for ($lp = 0; $lp < 8; $lp++) {
  26             push(@vexlist, sprintf("%s%02X%01X", $vex_class[$c], $m, $lp));
  27         }
  28     }
  29 }
  30 @disasm_prefixes = (@vexlist, @disasm_prefixes);
  31
  32 @bytecode_count = (0) x 256;
  33
  34 print STDERR "Reading insns.dat...\n";
  35
  36 @args   = ();
  37 undef $output;
  38 foreach $arg ( @ARGV ) {
  39     if ( $arg =~ /^\-/ ) {
  40         if  ( $arg =~ /^\-([abdin])$/ ) {
  41             $output = $1;
  42         } else {
  43             die "$0: Unknown option: ${arg}\n";
  44         }
  45     } else {
  46         push (@args, $arg);
  47     }
  48 }
  49
  50 $fname = "insns.dat" unless $fname = $args[0];
  51 open (F, $fname) || die "unable to open $fname";
  52
  53 %dinstables = ();
  54 @bytecode_list = ();
  55
  56 $line = 0;
  57 $insns = 0;
  58 while (<F>) {
  59     $line++;
  60     chomp;
  61     next if ( /^\s*(\;.*|)$/ );   # comments or blank lines
  62
  63     unless (/^\s*(\S+)\s+(\S+)\s+(\S+|\[.*\])\s+(\S+)\s*$/) {
  64         warn "line $line does not contain four fields\n";
  65         next;
  66     }
  67     @fields = ($1, $2, $3, $4);
  68     @field_list = ([@fields, 0]);
  69
  70     if ($fields[1] =~ /\*/) {
  71         # This instruction has relaxed form(s)
  72         if ($fields[2] !~ /^\[/) {
  73             warn "line $line has an * operand but uses raw bytecodes\n";
  74             next;
  75         }
  76
  77         $opmask = 0;
  78         @ops = split(/,/, $fields[1]);
  79         for ($oi = 0; $oi < scalar @ops; $oi++) {
  80             if ($ops[$oi] =~ /\*$/) {
  81                 if ($oi == 0) {
  82                     warn "line $line has a first operand with a *\n";
  83                     next;
  84                 }
  85                 $opmask |= 1 << $oi;
  86             }
  87         }
  88
  89         for ($oi = 1; $oi < (1 << scalar @ops); $oi++) {
  90             if (($oi & ~$opmask) == 0) {
  91                 my @xops = ();
  92                 my $omask = ~$oi;
  93                 for ($oj = 0; $oj < scalar(@ops); $oj++) {
  94                     if ($omask & 1) {
  95                         push(@xops, $ops[$oj]);
  96                     }
  97                     $omask >>= 1;
  98                 }
  99                 push(@field_list, [$fields[0], join(',', @xops),
 100                                    $fields[2], $fields[3], $oi]);
 101             }
 102         }
 103     }
 104
 105     foreach $fptr (@field_list) {
 106         @fields = @$fptr;
 107         ($formatted, $nd) = format_insn(@fields);
 108         if ($formatted) {
 109             $insns++;
 110             $aname = "aa_$fields[0]";
 111             push @$aname, $formatted;
 112         }
 113         if ( $fields[0] =~ /cc$/ ) {
 114             # Conditional instruction
 115             $k_opcodes_cc{$fields[0]}++;
 116         } else {
 117             # Unconditional instruction
 118             $k_opcodes{$fields[0]}++;
 119         }
 120         if ($formatted && !$nd) {
 121             push @big, $formatted;
 122             my @sseq = startseq($fields[2], $fields[4]);
 123             foreach $i (@sseq) {
 124                 if (!defined($dinstables{$i})) {
 125                     $dinstables{$i} = [];
 126                 }
 127                 push(@{$dinstables{$i}}, $#big);
 128             }
 129         }
 130     }
 131 }
 132
 133 close F;
 134
 135 #
 136 # Generate the bytecode array.  At this point, @bytecode_list contains
 137 # the full set of bytecodes.
 138 #
 139
 140 # Sort by descending length
 141 @bytecode_list = sort { scalar(@$b) <=> scalar(@$a) } @bytecode_list;
 142 @bytecode_array = ();
 143 %bytecode_pos = ();
 144 $bytecode_next = 0;
 145 foreach $bl (@bytecode_list) {
 146     my $h = hexstr(@$bl);
 147     next if (defined($bytecode_pos{$h}));
 148
 149     push(@bytecode_array, $bl);
 150     while ($h ne '') {
 151         $bytecode_pos{$h} = $bytecode_next;
 152         $h = substr($h, 2);
 153         $bytecode_next++;
 154     }
 155 }
 156 undef @bytecode_list;
 157
 158 @opcodes    = sort keys(%k_opcodes);
 159 @opcodes_cc = sort keys(%k_opcodes_cc);
 160
 161 if ( !defined($output) || $output eq 'b') {
 162     print STDERR "Writing insnsb.c...\n";
 163
 164     open B, ">insnsb.c";
 165
 166     print B "/* This file auto-generated from insns.dat by insns.pl" .
 167         " - don't edit it */\n\n";
 168
 169     print B "#include \"nasm.h\"\n";
 170     print B "#include \"insns.h\"\n\n";
 171
 172     print B "const uint8_t nasm_bytecodes[$bytecode_next] = {\n";
 173
 174     $p = 0;
 175     foreach $bl (@bytecode_array) {
 176         printf B "    /* %5d */ ", $p;
 177         foreach $d (@$bl) {
 178             printf B "%#o,", $d;
 179             $p++;
 180         }
 181         printf B "\n";
 182     }
 183     print B "};\n";
 184
 185     print B "\n";
 186     print B "/*\n";
 187     print B " * Bytecode frequencies (including reuse):\n";
 188     print B " *\n";
 189     for ($i = 0; $i < 32; $i++) {
 190         print B " *";
 191         for ($j = 0; $j < 256; $j += 32) {
 192             print B " |" if ($j);
 193             printf B " %3o:%4d", $i+$j, $bytecode_count[$i+$j];
 194         }
 195         print B "\n";
 196     }
 197     print B " */\n";
 198
 199     close B;
 200 }
 201
 202 if ( !defined($output) || $output eq 'a' ) {
 203     print STDERR "Writing insnsa.c...\n";
 204
 205     open A, ">insnsa.c";
 206
 207     print A "/* This file auto-generated from insns.dat by insns.pl" .
 208         " - don't edit it */\n\n";
 209
 210     print A "#include \"nasm.h\"\n";
 211     print A "#include \"insns.h\"\n\n";
 212
 213     foreach $i (@opcodes, @opcodes_cc) {
 214         print A "static const struct itemplate instrux_${i}[] = {\n";
 215         $aname = "aa_$i";
 216         foreach $j (@$aname) {
 217             print A "    ", codesubst($j), "\n";
 218         }
 219         print A "    ITEMPLATE_END\n};\n\n";
 220     }
 221     print A "const struct itemplate * const nasm_instructions[] = {\n";
 222     foreach $i (@opcodes, @opcodes_cc) {
 223         print A "    instrux_${i},\n";
 224     }
 225     print A "};\n";
 226
 227     close A;
 228 }
 229
 230 if ( !defined($output) || $output eq 'd' ) {
 231     print STDERR "Writing insnsd.c...\n";
 232
 233     open D, ">insnsd.c";
 234
 235     print D "/* This file auto-generated from insns.dat by insns.pl" .
 236         " - don't edit it */\n\n";
 237
 238     print D "#include \"nasm.h\"\n";
 239     print D "#include \"insns.h\"\n\n";
 240
 241     print D "static const struct itemplate instrux[] = {\n";
 242     $n = 0;
 243     foreach $j (@big) {
 244         printf D "    /* %4d */ %s\n", $n++, codesubst($j);
 245     }
 246     print D "};\n";
 247
 248     foreach $h (sort(keys(%dinstables))) {
 249         next if ($h eq '');     # Skip pseudo-instructions
 250         print D "\nstatic const struct itemplate * const itable_${h}[] = {\n";
 251         foreach $j (@{$dinstables{$h}}) {
 252             print D "    instrux + $j,\n";
 253         }
 254         print D "};\n";
 255     }
 256
 257     @prefix_list = ();
 258     foreach $h (@disasm_prefixes, '') {
 259         for ($c = 0; $c < 256; $c++) {
 260             $nn = sprintf("%s%02X", $h, $c);
 261             if ($is_prefix{$nn} || defined($dinstables{$nn})) {
 262                 # At least one entry in this prefix table
 263                 push(@prefix_list, $h);
 264                 $is_prefix{$h} = 1;
 265                 last;
 266             }
 267         }
 268     }
 269
 270     foreach $h (@prefix_list) {
 271         print D "\n";
 272         print D "static " unless ($h eq '');
 273         print D "const struct disasm_index ";
 274         print D ($h eq '') ? 'itable' : "itable_$h";
 275         print D "[256] = {\n";
 276         for ($c = 0; $c < 256; $c++) {
 277             $nn = sprintf("%s%02X", $h, $c);
 278             if ($is_prefix{$nn}) {
 279                 die "$fname: ambiguous decoding of $nn\n"
 280                     if (defined($dinstables{$nn}));
 281                 printf D "    { itable_%s, -1 },\n", $nn;
 282             } elsif (defined($dinstables{$nn})) {
 283                 printf D "    { itable_%s, %u },\n",
 284                 $nn, scalar(@{$dinstables{$nn}});
 285             } else {
 286                 printf D "    { NULL, 0 },\n";
 287             }
 288         }
 289         print D "};\n";
 290     }
 291
 292     printf D "\nconst struct disasm_index * const itable_vex[%d][32][8] =\n",
 293         $vex_classes;
 294     print D "{\n";
 295     for ($c = 0; $c < $vex_classes; $c++) {
 296         print D "    {\n";
 297         for ($m = 0; $m < 32; $m++) {
 298             print D "        {\n";
 299             for ($lp = 0; $lp < 8; $lp++) {
 300                 $vp = sprintf("%s%02X%01X", $vex_class[$c], $m, $lp);
 301                 if ($is_prefix{$vp}) {
 302                     printf D "            itable_%s,\n", $vp;
 303                 } else {
 304                     print  D "            NULL,\n";
 305                 }
 306             }
 307             print D "        },\n";
 308         }
 309         print D "    },\n";
 310     }
 311     print D "};\n";
 312
 313     close D;
 314 }
 315
 316 if ( !defined($output) || $output eq 'i' ) {
 317     print STDERR "Writing insnsi.h...\n";
 318
 319     open I, ">insnsi.h";
 320
 321     print I "/* This file is auto-generated from insns.dat by insns.pl" .
 322         " - don't edit it */\n\n";
 323     print I "/* This file in included by nasm.h */\n\n";
 324
 325     print I "/* Instruction names */\n\n";
 326     print I "#ifndef NASM_INSNSI_H\n";
 327     print I "#define NASM_INSNSI_H 1\n\n";
 328     print I "enum opcode {\n";
 329     $maxlen = 0;
 330     foreach $i (@opcodes, @opcodes_cc) {
 331         print I "\tI_${i},\n";
 332         $len = length($i);
 333         $len++ if ( $i =~ /cc$/ ); # Condition codes can be 3 characters long
 334         $maxlen = $len if ( $len > $maxlen );
 335     }
 336     print I "\tI_none = -1\n";
 337     print I "\n};\n\n";
 338     print I "#define MAX_INSLEN ", $maxlen, "\n";
 339     print I "#define FIRST_COND_OPCODE I_", $opcodes_cc[0], "\n\n";
 340     print I "#endif /* NASM_INSNSI_H */\n";
 341
 342     close I;
 343 }
 344
 345 if ( !defined($output) || $output eq 'n' ) {
 346     print STDERR "Writing insnsn.c...\n";
 347
 348     open N, ">insnsn.c";
 349
 350     print N "/* This file is auto-generated from insns.dat by insns.pl" .
 351         " - don't edit it */\n\n";
 352     print N "#include \"tables.h\"\n\n";
 353
 354     print N "const char * const nasm_insn_names[] = {";
 355     $first = 1;
 356     foreach $i (@opcodes, @opcodes_cc) {
 357         print N "," if ( !$first );
 358         $first = 0;
 359         $ilower = $i;
 360         $ilower =~ s/cc$//;     # Remove conditional cc suffix
 361         $ilower =~ tr/A-Z/a-z/; # Change to lower case (Perl 4 compatible)
 362         print N "\n\t\"${ilower}\"";
 363     }
 364     print N "\n};\n";
 365     close N;
 366 }
 367
 368 printf STDERR "Done: %d instructions\n", $insns;
 369
 370 # Count primary bytecodes, for statistics
 371 sub count_bytecodes(@) {
 372     my $skip = 0;
 373     foreach my $bc (@_) {
 374         if ($skip) {
 375             $skip--;
 376             next;
 377         }
 378         $bytecode_count[$bc]++;
 379         if ($bc >= 01 && $bc <= 04) {
 380             $skip = $bc;
 381         } elsif (($bc & ~03) == 010) {
 382             $skip = 1;
 383         } elsif (($bc & ~013) == 0144) {
 384             $skip = 1;
 385         } elsif ($bc == 0172) {
 386             $skip = 1;
 387         } elsif ($bc >= 0260 && $bc <= 0270) {
 388             $skip = 2;
 389         } elsif ($bc == 0330) {
 390             $skip = 1;
 391         }
 392     }
 393 }
 394
 395 sub format_insn($$$$$) {
 396     my ($opcode, $operands, $codes, $flags, $relax) = @_;
 397     my $num, $nd = 0;
 398     my @bytecode;
 399
 400     return (undef, undef) if $operands eq "ignore";
 401
 402     # format the operands
 403     $operands =~ s/\*//g;
 404     $operands =~ s/:/|colon,/g;
 405     $operands =~ s/mem(\d+)/mem|bits$1/g;
 406     $operands =~ s/mem/memory/g;
 407     $operands =~ s/memory_offs/mem_offs/g;
 408     $operands =~ s/imm(\d+)/imm|bits$1/g;
 409     $operands =~ s/imm/immediate/g;
 410     $operands =~ s/rm(\d+)/rm_gpr|bits$1/g;
 411     $operands =~ s/(mmx|xmm|ymm)rm/rm_$1/g;
 412     $operands =~ s/\=([0-9]+)/same_as|$1/g;
 413     if ($operands eq 'void') {
 414         @ops = ();
 415     } else {
 416         @ops = split(/\,/, $operands);
 417     }
 418     $num = scalar(@ops);
 419     while (scalar(@ops) < $MAX_OPERANDS) {
 420         push(@ops, '0');
 421     }
 422     $operands = join(',', @ops);
 423     $operands =~ tr/a-z/A-Z/;
 424
 425     # format the flags
 426     $flags =~ s/,/|IF_/g;
 427     $flags =~ s/(\|IF_ND|IF_ND\|)//, $nd = 1 if $flags =~ /IF_ND/;
 428     $flags = "IF_" . $flags;
 429
 430     @bytecode = (decodify($codes, $relax), 0);
 431     push(@bytecode_list, [@bytecode]);
 432     $codes = hexstr(@bytecode);
 433     count_bytecodes(@bytecode);
 434
 435     ("{I_$opcode, $num, {$operands}, \@\@CODES-$codes\@\@, $flags},", $nd);
 436 }
 437
 438 #
 439 # Look for @@CODES-xxx@@ sequences and replace them with the appropriate
 440 # offset into nasm_bytecodes
 441 #
 442 sub codesubst($) {
 443     my($s) = @_;
 444     my $n;
 445
 446     while ($s =~ /\@\@CODES-([0-9A-F]+)\@\@/) {
 447         my $pos = $bytecode_pos{$1};
 448         if (!defined($pos)) {
 449             die "$fname: no position assigned to byte code $1\n";
 450         }
 451         $s = $` . "nasm_bytecodes+${pos}" . "$'";
 452     }
 453     return $s;
 454 }
 455
 456 sub addprefix ($@) {
 457     my ($prefix, @list) = @_;
 458     my $x;
 459     my @l = ();
 460
 461     foreach $x (@list) {
 462         push(@l, sprintf("%s%02X", $prefix, $x));
 463     }
 464
 465     return @l;
 466 }
 467
 468 #
 469 # Turn a code string into a sequence of bytes
 470 #
 471 sub decodify($$) {
 472   # Although these are C-syntax strings, by convention they should have
 473   # only octal escapes (for directives) and hexadecimal escapes
 474   # (for verbatim bytes)
 475     my($codestr, $relax) = @_;
 476
 477     if ($codestr =~ /^\s*\[([^\]]*)\]\s*$/) {
 478         return byte_code_compile($1, $relax);
 479     }
 480
 481     my $c = $codestr;
 482     my @codes = ();
 483
 484     while ($c ne '') {
 485         if ($c =~ /^\\x([0-9a-f]+)(.*)$/i) {
 486             push(@codes, hex $1);
 487             $c = $2;
 488             next;
 489         } elsif ($c =~ /^\\([0-7]{1,3})(.*)$/) {
 490             push(@codes, oct $1);
 491             $c = $2;
 492             next;
 493         } else {
 494             die "$fname: unknown code format in \"$codestr\"\n";
 495         }
 496     }
 497
 498     return @codes;
 499 }
 500
 501 # Turn a numeric list into a hex string
 502 sub hexstr(@) {
 503     my $s = '';
 504     my $c;
 505
 506     foreach $c (@_) {
 507         $s .= sprintf("%02X", $c);
 508     }
 509     return $s;
 510 }
 511
 512 # Here we determine the range of possible starting bytes for a given
 513 # instruction. We need only consider the codes:
 514 # \[1234]      mean literal bytes, of course
 515 # \1[0123]     mean byte plus register value
 516 # \330         means byte plus condition code
 517 # \0 or \340   mean give up and return empty set
 518 # \34[4567]    mean PUSH/POP of segment registers: special case
 519 # \17[234]     skip is4 control byte
 520 # \26x \270    skip VEX control bytes
 521 sub startseq($$) {
 522     my ($codestr, $relax) = @_;
 523     my $word, @range;
 524     my @codes = ();
 525     my $c = $codestr;
 526     my $c0, $c1, $i;
 527     my $prefix = '';
 528
 529     @codes = decodify($codestr, $relax);
 530
 531     while ($c0 = shift(@codes)) {
 532         $c1 = $codes[0];
 533         if ($c0 >= 01 && $c0 <= 04) {
 534             # Fixed byte string
 535             my $fbs = $prefix;
 536             while (1) {
 537                 if ($c0 >= 01 && $c0 <= 04) {
 538                     while ($c0--) {
 539                         $fbs .= sprintf("%02X", shift(@codes));
 540                     }
 541                 } else {
 542                     last;
 543                 }
 544                 $c0 = shift(@codes);
 545             }
 546
 547             foreach $pfx (@disasm_prefixes) {
 548                 if (substr($fbs, 0, length($pfx)) eq $pfx) {
 549                     $prefix = $pfx;
 550                     $fbs = substr($fbs, length($pfx));
 551                     last;
 552                 }
 553             }
 554
 555             if ($fbs ne '') {
 556                 return ($prefix.substr($fbs,0,2));
 557             }
 558
 559             unshift(@codes, $c0);
 560         } elsif ($c0 >= 010 && $c0 <= 013) {
 561             return addprefix($prefix, $c1..($c1+7));
 562         } elsif (($c0 & ~013) == 0144) {
 563             return addprefix($prefix, $c1, $c1|2);
 564         } elsif ($c0 == 0330) {
 565             return addprefix($prefix, $c1..($c1+15));
 566         } elsif ($c0 == 0 || $c0 == 0340) {
 567             return $prefix;
 568         } elsif ($c0 == 0344) {
 569             return addprefix($prefix, 0x06, 0x0E, 0x16, 0x1E);
 570         } elsif ($c0 == 0345) {
 571             return addprefix($prefix, 0x07, 0x17, 0x1F);
 572         } elsif ($c0 == 0346) {
 573             return addprefix($prefix, 0xA0, 0xA8);
 574         } elsif ($c0 == 0347) {
 575             return addprefix($prefix, 0xA1, 0xA9);
 576         } elsif (($c0 & ~3) == 0260 || $c0 == 0270) {
 577             my $c,$m,$wlp;
 578             $m   = shift(@codes);
 579             $wlp = shift(@codes);
 580             $c = ($m >> 6);
 581             $m = $m & 31;
 582             $prefix .= sprintf('%s%02X%01X', $vex_class[$c], $m, $wlp & 7);
 583         } elsif ($c0 >= 0172 && $c0 <= 174) {
 584             shift(@codes);      # Skip is4 control byte
 585         } else {
 586             # We really need to be able to distinguish "forbidden"
 587             # and "ignorable" codes here
 588         }
 589     }
 590     return $prefix;
 591 }
 592
 593 #
 594 # This function takes a series of byte codes in a format which is more
 595 # typical of the Intel documentation, and encode it.
 596 #
 597 # The format looks like:
 598 #
 599 # [operands: opcodes]
 600 #
 601 # The operands word lists the order of the operands:
 602 #
 603 # r = register field in the modr/m
 604 # m = modr/m
 605 # v = VEX "v" field
 606 # d = DREX "dst" field
 607 # i = immediate
 608 # s = register field of is4/imz2 field
 609 # - = implicit (unencoded) operand
 610 #
 611 # For an operand that should be filled into more than one field,
 612 # enter it as e.g. "r+v".
 613 #
 614 sub byte_code_compile($$) {
 615     my($str, $relax) = @_;
 616     my $opr;
 617     my $opc;
 618     my @codes = ();
 619     my $litix = undef;
 620     my %oppos = ();
 621     my $i;
 622     my $op, $oq;
 623     my $opex;
 624
 625     unless ($str =~ /^(([^\s:]*)\:|)\s*(.*\S)\s*$/) {
 626         die "$fname: $line: cannot parse: [$str]\n";
 627     }
 628     $opr = "\L$2";
 629     $opc = "\L$3";
 630
 631     my $op = 0;
 632     for ($i = 0; $i < length($opr); $i++) {
 633         my $c = substr($opr,$i,1);
 634         if ($c eq '+') {
 635             $op--;
 636         } else {
 637             if ($relax & 1) {
 638                 $op--;
 639             }
 640             $relax >>= 1;
 641             $oppos{$c} = $op++;
 642         }
 643     }
 644
 645     $prefix_ok = 1;
 646     foreach $op (split(/\s*(?:\s|(?=[\/\\]))/, $opc)) {
 647         if ($op eq 'o16') {
 648             push(@codes, 0320);
 649         } elsif ($op eq 'o32') {
 650             push(@codes, 0321);
 651         } elsif ($op eq 'o64') {  # 64-bit operand size requiring REX.W
 652             push(@codes, 0324);
 653         } elsif ($op eq 'o64nw') { # Implied 64-bit operand size (no REX.W)
 654             push(@codes, 0323);
 655         } elsif ($op eq 'a16') {
 656             push(@codes, 0310);
 657         } elsif ($op eq 'a32') {
 658             push(@codes, 0311);
 659         } elsif ($op eq 'a64') {
 660             push(@codes, 0313);
 661         } elsif ($op eq '!osp') {
 662             push(@codes, 0364);
 663         } elsif ($op eq '!asp') {
 664             push(@codes, 0365);
 665         } elsif ($op eq 'rex.l') {
 666             push(@codes, 0334);
 667         } elsif ($op eq 'repe') {
 668             push(@codes, 0335);
 669         } elsif ($op eq 'nohi') { # Use spl/bpl/sil/dil even without REX
 670             push(@codes, 0325);
 671         } elsif ($prefix_ok && $op =~ /^(66|f2|f3|np)$/) {
 672             # 66/F2/F3 prefix used as an opcode extension, or np = no prefix
 673             if ($op eq '66') {
 674                 push(@codes, 0361);
 675             } elsif ($op eq 'f2') {
 676                 push(@codes, 0362);
 677             } elsif ($op eq 'f3') {
 678                 push(@codes, 0363);
 679             } else {
 680                 push(@codes, 0360);
 681             }
 682         } elsif ($op =~ /^[0-9a-f]{2}$/) {
 683             if (defined($litix) && $litix+$codes[$litix]+1 == scalar @codes &&
 684                 $codes[$litix] < 4) {
 685                 $codes[$litix]++;
 686                 push(@codes, hex $op);
 687             } else {
 688                 $litix = scalar(@codes);
 689                 push(@codes, 01, hex $op);
 690             }
 691             $prefix_ok = 0;
 692         } elsif ($op eq '/r') {
 693             if (!defined($oppos{'r'}) || !defined($oppos{'m'})) {
 694                 die "$fname: $line: $op requires r and m operands\n";
 695             }
 696             $opex = (($oppos{'m'} & 4) ? 06 : 0) |
 697                     (($oppos{'r'} & 4) ? 05 : 0);
 698             push(@codes, $opex) if ($opex);
 699             push(@codes, 0100 + (($oppos{'m'} & 3) << 3) + ($oppos{'r'} & 3));
 700             $prefix_ok = 0;
 701         } elsif ($op =~ m:^/([0-7])$:) {
 702             if (!defined($oppos{'m'})) {
 703                 die "$fname: $line: $op requires m operand\n";
 704             }
 705             push(@codes, 06) if ($oppos{'m'} & 4);
 706             push(@codes, 0200 + (($oppos{'m'} & 3) << 3) + $1);
 707             $prefix_ok = 0;
 708         } elsif ($op =~ /^(vex|xop)(|\..*)$/) {
 709             my $c = $vexmap{$1};
 710             my ($m,$w,$l,$p) = (undef,2,undef,0);
 711             my $has_nds = 0;
 712             my @subops = split(/\./, $op);
 713             shift @subops;      # Drop prefix
 714             foreach $oq (@subops) {
 715                 if ($oq eq '128' || $oq eq 'l0') {
 716                     $l = 0;
 717                 } elsif ($oq eq '256' || $oq eq 'l1') {
 718                     $l = 1;
 719                 } elsif ($oq eq 'w0') {
 720                     $w = 0;
 721                 } elsif ($oq eq 'w1') {
 722                     $w = 1;
 723                 } elsif ($oq eq 'wx') {
 724                     $w = 2;
 725                 } elsif ($oq eq 'ww') {
 726                     $w = 3;
 727                 } elsif ($oq eq 'p0') {
 728                     $p = 0;
 729                 } elsif ($oq eq '66' || $oq eq 'p1') {
 730                     $p = 1;
 731                 } elsif ($oq eq 'f3' || $oq eq 'p2') {
 732                     $p = 2;
 733                 } elsif ($oq eq 'f2' || $oq eq 'p3') {
 734                     $p = 3;
 735                 } elsif ($oq eq '0f') {
 736                     $m = 1;
 737                 } elsif ($oq eq '0f38') {
 738                     $m = 2;
 739                 } elsif ($oq eq '0f3a') {
 740                     $m = 3;
 741                 } elsif ($oq =~ /^m([0-9]+)$/) {
 742                     $m = $1+0;
 743                 } elsif ($oq eq 'nds' || $oq eq 'ndd' || $oq eq 'dds') {
 744                     if (!defined($oppos{'v'})) {
 745                         die "$fname: $line: vex.$oq without 'v' operand\n";
 746                     }
 747                     $has_nds = 1;
 748                 } else {
 749                     die "$fname: $line: undefined VEX subcode: $oq\n";
 750                 }
 751             }
 752             if (!defined($m) || !defined($w) || !defined($l) || !defined($p)) {
 753                 die "$fname: $line: missing fields in VEX specification\n";
 754             }
 755             if (defined($oppos{'v'}) && !$has_nds) {
 756                 die "$fname: $line: 'v' operand without vex.nds or vex.ndd\n";
 757             }
 758             push(@codes, defined($oppos{'v'}) ? 0260+($oppos{'v'} & 3) : 0270,
 759                  ($c << 6)+$m, ($w << 3)+($l << 2)+$p);
 760             $prefix_ok = 0;
 761         } elsif ($op =~ /^\/drex([01])$/) {
 762             my $oc0 = $1;
 763             if (!defined($oppos{'d'})) {
 764                 die "$fname: $line: DREX without a 'd' operand\n";
 765             }
 766             # Note the use of *unshift* here, as opposed to *push*.
 767             # This is because NASM want this byte code at the start of
 768             # the instruction sequence, but the AMD documentation puts
 769             # this at (roughly) the position of the drex byte itself.
 770             # This allows us to match the AMD documentation and still
 771             # do the right thing.
 772             unshift(@codes, 0160+($oppos{'d'} & 3)+($oc0 ? 4 : 0));
 773             unshift(@codes, 05) if ($oppos{'d'} & 4);
 774         } elsif ($op =~ /^(ib\,s|ib|ibx|ib\,w|iw|iwd|id|idx|iwdq|rel|rel8|rel16|rel32|iq|seg|ibw|ibd|ibd,s)$/) {
 775             if (!defined($oppos{'i'})) {
 776                 die "$fname: $line: $op without 'i' operand\n";
 777             }
 778             if ($op eq 'ib,s') { # Signed imm8
 779                 push(@codes, 05) if ($oppos{'i'} & 4);
 780                 push(@codes, 014+($oppos{'i'} & 3));
 781             } elsif ($op eq 'ib') { # imm8
 782                 push(@codes, 05) if ($oppos{'i'} & 4);
 783                 push(@codes, 020+($oppos{'i'} & 3));
 784             } elsif ($op eq 'ib,u') { # Unsigned imm8
 785                 push(@codes, 05) if ($oppos{'i'} & 4);
 786                 push(@codes, 024+($oppos{'i'} & 3));
 787             } elsif ($op eq 'iw') { # imm16
 788                 push(@codes, 05) if ($oppos{'i'} & 4);
 789                 push(@codes, 030+($oppos{'i'} & 3));
 790             } elsif ($op eq 'ibx') { # imm8 sign-extended to opsize
 791                 push(@codes, 05) if ($oppos{'i'} & 4);
 792                 push(@codes, 0274+($oppos{'i'} & 3));
 793             } elsif ($op eq 'iwd') { # imm16 or imm32, depending on opsize
 794                 push(@codes, 05) if ($oppos{'i'} & 4);
 795                 push(@codes, 034+($oppos{'i'} & 3));
 796             } elsif ($op eq 'id') { # imm32
 797                 push(@codes, 05) if ($oppos{'i'} & 4);
 798                 push(@codes, 040+($oppos{'i'} & 3));
 799             } elsif ($op eq 'idx') { # imm32 extended to 64 bits
 800                 push(@codes, 05) if ($oppos{'i'} & 4);
 801                 push(@codes, 0254+($oppos{'i'} & 3));
 802             } elsif ($op eq 'iwdq') { # imm16/32/64, depending on opsize
 803                 push(@codes, 05) if ($oppos{'i'} & 4);
 804                 push(@codes, 044+($oppos{'i'} & 3));
 805             } elsif ($op eq 'rel8') {
 806                 push(@codes, 05) if ($oppos{'i'} & 4);
 807                 push(@codes, 050+($oppos{'i'} & 3));
 808             } elsif ($op eq 'iq') {
 809                 push(@codes, 05) if ($oppos{'i'} & 4);
 810                 push(@codes, 054+($oppos{'i'} & 3));
 811             } elsif ($op eq 'rel16') {
 812                 push(@codes, 05) if ($oppos{'i'} & 4);
 813                 push(@codes, 060+($oppos{'i'} & 3));
 814             } elsif ($op eq 'rel') { # 16 or 32 bit relative operand
 815                 push(@codes, 05) if ($oppos{'i'} & 4);
 816                 push(@codes, 064+($oppos{'i'} & 3));
 817             } elsif ($op eq 'rel32') {
 818                 push(@codes, 05) if ($oppos{'i'} & 4);
 819                 push(@codes, 070+($oppos{'i'} & 3));
 820             } elsif ($op eq 'seg') {
 821                 push(@codes, 05) if ($oppos{'i'} & 4);
 822                 push(@codes, 074+($oppos{'i'} & 3));
 823             } elsif ($op eq 'ibw') { # imm16 that can be bytified
 824                 if (!defined($s_pos)) {
 825                     die "$fname: $line: $op without a +s byte\n";
 826                 }
 827                 $codes[$s_pos] += 0144;
 828                 push(@codes, 05) if ($oppos{'i'} & 4);
 829                 push(@codes, 0140+($oppos{'i'} & 3));
 830             } elsif ($op eq 'ibd') { # imm32 that can be bytified
 831                 if (!defined($s_pos)) {
 832                     die "$fname: $line: $op without a +s byte\n";
 833                 }
 834                 $codes[$s_pos] += 0154;
 835                 push(@codes, 05) if ($oppos{'i'} & 4);
 836                 push(@codes, 0150+($oppos{'i'} & 3));
 837             } elsif ($op eq 'ibd,s') {
 838                 # imm32 that can be bytified, sign extended to 64 bits
 839                 if (!defined($s_pos)) {
 840                     die "$fname: $line: $op without a +s byte\n";
 841                 }
 842                 $codes[$s_pos] += 0154;
 843                 push(@codes, 05) if ($oppos{'i'} & 4);
 844                 push(@codes, 0250+($oppos{'i'} & 3));
 845             }
 846             $prefix_ok = 0;
 847         } elsif ($op eq '/is4') {
 848             if (!defined($oppos{'s'})) {
 849                 die "$fname: $line: $op without 's' operand\n";
 850             }
 851             if (defined($oppos{'i'})) {
 852                 push(@codes, 0172, ($oppos{'s'} << 3)+$oppos{'i'});
 853             } else {
 854                 push(@codes, 0174, $oppos{'s'});
 855             }
 856             $prefix_ok = 0;
 857         } elsif ($op =~ /^\/is4\=([0-9]+)$/) {
 858             my $imm = $1;
 859             if (!defined($oppos{'s'})) {
 860                 die "$fname: $line: $op without 's' operand\n";
 861             }
 862             if ($imm < 0 || $imm > 15) {
 863                 die "$fname: $line: invalid imm4 value for $op: $imm\n";
 864             }
 865             push(@codes, 0173, ($oppos{'s'} << 4) + $imm);
 866             $prefix_ok = 0;
 867         } elsif ($op =~ /^([0-9a-f]{2})\+s$/) {
 868             if (!defined($oppos{'i'})) {
 869                 die "$fname: $line: $op without 'i' operand\n";
 870             }
 871             $s_pos = scalar @codes;
 872             push(@codes, 05) if ($oppos{'i'} & 4);
 873             push(@codes, $oppos{'i'} & 3, hex $1);
 874             $prefix_ok = 0;
 875         } elsif ($op =~ /^([0-9a-f]{2})\+c$/) {
 876             push(@codes, 0330, hex $1);
 877             $prefix_ok = 0;
 878         } elsif ($op =~ /^\\([0-7]+|x[0-9a-f]{2})$/) {
 879             # Escape to enter literal bytecodes
 880             push(@codes, oct $1);
 881         } else {
 882             die "$fname: $line: unknown operation: $op\n";
 883         }
 884     }
 885
 886     return @codes;
 887 }