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