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