3 # insns.pl produce insnsa.c, insnsd.c, insnsi.h, insnsn.c from insns.dat
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.
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);
14 # This should match MAX_OPERANDS from nasm.h
19 for ($m = 0; $m < 32; $m++) {
20 for ($lp = 0; $lp < 8; $lp++) {
21 push(@vexlist, sprintf("VEX%02X%01X", $m, $lp));
24 @disasm_prefixes = (@vexlist, @disasm_prefixes);
26 print STDERR "Reading insns.dat...\n";
30 foreach $arg ( @ARGV ) {
31 if ( $arg =~ /^\-/ ) {
32 if ( $arg =~ /^\-([abdin])$/ ) {
35 die "$0: Unknown option: ${arg}\n";
42 $fname = "insns.dat" unless $fname = $args[0];
43 open (F, $fname) || die "unable to open $fname";
53 next if ( /^\s*(\;.*|)$/ ); # comments or blank lines
55 unless (/^\s*(\S+)\s+(\S+)\s+(\S+|\[.*\])\s+(\S+)\s*$/) {
56 warn "line $line does not contain four fields\n";
59 @fields = ($1, $2, $3, $4);
60 ($formatted, $nd) = format_insn(@fields);
63 $aname = "aa_$fields[0]";
64 push @$aname, $formatted;
66 if ( $fields[0] =~ /cc$/ ) {
67 # Conditional instruction
68 $k_opcodes_cc{$fields[0]}++;
70 # Unconditional instruction
71 $k_opcodes{$fields[0]}++;
73 if ($formatted && !$nd) {
74 push @big, $formatted;
75 my @sseq = startseq($fields[2]);
77 if (!defined($dinstables{$i})) {
80 push(@{$dinstables{$i}}, $#big);
88 # Generate the bytecode array. At this point, @bytecode_list contains
89 # the full set of bytecodes.
92 # Sort by descending length
93 @bytecode_list = sort { scalar(@$b) <=> scalar(@$a) } @bytecode_list;
97 foreach $bl (@bytecode_list) {
99 next if (defined($bytecode_pos{$h}));
101 push(@bytecode_array, $bl);
103 $bytecode_pos{$h} = $bytecode_next;
108 undef @bytecode_list;
110 @opcodes = sort keys(%k_opcodes);
111 @opcodes_cc = sort keys(%k_opcodes_cc);
113 if ( !defined($output) || $output eq 'b') {
114 print STDERR "Writing insnsb.c...\n";
118 print B "/* This file auto-generated from insns.dat by insns.pl" .
119 " - don't edit it */\n\n";
121 print B "#include \"nasm.h\"\n";
122 print B "#include \"insns.h\"\n\n";
124 print B "const uint8_t nasm_bytecodes[$bytecode_next] = {\n";
127 foreach $bl (@bytecode_array) {
128 printf B " /* %5d */ ", $p;
140 if ( !defined($output) || $output eq 'a' ) {
141 print STDERR "Writing insnsa.c...\n";
145 print A "/* This file auto-generated from insns.dat by insns.pl" .
146 " - don't edit it */\n\n";
148 print A "#include \"nasm.h\"\n";
149 print A "#include \"insns.h\"\n\n";
151 foreach $i (@opcodes, @opcodes_cc) {
152 print A "static const struct itemplate instrux_${i}[] = {\n";
154 foreach $j (@$aname) {
155 print A " ", codesubst($j), "\n";
157 print A " ITEMPLATE_END\n};\n\n";
159 print A "const struct itemplate * const nasm_instructions[] = {\n";
160 foreach $i (@opcodes, @opcodes_cc) {
161 print A " instrux_${i},\n";
168 if ( !defined($output) || $output eq 'd' ) {
169 print STDERR "Writing insnsd.c...\n";
173 print D "/* This file auto-generated from insns.dat by insns.pl" .
174 " - don't edit it */\n\n";
176 print D "#include \"nasm.h\"\n";
177 print D "#include \"insns.h\"\n\n";
179 print D "static const struct itemplate instrux[] = {\n";
182 printf D " /* %4d */ %s\n", $n++, codesubst($j);
186 foreach $h (sort(keys(%dinstables))) {
187 print D "\nstatic const struct itemplate * const itable_${h}[] = {\n";
188 foreach $j (@{$dinstables{$h}}) {
189 print D " instrux + $j,\n";
195 foreach $h (@disasm_prefixes, '') {
196 for ($c = 0; $c < 256; $c++) {
197 $nn = sprintf("%s%02X", $h, $c);
198 if ($is_prefix{$nn} || defined($dinstables{$nn})) {
199 # At least one entry in this prefix table
200 push(@prefix_list, $h);
207 foreach $h (@prefix_list) {
209 print D "static " unless ($h eq '');
210 print D "const struct disasm_index ";
211 print D ($h eq '') ? 'itable' : "itable_$h";
212 print D "[256] = {\n";
213 for ($c = 0; $c < 256; $c++) {
214 $nn = sprintf("%s%02X", $h, $c);
215 if ($is_prefix{$nn}) {
216 die "$fname: ambiguous decoding of $nn\n"
217 if (defined($dinstables{$nn}));
218 printf D " { itable_%s, -1 },\n", $nn;
219 } elsif (defined($dinstables{$nn})) {
220 printf D " { itable_%s, %u },\n",
221 $nn, scalar(@{$dinstables{$nn}});
223 printf D " { NULL, 0 },\n";
229 print D "\nconst struct disasm_index * const itable_VEX[32][8] = {\n ";
230 for ($m = 0; $m < 32; $m++) {
232 for ($lp = 0; $lp < 8; $lp++) {
233 $vp = sprintf("VEX%02X%01X", $m, $lp);
234 if ($is_prefix{$vp}) {
235 printf D " itable_%s,\n", $vp;
247 if ( !defined($output) || $output eq 'i' ) {
248 print STDERR "Writing insnsi.h...\n";
252 print I "/* This file is auto-generated from insns.dat by insns.pl" .
253 " - don't edit it */\n\n";
254 print I "/* This file in included by nasm.h */\n\n";
256 print I "/* Instruction names */\n\n";
257 print I "#ifndef NASM_INSNSI_H\n";
258 print I "#define NASM_INSNSI_H 1\n\n";
259 print I "enum opcode {\n";
261 foreach $i (@opcodes, @opcodes_cc) {
262 print I "\tI_${i},\n";
264 $len++ if ( $i =~ /cc$/ ); # Condition codes can be 3 characters long
265 $maxlen = $len if ( $len > $maxlen );
267 print I "\tI_none = -1\n";
269 print I "#define MAX_INSLEN ", $maxlen, "\n";
270 print I "#define FIRST_COND_OPCODE I_", $opcodes_cc[0], "\n\n";
271 print I "#endif /* NASM_INSNSI_H */\n";
276 if ( !defined($output) || $output eq 'n' ) {
277 print STDERR "Writing insnsn.c...\n";
281 print N "/* This file is auto-generated from insns.dat by insns.pl" .
282 " - don't edit it */\n\n";
283 print N "#include \"tables.h\"\n\n";
285 print N "const char * const nasm_insn_names[] = {";
287 foreach $i (@opcodes, @opcodes_cc) {
288 print N "," if ( !$first );
291 $ilower =~ s/cc$//; # Remove conditional cc suffix
292 $ilower =~ tr/A-Z/a-z/; # Change to lower case (Perl 4 compatible)
293 print N "\n\t\"${ilower}\"";
299 printf STDERR "Done: %d instructions\n", $insns;
302 my ($opcode, $operands, $codes, $flags) = @_;
306 return (undef, undef) if $operands eq "ignore";
308 # format the operands
309 $operands =~ s/:/|colon,/g;
310 $operands =~ s/mem(\d+)/mem|bits$1/g;
311 $operands =~ s/mem/memory/g;
312 $operands =~ s/memory_offs/mem_offs/g;
313 $operands =~ s/imm(\d+)/imm|bits$1/g;
314 $operands =~ s/imm/immediate/g;
315 $operands =~ s/rm(\d+)/rm_gpr|bits$1/g;
316 $operands =~ s/(mmx|xmm|ymm)rm/rm_$1/g;
317 $operands =~ s/\=([0-9]+)/same_as|$1/g;
318 if ($operands eq 'void') {
321 @ops = split(/\,/, $operands);
324 while (scalar(@ops) < $MAX_OPERANDS) {
327 $operands = join(',', @ops);
328 $operands =~ tr/a-z/A-Z/;
331 $flags =~ s/,/|IF_/g;
332 $flags =~ s/(\|IF_ND|IF_ND\|)//, $nd = 1 if $flags =~ /IF_ND/;
333 $flags = "IF_" . $flags;
335 @bytecode = (decodify($codes), 0);
336 push(@bytecode_list, [@bytecode]);
337 $codes = hexstr(@bytecode);
339 ("{I_$opcode, $num, {$operands}, \@\@CODES-$codes\@\@, $flags},", $nd);
343 # Look for @@CODES-xxx@@ sequences and replace them with the appropriate
344 # offset into nasm_bytecodes
350 while ($s =~ /\@\@CODES-([0-9A-F]+)\@\@/) {
351 my $pos = $bytecode_pos{$1};
352 if (!defined($pos)) {
353 die "$fname: no position assigned to byte code $1\n";
355 $s = $` . "nasm_bytecodes+${pos}" . "$'";
361 my ($prefix, @list) = @_;
366 push(@l, sprintf("%s%02X", $prefix, $x));
373 # Turn a code string into a sequence of bytes
376 # Although these are C-syntax strings, by convention they should have
377 # only octal escapes (for directives) and hexadecimal escapes
378 # (for verbatim bytes)
381 if ($codestr =~ /^\s*\[([^\]]*)\]\s*$/) {
382 return byte_code_compile($1);
389 if ($c =~ /^\\x([0-9a-f]+)(.*)$/i) {
390 push(@codes, hex $1);
393 } elsif ($c =~ /^\\([0-7]{1,3})(.*)$/) {
394 push(@codes, oct $1);
398 die "$fname: unknown code format in \"$codestr\"\n";
405 # Turn a numeric list into a hex string
411 $s .= sprintf("%02X", $c);
416 # Here we determine the range of possible starting bytes for a given
417 # instruction. We need only consider the codes:
418 # \1 \2 \3 mean literal bytes, of course
419 # \4 \5 \6 \7 mean PUSH/POP of segment registers: special case
420 # \1[0123] mean byte plus register value
421 # \330 means byte plus condition code
422 # \0 or \340 mean give up and return empty set
423 # \17[234] skip is4 control byte
424 # \26x \270 skip VEX control bytes
433 @codes = decodify($codestr);
435 while ($c0 = shift(@codes)) {
437 if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
441 if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
443 $fbs .= sprintf("%02X", shift(@codes));
451 foreach $pfx (@disasm_prefixes) {
452 if (substr($fbs, 0, length($pfx)) eq $pfx) {
454 $fbs = substr($fbs, length($pfx));
460 return ($prefix.substr($fbs,0,2));
463 unshift(@codes, $c0);
464 } elsif ($c0 == 04) {
465 return addprefix($prefix, 0x07, 0x17, 0x1F);
466 } elsif ($c0 == 05) {
467 return addprefix($prefix, 0xA1, 0xA9);
468 } elsif ($c0 == 06) {
469 return addprefix($prefix, 0x06, 0x0E, 0x16, 0x1E);
470 } elsif ($c0 == 07) {
471 return addprefix($prefix, 0xA0, 0xA8);
472 } elsif ($c0 >= 010 && $c0 <= 013) {
473 return addprefix($prefix, $c1..($c1+7));
474 } elsif (($c0 & ~013) == 0144) {
475 return addprefix($prefix, $c1, $c1|2);
476 } elsif ($c0 == 0330) {
477 return addprefix($prefix, $c1..($c1+15));
478 } elsif ($c0 == 0 || $c0 == 0340) {
480 } elsif (($c0 & ~3) == 0260 || $c0 == 0270) {
483 $wlp = shift(@codes);
484 $prefix .= sprintf('VEX%02X%01X', $m, $wlp & 7);
485 } elsif ($c0 >= 0172 && $c0 <= 174) {
486 shift(@codes); # Skip is4 control byte
488 # We really need to be able to distinguish "forbidden"
489 # and "ignorable" codes here
496 # This function takes a series of byte codes in a format which is more
497 # typical of the Intel documentation, and encode it.
499 # The format looks like:
501 # [operands: opcodes]
503 # The operands word lists the order of the operands:
505 # r = register field in the modr/m
508 # d = DREX "dst" field
510 # s = register field of is4/imz2 field
511 # - = implicit (unencoded) operand
513 # For an operand that should be filled into more than one field,
514 # enter it as e.g. "r+v".
516 sub byte_code_compile($) {
526 unless ($str =~ /^(([^\s:]*)\:|)\s*(.*\S)\s*$/) {
527 die "$fname: $line: cannot parse: [$str]\n";
533 for ($i = 0; $i < length($opr); $i++) {
534 my $c = substr($opr,$i,1);
543 foreach $op (split(/\s*(?:\s|(?=[\/\\]))/, $opc)) {
546 } elsif ($op eq 'o32') {
548 } elsif ($op eq 'o64') { # 64-bit operand size requiring REX.W
550 } elsif ($op eq 'o64nw') { # Implied 64-bit operand size (no REX.W)
552 } elsif ($op eq 'a16') {
554 } elsif ($op eq 'a32') {
556 } elsif ($op eq 'a64') {
558 } elsif ($op eq '!osp') {
560 } elsif ($op eq '!asp') {
562 } elsif ($op eq 'rex.l') {
564 } elsif ($op eq 'repe') {
566 } elsif ($prefix_ok && $op =~ /^(66|f2|f3|np)$/) {
567 # 66/F2/F3 prefix used as an opcode extension, or np = no prefix
570 } elsif ($op eq 'f2') {
572 } elsif ($op eq 'f3') {
577 } elsif ($op =~ /^[0-9a-f]{2}$/) {
578 if (defined($litix) && $litix+$codes[$litix]+1 == scalar @codes) {
580 push(@codes, hex $op);
582 $litix = scalar(@codes);
583 push(@codes, 01, hex $op);
586 } elsif ($op eq '/r') {
587 if (!defined($oppos{'r'}) || !defined($oppos{'m'})) {
588 die "$fname: $line: $op requires r and m operands\n";
590 push(@codes, 0100 + ($oppos{'m'} << 3) + $oppos{'r'});
592 } elsif ($op =~ m:^/([0-7])$:) {
593 if (!defined($oppos{'m'})) {
594 die "$fname: $line: $op requires m operand\n";
596 push(@codes, 0200 + ($oppos{'m'} << 3) + $1);
598 } elsif ($op =~ /^vex(|\..*)$/) {
599 my ($m,$w,$l,$p) = (undef,2,undef,0);
601 foreach $oq (split(/\./, $op)) {
604 } elsif ($oq eq '128' || $oq eq 'l0') {
606 } elsif ($oq eq '256' || $oq eq 'l1') {
608 } elsif ($oq eq 'w0') {
610 } elsif ($oq eq 'w1') {
612 } elsif ($oq eq 'wx') {
614 } elsif ($oq eq 'ww') {
616 } elsif ($oq eq '66') {
618 } elsif ($oq eq 'f3') {
620 } elsif ($oq eq 'f2') {
622 } elsif ($oq eq '0f') {
624 } elsif ($oq eq '0f38') {
626 } elsif ($oq eq '0f3a') {
628 } elsif ($oq =~ /^m([0-9]+)$/) {
630 } elsif ($oq eq 'nds' || $oq eq 'ndd') {
631 if (!defined($oppos{'v'})) {
632 die "$fname: $line: vex.$oq without 'v' operand\n";
636 die "$fname: $line: undefined VEX subcode: $oq\n";
639 if (!defined($m) || !defined($w) || !defined($l) || !defined($p)) {
640 die "$fname: $line: missing fields in VEX specification\n";
642 if (defined($oppos{'v'}) && !$has_nds) {
643 die "$fname: $line: 'v' operand without vex.nds or vex.ndd\n";
645 push(@codes, defined($oppos{'v'}) ? 0260+$oppos{'v'} : 0270,
646 $m, ($w << 3)+($l << 2)+$p);
648 } elsif ($op =~ /^\/drex([01])$/) {
650 if (!defined($oppos{'d'})) {
651 die "$fname: $line: DREX without a 'd' operand\n";
653 # Note the use of *unshift* here, as opposed to *push*.
654 # This is because NASM want this byte code at the start of
655 # the instruction sequence, but the AMD documentation puts
656 # this at (roughly) the position of the drex byte itself.
657 # This allows us to match the AMD documentation and still
658 # do the right thing.
659 unshift(@codes, 0160+$oppos{'d'}+($oc0 ? 4 : 0));
660 } elsif ($op =~ /^(ib\,s|ib|ib\,w|iw|iwd|id|iwdq|rel|rel8|rel16|rel32|iq|seg|ibw|ibd|ibd,s)$/) {
661 if (!defined($oppos{'i'})) {
662 die "$fname: $line: $op without 'i' operand\n";
664 if ($op eq 'ib,s') { # Signed imm8
665 push(@codes, 014+$oppos{'i'});
666 } elsif ($op eq 'ib') { # imm8
667 push(@codes, 020+$oppos{'i'});
668 } elsif ($op eq 'ib,u') { # Unsigned imm8
669 push(@codes, 024+$oppos{'i'});
670 } elsif ($op eq 'iw') { # imm16
671 push(@codes, 030+$oppos{'i'});
672 } elsif ($op eq 'iwd') { # imm16 or imm32, depending on opsize
673 push(@codes, 034+$oppos{'i'});
674 } elsif ($op eq 'id') { # imm32
675 push(@codes, 040+$oppos{'i'});
676 } elsif ($op eq 'iwdq') { # imm16/32/64, depending on opsize
677 push(@codes, 044+$oppos{'i'});
678 } elsif ($op eq 'rel8') {
679 push(@codes, 050+$oppos{'i'});
680 } elsif ($op eq 'iq') {
681 push(@codes, 054+$oppos{'i'});
682 } elsif ($op eq 'rel16') {
683 push(@codes, 060+$oppos{'i'});
684 } elsif ($op eq 'rel') { # 16 or 32 bit relative operand
685 push(@codes, 064+$oppos{'i'});
686 } elsif ($op eq 'rel32') {
687 push(@codes, 070+$oppos{'i'});
688 } elsif ($op eq 'seg') {
689 push(@codes, 074+$oppos{'i'});
690 } elsif ($op eq 'ibw') { # imm16 that can be bytified
691 if (!defined($s_pos)) {
692 die "$fname: $line: $op without a +s byte\n";
694 $codes[$s_pos] += 0144;
695 push(@codes, 0140+$oppos{'i'});
696 } elsif ($op eq 'ibd') { # imm32 that can be bytified
697 if (!defined($s_pos)) {
698 die "$fname: $line: $op without a +s byte\n";
700 $codes[$s_pos] += 0154;
701 push(@codes, 0150+$oppos{'i'});
702 } elsif ($op eq 'ibd,s') {
703 # imm32 that can be bytified, sign extended to 64 bits
704 if (!defined($s_pos)) {
705 die "$fname: $line: $op without a +s byte\n";
707 $codes[$s_pos] += 0154;
708 push(@codes, 0250+$oppos{'i'});
711 } elsif ($op eq '/is4') {
712 if (!defined($oppos{'s'})) {
713 die "$fname: $line: $op without 's' operand\n";
715 if (defined($oppos{'i'})) {
716 push(@codes, 0172, ($oppos{'s'} << 3)+$oppos{'i'});
718 push(@codes, 0174, $oppos{'s'});
721 } elsif ($op =~ /^\/is4\=([0-9]+)$/) {
723 if (!defined($oppos{'s'})) {
724 die "$fname: $line: $op without 's' operand\n";
726 if ($imm < 0 || $imm > 15) {
727 die "$fname: $line: invalid imm4 value for $op: $imm\n";
729 push(@codes, 0173, ($oppos{'s'} << 4) + $imm);
731 } elsif ($op =~ /^([0-9a-f]{2})\+s$/) {
732 if (!defined($oppos{'i'})) {
733 die "$fname: $line: $op without 'i' operand\n";
735 $s_pos = scalar @codes;
736 push(@codes, $oppos{'i'}, hex $1);
738 } elsif ($op =~ /^([0-9a-f]{2})\+c$/) {
739 push(@codes, 0330, hex $1);
741 } elsif ($op =~ /^\\([0-7]+|x[0-9a-f]{2})$/) {
742 # Escape to enter literal bytecodes
743 push(@codes, oct $1);
745 die "$fname: $line: unknown operation: $op\n";