Generate a byte array instead of using strings for the byte codes

Generate a byte array instead of using C compiler strings for the byte
codes.  This has a few advantages:

- No need to special-case zero due to broken C compilers.
- Only insns.pl only ever reads the string, so we can invent our own
  syntax.
- Compaction.
- We can give it the proper, unsigned type.

diff --git a/Makefile.in b/Makefile.in
index 287340f..a0308c0 100644 (file)
--- a/Makefile.in
+++ b/Makefile.in
@@ -84,6 +84,8 @@ ndisasm$(X): $(NDISASM) $(XOBJS)
 # though, so it isn't necessary to have Perl just to recompile NASM
 # from the distribution.
 
+insnsb.c: insns.dat insns.pl
+       $(PERL) $(srcdir)/insns.pl -b $(srcdir)/insns.dat
 insnsa.c: insns.dat insns.pl
        $(PERL) $(srcdir)/insns.pl -a $(srcdir)/insns.dat
 insnsd.c: insns.dat insns.pl
@@ -142,7 +144,7 @@ pptok.c: pptok.dat pptok.pl perllib/phash.ph
 
 # This target generates all files that require perl.
 # This allows easier generation of distribution (see dist target).
-PERLREQ = macros.c insnsa.c insnsd.c insnsi.h insnsn.c \
+PERLREQ = macros.c insnsb.c insnsa.c insnsd.c insnsi.h insnsn.c \
          regs.c regs.h regflags.c regdis.c regvals.c tokhash.c tokens.h \
          version.h version.mac pptok.h pptok.c
 perlreq: $(PERLREQ)
@@ -238,10 +240,12 @@ float.$(O): float.c compiler.h config.h float.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h
 hashtbl.$(O): hashtbl.c compiler.h config.h hashtbl.h insnsi.h nasm.h \
  nasmlib.h regs.h version.h
-insnsa.$(O): insnsa.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
- regs.h tokens.h version.h
-insnsd.$(O): insnsd.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
+insnsa.$(O): insnsa.c compiler.h config.h insns.h insnsb.c insnsi.h nasm.h \
+ nasmlib.h regs.h tokens.h version.h
+insnsb.$(O): insnsb.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
  regs.h tokens.h version.h
+insnsd.$(O): insnsd.c compiler.h config.h insns.h insnsb.c insnsi.h nasm.h \
+ nasmlib.h regs.h tokens.h version.h
 insnsn.$(O): insnsn.c
 labels.$(O): labels.c compiler.h config.h hashtbl.h insnsi.h nasm.h \
  nasmlib.h regs.h version.h

diff --git a/Mkfiles/msvc.mak b/Mkfiles/msvc.mak
index bf8bc01..e154634 100644 (file)
--- a/Mkfiles/msvc.mak
+++ b/Mkfiles/msvc.mak
@@ -19,8 +19,8 @@ CFLAGS                = /O2 /Ox /Oy /W2
 BUILD_CFLAGS   = $(CFLAGS) /I$(srcdir)/inttypes
 INTERNAL_CFLAGS = /I$(srcdir) /I. /DHAVE__SNPRINTF /DHAVE__VSNPRINTF
 ALL_CFLAGS     = $(BUILD_CFLAGS) $(INTERNAL_CFLAGS)
-LDFLAGS                = 
-LIBS           = 
+LDFLAGS                =
+LIBS           =
 PERL           = perl -I$(srcdir)/perllib
 
 # Binary suffixes
@@ -58,6 +58,8 @@ ndisasm$(X): $(NDISASM)
 # though, so it isn't necessary to have Perl just to recompile NASM
 # from the distribution.
 
+insnsb.c: insns.dat insns.pl
+       $(PERL) $(srcdir)/insns.pl -b $(srcdir)/insns.dat
 insnsa.c: insns.dat insns.pl
        $(PERL) $(srcdir)/insns.pl -a $(srcdir)/insns.dat
 insnsd.c: insns.dat insns.pl
@@ -113,7 +115,7 @@ pptok.c: pptok.dat pptok.pl perllib/phash.ph
 
 # This target generates all files that require perl.
 # This allows easier generation of distribution (see dist target).
-PERLREQ = macros.c insnsa.c insnsd.c insnsi.h insnsn.c \
+PERLREQ = macros.c insnsb.c insnsa.c insnsd.c insnsi.h insnsn.c \
          regs.c regs.h regflags.c regdis.c regvals.c tokhash.c tokens.h \
          version.h version.mac pptok.h pptok.c
 perlreq: $(PERLREQ)
@@ -186,10 +188,12 @@ float.$(O): float.c compiler.h float.h insnsi.h nasm.h nasmlib.h regs.h \
  version.h
 hashtbl.$(O): hashtbl.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h
-insnsa.$(O): insnsa.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
- tokens.h version.h
-insnsd.$(O): insnsd.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
+insnsa.$(O): insnsa.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h \
+ regs.h tokens.h version.h
+insnsb.$(O): insnsb.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
  tokens.h version.h
+insnsd.$(O): insnsd.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h \
+ regs.h tokens.h version.h
 insnsn.$(O): insnsn.c
 labels.$(O): labels.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h regs.h \
  version.h

diff --git a/Mkfiles/netware.mak b/Mkfiles/netware.mak
index 6f6e502..639b138 100644 (file)
--- a/Mkfiles/netware.mak
+++ b/Mkfiles/netware.mak
@@ -151,10 +151,12 @@ float.o: float.c compiler.h config.h float.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h
 hashtbl.o: hashtbl.c compiler.h config.h hashtbl.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h
-insnsa.o: insnsa.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
- regs.h tokens.h version.h
-insnsd.o: insnsd.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
+insnsa.o: insnsa.c compiler.h config.h insns.h insnsb.c insnsi.h nasm.h \
+ nasmlib.h regs.h tokens.h version.h
+insnsb.o: insnsb.c compiler.h config.h insns.h insnsi.h nasm.h nasmlib.h \
  regs.h tokens.h version.h
+insnsd.o: insnsd.c compiler.h config.h insns.h insnsb.c insnsi.h nasm.h \
+ nasmlib.h regs.h tokens.h version.h
 insnsn.o: insnsn.c
 labels.o: labels.c compiler.h config.h hashtbl.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h

diff --git a/Mkfiles/openwcom.mak b/Mkfiles/openwcom.mak
index dbbc198..a9cc5c7 100644 (file)
--- a/Mkfiles/openwcom.mak
+++ b/Mkfiles/openwcom.mak
@@ -14,14 +14,14 @@ bindir              = $(prefix)\bin
 mandir         = $(prefix)\man
 
 CC             = wcl386
-DEBUG          = 
+DEBUG          =
 CFLAGS         = -6 -ox -wx -ze -fpi $(DEBUG)
 BUILD_CFLAGS   = $(CFLAGS) $(TARGET_FLAGS) # -I$(srcdir)/inttypes
 INTERNAL_CFLAGS = -I$(srcdir) -I. -DHAVE_SNPRINTF -DHAVE_VSNPRINTF
 ALL_CFLAGS     = $(BUILD_CFLAGS) $(INTERNAL_CFLAGS)
 LD             = $(CC)
 LDFLAGS                = $(ALL_CFLAGS)
-LIBS           = 
+LIBS           =
 PERL           = perl -I$(srcdir)/perllib
 
 STRIP          = wstrip
@@ -82,6 +82,8 @@ ndisasm$(X): $(NDISASM)
 # though, so it isn't necessary to have Perl just to recompile NASM
 # from the distribution.
 
+insnsb.c: insns.dat insns.pl
+       $(PERL) $(srcdir)/insns.pl -b $(srcdir)/insns.dat
 insnsa.c: insns.dat insns.pl
        $(PERL) $(srcdir)/insns.pl -a $(srcdir)/insns.dat
 insnsd.c: insns.dat insns.pl
@@ -137,7 +139,7 @@ pptok.c: pptok.dat pptok.pl perllib/phash.ph
 
 # This target generates all files that require perl.
 # This allows easier generation of distribution (see dist target).
-PERLREQ = macros.c insnsa.c insnsd.c insnsi.h insnsn.c &
+PERLREQ = macros.c insnsb.c insnsa.c insnsd.c insnsi.h insnsn.c &
          regs.c regs.h regflags.c regdis.c regvals.c tokhash.c tokens.h &
          version.h version.mac pptok.h pptok.c
 perlreq: $(PERLREQ)
@@ -212,10 +214,12 @@ float.$(O): float.c compiler.h float.h insnsi.h nasm.h nasmlib.h regs.h &
  version.h
 hashtbl.$(O): hashtbl.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h &
  regs.h version.h
-insnsa.$(O): insnsa.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h &
- tokens.h version.h
-insnsd.$(O): insnsd.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h &
+insnsa.$(O): insnsa.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h &
+ regs.h tokens.h version.h
+insnsb.$(O): insnsb.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h &
  tokens.h version.h
+insnsd.$(O): insnsd.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h &
+ regs.h tokens.h version.h
 insnsn.$(O): insnsn.c
 labels.$(O): labels.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h regs.h &
  version.h

diff --git a/Mkfiles/owlinux.mak b/Mkfiles/owlinux.mak
index aed1188..ce369ed 100644 (file)
--- a/Mkfiles/owlinux.mak
+++ b/Mkfiles/owlinux.mak
@@ -12,7 +12,7 @@
 # > building on DOS, Windows, or OS/2, as they share the same C
 # > library headers. But when cross-compiling from (or to) Linux, it
 # > is crucial.
-# > 
+# >
 # > This may be accomplished by setting the INCLUDE env var in the
 # > makefile, or setting OS2_INCLUDE, DOS_INCLUDE, NT_INCLUDE env vars
 # > *and* making sure that the proper -bt switch is used, or passing a
@@ -28,7 +28,7 @@ bindir                = $(prefix)/bin
 mandir         = $(prefix)/man
 
 CC             = wcl386
-DEBUG          = 
+DEBUG          =
 CFLAGS         = -6 -ox -wx -ze -fpi $(DEBUG)
 BUILD_CFLAGS   = $(CFLAGS) $(TARGET_FLAGS) # -I$(srcdir)/inttypes
 INTERNAL_CFLAGS = -I$(srcdir) -I. \
@@ -36,7 +36,7 @@ INTERNAL_CFLAGS = -I$(srcdir) -I. \
 ALL_CFLAGS     = $(BUILD_CFLAGS) $(INTERNAL_CFLAGS)
 LD             = $(CC)
 LDFLAGS                = $(ALL_CFLAGS)
-LIBS           = 
+LIBS           =
 PERL           = perl -I$(srcdir)/perllib
 
 STRIP          = wstrip
@@ -92,6 +92,8 @@ ndisasm$(X): $(NDISASM)
 # though, so it isn't necessary to have Perl just to recompile NASM
 # from the distribution.
 
+insnsb.c: insns.dat insns.pl
+       $(PERL) $(srcdir)/insns.pl -b $(srcdir)/insns.dat
 insnsa.c: insns.dat insns.pl
        $(PERL) $(srcdir)/insns.pl -a $(srcdir)/insns.dat
 insnsd.c: insns.dat insns.pl
@@ -147,7 +149,7 @@ pptok.c: pptok.dat pptok.pl perllib/phash.ph
 
 # This target generates all files that require perl.
 # This allows easier generation of distribution (see dist target).
-PERLREQ = macros.c insnsa.c insnsd.c insnsi.h insnsn.c \
+PERLREQ = macros.c insnsb.c insnsa.c insnsd.c insnsi.h insnsn.c \
          regs.c regs.h regflags.c regdis.c regvals.c tokhash.c tokens.h \
          version.h version.mac pptok.h pptok.c
 perlreq: $(PERLREQ)
@@ -222,10 +224,12 @@ float.$(O): float.c compiler.h float.h insnsi.h nasm.h nasmlib.h regs.h \
  version.h
 hashtbl.$(O): hashtbl.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h \
  regs.h version.h
-insnsa.$(O): insnsa.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
- tokens.h version.h
-insnsd.$(O): insnsd.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
+insnsa.$(O): insnsa.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h \
+ regs.h tokens.h version.h
+insnsb.$(O): insnsb.c compiler.h insns.h insnsi.h nasm.h nasmlib.h regs.h \
  tokens.h version.h
+insnsd.$(O): insnsd.c compiler.h insns.h insnsb.c insnsi.h nasm.h nasmlib.h \
+ regs.h tokens.h version.h
 insnsn.$(O): insnsn.c
 labels.$(O): labels.c compiler.h hashtbl.h insnsi.h nasm.h nasmlib.h regs.h \
  version.h

diff --git a/assemble.c b/assemble.c
index 56a05e6..104b42c 100644 (file)
--- a/assemble.c
+++ b/assemble.c
@@ -130,8 +130,8 @@ static efunc errfunc;
 static struct ofmt *outfmt;
 static ListGen *list;
 
-static int64_t calcsize(int32_t, int64_t, int, insn *, const char *);
-static void gencode(int32_t, int64_t, int, insn *, const char *, int64_t);
+static int64_t calcsize(int32_t, int64_t, int, insn *, const uint8_t *);
+static void gencode(int32_t, int64_t, int, insn *, const uint8_t *, int64_t);
 static int matches(const struct itemplate *, insn *, int bits);
 static int32_t regflag(const operand *);
 static int32_t regval(const operand *);
@@ -231,7 +231,7 @@ static void out(int64_t offset, int32_t segto, const void *data,
 }
 
 static int jmp_match(int32_t segment, int64_t offset, int bits,
-                     insn * ins, const char *code)
+                     insn * ins, const uint8_t *code)
 {
     int64_t isize;
     uint8_t c = code[0];
@@ -462,7 +462,7 @@ int64_t assemble(int32_t segment, int64_t offset, int bits, uint32_t cp,
             m += jmp_match(segment, offset, bits, instruction, temp->code);
 
         if (m == 100) {         /* matches! */
-            const char *codes = temp->code;
+            const uint8_t *codes = temp->code;
             int64_t insn_size = calcsize(segment, offset, bits,
                                       instruction, codes);
             itimes = instruction->times;
@@ -737,7 +737,7 @@ int64_t insn_size(int32_t segment, int64_t offset, int bits, uint32_t cp,
         if (m == 100) {
             /* we've matched an instruction. */
             int64_t isize;
-            const char *codes = temp->code;
+            const uint8_t *codes = temp->code;
             int j;
 
             isize = calcsize(segment, offset, bits, instruction, codes);
@@ -826,7 +826,7 @@ static bool is_sbyte64(insn * ins, int op)
     return v32 >= -128 && v32 <= 127;
 }
 static int64_t calcsize(int32_t segment, int64_t offset, int bits,
-                     insn * ins, const char *codes)
+                     insn * ins, const uint8_t *codes)
 {
     int64_t length = 0;
     uint8_t c;
@@ -1200,7 +1200,7 @@ static int64_t calcsize(int32_t segment, int64_t offset, int bits,
     }
 
 static void gencode(int32_t segment, int64_t offset, int bits,
-                    insn * ins, const char *codes, int64_t insn_end)
+                    insn * ins, const uint8_t *codes, int64_t insn_end)
 {
     static char condval[] = {   /* conditional opcodes */
         0x7, 0x3, 0x2, 0x6, 0x2, 0x4, 0xF, 0xD, 0xC, 0xE, 0x6, 0x2,

diff --git a/insns.h b/insns.h
index 35d05b5..49cdc8c 100644 (file)
--- a/insns.h
+++ b/insns.h
@@ -16,7 +16,7 @@ struct itemplate {
     enum opcode opcode;                /* the token, passed from "parser.c" */
     int operands;              /* number of operands */
     opflags_t opd[MAX_OPERANDS]; /* bit flags for operand types */
-    const char *code;          /* the code it assembles to */
+    const uint8_t *code;       /* the code it assembles to */
     uint32_t flags;            /* some flags */
 };
 

diff --git a/insns.pl b/insns.pl
index 03c840a..eff70f5 100644 (file)
--- a/insns.pl
+++ b/insns.pl
@@ -17,7 +17,7 @@ print STDERR "Reading insns.dat...\n";
 undef $output;
 foreach $arg ( @ARGV ) {
     if ( $arg =~ /^\-/ ) {
-       if  ( $arg =~ /^\-([adin])$/ ) {
+       if  ( $arg =~ /^\-([abdin])$/ ) {
            $output = $1;
        } else {
            die "$0: Unknown option: ${arg}\n";
@@ -31,6 +31,7 @@ $fname = "insns.dat" unless $fname = $args[0];
 open (F, $fname) || die "unable to open $fname";
 
 %dinstables = ();
+@bytecode_list = ();
 
 $line = 0;
 $insns = 0;
@@ -68,9 +69,59 @@ while (<F>) {
 
 close F;
 
+#
+# Generate the bytecode array.  At this point, @bytecode_list contains
+# the full set of bytecodes.
+#
+
+# Sort by descending length
+@bytecode_list = sort { scalar(@$b) <=> scalar(@$a) } @bytecode_list;
+@bytecode_array = ();
+%bytecode_pos = ();
+$bytecode_next = 0;
+foreach $bl (@bytecode_list) {
+    my $h = hexstr(@$bl);
+    next if (defined($bytecode_pos{$h}));
+
+    push(@bytecode_array, $bl);
+    while ($h ne '') {
+       $bytecode_pos{$h} = $bytecode_next;
+       $h = substr($h, 2);
+       $bytecode_next++;
+    }
+}
+undef @bytecode_list;
+
 @opcodes    = sort keys(%k_opcodes);
 @opcodes_cc = sort keys(%k_opcodes_cc);
 
+if ( !defined($output) || $output eq 'b') {
+    print STDERR "Writing insnsb.c...\n";
+
+    open B, ">insnsb.c";
+    
+    print B "/* This file auto-generated from insns.dat by insns.pl" .
+        " - don't edit it */\n\n";
+
+    print B "#include \"nasm.h\"\n";
+    print B "#include \"insns.h\"\n\n";
+
+    print B "static const uint8_t nasm_bytecodes[$bytecode_next] = {\n";
+
+    $p = 0;
+    foreach $bl (@bytecode_array) {
+       printf B "    /* %4d */ ", $p;
+       foreach $d (@$bl) {
+           printf B "%#o,", $d;
+           $p++;
+       }
+       printf B "\n";
+    }
+    print B "};\n";
+
+    close B;
+}
+    
 if ( !defined($output) || $output eq 'a' ) {
     print STDERR "Writing insnsa.c...\n";
 
@@ -78,15 +129,14 @@ if ( !defined($output) || $output eq 'a' ) {
 
     print A "/* This file auto-generated from insns.dat by insns.pl" .
         " - don't edit it */\n\n";
-    print A "#include \"nasm.h\"\n";
-    print A "#include \"insns.h\"\n";
-    print A "\n";
+
+    print A "#include \"insnsb.c\"\n\n";
 
     foreach $i (@opcodes, @opcodes_cc) {
        print A "static const struct itemplate instrux_${i}[] = {\n";
        $aname = "aa_$i";
        foreach $j (@$aname) {
-           print A "    $j\n";
+           print A "    ", codesubst($j), "\n";
        }
        print A "    ITEMPLATE_END\n};\n\n";
     }
@@ -106,14 +156,13 @@ if ( !defined($output) || $output eq 'd' ) {
 
     print D "/* This file auto-generated from insns.dat by insns.pl" .
         " - don't edit it */\n\n";
-    print D "#include \"nasm.h\"\n";
-    print D "#include \"insns.h\"\n";
-    print D "\n";
+
+    print D "#include \"insnsb.c\"\n\n";
 
     print D "static const struct itemplate instrux[] = {\n";
     $n = 0;
     foreach $j (@big) {
-       printf D "    /* %4d */ %s\n", $n++, $j;
+       printf D "    /* %4d */ %s\n", $n++, codesubst($j);
     }
     print D "};\n";
 
@@ -230,6 +279,7 @@ printf STDERR "Done: %d instructions\n", $insns;
 sub format {
     my ($opcode, $operands, $codes, $flags) = @_;
     my $num, $nd = 0;
+    my @bytecode;
 
     return (undef, undef) if $operands eq "ignore";
 
@@ -260,7 +310,29 @@ sub format {
     $flags =~ s/(\|IF_ND|IF_ND\|)//, $nd = 1 if $flags =~ /IF_ND/;
     $flags = "IF_" . $flags;
 
-    ("{I_$opcode, $num, {$operands}, \"$codes\", $flags},", $nd);
+    @bytecode = (decodify($codes), 0);
+    push(@bytecode_list, [@bytecode]);
+    $codes = hexstr(@bytecode);
+
+    ("{I_$opcode, $num, {$operands}, \@\@CODES-$codes\@\@, $flags},", $nd);
+}
+
+#
+# Look for @@CODES-xxx@@ sequences and replace them with the appropriate
+# offset into nasm_bytecodes
+#
+sub codesubst($) {
+    my($s) = @_;
+    my $n;
+
+    while ($s =~ /\@\@CODES-([0-9A-F]+)\@\@/) {
+       my $pos = $bytecode_pos{$1};
+       if (!defined($pos)) {
+           die "$0: no position assigned to byte code $1\n";
+       }
+       $s = $` . "nasm_bytecodes+${pos}" . "$'";
+    }
+    return $s;
 }
 
 sub addprefix ($@) {
@@ -303,6 +375,17 @@ sub decodify($) {
     return @codes;
 }
 
+# Turn a numeric list into a hex string
+sub hexstr(@) {
+    my $s = '';
+    my $c;
+
+    foreach $c (@_) {
+       $s .= sprintf("%02X", $c);
+    }
+    return $s;
+}
+
 # Here we determine the range of possible starting bytes for a given
 # instruction. We need only consider the codes:
 # \1 \2 \3     mean literal bytes, of course