Peter Johnson [Sat, 1 Dec 2001 06:15:58 +0000 (06:15 -0000)]
Don't free EA when simplifying shift_op case. Not only is this code completely
wrong (the ModRM byte definitely must be kept), but it had a memory leak too
(it just freed the ea without calling expr_delete())!
Peter Johnson [Mon, 26 Nov 2001 17:37:09 +0000 (17:37 -0000)]
Switch from using ternary tree to Hash Array Mapped Trie (HAMT), which has
*much* less overhead.
XXX: current implementation of HAMT is *not* portable due to pointer alignment
restrictions (it uses the LSB of a pointer to store a flag). Need to write a
portable (if not so space-efficient) equivalent.
Peter Johnson [Wed, 21 Nov 2001 08:25:09 +0000 (08:25 -0000)]
Massive cleanup of debugging output (via _print() functions). All now take
FILE *'s to print to somewhere other than stdout, and the formatting is
improved through the use of a global indent_level.
Changes to main() include the ability to specify an output file.
Peter Johnson [Wed, 21 Nov 2001 03:17:43 +0000 (03:17 -0000)]
Polish up most directive support. This necessitated adding objfmt-specific
data pointers to symrec for extern/common/etc. Also fixed parsing of direxpr;
it didn't allow exprs with ID's before.
Peter Johnson [Tue, 20 Nov 2001 21:02:17 +0000 (21:02 -0000)]
Don't do REG_E?AX combining with reg8/reg8x instructions, to allow things like
"ADD AX, byte 8" to generate the same code as NASM does (OT: this doesn't save
any space because the reg8x form needs a ModRM, but if it's what the user
wants, do it).
Peter Johnson [Mon, 19 Nov 2001 06:47:14 +0000 (06:47 -0000)]
Print "undefined symbol" message for all undefined symbols. Also make a little
nicer output by only outputting the "reported only once" informational message
once, for the first (line-wise) error.
Note: this isn't foolproof, as different files have different line countings...
need a monotonically increasing "line" counter to fix this. This is a problem
in more areas than just here, though.
Peter Johnson [Mon, 19 Nov 2001 04:25:44 +0000 (04:25 -0000)]
Don't try to redefine a label for an already existing segment. Thus, add a
return flag from sections_switch_general() to indicate if the segment is new,
which is checked in objfmt_sections_switch().
Peter Johnson [Mon, 19 Nov 2001 04:15:22 +0000 (04:15 -0000)]
Fix a memory re-free issue in nasm_parser_directive(). vp->param was passed
to a function that keeps the pointer, and then the pointer is freed. Whoops!
Peter Johnson [Mon, 19 Nov 2001 03:27:56 +0000 (03:27 -0000)]
- Greatly enhance capabilities of directives (they're actually parsed now).
objfmt interface changed a lot due to this.
- New variables cur_objfmt and cur_parser that define what objfmt/parser are
active. They're set in main() by searching through the list of objfmts and
parsers. This necessitated libyasm.a to be listed twice in LDADD.
Peter Johnson [Sun, 18 Nov 2001 07:06:05 +0000 (07:06 -0000)]
Add symrec_delete() for deleting non-table symbols. Use it from expr_delete().
Also allow the sect for labels to be NULL (special case for new directive
handling code).
Peter Johnson [Sat, 17 Nov 2001 08:39:14 +0000 (08:39 -0000)]
Decided to go ahead and commit these little NASM-syntax test files I've been
using for testing various things during development. Many don't even actually
assemble, as they're testing error conditions. These all need to be integrated
into a test suite at some point to be truly useful.
Peter Johnson [Sat, 17 Nov 2001 08:33:23 +0000 (08:33 -0000)]
Modify so that almost everything passes LCLint with the options in lclint.sh.
This is actually worthwhile; I found and fixed a few bugs/edge cases while
doing this.
For more information on LCLint, see <http://lclint.cs.virginia.edu/>.
Peter Johnson [Fri, 16 Nov 2001 22:05:24 +0000 (22:05 -0000)]
Fix some nits found by lclint. Still something weird going on with list2 in
mergesort(), according to lclint, but I haven't figured out if it's real or
not.
Peter Johnson [Sat, 3 Nov 2001 06:22:55 +0000 (06:22 -0000)]
Start framework for separating architecture-dependent functions and data
structures from the rest of the assembler. We're not trying to write GAS
here (yet :), but doing this will make large parts of the codebase a lot
clearer.
Peter Johnson [Sat, 3 Nov 2001 05:17:51 +0000 (05:17 -0000)]
Add coretype.h, make util.h do a lot more, use util.h to simplify include
sections of C files. Also remove IdPath from top comment in files where
RCSID() is used. Move RCSID() to immediately after util.h include.
Peter Johnson [Sat, 3 Nov 2001 04:13:42 +0000 (04:13 -0000)]
bytecode_print(): Change how NULL values are output for imm and multiple.
bytecode_parser_finalize_insn(): Add some preliminary imm handling. Handle
shift_op special-casing (,1 hack).
Peter Johnson [Thu, 1 Nov 2001 03:51:26 +0000 (03:51 -0000)]
Optimize expression building a bit in expr_new() by eliminating IDENT's here.
This saves a lot of extra unnecessary recursions when the expr gets to the
simplification stage.
Peter Johnson [Mon, 29 Oct 2001 00:01:31 +0000 (00:01 -0000)]
Add option for malloc debugging using dmalloc library (http://dmalloc.com/).
Remove unnecessary stdlib includes (because we use xmalloc instead of malloc).
Peter Johnson [Sun, 28 Oct 2001 09:22:13 +0000 (09:22 -0000)]
*COMPLETELY* rewrite most of expression handling.
Features:
- More complex memory expressions like [bx-(bx-di)] are now calculated through
as valid.
- Much better simplification.
Internal changes:
- More flexible n-way tree (rather than old binary-only).
- Many new functions to simplify and deal with very complex expressions.
Peter Johnson [Mon, 15 Oct 2001 04:32:16 +0000 (04:32 -0000)]
Allow both signed and unsigned interpretations for get_int and check_size.
Change argument order on get_int and check_size to better match other
functions (intnum * first arg).
Peter Johnson [Fri, 12 Oct 2001 03:36:10 +0000 (03:36 -0000)]
Add parsing of memory expressions. Currently adds a lot of duplication for
the expressions in the yacc code, but I don't see an easier way right now.
Also, no validity checking is done.
Peter Johnson [Fri, 12 Oct 2001 02:34:30 +0000 (02:34 -0000)]
Add expr_contains_float() and symrec_get_equ() functions. These are the first
of many functions like this which will be needed for advanced validity checking
in the finalization of the parser stage.
Fix yacc grammar to not use the (incorrect) expr_no_fltstr, and change memexp
to memexpr and its handling to prepare for memory expressions.
Peter Johnson [Fri, 12 Oct 2001 02:29:31 +0000 (02:29 -0000)]
Clarify and fix some usages of valid/need flags for modrm and sib bytes.
In bytecode_print(), check for NULL ea displacement and print out modrm and
sib bytes in octal (because it's easier to read the 2/3/3 format in octal).
Peter Johnson [Thu, 11 Oct 2001 05:34:50 +0000 (05:34 -0000)]
Fix LOTS of little nits. Primary changes:
- New intnum module to replace using just unsigned long (for >32 bit values).
Read values using BitVector instead of strtoul().
- Brought both floatnum and intnum into expr (still need to integrate calc for
floatnum).
- Add traversal function to ternary tree (for use by symrec_foreach).
- Change floatnum output interface from single/double/extended to byte size.
- Try to make floatnum/intnum interfaces very similar. Still needs work.
- Add/fix *_print functions, including interating ones for sections/bytecodes.
- Fix handling of $ and $$. Both now are properly parsed and bytecode'd.
- Print out all sections and symbol table in main() in preparation for work
on "cleanup" functionality between parser and optimizer stages (really part
of the parser stage).
projects / yasm / log