1 \# --------------------------------------------------------------------------
3 \# Copyright 1996-2013 The NASM Authors - All Rights Reserved
4 \# See the file AUTHORS included with the NASM distribution for
5 \# the specific copyright holders.
7 \# Redistribution and use in source and binary forms, with or without
8 \# modification, are permitted provided that the following
11 \# * Redistributions of source code must retain the above copyright
12 \# notice, this list of conditions and the following disclaimer.
13 \# * Redistributions in binary form must reproduce the above
14 \# copyright notice, this list of conditions and the following
15 \# disclaimer in the documentation and/or other materials provided
16 \# with the distribution.
18 \# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
19 \# CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
20 \# INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
21 \# MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 \# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23 \# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 \# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 \# NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 \# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 \# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 \# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
29 \# OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
30 \# EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 \# --------------------------------------------------------------------------
34 \# Source code to NASM documentation
36 \M{category}{Programming}
37 \M{title}{NASM - The Netwide Assembler}
39 \M{author}{The NASM Development Team}
40 \M{copyright_tail}{-- All Rights Reserved}
41 \M{license}{This document is redistributable under the license given in the file "LICENSE" distributed in the NASM archive.}
42 \M{summary}{This file documents NASM, the Netwide Assembler: an assembler targetting the Intel x86 series of processors, with portable source.}
45 \M{infotitle}{The Netwide Assembler for x86}
46 \M{epslogo}{nasmlogo.eps}
53 \IR{-MD} \c{-MD} option
54 \IR{-MF} \c{-MF} option
55 \IR{-MG} \c{-MG} option
56 \IR{-MP} \c{-MP} option
57 \IR{-MQ} \c{-MQ} option
58 \IR{-MT} \c{-MT} option
79 \IR{!=} \c{!=} operator
80 \IR{$, here} \c{$}, Here token
81 \IR{$, prefix} \c{$}, prefix
84 \IR{%%} \c{%%} operator
85 \IR{%+1} \c{%+1} and \c{%-1} syntax
87 \IR{%0} \c{%0} parameter count
89 \IR{&&} \c{&&} operator
91 \IR{..@} \c{..@} symbol prefix
93 \IR{//} \c{//} operator
95 \IR{<<} \c{<<} operator
96 \IR{<=} \c{<=} operator
97 \IR{<>} \c{<>} operator
99 \IR{==} \c{==} operator
100 \IR{>} \c{>} operator
101 \IR{>=} \c{>=} operator
102 \IR{>>} \c{>>} operator
103 \IR{?} \c{?} MASM syntax
104 \IR{^} \c{^} operator
105 \IR{^^} \c{^^} operator
106 \IR{|} \c{|} operator
107 \IR{||} \c{||} operator
108 \IR{~} \c{~} operator
109 \IR{%$} \c{%$} and \c{%$$} prefixes
111 \IR{+ opaddition} \c{+} operator, binary
112 \IR{+ opunary} \c{+} operator, unary
113 \IR{+ modifier} \c{+} modifier
114 \IR{- opsubtraction} \c{-} operator, binary
115 \IR{- opunary} \c{-} operator, unary
116 \IR{! opunary} \c{!} operator, unary
117 \IR{alignment, in bin sections} alignment, in \c{bin} sections
118 \IR{alignment, in elf sections} alignment, in \c{elf} sections
119 \IR{alignment, in win32 sections} alignment, in \c{win32} sections
120 \IR{alignment, of elf common variables} alignment, of \c{elf} common
122 \IR{alignment, in obj sections} alignment, in \c{obj} sections
123 \IR{a.out, bsd version} \c{a.out}, BSD version
124 \IR{a.out, linux version} \c{a.out}, Linux version
125 \IR{autoconf} Autoconf
127 \IR{bitwise and} bitwise AND
128 \IR{bitwise or} bitwise OR
129 \IR{bitwise xor} bitwise XOR
130 \IR{block ifs} block IFs
131 \IR{borland pascal} Borland, Pascal
132 \IR{borland's win32 compilers} Borland, Win32 compilers
133 \IR{braces, after % sign} braces, after \c{%} sign
135 \IR{c calling convention} C calling convention
136 \IR{c symbol names} C symbol names
137 \IA{critical expressions}{critical expression}
138 \IA{command line}{command-line}
139 \IA{case sensitivity}{case sensitive}
140 \IA{case-sensitive}{case sensitive}
141 \IA{case-insensitive}{case sensitive}
142 \IA{character constants}{character constant}
143 \IR{common object file format} Common Object File Format
144 \IR{common variables, alignment in elf} common variables, alignment
146 \IR{common, elf extensions to} \c{COMMON}, \c{elf} extensions to
147 \IR{common, obj extensions to} \c{COMMON}, \c{obj} extensions to
148 \IR{declaring structure} declaring structures
149 \IR{default-wrt mechanism} default-\c{WRT} mechanism
152 \IR{dll symbols, exporting} DLL symbols, exporting
153 \IR{dll symbols, importing} DLL symbols, importing
155 \IR{dos archive} DOS archive
156 \IR{dos source archive} DOS source archive
157 \IA{effective address}{effective addresses}
158 \IA{effective-address}{effective addresses}
160 \IR{elf, 16-bit code and} ELF, 16-bit code and
161 \IR{elf shared libraries} ELF, shared libraries
164 \IR{elfx32} \c{elfx32}
165 \IR{executable and linkable format} Executable and Linkable Format
166 \IR{extern, obj extensions to} \c{EXTERN}, \c{obj} extensions to
167 \IR{extern, rdf extensions to} \c{EXTERN}, \c{rdf} extensions to
168 \IR{floating-point, constants} floating-point, constants
169 \IR{floating-point, packed bcd constants} floating-point, packed BCD constants
171 \IR{freelink} FreeLink
172 \IR{functions, c calling convention} functions, C calling convention
173 \IR{functions, pascal calling convention} functions, Pascal calling
175 \IR{global, aoutb extensions to} \c{GLOBAL}, \c{aoutb} extensions to
176 \IR{global, elf extensions to} \c{GLOBAL}, \c{elf} extensions to
177 \IR{global, rdf extensions to} \c{GLOBAL}, \c{rdf} extensions to
179 \IR{got relocations} \c{GOT} relocations
180 \IR{gotoff relocation} \c{GOTOFF} relocations
181 \IR{gotpc relocation} \c{GOTPC} relocations
182 \IR{intel number formats} Intel number formats
183 \IR{linux, elf} Linux, ELF
184 \IR{linux, a.out} Linux, \c{a.out}
185 \IR{linux, as86} Linux, \c{as86}
186 \IR{logical and} logical AND
187 \IR{logical or} logical OR
188 \IR{logical xor} logical XOR
189 \IR{mach object file format} Mach, object file format
191 \IR{macho32} \c{macho32}
192 \IR{macho64} \c{macho64}
195 \IA{memory reference}{memory references}
197 \IA{misc directory}{misc subdirectory}
198 \IR{misc subdirectory} \c{misc} subdirectory
199 \IR{microsoft omf} Microsoft OMF
200 \IR{mmx registers} MMX registers
201 \IA{modr/m}{modr/m byte}
202 \IR{modr/m byte} ModR/M byte
204 \IR{ms-dos device drivers} MS-DOS device drivers
205 \IR{multipush} \c{multipush} macro
207 \IR{nasm version} NASM version
211 \IR{operating system} operating system
213 \IR{pascal calling convention}Pascal calling convention
214 \IR{passes} passes, assembly
219 \IR{plt} \c{PLT} relocations
220 \IA{pre-defining macros}{pre-define}
221 \IA{preprocessor expressions}{preprocessor, expressions}
222 \IA{preprocessor loops}{preprocessor, loops}
223 \IA{preprocessor variables}{preprocessor, variables}
224 \IA{rdoff subdirectory}{rdoff}
225 \IR{rdoff} \c{rdoff} subdirectory
226 \IR{relocatable dynamic object file format} Relocatable Dynamic
228 \IR{relocations, pic-specific} relocations, PIC-specific
229 \IA{repeating}{repeating code}
230 \IR{section alignment, in elf} section alignment, in \c{elf}
231 \IR{section alignment, in bin} section alignment, in \c{bin}
232 \IR{section alignment, in obj} section alignment, in \c{obj}
233 \IR{section alignment, in win32} section alignment, in \c{win32}
234 \IR{section, elf extensions to} \c{SECTION}, \c{elf} extensions to
235 \IR{section, win32 extensions to} \c{SECTION}, \c{win32} extensions to
236 \IR{segment alignment, in bin} segment alignment, in \c{bin}
237 \IR{segment alignment, in obj} segment alignment, in \c{obj}
238 \IR{segment, obj extensions to} \c{SEGMENT}, \c{elf} extensions to
239 \IR{segment names, borland pascal} segment names, Borland Pascal
240 \IR{shift command} \c{shift} command
242 \IR{sib byte} SIB byte
243 \IR{align, smart} \c{ALIGN}, smart
244 \IA{sectalign}{sectalign}
245 \IR{solaris x86} Solaris x86
246 \IA{standard section names}{standardized section names}
247 \IR{symbols, exporting from dlls} symbols, exporting from DLLs
248 \IR{symbols, importing from dlls} symbols, importing from DLLs
249 \IR{test subdirectory} \c{test} subdirectory
251 \IR{underscore, in c symbols} underscore, in C symbols
257 \IA{sco unix}{unix, sco}
258 \IR{unix, sco} Unix, SCO
259 \IA{unix source archive}{unix, source archive}
260 \IR{unix, source archive} Unix, source archive
261 \IA{unix system v}{unix, system v}
262 \IR{unix, system v} Unix, System V
263 \IR{unixware} UnixWare
265 \IR{version number of nasm} version number of NASM
266 \IR{visual c++} Visual C++
267 \IR{www page} WWW page
271 \IR{windows 95} Windows 95
272 \IR{windows nt} Windows NT
273 \# \IC{program entry point}{entry point, program}
274 \# \IC{program entry point}{start point, program}
275 \# \IC{MS-DOS device drivers}{device drivers, MS-DOS}
276 \# \IC{16-bit mode, versus 32-bit mode}{32-bit mode, versus 16-bit mode}
277 \# \IC{c symbol names}{symbol names, in C}
280 \C{intro} Introduction
282 \H{whatsnasm} What Is NASM?
284 The Netwide Assembler, NASM, is an 80x86 and x86-64 assembler designed
285 for portability and modularity. It supports a range of object file
286 formats, including Linux and \c{*BSD} \c{a.out}, \c{ELF}, \c{COFF},
287 \c{Mach-O}, Microsoft 16-bit \c{OBJ}, \c{Win32} and \c{Win64}. It will
288 also output plain binary files. Its syntax is designed to be simple
289 and easy to understand, similar to Intel's but less complex. It
290 supports all currently known x86 architectural extensions, and has
291 strong support for macros.
294 \S{yaasm} Why Yet Another Assembler?
296 The Netwide Assembler grew out of an idea on \i\c{comp.lang.asm.x86}
297 (or possibly \i\c{alt.lang.asm} - I forget which), which was
298 essentially that there didn't seem to be a good \e{free} x86-series
299 assembler around, and that maybe someone ought to write one.
301 \b \i\c{a86} is good, but not free, and in particular you don't get any
302 32-bit capability until you pay. It's DOS only, too.
304 \b \i\c{gas} is free, and ports over to DOS and Unix, but it's not
305 very good, since it's designed to be a back end to \i\c{gcc}, which
306 always feeds it correct code. So its error checking is minimal. Also,
307 its syntax is horrible, from the point of view of anyone trying to
308 actually \e{write} anything in it. Plus you can't write 16-bit code in
311 \b \i\c{as86} is specific to Minix and Linux, and (my version at least)
312 doesn't seem to have much (or any) documentation.
314 \b \i\c{MASM} isn't very good, and it's (was) expensive, and it runs only under
317 \b \i\c{TASM} is better, but still strives for MASM compatibility,
318 which means millions of directives and tons of red tape. And its syntax
319 is essentially MASM's, with the contradictions and quirks that
320 entails (although it sorts out some of those by means of Ideal mode.)
321 It's expensive too. And it's DOS-only.
323 So here, for your coding pleasure, is NASM. At present it's
324 still in prototype stage - we don't promise that it can outperform
325 any of these assemblers. But please, \e{please} send us bug reports,
326 fixes, helpful information, and anything else you can get your hands
327 on (and thanks to the many people who've done this already! You all
328 know who you are), and we'll improve it out of all recognition.
332 \S{legal} \i{License} Conditions
334 Please see the file \c{LICENSE}, supplied as part of any NASM
335 distribution archive, for the license conditions under which you may
336 use NASM. NASM is now under the so-called 2-clause BSD license, also
337 known as the simplified BSD license.
339 Copyright 1996-2011 the NASM Authors - All rights reserved.
341 Redistribution and use in source and binary forms, with or without
342 modification, are permitted provided that the following conditions are
345 \b Redistributions of source code must retain the above copyright
346 notice, this list of conditions and the following disclaimer.
348 \b Redistributions in binary form must reproduce the above copyright
349 notice, this list of conditions and the following disclaimer in the
350 documentation and/or other materials provided with the distribution.
352 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
353 CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
354 INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
355 MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
356 DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
357 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
358 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
359 NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
360 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
361 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
362 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
363 OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
364 EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
367 \H{contact} Contact Information
369 The current version of NASM (since about 0.98.08) is maintained by a
370 team of developers, accessible through the \c{nasm-devel} mailing list
371 (see below for the link).
372 If you want to report a bug, please read \k{bugs} first.
374 NASM has a \i{website} at
375 \W{http://www.nasm.us/}\c{http://www.nasm.us/}. If it's not there,
378 \i{New releases}, \i{release candidates}, and \I{snapshots, daily
379 development}\i{daily development snapshots} of NASM are available from
380 the official web site.
382 Announcements are posted to
383 \W{news:comp.lang.asm.x86}\i\c{comp.lang.asm.x86},
385 \W{http://www.freshmeat.net/}\c{http://www.freshmeat.net/}.
387 If you want information about the current development status, please
388 subscribe to the \i\c{nasm-devel} email list; see link from the
392 \H{install} Installation
394 \S{instdos} \i{Installing} NASM under MS-\i{DOS} or Windows
396 Once you've obtained the appropriate archive for NASM,
397 \i\c{nasm-XXX-dos.zip} or \i\c{nasm-XXX-win32.zip} (where \c{XXX}
398 denotes the version number of NASM contained in the archive), unpack
399 it into its own directory (for example \c{c:\\nasm}).
401 The archive will contain a set of executable files: the NASM
402 executable file \i\c{nasm.exe}, the NDISASM executable file
403 \i\c{ndisasm.exe}, and possibly additional utilities to handle the
406 The only file NASM needs to run is its own executable, so copy
407 \c{nasm.exe} to a directory on your PATH, or alternatively edit
408 \i\c{autoexec.bat} to add the \c{nasm} directory to your
409 \i\c{PATH} (to do that under Windows XP, go to Start > Control Panel >
410 System > Advanced > Environment Variables; these instructions may work
411 under other versions of Windows as well.)
413 That's it - NASM is installed. You don't need the nasm directory
414 to be present to run NASM (unless you've added it to your \c{PATH}),
415 so you can delete it if you need to save space; however, you may
416 want to keep the documentation or test programs.
418 If you've downloaded the \i{DOS source archive}, \i\c{nasm-XXX.zip},
419 the \c{nasm} directory will also contain the full NASM \i{source
420 code}, and a selection of \i{Makefiles} you can (hopefully) use to
421 rebuild your copy of NASM from scratch. See the file \c{INSTALL} in
424 Note that a number of files are generated from other files by Perl
425 scripts. Although the NASM source distribution includes these
426 generated files, you will need to rebuild them (and hence, will need a
427 Perl interpreter) if you change insns.dat, standard.mac or the
428 documentation. It is possible future source distributions may not
429 include these files at all. Ports of \i{Perl} for a variety of
430 platforms, including DOS and Windows, are available from
431 \W{http://www.cpan.org/ports/}\i{www.cpan.org}.
434 \S{instdos} Installing NASM under \i{Unix}
436 Once you've obtained the \i{Unix source archive} for NASM,
437 \i\c{nasm-XXX.tar.gz} (where \c{XXX} denotes the version number of
438 NASM contained in the archive), unpack it into a directory such
439 as \c{/usr/local/src}. The archive, when unpacked, will create its
440 own subdirectory \c{nasm-XXX}.
442 NASM is an \I{Autoconf}\I\c{configure}auto-configuring package: once
443 you've unpacked it, \c{cd} to the directory it's been unpacked into
444 and type \c{./configure}. This shell script will find the best C
445 compiler to use for building NASM and set up \i{Makefiles}
448 Once NASM has auto-configured, you can type \i\c{make} to build the
449 \c{nasm} and \c{ndisasm} binaries, and then \c{make install} to
450 install them in \c{/usr/local/bin} and install the \i{man pages}
451 \i\c{nasm.1} and \i\c{ndisasm.1} in \c{/usr/local/man/man1}.
452 Alternatively, you can give options such as \c{--prefix} to the
453 configure script (see the file \i\c{INSTALL} for more details), or
454 install the programs yourself.
456 NASM also comes with a set of utilities for handling the \c{RDOFF}
457 custom object-file format, which are in the \i\c{rdoff} subdirectory
458 of the NASM archive. You can build these with \c{make rdf} and
459 install them with \c{make rdf_install}, if you want them.
462 \C{running} Running NASM
464 \H{syntax} NASM \i{Command-Line} Syntax
466 To assemble a file, you issue a command of the form
468 \c nasm -f <format> <filename> [-o <output>]
472 \c nasm -f elf myfile.asm
474 will assemble \c{myfile.asm} into an \c{ELF} object file \c{myfile.o}. And
476 \c nasm -f bin myfile.asm -o myfile.com
478 will assemble \c{myfile.asm} into a raw binary file \c{myfile.com}.
480 To produce a listing file, with the hex codes output from NASM
481 displayed on the left of the original sources, use the \c{-l} option
482 to give a listing file name, for example:
484 \c nasm -f coff myfile.asm -l myfile.lst
486 To get further usage instructions from NASM, try typing
490 As \c{-hf}, this will also list the available output file formats, and what they
493 If you use Linux but aren't sure whether your system is \c{a.out}
498 (in the directory in which you put the NASM binary when you
499 installed it). If it says something like
501 \c nasm: ELF 32-bit LSB executable i386 (386 and up) Version 1
503 then your system is \c{ELF}, and you should use the option \c{-f elf}
504 when you want NASM to produce Linux object files. If it says
506 \c nasm: Linux/i386 demand-paged executable (QMAGIC)
508 or something similar, your system is \c{a.out}, and you should use
509 \c{-f aout} instead (Linux \c{a.out} systems have long been obsolete,
510 and are rare these days.)
512 Like Unix compilers and assemblers, NASM is silent unless it
513 goes wrong: you won't see any output at all, unless it gives error
517 \S{opt-o} The \i\c{-o} Option: Specifying the Output File Name
519 NASM will normally choose the name of your output file for you;
520 precisely how it does this is dependent on the object file format.
521 For Microsoft object file formats (\c{obj}, \c{win32} and \c{win64}),
522 it will remove the \c{.asm} \i{extension} (or whatever extension you
523 like to use - NASM doesn't care) from your source file name and
524 substitute \c{.obj}. For Unix object file formats (\c{aout}, \c{as86},
525 \c{coff}, \c{elf32}, \c{elf64}, \c{elfx32}, \c{ieee}, \c{macho32} and
526 \c{macho64}) it will substitute \c{.o}. For \c{dbg}, \c{rdf}, \c{ith}
527 and \c{srec}, it will use \c{.dbg}, \c{.rdf}, \c{.ith} and \c{.srec},
528 respectively, and for the \c{bin} format it will simply remove the
529 extension, so that \c{myfile.asm} produces the output file \c{myfile}.
531 If the output file already exists, NASM will overwrite it, unless it
532 has the same name as the input file, in which case it will give a
533 warning and use \i\c{nasm.out} as the output file name instead.
535 For situations in which this behaviour is unacceptable, NASM
536 provides the \c{-o} command-line option, which allows you to specify
537 your desired output file name. You invoke \c{-o} by following it
538 with the name you wish for the output file, either with or without
539 an intervening space. For example:
541 \c nasm -f bin program.asm -o program.com
542 \c nasm -f bin driver.asm -odriver.sys
544 Note that this is a small o, and is different from a capital O , which
545 is used to specify the number of optimisation passes required. See \k{opt-O}.
548 \S{opt-f} The \i\c{-f} Option: Specifying the \i{Output File Format}
550 If you do not supply the \c{-f} option to NASM, it will choose an
551 output file format for you itself. In the distribution versions of
552 NASM, the default is always \i\c{bin}; if you've compiled your own
553 copy of NASM, you can redefine \i\c{OF_DEFAULT} at compile time and
554 choose what you want the default to be.
556 Like \c{-o}, the intervening space between \c{-f} and the output
557 file format is optional; so \c{-f elf} and \c{-felf} are both valid.
559 A complete list of the available output file formats can be given by
560 issuing the command \i\c{nasm -hf}.
563 \S{opt-l} The \i\c{-l} Option: Generating a \i{Listing File}
565 If you supply the \c{-l} option to NASM, followed (with the usual
566 optional space) by a file name, NASM will generate a
567 \i{source-listing file} for you, in which addresses and generated
568 code are listed on the left, and the actual source code, with
569 expansions of multi-line macros (except those which specifically
570 request no expansion in source listings: see \k{nolist}) on the
573 \c nasm -f elf myfile.asm -l myfile.lst
575 If a list file is selected, you may turn off listing for a
576 section of your source with \c{[list -]}, and turn it back on
577 with \c{[list +]}, (the default, obviously). There is no "user
578 form" (without the brackets). This can be used to list only
579 sections of interest, avoiding excessively long listings.
582 \S{opt-M} The \i\c{-M} Option: Generate \i{Makefile Dependencies}
584 This option can be used to generate makefile dependencies on stdout.
585 This can be redirected to a file for further processing. For example:
587 \c nasm -M myfile.asm > myfile.dep
590 \S{opt-MG} The \i\c{-MG} Option: Generate \i{Makefile Dependencies}
592 This option can be used to generate makefile dependencies on stdout.
593 This differs from the \c{-M} option in that if a nonexisting file is
594 encountered, it is assumed to be a generated file and is added to the
595 dependency list without a prefix.
598 \S{opt-MF} The \i\c\{-MF} Option: Set Makefile Dependency File
600 This option can be used with the \c{-M} or \c{-MG} options to send the
601 output to a file, rather than to stdout. For example:
603 \c nasm -M -MF myfile.dep myfile.asm
606 \S{opt-MD} The \i\c{-MD} Option: Assemble and Generate Dependencies
608 The \c{-MD} option acts as the combination of the \c{-M} and \c{-MF}
609 options (i.e. a filename has to be specified.) However, unlike the
610 \c{-M} or \c{-MG} options, \c{-MD} does \e{not} inhibit the normal
611 operation of the assembler. Use this to automatically generate
612 updated dependencies with every assembly session. For example:
614 \c nasm -f elf -o myfile.o -MD myfile.dep myfile.asm
617 \S{opt-MT} The \i\c{-MT} Option: Dependency Target Name
619 The \c{-MT} option can be used to override the default name of the
620 dependency target. This is normally the same as the output filename,
621 specified by the \c{-o} option.
624 \S{opt-MQ} The \i\c{-MQ} Option: Dependency Target Name (Quoted)
626 The \c{-MQ} option acts as the \c{-MT} option, except it tries to
627 quote characters that have special meaning in Makefile syntax. This
628 is not foolproof, as not all characters with special meaning are
629 quotable in Make. The default output (if no \c{-MT} or \c{-MQ} option
630 is specified) is automatically quoted.
633 \S{opt-MP} The \i\c{-MP} Option: Emit phony targets
635 When used with any of the dependency generation options, the \c{-MP}
636 option causes NASM to emit a phony target without dependencies for
637 each header file. This prevents Make from complaining if a header
638 file has been removed.
641 \S{opt-F} The \i\c{-F} Option: Selecting a \i{Debug Information Format}
643 This option is used to select the format of the debug information
644 emitted into the output file, to be used by a debugger (or \e{will}
645 be). Prior to version 2.03.01, the use of this switch did \e{not} enable
646 output of the selected debug info format. Use \c{-g}, see \k{opt-g},
647 to enable output. Versions 2.03.01 and later automatically enable \c{-g}
648 if \c{-F} is specified.
650 A complete list of the available debug file formats for an output
651 format can be seen by issuing the command \c{nasm -f <format> -y}. Not
652 all output formats currently support debugging output. See \k{opt-y}.
654 This should not be confused with the \c{-f dbg} output format option which
655 is not built into NASM by default. For information on how
656 to enable it when building from the sources, see \k{dbgfmt}.
659 \S{opt-g} The \i\c{-g} Option: Enabling \i{Debug Information}.
661 This option can be used to generate debugging information in the specified
662 format. See \k{opt-F}. Using \c{-g} without \c{-F} results in emitting
663 debug info in the default format, if any, for the selected output format.
664 If no debug information is currently implemented in the selected output
665 format, \c{-g} is \e{silently ignored}.
668 \S{opt-X} The \i\c{-X} Option: Selecting an \i{Error Reporting Format}
670 This option can be used to select an error reporting format for any
671 error messages that might be produced by NASM.
673 Currently, two error reporting formats may be selected. They are
674 the \c{-Xvc} option and the \c{-Xgnu} option. The GNU format is
675 the default and looks like this:
677 \c filename.asm:65: error: specific error message
679 where \c{filename.asm} is the name of the source file in which the
680 error was detected, \c{65} is the source file line number on which
681 the error was detected, \c{error} is the severity of the error (this
682 could be \c{warning}), and \c{specific error message} is a more
683 detailed text message which should help pinpoint the exact problem.
685 The other format, specified by \c{-Xvc} is the style used by Microsoft
686 Visual C++ and some other programs. It looks like this:
688 \c filename.asm(65) : error: specific error message
690 where the only difference is that the line number is in parentheses
691 instead of being delimited by colons.
693 See also the \c{Visual C++} output format, \k{win32fmt}.
695 \S{opt-Z} The \i\c{-Z} Option: Send Errors to a File
697 Under \I{DOS}\c{MS-DOS} it can be difficult (though there are ways) to
698 redirect the standard-error output of a program to a file. Since
699 NASM usually produces its warning and \i{error messages} on
700 \i\c{stderr}, this can make it hard to capture the errors if (for
701 example) you want to load them into an editor.
703 NASM therefore provides the \c{-Z} option, taking a filename argument
704 which causes errors to be sent to the specified files rather than
705 standard error. Therefore you can \I{redirecting errors}redirect
706 the errors into a file by typing
708 \c nasm -Z myfile.err -f obj myfile.asm
710 In earlier versions of NASM, this option was called \c{-E}, but it was
711 changed since \c{-E} is an option conventionally used for
712 preprocessing only, with disastrous results. See \k{opt-E}.
714 \S{opt-s} The \i\c{-s} Option: Send Errors to \i\c{stdout}
716 The \c{-s} option redirects \i{error messages} to \c{stdout} rather
717 than \c{stderr}, so it can be redirected under \I{DOS}\c{MS-DOS}. To
718 assemble the file \c{myfile.asm} and pipe its output to the \c{more}
719 program, you can type:
721 \c nasm -s -f obj myfile.asm | more
723 See also the \c{-Z} option, \k{opt-Z}.
726 \S{opt-i} The \i\c{-i}\I\c{-I} Option: Include File Search Directories
728 When NASM sees the \i\c{%include} or \i\c{%pathsearch} directive in a
729 source file (see \k{include}, \k{pathsearch} or \k{incbin}), it will
730 search for the given file not only in the current directory, but also
731 in any directories specified on the command line by the use of the
732 \c{-i} option. Therefore you can include files from a \i{macro
733 library}, for example, by typing
735 \c nasm -ic:\macrolib\ -f obj myfile.asm
737 (As usual, a space between \c{-i} and the path name is allowed, and
740 NASM, in the interests of complete source-code portability, does not
741 understand the file naming conventions of the OS it is running on;
742 the string you provide as an argument to the \c{-i} option will be
743 prepended exactly as written to the name of the include file.
744 Therefore the trailing backslash in the above example is necessary.
745 Under Unix, a trailing forward slash is similarly necessary.
747 (You can use this to your advantage, if you're really \i{perverse},
748 by noting that the option \c{-ifoo} will cause \c{%include "bar.i"}
749 to search for the file \c{foobar.i}...)
751 If you want to define a \e{standard} \i{include search path},
752 similar to \c{/usr/include} on Unix systems, you should place one or
753 more \c{-i} directives in the \c{NASMENV} environment variable (see
756 For Makefile compatibility with many C compilers, this option can also
757 be specified as \c{-I}.
760 \S{opt-p} The \i\c{-p}\I\c{-P} Option: \I{pre-including files}Pre-Include a File
762 \I\c{%include}NASM allows you to specify files to be
763 \e{pre-included} into your source file, by the use of the \c{-p}
766 \c nasm myfile.asm -p myinc.inc
768 is equivalent to running \c{nasm myfile.asm} and placing the
769 directive \c{%include "myinc.inc"} at the start of the file.
771 For consistency with the \c{-I}, \c{-D} and \c{-U} options, this
772 option can also be specified as \c{-P}.
775 \S{opt-d} The \i\c{-d}\I\c{-D} Option: \I{pre-defining macros}Pre-Define a Macro
777 \I\c{%define}Just as the \c{-p} option gives an alternative to placing
778 \c{%include} directives at the start of a source file, the \c{-d}
779 option gives an alternative to placing a \c{%define} directive. You
782 \c nasm myfile.asm -dFOO=100
784 as an alternative to placing the directive
788 at the start of the file. You can miss off the macro value, as well:
789 the option \c{-dFOO} is equivalent to coding \c{%define FOO}. This
790 form of the directive may be useful for selecting \i{assembly-time
791 options} which are then tested using \c{%ifdef}, for example
794 For Makefile compatibility with many C compilers, this option can also
795 be specified as \c{-D}.
798 \S{opt-u} The \i\c{-u}\I\c{-U} Option: \I{Undefining macros}Undefine a Macro
800 \I\c{%undef}The \c{-u} option undefines a macro that would otherwise
801 have been pre-defined, either automatically or by a \c{-p} or \c{-d}
802 option specified earlier on the command lines.
804 For example, the following command line:
806 \c nasm myfile.asm -dFOO=100 -uFOO
808 would result in \c{FOO} \e{not} being a predefined macro in the
809 program. This is useful to override options specified at a different
812 For Makefile compatibility with many C compilers, this option can also
813 be specified as \c{-U}.
816 \S{opt-E} The \i\c{-E}\I{-e} Option: Preprocess Only
818 NASM allows the \i{preprocessor} to be run on its own, up to a
819 point. Using the \c{-E} option (which requires no arguments) will
820 cause NASM to preprocess its input file, expand all the macro
821 references, remove all the comments and preprocessor directives, and
822 print the resulting file on standard output (or save it to a file,
823 if the \c{-o} option is also used).
825 This option cannot be applied to programs which require the
826 preprocessor to evaluate \I{preprocessor expressions}\i{expressions}
827 which depend on the values of symbols: so code such as
829 \c %assign tablesize ($-tablestart)
831 will cause an error in \i{preprocess-only mode}.
833 For compatiblity with older version of NASM, this option can also be
834 written \c{-e}. \c{-E} in older versions of NASM was the equivalent
835 of the current \c{-Z} option, \k{opt-Z}.
837 \S{opt-a} The \i\c{-a} Option: Don't Preprocess At All
839 If NASM is being used as the back end to a compiler, it might be
840 desirable to \I{suppressing preprocessing}suppress preprocessing
841 completely and assume the compiler has already done it, to save time
842 and increase compilation speeds. The \c{-a} option, requiring no
843 argument, instructs NASM to replace its powerful \i{preprocessor}
844 with a \i{stub preprocessor} which does nothing.
847 \S{opt-O} The \i\c{-O} Option: Specifying \i{Multipass Optimization}
849 Using the \c{-O} option, you can tell NASM to carry out different
850 levels of optimization. The syntax is:
852 \b \c{-O0}: No optimization. All operands take their long forms,
853 if a short form is not specified, except conditional jumps.
854 This is intended to match NASM 0.98 behavior.
856 \b \c{-O1}: Minimal optimization. As above, but immediate operands
857 which will fit in a signed byte are optimized,
858 unless the long form is specified. Conditional jumps default
859 to the long form unless otherwise specified.
861 \b \c{-Ox} (where \c{x} is the actual letter \c{x}): Multipass optimization.
862 Minimize branch offsets and signed immediate bytes,
863 overriding size specification unless the \c{strict} keyword
864 has been used (see \k{strict}). For compatibility with earlier
865 releases, the letter \c{x} may also be any number greater than
866 one. This number has no effect on the actual number of passes.
868 The \c{-Ox} mode is recommended for most uses, and is the default
871 Note that this is a capital \c{O}, and is different from a small \c{o}, which
872 is used to specify the output file name. See \k{opt-o}.
875 \S{opt-t} The \i\c{-t} Option: Enable TASM Compatibility Mode
877 NASM includes a limited form of compatibility with Borland's \i\c{TASM}.
878 When NASM's \c{-t} option is used, the following changes are made:
880 \b local labels may be prefixed with \c{@@} instead of \c{.}
882 \b size override is supported within brackets. In TASM compatible mode,
883 a size override inside square brackets changes the size of the operand,
884 and not the address type of the operand as it does in NASM syntax. E.g.
885 \c{mov eax,[DWORD val]} is valid syntax in TASM compatibility mode.
886 Note that you lose the ability to override the default address type for
889 \b unprefixed forms of some directives supported (\c{arg}, \c{elif},
890 \c{else}, \c{endif}, \c{if}, \c{ifdef}, \c{ifdifi}, \c{ifndef},
891 \c{include}, \c{local})
893 \S{opt-w} The \i\c{-w} and \i\c{-W} Options: Enable or Disable Assembly \i{Warnings}
895 NASM can observe many conditions during the course of assembly which
896 are worth mentioning to the user, but not a sufficiently severe
897 error to justify NASM refusing to generate an output file. These
898 conditions are reported like errors, but come up with the word
899 `warning' before the message. Warnings do not prevent NASM from
900 generating an output file and returning a success status to the
903 Some conditions are even less severe than that: they are only
904 sometimes worth mentioning to the user. Therefore NASM supports the
905 \c{-w} command-line option, which enables or disables certain
906 classes of assembly warning. Such warning classes are described by a
907 name, for example \c{orphan-labels}; you can enable warnings of
908 this class by the command-line option \c{-w+orphan-labels} and
909 disable it by \c{-w-orphan-labels}.
911 The \i{suppressible warning} classes are:
913 \b \i\c{macro-params} covers warnings about \i{multi-line macros}
914 being invoked with the wrong number of parameters. This warning
915 class is enabled by default; see \k{mlmacover} for an example of why
916 you might want to disable it.
918 \b \i\c{macro-selfref} warns if a macro references itself. This
919 warning class is disabled by default.
921 \b\i\c{macro-defaults} warns when a macro has more default
922 parameters than optional parameters. This warning class
923 is enabled by default; see \k{mlmacdef} for why you might want to disable it.
925 \b \i\c{orphan-labels} covers warnings about source lines which
926 contain no instruction but define a label without a trailing colon.
927 NASM warns about this somewhat obscure condition by default;
928 see \k{syntax} for more information.
930 \b \i\c{number-overflow} covers warnings about numeric constants which
931 don't fit in 64 bits. This warning class is enabled by default.
933 \b \i\c{gnu-elf-extensions} warns if 8-bit or 16-bit relocations
934 are used in \c{-f elf} format. The GNU extensions allow this.
935 This warning class is disabled by default.
937 \b \i\c{float-overflow} warns about floating point overflow.
940 \b \i\c{float-denorm} warns about floating point denormals.
943 \b \i\c{float-underflow} warns about floating point underflow.
946 \b \i\c{float-toolong} warns about too many digits in floating-point numbers.
949 \b \i\c{user} controls \c{%warning} directives (see \k{pperror}).
952 \b \i\c{lock} warns about \c{LOCK} prefixes on unlockable instructions.
955 \b \i\c{hle} warns about invalid use of the HLE \c{XACQUIRE} or \c{XRELEASE}
959 \b \i\c{bnd} warns about ineffective use of the \c{BND} prefix when a relaxed
960 form of jmp instruction becomes jmp short form.
963 \b \i\c{error} causes warnings to be treated as errors. Disabled by
966 \b \i\c{all} is an alias for \e{all} suppressible warning classes (not
967 including \c{error}). Thus, \c{-w+all} enables all available warnings.
969 In addition, you can set warning classes across sections.
970 Warning classes may be enabled with \i\c{[warning +warning-name]},
971 disabled with \i\c{[warning -warning-name]} or reset to their
972 original value with \i\c{[warning *warning-name]}. No "user form"
973 (without the brackets) exists.
975 Since version 2.00, NASM has also supported the gcc-like syntax
976 \c{-Wwarning} and \c{-Wno-warning} instead of \c{-w+warning} and
977 \c{-w-warning}, respectively.
980 \S{opt-v} The \i\c{-v} Option: Display \i{Version} Info
982 Typing \c{NASM -v} will display the version of NASM which you are using,
983 and the date on which it was compiled.
985 You will need the version number if you report a bug.
987 \S{opt-y} The \i\c{-y} Option: Display Available Debug Info Formats
989 Typing \c{nasm -f <option> -y} will display a list of the available
990 debug info formats for the given output format. The default format
991 is indicated by an asterisk. For example:
995 \c valid debug formats for 'elf32' output format are
996 \c ('*' denotes default):
997 \c * stabs ELF32 (i386) stabs debug format for Linux
998 \c dwarf elf32 (i386) dwarf debug format for Linux
1001 \S{opt-pfix} The \i\c{--prefix} and \i\c{--postfix} Options.
1003 The \c{--prefix} and \c{--postfix} options prepend or append
1004 (respectively) the given argument to all \c{global} or
1005 \c{extern} variables. E.g. \c{--prefix _} will prepend the
1006 underscore to all global and external variables, as C sometimes
1007 (but not always) likes it.
1010 \S{nasmenv} The \i\c{NASMENV} \i{Environment} Variable
1012 If you define an environment variable called \c{NASMENV}, the program
1013 will interpret it as a list of extra command-line options, which are
1014 processed before the real command line. You can use this to define
1015 standard search directories for include files, by putting \c{-i}
1016 options in the \c{NASMENV} variable.
1018 The value of the variable is split up at white space, so that the
1019 value \c{-s -ic:\\nasmlib\\} will be treated as two separate options.
1020 However, that means that the value \c{-dNAME="my name"} won't do
1021 what you might want, because it will be split at the space and the
1022 NASM command-line processing will get confused by the two
1023 nonsensical words \c{-dNAME="my} and \c{name"}.
1025 To get round this, NASM provides a feature whereby, if you begin the
1026 \c{NASMENV} environment variable with some character that isn't a minus
1027 sign, then NASM will treat this character as the \i{separator
1028 character} for options. So setting the \c{NASMENV} variable to the
1029 value \c{!-s!-ic:\\nasmlib\\} is equivalent to setting it to \c{-s
1030 -ic:\\nasmlib\\}, but \c{!-dNAME="my name"} will work.
1032 This environment variable was previously called \c{NASM}. This was
1033 changed with version 0.98.31.
1036 \H{qstart} \i{Quick Start} for \i{MASM} Users
1038 If you're used to writing programs with MASM, or with \i{TASM} in
1039 MASM-compatible (non-Ideal) mode, or with \i\c{a86}, this section
1040 attempts to outline the major differences between MASM's syntax and
1041 NASM's. If you're not already used to MASM, it's probably worth
1042 skipping this section.
1045 \S{qscs} NASM Is \I{case sensitivity}Case-Sensitive
1047 One simple difference is that NASM is case-sensitive. It makes a
1048 difference whether you call your label \c{foo}, \c{Foo} or \c{FOO}.
1049 If you're assembling to \c{DOS} or \c{OS/2} \c{.OBJ} files, you can
1050 invoke the \i\c{UPPERCASE} directive (documented in \k{objfmt}) to
1051 ensure that all symbols exported to other code modules are forced
1052 to be upper case; but even then, \e{within} a single module, NASM
1053 will distinguish between labels differing only in case.
1056 \S{qsbrackets} NASM Requires \i{Square Brackets} For \i{Memory References}
1058 NASM was designed with simplicity of syntax in mind. One of the
1059 \i{design goals} of NASM is that it should be possible, as far as is
1060 practical, for the user to look at a single line of NASM code
1061 and tell what opcode is generated by it. You can't do this in MASM:
1062 if you declare, for example,
1067 then the two lines of code
1072 generate completely different opcodes, despite having
1073 identical-looking syntaxes.
1075 NASM avoids this undesirable situation by having a much simpler
1076 syntax for memory references. The rule is simply that any access to
1077 the \e{contents} of a memory location requires square brackets
1078 around the address, and any access to the \e{address} of a variable
1079 doesn't. So an instruction of the form \c{mov ax,foo} will
1080 \e{always} refer to a compile-time constant, whether it's an \c{EQU}
1081 or the address of a variable; and to access the \e{contents} of the
1082 variable \c{bar}, you must code \c{mov ax,[bar]}.
1084 This also means that NASM has no need for MASM's \i\c{OFFSET}
1085 keyword, since the MASM code \c{mov ax,offset bar} means exactly the
1086 same thing as NASM's \c{mov ax,bar}. If you're trying to get
1087 large amounts of MASM code to assemble sensibly under NASM, you
1088 can always code \c{%idefine offset} to make the preprocessor treat
1089 the \c{OFFSET} keyword as a no-op.
1091 This issue is even more confusing in \i\c{a86}, where declaring a
1092 label with a trailing colon defines it to be a `label' as opposed to
1093 a `variable' and causes \c{a86} to adopt NASM-style semantics; so in
1094 \c{a86}, \c{mov ax,var} has different behaviour depending on whether
1095 \c{var} was declared as \c{var: dw 0} (a label) or \c{var dw 0} (a
1096 word-size variable). NASM is very simple by comparison:
1097 \e{everything} is a label.
1099 NASM, in the interests of simplicity, also does not support the
1100 \i{hybrid syntaxes} supported by MASM and its clones, such as
1101 \c{mov ax,table[bx]}, where a memory reference is denoted by one
1102 portion outside square brackets and another portion inside. The
1103 correct syntax for the above is \c{mov ax,[table+bx]}. Likewise,
1104 \c{mov ax,es:[di]} is wrong and \c{mov ax,[es:di]} is right.
1107 \S{qstypes} NASM Doesn't Store \i{Variable Types}
1109 NASM, by design, chooses not to remember the types of variables you
1110 declare. Whereas MASM will remember, on seeing \c{var dw 0}, that
1111 you declared \c{var} as a word-size variable, and will then be able
1112 to fill in the \i{ambiguity} in the size of the instruction \c{mov
1113 var,2}, NASM will deliberately remember nothing about the symbol
1114 \c{var} except where it begins, and so you must explicitly code
1115 \c{mov word [var],2}.
1117 For this reason, NASM doesn't support the \c{LODS}, \c{MOVS},
1118 \c{STOS}, \c{SCAS}, \c{CMPS}, \c{INS}, or \c{OUTS} instructions,
1119 but only supports the forms such as \c{LODSB}, \c{MOVSW}, and
1120 \c{SCASD}, which explicitly specify the size of the components of
1121 the strings being manipulated.
1124 \S{qsassume} NASM Doesn't \i\c{ASSUME}
1126 As part of NASM's drive for simplicity, it also does not support the
1127 \c{ASSUME} directive. NASM will not keep track of what values you
1128 choose to put in your segment registers, and will never
1129 \e{automatically} generate a \i{segment override} prefix.
1132 \S{qsmodel} NASM Doesn't Support \i{Memory Models}
1134 NASM also does not have any directives to support different 16-bit
1135 memory models. The programmer has to keep track of which functions
1136 are supposed to be called with a \i{far call} and which with a
1137 \i{near call}, and is responsible for putting the correct form of
1138 \c{RET} instruction (\c{RETN} or \c{RETF}; NASM accepts \c{RET}
1139 itself as an alternate form for \c{RETN}); in addition, the
1140 programmer is responsible for coding CALL FAR instructions where
1141 necessary when calling \e{external} functions, and must also keep
1142 track of which external variable definitions are far and which are
1146 \S{qsfpu} \i{Floating-Point} Differences
1148 NASM uses different names to refer to floating-point registers from
1149 MASM: where MASM would call them \c{ST(0)}, \c{ST(1)} and so on, and
1150 \i\c{a86} would call them simply \c{0}, \c{1} and so on, NASM
1151 chooses to call them \c{st0}, \c{st1} etc.
1153 As of version 0.96, NASM now treats the instructions with
1154 \i{`nowait'} forms in the same way as MASM-compatible assemblers.
1155 The idiosyncratic treatment employed by 0.95 and earlier was based
1156 on a misunderstanding by the authors.
1159 \S{qsother} Other Differences
1161 For historical reasons, NASM uses the keyword \i\c{TWORD} where MASM
1162 and compatible assemblers use \i\c{TBYTE}.
1164 NASM does not declare \i{uninitialized storage} in the same way as
1165 MASM: where a MASM programmer might use \c{stack db 64 dup (?)},
1166 NASM requires \c{stack resb 64}, intended to be read as `reserve 64
1167 bytes'. For a limited amount of compatibility, since NASM treats
1168 \c{?} as a valid character in symbol names, you can code \c{? equ 0}
1169 and then writing \c{dw ?} will at least do something vaguely useful.
1170 \I\c{RESB}\i\c{DUP} is still not a supported syntax, however.
1172 In addition to all of this, macros and directives work completely
1173 differently to MASM. See \k{preproc} and \k{directive} for further
1177 \C{lang} The NASM Language
1179 \H{syntax} Layout of a NASM Source Line
1181 Like most assemblers, each NASM source line contains (unless it
1182 is a macro, a preprocessor directive or an assembler directive: see
1183 \k{preproc} and \k{directive}) some combination of the four fields
1185 \c label: instruction operands ; comment
1187 As usual, most of these fields are optional; the presence or absence
1188 of any combination of a label, an instruction and a comment is allowed.
1189 Of course, the operand field is either required or forbidden by the
1190 presence and nature of the instruction field.
1192 NASM uses backslash (\\) as the line continuation character; if a line
1193 ends with backslash, the next line is considered to be a part of the
1194 backslash-ended line.
1196 NASM places no restrictions on white space within a line: labels may
1197 have white space before them, or instructions may have no space
1198 before them, or anything. The \i{colon} after a label is also
1199 optional. (Note that this means that if you intend to code \c{lodsb}
1200 alone on a line, and type \c{lodab} by accident, then that's still a
1201 valid source line which does nothing but define a label. Running
1202 NASM with the command-line option
1203 \I{orphan-labels}\c{-w+orphan-labels} will cause it to warn you if
1204 you define a label alone on a line without a \i{trailing colon}.)
1206 \i{Valid characters} in labels are letters, numbers, \c{_}, \c{$},
1207 \c{#}, \c{@}, \c{~}, \c{.}, and \c{?}. The only characters which may
1208 be used as the \e{first} character of an identifier are letters,
1209 \c{.} (with special meaning: see \k{locallab}), \c{_} and \c{?}.
1210 An identifier may also be prefixed with a \I{$, prefix}\c{$} to
1211 indicate that it is intended to be read as an identifier and not a
1212 reserved word; thus, if some other module you are linking with
1213 defines a symbol called \c{eax}, you can refer to \c{$eax} in NASM
1214 code to distinguish the symbol from the register. Maximum length of
1215 an identifier is 4095 characters.
1217 The instruction field may contain any machine instruction: Pentium
1218 and P6 instructions, FPU instructions, MMX instructions and even
1219 undocumented instructions are all supported. The instruction may be
1220 prefixed by \c{LOCK}, \c{REP}, \c{REPE}/\c{REPZ}, \c{REPNE}/\c{REPNZ},
1221 \c{XACQUIRE}/\c{XRELEASE} or \c{BND}/\c{NOBND}, in the usual way. Explicit
1222 \I{address-size prefixes}address-size and \i{operand-size prefixes} \i\c{A16},
1223 \i\c{A32}, \i\c{A64}, \i\c{O16} and \i\c{O32}, \i\c{O64} are provided - one example of their use
1224 is given in \k{mixsize}. You can also use the name of a \I{segment
1225 override}segment register as an instruction prefix: coding
1226 \c{es mov [bx],ax} is equivalent to coding \c{mov [es:bx],ax}. We
1227 recommend the latter syntax, since it is consistent with other
1228 syntactic features of the language, but for instructions such as
1229 \c{LODSB}, which has no operands and yet can require a segment
1230 override, there is no clean syntactic way to proceed apart from
1233 An instruction is not required to use a prefix: prefixes such as
1234 \c{CS}, \c{A32}, \c{LOCK} or \c{REPE} can appear on a line by
1235 themselves, and NASM will just generate the prefix bytes.
1237 In addition to actual machine instructions, NASM also supports a
1238 number of pseudo-instructions, described in \k{pseudop}.
1240 Instruction \i{operands} may take a number of forms: they can be
1241 registers, described simply by the register name (e.g. \c{ax},
1242 \c{bp}, \c{ebx}, \c{cr0}: NASM does not use the \c{gas}-style
1243 syntax in which register names must be prefixed by a \c{%} sign), or
1244 they can be \i{effective addresses} (see \k{effaddr}), constants
1245 (\k{const}) or expressions (\k{expr}).
1247 For x87 \i{floating-point} instructions, NASM accepts a wide range of
1248 syntaxes: you can use two-operand forms like MASM supports, or you
1249 can use NASM's native single-operand forms in most cases.
1251 \# all forms of each supported instruction are given in
1253 For example, you can code:
1255 \c fadd st1 ; this sets st0 := st0 + st1
1256 \c fadd st0,st1 ; so does this
1258 \c fadd st1,st0 ; this sets st1 := st1 + st0
1259 \c fadd to st1 ; so does this
1261 Almost any x87 floating-point instruction that references memory must
1262 use one of the prefixes \i\c{DWORD}, \i\c{QWORD} or \i\c{TWORD} to
1263 indicate what size of \i{memory operand} it refers to.
1266 \H{pseudop} \i{Pseudo-Instructions}
1268 Pseudo-instructions are things which, though not real x86 machine
1269 instructions, are used in the instruction field anyway because that's
1270 the most convenient place to put them. The current pseudo-instructions
1271 are \i\c{DB}, \i\c{DW}, \i\c{DD}, \i\c{DQ}, \i\c{DT}, \i\c{DO},
1272 \i\c{DY} and \i\c\{DZ}; their \i{uninitialized} counterparts
1273 \i\c{RESB}, \i\c{RESW}, \i\c{RESD}, \i\c{RESQ}, \i\c{REST},
1274 \i\c{RESO}, \i\c{RESY} and \i\c\{RESZ}; the \i\c{INCBIN} command, the
1275 \i\c{EQU} command, and the \i\c{TIMES} prefix.
1278 \S{db} \c{DB} and Friends: Declaring Initialized Data
1280 \i\c{DB}, \i\c{DW}, \i\c{DD}, \i\c{DQ}, \i\c{DT}, \i\c{DO}, \i\c{DY}
1281 and \i\c{DZ} are used, much as in MASM, to declare initialized data in
1282 the output file. They can be invoked in a wide range of ways:
1283 \I{floating-point}\I{character constant}\I{string constant}
1285 \c db 0x55 ; just the byte 0x55
1286 \c db 0x55,0x56,0x57 ; three bytes in succession
1287 \c db 'a',0x55 ; character constants are OK
1288 \c db 'hello',13,10,'$' ; so are string constants
1289 \c dw 0x1234 ; 0x34 0x12
1290 \c dw 'a' ; 0x61 0x00 (it's just a number)
1291 \c dw 'ab' ; 0x61 0x62 (character constant)
1292 \c dw 'abc' ; 0x61 0x62 0x63 0x00 (string)
1293 \c dd 0x12345678 ; 0x78 0x56 0x34 0x12
1294 \c dd 1.234567e20 ; floating-point constant
1295 \c dq 0x123456789abcdef0 ; eight byte constant
1296 \c dq 1.234567e20 ; double-precision float
1297 \c dt 1.234567e20 ; extended-precision float
1299 \c{DT}, \c{DO}, \c{DY} and \c{DZ} do not accept \i{numeric constants}
1303 \S{resb} \c{RESB} and Friends: Declaring \i{Uninitialized} Data
1305 \i\c{RESB}, \i\c{RESW}, \i\c{RESD}, \i\c{RESQ}, \i\c{REST},
1306 \i\c{RESO}, \i\c{RESY} and \i\c\{RESZ} are designed to be used in the
1307 BSS section of a module: they declare \e{uninitialized} storage
1308 space. Each takes a single operand, which is the number of bytes,
1309 words, doublewords or whatever to reserve. As stated in \k{qsother},
1310 NASM does not support the MASM/TASM syntax of reserving uninitialized
1311 space by writing \I\c{?}\c{DW ?} or similar things: this is what it
1312 does instead. The operand to a \c{RESB}-type pseudo-instruction is a
1313 \i\e{critical expression}: see \k{crit}.
1317 \c buffer: resb 64 ; reserve 64 bytes
1318 \c wordvar: resw 1 ; reserve a word
1319 \c realarray resq 10 ; array of ten reals
1320 \c ymmval: resy 1 ; one YMM register
1321 \c zmmvals: resz 32 ; 32 ZMM registers
1323 \S{incbin} \i\c{INCBIN}: Including External \i{Binary Files}
1325 \c{INCBIN} is borrowed from the old Amiga assembler \i{DevPac}: it
1326 includes a binary file verbatim into the output file. This can be
1327 handy for (for example) including \i{graphics} and \i{sound} data
1328 directly into a game executable file. It can be called in one of
1331 \c incbin "file.dat" ; include the whole file
1332 \c incbin "file.dat",1024 ; skip the first 1024 bytes
1333 \c incbin "file.dat",1024,512 ; skip the first 1024, and
1334 \c ; actually include at most 512
1336 \c{INCBIN} is both a directive and a standard macro; the standard
1337 macro version searches for the file in the include file search path
1338 and adds the file to the dependency lists. This macro can be
1339 overridden if desired.
1342 \S{equ} \i\c{EQU}: Defining Constants
1344 \c{EQU} defines a symbol to a given constant value: when \c{EQU} is
1345 used, the source line must contain a label. The action of \c{EQU} is
1346 to define the given label name to the value of its (only) operand.
1347 This definition is absolute, and cannot change later. So, for
1350 \c message db 'hello, world'
1351 \c msglen equ $-message
1353 defines \c{msglen} to be the constant 12. \c{msglen} may not then be
1354 redefined later. This is not a \i{preprocessor} definition either:
1355 the value of \c{msglen} is evaluated \e{once}, using the value of
1356 \c{$} (see \k{expr} for an explanation of \c{$}) at the point of
1357 definition, rather than being evaluated wherever it is referenced
1358 and using the value of \c{$} at the point of reference.
1361 \S{times} \i\c{TIMES}: \i{Repeating} Instructions or Data
1363 The \c{TIMES} prefix causes the instruction to be assembled multiple
1364 times. This is partly present as NASM's equivalent of the \i\c{DUP}
1365 syntax supported by \i{MASM}-compatible assemblers, in that you can
1368 \c zerobuf: times 64 db 0
1370 or similar things; but \c{TIMES} is more versatile than that. The
1371 argument to \c{TIMES} is not just a numeric constant, but a numeric
1372 \e{expression}, so you can do things like
1374 \c buffer: db 'hello, world'
1375 \c times 64-$+buffer db ' '
1377 which will store exactly enough spaces to make the total length of
1378 \c{buffer} up to 64. Finally, \c{TIMES} can be applied to ordinary
1379 instructions, so you can code trivial \i{unrolled loops} in it:
1383 Note that there is no effective difference between \c{times 100 resb
1384 1} and \c{resb 100}, except that the latter will be assembled about
1385 100 times faster due to the internal structure of the assembler.
1387 The operand to \c{TIMES} is a critical expression (\k{crit}).
1389 Note also that \c{TIMES} can't be applied to \i{macros}: the reason
1390 for this is that \c{TIMES} is processed after the macro phase, which
1391 allows the argument to \c{TIMES} to contain expressions such as
1392 \c{64-$+buffer} as above. To repeat more than one line of code, or a
1393 complex macro, use the preprocessor \i\c{%rep} directive.
1396 \H{effaddr} Effective Addresses
1398 An \i{effective address} is any operand to an instruction which
1399 \I{memory reference}references memory. Effective addresses, in NASM,
1400 have a very simple syntax: they consist of an expression evaluating
1401 to the desired address, enclosed in \i{square brackets}. For
1406 \c mov ax,[wordvar+1]
1407 \c mov ax,[es:wordvar+bx]
1409 Anything not conforming to this simple system is not a valid memory
1410 reference in NASM, for example \c{es:wordvar[bx]}.
1412 More complicated effective addresses, such as those involving more
1413 than one register, work in exactly the same way:
1415 \c mov eax,[ebx*2+ecx+offset]
1418 NASM is capable of doing \i{algebra} on these effective addresses,
1419 so that things which don't necessarily \e{look} legal are perfectly
1422 \c mov eax,[ebx*5] ; assembles as [ebx*4+ebx]
1423 \c mov eax,[label1*2-label2] ; ie [label1+(label1-label2)]
1425 Some forms of effective address have more than one assembled form;
1426 in most such cases NASM will generate the smallest form it can. For
1427 example, there are distinct assembled forms for the 32-bit effective
1428 addresses \c{[eax*2+0]} and \c{[eax+eax]}, and NASM will generally
1429 generate the latter on the grounds that the former requires four
1430 bytes to store a zero offset.
1432 NASM has a hinting mechanism which will cause \c{[eax+ebx]} and
1433 \c{[ebx+eax]} to generate different opcodes; this is occasionally
1434 useful because \c{[esi+ebp]} and \c{[ebp+esi]} have different
1435 default segment registers.
1437 However, you can force NASM to generate an effective address in a
1438 particular form by the use of the keywords \c{BYTE}, \c{WORD},
1439 \c{DWORD} and \c{NOSPLIT}. If you need \c{[eax+3]} to be assembled
1440 using a double-word offset field instead of the one byte NASM will
1441 normally generate, you can code \c{[dword eax+3]}. Similarly, you
1442 can force NASM to use a byte offset for a small value which it
1443 hasn't seen on the first pass (see \k{crit} for an example of such a
1444 code fragment) by using \c{[byte eax+offset]}. As special cases,
1445 \c{[byte eax]} will code \c{[eax+0]} with a byte offset of zero, and
1446 \c{[dword eax]} will code it with a double-word offset of zero. The
1447 normal form, \c{[eax]}, will be coded with no offset field.
1449 The form described in the previous paragraph is also useful if you
1450 are trying to access data in a 32-bit segment from within 16 bit code.
1451 For more information on this see the section on mixed-size addressing
1452 (\k{mixaddr}). In particular, if you need to access data with a known
1453 offset that is larger than will fit in a 16-bit value, if you don't
1454 specify that it is a dword offset, nasm will cause the high word of
1455 the offset to be lost.
1457 Similarly, NASM will split \c{[eax*2]} into \c{[eax+eax]} because
1458 that allows the offset field to be absent and space to be saved; in
1459 fact, it will also split \c{[eax*2+offset]} into
1460 \c{[eax+eax+offset]}. You can combat this behaviour by the use of
1461 the \c{NOSPLIT} keyword: \c{[nosplit eax*2]} will force
1462 \c{[eax*2+0]} to be generated literally. \c{[nosplit eax*1]} also has the
1463 same effect. In another way, a split EA form \c{[0, eax*2]} can be used, too.
1464 However, \c{NOSPLIT} in \c{[nosplit eax+eax]} will be ignored because user's
1465 intention here is considered as \c{[eax+eax]}.
1467 In 64-bit mode, NASM will by default generate absolute addresses. The
1468 \i\c{REL} keyword makes it produce \c{RIP}-relative addresses. Since
1469 this is frequently the normally desired behaviour, see the \c{DEFAULT}
1470 directive (\k{default}). The keyword \i\c{ABS} overrides \i\c{REL}.
1472 A new form of split effective addres syntax is also supported. This is
1473 mainly intended for mib operands as used by MPX instructions, but can
1474 be used for any memory reference. The basic concept of this form is
1475 splitting base and index.
1477 \c mov eax,[ebx+8,ecx*4] ; ebx=base, ecx=index, 4=scale, 8=disp
1479 For mib operands, there are several ways of writing effective address depending
1480 on the tools. NASM supports all currently possible ways of mib syntax:
1483 \c ; next 5 lines are parsed same
1484 \c ; base=rax, index=rbx, scale=1, displacement=3
1485 \c bndstx [rax+0x3,rbx], bnd0 ; NASM - split EA
1486 \c bndstx [rbx*1+rax+0x3], bnd0 ; GAS - '*1' indecates an index reg
1487 \c bndstx [rax+rbx+3], bnd0 ; GAS - without hints
1488 \c bndstx [rax+0x3], bnd0, rbx ; ICC-1
1489 \c bndstx [rax+0x3], rbx, bnd0 ; ICC-2
1491 When broadcasting decorator is used, the opsize keyword should match
1492 the size of each element.
1494 \c VDIVPS zmm4, zmm5, dword [rbx]{1to16} ; single-precision float
1495 \c VDIVPS zmm4, zmm5, zword [rbx] ; packed 512 bit memory
1498 \H{const} \i{Constants}
1500 NASM understands four different types of constant: numeric,
1501 character, string and floating-point.
1504 \S{numconst} \i{Numeric Constants}
1506 A numeric constant is simply a number. NASM allows you to specify
1507 numbers in a variety of number bases, in a variety of ways: you can
1508 suffix \c{H} or \c{X}, \c{D} or \c{T}, \c{Q} or \c{O}, and \c{B} or
1509 \c{Y} for \i{hexadecimal}, \i{decimal}, \i{octal} and \i{binary}
1510 respectively, or you can prefix \c{0x}, for hexadecimal in the style
1511 of C, or you can prefix \c{$} for hexadecimal in the style of Borland
1512 Pascal or Motorola Assemblers. Note, though, that the \I{$,
1513 prefix}\c{$} prefix does double duty as a prefix on identifiers (see
1514 \k{syntax}), so a hex number prefixed with a \c{$} sign must have a
1515 digit after the \c{$} rather than a letter. In addition, current
1516 versions of NASM accept the prefix \c{0h} for hexadecimal, \c{0d} or
1517 \c{0t} for decimal, \c{0o} or \c{0q} for octal, and \c{0b} or \c{0y}
1518 for binary. Please note that unlike C, a \c{0} prefix by itself does
1519 \e{not} imply an octal constant!
1521 Numeric constants can have underscores (\c{_}) interspersed to break
1524 Some examples (all producing exactly the same code):
1526 \c mov ax,200 ; decimal
1527 \c mov ax,0200 ; still decimal
1528 \c mov ax,0200d ; explicitly decimal
1529 \c mov ax,0d200 ; also decimal
1530 \c mov ax,0c8h ; hex
1531 \c mov ax,$0c8 ; hex again: the 0 is required
1532 \c mov ax,0xc8 ; hex yet again
1533 \c mov ax,0hc8 ; still hex
1534 \c mov ax,310q ; octal
1535 \c mov ax,310o ; octal again
1536 \c mov ax,0o310 ; octal yet again
1537 \c mov ax,0q310 ; octal yet again
1538 \c mov ax,11001000b ; binary
1539 \c mov ax,1100_1000b ; same binary constant
1540 \c mov ax,1100_1000y ; same binary constant once more
1541 \c mov ax,0b1100_1000 ; same binary constant yet again
1542 \c mov ax,0y1100_1000 ; same binary constant yet again
1544 \S{strings} \I{Strings}\i{Character Strings}
1546 A character string consists of up to eight characters enclosed in
1547 either single quotes (\c{'...'}), double quotes (\c{"..."}) or
1548 backquotes (\c{`...`}). Single or double quotes are equivalent to
1549 NASM (except of course that surrounding the constant with single
1550 quotes allows double quotes to appear within it and vice versa); the
1551 contents of those are represented verbatim. Strings enclosed in
1552 backquotes support C-style \c{\\}-escapes for special characters.
1555 The following \i{escape sequences} are recognized by backquoted strings:
1557 \c \' single quote (')
1558 \c \" double quote (")
1560 \c \\\ backslash (\)
1561 \c \? question mark (?)
1569 \c \e ESC (ASCII 27)
1570 \c \377 Up to 3 octal digits - literal byte
1571 \c \xFF Up to 2 hexadecimal digits - literal byte
1572 \c \u1234 4 hexadecimal digits - Unicode character
1573 \c \U12345678 8 hexadecimal digits - Unicode character
1575 All other escape sequences are reserved. Note that \c{\\0}, meaning a
1576 \c{NUL} character (ASCII 0), is a special case of the octal escape
1579 \i{Unicode} characters specified with \c{\\u} or \c{\\U} are converted to
1580 \i{UTF-8}. For example, the following lines are all equivalent:
1582 \c db `\u263a` ; UTF-8 smiley face
1583 \c db `\xe2\x98\xba` ; UTF-8 smiley face
1584 \c db 0E2h, 098h, 0BAh ; UTF-8 smiley face
1587 \S{chrconst} \i{Character Constants}
1589 A character constant consists of a string up to eight bytes long, used
1590 in an expression context. It is treated as if it was an integer.
1592 A character constant with more than one byte will be arranged
1593 with \i{little-endian} order in mind: if you code
1597 then the constant generated is not \c{0x61626364}, but
1598 \c{0x64636261}, so that if you were then to store the value into
1599 memory, it would read \c{abcd} rather than \c{dcba}. This is also
1600 the sense of character constants understood by the Pentium's
1601 \i\c{CPUID} instruction.
1604 \S{strconst} \i{String Constants}
1606 String constants are character strings used in the context of some
1607 pseudo-instructions, namely the
1608 \I\c{DW}\I\c{DD}\I\c{DQ}\I\c{DT}\I\c{DO}\I\c{DY}\i\c{DB} family and
1609 \i\c{INCBIN} (where it represents a filename.) They are also used in
1610 certain preprocessor directives.
1612 A string constant looks like a character constant, only longer. It
1613 is treated as a concatenation of maximum-size character constants
1614 for the conditions. So the following are equivalent:
1616 \c db 'hello' ; string constant
1617 \c db 'h','e','l','l','o' ; equivalent character constants
1619 And the following are also equivalent:
1621 \c dd 'ninechars' ; doubleword string constant
1622 \c dd 'nine','char','s' ; becomes three doublewords
1623 \c db 'ninechars',0,0,0 ; and really looks like this
1625 Note that when used in a string-supporting context, quoted strings are
1626 treated as a string constants even if they are short enough to be a
1627 character constant, because otherwise \c{db 'ab'} would have the same
1628 effect as \c{db 'a'}, which would be silly. Similarly, three-character
1629 or four-character constants are treated as strings when they are
1630 operands to \c{DW}, and so forth.
1632 \S{unicode} \I{UTF-16}\I{UTF-32}\i{Unicode} Strings
1634 The special operators \i\c{__utf16__}, \i\c{__utf16le__},
1635 \i\c{__utf16be__}, \i\c{__utf32__}, \i\c{__utf32le__} and
1636 \i\c{__utf32be__} allows definition of Unicode strings. They take a
1637 string in UTF-8 format and converts it to UTF-16 or UTF-32,
1638 respectively. Unless the \c{be} forms are specified, the output is
1643 \c %define u(x) __utf16__(x)
1644 \c %define w(x) __utf32__(x)
1646 \c dw u('C:\WINDOWS'), 0 ; Pathname in UTF-16
1647 \c dd w(`A + B = \u206a`), 0 ; String in UTF-32
1649 The UTF operators can be applied either to strings passed to the
1650 \c{DB} family instructions, or to character constants in an expression
1653 \S{fltconst} \I{floating-point, constants}Floating-Point Constants
1655 \i{Floating-point} constants are acceptable only as arguments to
1656 \i\c{DB}, \i\c{DW}, \i\c{DD}, \i\c{DQ}, \i\c{DT}, and \i\c{DO}, or as
1657 arguments to the special operators \i\c{__float8__},
1658 \i\c{__float16__}, \i\c{__float32__}, \i\c{__float64__},
1659 \i\c{__float80m__}, \i\c{__float80e__}, \i\c{__float128l__}, and
1660 \i\c{__float128h__}.
1662 Floating-point constants are expressed in the traditional form:
1663 digits, then a period, then optionally more digits, then optionally an
1664 \c{E} followed by an exponent. The period is mandatory, so that NASM
1665 can distinguish between \c{dd 1}, which declares an integer constant,
1666 and \c{dd 1.0} which declares a floating-point constant.
1668 NASM also support C99-style hexadecimal floating-point: \c{0x},
1669 hexadecimal digits, period, optionally more hexadeximal digits, then
1670 optionally a \c{P} followed by a \e{binary} (not hexadecimal) exponent
1671 in decimal notation. As an extension, NASM additionally supports the
1672 \c{0h} and \c{$} prefixes for hexadecimal, as well binary and octal
1673 floating-point, using the \c{0b} or \c{0y} and \c{0o} or \c{0q}
1674 prefixes, respectively.
1676 Underscores to break up groups of digits are permitted in
1677 floating-point constants as well.
1681 \c db -0.2 ; "Quarter precision"
1682 \c dw -0.5 ; IEEE 754r/SSE5 half precision
1683 \c dd 1.2 ; an easy one
1684 \c dd 1.222_222_222 ; underscores are permitted
1685 \c dd 0x1p+2 ; 1.0x2^2 = 4.0
1686 \c dq 0x1p+32 ; 1.0x2^32 = 4 294 967 296.0
1687 \c dq 1.e10 ; 10 000 000 000.0
1688 \c dq 1.e+10 ; synonymous with 1.e10
1689 \c dq 1.e-10 ; 0.000 000 000 1
1690 \c dt 3.141592653589793238462 ; pi
1691 \c do 1.e+4000 ; IEEE 754r quad precision
1693 The 8-bit "quarter-precision" floating-point format is
1694 sign:exponent:mantissa = 1:4:3 with an exponent bias of 7. This
1695 appears to be the most frequently used 8-bit floating-point format,
1696 although it is not covered by any formal standard. This is sometimes
1697 called a "\i{minifloat}."
1699 The special operators are used to produce floating-point numbers in
1700 other contexts. They produce the binary representation of a specific
1701 floating-point number as an integer, and can use anywhere integer
1702 constants are used in an expression. \c{__float80m__} and
1703 \c{__float80e__} produce the 64-bit mantissa and 16-bit exponent of an
1704 80-bit floating-point number, and \c{__float128l__} and
1705 \c{__float128h__} produce the lower and upper 64-bit halves of a 128-bit
1706 floating-point number, respectively.
1710 \c mov rax,__float64__(3.141592653589793238462)
1712 ... would assign the binary representation of pi as a 64-bit floating
1713 point number into \c{RAX}. This is exactly equivalent to:
1715 \c mov rax,0x400921fb54442d18
1717 NASM cannot do compile-time arithmetic on floating-point constants.
1718 This is because NASM is designed to be portable - although it always
1719 generates code to run on x86 processors, the assembler itself can
1720 run on any system with an ANSI C compiler. Therefore, the assembler
1721 cannot guarantee the presence of a floating-point unit capable of
1722 handling the \i{Intel number formats}, and so for NASM to be able to
1723 do floating arithmetic it would have to include its own complete set
1724 of floating-point routines, which would significantly increase the
1725 size of the assembler for very little benefit.
1727 The special tokens \i\c{__Infinity__}, \i\c{__QNaN__} (or
1728 \i\c{__NaN__}) and \i\c{__SNaN__} can be used to generate
1729 \I{infinity}infinities, quiet \i{NaN}s, and signalling NaNs,
1730 respectively. These are normally used as macros:
1732 \c %define Inf __Infinity__
1733 \c %define NaN __QNaN__
1735 \c dq +1.5, -Inf, NaN ; Double-precision constants
1737 The \c{%use fp} standard macro package contains a set of convenience
1738 macros. See \k{pkg_fp}.
1740 \S{bcdconst} \I{floating-point, packed BCD constants}Packed BCD Constants
1742 x87-style packed BCD constants can be used in the same contexts as
1743 80-bit floating-point numbers. They are suffixed with \c{p} or
1744 prefixed with \c{0p}, and can include up to 18 decimal digits.
1746 As with other numeric constants, underscores can be used to separate
1751 \c dt 12_345_678_901_245_678p
1752 \c dt -12_345_678_901_245_678p
1757 \H{expr} \i{Expressions}
1759 Expressions in NASM are similar in syntax to those in C. Expressions
1760 are evaluated as 64-bit integers which are then adjusted to the
1763 NASM supports two special tokens in expressions, allowing
1764 calculations to involve the current assembly position: the
1765 \I{$, here}\c{$} and \i\c{$$} tokens. \c{$} evaluates to the assembly
1766 position at the beginning of the line containing the expression; so
1767 you can code an \i{infinite loop} using \c{JMP $}. \c{$$} evaluates
1768 to the beginning of the current section; so you can tell how far
1769 into the section you are by using \c{($-$$)}.
1771 The arithmetic \i{operators} provided by NASM are listed here, in
1772 increasing order of \i{precedence}.
1775 \S{expor} \i\c{|}: \i{Bitwise OR} Operator
1777 The \c{|} operator gives a bitwise OR, exactly as performed by the
1778 \c{OR} machine instruction. Bitwise OR is the lowest-priority
1779 arithmetic operator supported by NASM.
1782 \S{expxor} \i\c{^}: \i{Bitwise XOR} Operator
1784 \c{^} provides the bitwise XOR operation.
1787 \S{expand} \i\c{&}: \i{Bitwise AND} Operator
1789 \c{&} provides the bitwise AND operation.
1792 \S{expshift} \i\c{<<} and \i\c{>>}: \i{Bit Shift} Operators
1794 \c{<<} gives a bit-shift to the left, just as it does in C. So \c{5<<3}
1795 evaluates to 5 times 8, or 40. \c{>>} gives a bit-shift to the
1796 right; in NASM, such a shift is \e{always} unsigned, so that
1797 the bits shifted in from the left-hand end are filled with zero
1798 rather than a sign-extension of the previous highest bit.
1801 \S{expplmi} \I{+ opaddition}\c{+} and \I{- opsubtraction}\c{-}:
1802 \i{Addition} and \i{Subtraction} Operators
1804 The \c{+} and \c{-} operators do perfectly ordinary addition and
1808 \S{expmul} \i\c{*}, \i\c{/}, \i\c{//}, \i\c{%} and \i\c{%%}:
1809 \i{Multiplication} and \i{Division}
1811 \c{*} is the multiplication operator. \c{/} and \c{//} are both
1812 division operators: \c{/} is \i{unsigned division} and \c{//} is
1813 \i{signed division}. Similarly, \c{%} and \c{%%} provide \I{unsigned
1814 modulo}\I{modulo operators}unsigned and
1815 \i{signed modulo} operators respectively.
1817 NASM, like ANSI C, provides no guarantees about the sensible
1818 operation of the signed modulo operator.
1820 Since the \c{%} character is used extensively by the macro
1821 \i{preprocessor}, you should ensure that both the signed and unsigned
1822 modulo operators are followed by white space wherever they appear.
1825 \S{expmul} \i{Unary Operators}
1827 The highest-priority operators in NASM's expression grammar are those
1828 which only apply to one argument. These are \I{+ opunary}\c{+}, \I{-
1829 opunary}\c{-}, \i\c{~}, \I{! opunary}\c{!}, \i\c{SEG}, and the
1830 \i{integer functions} operators.
1832 \c{-} negates its operand, \c{+} does nothing (it's provided for
1833 symmetry with \c{-}), \c{~} computes the \i{one's complement} of its
1834 operand, \c{!} is the \i{logical negation} operator.
1836 \c{SEG} provides the \i{segment address}
1837 of its operand (explained in more detail in \k{segwrt}).
1839 A set of additional operators with leading and trailing double
1840 underscores are used to implement the integer functions of the
1841 \c{ifunc} macro package, see \k{pkg_ifunc}.
1844 \H{segwrt} \i\c{SEG} and \i\c{WRT}
1846 When writing large 16-bit programs, which must be split into
1847 multiple \i{segments}, it is often necessary to be able to refer to
1848 the \I{segment address}segment part of the address of a symbol. NASM
1849 supports the \c{SEG} operator to perform this function.
1851 The \c{SEG} operator returns the \i\e{preferred} segment base of a
1852 symbol, defined as the segment base relative to which the offset of
1853 the symbol makes sense. So the code
1855 \c mov ax,seg symbol
1859 will load \c{ES:BX} with a valid pointer to the symbol \c{symbol}.
1861 Things can be more complex than this: since 16-bit segments and
1862 \i{groups} may \I{overlapping segments}overlap, you might occasionally
1863 want to refer to some symbol using a different segment base from the
1864 preferred one. NASM lets you do this, by the use of the \c{WRT}
1865 (With Reference To) keyword. So you can do things like
1867 \c mov ax,weird_seg ; weird_seg is a segment base
1869 \c mov bx,symbol wrt weird_seg
1871 to load \c{ES:BX} with a different, but functionally equivalent,
1872 pointer to the symbol \c{symbol}.
1874 NASM supports far (inter-segment) calls and jumps by means of the
1875 syntax \c{call segment:offset}, where \c{segment} and \c{offset}
1876 both represent immediate values. So to call a far procedure, you
1877 could code either of
1879 \c call (seg procedure):procedure
1880 \c call weird_seg:(procedure wrt weird_seg)
1882 (The parentheses are included for clarity, to show the intended
1883 parsing of the above instructions. They are not necessary in
1886 NASM supports the syntax \I\c{CALL FAR}\c{call far procedure} as a
1887 synonym for the first of the above usages. \c{JMP} works identically
1888 to \c{CALL} in these examples.
1890 To declare a \i{far pointer} to a data item in a data segment, you
1893 \c dw symbol, seg symbol
1895 NASM supports no convenient synonym for this, though you can always
1896 invent one using the macro processor.
1899 \H{strict} \i\c{STRICT}: Inhibiting Optimization
1901 When assembling with the optimizer set to level 2 or higher (see
1902 \k{opt-O}), NASM will use size specifiers (\c{BYTE}, \c{WORD},
1903 \c{DWORD}, \c{QWORD}, \c{TWORD}, \c{OWORD}, \c{YWORD} or \c{ZWORD}),
1904 but will give them the smallest possible size. The keyword \c{STRICT}
1905 can be used to inhibit optimization and force a particular operand to
1906 be emitted in the specified size. For example, with the optimizer on,
1907 and in \c{BITS 16} mode,
1911 is encoded in three bytes \c{66 6A 21}, whereas
1913 \c push strict dword 33
1915 is encoded in six bytes, with a full dword immediate operand \c{66 68
1918 With the optimizer off, the same code (six bytes) is generated whether
1919 the \c{STRICT} keyword was used or not.
1922 \H{crit} \i{Critical Expressions}
1924 Although NASM has an optional multi-pass optimizer, there are some
1925 expressions which must be resolvable on the first pass. These are
1926 called \e{Critical Expressions}.
1928 The first pass is used to determine the size of all the assembled
1929 code and data, so that the second pass, when generating all the
1930 code, knows all the symbol addresses the code refers to. So one
1931 thing NASM can't handle is code whose size depends on the value of a
1932 symbol declared after the code in question. For example,
1934 \c times (label-$) db 0
1935 \c label: db 'Where am I?'
1937 The argument to \i\c{TIMES} in this case could equally legally
1938 evaluate to anything at all; NASM will reject this example because
1939 it cannot tell the size of the \c{TIMES} line when it first sees it.
1940 It will just as firmly reject the slightly \I{paradox}paradoxical
1943 \c times (label-$+1) db 0
1944 \c label: db 'NOW where am I?'
1946 in which \e{any} value for the \c{TIMES} argument is by definition
1949 NASM rejects these examples by means of a concept called a
1950 \e{critical expression}, which is defined to be an expression whose
1951 value is required to be computable in the first pass, and which must
1952 therefore depend only on symbols defined before it. The argument to
1953 the \c{TIMES} prefix is a critical expression.
1955 \H{locallab} \i{Local Labels}
1957 NASM gives special treatment to symbols beginning with a \i{period}.
1958 A label beginning with a single period is treated as a \e{local}
1959 label, which means that it is associated with the previous non-local
1960 label. So, for example:
1962 \c label1 ; some code
1970 \c label2 ; some code
1978 In the above code fragment, each \c{JNE} instruction jumps to the
1979 line immediately before it, because the two definitions of \c{.loop}
1980 are kept separate by virtue of each being associated with the
1981 previous non-local label.
1983 This form of local label handling is borrowed from the old Amiga
1984 assembler \i{DevPac}; however, NASM goes one step further, in
1985 allowing access to local labels from other parts of the code. This
1986 is achieved by means of \e{defining} a local label in terms of the
1987 previous non-local label: the first definition of \c{.loop} above is
1988 really defining a symbol called \c{label1.loop}, and the second
1989 defines a symbol called \c{label2.loop}. So, if you really needed
1992 \c label3 ; some more code
1997 Sometimes it is useful - in a macro, for instance - to be able to
1998 define a label which can be referenced from anywhere but which
1999 doesn't interfere with the normal local-label mechanism. Such a
2000 label can't be non-local because it would interfere with subsequent
2001 definitions of, and references to, local labels; and it can't be
2002 local because the macro that defined it wouldn't know the label's
2003 full name. NASM therefore introduces a third type of label, which is
2004 probably only useful in macro definitions: if a label begins with
2005 the \I{label prefix}special prefix \i\c{..@}, then it does nothing
2006 to the local label mechanism. So you could code
2008 \c label1: ; a non-local label
2009 \c .local: ; this is really label1.local
2010 \c ..@foo: ; this is a special symbol
2011 \c label2: ; another non-local label
2012 \c .local: ; this is really label2.local
2014 \c jmp ..@foo ; this will jump three lines up
2016 NASM has the capacity to define other special symbols beginning with
2017 a double period: for example, \c{..start} is used to specify the
2018 entry point in the \c{obj} output format (see \k{dotdotstart}),
2019 \c{..imagebase} is used to find out the offset from a base address
2020 of the current image in the \c{win64} output format (see \k{win64pic}).
2021 So just keep in mind that symbols beginning with a double period are
2025 \C{preproc} The NASM \i{Preprocessor}
2027 NASM contains a powerful \i{macro processor}, which supports
2028 conditional assembly, multi-level file inclusion, two forms of macro
2029 (single-line and multi-line), and a `context stack' mechanism for
2030 extra macro power. Preprocessor directives all begin with a \c{%}
2033 The preprocessor collapses all lines which end with a backslash (\\)
2034 character into a single line. Thus:
2036 \c %define THIS_VERY_LONG_MACRO_NAME_IS_DEFINED_TO \\
2039 will work like a single-line macro without the backslash-newline
2042 \H{slmacro} \i{Single-Line Macros}
2044 \S{define} The Normal Way: \I\c{%idefine}\i\c{%define}
2046 Single-line macros are defined using the \c{%define} preprocessor
2047 directive. The definitions work in a similar way to C; so you can do
2050 \c %define ctrl 0x1F &
2051 \c %define param(a,b) ((a)+(a)*(b))
2053 \c mov byte [param(2,ebx)], ctrl 'D'
2055 which will expand to
2057 \c mov byte [(2)+(2)*(ebx)], 0x1F & 'D'
2059 When the expansion of a single-line macro contains tokens which
2060 invoke another macro, the expansion is performed at invocation time,
2061 not at definition time. Thus the code
2063 \c %define a(x) 1+b(x)
2068 will evaluate in the expected way to \c{mov ax,1+2*8}, even though
2069 the macro \c{b} wasn't defined at the time of definition of \c{a}.
2071 Macros defined with \c{%define} are \i{case sensitive}: after
2072 \c{%define foo bar}, only \c{foo} will expand to \c{bar}: \c{Foo} or
2073 \c{FOO} will not. By using \c{%idefine} instead of \c{%define} (the
2074 `i' stands for `insensitive') you can define all the case variants
2075 of a macro at once, so that \c{%idefine foo bar} would cause
2076 \c{foo}, \c{Foo}, \c{FOO}, \c{fOO} and so on all to expand to
2079 There is a mechanism which detects when a macro call has occurred as
2080 a result of a previous expansion of the same macro, to guard against
2081 \i{circular references} and infinite loops. If this happens, the
2082 preprocessor will only expand the first occurrence of the macro.
2085 \c %define a(x) 1+a(x)
2089 the macro \c{a(3)} will expand once, becoming \c{1+a(3)}, and will
2090 then expand no further. This behaviour can be useful: see \k{32c}
2091 for an example of its use.
2093 You can \I{overloading, single-line macros}overload single-line
2094 macros: if you write
2096 \c %define foo(x) 1+x
2097 \c %define foo(x,y) 1+x*y
2099 the preprocessor will be able to handle both types of macro call,
2100 by counting the parameters you pass; so \c{foo(3)} will become
2101 \c{1+3} whereas \c{foo(ebx,2)} will become \c{1+ebx*2}. However, if
2106 then no other definition of \c{foo} will be accepted: a macro with
2107 no parameters prohibits the definition of the same name as a macro
2108 \e{with} parameters, and vice versa.
2110 This doesn't prevent single-line macros being \e{redefined}: you can
2111 perfectly well define a macro with
2115 and then re-define it later in the same source file with
2119 Then everywhere the macro \c{foo} is invoked, it will be expanded
2120 according to the most recent definition. This is particularly useful
2121 when defining single-line macros with \c{%assign} (see \k{assign}).
2123 You can \i{pre-define} single-line macros using the `-d' option on
2124 the NASM command line: see \k{opt-d}.
2127 \S{xdefine} Resolving \c{%define}: \I\c{%ixdefine}\i\c{%xdefine}
2129 To have a reference to an embedded single-line macro resolved at the
2130 time that the embedding macro is \e{defined}, as opposed to when the
2131 embedding macro is \e{expanded}, you need a different mechanism to the
2132 one offered by \c{%define}. The solution is to use \c{%xdefine}, or
2133 it's \I{case sensitive}case-insensitive counterpart \c{%ixdefine}.
2135 Suppose you have the following code:
2138 \c %define isFalse isTrue
2147 In this case, \c{val1} is equal to 0, and \c{val2} is equal to 1.
2148 This is because, when a single-line macro is defined using
2149 \c{%define}, it is expanded only when it is called. As \c{isFalse}
2150 expands to \c{isTrue}, the expansion will be the current value of
2151 \c{isTrue}. The first time it is called that is 0, and the second
2154 If you wanted \c{isFalse} to expand to the value assigned to the
2155 embedded macro \c{isTrue} at the time that \c{isFalse} was defined,
2156 you need to change the above code to use \c{%xdefine}.
2158 \c %xdefine isTrue 1
2159 \c %xdefine isFalse isTrue
2160 \c %xdefine isTrue 0
2164 \c %xdefine isTrue 1
2168 Now, each time that \c{isFalse} is called, it expands to 1,
2169 as that is what the embedded macro \c{isTrue} expanded to at
2170 the time that \c{isFalse} was defined.
2173 \S{indmacro} \i{Macro Indirection}: \I\c{%[}\c{%[...]}
2175 The \c{%[...]} construct can be used to expand macros in contexts
2176 where macro expansion would otherwise not occur, including in the
2177 names other macros. For example, if you have a set of macros named
2178 \c{Foo16}, \c{Foo32} and \c{Foo64}, you could write:
2180 \c mov ax,Foo%[__BITS__] ; The Foo value
2182 to use the builtin macro \c{__BITS__} (see \k{bitsm}) to automatically
2183 select between them. Similarly, the two statements:
2185 \c %xdefine Bar Quux ; Expands due to %xdefine
2186 \c %define Bar %[Quux] ; Expands due to %[...]
2188 have, in fact, exactly the same effect.
2190 \c{%[...]} concatenates to adjacent tokens in the same way that
2191 multi-line macro parameters do, see \k{concat} for details.
2194 \S{concat%+} Concatenating Single Line Macro Tokens: \i\c{%+}
2196 Individual tokens in single line macros can be concatenated, to produce
2197 longer tokens for later processing. This can be useful if there are
2198 several similar macros that perform similar functions.
2200 Please note that a space is required after \c{%+}, in order to
2201 disambiguate it from the syntax \c{%+1} used in multiline macros.
2203 As an example, consider the following:
2205 \c %define BDASTART 400h ; Start of BIOS data area
2207 \c struc tBIOSDA ; its structure
2213 Now, if we need to access the elements of tBIOSDA in different places,
2216 \c mov ax,BDASTART + tBIOSDA.COM1addr
2217 \c mov bx,BDASTART + tBIOSDA.COM2addr
2219 This will become pretty ugly (and tedious) if used in many places, and
2220 can be reduced in size significantly by using the following macro:
2222 \c ; Macro to access BIOS variables by their names (from tBDA):
2224 \c %define BDA(x) BDASTART + tBIOSDA. %+ x
2226 Now the above code can be written as:
2228 \c mov ax,BDA(COM1addr)
2229 \c mov bx,BDA(COM2addr)
2231 Using this feature, we can simplify references to a lot of macros (and,
2232 in turn, reduce typing errors).
2235 \S{selfref%?} The Macro Name Itself: \i\c{%?} and \i\c{%??}
2237 The special symbols \c{%?} and \c{%??} can be used to reference the
2238 macro name itself inside a macro expansion, this is supported for both
2239 single-and multi-line macros. \c{%?} refers to the macro name as
2240 \e{invoked}, whereas \c{%??} refers to the macro name as
2241 \e{declared}. The two are always the same for case-sensitive
2242 macros, but for case-insensitive macros, they can differ.
2246 \c %idefine Foo mov %?,%??
2258 \c %idefine keyword $%?
2260 can be used to make a keyword "disappear", for example in case a new
2261 instruction has been used as a label in older code. For example:
2263 \c %idefine pause $%? ; Hide the PAUSE instruction
2266 \S{undef} Undefining Single-Line Macros: \i\c{%undef}
2268 Single-line macros can be removed with the \c{%undef} directive. For
2269 example, the following sequence:
2276 will expand to the instruction \c{mov eax, foo}, since after
2277 \c{%undef} the macro \c{foo} is no longer defined.
2279 Macros that would otherwise be pre-defined can be undefined on the
2280 command-line using the `-u' option on the NASM command line: see
2284 \S{assign} \i{Preprocessor Variables}: \i\c{%assign}
2286 An alternative way to define single-line macros is by means of the
2287 \c{%assign} command (and its \I{case sensitive}case-insensitive
2288 counterpart \i\c{%iassign}, which differs from \c{%assign} in
2289 exactly the same way that \c{%idefine} differs from \c{%define}).
2291 \c{%assign} is used to define single-line macros which take no
2292 parameters and have a numeric value. This value can be specified in
2293 the form of an expression, and it will be evaluated once, when the
2294 \c{%assign} directive is processed.
2296 Like \c{%define}, macros defined using \c{%assign} can be re-defined
2297 later, so you can do things like
2301 to increment the numeric value of a macro.
2303 \c{%assign} is useful for controlling the termination of \c{%rep}
2304 preprocessor loops: see \k{rep} for an example of this. Another
2305 use for \c{%assign} is given in \k{16c} and \k{32c}.
2307 The expression passed to \c{%assign} is a \i{critical expression}
2308 (see \k{crit}), and must also evaluate to a pure number (rather than
2309 a relocatable reference such as a code or data address, or anything
2310 involving a register).
2313 \S{defstr} Defining Strings: \I\c{%idefstr}\i\c{%defstr}
2315 \c{%defstr}, and its case-insensitive counterpart \c{%idefstr}, define
2316 or redefine a single-line macro without parameters but converts the
2317 entire right-hand side, after macro expansion, to a quoted string
2322 \c %defstr test TEST
2326 \c %define test 'TEST'
2328 This can be used, for example, with the \c{%!} construct (see
2331 \c %defstr PATH %!PATH ; The operating system PATH variable
2334 \S{deftok} Defining Tokens: \I\c{%ideftok}\i\c{%deftok}
2336 \c{%deftok}, and its case-insensitive counterpart \c{%ideftok}, define
2337 or redefine a single-line macro without parameters but converts the
2338 second parameter, after string conversion, to a sequence of tokens.
2342 \c %deftok test 'TEST'
2346 \c %define test TEST
2349 \H{strlen} \i{String Manipulation in Macros}
2351 It's often useful to be able to handle strings in macros. NASM
2352 supports a few simple string handling macro operators from which
2353 more complex operations can be constructed.
2355 All the string operators define or redefine a value (either a string
2356 or a numeric value) to a single-line macro. When producing a string
2357 value, it may change the style of quoting of the input string or
2358 strings, and possibly use \c{\\}-escapes inside \c{`}-quoted strings.
2360 \S{strcat} \i{Concatenating Strings}: \i\c{%strcat}
2362 The \c{%strcat} operator concatenates quoted strings and assign them to
2363 a single-line macro.
2367 \c %strcat alpha "Alpha: ", '12" screen'
2369 ... would assign the value \c{'Alpha: 12" screen'} to \c{alpha}.
2372 \c %strcat beta '"foo"\', "'bar'"
2374 ... would assign the value \c{`"foo"\\\\'bar'`} to \c{beta}.
2376 The use of commas to separate strings is permitted but optional.
2379 \S{strlen} \i{String Length}: \i\c{%strlen}
2381 The \c{%strlen} operator assigns the length of a string to a macro.
2384 \c %strlen charcnt 'my string'
2386 In this example, \c{charcnt} would receive the value 9, just as
2387 if an \c{%assign} had been used. In this example, \c{'my string'}
2388 was a literal string but it could also have been a single-line
2389 macro that expands to a string, as in the following example:
2391 \c %define sometext 'my string'
2392 \c %strlen charcnt sometext
2394 As in the first case, this would result in \c{charcnt} being
2395 assigned the value of 9.
2398 \S{substr} \i{Extracting Substrings}: \i\c{%substr}
2400 Individual letters or substrings in strings can be extracted using the
2401 \c{%substr} operator. An example of its use is probably more useful
2402 than the description:
2404 \c %substr mychar 'xyzw' 1 ; equivalent to %define mychar 'x'
2405 \c %substr mychar 'xyzw' 2 ; equivalent to %define mychar 'y'
2406 \c %substr mychar 'xyzw' 3 ; equivalent to %define mychar 'z'
2407 \c %substr mychar 'xyzw' 2,2 ; equivalent to %define mychar 'yz'
2408 \c %substr mychar 'xyzw' 2,-1 ; equivalent to %define mychar 'yzw'
2409 \c %substr mychar 'xyzw' 2,-2 ; equivalent to %define mychar 'yz'
2411 As with \c{%strlen} (see \k{strlen}), the first parameter is the
2412 single-line macro to be created and the second is the string. The
2413 third parameter specifies the first character to be selected, and the
2414 optional fourth parameter preceeded by comma) is the length. Note
2415 that the first index is 1, not 0 and the last index is equal to the
2416 value that \c{%strlen} would assign given the same string. Index
2417 values out of range result in an empty string. A negative length
2418 means "until N-1 characters before the end of string", i.e. \c{-1}
2419 means until end of string, \c{-2} until one character before, etc.
2422 \H{mlmacro} \i{Multi-Line Macros}: \I\c{%imacro}\i\c{%macro}
2424 Multi-line macros are much more like the type of macro seen in MASM
2425 and TASM: a multi-line macro definition in NASM looks something like
2428 \c %macro prologue 1
2436 This defines a C-like function prologue as a macro: so you would
2437 invoke the macro with a call such as
2439 \c myfunc: prologue 12
2441 which would expand to the three lines of code
2447 The number \c{1} after the macro name in the \c{%macro} line defines
2448 the number of parameters the macro \c{prologue} expects to receive.
2449 The use of \c{%1} inside the macro definition refers to the first
2450 parameter to the macro call. With a macro taking more than one
2451 parameter, subsequent parameters would be referred to as \c{%2},
2454 Multi-line macros, like single-line macros, are \i{case-sensitive},
2455 unless you define them using the alternative directive \c{%imacro}.
2457 If you need to pass a comma as \e{part} of a parameter to a
2458 multi-line macro, you can do that by enclosing the entire parameter
2459 in \I{braces, around macro parameters}braces. So you could code
2468 \c silly 'a', letter_a ; letter_a: db 'a'
2469 \c silly 'ab', string_ab ; string_ab: db 'ab'
2470 \c silly {13,10}, crlf ; crlf: db 13,10
2473 \S{mlmacover} Overloading Multi-Line Macros\I{overloading, multi-line macros}
2475 As with single-line macros, multi-line macros can be overloaded by
2476 defining the same macro name several times with different numbers of
2477 parameters. This time, no exception is made for macros with no
2478 parameters at all. So you could define
2480 \c %macro prologue 0
2487 to define an alternative form of the function prologue which
2488 allocates no local stack space.
2490 Sometimes, however, you might want to `overload' a machine
2491 instruction; for example, you might want to define
2500 so that you could code
2502 \c push ebx ; this line is not a macro call
2503 \c push eax,ecx ; but this one is
2505 Ordinarily, NASM will give a warning for the first of the above two
2506 lines, since \c{push} is now defined to be a macro, and is being
2507 invoked with a number of parameters for which no definition has been
2508 given. The correct code will still be generated, but the assembler
2509 will give a warning. This warning can be disabled by the use of the
2510 \c{-w-macro-params} command-line option (see \k{opt-w}).
2513 \S{maclocal} \i{Macro-Local Labels}
2515 NASM allows you to define labels within a multi-line macro
2516 definition in such a way as to make them local to the macro call: so
2517 calling the same macro multiple times will use a different label
2518 each time. You do this by prefixing \i\c{%%} to the label name. So
2519 you can invent an instruction which executes a \c{RET} if the \c{Z}
2520 flag is set by doing this:
2530 You can call this macro as many times as you want, and every time
2531 you call it NASM will make up a different `real' name to substitute
2532 for the label \c{%%skip}. The names NASM invents are of the form
2533 \c{..@2345.skip}, where the number 2345 changes with every macro
2534 call. The \i\c{..@} prefix prevents macro-local labels from
2535 interfering with the local label mechanism, as described in
2536 \k{locallab}. You should avoid defining your own labels in this form
2537 (the \c{..@} prefix, then a number, then another period) in case
2538 they interfere with macro-local labels.
2541 \S{mlmacgre} \i{Greedy Macro Parameters}
2543 Occasionally it is useful to define a macro which lumps its entire
2544 command line into one parameter definition, possibly after
2545 extracting one or two smaller parameters from the front. An example
2546 might be a macro to write a text string to a file in MS-DOS, where
2547 you might want to be able to write
2549 \c writefile [filehandle],"hello, world",13,10
2551 NASM allows you to define the last parameter of a macro to be
2552 \e{greedy}, meaning that if you invoke the macro with more
2553 parameters than it expects, all the spare parameters get lumped into
2554 the last defined one along with the separating commas. So if you
2557 \c %macro writefile 2+
2563 \c mov cx,%%endstr-%%str
2570 then the example call to \c{writefile} above will work as expected:
2571 the text before the first comma, \c{[filehandle]}, is used as the
2572 first macro parameter and expanded when \c{%1} is referred to, and
2573 all the subsequent text is lumped into \c{%2} and placed after the
2576 The greedy nature of the macro is indicated to NASM by the use of
2577 the \I{+ modifier}\c{+} sign after the parameter count on the
2580 If you define a greedy macro, you are effectively telling NASM how
2581 it should expand the macro given \e{any} number of parameters from
2582 the actual number specified up to infinity; in this case, for
2583 example, NASM now knows what to do when it sees a call to
2584 \c{writefile} with 2, 3, 4 or more parameters. NASM will take this
2585 into account when overloading macros, and will not allow you to
2586 define another form of \c{writefile} taking 4 parameters (for
2589 Of course, the above macro could have been implemented as a
2590 non-greedy macro, in which case the call to it would have had to
2593 \c writefile [filehandle], {"hello, world",13,10}
2595 NASM provides both mechanisms for putting \i{commas in macro
2596 parameters}, and you choose which one you prefer for each macro
2599 See \k{sectmac} for a better way to write the above macro.
2601 \S{mlmacrange} \i{Macro Parameters Range}
2603 NASM allows you to expand parameters via special construction \c{%\{x:y\}}
2604 where \c{x} is the first parameter index and \c{y} is the last. Any index can
2605 be either negative or positive but must never be zero.
2615 expands to \c{3,4,5} range.
2617 Even more, the parameters can be reversed so that
2625 expands to \c{5,4,3} range.
2627 But even this is not the last. The parameters can be addressed via negative
2628 indices so NASM will count them reversed. The ones who know Python may see
2637 expands to \c{6,5,4} range.
2639 Note that NASM uses \i{comma} to separate parameters being expanded.
2641 By the way, here is a trick - you might use the index \c{%{-1:-1}}
2642 which gives you the \i{last} argument passed to a macro.
2644 \S{mlmacdef} \i{Default Macro Parameters}
2646 NASM also allows you to define a multi-line macro with a \e{range}
2647 of allowable parameter counts. If you do this, you can specify
2648 defaults for \i{omitted parameters}. So, for example:
2650 \c %macro die 0-1 "Painful program death has occurred."
2658 This macro (which makes use of the \c{writefile} macro defined in
2659 \k{mlmacgre}) can be called with an explicit error message, which it
2660 will display on the error output stream before exiting, or it can be
2661 called with no parameters, in which case it will use the default
2662 error message supplied in the macro definition.
2664 In general, you supply a minimum and maximum number of parameters
2665 for a macro of this type; the minimum number of parameters are then
2666 required in the macro call, and then you provide defaults for the
2667 optional ones. So if a macro definition began with the line
2669 \c %macro foobar 1-3 eax,[ebx+2]
2671 then it could be called with between one and three parameters, and
2672 \c{%1} would always be taken from the macro call. \c{%2}, if not
2673 specified by the macro call, would default to \c{eax}, and \c{%3} if
2674 not specified would default to \c{[ebx+2]}.
2676 You can provide extra information to a macro by providing
2677 too many default parameters:
2679 \c %macro quux 1 something
2681 This will trigger a warning by default; see \k{opt-w} for
2683 When \c{quux} is invoked, it receives not one but two parameters.
2684 \c{something} can be referred to as \c{%2}. The difference
2685 between passing \c{something} this way and writing \c{something}
2686 in the macro body is that with this way \c{something} is evaluated
2687 when the macro is defined, not when it is expanded.
2689 You may omit parameter defaults from the macro definition, in which
2690 case the parameter default is taken to be blank. This can be useful
2691 for macros which can take a variable number of parameters, since the
2692 \i\c{%0} token (see \k{percent0}) allows you to determine how many
2693 parameters were really passed to the macro call.
2695 This defaulting mechanism can be combined with the greedy-parameter
2696 mechanism; so the \c{die} macro above could be made more powerful,
2697 and more useful, by changing the first line of the definition to
2699 \c %macro die 0-1+ "Painful program death has occurred.",13,10
2701 The maximum parameter count can be infinite, denoted by \c{*}. In
2702 this case, of course, it is impossible to provide a \e{full} set of
2703 default parameters. Examples of this usage are shown in \k{rotate}.
2706 \S{percent0} \i\c{%0}: \I{counting macro parameters}Macro Parameter Counter
2708 The parameter reference \c{%0} will return a numeric constant giving the
2709 number of parameters received, that is, if \c{%0} is n then \c{%}n is the
2710 last parameter. \c{%0} is mostly useful for macros that can take a variable
2711 number of parameters. It can be used as an argument to \c{%rep}
2712 (see \k{rep}) in order to iterate through all the parameters of a macro.
2713 Examples are given in \k{rotate}.
2716 \S{percent00} \i\c{%00}: \I{label preceeding macro}Label Preceeding Macro
2718 \c{%00} will return the label preceeding the macro invocation, if any. The
2719 label must be on the same line as the macro invocation, may be a local label
2720 (see \k{locallab}), and need not end in a colon.
2723 \S{rotate} \i\c{%rotate}: \i{Rotating Macro Parameters}
2725 Unix shell programmers will be familiar with the \I{shift
2726 command}\c{shift} shell command, which allows the arguments passed
2727 to a shell script (referenced as \c{$1}, \c{$2} and so on) to be
2728 moved left by one place, so that the argument previously referenced
2729 as \c{$2} becomes available as \c{$1}, and the argument previously
2730 referenced as \c{$1} is no longer available at all.
2732 NASM provides a similar mechanism, in the form of \c{%rotate}. As
2733 its name suggests, it differs from the Unix \c{shift} in that no
2734 parameters are lost: parameters rotated off the left end of the
2735 argument list reappear on the right, and vice versa.
2737 \c{%rotate} is invoked with a single numeric argument (which may be
2738 an expression). The macro parameters are rotated to the left by that
2739 many places. If the argument to \c{%rotate} is negative, the macro
2740 parameters are rotated to the right.
2742 \I{iterating over macro parameters}So a pair of macros to save and
2743 restore a set of registers might work as follows:
2745 \c %macro multipush 1-*
2754 This macro invokes the \c{PUSH} instruction on each of its arguments
2755 in turn, from left to right. It begins by pushing its first
2756 argument, \c{%1}, then invokes \c{%rotate} to move all the arguments
2757 one place to the left, so that the original second argument is now
2758 available as \c{%1}. Repeating this procedure as many times as there
2759 were arguments (achieved by supplying \c{%0} as the argument to
2760 \c{%rep}) causes each argument in turn to be pushed.
2762 Note also the use of \c{*} as the maximum parameter count,
2763 indicating that there is no upper limit on the number of parameters
2764 you may supply to the \i\c{multipush} macro.
2766 It would be convenient, when using this macro, to have a \c{POP}
2767 equivalent, which \e{didn't} require the arguments to be given in
2768 reverse order. Ideally, you would write the \c{multipush} macro
2769 call, then cut-and-paste the line to where the pop needed to be
2770 done, and change the name of the called macro to \c{multipop}, and
2771 the macro would take care of popping the registers in the opposite
2772 order from the one in which they were pushed.
2774 This can be done by the following definition:
2776 \c %macro multipop 1-*
2785 This macro begins by rotating its arguments one place to the
2786 \e{right}, so that the original \e{last} argument appears as \c{%1}.
2787 This is then popped, and the arguments are rotated right again, so
2788 the second-to-last argument becomes \c{%1}. Thus the arguments are
2789 iterated through in reverse order.
2792 \S{concat} \i{Concatenating Macro Parameters}
2794 NASM can concatenate macro parameters and macro indirection constructs
2795 on to other text surrounding them. This allows you to declare a family
2796 of symbols, for example, in a macro definition. If, for example, you
2797 wanted to generate a table of key codes along with offsets into the
2798 table, you could code something like
2800 \c %macro keytab_entry 2
2802 \c keypos%1 equ $-keytab
2808 \c keytab_entry F1,128+1
2809 \c keytab_entry F2,128+2
2810 \c keytab_entry Return,13
2812 which would expand to
2815 \c keyposF1 equ $-keytab
2817 \c keyposF2 equ $-keytab
2819 \c keyposReturn equ $-keytab
2822 You can just as easily concatenate text on to the other end of a
2823 macro parameter, by writing \c{%1foo}.
2825 If you need to append a \e{digit} to a macro parameter, for example
2826 defining labels \c{foo1} and \c{foo2} when passed the parameter
2827 \c{foo}, you can't code \c{%11} because that would be taken as the
2828 eleventh macro parameter. Instead, you must code
2829 \I{braces, after % sign}\c{%\{1\}1}, which will separate the first
2830 \c{1} (giving the number of the macro parameter) from the second
2831 (literal text to be concatenated to the parameter).
2833 This concatenation can also be applied to other preprocessor in-line
2834 objects, such as macro-local labels (\k{maclocal}) and context-local
2835 labels (\k{ctxlocal}). In all cases, ambiguities in syntax can be
2836 resolved by enclosing everything after the \c{%} sign and before the
2837 literal text in braces: so \c{%\{%foo\}bar} concatenates the text
2838 \c{bar} to the end of the real name of the macro-local label
2839 \c{%%foo}. (This is unnecessary, since the form NASM uses for the
2840 real names of macro-local labels means that the two usages
2841 \c{%\{%foo\}bar} and \c{%%foobar} would both expand to the same
2842 thing anyway; nevertheless, the capability is there.)
2844 The single-line macro indirection construct, \c{%[...]}
2845 (\k{indmacro}), behaves the same way as macro parameters for the
2846 purpose of concatenation.
2848 See also the \c{%+} operator, \k{concat%+}.
2851 \S{mlmaccc} \i{Condition Codes as Macro Parameters}
2853 NASM can give special treatment to a macro parameter which contains
2854 a condition code. For a start, you can refer to the macro parameter
2855 \c{%1} by means of the alternative syntax \i\c{%+1}, which informs
2856 NASM that this macro parameter is supposed to contain a condition
2857 code, and will cause the preprocessor to report an error message if
2858 the macro is called with a parameter which is \e{not} a valid
2861 Far more usefully, though, you can refer to the macro parameter by
2862 means of \i\c{%-1}, which NASM will expand as the \e{inverse}
2863 condition code. So the \c{retz} macro defined in \k{maclocal} can be
2864 replaced by a general \i{conditional-return macro} like this:
2874 This macro can now be invoked using calls like \c{retc ne}, which
2875 will cause the conditional-jump instruction in the macro expansion
2876 to come out as \c{JE}, or \c{retc po} which will make the jump a
2879 The \c{%+1} macro-parameter reference is quite happy to interpret
2880 the arguments \c{CXZ} and \c{ECXZ} as valid condition codes;
2881 however, \c{%-1} will report an error if passed either of these,
2882 because no inverse condition code exists.
2885 \S{nolist} \i{Disabling Listing Expansion}\I\c{.nolist}
2887 When NASM is generating a listing file from your program, it will
2888 generally expand multi-line macros by means of writing the macro
2889 call and then listing each line of the expansion. This allows you to
2890 see which instructions in the macro expansion are generating what
2891 code; however, for some macros this clutters the listing up
2894 NASM therefore provides the \c{.nolist} qualifier, which you can
2895 include in a macro definition to inhibit the expansion of the macro
2896 in the listing file. The \c{.nolist} qualifier comes directly after
2897 the number of parameters, like this:
2899 \c %macro foo 1.nolist
2903 \c %macro bar 1-5+.nolist a,b,c,d,e,f,g,h
2905 \S{unmacro} Undefining Multi-Line Macros: \i\c{%unmacro}
2907 Multi-line macros can be removed with the \c{%unmacro} directive.
2908 Unlike the \c{%undef} directive, however, \c{%unmacro} takes an
2909 argument specification, and will only remove \i{exact matches} with
2910 that argument specification.
2919 removes the previously defined macro \c{foo}, but
2926 does \e{not} remove the macro \c{bar}, since the argument
2927 specification does not match exactly.
2930 \H{condasm} \i{Conditional Assembly}\I\c{%if}
2932 Similarly to the C preprocessor, NASM allows sections of a source
2933 file to be assembled only if certain conditions are met. The general
2934 syntax of this feature looks like this:
2937 \c ; some code which only appears if <condition> is met
2938 \c %elif<condition2>
2939 \c ; only appears if <condition> is not met but <condition2> is
2941 \c ; this appears if neither <condition> nor <condition2> was met
2944 The inverse forms \i\c{%ifn} and \i\c{%elifn} are also supported.
2946 The \i\c{%else} clause is optional, as is the \i\c{%elif} clause.
2947 You can have more than one \c{%elif} clause as well.
2949 There are a number of variants of the \c{%if} directive. Each has its
2950 corresponding \c{%elif}, \c{%ifn}, and \c{%elifn} directives; for
2951 example, the equivalents to the \c{%ifdef} directive are \c{%elifdef},
2952 \c{%ifndef}, and \c{%elifndef}.
2954 \S{ifdef} \i\c{%ifdef}: Testing Single-Line Macro Existence\I{testing,
2955 single-line macro existence}
2957 Beginning a conditional-assembly block with the line \c{%ifdef
2958 MACRO} will assemble the subsequent code if, and only if, a
2959 single-line macro called \c{MACRO} is defined. If not, then the
2960 \c{%elif} and \c{%else} blocks (if any) will be processed instead.
2962 For example, when debugging a program, you might want to write code
2965 \c ; perform some function
2967 \c writefile 2,"Function performed successfully",13,10
2969 \c ; go and do something else
2971 Then you could use the command-line option \c{-dDEBUG} to create a
2972 version of the program which produced debugging messages, and remove
2973 the option to generate the final release version of the program.
2975 You can test for a macro \e{not} being defined by using
2976 \i\c{%ifndef} instead of \c{%ifdef}. You can also test for macro
2977 definitions in \c{%elif} blocks by using \i\c{%elifdef} and
2981 \S{ifmacro} \i\c{%ifmacro}: Testing Multi-Line Macro
2982 Existence\I{testing, multi-line macro existence}
2984 The \c{%ifmacro} directive operates in the same way as the \c{%ifdef}
2985 directive, except that it checks for the existence of a multi-line macro.
2987 For example, you may be working with a large project and not have control
2988 over the macros in a library. You may want to create a macro with one
2989 name if it doesn't already exist, and another name if one with that name
2992 The \c{%ifmacro} is considered true if defining a macro with the given name
2993 and number of arguments would cause a definitions conflict. For example:
2995 \c %ifmacro MyMacro 1-3
2997 \c %error "MyMacro 1-3" causes a conflict with an existing macro.
3001 \c %macro MyMacro 1-3
3003 \c ; insert code to define the macro
3009 This will create the macro "MyMacro 1-3" if no macro already exists which
3010 would conflict with it, and emits a warning if there would be a definition
3013 You can test for the macro not existing by using the \i\c{%ifnmacro} instead
3014 of \c{%ifmacro}. Additional tests can be performed in \c{%elif} blocks by using
3015 \i\c{%elifmacro} and \i\c{%elifnmacro}.
3018 \S{ifctx} \i\c{%ifctx}: Testing the Context Stack\I{testing, context
3021 The conditional-assembly construct \c{%ifctx} will cause the
3022 subsequent code to be assembled if and only if the top context on
3023 the preprocessor's context stack has the same name as one of the arguments.
3024 As with \c{%ifdef}, the inverse and \c{%elif} forms \i\c{%ifnctx},
3025 \i\c{%elifctx} and \i\c{%elifnctx} are also supported.
3027 For more details of the context stack, see \k{ctxstack}. For a
3028 sample use of \c{%ifctx}, see \k{blockif}.
3031 \S{if} \i\c{%if}: Testing Arbitrary Numeric Expressions\I{testing,
3032 arbitrary numeric expressions}
3034 The conditional-assembly construct \c{%if expr} will cause the
3035 subsequent code to be assembled if and only if the value of the
3036 numeric expression \c{expr} is non-zero. An example of the use of
3037 this feature is in deciding when to break out of a \c{%rep}
3038 preprocessor loop: see \k{rep} for a detailed example.
3040 The expression given to \c{%if}, and its counterpart \i\c{%elif}, is
3041 a critical expression (see \k{crit}).
3043 \c{%if} extends the normal NASM expression syntax, by providing a
3044 set of \i{relational operators} which are not normally available in
3045 expressions. The operators \i\c{=}, \i\c{<}, \i\c{>}, \i\c{<=},
3046 \i\c{>=} and \i\c{<>} test equality, less-than, greater-than,
3047 less-or-equal, greater-or-equal and not-equal respectively. The
3048 C-like forms \i\c{==} and \i\c{!=} are supported as alternative
3049 forms of \c{=} and \c{<>}. In addition, low-priority logical
3050 operators \i\c{&&}, \i\c{^^} and \i\c{||} are provided, supplying
3051 \i{logical AND}, \i{logical XOR} and \i{logical OR}. These work like
3052 the C logical operators (although C has no logical XOR), in that
3053 they always return either 0 or 1, and treat any non-zero input as 1
3054 (so that \c{^^}, for example, returns 1 if exactly one of its inputs
3055 is zero, and 0 otherwise). The relational operators also return 1
3056 for true and 0 for false.
3058 Like other \c{%if} constructs, \c{%if} has a counterpart
3059 \i\c{%elif}, and negative forms \i\c{%ifn} and \i\c{%elifn}.
3061 \S{ifidn} \i\c{%ifidn} and \i\c{%ifidni}: Testing Exact Text
3062 Identity\I{testing, exact text identity}
3064 The construct \c{%ifidn text1,text2} will cause the subsequent code
3065 to be assembled if and only if \c{text1} and \c{text2}, after
3066 expanding single-line macros, are identical pieces of text.
3067 Differences in white space are not counted.
3069 \c{%ifidni} is similar to \c{%ifidn}, but is \i{case-insensitive}.
3071 For example, the following macro pushes a register or number on the
3072 stack, and allows you to treat \c{IP} as a real register:
3074 \c %macro pushparam 1
3085 Like other \c{%if} constructs, \c{%ifidn} has a counterpart
3086 \i\c{%elifidn}, and negative forms \i\c{%ifnidn} and \i\c{%elifnidn}.
3087 Similarly, \c{%ifidni} has counterparts \i\c{%elifidni},
3088 \i\c{%ifnidni} and \i\c{%elifnidni}.
3090 \S{iftyp} \i\c{%ifid}, \i\c{%ifnum}, \i\c{%ifstr}: Testing Token
3091 Types\I{testing, token types}
3093 Some macros will want to perform different tasks depending on
3094 whether they are passed a number, a string, or an identifier. For
3095 example, a string output macro might want to be able to cope with
3096 being passed either a string constant or a pointer to an existing
3099 The conditional assembly construct \c{%ifid}, taking one parameter
3100 (which may be blank), assembles the subsequent code if and only if
3101 the first token in the parameter exists and is an identifier.
3102 \c{%ifnum} works similarly, but tests for the token being a numeric
3103 constant; \c{%ifstr} tests for it being a string.
3105 For example, the \c{writefile} macro defined in \k{mlmacgre} can be
3106 extended to take advantage of \c{%ifstr} in the following fashion:
3108 \c %macro writefile 2-3+
3117 \c %%endstr: mov dx,%%str
3118 \c mov cx,%%endstr-%%str
3129 Then the \c{writefile} macro can cope with being called in either of
3130 the following two ways:
3132 \c writefile [file], strpointer, length
3133 \c writefile [file], "hello", 13, 10
3135 In the first, \c{strpointer} is used as the address of an
3136 already-declared string, and \c{length} is used as its length; in
3137 the second, a string is given to the macro, which therefore declares
3138 it itself and works out the address and length for itself.
3140 Note the use of \c{%if} inside the \c{%ifstr}: this is to detect
3141 whether the macro was passed two arguments (so the string would be a
3142 single string constant, and \c{db %2} would be adequate) or more (in
3143 which case, all but the first two would be lumped together into
3144 \c{%3}, and \c{db %2,%3} would be required).
3146 The usual \I\c{%elifid}\I\c{%elifnum}\I\c{%elifstr}\c{%elif}...,
3147 \I\c{%ifnid}\I\c{%ifnnum}\I\c{%ifnstr}\c{%ifn}..., and
3148 \I\c{%elifnid}\I\c{%elifnnum}\I\c{%elifnstr}\c{%elifn}... versions
3149 exist for each of \c{%ifid}, \c{%ifnum} and \c{%ifstr}.
3151 \S{iftoken} \i\c{%iftoken}: Test for a Single Token
3153 Some macros will want to do different things depending on if it is
3154 passed a single token (e.g. paste it to something else using \c{%+})
3155 versus a multi-token sequence.
3157 The conditional assembly construct \c{%iftoken} assembles the
3158 subsequent code if and only if the expanded parameters consist of
3159 exactly one token, possibly surrounded by whitespace.
3165 will assemble the subsequent code, but
3169 will not, since \c{-1} contains two tokens: the unary minus operator
3170 \c{-}, and the number \c{1}.
3172 The usual \i\c{%eliftoken}, \i\c\{%ifntoken}, and \i\c{%elifntoken}
3173 variants are also provided.
3175 \S{ifempty} \i\c{%ifempty}: Test for Empty Expansion
3177 The conditional assembly construct \c{%ifempty} assembles the
3178 subsequent code if and only if the expanded parameters do not contain
3179 any tokens at all, whitespace excepted.
3181 The usual \i\c{%elifempty}, \i\c\{%ifnempty}, and \i\c{%elifnempty}
3182 variants are also provided.
3184 \S{ifenv} \i\c{%ifenv}: Test If Environment Variable Exists
3186 The conditional assembly construct \c{%ifenv} assembles the
3187 subsequent code if and only if the environment variable referenced by
3188 the \c{%!<env>} directive exists.
3190 The usual \i\c{%elifenv}, \i\c\{%ifnenv}, and \i\c{%elifnenv}
3191 variants are also provided.
3193 Just as for \c{%!<env>} the argument should be written as a string if
3194 it contains characters that would not be legal in an identifier. See
3197 \H{rep} \i{Preprocessor Loops}\I{repeating code}: \i\c{%rep}
3199 NASM's \c{TIMES} prefix, though useful, cannot be used to invoke a
3200 multi-line macro multiple times, because it is processed by NASM
3201 after macros have already been expanded. Therefore NASM provides
3202 another form of loop, this time at the preprocessor level: \c{%rep}.
3204 The directives \c{%rep} and \i\c{%endrep} (\c{%rep} takes a numeric
3205 argument, which can be an expression; \c{%endrep} takes no
3206 arguments) can be used to enclose a chunk of code, which is then
3207 replicated as many times as specified by the preprocessor:
3211 \c inc word [table+2*i]
3215 This will generate a sequence of 64 \c{INC} instructions,
3216 incrementing every word of memory from \c{[table]} to
3219 For more complex termination conditions, or to break out of a repeat
3220 loop part way along, you can use the \i\c{%exitrep} directive to
3221 terminate the loop, like this:
3236 \c fib_number equ ($-fibonacci)/2
3238 This produces a list of all the Fibonacci numbers that will fit in
3239 16 bits. Note that a maximum repeat count must still be given to
3240 \c{%rep}. This is to prevent the possibility of NASM getting into an
3241 infinite loop in the preprocessor, which (on multitasking or
3242 multi-user systems) would typically cause all the system memory to
3243 be gradually used up and other applications to start crashing.
3245 Note a maximum repeat count is limited by 62 bit number, though it
3246 is hardly possible that you ever need anything bigger.
3249 \H{files} Source Files and Dependencies
3251 These commands allow you to split your sources into multiple files.
3253 \S{include} \i\c{%include}: \i{Including Other Files}
3255 Using, once again, a very similar syntax to the C preprocessor,
3256 NASM's preprocessor lets you include other source files into your
3257 code. This is done by the use of the \i\c{%include} directive:
3259 \c %include "macros.mac"
3261 will include the contents of the file \c{macros.mac} into the source
3262 file containing the \c{%include} directive.
3264 Include files are \I{searching for include files}searched for in the
3265 current directory (the directory you're in when you run NASM, as
3266 opposed to the location of the NASM executable or the location of
3267 the source file), plus any directories specified on the NASM command
3268 line using the \c{-i} option.
3270 The standard C idiom for preventing a file being included more than
3271 once is just as applicable in NASM: if the file \c{macros.mac} has
3274 \c %ifndef MACROS_MAC
3275 \c %define MACROS_MAC
3276 \c ; now define some macros
3279 then including the file more than once will not cause errors,
3280 because the second time the file is included nothing will happen
3281 because the macro \c{MACROS_MAC} will already be defined.
3283 You can force a file to be included even if there is no \c{%include}
3284 directive that explicitly includes it, by using the \i\c{-p} option
3285 on the NASM command line (see \k{opt-p}).
3288 \S{pathsearch} \i\c{%pathsearch}: Search the Include Path
3290 The \c{%pathsearch} directive takes a single-line macro name and a
3291 filename, and declare or redefines the specified single-line macro to
3292 be the include-path-resolved version of the filename, if the file
3293 exists (otherwise, it is passed unchanged.)
3297 \c %pathsearch MyFoo "foo.bin"
3299 ... with \c{-Ibins/} in the include path may end up defining the macro
3300 \c{MyFoo} to be \c{"bins/foo.bin"}.
3303 \S{depend} \i\c{%depend}: Add Dependent Files
3305 The \c{%depend} directive takes a filename and adds it to the list of
3306 files to be emitted as dependency generation when the \c{-M} options
3307 and its relatives (see \k{opt-M}) are used. It produces no output.
3309 This is generally used in conjunction with \c{%pathsearch}. For
3310 example, a simplified version of the standard macro wrapper for the
3311 \c{INCBIN} directive looks like:
3313 \c %imacro incbin 1-2+ 0
3314 \c %pathsearch dep %1
3319 This first resolves the location of the file into the macro \c{dep},
3320 then adds it to the dependency lists, and finally issues the
3321 assembler-level \c{INCBIN} directive.
3324 \S{use} \i\c{%use}: Include Standard Macro Package
3326 The \c{%use} directive is similar to \c{%include}, but rather than
3327 including the contents of a file, it includes a named standard macro
3328 package. The standard macro packages are part of NASM, and are
3329 described in \k{macropkg}.
3331 Unlike the \c{%include} directive, package names for the \c{%use}
3332 directive do not require quotes, but quotes are permitted. In NASM
3333 2.04 and 2.05 the unquoted form would be macro-expanded; this is no
3334 longer true. Thus, the following lines are equivalent:
3339 Standard macro packages are protected from multiple inclusion. When a
3340 standard macro package is used, a testable single-line macro of the
3341 form \c{__USE_}\e{package}\c{__} is also defined, see \k{use_def}.
3343 \H{ctxstack} The \i{Context Stack}
3345 Having labels that are local to a macro definition is sometimes not
3346 quite powerful enough: sometimes you want to be able to share labels
3347 between several macro calls. An example might be a \c{REPEAT} ...
3348 \c{UNTIL} loop, in which the expansion of the \c{REPEAT} macro
3349 would need to be able to refer to a label which the \c{UNTIL} macro
3350 had defined. However, for such a macro you would also want to be
3351 able to nest these loops.
3353 NASM provides this level of power by means of a \e{context stack}.
3354 The preprocessor maintains a stack of \e{contexts}, each of which is
3355 characterized by a name. You add a new context to the stack using
3356 the \i\c{%push} directive, and remove one using \i\c{%pop}. You can
3357 define labels that are local to a particular context on the stack.
3360 \S{pushpop} \i\c{%push} and \i\c{%pop}: \I{creating
3361 contexts}\I{removing contexts}Creating and Removing Contexts
3363 The \c{%push} directive is used to create a new context and place it
3364 on the top of the context stack. \c{%push} takes an optional argument,
3365 which is the name of the context. For example:
3369 This pushes a new context called \c{foobar} on the stack. You can have
3370 several contexts on the stack with the same name: they can still be
3371 distinguished. If no name is given, the context is unnamed (this is
3372 normally used when both the \c{%push} and the \c{%pop} are inside a
3373 single macro definition.)
3375 The directive \c{%pop}, taking one optional argument, removes the top
3376 context from the context stack and destroys it, along with any
3377 labels associated with it. If an argument is given, it must match the
3378 name of the current context, otherwise it will issue an error.
3381 \S{ctxlocal} \i{Context-Local Labels}
3383 Just as the usage \c{%%foo} defines a label which is local to the
3384 particular macro call in which it is used, the usage \I{%$}\c{%$foo}
3385 is used to define a label which is local to the context on the top
3386 of the context stack. So the \c{REPEAT} and \c{UNTIL} example given
3387 above could be implemented by means of:
3403 and invoked by means of, for example,
3411 which would scan every fourth byte of a string in search of the byte
3414 If you need to define, or access, labels local to the context
3415 \e{below} the top one on the stack, you can use \I{%$$}\c{%$$foo}, or
3416 \c{%$$$foo} for the context below that, and so on.
3419 \S{ctxdefine} \i{Context-Local Single-Line Macros}
3421 NASM also allows you to define single-line macros which are local to
3422 a particular context, in just the same way:
3424 \c %define %$localmac 3
3426 will define the single-line macro \c{%$localmac} to be local to the
3427 top context on the stack. Of course, after a subsequent \c{%push},
3428 it can then still be accessed by the name \c{%$$localmac}.
3431 \S{ctxfallthrough} \i{Context Fall-Through Lookup}
3433 Context fall-through lookup (automatic searching of outer contexts)
3434 is a feature that was added in NASM version 0.98.03. Unfortunately,
3435 this feature is unintuitive and can result in buggy code that would
3436 have otherwise been prevented by NASM's error reporting. As a result,
3437 this feature has been \e{deprecated}. NASM version 2.09 will issue a
3438 warning when usage of this \e{deprecated} feature is detected. Starting
3439 with NASM version 2.10, usage of this \e{deprecated} feature will simply
3440 result in an \e{expression syntax error}.
3442 An example usage of this \e{deprecated} feature follows:
3446 \c %assign %$external 1
3448 \c %assign %$internal 1
3449 \c mov eax, %$external
3450 \c mov eax, %$internal
3455 As demonstrated, \c{%$external} is being defined in the \c{ctx1}
3456 context and referenced within the \c{ctx2} context. With context
3457 fall-through lookup, referencing an undefined context-local macro
3458 like this implicitly searches through all outer contexts until a match
3459 is made or isn't found in any context. As a result, \c{%$external}
3460 referenced within the \c{ctx2} context would implicitly use \c{%$external}
3461 as defined in \c{ctx1}. Most people would expect NASM to issue an error in
3462 this situation because \c{%$external} was never defined within \c{ctx2} and also
3463 isn't qualified with the proper context depth, \c{%$$external}.
3465 Here is a revision of the above example with proper context depth:
3469 \c %assign %$external 1
3471 \c %assign %$internal 1
3472 \c mov eax, %$$external
3473 \c mov eax, %$internal
3478 As demonstrated, \c{%$external} is still being defined in the \c{ctx1}
3479 context and referenced within the \c{ctx2} context. However, the
3480 reference to \c{%$external} within \c{ctx2} has been fully qualified with
3481 the proper context depth, \c{%$$external}, and thus is no longer ambiguous,
3482 unintuitive or erroneous.
3485 \S{ctxrepl} \i\c{%repl}: \I{renaming contexts}Renaming a Context
3487 If you need to change the name of the top context on the stack (in
3488 order, for example, to have it respond differently to \c{%ifctx}),
3489 you can execute a \c{%pop} followed by a \c{%push}; but this will
3490 have the side effect of destroying all context-local labels and
3491 macros associated with the context that was just popped.
3493 NASM provides the directive \c{%repl}, which \e{replaces} a context
3494 with a different name, without touching the associated macros and
3495 labels. So you could replace the destructive code
3500 with the non-destructive version \c{%repl newname}.
3503 \S{blockif} Example Use of the \i{Context Stack}: \i{Block IFs}
3505 This example makes use of almost all the context-stack features,
3506 including the conditional-assembly construct \i\c{%ifctx}, to
3507 implement a block IF statement as a set of macros.
3523 \c %error "expected `if' before `else'"
3537 \c %error "expected `if' or `else' before `endif'"
3542 This code is more robust than the \c{REPEAT} and \c{UNTIL} macros
3543 given in \k{ctxlocal}, because it uses conditional assembly to check
3544 that the macros are issued in the right order (for example, not
3545 calling \c{endif} before \c{if}) and issues a \c{%error} if they're
3548 In addition, the \c{endif} macro has to be able to cope with the two
3549 distinct cases of either directly following an \c{if}, or following
3550 an \c{else}. It achieves this, again, by using conditional assembly
3551 to do different things depending on whether the context on top of
3552 the stack is \c{if} or \c{else}.
3554 The \c{else} macro has to preserve the context on the stack, in
3555 order to have the \c{%$ifnot} referred to by the \c{if} macro be the
3556 same as the one defined by the \c{endif} macro, but has to change
3557 the context's name so that \c{endif} will know there was an
3558 intervening \c{else}. It does this by the use of \c{%repl}.
3560 A sample usage of these macros might look like:
3582 The block-\c{IF} macros handle nesting quite happily, by means of
3583 pushing another context, describing the inner \c{if}, on top of the
3584 one describing the outer \c{if}; thus \c{else} and \c{endif} always
3585 refer to the last unmatched \c{if} or \c{else}.
3588 \H{stackrel} \i{Stack Relative Preprocessor Directives}
3590 The following preprocessor directives provide a way to use
3591 labels to refer to local variables allocated on the stack.
3593 \b\c{%arg} (see \k{arg})
3595 \b\c{%stacksize} (see \k{stacksize})
3597 \b\c{%local} (see \k{local})
3600 \S{arg} \i\c{%arg} Directive
3602 The \c{%arg} directive is used to simplify the handling of
3603 parameters passed on the stack. Stack based parameter passing
3604 is used by many high level languages, including C, C++ and Pascal.
3606 While NASM has macros which attempt to duplicate this
3607 functionality (see \k{16cmacro}), the syntax is not particularly
3608 convenient to use and is not TASM compatible. Here is an example
3609 which shows the use of \c{%arg} without any external macros:
3613 \c %push mycontext ; save the current context
3614 \c %stacksize large ; tell NASM to use bp
3615 \c %arg i:word, j_ptr:word
3622 \c %pop ; restore original context
3624 This is similar to the procedure defined in \k{16cmacro} and adds
3625 the value in i to the value pointed to by j_ptr and returns the
3626 sum in the ax register. See \k{pushpop} for an explanation of
3627 \c{push} and \c{pop} and the use of context stacks.
3630 \S{stacksize} \i\c{%stacksize} Directive
3632 The \c{%stacksize} directive is used in conjunction with the
3633 \c{%arg} (see \k{arg}) and the \c{%local} (see \k{local}) directives.
3634 It tells NASM the default size to use for subsequent \c{%arg} and
3635 \c{%local} directives. The \c{%stacksize} directive takes one
3636 required argument which is one of \c{flat}, \c{flat64}, \c{large} or \c{small}.
3640 This form causes NASM to use stack-based parameter addressing
3641 relative to \c{ebp} and it assumes that a near form of call was used
3642 to get to this label (i.e. that \c{eip} is on the stack).
3644 \c %stacksize flat64
3646 This form causes NASM to use stack-based parameter addressing
3647 relative to \c{rbp} and it assumes that a near form of call was used
3648 to get to this label (i.e. that \c{rip} is on the stack).
3652 This form uses \c{bp} to do stack-based parameter addressing and
3653 assumes that a far form of call was used to get to this address
3654 (i.e. that \c{ip} and \c{cs} are on the stack).
3658 This form also uses \c{bp} to address stack parameters, but it is
3659 different from \c{large} because it also assumes that the old value
3660 of bp is pushed onto the stack (i.e. it expects an \c{ENTER}
3661 instruction). In other words, it expects that \c{bp}, \c{ip} and
3662 \c{cs} are on the top of the stack, underneath any local space which
3663 may have been allocated by \c{ENTER}. This form is probably most
3664 useful when used in combination with the \c{%local} directive
3668 \S{local} \i\c{%local} Directive
3670 The \c{%local} directive is used to simplify the use of local
3671 temporary stack variables allocated in a stack frame. Automatic
3672 local variables in C are an example of this kind of variable. The
3673 \c{%local} directive is most useful when used with the \c{%stacksize}
3674 (see \k{stacksize} and is also compatible with the \c{%arg} directive
3675 (see \k{arg}). It allows simplified reference to variables on the
3676 stack which have been allocated typically by using the \c{ENTER}
3678 \# (see \k{insENTER} for a description of that instruction).
3679 An example of its use is the following:
3683 \c %push mycontext ; save the current context
3684 \c %stacksize small ; tell NASM to use bp
3685 \c %assign %$localsize 0 ; see text for explanation
3686 \c %local old_ax:word, old_dx:word
3688 \c enter %$localsize,0 ; see text for explanation
3689 \c mov [old_ax],ax ; swap ax & bx
3690 \c mov [old_dx],dx ; and swap dx & cx
3695 \c leave ; restore old bp
3698 \c %pop ; restore original context
3700 The \c{%$localsize} variable is used internally by the
3701 \c{%local} directive and \e{must} be defined within the
3702 current context before the \c{%local} directive may be used.
3703 Failure to do so will result in one expression syntax error for
3704 each \c{%local} variable declared. It then may be used in
3705 the construction of an appropriately sized ENTER instruction
3706 as shown in the example.
3709 \H{pperror} Reporting \i{User-Defined Errors}: \i\c{%error}, \i\c{%warning}, \i\c{%fatal}
3711 The preprocessor directive \c{%error} will cause NASM to report an
3712 error if it occurs in assembled code. So if other users are going to
3713 try to assemble your source files, you can ensure that they define the
3714 right macros by means of code like this:
3719 \c ; do some different setup
3721 \c %error "Neither F1 nor F2 was defined."
3724 Then any user who fails to understand the way your code is supposed
3725 to be assembled will be quickly warned of their mistake, rather than
3726 having to wait until the program crashes on being run and then not
3727 knowing what went wrong.
3729 Similarly, \c{%warning} issues a warning, but allows assembly to continue:
3734 \c ; do some different setup
3736 \c %warning "Neither F1 nor F2 was defined, assuming F1."
3740 \c{%error} and \c{%warning} are issued only on the final assembly
3741 pass. This makes them safe to use in conjunction with tests that
3742 depend on symbol values.
3744 \c{%fatal} terminates assembly immediately, regardless of pass. This
3745 is useful when there is no point in continuing the assembly further,
3746 and doing so is likely just going to cause a spew of confusing error
3749 It is optional for the message string after \c{%error}, \c{%warning}
3750 or \c{%fatal} to be quoted. If it is \e{not}, then single-line macros
3751 are expanded in it, which can be used to display more information to
3752 the user. For example:
3755 \c %assign foo_over foo-64
3756 \c %error foo is foo_over bytes too large
3760 \H{otherpreproc} \i{Other Preprocessor Directives}
3762 NASM also has preprocessor directives which allow access to
3763 information from external sources. Currently they include:
3765 \b\c{%line} enables NASM to correctly handle the output of another
3766 preprocessor (see \k{line}).
3768 \b\c{%!} enables NASM to read in the value of an environment variable,
3769 which can then be used in your program (see \k{getenv}).
3771 \S{line} \i\c{%line} Directive
3773 The \c{%line} directive is used to notify NASM that the input line
3774 corresponds to a specific line number in another file. Typically
3775 this other file would be an original source file, with the current
3776 NASM input being the output of a pre-processor. The \c{%line}
3777 directive allows NASM to output messages which indicate the line
3778 number of the original source file, instead of the file that is being
3781 This preprocessor directive is not generally of use to programmers,
3782 by may be of interest to preprocessor authors. The usage of the
3783 \c{%line} preprocessor directive is as follows:
3785 \c %line nnn[+mmm] [filename]
3787 In this directive, \c{nnn} identifies the line of the original source
3788 file which this line corresponds to. \c{mmm} is an optional parameter
3789 which specifies a line increment value; each line of the input file
3790 read in is considered to correspond to \c{mmm} lines of the original
3791 source file. Finally, \c{filename} is an optional parameter which
3792 specifies the file name of the original source file.
3794 After reading a \c{%line} preprocessor directive, NASM will report
3795 all file name and line numbers relative to the values specified
3799 \S{getenv} \i\c{%!}\c{<env>}: Read an environment variable.
3801 The \c{%!<env>} directive makes it possible to read the value of an
3802 environment variable at assembly time. This could, for example, be used
3803 to store the contents of an environment variable into a string, which
3804 could be used at some other point in your code.
3806 For example, suppose that you have an environment variable \c{FOO}, and
3807 you want the contents of \c{FOO} to be embedded in your program. You
3808 could do that as follows:
3810 \c %defstr FOO %!FOO
3812 See \k{defstr} for notes on the \c{%defstr} directive.
3814 If the name of the environment variable contains non-identifier
3815 characters, you can use string quotes to surround the name of the
3816 variable, for example:
3818 \c %defstr C_colon %!'C:'
3821 \H{comment} Comment Blocks: \i\c{%comment}
3823 The \c{%comment} and \c{%endcomment} directives are used to specify
3824 a block of commented (i.e. unprocessed) code/text. Everything between
3825 \c{%comment} and \c{%endcomment} will be ignored by the preprocessor.
3828 \c ; some code, text or data to be ignored
3832 \H{stdmac} \i{Standard Macros}
3834 NASM defines a set of standard macros, which are already defined
3835 when it starts to process any source file. If you really need a
3836 program to be assembled with no pre-defined macros, you can use the
3837 \i\c{%clear} directive to empty the preprocessor of everything but
3838 context-local preprocessor variables and single-line macros.
3840 Most \i{user-level assembler directives} (see \k{directive}) are
3841 implemented as macros which invoke primitive directives; these are
3842 described in \k{directive}. The rest of the standard macro set is
3846 \S{stdmacver} \i{NASM Version} Macros
3848 The single-line macros \i\c{__NASM_MAJOR__}, \i\c{__NASM_MINOR__},
3849 \i\c{__NASM_SUBMINOR__} and \i\c{___NASM_PATCHLEVEL__} expand to the
3850 major, minor, subminor and patch level parts of the \i{version
3851 number of NASM} being used. So, under NASM 0.98.32p1 for
3852 example, \c{__NASM_MAJOR__} would be defined to be 0, \c{__NASM_MINOR__}
3853 would be defined as 98, \c{__NASM_SUBMINOR__} would be defined to 32,
3854 and \c{___NASM_PATCHLEVEL__} would be defined as 1.
3856 Additionally, the macro \i\c{__NASM_SNAPSHOT__} is defined for
3857 automatically generated snapshot releases \e{only}.
3860 \S{stdmacverid} \i\c{__NASM_VERSION_ID__}: \i{NASM Version ID}
3862 The single-line macro \c{__NASM_VERSION_ID__} expands to a dword integer
3863 representing the full version number of the version of nasm being used.
3864 The value is the equivalent to \c{__NASM_MAJOR__}, \c{__NASM_MINOR__},
3865 \c{__NASM_SUBMINOR__} and \c{___NASM_PATCHLEVEL__} concatenated to
3866 produce a single doubleword. Hence, for 0.98.32p1, the returned number
3867 would be equivalent to:
3875 Note that the above lines are generate exactly the same code, the second
3876 line is used just to give an indication of the order that the separate
3877 values will be present in memory.
3880 \S{stdmacverstr} \i\c{__NASM_VER__}: \i{NASM Version string}
3882 The single-line macro \c{__NASM_VER__} expands to a string which defines
3883 the version number of nasm being used. So, under NASM 0.98.32 for example,
3892 \S{fileline} \i\c{__FILE__} and \i\c{__LINE__}: File Name and Line Number
3894 Like the C preprocessor, NASM allows the user to find out the file
3895 name and line number containing the current instruction. The macro
3896 \c{__FILE__} expands to a string constant giving the name of the
3897 current input file (which may change through the course of assembly
3898 if \c{%include} directives are used), and \c{__LINE__} expands to a
3899 numeric constant giving the current line number in the input file.
3901 These macros could be used, for example, to communicate debugging
3902 information to a macro, since invoking \c{__LINE__} inside a macro
3903 definition (either single-line or multi-line) will return the line
3904 number of the macro \e{call}, rather than \e{definition}. So to
3905 determine where in a piece of code a crash is occurring, for
3906 example, one could write a routine \c{stillhere}, which is passed a
3907 line number in \c{EAX} and outputs something like `line 155: still
3908 here'. You could then write a macro
3910 \c %macro notdeadyet 0
3919 and then pepper your code with calls to \c{notdeadyet} until you
3920 find the crash point.
3923 \S{bitsm} \i\c{__BITS__}: Current BITS Mode
3925 The \c{__BITS__} standard macro is updated every time that the BITS mode is
3926 set using the \c{BITS XX} or \c{[BITS XX]} directive, where XX is a valid mode
3927 number of 16, 32 or 64. \c{__BITS__} receives the specified mode number and
3928 makes it globally available. This can be very useful for those who utilize
3929 mode-dependent macros.
3931 \S{ofmtm} \i\c{__OUTPUT_FORMAT__}: Current Output Format
3933 The \c{__OUTPUT_FORMAT__} standard macro holds the current Output Format,
3934 as given by the \c{-f} option or NASM's default. Type \c{nasm -hf} for a
3937 \c %ifidn __OUTPUT_FORMAT__, win32
3938 \c %define NEWLINE 13, 10
3939 \c %elifidn __OUTPUT_FORMAT__, elf32
3940 \c %define NEWLINE 10
3944 \S{datetime} Assembly Date and Time Macros
3946 NASM provides a variety of macros that represent the timestamp of the
3949 \b The \i\c{__DATE__} and \i\c{__TIME__} macros give the assembly date and
3950 time as strings, in ISO 8601 format (\c{"YYYY-MM-DD"} and \c{"HH:MM:SS"},
3953 \b The \i\c{__DATE_NUM__} and \i\c{__TIME_NUM__} macros give the assembly
3954 date and time in numeric form; in the format \c{YYYYMMDD} and
3955 \c{HHMMSS} respectively.
3957 \b The \i\c{__UTC_DATE__} and \i\c{__UTC_TIME__} macros give the assembly
3958 date and time in universal time (UTC) as strings, in ISO 8601 format
3959 (\c{"YYYY-MM-DD"} and \c{"HH:MM:SS"}, respectively.) If the host
3960 platform doesn't provide UTC time, these macros are undefined.
3962 \b The \i\c{__UTC_DATE_NUM__} and \i\c{__UTC_TIME_NUM__} macros give the
3963 assembly date and time universal time (UTC) in numeric form; in the
3964 format \c{YYYYMMDD} and \c{HHMMSS} respectively. If the
3965 host platform doesn't provide UTC time, these macros are
3968 \b The \c{__POSIX_TIME__} macro is defined as a number containing the
3969 number of seconds since the POSIX epoch, 1 January 1970 00:00:00 UTC;
3970 excluding any leap seconds. This is computed using UTC time if
3971 available on the host platform, otherwise it is computed using the
3972 local time as if it was UTC.
3974 All instances of time and date macros in the same assembly session
3975 produce consistent output. For example, in an assembly session
3976 started at 42 seconds after midnight on January 1, 2010 in Moscow
3977 (timezone UTC+3) these macros would have the following values,
3978 assuming, of course, a properly configured environment with a correct
3981 \c __DATE__ "2010-01-01"
3982 \c __TIME__ "00:00:42"
3983 \c __DATE_NUM__ 20100101
3984 \c __TIME_NUM__ 000042
3985 \c __UTC_DATE__ "2009-12-31"
3986 \c __UTC_TIME__ "21:00:42"
3987 \c __UTC_DATE_NUM__ 20091231
3988 \c __UTC_TIME_NUM__ 210042
3989 \c __POSIX_TIME__ 1262293242
3992 \S{use_def} \I\c{__USE_*__}\c{__USE_}\e{package}\c{__}: Package
3995 When a standard macro package (see \k{macropkg}) is included with the
3996 \c{%use} directive (see \k{use}), a single-line macro of the form
3997 \c{__USE_}\e{package}\c{__} is automatically defined. This allows
3998 testing if a particular package is invoked or not.
4000 For example, if the \c{altreg} package is included (see
4001 \k{pkg_altreg}), then the macro \c{__USE_ALTREG__} is defined.
4004 \S{pass_macro} \i\c{__PASS__}: Assembly Pass
4006 The macro \c{__PASS__} is defined to be \c{1} on preparatory passes,
4007 and \c{2} on the final pass. In preprocess-only mode, it is set to
4008 \c{3}, and when running only to generate dependencies (due to the
4009 \c{-M} or \c{-MG} option, see \k{opt-M}) it is set to \c{0}.
4011 \e{Avoid using this macro if at all possible. It is tremendously easy
4012 to generate very strange errors by misusing it, and the semantics may
4013 change in future versions of NASM.}
4016 \S{struc} \i\c{STRUC} and \i\c{ENDSTRUC}: \i{Declaring Structure} Data Types
4018 The core of NASM contains no intrinsic means of defining data
4019 structures; instead, the preprocessor is sufficiently powerful that
4020 data structures can be implemented as a set of macros. The macros
4021 \c{STRUC} and \c{ENDSTRUC} are used to define a structure data type.
4023 \c{STRUC} takes one or two parameters. The first parameter is the name
4024 of the data type. The second, optional parameter is the base offset of
4025 the structure. The name of the data type is defined as a symbol with
4026 the value of the base offset, and the name of the data type with the
4027 suffix \c{_size} appended to it is defined as an \c{EQU} giving the
4028 size of the structure. Once \c{STRUC} has been issued, you are
4029 defining the structure, and should define fields using the \c{RESB}
4030 family of pseudo-instructions, and then invoke \c{ENDSTRUC} to finish
4033 For example, to define a structure called \c{mytype} containing a
4034 longword, a word, a byte and a string of bytes, you might code
4045 The above code defines six symbols: \c{mt_long} as 0 (the offset
4046 from the beginning of a \c{mytype} structure to the longword field),
4047 \c{mt_word} as 4, \c{mt_byte} as 6, \c{mt_str} as 7, \c{mytype_size}
4048 as 39, and \c{mytype} itself as zero.
4050 The reason why the structure type name is defined at zero by default
4051 is a side effect of allowing structures to work with the local label
4052 mechanism: if your structure members tend to have the same names in
4053 more than one structure, you can define the above structure like this:
4064 This defines the offsets to the structure fields as \c{mytype.long},
4065 \c{mytype.word}, \c{mytype.byte} and \c{mytype.str}.
4067 NASM, since it has no \e{intrinsic} structure support, does not
4068 support any form of period notation to refer to the elements of a
4069 structure once you have one (except the above local-label notation),
4070 so code such as \c{mov ax,[mystruc.mt_word]} is not valid.
4071 \c{mt_word} is a constant just like any other constant, so the
4072 correct syntax is \c{mov ax,[mystruc+mt_word]} or \c{mov
4073 ax,[mystruc+mytype.word]}.
4075 Sometimes you only have the address of the structure displaced by an
4076 offset. For example, consider this standard stack frame setup:
4082 In this case, you could access an element by subtracting the offset:
4084 \c mov [ebp - 40 + mytype.word], ax
4086 However, if you do not want to repeat this offset, you can use -40 as
4089 \c struc mytype, -40
4091 And access an element this way:
4093 \c mov [ebp + mytype.word], ax
4096 \S{istruc} \i\c{ISTRUC}, \i\c{AT} and \i\c{IEND}: Declaring
4097 \i{Instances of Structures}
4099 Having defined a structure type, the next thing you typically want
4100 to do is to declare instances of that structure in your data
4101 segment. NASM provides an easy way to do this in the \c{ISTRUC}
4102 mechanism. To declare a structure of type \c{mytype} in a program,
4103 you code something like this:
4108 \c at mt_long, dd 123456
4109 \c at mt_word, dw 1024
4110 \c at mt_byte, db 'x'
4111 \c at mt_str, db 'hello, world', 13, 10, 0
4115 The function of the \c{AT} macro is to make use of the \c{TIMES}
4116 prefix to advance the assembly position to the correct point for the
4117 specified structure field, and then to declare the specified data.
4118 Therefore the structure fields must be declared in the same order as
4119 they were specified in the structure definition.
4121 If the data to go in a structure field requires more than one source
4122 line to specify, the remaining source lines can easily come after
4123 the \c{AT} line. For example:
4125 \c at mt_str, db 123,134,145,156,167,178,189
4128 Depending on personal taste, you can also omit the code part of the
4129 \c{AT} line completely, and start the structure field on the next
4133 \c db 'hello, world'
4137 \S{align} \i\c{ALIGN} and \i\c{ALIGNB}: Data Alignment
4139 The \c{ALIGN} and \c{ALIGNB} macros provides a convenient way to
4140 align code or data on a word, longword, paragraph or other boundary.
4141 (Some assemblers call this directive \i\c{EVEN}.) The syntax of the
4142 \c{ALIGN} and \c{ALIGNB} macros is
4144 \c align 4 ; align on 4-byte boundary
4145 \c align 16 ; align on 16-byte boundary
4146 \c align 8,db 0 ; pad with 0s rather than NOPs
4147 \c align 4,resb 1 ; align to 4 in the BSS
4148 \c alignb 4 ; equivalent to previous line
4150 Both macros require their first argument to be a power of two; they
4151 both compute the number of additional bytes required to bring the
4152 length of the current section up to a multiple of that power of two,
4153 and then apply the \c{TIMES} prefix to their second argument to
4154 perform the alignment.
4156 If the second argument is not specified, the default for \c{ALIGN}
4157 is \c{NOP}, and the default for \c{ALIGNB} is \c{RESB 1}. So if the
4158 second argument is specified, the two macros are equivalent.
4159 Normally, you can just use \c{ALIGN} in code and data sections and
4160 \c{ALIGNB} in BSS sections, and never need the second argument
4161 except for special purposes.
4163 \c{ALIGN} and \c{ALIGNB}, being simple macros, perform no error
4164 checking: they cannot warn you if their first argument fails to be a
4165 power of two, or if their second argument generates more than one
4166 byte of code. In each of these cases they will silently do the wrong
4169 \c{ALIGNB} (or \c{ALIGN} with a second argument of \c{RESB 1}) can
4170 be used within structure definitions:
4187 This will ensure that the structure members are sensibly aligned
4188 relative to the base of the structure.
4190 A final caveat: \c{ALIGN} and \c{ALIGNB} work relative to the
4191 beginning of the \e{section}, not the beginning of the address space
4192 in the final executable. Aligning to a 16-byte boundary when the
4193 section you're in is only guaranteed to be aligned to a 4-byte
4194 boundary, for example, is a waste of effort. Again, NASM does not
4195 check that the section's alignment characteristics are sensible for
4196 the use of \c{ALIGN} or \c{ALIGNB}.
4198 Both \c{ALIGN} and \c{ALIGNB} do call \c{SECTALIGN} macro implicitly.
4199 See \k{sectalign} for details.
4201 See also the \c{smartalign} standard macro package, \k{pkg_smartalign}.
4204 \S{sectalign} \i\c{SECTALIGN}: Section Alignment
4206 The \c{SECTALIGN} macros provides a way to modify alignment attribute
4207 of output file section. Unlike the \c{align=} attribute (which is allowed
4208 at section definition only) the \c{SECTALIGN} macro may be used at any time.
4210 For example the directive
4214 sets the section alignment requirements to 16 bytes. Once increased it can
4215 not be decreased, the magnitude may grow only.
4217 Note that \c{ALIGN} (see \k{align}) calls the \c{SECTALIGN} macro implicitly
4218 so the active section alignment requirements may be updated. This is by default
4219 behaviour, if for some reason you want the \c{ALIGN} do not call \c{SECTALIGN}
4220 at all use the directive
4224 It is still possible to turn in on again by
4229 \C{macropkg} \i{Standard Macro Packages}
4231 The \i\c{%use} directive (see \k{use}) includes one of the standard
4232 macro packages included with the NASM distribution and compiled into
4233 the NASM binary. It operates like the \c{%include} directive (see
4234 \k{include}), but the included contents is provided by NASM itself.
4236 The names of standard macro packages are case insensitive, and can be
4240 \H{pkg_altreg} \i\c{altreg}: \i{Alternate Register Names}
4242 The \c{altreg} standard macro package provides alternate register
4243 names. It provides numeric register names for all registers (not just
4244 \c{R8}-\c{R15}), the Intel-defined aliases \c{R8L}-\c{R15L} for the
4245 low bytes of register (as opposed to the NASM/AMD standard names
4246 \c{R8B}-\c{R15B}), and the names \c{R0H}-\c{R3H} (by analogy with
4247 \c{R0L}-\c{R3L}) for \c{AH}, \c{CH}, \c{DH}, and \c{BH}.
4254 \c mov r0l,r3h ; mov al,bh
4260 \H{pkg_smartalign} \i\c{smartalign}\I{align, smart}: Smart \c{ALIGN} Macro
4262 The \c{smartalign} standard macro package provides for an \i\c{ALIGN}
4263 macro which is more powerful than the default (and
4264 backwards-compatible) one (see \k{align}). When the \c{smartalign}
4265 package is enabled, when \c{ALIGN} is used without a second argument,
4266 NASM will generate a sequence of instructions more efficient than a
4267 series of \c{NOP}. Furthermore, if the padding exceeds a specific
4268 threshold, then NASM will generate a jump over the entire padding
4271 The specific instructions generated can be controlled with the
4272 new \i\c{ALIGNMODE} macro. This macro takes two parameters: one mode,
4273 and an optional jump threshold override. If (for any reason) you need
4274 to turn off the jump completely just set jump threshold value to -1
4275 (or set it to \c{nojmp}). The following modes are possible:
4277 \b \c{generic}: Works on all x86 CPUs and should have reasonable
4278 performance. The default jump threshold is 8. This is the
4281 \b \c{nop}: Pad out with \c{NOP} instructions. The only difference
4282 compared to the standard \c{ALIGN} macro is that NASM can still jump
4283 over a large padding area. The default jump threshold is 16.
4285 \b \c{k7}: Optimize for the AMD K7 (Athlon/Althon XP). These
4286 instructions should still work on all x86 CPUs. The default jump
4289 \b \c{k8}: Optimize for the AMD K8 (Opteron/Althon 64). These
4290 instructions should still work on all x86 CPUs. The default jump
4293 \b \c{p6}: Optimize for Intel CPUs. This uses the long \c{NOP}
4294 instructions first introduced in Pentium Pro. This is incompatible
4295 with all CPUs of family 5 or lower, as well as some VIA CPUs and
4296 several virtualization solutions. The default jump threshold is 16.
4298 The macro \i\c{__ALIGNMODE__} is defined to contain the current
4299 alignment mode. A number of other macros beginning with \c{__ALIGN_}
4300 are used internally by this macro package.
4303 \H{pkg_fp} \i\c\{fp}: Floating-point macros
4305 This packages contains the following floating-point convenience macros:
4307 \c %define Inf __Infinity__
4308 \c %define NaN __QNaN__
4309 \c %define QNaN __QNaN__
4310 \c %define SNaN __SNaN__
4312 \c %define float8(x) __float8__(x)
4313 \c %define float16(x) __float16__(x)
4314 \c %define float32(x) __float32__(x)
4315 \c %define float64(x) __float64__(x)
4316 \c %define float80m(x) __float80m__(x)
4317 \c %define float80e(x) __float80e__(x)
4318 \c %define float128l(x) __float128l__(x)
4319 \c %define float128h(x) __float128h__(x)
4322 \H{pkg_ifunc} \i\c{ifunc}: \i{Integer functions}
4324 This package contains a set of macros which implement integer
4325 functions. These are actually implemented as special operators, but
4326 are most conveniently accessed via this macro package.
4328 The macros provided are:
4330 \S{ilog2} \i{Integer logarithms}
4332 These functions calculate the integer logarithm base 2 of their
4333 argument, considered as an unsigned integer. The only differences
4334 between the functions is their behavior if the argument provided is
4337 The function \i\c{ilog2e()} (alias \i\c{ilog2()}) generate an error if
4338 the argument is not a power of two.
4340 The function \i\c{ilog2w()} generate a warning if the argument is not
4343 The function \i\c{ilog2f()} rounds the argument down to the nearest
4344 power of two; if the argument is zero it returns zero.
4346 The function \i\c{ilog2c()} rounds the argument up to the nearest
4350 \C{directive} \i{Assembler Directives}
4352 NASM, though it attempts to avoid the bureaucracy of assemblers like
4353 MASM and TASM, is nevertheless forced to support a \e{few}
4354 directives. These are described in this chapter.
4356 NASM's directives come in two types: \I{user-level
4357 directives}\e{user-level} directives and \I{primitive
4358 directives}\e{primitive} directives. Typically, each directive has a
4359 user-level form and a primitive form. In almost all cases, we
4360 recommend that users use the user-level forms of the directives,
4361 which are implemented as macros which call the primitive forms.
4363 Primitive directives are enclosed in square brackets; user-level
4366 In addition to the universal directives described in this chapter,
4367 each object file format can optionally supply extra directives in
4368 order to control particular features of that file format. These
4369 \I{format-specific directives}\e{format-specific} directives are
4370 documented along with the formats that implement them, in \k{outfmt}.
4373 \H{bits} \i\c{BITS}: Specifying Target \i{Processor Mode}
4375 The \c{BITS} directive specifies whether NASM should generate code
4376 \I{16-bit mode, versus 32-bit mode}designed to run on a processor
4377 operating in 16-bit mode, 32-bit mode or 64-bit mode. The syntax is
4378 \c{BITS XX}, where XX is 16, 32 or 64.
4380 In most cases, you should not need to use \c{BITS} explicitly. The
4381 \c{aout}, \c{coff}, \c{elf}, \c{macho}, \c{win32} and \c{win64}
4382 object formats, which are designed for use in 32-bit or 64-bit
4383 operating systems, all cause NASM to select 32-bit or 64-bit mode,
4384 respectively, by default. The \c{obj} object format allows you
4385 to specify each segment you define as either \c{USE16} or \c{USE32},
4386 and NASM will set its operating mode accordingly, so the use of the
4387 \c{BITS} directive is once again unnecessary.
4389 The most likely reason for using the \c{BITS} directive is to write
4390 32-bit or 64-bit code in a flat binary file; this is because the \c{bin}
4391 output format defaults to 16-bit mode in anticipation of it being
4392 used most frequently to write DOS \c{.COM} programs, DOS \c{.SYS}
4393 device drivers and boot loader software.
4395 You do \e{not} need to specify \c{BITS 32} merely in order to use
4396 32-bit instructions in a 16-bit DOS program; if you do, the
4397 assembler will generate incorrect code because it will be writing
4398 code targeted at a 32-bit platform, to be run on a 16-bit one.
4400 When NASM is in \c{BITS 16} mode, instructions which use 32-bit
4401 data are prefixed with an 0x66 byte, and those referring to 32-bit
4402 addresses have an 0x67 prefix. In \c{BITS 32} mode, the reverse is
4403 true: 32-bit instructions require no prefixes, whereas instructions
4404 using 16-bit data need an 0x66 and those working on 16-bit addresses
4407 When NASM is in \c{BITS 64} mode, most instructions operate the same
4408 as they do for \c{BITS 32} mode. However, there are 8 more general and
4409 SSE registers, and 16-bit addressing is no longer supported.
4411 The default address size is 64 bits; 32-bit addressing can be selected
4412 with the 0x67 prefix. The default operand size is still 32 bits,
4413 however, and the 0x66 prefix selects 16-bit operand size. The \c{REX}
4414 prefix is used both to select 64-bit operand size, and to access the
4415 new registers. NASM automatically inserts REX prefixes when
4418 When the \c{REX} prefix is used, the processor does not know how to
4419 address the AH, BH, CH or DH (high 8-bit legacy) registers. Instead,
4420 it is possible to access the the low 8-bits of the SP, BP SI and DI
4421 registers as SPL, BPL, SIL and DIL, respectively; but only when the
4424 The \c{BITS} directive has an exactly equivalent primitive form,
4425 \c{[BITS 16]}, \c{[BITS 32]} and \c{[BITS 64]}. The user-level form is
4426 a macro which has no function other than to call the primitive form.
4428 Note that the space is neccessary, e.g. \c{BITS32} will \e{not} work!
4430 \S{USE16 & USE32} \i\c{USE16} & \i\c{USE32}: Aliases for BITS
4432 The `\c{USE16}' and `\c{USE32}' directives can be used in place of
4433 `\c{BITS 16}' and `\c{BITS 32}', for compatibility with other assemblers.
4436 \H{default} \i\c{DEFAULT}: Change the assembler defaults
4438 The \c{DEFAULT} directive changes the assembler defaults. Normally,
4439 NASM defaults to a mode where the programmer is expected to explicitly
4440 specify most features directly. However, this is occationally
4441 obnoxious, as the explicit form is pretty much the only one one wishes
4444 Currently, \c{DEFAULT} can set \c{REL} & \c{ABS} and \c{BND} & \c{NOBND}.
4446 \S{REL & ABS} \i\c{REL} & \i\c{ABS}: RIP-relative addressing
4448 This sets whether registerless instructions in 64-bit mode are \c{RIP}-relative
4449 or not. By default, they are absolute unless overridden with the \i\c{REL}
4450 specifier (see \k{effaddr}). However, if \c{DEFAULT REL} is
4451 specified, \c{REL} is default, unless overridden with the \c{ABS}
4452 specifier, \e{except when used with an FS or GS segment override}.
4454 The special handling of \c{FS} and \c{GS} overrides are due to the
4455 fact that these registers are generally used as thread pointers or
4456 other special functions in 64-bit mode, and generating
4457 \c{RIP}-relative addresses would be extremely confusing.
4459 \c{DEFAULT REL} is disabled with \c{DEFAULT ABS}.
4461 \S{BND & NOBND} \i\c{BND} & \i\c{NOBND}: \c{BND} prefix
4463 If \c{DEFAULT BND} is set, all bnd-prefix available instructions following
4464 this directive are prefixed with bnd. To override it, \c{NOBND} prefix can
4468 \c call foo ; BND will be prefixed
4469 \c nobnd call foo ; BND will NOT be prefixed
4471 \c{DEFAULT NOBND} can disable \c{DEFAULT BND} and then \c{BND} prefix will be
4472 added only when explicitly specified in code.
4474 \H{section} \i\c{SECTION} or \i\c{SEGMENT}: Changing and \i{Defining
4477 \I{changing sections}\I{switching between sections}The \c{SECTION}
4478 directive (\c{SEGMENT} is an exactly equivalent synonym) changes
4479 which section of the output file the code you write will be
4480 assembled into. In some object file formats, the number and names of
4481 sections are fixed; in others, the user may make up as many as they
4482 wish. Hence \c{SECTION} may sometimes give an error message, or may
4483 define a new section, if you try to switch to a section that does
4486 The Unix object formats, and the \c{bin} object format (but see
4487 \k{multisec}, all support
4488 the \i{standardized section names} \c{.text}, \c{.data} and \c{.bss}
4489 for the code, data and uninitialized-data sections. The \c{obj}
4490 format, by contrast, does not recognize these section names as being
4491 special, and indeed will strip off the leading period of any section
4495 \S{sectmac} The \i\c{__SECT__} Macro
4497 The \c{SECTION} directive is unusual in that its user-level form
4498 functions differently from its primitive form. The primitive form,
4499 \c{[SECTION xyz]}, simply switches the current target section to the
4500 one given. The user-level form, \c{SECTION xyz}, however, first
4501 defines the single-line macro \c{__SECT__} to be the primitive
4502 \c{[SECTION]} directive which it is about to issue, and then issues
4503 it. So the user-level directive
4507 expands to the two lines
4509 \c %define __SECT__ [SECTION .text]
4512 Users may find it useful to make use of this in their own macros.
4513 For example, the \c{writefile} macro defined in \k{mlmacgre} can be
4514 usefully rewritten in the following more sophisticated form:
4516 \c %macro writefile 2+
4526 \c mov cx,%%endstr-%%str
4533 This form of the macro, once passed a string to output, first
4534 switches temporarily to the data section of the file, using the
4535 primitive form of the \c{SECTION} directive so as not to modify
4536 \c{__SECT__}. It then declares its string in the data section, and
4537 then invokes \c{__SECT__} to switch back to \e{whichever} section
4538 the user was previously working in. It thus avoids the need, in the
4539 previous version of the macro, to include a \c{JMP} instruction to
4540 jump over the data, and also does not fail if, in a complicated
4541 \c{OBJ} format module, the user could potentially be assembling the
4542 code in any of several separate code sections.
4545 \H{absolute} \i\c{ABSOLUTE}: Defining Absolute Labels
4547 The \c{ABSOLUTE} directive can be thought of as an alternative form
4548 of \c{SECTION}: it causes the subsequent code to be directed at no
4549 physical section, but at the hypothetical section starting at the
4550 given absolute address. The only instructions you can use in this
4551 mode are the \c{RESB} family.
4553 \c{ABSOLUTE} is used as follows:
4561 This example describes a section of the PC BIOS data area, at
4562 segment address 0x40: the above code defines \c{kbuf_chr} to be
4563 0x1A, \c{kbuf_free} to be 0x1C, and \c{kbuf} to be 0x1E.
4565 The user-level form of \c{ABSOLUTE}, like that of \c{SECTION},
4566 redefines the \i\c{__SECT__} macro when it is invoked.
4568 \i\c{STRUC} and \i\c{ENDSTRUC} are defined as macros which use
4569 \c{ABSOLUTE} (and also \c{__SECT__}).
4571 \c{ABSOLUTE} doesn't have to take an absolute constant as an
4572 argument: it can take an expression (actually, a \i{critical
4573 expression}: see \k{crit}) and it can be a value in a segment. For
4574 example, a TSR can re-use its setup code as run-time BSS like this:
4576 \c org 100h ; it's a .COM program
4578 \c jmp setup ; setup code comes last
4580 \c ; the resident part of the TSR goes here
4582 \c ; now write the code that installs the TSR here
4586 \c runtimevar1 resw 1
4587 \c runtimevar2 resd 20
4591 This defines some variables `on top of' the setup code, so that
4592 after the setup has finished running, the space it took up can be
4593 re-used as data storage for the running TSR. The symbol `tsr_end'
4594 can be used to calculate the total size of the part of the TSR that
4595 needs to be made resident.
4598 \H{extern} \i\c{EXTERN}: \i{Importing Symbols} from Other Modules
4600 \c{EXTERN} is similar to the MASM directive \c{EXTRN} and the C
4601 keyword \c{extern}: it is used to declare a symbol which is not
4602 defined anywhere in the module being assembled, but is assumed to be
4603 defined in some other module and needs to be referred to by this
4604 one. Not every object-file format can support external variables:
4605 the \c{bin} format cannot.
4607 The \c{EXTERN} directive takes as many arguments as you like. Each
4608 argument is the name of a symbol:
4611 \c extern _sscanf,_fscanf
4613 Some object-file formats provide extra features to the \c{EXTERN}
4614 directive. In all cases, the extra features are used by suffixing a
4615 colon to the symbol name followed by object-format specific text.
4616 For example, the \c{obj} format allows you to declare that the
4617 default segment base of an external should be the group \c{dgroup}
4618 by means of the directive
4620 \c extern _variable:wrt dgroup
4622 The primitive form of \c{EXTERN} differs from the user-level form
4623 only in that it can take only one argument at a time: the support
4624 for multiple arguments is implemented at the preprocessor level.
4626 You can declare the same variable as \c{EXTERN} more than once: NASM
4627 will quietly ignore the second and later redeclarations. You can't
4628 declare a variable as \c{EXTERN} as well as something else, though.
4631 \H{global} \i\c{GLOBAL}: \i{Exporting Symbols} to Other Modules
4633 \c{GLOBAL} is the other end of \c{EXTERN}: if one module declares a
4634 symbol as \c{EXTERN} and refers to it, then in order to prevent
4635 linker errors, some other module must actually \e{define} the
4636 symbol and declare it as \c{GLOBAL}. Some assemblers use the name
4637 \i\c{PUBLIC} for this purpose.
4639 The \c{GLOBAL} directive applying to a symbol must appear \e{before}
4640 the definition of the symbol.
4642 \c{GLOBAL} uses the same syntax as \c{EXTERN}, except that it must
4643 refer to symbols which \e{are} defined in the same module as the
4644 \c{GLOBAL} directive. For example:
4650 \c{GLOBAL}, like \c{EXTERN}, allows object formats to define private
4651 extensions by means of a colon. The \c{elf} object format, for
4652 example, lets you specify whether global data items are functions or
4655 \c global hashlookup:function, hashtable:data
4657 Like \c{EXTERN}, the primitive form of \c{GLOBAL} differs from the
4658 user-level form only in that it can take only one argument at a
4662 \H{common} \i\c{COMMON}: Defining Common Data Areas
4664 The \c{COMMON} directive is used to declare \i\e{common variables}.
4665 A common variable is much like a global variable declared in the
4666 uninitialized data section, so that
4670 is similar in function to
4677 The difference is that if more than one module defines the same
4678 common variable, then at link time those variables will be
4679 \e{merged}, and references to \c{intvar} in all modules will point
4680 at the same piece of memory.
4682 Like \c{GLOBAL} and \c{EXTERN}, \c{COMMON} supports object-format
4683 specific extensions. For example, the \c{obj} format allows common
4684 variables to be NEAR or FAR, and the \c{elf} format allows you to
4685 specify the alignment requirements of a common variable:
4687 \c common commvar 4:near ; works in OBJ
4688 \c common intarray 100:4 ; works in ELF: 4 byte aligned
4690 Once again, like \c{EXTERN} and \c{GLOBAL}, the primitive form of
4691 \c{COMMON} differs from the user-level form only in that it can take
4692 only one argument at a time.
4695 \H{CPU} \i\c{CPU}: Defining CPU Dependencies
4697 The \i\c{CPU} directive restricts assembly to those instructions which
4698 are available on the specified CPU.
4702 \b\c{CPU 8086} Assemble only 8086 instruction set
4704 \b\c{CPU 186} Assemble instructions up to the 80186 instruction set
4706 \b\c{CPU 286} Assemble instructions up to the 286 instruction set
4708 \b\c{CPU 386} Assemble instructions up to the 386 instruction set
4710 \b\c{CPU 486} 486 instruction set
4712 \b\c{CPU 586} Pentium instruction set
4714 \b\c{CPU PENTIUM} Same as 586
4716 \b\c{CPU 686} P6 instruction set
4718 \b\c{CPU PPRO} Same as 686
4720 \b\c{CPU P2} Same as 686
4722 \b\c{CPU P3} Pentium III (Katmai) instruction sets
4724 \b\c{CPU KATMAI} Same as P3
4726 \b\c{CPU P4} Pentium 4 (Willamette) instruction set
4728 \b\c{CPU WILLAMETTE} Same as P4
4730 \b\c{CPU PRESCOTT} Prescott instruction set
4732 \b\c{CPU X64} x86-64 (x64/AMD64/Intel 64) instruction set
4734 \b\c{CPU IA64} IA64 CPU (in x86 mode) instruction set
4736 All options are case insensitive. All instructions will be selected
4737 only if they apply to the selected CPU or lower. By default, all
4738 instructions are available.
4741 \H{FLOAT} \i\c{FLOAT}: Handling of \I{floating-point, constants}floating-point constants
4743 By default, floating-point constants are rounded to nearest, and IEEE
4744 denormals are supported. The following options can be set to alter
4747 \b\c{FLOAT DAZ} Flush denormals to zero
4749 \b\c{FLOAT NODAZ} Do not flush denormals to zero (default)
4751 \b\c{FLOAT NEAR} Round to nearest (default)
4753 \b\c{FLOAT UP} Round up (toward +Infinity)
4755 \b\c{FLOAT DOWN} Round down (toward -Infinity)
4757 \b\c{FLOAT ZERO} Round toward zero
4759 \b\c{FLOAT DEFAULT} Restore default settings
4761 The standard macros \i\c{__FLOAT_DAZ__}, \i\c{__FLOAT_ROUND__}, and
4762 \i\c{__FLOAT__} contain the current state, as long as the programmer
4763 has avoided the use of the brackeded primitive form, (\c{[FLOAT]}).
4765 \c{__FLOAT__} contains the full set of floating-point settings; this
4766 value can be saved away and invoked later to restore the setting.
4769 \C{outfmt} \i{Output Formats}
4771 NASM is a portable assembler, designed to be able to compile on any
4772 ANSI C-supporting platform and produce output to run on a variety of
4773 Intel x86 operating systems. For this reason, it has a large number
4774 of available output formats, selected using the \i\c{-f} option on
4775 the NASM \i{command line}. Each of these formats, along with its
4776 extensions to the base NASM syntax, is detailed in this chapter.
4778 As stated in \k{opt-o}, NASM chooses a \i{default name} for your
4779 output file based on the input file name and the chosen output
4780 format. This will be generated by removing the \i{extension}
4781 (\c{.asm}, \c{.s}, or whatever you like to use) from the input file
4782 name, and substituting an extension defined by the output format.
4783 The extensions are given with each format below.
4786 \H{binfmt} \i\c{bin}: \i{Flat-Form Binary}\I{pure binary} Output
4788 The \c{bin} format does not produce object files: it generates
4789 nothing in the output file except the code you wrote. Such `pure
4790 binary' files are used by \i{MS-DOS}: \i\c{.COM} executables and
4791 \i\c{.SYS} device drivers are pure binary files. Pure binary output
4792 is also useful for \i{operating system} and \i{boot loader}
4795 The \c{bin} format supports \i{multiple section names}. For details of
4796 how NASM handles sections in the \c{bin} format, see \k{multisec}.
4798 Using the \c{bin} format puts NASM by default into 16-bit mode (see
4799 \k{bits}). In order to use \c{bin} to write 32-bit or 64-bit code,
4800 such as an OS kernel, you need to explicitly issue the \I\c{BITS}\c{BITS 32}
4801 or \I\c{BITS}\c{BITS 64} directive.
4803 \c{bin} has no default output file name extension: instead, it
4804 leaves your file name as it is once the original extension has been
4805 removed. Thus, the default is for NASM to assemble \c{binprog.asm}
4806 into a binary file called \c{binprog}.
4809 \S{org} \i\c{ORG}: Binary File \i{Program Origin}
4811 The \c{bin} format provides an additional directive to the list
4812 given in \k{directive}: \c{ORG}. The function of the \c{ORG}
4813 directive is to specify the origin address which NASM will assume
4814 the program begins at when it is loaded into memory.
4816 For example, the following code will generate the longword
4823 Unlike the \c{ORG} directive provided by MASM-compatible assemblers,
4824 which allows you to jump around in the object file and overwrite
4825 code you have already generated, NASM's \c{ORG} does exactly what
4826 the directive says: \e{origin}. Its sole function is to specify one
4827 offset which is added to all internal address references within the
4828 section; it does not permit any of the trickery that MASM's version
4829 does. See \k{proborg} for further comments.
4832 \S{binseg} \c{bin} Extensions to the \c{SECTION}
4833 Directive\I{SECTION, bin extensions to}
4835 The \c{bin} output format extends the \c{SECTION} (or \c{SEGMENT})
4836 directive to allow you to specify the alignment requirements of
4837 segments. This is done by appending the \i\c{ALIGN} qualifier to the
4838 end of the section-definition line. For example,
4840 \c section .data align=16
4842 switches to the section \c{.data} and also specifies that it must be
4843 aligned on a 16-byte boundary.
4845 The parameter to \c{ALIGN} specifies how many low bits of the
4846 section start address must be forced to zero. The alignment value
4847 given may be any power of two.\I{section alignment, in
4848 bin}\I{segment alignment, in bin}\I{alignment, in bin sections}
4851 \S{multisec} \i{Multisection}\I{bin, multisection} Support for the \c{bin} Format
4853 The \c{bin} format allows the use of multiple sections, of arbitrary names,
4854 besides the "known" \c{.text}, \c{.data}, and \c{.bss} names.
4856 \b Sections may be designated \i\c{progbits} or \i\c{nobits}. Default
4857 is \c{progbits} (except \c{.bss}, which defaults to \c{nobits},
4860 \b Sections can be aligned at a specified boundary following the previous
4861 section with \c{align=}, or at an arbitrary byte-granular position with
4864 \b Sections can be given a virtual start address, which will be used
4865 for the calculation of all memory references within that section
4868 \b Sections can be ordered using \i\c{follows=}\c{<section>} or
4869 \i\c{vfollows=}\c{<section>} as an alternative to specifying an explicit
4872 \b Arguments to \c{org}, \c{start}, \c{vstart}, and \c{align=} are
4873 critical expressions. See \k{crit}. E.g. \c{align=(1 << ALIGN_SHIFT)}
4874 - \c{ALIGN_SHIFT} must be defined before it is used here.
4876 \b Any code which comes before an explicit \c{SECTION} directive
4877 is directed by default into the \c{.text} section.
4879 \b If an \c{ORG} statement is not given, \c{ORG 0} is used
4882 \b The \c{.bss} section will be placed after the last \c{progbits}
4883 section, unless \c{start=}, \c{vstart=}, \c{follows=}, or \c{vfollows=}
4886 \b All sections are aligned on dword boundaries, unless a different
4887 alignment has been specified.
4889 \b Sections may not overlap.
4891 \b NASM creates the \c{section.<secname>.start} for each section,
4892 which may be used in your code.
4894 \S{map}\i{Map Files}
4896 Map files can be generated in \c{-f bin} format by means of the \c{[map]}
4897 option. Map types of \c{all} (default), \c{brief}, \c{sections}, \c{segments},
4898 or \c{symbols} may be specified. Output may be directed to \c{stdout}
4899 (default), \c{stderr}, or a specified file. E.g.
4900 \c{[map symbols myfile.map]}. No "user form" exists, the square
4901 brackets must be used.
4904 \H{ithfmt} \i\c{ith}: \i{Intel Hex} Output
4906 The \c{ith} file format produces Intel hex-format files. Just as the
4907 \c{bin} format, this is a flat memory image format with no support for
4908 relocation or linking. It is usually used with ROM programmers and
4911 All extensions supported by the \c{bin} file format is also supported by
4912 the \c{ith} file format.
4914 \c{ith} provides a default output file-name extension of \c{.ith}.
4917 \H{srecfmt} \i\c{srec}: \i{Motorola S-Records} Output
4919 The \c{srec} file format produces Motorola S-records files. Just as the
4920 \c{bin} format, this is a flat memory image format with no support for
4921 relocation or linking. It is usually used with ROM programmers and
4924 All extensions supported by the \c{bin} file format is also supported by
4925 the \c{srec} file format.
4927 \c{srec} provides a default output file-name extension of \c{.srec}.
4930 \H{objfmt} \i\c{obj}: \i{Microsoft OMF}\I{OMF} Object Files
4932 The \c{obj} file format (NASM calls it \c{obj} rather than \c{omf}
4933 for historical reasons) is the one produced by \i{MASM} and
4934 \i{TASM}, which is typically fed to 16-bit DOS linkers to produce
4935 \i\c{.EXE} files. It is also the format used by \i{OS/2}.
4937 \c{obj} provides a default output file-name extension of \c{.obj}.
4939 \c{obj} is not exclusively a 16-bit format, though: NASM has full
4940 support for the 32-bit extensions to the format. In particular,
4941 32-bit \c{obj} format files are used by \i{Borland's Win32
4942 compilers}, instead of using Microsoft's newer \i\c{win32} object
4945 The \c{obj} format does not define any special segment names: you
4946 can call your segments anything you like. Typical names for segments
4947 in \c{obj} format files are \c{CODE}, \c{DATA} and \c{BSS}.
4949 If your source file contains code before specifying an explicit
4950 \c{SEGMENT} directive, then NASM will invent its own segment called
4951 \i\c{__NASMDEFSEG} for you.
4953 When you define a segment in an \c{obj} file, NASM defines the
4954 segment name as a symbol as well, so that you can access the segment
4955 address of the segment. So, for example:
4964 \c mov ax,data ; get segment address of data
4965 \c mov ds,ax ; and move it into DS
4966 \c inc word [dvar] ; now this reference will work
4969 The \c{obj} format also enables the use of the \i\c{SEG} and
4970 \i\c{WRT} operators, so that you can write code which does things
4975 \c mov ax,seg foo ; get preferred segment of foo
4977 \c mov ax,data ; a different segment
4979 \c mov ax,[ds:foo] ; this accesses `foo'
4980 \c mov [es:foo wrt data],bx ; so does this
4983 \S{objseg} \c{obj} Extensions to the \c{SEGMENT}
4984 Directive\I{SEGMENT, obj extensions to}
4986 The \c{obj} output format extends the \c{SEGMENT} (or \c{SECTION})
4987 directive to allow you to specify various properties of the segment
4988 you are defining. This is done by appending extra qualifiers to the
4989 end of the segment-definition line. For example,
4991 \c segment code private align=16
4993 defines the segment \c{code}, but also declares it to be a private
4994 segment, and requires that the portion of it described in this code
4995 module must be aligned on a 16-byte boundary.
4997 The available qualifiers are:
4999 \b \i\c{PRIVATE}, \i\c{PUBLIC}, \i\c{COMMON} and \i\c{STACK} specify
5000 the combination characteristics of the segment. \c{PRIVATE} segments
5001 do not get combined with any others by the linker; \c{PUBLIC} and
5002 \c{STACK} segments get concatenated together at link time; and
5003 \c{COMMON} segments all get overlaid on top of each other rather
5004 than stuck end-to-end.
5006 \b \i\c{ALIGN} is used, as shown above, to specify how many low bits
5007 of the segment start address must be forced to zero. The alignment
5008 value given may be any power of two from 1 to 4096; in reality, the
5009 only values supported are 1, 2, 4, 16, 256 and 4096, so if 8 is
5010 specified it will be rounded up to 16, and 32, 64 and 128 will all
5011 be rounded up to 256, and so on. Note that alignment to 4096-byte
5012 boundaries is a \i{PharLap} extension to the format and may not be
5013 supported by all linkers.\I{section alignment, in OBJ}\I{segment
5014 alignment, in OBJ}\I{alignment, in OBJ sections}
5016 \b \i\c{CLASS} can be used to specify the segment class; this feature
5017 indicates to the linker that segments of the same class should be
5018 placed near each other in the output file. The class name can be any
5019 word, e.g. \c{CLASS=CODE}.
5021 \b \i\c{OVERLAY}, like \c{CLASS}, is specified with an arbitrary word
5022 as an argument, and provides overlay information to an
5023 overlay-capable linker.
5025 \b Segments can be declared as \i\c{USE16} or \i\c{USE32}, which has
5026 the effect of recording the choice in the object file and also
5027 ensuring that NASM's default assembly mode when assembling in that
5028 segment is 16-bit or 32-bit respectively.
5030 \b When writing \i{OS/2} object files, you should declare 32-bit
5031 segments as \i\c{FLAT}, which causes the default segment base for
5032 anything in the segment to be the special group \c{FLAT}, and also
5033 defines the group if it is not already defined.
5035 \b The \c{obj} file format also allows segments to be declared as
5036 having a pre-defined absolute segment address, although no linkers
5037 are currently known to make sensible use of this feature;
5038 nevertheless, NASM allows you to declare a segment such as
5039 \c{SEGMENT SCREEN ABSOLUTE=0xB800} if you need to. The \i\c{ABSOLUTE}
5040 and \c{ALIGN} keywords are mutually exclusive.
5042 NASM's default segment attributes are \c{PUBLIC}, \c{ALIGN=1}, no
5043 class, no overlay, and \c{USE16}.
5046 \S{group} \i\c{GROUP}: Defining Groups of Segments\I{segments, groups of}
5048 The \c{obj} format also allows segments to be grouped, so that a
5049 single segment register can be used to refer to all the segments in
5050 a group. NASM therefore supplies the \c{GROUP} directive, whereby
5059 \c ; some uninitialized data
5061 \c group dgroup data bss
5063 which will define a group called \c{dgroup} to contain the segments
5064 \c{data} and \c{bss}. Like \c{SEGMENT}, \c{GROUP} causes the group
5065 name to be defined as a symbol, so that you can refer to a variable
5066 \c{var} in the \c{data} segment as \c{var wrt data} or as \c{var wrt
5067 dgroup}, depending on which segment value is currently in your
5070 If you just refer to \c{var}, however, and \c{var} is declared in a
5071 segment which is part of a group, then NASM will default to giving
5072 you the offset of \c{var} from the beginning of the \e{group}, not
5073 the \e{segment}. Therefore \c{SEG var}, also, will return the group
5074 base rather than the segment base.
5076 NASM will allow a segment to be part of more than one group, but
5077 will generate a warning if you do this. Variables declared in a
5078 segment which is part of more than one group will default to being
5079 relative to the first group that was defined to contain the segment.
5081 A group does not have to contain any segments; you can still make
5082 \c{WRT} references to a group which does not contain the variable
5083 you are referring to. OS/2, for example, defines the special group
5084 \c{FLAT} with no segments in it.
5087 \S{uppercase} \i\c{UPPERCASE}: Disabling Case Sensitivity in Output
5089 Although NASM itself is \i{case sensitive}, some OMF linkers are
5090 not; therefore it can be useful for NASM to output single-case
5091 object files. The \c{UPPERCASE} format-specific directive causes all
5092 segment, group and symbol names that are written to the object file
5093 to be forced to upper case just before being written. Within a
5094 source file, NASM is still case-sensitive; but the object file can
5095 be written entirely in upper case if desired.
5097 \c{UPPERCASE} is used alone on a line; it requires no parameters.
5100 \S{import} \i\c{IMPORT}: Importing DLL Symbols\I{DLL symbols,
5101 importing}\I{symbols, importing from DLLs}
5103 The \c{IMPORT} format-specific directive defines a symbol to be
5104 imported from a DLL, for use if you are writing a DLL's \i{import
5105 library} in NASM. You still need to declare the symbol as \c{EXTERN}
5106 as well as using the \c{IMPORT} directive.
5108 The \c{IMPORT} directive takes two required parameters, separated by
5109 white space, which are (respectively) the name of the symbol you
5110 wish to import and the name of the library you wish to import it
5113 \c import WSAStartup wsock32.dll
5115 A third optional parameter gives the name by which the symbol is
5116 known in the library you are importing it from, in case this is not
5117 the same as the name you wish the symbol to be known by to your code
5118 once you have imported it. For example:
5120 \c import asyncsel wsock32.dll WSAAsyncSelect
5123 \S{export} \i\c{EXPORT}: Exporting DLL Symbols\I{DLL symbols,
5124 exporting}\I{symbols, exporting from DLLs}
5126 The \c{EXPORT} format-specific directive defines a global symbol to
5127 be exported as a DLL symbol, for use if you are writing a DLL in
5128 NASM. You still need to declare the symbol as \c{GLOBAL} as well as
5129 using the \c{EXPORT} directive.
5131 \c{EXPORT} takes one required parameter, which is the name of the
5132 symbol you wish to export, as it was defined in your source file. An
5133 optional second parameter (separated by white space from the first)
5134 gives the \e{external} name of the symbol: the name by which you
5135 wish the symbol to be known to programs using the DLL. If this name
5136 is the same as the internal name, you may leave the second parameter
5139 Further parameters can be given to define attributes of the exported
5140 symbol. These parameters, like the second, are separated by white
5141 space. If further parameters are given, the external name must also
5142 be specified, even if it is the same as the internal name. The
5143 available attributes are:
5145 \b \c{resident} indicates that the exported name is to be kept
5146 resident by the system loader. This is an optimisation for
5147 frequently used symbols imported by name.
5149 \b \c{nodata} indicates that the exported symbol is a function which
5150 does not make use of any initialized data.
5152 \b \c{parm=NNN}, where \c{NNN} is an integer, sets the number of
5153 parameter words for the case in which the symbol is a call gate
5154 between 32-bit and 16-bit segments.
5156 \b An attribute which is just a number indicates that the symbol
5157 should be exported with an identifying number (ordinal), and gives
5163 \c export myfunc TheRealMoreFormalLookingFunctionName
5164 \c export myfunc myfunc 1234 ; export by ordinal
5165 \c export myfunc myfunc resident parm=23 nodata
5168 \S{dotdotstart} \i\c{..start}: Defining the \i{Program Entry
5171 \c{OMF} linkers require exactly one of the object files being linked to
5172 define the program entry point, where execution will begin when the
5173 program is run. If the object file that defines the entry point is
5174 assembled using NASM, you specify the entry point by declaring the
5175 special symbol \c{..start} at the point where you wish execution to
5179 \S{objextern} \c{obj} Extensions to the \c{EXTERN}
5180 Directive\I{EXTERN, obj extensions to}
5182 If you declare an external symbol with the directive
5186 then references such as \c{mov ax,foo} will give you the offset of
5187 \c{foo} from its preferred segment base (as specified in whichever
5188 module \c{foo} is actually defined in). So to access the contents of
5189 \c{foo} you will usually need to do something like
5191 \c mov ax,seg foo ; get preferred segment base
5192 \c mov es,ax ; move it into ES
5193 \c mov ax,[es:foo] ; and use offset `foo' from it
5195 This is a little unwieldy, particularly if you know that an external
5196 is going to be accessible from a given segment or group, say
5197 \c{dgroup}. So if \c{DS} already contained \c{dgroup}, you could
5200 \c mov ax,[foo wrt dgroup]
5202 However, having to type this every time you want to access \c{foo}
5203 can be a pain; so NASM allows you to declare \c{foo} in the
5206 \c extern foo:wrt dgroup
5208 This form causes NASM to pretend that the preferred segment base of
5209 \c{foo} is in fact \c{dgroup}; so the expression \c{seg foo} will
5210 now return \c{dgroup}, and the expression \c{foo} is equivalent to
5213 This \I{default-WRT mechanism}default-\c{WRT} mechanism can be used
5214 to make externals appear to be relative to any group or segment in
5215 your program. It can also be applied to common variables: see
5219 \S{objcommon} \c{obj} Extensions to the \c{COMMON}
5220 Directive\I{COMMON, obj extensions to}
5222 The \c{obj} format allows common variables to be either near\I{near
5223 common variables} or far\I{far common variables}; NASM allows you to
5224 specify which your variables should be by the use of the syntax
5226 \c common nearvar 2:near ; `nearvar' is a near common
5227 \c common farvar 10:far ; and `farvar' is far
5229 Far common variables may be greater in size than 64Kb, and so the
5230 OMF specification says that they are declared as a number of
5231 \e{elements} of a given size. So a 10-byte far common variable could
5232 be declared as ten one-byte elements, five two-byte elements, two
5233 five-byte elements or one ten-byte element.
5235 Some \c{OMF} linkers require the \I{element size, in common
5236 variables}\I{common variables, element size}element size, as well as
5237 the variable size, to match when resolving common variables declared
5238 in more than one module. Therefore NASM must allow you to specify
5239 the element size on your far common variables. This is done by the
5242 \c common c_5by2 10:far 5 ; two five-byte elements
5243 \c common c_2by5 10:far 2 ; five two-byte elements
5245 If no element size is specified, the default is 1. Also, the \c{FAR}
5246 keyword is not required when an element size is specified, since
5247 only far commons may have element sizes at all. So the above
5248 declarations could equivalently be
5250 \c common c_5by2 10:5 ; two five-byte elements
5251 \c common c_2by5 10:2 ; five two-byte elements
5253 In addition to these extensions, the \c{COMMON} directive in \c{obj}
5254 also supports default-\c{WRT} specification like \c{EXTERN} does
5255 (explained in \k{objextern}). So you can also declare things like
5257 \c common foo 10:wrt dgroup
5258 \c common bar 16:far 2:wrt data
5259 \c common baz 24:wrt data:6
5262 \H{win32fmt} \i\c{win32}: Microsoft Win32 Object Files
5264 The \c{win32} output format generates Microsoft Win32 object files,
5265 suitable for passing to Microsoft linkers such as \i{Visual C++}.
5266 Note that Borland Win32 compilers do not use this format, but use
5267 \c{obj} instead (see \k{objfmt}).
5269 \c{win32} provides a default output file-name extension of \c{.obj}.
5271 Note that although Microsoft say that Win32 object files follow the
5272 \c{COFF} (Common Object File Format) standard, the object files produced
5273 by Microsoft Win32 compilers are not compatible with COFF linkers
5274 such as DJGPP's, and vice versa. This is due to a difference of
5275 opinion over the precise semantics of PC-relative relocations. To
5276 produce COFF files suitable for DJGPP, use NASM's \c{coff} output
5277 format; conversely, the \c{coff} format does not produce object
5278 files that Win32 linkers can generate correct output from.
5281 \S{win32sect} \c{win32} Extensions to the \c{SECTION}
5282 Directive\I{SECTION, win32 extensions to}
5284 Like the \c{obj} format, \c{win32} allows you to specify additional
5285 information on the \c{SECTION} directive line, to control the type
5286 and properties of sections you declare. Section types and properties
5287 are generated automatically by NASM for the \i{standard section names}
5288 \c{.text}, \c{.data} and \c{.bss}, but may still be overridden by
5291 The available qualifiers are:
5293 \b \c{code}, or equivalently \c{text}, defines the section to be a
5294 code section. This marks the section as readable and executable, but
5295 not writable, and also indicates to the linker that the type of the
5298 \b \c{data} and \c{bss} define the section to be a data section,
5299 analogously to \c{code}. Data sections are marked as readable and
5300 writable, but not executable. \c{data} declares an initialized data
5301 section, whereas \c{bss} declares an uninitialized data section.
5303 \b \c{rdata} declares an initialized data section that is readable
5304 but not writable. Microsoft compilers use this section to place
5307 \b \c{info} defines the section to be an \i{informational section},
5308 which is not included in the executable file by the linker, but may
5309 (for example) pass information \e{to} the linker. For example,
5310 declaring an \c{info}-type section called \i\c{.drectve} causes the
5311 linker to interpret the contents of the section as command-line
5314 \b \c{align=}, used with a trailing number as in \c{obj}, gives the
5315 \I{section alignment, in win32}\I{alignment, in win32
5316 sections}alignment requirements of the section. The maximum you may
5317 specify is 64: the Win32 object file format contains no means to
5318 request a greater section alignment than this. If alignment is not
5319 explicitly specified, the defaults are 16-byte alignment for code
5320 sections, 8-byte alignment for rdata sections and 4-byte alignment
5321 for data (and BSS) sections.
5322 Informational sections get a default alignment of 1 byte (no
5323 alignment), though the value does not matter.
5325 The defaults assumed by NASM if you do not specify the above
5328 \c section .text code align=16
5329 \c section .data data align=4
5330 \c section .rdata rdata align=8
5331 \c section .bss bss align=4
5333 Any other section name is treated by default like \c{.text}.
5335 \S{win32safeseh} \c{win32}: Safe Structured Exception Handling
5337 Among other improvements in Windows XP SP2 and Windows Server 2003
5338 Microsoft has introduced concept of "safe structured exception
5339 handling." General idea is to collect handlers' entry points in
5340 designated read-only table and have alleged entry point verified
5341 against this table prior exception control is passed to the handler. In
5342 order for an executable module to be equipped with such "safe exception
5343 handler table," all object modules on linker command line has to comply
5344 with certain criteria. If one single module among them does not, then
5345 the table in question is omitted and above mentioned run-time checks
5346 will not be performed for application in question. Table omission is by
5347 default silent and therefore can be easily overlooked. One can instruct
5348 linker to refuse to produce binary without such table by passing
5349 \c{/safeseh} command line option.
5351 Without regard to this run-time check merits it's natural to expect
5352 NASM to be capable of generating modules suitable for \c{/safeseh}
5353 linking. From developer's viewpoint the problem is two-fold:
5355 \b how to adapt modules not deploying exception handlers of their own;
5357 \b how to adapt/develop modules utilizing custom exception handling;
5359 Former can be easily achieved with any NASM version by adding following
5360 line to source code:
5364 As of version 2.03 NASM adds this absolute symbol automatically. If
5365 it's not already present to be precise. I.e. if for whatever reason
5366 developer would choose to assign another value in source file, it would
5367 still be perfectly possible.
5369 Registering custom exception handler on the other hand requires certain
5370 "magic." As of version 2.03 additional directive is implemented,
5371 \c{safeseh}, which instructs the assembler to produce appropriately
5372 formatted input data for above mentioned "safe exception handler
5373 table." Its typical use would be:
5376 \c extern _MessageBoxA@16
5377 \c %if __NASM_VERSION_ID__ >= 0x02030000
5378 \c safeseh handler ; register handler as "safe handler"
5381 \c push DWORD 1 ; MB_OKCANCEL
5382 \c push DWORD caption
5385 \c call _MessageBoxA@16
5386 \c sub eax,1 ; incidentally suits as return value
5387 \c ; for exception handler
5391 \c push DWORD handler
5392 \c push DWORD [fs:0]
5393 \c mov DWORD [fs:0],esp ; engage exception handler
5395 \c mov eax,DWORD[eax] ; cause exception
5396 \c pop DWORD [fs:0] ; disengage exception handler
5399 \c text: db 'OK to rethrow, CANCEL to generate core dump',0
5400 \c caption:db 'SEGV',0
5402 \c section .drectve info
5403 \c db '/defaultlib:user32.lib /defaultlib:msvcrt.lib '
5405 As you might imagine, it's perfectly possible to produce .exe binary
5406 with "safe exception handler table" and yet engage unregistered
5407 exception handler. Indeed, handler is engaged by simply manipulating
5408 \c{[fs:0]} location at run-time, something linker has no power over,
5409 run-time that is. It should be explicitly mentioned that such failure
5410 to register handler's entry point with \c{safeseh} directive has
5411 undesired side effect at run-time. If exception is raised and
5412 unregistered handler is to be executed, the application is abruptly
5413 terminated without any notification whatsoever. One can argue that
5414 system could at least have logged some kind "non-safe exception
5415 handler in x.exe at address n" message in event log, but no, literally
5416 no notification is provided and user is left with no clue on what
5417 caused application failure.
5419 Finally, all mentions of linker in this paragraph refer to Microsoft
5420 linker version 7.x and later. Presence of \c{@feat.00} symbol and input
5421 data for "safe exception handler table" causes no backward
5422 incompatibilities and "safeseh" modules generated by NASM 2.03 and
5423 later can still be linked by earlier versions or non-Microsoft linkers.
5426 \H{win64fmt} \i\c{win64}: Microsoft Win64 Object Files
5428 The \c{win64} output format generates Microsoft Win64 object files,
5429 which is nearly 100% identical to the \c{win32} object format (\k{win32fmt})
5430 with the exception that it is meant to target 64-bit code and the x86-64
5431 platform altogether. This object file is used exactly the same as the \c{win32}
5432 object format (\k{win32fmt}), in NASM, with regard to this exception.
5434 \S{win64pic} \c{win64}: Writing Position-Independent Code
5436 While \c{REL} takes good care of RIP-relative addressing, there is one
5437 aspect that is easy to overlook for a Win64 programmer: indirect
5438 references. Consider a switch dispatch table:
5440 \c jmp qword [dsptch+rax*8]
5446 Even a novice Win64 assembler programmer will soon realize that the code
5447 is not 64-bit savvy. Most notably linker will refuse to link it with
5449 \c 'ADDR32' relocation to '.text' invalid without /LARGEADDRESSAWARE:NO
5451 So [s]he will have to split jmp instruction as following:
5453 \c lea rbx,[rel dsptch]
5454 \c jmp qword [rbx+rax*8]
5456 What happens behind the scene is that effective address in \c{lea} is
5457 encoded relative to instruction pointer, or in perfectly
5458 position-independent manner. But this is only part of the problem!
5459 Trouble is that in .dll context \c{caseN} relocations will make their
5460 way to the final module and might have to be adjusted at .dll load
5461 time. To be specific when it can't be loaded at preferred address. And
5462 when this occurs, pages with such relocations will be rendered private
5463 to current process, which kind of undermines the idea of sharing .dll.
5464 But no worry, it's trivial to fix:
5466 \c lea rbx,[rel dsptch]
5467 \c add rbx,[rbx+rax*8]
5470 \c dsptch: dq case0-dsptch
5474 NASM version 2.03 and later provides another alternative, \c{wrt
5475 ..imagebase} operator, which returns offset from base address of the
5476 current image, be it .exe or .dll module, therefore the name. For those
5477 acquainted with PE-COFF format base address denotes start of
5478 \c{IMAGE_DOS_HEADER} structure. Here is how to implement switch with
5479 these image-relative references:
5481 \c lea rbx,[rel dsptch]
5482 \c mov eax,[rbx+rax*4]
5483 \c sub rbx,dsptch wrt ..imagebase
5487 \c dsptch: dd case0 wrt ..imagebase
5488 \c dd case1 wrt ..imagebase
5490 One can argue that the operator is redundant. Indeed, snippet before
5491 last works just fine with any NASM version and is not even Windows
5492 specific... The real reason for implementing \c{wrt ..imagebase} will
5493 become apparent in next paragraph.
5495 It should be noted that \c{wrt ..imagebase} is defined as 32-bit
5498 \c dd label wrt ..imagebase ; ok
5499 \c dq label wrt ..imagebase ; bad
5500 \c mov eax,label wrt ..imagebase ; ok
5501 \c mov rax,label wrt ..imagebase ; bad
5503 \S{win64seh} \c{win64}: Structured Exception Handling
5505 Structured exception handing in Win64 is completely different matter
5506 from Win32. Upon exception program counter value is noted, and
5507 linker-generated table comprising start and end addresses of all the
5508 functions [in given executable module] is traversed and compared to the
5509 saved program counter. Thus so called \c{UNWIND_INFO} structure is
5510 identified. If it's not found, then offending subroutine is assumed to
5511 be "leaf" and just mentioned lookup procedure is attempted for its
5512 caller. In Win64 leaf function is such function that does not call any
5513 other function \e{nor} modifies any Win64 non-volatile registers,
5514 including stack pointer. The latter ensures that it's possible to
5515 identify leaf function's caller by simply pulling the value from the
5518 While majority of subroutines written in assembler are not calling any
5519 other function, requirement for non-volatile registers' immutability
5520 leaves developer with not more than 7 registers and no stack frame,
5521 which is not necessarily what [s]he counted with. Customarily one would
5522 meet the requirement by saving non-volatile registers on stack and
5523 restoring them upon return, so what can go wrong? If [and only if] an
5524 exception is raised at run-time and no \c{UNWIND_INFO} structure is
5525 associated with such "leaf" function, the stack unwind procedure will
5526 expect to find caller's return address on the top of stack immediately
5527 followed by its frame. Given that developer pushed caller's
5528 non-volatile registers on stack, would the value on top point at some
5529 code segment or even addressable space? Well, developer can attempt
5530 copying caller's return address to the top of stack and this would
5531 actually work in some very specific circumstances. But unless developer
5532 can guarantee that these circumstances are always met, it's more
5533 appropriate to assume worst case scenario, i.e. stack unwind procedure
5534 going berserk. Relevant question is what happens then? Application is
5535 abruptly terminated without any notification whatsoever. Just like in
5536 Win32 case, one can argue that system could at least have logged
5537 "unwind procedure went berserk in x.exe at address n" in event log, but
5538 no, no trace of failure is left.
5540 Now, when we understand significance of the \c{UNWIND_INFO} structure,
5541 let's discuss what's in it and/or how it's processed. First of all it
5542 is checked for presence of reference to custom language-specific
5543 exception handler. If there is one, then it's invoked. Depending on the
5544 return value, execution flow is resumed (exception is said to be
5545 "handled"), \e{or} rest of \c{UNWIND_INFO} structure is processed as
5546 following. Beside optional reference to custom handler, it carries
5547 information about current callee's stack frame and where non-volatile
5548 registers are saved. Information is detailed enough to be able to
5549 reconstruct contents of caller's non-volatile registers upon call to
5550 current callee. And so caller's context is reconstructed, and then
5551 unwind procedure is repeated, i.e. another \c{UNWIND_INFO} structure is
5552 associated, this time, with caller's instruction pointer, which is then
5553 checked for presence of reference to language-specific handler, etc.
5554 The procedure is recursively repeated till exception is handled. As
5555 last resort system "handles" it by generating memory core dump and
5556 terminating the application.
5558 As for the moment of this writing NASM unfortunately does not
5559 facilitate generation of above mentioned detailed information about
5560 stack frame layout. But as of version 2.03 it implements building
5561 blocks for generating structures involved in stack unwinding. As
5562 simplest example, here is how to deploy custom exception handler for
5567 \c extern MessageBoxA
5573 \c mov r9,1 ; MB_OKCANCEL
5575 \c sub eax,1 ; incidentally suits as return value
5576 \c ; for exception handler
5582 \c mov rax,QWORD[rax] ; cause exception
5585 \c text: db 'OK to rethrow, CANCEL to generate core dump',0
5586 \c caption:db 'SEGV',0
5588 \c section .pdata rdata align=4
5589 \c dd main wrt ..imagebase
5590 \c dd main_end wrt ..imagebase
5591 \c dd xmain wrt ..imagebase
5592 \c section .xdata rdata align=8
5593 \c xmain: db 9,0,0,0
5594 \c dd handler wrt ..imagebase
5595 \c section .drectve info
5596 \c db '/defaultlib:user32.lib /defaultlib:msvcrt.lib '
5598 What you see in \c{.pdata} section is element of the "table comprising
5599 start and end addresses of function" along with reference to associated
5600 \c{UNWIND_INFO} structure. And what you see in \c{.xdata} section is
5601 \c{UNWIND_INFO} structure describing function with no frame, but with
5602 designated exception handler. References are \e{required} to be
5603 image-relative (which is the real reason for implementing \c{wrt
5604 ..imagebase} operator). It should be noted that \c{rdata align=n}, as
5605 well as \c{wrt ..imagebase}, are optional in these two segments'
5606 contexts, i.e. can be omitted. Latter means that \e{all} 32-bit
5607 references, not only above listed required ones, placed into these two
5608 segments turn out image-relative. Why is it important to understand?
5609 Developer is allowed to append handler-specific data to \c{UNWIND_INFO}
5610 structure, and if [s]he adds a 32-bit reference, then [s]he will have
5611 to remember to adjust its value to obtain the real pointer.
5613 As already mentioned, in Win64 terms leaf function is one that does not
5614 call any other function \e{nor} modifies any non-volatile register,
5615 including stack pointer. But it's not uncommon that assembler
5616 programmer plans to utilize every single register and sometimes even
5617 have variable stack frame. Is there anything one can do with bare
5618 building blocks? I.e. besides manually composing fully-fledged
5619 \c{UNWIND_INFO} structure, which would surely be considered
5620 error-prone? Yes, there is. Recall that exception handler is called
5621 first, before stack layout is analyzed. As it turned out, it's
5622 perfectly possible to manipulate current callee's context in custom
5623 handler in manner that permits further stack unwinding. General idea is
5624 that handler would not actually "handle" the exception, but instead
5625 restore callee's context, as it was at its entry point and thus mimic
5626 leaf function. In other words, handler would simply undertake part of
5627 unwinding procedure. Consider following example:
5630 \c mov rax,rsp ; copy rsp to volatile register
5631 \c push r15 ; save non-volatile registers
5634 \c mov r11,rsp ; prepare variable stack frame
5637 \c mov QWORD[r11],rax ; check for exceptions
5638 \c mov rsp,r11 ; allocate stack frame
5639 \c mov QWORD[rsp],rax ; save original rsp value
5642 \c mov r11,QWORD[rsp] ; pull original rsp value
5643 \c mov rbp,QWORD[r11-24]
5644 \c mov rbx,QWORD[r11-16]
5645 \c mov r15,QWORD[r11-8]
5646 \c mov rsp,r11 ; destroy frame
5649 The keyword is that up to \c{magic_point} original \c{rsp} value
5650 remains in chosen volatile register and no non-volatile register,
5651 except for \c{rsp}, is modified. While past \c{magic_point} \c{rsp}
5652 remains constant till the very end of the \c{function}. In this case
5653 custom language-specific exception handler would look like this:
5655 \c EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
5656 \c CONTEXT *context,DISPATCHER_CONTEXT *disp)
5658 \c if (context->Rip<(ULONG64)magic_point)
5659 \c rsp = (ULONG64 *)context->Rax;
5661 \c { rsp = ((ULONG64 **)context->Rsp)[0];
5662 \c context->Rbp = rsp[-3];
5663 \c context->Rbx = rsp[-2];
5664 \c context->R15 = rsp[-1];
5666 \c context->Rsp = (ULONG64)rsp;
5668 \c memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
5669 \c RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
5670 \c dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
5671 \c &disp->HandlerData,&disp->EstablisherFrame,NULL);
5672 \c return ExceptionContinueSearch;
5675 As custom handler mimics leaf function, corresponding \c{UNWIND_INFO}
5676 structure does not have to contain any information about stack frame
5679 \H{cofffmt} \i\c{coff}: \i{Common Object File Format}
5681 The \c{coff} output type produces \c{COFF} object files suitable for
5682 linking with the \i{DJGPP} linker.
5684 \c{coff} provides a default output file-name extension of \c{.o}.
5686 The \c{coff} format supports the same extensions to the \c{SECTION}
5687 directive as \c{win32} does, except that the \c{align} qualifier and
5688 the \c{info} section type are not supported.
5690 \H{machofmt} \I{Mach-O}\i\c{macho32} and \i\c{macho64}: \i{Mach Object File Format}
5692 The \c{macho32} and \c{macho64} output formts produces \c{Mach-O}
5693 object files suitable for linking with the \i{MacOS X} linker.
5694 \i\c{macho} is a synonym for \c{macho32}.
5696 \c{macho} provides a default output file-name extension of \c{.o}.
5698 \H{elffmt} \i\c{elf32}, \i\c{elf64}, \i\c{elfx32}: \I{ELF}\I{linux, elf}\i{Executable and Linkable
5699 Format} Object Files
5701 The \c{elf32}, \c{elf64} and \c{elfx32} output formats generate
5702 \c{ELF32 and ELF64} (Executable and Linkable Format) object files, as
5703 used by Linux as well as \i{Unix System V}, including \i{Solaris x86},
5704 \i{UnixWare} and \i{SCO Unix}. \c{elf} provides a default output
5705 file-name extension of \c{.o}. \c{elf} is a synonym for \c{elf32}.
5707 The \c{elfx32} format is used for the \i{x32} ABI, which is a 32-bit
5708 ABI with the CPU in 64-bit mode.
5710 \S{abisect} ELF specific directive \i\c{osabi}
5712 The ELF header specifies the application binary interface for the target operating system (OSABI).
5713 This field can be set by using the \c{osabi} directive with the numeric value (0-255) of the target
5714 system. If this directive is not used, the default value will be "UNIX System V ABI" (0) which will work on
5715 most systems which support ELF.
5717 \S{elfsect} \c{elf} Extensions to the \c{SECTION}
5718 Directive\I{SECTION, elf extensions to}
5720 Like the \c{obj} format, \c{elf} allows you to specify additional
5721 information on the \c{SECTION} directive line, to control the type
5722 and properties of sections you declare. Section types and properties
5723 are generated automatically by NASM for the \i{standard section
5724 names}, but may still be
5725 overridden by these qualifiers.
5727 The available qualifiers are:
5729 \b \i\c{alloc} defines the section to be one which is loaded into
5730 memory when the program is run. \i\c{noalloc} defines it to be one
5731 which is not, such as an informational or comment section.
5733 \b \i\c{exec} defines the section to be one which should have execute
5734 permission when the program is run. \i\c{noexec} defines it as one
5737 \b \i\c{write} defines the section to be one which should be writable
5738 when the program is run. \i\c{nowrite} defines it as one which should
5741 \b \i\c{progbits} defines the section to be one with explicit contents
5742 stored in the object file: an ordinary code or data section, for
5743 example, \i\c{nobits} defines the section to be one with no explicit
5744 contents given, such as a BSS section.
5746 \b \c{align=}, used with a trailing number as in \c{obj}, gives the
5747 \I{section alignment, in elf}\I{alignment, in elf sections}alignment
5748 requirements of the section.
5750 \b \i\c{tls} defines the section to be one which contains
5751 thread local variables.
5753 The defaults assumed by NASM if you do not specify the above
5756 \I\c{.text} \I\c{.rodata} \I\c{.lrodata} \I\c{.data} \I\c{.ldata}
5757 \I\c{.bss} \I\c{.lbss} \I\c{.tdata} \I\c{.tbss} \I\c\{.comment}
5759 \c section .text progbits alloc exec nowrite align=16
5760 \c section .rodata progbits alloc noexec nowrite align=4
5761 \c section .lrodata progbits alloc noexec nowrite align=4
5762 \c section .data progbits alloc noexec write align=4
5763 \c section .ldata progbits alloc noexec write align=4
5764 \c section .bss nobits alloc noexec write align=4
5765 \c section .lbss nobits alloc noexec write align=4
5766 \c section .tdata progbits alloc noexec write align=4 tls
5767 \c section .tbss nobits alloc noexec write align=4 tls
5768 \c section .comment progbits noalloc noexec nowrite align=1
5769 \c section other progbits alloc noexec nowrite align=1
5771 (Any section name other than those in the above table
5772 is treated by default like \c{other} in the above table.
5773 Please note that section names are case sensitive.)
5776 \S{elfwrt} \i{Position-Independent Code}\I{PIC}: \c{elf} Special
5777 Symbols and \i\c{WRT}
5779 The \c{ELF} specification contains enough features to allow
5780 position-independent code (PIC) to be written, which makes \i{ELF
5781 shared libraries} very flexible. However, it also means NASM has to
5782 be able to generate a variety of ELF specific relocation types in ELF
5783 object files, if it is to be an assembler which can write PIC.
5785 Since \c{ELF} does not support segment-base references, the \c{WRT}
5786 operator is not used for its normal purpose; therefore NASM's
5787 \c{elf} output format makes use of \c{WRT} for a different purpose,
5788 namely the PIC-specific \I{relocations, PIC-specific}relocation
5791 \c{elf} defines five special symbols which you can use as the
5792 right-hand side of the \c{WRT} operator to obtain PIC relocation
5793 types. They are \i\c{..gotpc}, \i\c{..gotoff}, \i\c{..got},
5794 \i\c{..plt} and \i\c{..sym}. Their functions are summarized here:
5796 \b Referring to the symbol marking the global offset table base
5797 using \c{wrt ..gotpc} will end up giving the distance from the
5798 beginning of the current section to the global offset table.
5799 (\i\c{_GLOBAL_OFFSET_TABLE_} is the standard symbol name used to
5800 refer to the \i{GOT}.) So you would then need to add \i\c{$$} to the
5801 result to get the real address of the GOT.
5803 \b Referring to a location in one of your own sections using \c{wrt
5804 ..gotoff} will give the distance from the beginning of the GOT to
5805 the specified location, so that adding on the address of the GOT
5806 would give the real address of the location you wanted.
5808 \b Referring to an external or global symbol using \c{wrt ..got}
5809 causes the linker to build an entry \e{in} the GOT containing the
5810 address of the symbol, and the reference gives the distance from the
5811 beginning of the GOT to the entry; so you can add on the address of
5812 the GOT, load from the resulting address, and end up with the
5813 address of the symbol.
5815 \b Referring to a procedure name using \c{wrt ..plt} causes the
5816 linker to build a \i{procedure linkage table} entry for the symbol,
5817 and the reference gives the address of the \i{PLT} entry. You can
5818 only use this in contexts which would generate a PC-relative
5819 relocation normally (i.e. as the destination for \c{CALL} or
5820 \c{JMP}), since ELF contains no relocation type to refer to PLT
5823 \b Referring to a symbol name using \c{wrt ..sym} causes NASM to
5824 write an ordinary relocation, but instead of making the relocation
5825 relative to the start of the section and then adding on the offset
5826 to the symbol, it will write a relocation record aimed directly at
5827 the symbol in question. The distinction is a necessary one due to a
5828 peculiarity of the dynamic linker.
5830 A fuller explanation of how to use these relocation types to write
5831 shared libraries entirely in NASM is given in \k{picdll}.
5833 \S{elftls} \i{Thread Local Storage}\I{TLS}: \c{elf} Special
5834 Symbols and \i\c{WRT}
5836 \b In ELF32 mode, referring to an external or global symbol using
5837 \c{wrt ..tlsie} \I\c{..tlsie}
5838 causes the linker to build an entry \e{in} the GOT containing the
5839 offset of the symbol within the TLS block, so you can access the value
5840 of the symbol with code such as:
5842 \c mov eax,[tid wrt ..tlsie]
5846 \b In ELF64 or ELFx32 mode, referring to an external or global symbol using
5847 \c{wrt ..gottpoff} \I\c{..gottpoff}
5848 causes the linker to build an entry \e{in} the GOT containing the
5849 offset of the symbol within the TLS block, so you can access the value
5850 of the symbol with code such as:
5852 \c mov rax,[rel tid wrt ..gottpoff]
5856 \S{elfglob} \c{elf} Extensions to the \c{GLOBAL} Directive\I{GLOBAL,
5857 elf extensions to}\I{GLOBAL, aoutb extensions to}
5859 \c{ELF} object files can contain more information about a global symbol
5860 than just its address: they can contain the \I{symbol sizes,
5861 specifying}\I{size, of symbols}size of the symbol and its \I{symbol
5862 types, specifying}\I{type, of symbols}type as well. These are not
5863 merely debugger conveniences, but are actually necessary when the
5864 program being written is a \i{shared library}. NASM therefore
5865 supports some extensions to the \c{GLOBAL} directive, allowing you
5866 to specify these features.
5868 You can specify whether a global variable is a function or a data
5869 object by suffixing the name with a colon and the word
5870 \i\c{function} or \i\c{data}. (\i\c{object} is a synonym for
5871 \c{data}.) For example:
5873 \c global hashlookup:function, hashtable:data
5875 exports the global symbol \c{hashlookup} as a function and
5876 \c{hashtable} as a data object.
5878 Optionally, you can control the ELF visibility of the symbol. Just
5879 add one of the visibility keywords: \i\c{default}, \i\c{internal},
5880 \i\c{hidden}, or \i\c{protected}. The default is \i\c{default} of
5881 course. For example, to make \c{hashlookup} hidden:
5883 \c global hashlookup:function hidden
5885 You can also specify the size of the data associated with the
5886 symbol, as a numeric expression (which may involve labels, and even
5887 forward references) after the type specifier. Like this:
5889 \c global hashtable:data (hashtable.end - hashtable)
5892 \c db this,that,theother ; some data here
5895 This makes NASM automatically calculate the length of the table and
5896 place that information into the \c{ELF} symbol table.
5898 Declaring the type and size of global symbols is necessary when
5899 writing shared library code. For more information, see
5903 \S{elfcomm} \c{elf} Extensions to the \c{COMMON} Directive
5904 \I{COMMON, elf extensions to}
5906 \c{ELF} also allows you to specify alignment requirements \I{common
5907 variables, alignment in elf}\I{alignment, of elf common variables}on
5908 common variables. This is done by putting a number (which must be a
5909 power of two) after the name and size of the common variable,
5910 separated (as usual) by a colon. For example, an array of
5911 doublewords would benefit from 4-byte alignment:
5913 \c common dwordarray 128:4
5915 This declares the total size of the array to be 128 bytes, and
5916 requires that it be aligned on a 4-byte boundary.
5919 \S{elf16} 16-bit code and ELF
5920 \I{ELF, 16-bit code and}
5922 The \c{ELF32} specification doesn't provide relocations for 8- and
5923 16-bit values, but the GNU \c{ld} linker adds these as an extension.
5924 NASM can generate GNU-compatible relocations, to allow 16-bit code to
5925 be linked as ELF using GNU \c{ld}. If NASM is used with the
5926 \c{-w+gnu-elf-extensions} option, a warning is issued when one of
5927 these relocations is generated.
5929 \S{elfdbg} Debug formats and ELF
5930 \I{ELF, Debug formats and}
5932 ELF provides debug information in \c{STABS} and \c{DWARF} formats.
5933 Line number information is generated for all executable sections, but please
5934 note that only the ".text" section is executable by default.
5936 \H{aoutfmt} \i\c{aout}: Linux \I{a.out, Linux version}\I{linux, a.out}\c{a.out} Object Files
5938 The \c{aout} format generates \c{a.out} object files, in the form used
5939 by early Linux systems (current Linux systems use ELF, see
5940 \k{elffmt}.) These differ from other \c{a.out} object files in that
5941 the magic number in the first four bytes of the file is
5942 different; also, some implementations of \c{a.out}, for example
5943 NetBSD's, support position-independent code, which Linux's
5944 implementation does not.
5946 \c{a.out} provides a default output file-name extension of \c{.o}.
5948 \c{a.out} is a very simple object format. It supports no special
5949 directives, no special symbols, no use of \c{SEG} or \c{WRT}, and no
5950 extensions to any standard directives. It supports only the three
5951 \i{standard section names} \i\c{.text}, \i\c{.data} and \i\c{.bss}.
5954 \H{aoutfmt} \i\c{aoutb}: \i{NetBSD}/\i{FreeBSD}/\i{OpenBSD}
5955 \I{a.out, BSD version}\c{a.out} Object Files
5957 The \c{aoutb} format generates \c{a.out} object files, in the form
5958 used by the various free \c{BSD Unix} clones, \c{NetBSD}, \c{FreeBSD}
5959 and \c{OpenBSD}. For simple object files, this object format is exactly
5960 the same as \c{aout} except for the magic number in the first four bytes
5961 of the file. However, the \c{aoutb} format supports
5962 \I{PIC}\i{position-independent code} in the same way as the \c{elf}
5963 format, so you can use it to write \c{BSD} \i{shared libraries}.
5965 \c{aoutb} provides a default output file-name extension of \c{.o}.
5967 \c{aoutb} supports no special directives, no special symbols, and
5968 only the three \i{standard section names} \i\c{.text}, \i\c{.data}
5969 and \i\c{.bss}. However, it also supports the same use of \i\c{WRT} as
5970 \c{elf} does, to provide position-independent code relocation types.
5971 See \k{elfwrt} for full documentation of this feature.
5973 \c{aoutb} also supports the same extensions to the \c{GLOBAL}
5974 directive as \c{elf} does: see \k{elfglob} for documentation of
5978 \H{as86fmt} \c{as86}: \i{Minix}/Linux\I{linux, as86} \i\c{as86} Object Files
5980 The Minix/Linux 16-bit assembler \c{as86} has its own non-standard
5981 object file format. Although its companion linker \i\c{ld86} produces
5982 something close to ordinary \c{a.out} binaries as output, the object
5983 file format used to communicate between \c{as86} and \c{ld86} is not
5986 NASM supports this format, just in case it is useful, as \c{as86}.
5987 \c{as86} provides a default output file-name extension of \c{.o}.
5989 \c{as86} is a very simple object format (from the NASM user's point
5990 of view). It supports no special directives, no use of \c{SEG} or \c{WRT},
5991 and no extensions to any standard directives. It supports only the three
5992 \i{standard section names} \i\c{.text}, \i\c{.data} and \i\c{.bss}. The
5993 only special symbol supported is \c{..start}.
5996 \H{rdffmt} \I{RDOFF}\i\c{rdf}: \i{Relocatable Dynamic Object File
5999 The \c{rdf} output format produces \c{RDOFF} object files. \c{RDOFF}
6000 (Relocatable Dynamic Object File Format) is a home-grown object-file
6001 format, designed alongside NASM itself and reflecting in its file
6002 format the internal structure of the assembler.
6004 \c{RDOFF} is not used by any well-known operating systems. Those
6005 writing their own systems, however, may well wish to use \c{RDOFF}
6006 as their object format, on the grounds that it is designed primarily
6007 for simplicity and contains very little file-header bureaucracy.
6009 The Unix NASM archive, and the DOS archive which includes sources,
6010 both contain an \I{rdoff subdirectory}\c{rdoff} subdirectory holding
6011 a set of RDOFF utilities: an RDF linker, an \c{RDF} static-library
6012 manager, an RDF file dump utility, and a program which will load and
6013 execute an RDF executable under Linux.
6015 \c{rdf} supports only the \i{standard section names} \i\c{.text},
6016 \i\c{.data} and \i\c{.bss}.
6019 \S{rdflib} Requiring a Library: The \i\c{LIBRARY} Directive
6021 \c{RDOFF} contains a mechanism for an object file to demand a given
6022 library to be linked to the module, either at load time or run time.
6023 This is done by the \c{LIBRARY} directive, which takes one argument
6024 which is the name of the module:
6026 \c library mylib.rdl
6029 \S{rdfmod} Specifying a Module Name: The \i\c{MODULE} Directive
6031 Special \c{RDOFF} header record is used to store the name of the module.
6032 It can be used, for example, by run-time loader to perform dynamic
6033 linking. \c{MODULE} directive takes one argument which is the name
6038 Note that when you statically link modules and tell linker to strip
6039 the symbols from output file, all module names will be stripped too.
6040 To avoid it, you should start module names with \I{$, prefix}\c{$}, like:
6042 \c module $kernel.core
6045 \S{rdfglob} \c{rdf} Extensions to the \c{GLOBAL} Directive\I{GLOBAL,
6048 \c{RDOFF} global symbols can contain additional information needed by
6049 the static linker. You can mark a global symbol as exported, thus
6050 telling the linker do not strip it from target executable or library
6051 file. Like in \c{ELF}, you can also specify whether an exported symbol
6052 is a procedure (function) or data object.
6054 Suffixing the name with a colon and the word \i\c{export} you make the
6057 \c global sys_open:export
6059 To specify that exported symbol is a procedure (function), you add the
6060 word \i\c{proc} or \i\c{function} after declaration:
6062 \c global sys_open:export proc
6064 Similarly, to specify exported data object, add the word \i\c{data}
6065 or \i\c{object} to the directive:
6067 \c global kernel_ticks:export data
6070 \S{rdfimpt} \c{rdf} Extensions to the \c{EXTERN} Directive\I{EXTERN,
6073 By default the \c{EXTERN} directive in \c{RDOFF} declares a "pure external"
6074 symbol (i.e. the static linker will complain if such a symbol is not resolved).
6075 To declare an "imported" symbol, which must be resolved later during a dynamic
6076 linking phase, \c{RDOFF} offers an additional \c{import} modifier. As in
6077 \c{GLOBAL}, you can also specify whether an imported symbol is a procedure
6078 (function) or data object. For example:
6081 \c extern _open:import
6082 \c extern _printf:import proc
6083 \c extern _errno:import data
6085 Here the directive \c{LIBRARY} is also included, which gives the dynamic linker
6086 a hint as to where to find requested symbols.
6089 \H{dbgfmt} \i\c{dbg}: Debugging Format
6091 The \c{dbg} output format is not built into NASM in the default
6092 configuration. If you are building your own NASM executable from the
6093 sources, you can define \i\c{OF_DBG} in \c{output/outform.h} or on the
6094 compiler command line, and obtain the \c{dbg} output format.
6096 The \c{dbg} format does not output an object file as such; instead,
6097 it outputs a text file which contains a complete list of all the
6098 transactions between the main body of NASM and the output-format
6099 back end module. It is primarily intended to aid people who want to
6100 write their own output drivers, so that they can get a clearer idea
6101 of the various requests the main program makes of the output driver,
6102 and in what order they happen.
6104 For simple files, one can easily use the \c{dbg} format like this:
6106 \c nasm -f dbg filename.asm
6108 which will generate a diagnostic file called \c{filename.dbg}.
6109 However, this will not work well on files which were designed for a
6110 different object format, because each object format defines its own
6111 macros (usually user-level forms of directives), and those macros
6112 will not be defined in the \c{dbg} format. Therefore it can be
6113 useful to run NASM twice, in order to do the preprocessing with the
6114 native object format selected:
6116 \c nasm -e -f rdf -o rdfprog.i rdfprog.asm
6117 \c nasm -a -f dbg rdfprog.i
6119 This preprocesses \c{rdfprog.asm} into \c{rdfprog.i}, keeping the
6120 \c{rdf} object format selected in order to make sure RDF special
6121 directives are converted into primitive form correctly. Then the
6122 preprocessed source is fed through the \c{dbg} format to generate
6123 the final diagnostic output.
6125 This workaround will still typically not work for programs intended
6126 for \c{obj} format, because the \c{obj} \c{SEGMENT} and \c{GROUP}
6127 directives have side effects of defining the segment and group names
6128 as symbols; \c{dbg} will not do this, so the program will not
6129 assemble. You will have to work around that by defining the symbols
6130 yourself (using \c{EXTERN}, for example) if you really need to get a
6131 \c{dbg} trace of an \c{obj}-specific source file.
6133 \c{dbg} accepts any section name and any directives at all, and logs
6134 them all to its output file.
6137 \C{16bit} Writing 16-bit Code (DOS, Windows 3/3.1)
6139 This chapter attempts to cover some of the common issues encountered
6140 when writing 16-bit code to run under \c{MS-DOS} or \c{Windows 3.x}. It
6141 covers how to link programs to produce \c{.EXE} or \c{.COM} files,
6142 how to write \c{.SYS} device drivers, and how to interface assembly
6143 language code with 16-bit C compilers and with Borland Pascal.
6146 \H{exefiles} Producing \i\c{.EXE} Files
6148 Any large program written under DOS needs to be built as a \c{.EXE}
6149 file: only \c{.EXE} files have the necessary internal structure
6150 required to span more than one 64K segment. \i{Windows} programs,
6151 also, have to be built as \c{.EXE} files, since Windows does not
6152 support the \c{.COM} format.
6154 In general, you generate \c{.EXE} files by using the \c{obj} output
6155 format to produce one or more \i\c{.OBJ} files, and then linking
6156 them together using a linker. However, NASM also supports the direct
6157 generation of simple DOS \c{.EXE} files using the \c{bin} output
6158 format (by using \c{DB} and \c{DW} to construct the \c{.EXE} file
6159 header), and a macro package is supplied to do this. Thanks to
6160 Yann Guidon for contributing the code for this.
6162 NASM may also support \c{.EXE} natively as another output format in
6166 \S{objexe} Using the \c{obj} Format To Generate \c{.EXE} Files
6168 This section describes the usual method of generating \c{.EXE} files
6169 by linking \c{.OBJ} files together.
6171 Most 16-bit programming language packages come with a suitable
6172 linker; if you have none of these, there is a free linker called
6173 \i{VAL}\I{linker, free}, available in \c{LZH} archive format from
6174 \W{ftp://x2ftp.oulu.fi/pub/msdos/programming/lang/}\i\c{x2ftp.oulu.fi}.
6175 An LZH archiver can be found at
6176 \W{ftp://ftp.simtel.net/pub/simtelnet/msdos/arcers}\i\c{ftp.simtel.net}.
6177 There is another `free' linker (though this one doesn't come with
6178 sources) called \i{FREELINK}, available from
6179 \W{http://www.pcorner.com/tpc/old/3-101.html}\i\c{www.pcorner.com}.
6180 A third, \i\c{djlink}, written by DJ Delorie, is available at
6181 \W{http://www.delorie.com/djgpp/16bit/djlink/}\i\c{www.delorie.com}.
6182 A fourth linker, \i\c{ALINK}, written by Anthony A.J. Williams, is
6183 available at \W{http://alink.sourceforge.net}\i\c{alink.sourceforge.net}.
6185 When linking several \c{.OBJ} files into a \c{.EXE} file, you should
6186 ensure that exactly one of them has a start point defined (using the
6187 \I{program entry point}\i\c{..start} special symbol defined by the
6188 \c{obj} format: see \k{dotdotstart}). If no module defines a start
6189 point, the linker will not know what value to give the entry-point
6190 field in the output file header; if more than one defines a start
6191 point, the linker will not know \e{which} value to use.
6193 An example of a NASM source file which can be assembled to a
6194 \c{.OBJ} file and linked on its own to a \c{.EXE} is given here. It
6195 demonstrates the basic principles of defining a stack, initialising
6196 the segment registers, and declaring a start point. This file is
6197 also provided in the \I{test subdirectory}\c{test} subdirectory of
6198 the NASM archives, under the name \c{objexe.asm}.
6209 This initial piece of code sets up \c{DS} to point to the data
6210 segment, and initializes \c{SS} and \c{SP} to point to the top of
6211 the provided stack. Notice that interrupts are implicitly disabled
6212 for one instruction after a move into \c{SS}, precisely for this
6213 situation, so that there's no chance of an interrupt occurring
6214 between the loads of \c{SS} and \c{SP} and not having a stack to
6217 Note also that the special symbol \c{..start} is defined at the
6218 beginning of this code, which means that will be the entry point
6219 into the resulting executable file.
6225 The above is the main program: load \c{DS:DX} with a pointer to the
6226 greeting message (\c{hello} is implicitly relative to the segment
6227 \c{data}, which was loaded into \c{DS} in the setup code, so the
6228 full pointer is valid), and call the DOS print-string function.
6233 This terminates the program using another DOS system call.
6237 \c hello: db 'hello, world', 13, 10, '$'
6239 The data segment contains the string we want to display.
6241 \c segment stack stack
6245 The above code declares a stack segment containing 64 bytes of
6246 uninitialized stack space, and points \c{stacktop} at the top of it.
6247 The directive \c{segment stack stack} defines a segment \e{called}
6248 \c{stack}, and also of \e{type} \c{STACK}. The latter is not
6249 necessary to the correct running of the program, but linkers are
6250 likely to issue warnings or errors if your program has no segment of
6253 The above file, when assembled into a \c{.OBJ} file, will link on
6254 its own to a valid \c{.EXE} file, which when run will print `hello,
6255 world' and then exit.
6258 \S{binexe} Using the \c{bin} Format To Generate \c{.EXE} Files
6260 The \c{.EXE} file format is simple enough that it's possible to
6261 build a \c{.EXE} file by writing a pure-binary program and sticking
6262 a 32-byte header on the front. This header is simple enough that it
6263 can be generated using \c{DB} and \c{DW} commands by NASM itself, so
6264 that you can use the \c{bin} output format to directly generate
6267 Included in the NASM archives, in the \I{misc subdirectory}\c{misc}
6268 subdirectory, is a file \i\c{exebin.mac} of macros. It defines three
6269 macros: \i\c{EXE_begin}, \i\c{EXE_stack} and \i\c{EXE_end}.
6271 To produce a \c{.EXE} file using this method, you should start by
6272 using \c{%include} to load the \c{exebin.mac} macro package into
6273 your source file. You should then issue the \c{EXE_begin} macro call
6274 (which takes no arguments) to generate the file header data. Then
6275 write code as normal for the \c{bin} format - you can use all three
6276 standard sections \c{.text}, \c{.data} and \c{.bss}. At the end of
6277 the file you should call the \c{EXE_end} macro (again, no arguments),
6278 which defines some symbols to mark section sizes, and these symbols
6279 are referred to in the header code generated by \c{EXE_begin}.
6281 In this model, the code you end up writing starts at \c{0x100}, just
6282 like a \c{.COM} file - in fact, if you strip off the 32-byte header
6283 from the resulting \c{.EXE} file, you will have a valid \c{.COM}
6284 program. All the segment bases are the same, so you are limited to a
6285 64K program, again just like a \c{.COM} file. Note that an \c{ORG}
6286 directive is issued by the \c{EXE_begin} macro, so you should not
6287 explicitly issue one of your own.
6289 You can't directly refer to your segment base value, unfortunately,
6290 since this would require a relocation in the header, and things
6291 would get a lot more complicated. So you should get your segment
6292 base by copying it out of \c{CS} instead.
6294 On entry to your \c{.EXE} file, \c{SS:SP} are already set up to
6295 point to the top of a 2Kb stack. You can adjust the default stack
6296 size of 2Kb by calling the \c{EXE_stack} macro. For example, to
6297 change the stack size of your program to 64 bytes, you would call
6300 A sample program which generates a \c{.EXE} file in this way is
6301 given in the \c{test} subdirectory of the NASM archive, as
6305 \H{comfiles} Producing \i\c{.COM} Files
6307 While large DOS programs must be written as \c{.EXE} files, small
6308 ones are often better written as \c{.COM} files. \c{.COM} files are
6309 pure binary, and therefore most easily produced using the \c{bin}
6313 \S{combinfmt} Using the \c{bin} Format To Generate \c{.COM} Files
6315 \c{.COM} files expect to be loaded at offset \c{100h} into their
6316 segment (though the segment may change). Execution then begins at
6317 \I\c{ORG}\c{100h}, i.e. right at the start of the program. So to
6318 write a \c{.COM} program, you would create a source file looking
6326 \c ; put your code here
6330 \c ; put data items here
6334 \c ; put uninitialized data here
6336 The \c{bin} format puts the \c{.text} section first in the file, so
6337 you can declare data or BSS items before beginning to write code if
6338 you want to and the code will still end up at the front of the file
6341 The BSS (uninitialized data) section does not take up space in the
6342 \c{.COM} file itself: instead, addresses of BSS items are resolved
6343 to point at space beyond the end of the file, on the grounds that
6344 this will be free memory when the program is run. Therefore you
6345 should not rely on your BSS being initialized to all zeros when you
6348 To assemble the above program, you should use a command line like
6350 \c nasm myprog.asm -fbin -o myprog.com
6352 The \c{bin} format would produce a file called \c{myprog} if no
6353 explicit output file name were specified, so you have to override it
6354 and give the desired file name.
6357 \S{comobjfmt} Using the \c{obj} Format To Generate \c{.COM} Files
6359 If you are writing a \c{.COM} program as more than one module, you
6360 may wish to assemble several \c{.OBJ} files and link them together
6361 into a \c{.COM} program. You can do this, provided you have a linker
6362 capable of outputting \c{.COM} files directly (\i{TLINK} does this),
6363 or alternatively a converter program such as \i\c{EXE2BIN} to
6364 transform the \c{.EXE} file output from the linker into a \c{.COM}
6367 If you do this, you need to take care of several things:
6369 \b The first object file containing code should start its code
6370 segment with a line like \c{RESB 100h}. This is to ensure that the
6371 code begins at offset \c{100h} relative to the beginning of the code
6372 segment, so that the linker or converter program does not have to
6373 adjust address references within the file when generating the
6374 \c{.COM} file. Other assemblers use an \i\c{ORG} directive for this
6375 purpose, but \c{ORG} in NASM is a format-specific directive to the
6376 \c{bin} output format, and does not mean the same thing as it does
6377 in MASM-compatible assemblers.
6379 \b You don't need to define a stack segment.
6381 \b All your segments should be in the same group, so that every time
6382 your code or data references a symbol offset, all offsets are
6383 relative to the same segment base. This is because, when a \c{.COM}
6384 file is loaded, all the segment registers contain the same value.
6387 \H{sysfiles} Producing \i\c{.SYS} Files
6389 \i{MS-DOS device drivers} - \c{.SYS} files - are pure binary files,
6390 similar to \c{.COM} files, except that they start at origin zero
6391 rather than \c{100h}. Therefore, if you are writing a device driver
6392 using the \c{bin} format, you do not need the \c{ORG} directive,
6393 since the default origin for \c{bin} is zero. Similarly, if you are
6394 using \c{obj}, you do not need the \c{RESB 100h} at the start of
6397 \c{.SYS} files start with a header structure, containing pointers to
6398 the various routines inside the driver which do the work. This
6399 structure should be defined at the start of the code segment, even
6400 though it is not actually code.
6402 For more information on the format of \c{.SYS} files, and the data
6403 which has to go in the header structure, a list of books is given in
6404 the Frequently Asked Questions list for the newsgroup
6405 \W{news:comp.os.msdos.programmer}\i\c{comp.os.msdos.programmer}.
6408 \H{16c} Interfacing to 16-bit C Programs
6410 This section covers the basics of writing assembly routines that
6411 call, or are called from, C programs. To do this, you would
6412 typically write an assembly module as a \c{.OBJ} file, and link it
6413 with your C modules to produce a \i{mixed-language program}.
6416 \S{16cunder} External Symbol Names
6418 \I{C symbol names}\I{underscore, in C symbols}C compilers have the
6419 convention that the names of all global symbols (functions or data)
6420 they define are formed by prefixing an underscore to the name as it
6421 appears in the C program. So, for example, the function a C
6422 programmer thinks of as \c{printf} appears to an assembly language
6423 programmer as \c{_printf}. This means that in your assembly
6424 programs, you can define symbols without a leading underscore, and
6425 not have to worry about name clashes with C symbols.
6427 If you find the underscores inconvenient, you can define macros to
6428 replace the \c{GLOBAL} and \c{EXTERN} directives as follows:
6444 (These forms of the macros only take one argument at a time; a
6445 \c{%rep} construct could solve this.)
6447 If you then declare an external like this:
6451 then the macro will expand it as
6454 \c %define printf _printf
6456 Thereafter, you can reference \c{printf} as if it was a symbol, and
6457 the preprocessor will put the leading underscore on where necessary.
6459 The \c{cglobal} macro works similarly. You must use \c{cglobal}
6460 before defining the symbol in question, but you would have had to do
6461 that anyway if you used \c{GLOBAL}.
6463 Also see \k{opt-pfix}.
6465 \S{16cmodels} \i{Memory Models}
6467 NASM contains no mechanism to support the various C memory models
6468 directly; you have to keep track yourself of which one you are
6469 writing for. This means you have to keep track of the following
6472 \b In models using a single code segment (tiny, small and compact),
6473 functions are near. This means that function pointers, when stored
6474 in data segments or pushed on the stack as function arguments, are
6475 16 bits long and contain only an offset field (the \c{CS} register
6476 never changes its value, and always gives the segment part of the
6477 full function address), and that functions are called using ordinary
6478 near \c{CALL} instructions and return using \c{RETN} (which, in
6479 NASM, is synonymous with \c{RET} anyway). This means both that you
6480 should write your own routines to return with \c{RETN}, and that you
6481 should call external C routines with near \c{CALL} instructions.
6483 \b In models using more than one code segment (medium, large and
6484 huge), functions are far. This means that function pointers are 32
6485 bits long (consisting of a 16-bit offset followed by a 16-bit
6486 segment), and that functions are called using \c{CALL FAR} (or
6487 \c{CALL seg:offset}) and return using \c{RETF}. Again, you should
6488 therefore write your own routines to return with \c{RETF} and use
6489 \c{CALL FAR} to call external routines.
6491 \b In models using a single data segment (tiny, small and medium),
6492 data pointers are 16 bits long, containing only an offset field (the
6493 \c{DS} register doesn't change its value, and always gives the
6494 segment part of the full data item address).
6496 \b In models using more than one data segment (compact, large and
6497 huge), data pointers are 32 bits long, consisting of a 16-bit offset
6498 followed by a 16-bit segment. You should still be careful not to
6499 modify \c{DS} in your routines without restoring it afterwards, but
6500 \c{ES} is free for you to use to access the contents of 32-bit data
6501 pointers you are passed.
6503 \b The huge memory model allows single data items to exceed 64K in
6504 size. In all other memory models, you can access the whole of a data
6505 item just by doing arithmetic on the offset field of the pointer you
6506 are given, whether a segment field is present or not; in huge model,
6507 you have to be more careful of your pointer arithmetic.
6509 \b In most memory models, there is a \e{default} data segment, whose
6510 segment address is kept in \c{DS} throughout the program. This data
6511 segment is typically the same segment as the stack, kept in \c{SS},
6512 so that functions' local variables (which are stored on the stack)
6513 and global data items can both be accessed easily without changing
6514 \c{DS}. Particularly large data items are typically stored in other
6515 segments. However, some memory models (though not the standard
6516 ones, usually) allow the assumption that \c{SS} and \c{DS} hold the
6517 same value to be removed. Be careful about functions' local
6518 variables in this latter case.
6520 In models with a single code segment, the segment is called
6521 \i\c{_TEXT}, so your code segment must also go by this name in order
6522 to be linked into the same place as the main code segment. In models
6523 with a single data segment, or with a default data segment, it is
6527 \S{16cfunc} Function Definitions and Function Calls
6529 \I{functions, C calling convention}The \i{C calling convention} in
6530 16-bit programs is as follows. In the following description, the
6531 words \e{caller} and \e{callee} are used to denote the function
6532 doing the calling and the function which gets called.
6534 \b The caller pushes the function's parameters on the stack, one
6535 after another, in reverse order (right to left, so that the first
6536 argument specified to the function is pushed last).
6538 \b The caller then executes a \c{CALL} instruction to pass control
6539 to the callee. This \c{CALL} is either near or far depending on the
6542 \b The callee receives control, and typically (although this is not
6543 actually necessary, in functions which do not need to access their
6544 parameters) starts by saving the value of \c{SP} in \c{BP} so as to
6545 be able to use \c{BP} as a base pointer to find its parameters on
6546 the stack. However, the caller was probably doing this too, so part
6547 of the calling convention states that \c{BP} must be preserved by
6548 any C function. Hence the callee, if it is going to set up \c{BP} as
6549 a \i\e{frame pointer}, must push the previous value first.
6551 \b The callee may then access its parameters relative to \c{BP}.
6552 The word at \c{[BP]} holds the previous value of \c{BP} as it was
6553 pushed; the next word, at \c{[BP+2]}, holds the offset part of the
6554 return address, pushed implicitly by \c{CALL}. In a small-model
6555 (near) function, the parameters start after that, at \c{[BP+4]}; in
6556 a large-model (far) function, the segment part of the return address
6557 lives at \c{[BP+4]}, and the parameters begin at \c{[BP+6]}. The
6558 leftmost parameter of the function, since it was pushed last, is
6559 accessible at this offset from \c{BP}; the others follow, at
6560 successively greater offsets. Thus, in a function such as \c{printf}
6561 which takes a variable number of parameters, the pushing of the
6562 parameters in reverse order means that the function knows where to
6563 find its first parameter, which tells it the number and type of the
6566 \b The callee may also wish to decrease \c{SP} further, so as to
6567 allocate space on the stack for local variables, which will then be
6568 accessible at negative offsets from \c{BP}.
6570 \b The callee, if it wishes to return a value to the caller, should
6571 leave the value in \c{AL}, \c{AX} or \c{DX:AX} depending on the size
6572 of the value. Floating-point results are sometimes (depending on the
6573 compiler) returned in \c{ST0}.
6575 \b Once the callee has finished processing, it restores \c{SP} from
6576 \c{BP} if it had allocated local stack space, then pops the previous
6577 value of \c{BP}, and returns via \c{RETN} or \c{RETF} depending on
6580 \b When the caller regains control from the callee, the function
6581 parameters are still on the stack, so it typically adds an immediate
6582 constant to \c{SP} to remove them (instead of executing a number of
6583 slow \c{POP} instructions). Thus, if a function is accidentally
6584 called with the wrong number of parameters due to a prototype
6585 mismatch, the stack will still be returned to a sensible state since
6586 the caller, which \e{knows} how many parameters it pushed, does the
6589 It is instructive to compare this calling convention with that for
6590 Pascal programs (described in \k{16bpfunc}). Pascal has a simpler
6591 convention, since no functions have variable numbers of parameters.
6592 Therefore the callee knows how many parameters it should have been
6593 passed, and is able to deallocate them from the stack itself by
6594 passing an immediate argument to the \c{RET} or \c{RETF}
6595 instruction, so the caller does not have to do it. Also, the
6596 parameters are pushed in left-to-right order, not right-to-left,
6597 which means that a compiler can give better guarantees about
6598 sequence points without performance suffering.
6600 Thus, you would define a function in C style in the following way.
6601 The following example is for small model:
6608 \c sub sp,0x40 ; 64 bytes of local stack space
6609 \c mov bx,[bp+4] ; first parameter to function
6613 \c mov sp,bp ; undo "sub sp,0x40" above
6617 For a large-model function, you would replace \c{RET} by \c{RETF},
6618 and look for the first parameter at \c{[BP+6]} instead of
6619 \c{[BP+4]}. Of course, if one of the parameters is a pointer, then
6620 the offsets of \e{subsequent} parameters will change depending on
6621 the memory model as well: far pointers take up four bytes on the
6622 stack when passed as a parameter, whereas near pointers take up two.
6624 At the other end of the process, to call a C function from your
6625 assembly code, you would do something like this:
6629 \c ; and then, further down...
6631 \c push word [myint] ; one of my integer variables
6632 \c push word mystring ; pointer into my data segment
6634 \c add sp,byte 4 ; `byte' saves space
6636 \c ; then those data items...
6641 \c mystring db 'This number -> %d <- should be 1234',10,0
6643 This piece of code is the small-model assembly equivalent of the C
6646 \c int myint = 1234;
6647 \c printf("This number -> %d <- should be 1234\n", myint);
6649 In large model, the function-call code might look more like this. In
6650 this example, it is assumed that \c{DS} already holds the segment
6651 base of the segment \c{_DATA}. If not, you would have to initialize
6654 \c push word [myint]
6655 \c push word seg mystring ; Now push the segment, and...
6656 \c push word mystring ; ... offset of "mystring"
6660 The integer value still takes up one word on the stack, since large
6661 model does not affect the size of the \c{int} data type. The first
6662 argument (pushed last) to \c{printf}, however, is a data pointer,
6663 and therefore has to contain a segment and offset part. The segment
6664 should be stored second in memory, and therefore must be pushed
6665 first. (Of course, \c{PUSH DS} would have been a shorter instruction
6666 than \c{PUSH WORD SEG mystring}, if \c{DS} was set up as the above
6667 example assumed.) Then the actual call becomes a far call, since
6668 functions expect far calls in large model; and \c{SP} has to be
6669 increased by 6 rather than 4 afterwards to make up for the extra
6673 \S{16cdata} Accessing Data Items
6675 To get at the contents of C variables, or to declare variables which
6676 C can access, you need only declare the names as \c{GLOBAL} or
6677 \c{EXTERN}. (Again, the names require leading underscores, as stated
6678 in \k{16cunder}.) Thus, a C variable declared as \c{int i} can be
6679 accessed from assembler as
6685 And to declare your own integer variable which C programs can access
6686 as \c{extern int j}, you do this (making sure you are assembling in
6687 the \c{_DATA} segment, if necessary):
6693 To access a C array, you need to know the size of the components of
6694 the array. For example, \c{int} variables are two bytes long, so if
6695 a C program declares an array as \c{int a[10]}, you can access
6696 \c{a[3]} by coding \c{mov ax,[_a+6]}. (The byte offset 6 is obtained
6697 by multiplying the desired array index, 3, by the size of the array
6698 element, 2.) The sizes of the C base types in 16-bit compilers are:
6699 1 for \c{char}, 2 for \c{short} and \c{int}, 4 for \c{long} and
6700 \c{float}, and 8 for \c{double}.
6702 To access a C \i{data structure}, you need to know the offset from
6703 the base of the structure to the field you are interested in. You
6704 can either do this by converting the C structure definition into a
6705 NASM structure definition (using \i\c{STRUC}), or by calculating the
6706 one offset and using just that.
6708 To do either of these, you should read your C compiler's manual to
6709 find out how it organizes data structures. NASM gives no special
6710 alignment to structure members in its own \c{STRUC} macro, so you
6711 have to specify alignment yourself if the C compiler generates it.
6712 Typically, you might find that a structure like
6719 might be four bytes long rather than three, since the \c{int} field
6720 would be aligned to a two-byte boundary. However, this sort of
6721 feature tends to be a configurable option in the C compiler, either
6722 using command-line options or \c{#pragma} lines, so you have to find
6723 out how your own compiler does it.
6726 \S{16cmacro} \i\c{c16.mac}: Helper Macros for the 16-bit C Interface
6728 Included in the NASM archives, in the \I{misc subdirectory}\c{misc}
6729 directory, is a file \c{c16.mac} of macros. It defines three macros:
6730 \i\c{proc}, \i\c{arg} and \i\c{endproc}. These are intended to be
6731 used for C-style procedure definitions, and they automate a lot of
6732 the work involved in keeping track of the calling convention.
6734 (An alternative, TASM compatible form of \c{arg} is also now built
6735 into NASM's preprocessor. See \k{stackrel} for details.)
6737 An example of an assembly function using the macro set is given
6744 \c mov ax,[bp + %$i]
6745 \c mov bx,[bp + %$j]
6750 This defines \c{_nearproc} to be a procedure taking two arguments,
6751 the first (\c{i}) an integer and the second (\c{j}) a pointer to an
6752 integer. It returns \c{i + *j}.
6754 Note that the \c{arg} macro has an \c{EQU} as the first line of its
6755 expansion, and since the label before the macro call gets prepended
6756 to the first line of the expanded macro, the \c{EQU} works, defining
6757 \c{%$i} to be an offset from \c{BP}. A context-local variable is
6758 used, local to the context pushed by the \c{proc} macro and popped
6759 by the \c{endproc} macro, so that the same argument name can be used
6760 in later procedures. Of course, you don't \e{have} to do that.
6762 The macro set produces code for near functions (tiny, small and
6763 compact-model code) by default. You can have it generate far
6764 functions (medium, large and huge-model code) by means of coding
6765 \I\c{FARCODE}\c{%define FARCODE}. This changes the kind of return
6766 instruction generated by \c{endproc}, and also changes the starting
6767 point for the argument offsets. The macro set contains no intrinsic
6768 dependency on whether data pointers are far or not.
6770 \c{arg} can take an optional parameter, giving the size of the
6771 argument. If no size is given, 2 is assumed, since it is likely that
6772 many function parameters will be of type \c{int}.
6774 The large-model equivalent of the above function would look like this:
6782 \c mov ax,[bp + %$i]
6783 \c mov bx,[bp + %$j]
6784 \c mov es,[bp + %$j + 2]
6789 This makes use of the argument to the \c{arg} macro to define a
6790 parameter of size 4, because \c{j} is now a far pointer. When we
6791 load from \c{j}, we must load a segment and an offset.
6794 \H{16bp} Interfacing to \i{Borland Pascal} Programs
6796 Interfacing to Borland Pascal programs is similar in concept to
6797 interfacing to 16-bit C programs. The differences are:
6799 \b The leading underscore required for interfacing to C programs is
6800 not required for Pascal.
6802 \b The memory model is always large: functions are far, data
6803 pointers are far, and no data item can be more than 64K long.
6804 (Actually, some functions are near, but only those functions that
6805 are local to a Pascal unit and never called from outside it. All
6806 assembly functions that Pascal calls, and all Pascal functions that
6807 assembly routines are able to call, are far.) However, all static
6808 data declared in a Pascal program goes into the default data
6809 segment, which is the one whose segment address will be in \c{DS}
6810 when control is passed to your assembly code. The only things that
6811 do not live in the default data segment are local variables (they
6812 live in the stack segment) and dynamically allocated variables. All
6813 data \e{pointers}, however, are far.
6815 \b The function calling convention is different - described below.
6817 \b Some data types, such as strings, are stored differently.
6819 \b There are restrictions on the segment names you are allowed to
6820 use - Borland Pascal will ignore code or data declared in a segment
6821 it doesn't like the name of. The restrictions are described below.
6824 \S{16bpfunc} The Pascal Calling Convention
6826 \I{functions, Pascal calling convention}\I{Pascal calling
6827 convention}The 16-bit Pascal calling convention is as follows. In
6828 the following description, the words \e{caller} and \e{callee} are
6829 used to denote the function doing the calling and the function which
6832 \b The caller pushes the function's parameters on the stack, one
6833 after another, in normal order (left to right, so that the first
6834 argument specified to the function is pushed first).
6836 \b The caller then executes a far \c{CALL} instruction to pass
6837 control to the callee.
6839 \b The callee receives control, and typically (although this is not
6840 actually necessary, in functions which do not need to access their
6841 parameters) starts by saving the value of \c{SP} in \c{BP} so as to
6842 be able to use \c{BP} as a base pointer to find its parameters on
6843 the stack. However, the caller was probably doing this too, so part
6844 of the calling convention states that \c{BP} must be preserved by
6845 any function. Hence the callee, if it is going to set up \c{BP} as a
6846 \i{frame pointer}, must push the previous value first.
6848 \b The callee may then access its parameters relative to \c{BP}.
6849 The word at \c{[BP]} holds the previous value of \c{BP} as it was
6850 pushed. The next word, at \c{[BP+2]}, holds the offset part of the
6851 return address, and the next one at \c{[BP+4]} the segment part. The
6852 parameters begin at \c{[BP+6]}. The rightmost parameter of the
6853 function, since it was pushed last, is accessible at this offset
6854 from \c{BP}; the others follow, at successively greater offsets.
6856 \b The callee may also wish to decrease \c{SP} further, so as to
6857 allocate space on the stack for local variables, which will then be
6858 accessible at negative offsets from \c{BP}.
6860 \b The callee, if it wishes to return a value to the caller, should
6861 leave the value in \c{AL}, \c{AX} or \c{DX:AX} depending on the size
6862 of the value. Floating-point results are returned in \c{ST0}.
6863 Results of type \c{Real} (Borland's own custom floating-point data
6864 type, not handled directly by the FPU) are returned in \c{DX:BX:AX}.
6865 To return a result of type \c{String}, the caller pushes a pointer
6866 to a temporary string before pushing the parameters, and the callee
6867 places the returned string value at that location. The pointer is
6868 not a parameter, and should not be removed from the stack by the
6869 \c{RETF} instruction.
6871 \b Once the callee has finished processing, it restores \c{SP} from
6872 \c{BP} if it had allocated local stack space, then pops the previous
6873 value of \c{BP}, and returns via \c{RETF}. It uses the form of
6874 \c{RETF} with an immediate parameter, giving the number of bytes
6875 taken up by the parameters on the stack. This causes the parameters
6876 to be removed from the stack as a side effect of the return
6879 \b When the caller regains control from the callee, the function
6880 parameters have already been removed from the stack, so it needs to
6883 Thus, you would define a function in Pascal style, taking two
6884 \c{Integer}-type parameters, in the following way:
6890 \c sub sp,0x40 ; 64 bytes of local stack space
6891 \c mov bx,[bp+8] ; first parameter to function
6892 \c mov bx,[bp+6] ; second parameter to function
6896 \c mov sp,bp ; undo "sub sp,0x40" above
6898 \c retf 4 ; total size of params is 4
6900 At the other end of the process, to call a Pascal function from your
6901 assembly code, you would do something like this:
6905 \c ; and then, further down...
6907 \c push word seg mystring ; Now push the segment, and...
6908 \c push word mystring ; ... offset of "mystring"
6909 \c push word [myint] ; one of my variables
6910 \c call far SomeFunc
6912 This is equivalent to the Pascal code
6914 \c procedure SomeFunc(String: PChar; Int: Integer);
6915 \c SomeFunc(@mystring, myint);
6918 \S{16bpseg} Borland Pascal \I{segment names, Borland Pascal}Segment
6921 Since Borland Pascal's internal unit file format is completely
6922 different from \c{OBJ}, it only makes a very sketchy job of actually
6923 reading and understanding the various information contained in a
6924 real \c{OBJ} file when it links that in. Therefore an object file
6925 intended to be linked to a Pascal program must obey a number of
6928 \b Procedures and functions must be in a segment whose name is
6929 either \c{CODE}, \c{CSEG}, or something ending in \c{_TEXT}.
6931 \b initialized data must be in a segment whose name is either
6932 \c{CONST} or something ending in \c{_DATA}.
6934 \b Uninitialized data must be in a segment whose name is either
6935 \c{DATA}, \c{DSEG}, or something ending in \c{_BSS}.
6937 \b Any other segments in the object file are completely ignored.
6938 \c{GROUP} directives and segment attributes are also ignored.
6941 \S{16bpmacro} Using \i\c{c16.mac} With Pascal Programs
6943 The \c{c16.mac} macro package, described in \k{16cmacro}, can also
6944 be used to simplify writing functions to be called from Pascal
6945 programs, if you code \I\c{PASCAL}\c{%define PASCAL}. This
6946 definition ensures that functions are far (it implies
6947 \i\c{FARCODE}), and also causes procedure return instructions to be
6948 generated with an operand.
6950 Defining \c{PASCAL} does not change the code which calculates the
6951 argument offsets; you must declare your function's arguments in
6952 reverse order. For example:
6960 \c mov ax,[bp + %$i]
6961 \c mov bx,[bp + %$j]
6962 \c mov es,[bp + %$j + 2]
6967 This defines the same routine, conceptually, as the example in
6968 \k{16cmacro}: it defines a function taking two arguments, an integer
6969 and a pointer to an integer, which returns the sum of the integer
6970 and the contents of the pointer. The only difference between this
6971 code and the large-model C version is that \c{PASCAL} is defined
6972 instead of \c{FARCODE}, and that the arguments are declared in
6976 \C{32bit} Writing 32-bit Code (Unix, Win32, DJGPP)
6978 This chapter attempts to cover some of the common issues involved
6979 when writing 32-bit code, to run under \i{Win32} or Unix, or to be
6980 linked with C code generated by a Unix-style C compiler such as
6981 \i{DJGPP}. It covers how to write assembly code to interface with
6982 32-bit C routines, and how to write position-independent code for
6985 Almost all 32-bit code, and in particular all code running under
6986 \c{Win32}, \c{DJGPP} or any of the PC Unix variants, runs in \I{flat
6987 memory model}\e{flat} memory model. This means that the segment registers
6988 and paging have already been set up to give you the same 32-bit 4Gb
6989 address space no matter what segment you work relative to, and that
6990 you should ignore all segment registers completely. When writing
6991 flat-model application code, you never need to use a segment
6992 override or modify any segment register, and the code-section
6993 addresses you pass to \c{CALL} and \c{JMP} live in the same address
6994 space as the data-section addresses you access your variables by and
6995 the stack-section addresses you access local variables and procedure
6996 parameters by. Every address is 32 bits long and contains only an
7000 \H{32c} Interfacing to 32-bit C Programs
7002 A lot of the discussion in \k{16c}, about interfacing to 16-bit C
7003 programs, still applies when working in 32 bits. The absence of
7004 memory models or segmentation worries simplifies things a lot.
7007 \S{32cunder} External Symbol Names
7009 Most 32-bit C compilers share the convention used by 16-bit
7010 compilers, that the names of all global symbols (functions or data)
7011 they define are formed by prefixing an underscore to the name as it
7012 appears in the C program. However, not all of them do: the \c{ELF}
7013 specification states that C symbols do \e{not} have a leading
7014 underscore on their assembly-language names.
7016 The older Linux \c{a.out} C compiler, all \c{Win32} compilers,
7017 \c{DJGPP}, and \c{NetBSD} and \c{FreeBSD}, all use the leading
7018 underscore; for these compilers, the macros \c{cextern} and
7019 \c{cglobal}, as given in \k{16cunder}, will still work. For \c{ELF},
7020 though, the leading underscore should not be used.
7022 See also \k{opt-pfix}.
7024 \S{32cfunc} Function Definitions and Function Calls
7026 \I{functions, C calling convention}The \i{C calling convention}
7027 in 32-bit programs is as follows. In the following description,
7028 the words \e{caller} and \e{callee} are used to denote
7029 the function doing the calling and the function which gets called.
7031 \b The caller pushes the function's parameters on the stack, one
7032 after another, in reverse order (right to left, so that the first
7033 argument specified to the function is pushed last).
7035 \b The caller then executes a near \c{CALL} instruction to pass
7036 control to the callee.
7038 \b The callee receives control, and typically (although this is not
7039 actually necessary, in functions which do not need to access their
7040 parameters) starts by saving the value of \c{ESP} in \c{EBP} so as
7041 to be able to use \c{EBP} as a base pointer to find its parameters
7042 on the stack. However, the caller was probably doing this too, so
7043 part of the calling convention states that \c{EBP} must be preserved
7044 by any C function. Hence the callee, if it is going to set up
7045 \c{EBP} as a \i{frame pointer}, must push the previous value first.
7047 \b The callee may then access its parameters relative to \c{EBP}.
7048 The doubleword at \c{[EBP]} holds the previous value of \c{EBP} as
7049 it was pushed; the next doubleword, at \c{[EBP+4]}, holds the return
7050 address, pushed implicitly by \c{CALL}. The parameters start after
7051 that, at \c{[EBP+8]}. The leftmost parameter of the function, since
7052 it was pushed last, is accessible at this offset from \c{EBP}; the
7053 others follow, at successively greater offsets. Thus, in a function
7054 such as \c{printf} which takes a variable number of parameters, the
7055 pushing of the parameters in reverse order means that the function
7056 knows where to find its first parameter, which tells it the number
7057 and type of the remaining ones.
7059 \b The callee may also wish to decrease \c{ESP} further, so as to
7060 allocate space on the stack for local variables, which will then be
7061 accessible at negative offsets from \c{EBP}.
7063 \b The callee, if it wishes to return a value to the caller, should
7064 leave the value in \c{AL}, \c{AX} or \c{EAX} depending on the size
7065 of the value. Floating-point results are typically returned in
7068 \b Once the callee has finished processing, it restores \c{ESP} from
7069 \c{EBP} if it had allocated local stack space, then pops the previous
7070 value of \c{EBP}, and returns via \c{RET} (equivalently, \c{RETN}).
7072 \b When the caller regains control from the callee, the function
7073 parameters are still on the stack, so it typically adds an immediate
7074 constant to \c{ESP} to remove them (instead of executing a number of
7075 slow \c{POP} instructions). Thus, if a function is accidentally
7076 called with the wrong number of parameters due to a prototype
7077 mismatch, the stack will still be returned to a sensible state since
7078 the caller, which \e{knows} how many parameters it pushed, does the
7081 There is an alternative calling convention used by Win32 programs
7082 for Windows API calls, and also for functions called \e{by} the
7083 Windows API such as window procedures: they follow what Microsoft
7084 calls the \c{__stdcall} convention. This is slightly closer to the
7085 Pascal convention, in that the callee clears the stack by passing a
7086 parameter to the \c{RET} instruction. However, the parameters are
7087 still pushed in right-to-left order.
7089 Thus, you would define a function in C style in the following way:
7096 \c sub esp,0x40 ; 64 bytes of local stack space
7097 \c mov ebx,[ebp+8] ; first parameter to function
7101 \c leave ; mov esp,ebp / pop ebp
7104 At the other end of the process, to call a C function from your
7105 assembly code, you would do something like this:
7109 \c ; and then, further down...
7111 \c push dword [myint] ; one of my integer variables
7112 \c push dword mystring ; pointer into my data segment
7114 \c add esp,byte 8 ; `byte' saves space
7116 \c ; then those data items...
7121 \c mystring db 'This number -> %d <- should be 1234',10,0
7123 This piece of code is the assembly equivalent of the C code
7125 \c int myint = 1234;
7126 \c printf("This number -> %d <- should be 1234\n", myint);
7129 \S{32cdata} Accessing Data Items
7131 To get at the contents of C variables, or to declare variables which
7132 C can access, you need only declare the names as \c{GLOBAL} or
7133 \c{EXTERN}. (Again, the names require leading underscores, as stated
7134 in \k{32cunder}.) Thus, a C variable declared as \c{int i} can be
7135 accessed from assembler as
7140 And to declare your own integer variable which C programs can access
7141 as \c{extern int j}, you do this (making sure you are assembling in
7142 the \c{_DATA} segment, if necessary):
7147 To access a C array, you need to know the size of the components of
7148 the array. For example, \c{int} variables are four bytes long, so if
7149 a C program declares an array as \c{int a[10]}, you can access
7150 \c{a[3]} by coding \c{mov ax,[_a+12]}. (The byte offset 12 is obtained
7151 by multiplying the desired array index, 3, by the size of the array
7152 element, 4.) The sizes of the C base types in 32-bit compilers are:
7153 1 for \c{char}, 2 for \c{short}, 4 for \c{int}, \c{long} and
7154 \c{float}, and 8 for \c{double}. Pointers, being 32-bit addresses,
7155 are also 4 bytes long.
7157 To access a C \i{data structure}, you need to know the offset from
7158 the base of the structure to the field you are interested in. You
7159 can either do this by converting the C structure definition into a
7160 NASM structure definition (using \c{STRUC}), or by calculating the
7161 one offset and using just that.
7163 To do either of these, you should read your C compiler's manual to
7164 find out how it organizes data structures. NASM gives no special
7165 alignment to structure members in its own \i\c{STRUC} macro, so you
7166 have to specify alignment yourself if the C compiler generates it.
7167 Typically, you might find that a structure like
7174 might be eight bytes long rather than five, since the \c{int} field
7175 would be aligned to a four-byte boundary. However, this sort of
7176 feature is sometimes a configurable option in the C compiler, either
7177 using command-line options or \c{#pragma} lines, so you have to find
7178 out how your own compiler does it.
7181 \S{32cmacro} \i\c{c32.mac}: Helper Macros for the 32-bit C Interface
7183 Included in the NASM archives, in the \I{misc directory}\c{misc}
7184 directory, is a file \c{c32.mac} of macros. It defines three macros:
7185 \i\c{proc}, \i\c{arg} and \i\c{endproc}. These are intended to be
7186 used for C-style procedure definitions, and they automate a lot of
7187 the work involved in keeping track of the calling convention.
7189 An example of an assembly function using the macro set is given
7196 \c mov eax,[ebp + %$i]
7197 \c mov ebx,[ebp + %$j]
7202 This defines \c{_proc32} to be a procedure taking two arguments, the
7203 first (\c{i}) an integer and the second (\c{j}) a pointer to an
7204 integer. It returns \c{i + *j}.
7206 Note that the \c{arg} macro has an \c{EQU} as the first line of its
7207 expansion, and since the label before the macro call gets prepended
7208 to the first line of the expanded macro, the \c{EQU} works, defining
7209 \c{%$i} to be an offset from \c{BP}. A context-local variable is
7210 used, local to the context pushed by the \c{proc} macro and popped
7211 by the \c{endproc} macro, so that the same argument name can be used
7212 in later procedures. Of course, you don't \e{have} to do that.
7214 \c{arg} can take an optional parameter, giving the size of the
7215 argument. If no size is given, 4 is assumed, since it is likely that
7216 many function parameters will be of type \c{int} or pointers.
7219 \H{picdll} Writing NetBSD/FreeBSD/OpenBSD and Linux/ELF \i{Shared
7222 \c{ELF} replaced the older \c{a.out} object file format under Linux
7223 because it contains support for \i{position-independent code}
7224 (\i{PIC}), which makes writing shared libraries much easier. NASM
7225 supports the \c{ELF} position-independent code features, so you can
7226 write Linux \c{ELF} shared libraries in NASM.
7228 \i{NetBSD}, and its close cousins \i{FreeBSD} and \i{OpenBSD}, take
7229 a different approach by hacking PIC support into the \c{a.out}
7230 format. NASM supports this as the \i\c{aoutb} output format, so you
7231 can write \i{BSD} shared libraries in NASM too.
7233 The operating system loads a PIC shared library by memory-mapping
7234 the library file at an arbitrarily chosen point in the address space
7235 of the running process. The contents of the library's code section
7236 must therefore not depend on where it is loaded in memory.
7238 Therefore, you cannot get at your variables by writing code like
7241 \c mov eax,[myvar] ; WRONG
7243 Instead, the linker provides an area of memory called the
7244 \i\e{global offset table}, or \i{GOT}; the GOT is situated at a
7245 constant distance from your library's code, so if you can find out
7246 where your library is loaded (which is typically done using a
7247 \c{CALL} and \c{POP} combination), you can obtain the address of the
7248 GOT, and you can then load the addresses of your variables out of
7249 linker-generated entries in the GOT.
7251 The \e{data} section of a PIC shared library does not have these
7252 restrictions: since the data section is writable, it has to be
7253 copied into memory anyway rather than just paged in from the library
7254 file, so as long as it's being copied it can be relocated too. So
7255 you can put ordinary types of relocation in the data section without
7256 too much worry (but see \k{picglobal} for a caveat).
7259 \S{picgot} Obtaining the Address of the GOT
7261 Each code module in your shared library should define the GOT as an
7264 \c extern _GLOBAL_OFFSET_TABLE_ ; in ELF
7265 \c extern __GLOBAL_OFFSET_TABLE_ ; in BSD a.out
7267 At the beginning of any function in your shared library which plans
7268 to access your data or BSS sections, you must first calculate the
7269 address of the GOT. This is typically done by writing the function
7278 \c add ebx,_GLOBAL_OFFSET_TABLE_+$$-.get_GOT wrt ..gotpc
7280 \c ; the function body comes here
7287 (For BSD, again, the symbol \c{_GLOBAL_OFFSET_TABLE} requires a
7288 second leading underscore.)
7290 The first two lines of this function are simply the standard C
7291 prologue to set up a stack frame, and the last three lines are
7292 standard C function epilogue. The third line, and the fourth to last
7293 line, save and restore the \c{EBX} register, because PIC shared
7294 libraries use this register to store the address of the GOT.
7296 The interesting bit is the \c{CALL} instruction and the following
7297 two lines. The \c{CALL} and \c{POP} combination obtains the address
7298 of the label \c{.get_GOT}, without having to know in advance where
7299 the program was loaded (since the \c{CALL} instruction is encoded
7300 relative to the current position). The \c{ADD} instruction makes use
7301 of one of the special PIC relocation types: \i{GOTPC relocation}.
7302 With the \i\c{WRT ..gotpc} qualifier specified, the symbol
7303 referenced (here \c{_GLOBAL_OFFSET_TABLE_}, the special symbol
7304 assigned to the GOT) is given as an offset from the beginning of the
7305 section. (Actually, \c{ELF} encodes it as the offset from the operand
7306 field of the \c{ADD} instruction, but NASM simplifies this
7307 deliberately, so you do things the same way for both \c{ELF} and
7308 \c{BSD}.) So the instruction then \e{adds} the beginning of the section,
7309 to get the real address of the GOT, and subtracts the value of
7310 \c{.get_GOT} which it knows is in \c{EBX}. Therefore, by the time
7311 that instruction has finished, \c{EBX} contains the address of the GOT.
7313 If you didn't follow that, don't worry: it's never necessary to
7314 obtain the address of the GOT by any other means, so you can put
7315 those three instructions into a macro and safely ignore them:
7322 \c add ebx,_GLOBAL_OFFSET_TABLE_+$$-%%getgot wrt ..gotpc
7326 \S{piclocal} Finding Your Local Data Items
7328 Having got the GOT, you can then use it to obtain the addresses of
7329 your data items. Most variables will reside in the sections you have
7330 declared; they can be accessed using the \I{GOTOFF
7331 relocation}\c{..gotoff} special \I\c{WRT ..gotoff}\c{WRT} type. The
7332 way this works is like this:
7334 \c lea eax,[ebx+myvar wrt ..gotoff]
7336 The expression \c{myvar wrt ..gotoff} is calculated, when the shared
7337 library is linked, to be the offset to the local variable \c{myvar}
7338 from the beginning of the GOT. Therefore, adding it to \c{EBX} as
7339 above will place the real address of \c{myvar} in \c{EAX}.
7341 If you declare variables as \c{GLOBAL} without specifying a size for
7342 them, they are shared between code modules in the library, but do
7343 not get exported from the library to the program that loaded it.
7344 They will still be in your ordinary data and BSS sections, so you
7345 can access them in the same way as local variables, using the above
7346 \c{..gotoff} mechanism.
7348 Note that due to a peculiarity of the way BSD \c{a.out} format
7349 handles this relocation type, there must be at least one non-local
7350 symbol in the same section as the address you're trying to access.
7353 \S{picextern} Finding External and Common Data Items
7355 If your library needs to get at an external variable (external to
7356 the \e{library}, not just to one of the modules within it), you must
7357 use the \I{GOT relocations}\I\c{WRT ..got}\c{..got} type to get at
7358 it. The \c{..got} type, instead of giving you the offset from the
7359 GOT base to the variable, gives you the offset from the GOT base to
7360 a GOT \e{entry} containing the address of the variable. The linker
7361 will set up this GOT entry when it builds the library, and the
7362 dynamic linker will place the correct address in it at load time. So
7363 to obtain the address of an external variable \c{extvar} in \c{EAX},
7366 \c mov eax,[ebx+extvar wrt ..got]
7368 This loads the address of \c{extvar} out of an entry in the GOT. The
7369 linker, when it builds the shared library, collects together every
7370 relocation of type \c{..got}, and builds the GOT so as to ensure it
7371 has every necessary entry present.
7373 Common variables must also be accessed in this way.
7376 \S{picglobal} Exporting Symbols to the Library User
7378 If you want to export symbols to the user of the library, you have
7379 to declare whether they are functions or data, and if they are data,
7380 you have to give the size of the data item. This is because the
7381 dynamic linker has to build \I{PLT}\i{procedure linkage table}
7382 entries for any exported functions, and also moves exported data
7383 items away from the library's data section in which they were
7386 So to export a function to users of the library, you must use
7388 \c global func:function ; declare it as a function
7394 And to export a data item such as an array, you would have to code
7396 \c global array:data array.end-array ; give the size too
7401 Be careful: If you export a variable to the library user, by
7402 declaring it as \c{GLOBAL} and supplying a size, the variable will
7403 end up living in the data section of the main program, rather than
7404 in your library's data section, where you declared it. So you will
7405 have to access your own global variable with the \c{..got} mechanism
7406 rather than \c{..gotoff}, as if it were external (which,
7407 effectively, it has become).
7409 Equally, if you need to store the address of an exported global in
7410 one of your data sections, you can't do it by means of the standard
7413 \c dataptr: dd global_data_item ; WRONG
7415 NASM will interpret this code as an ordinary relocation, in which
7416 \c{global_data_item} is merely an offset from the beginning of the
7417 \c{.data} section (or whatever); so this reference will end up
7418 pointing at your data section instead of at the exported global
7419 which resides elsewhere.
7421 Instead of the above code, then, you must write
7423 \c dataptr: dd global_data_item wrt ..sym
7425 which makes use of the special \c{WRT} type \I\c{WRT ..sym}\c{..sym}
7426 to instruct NASM to search the symbol table for a particular symbol
7427 at that address, rather than just relocating by section base.
7429 Either method will work for functions: referring to one of your
7430 functions by means of
7432 \c funcptr: dd my_function
7434 will give the user the address of the code you wrote, whereas
7436 \c funcptr: dd my_function wrt ..sym
7438 will give the address of the procedure linkage table for the
7439 function, which is where the calling program will \e{believe} the
7440 function lives. Either address is a valid way to call the function.
7443 \S{picproc} Calling Procedures Outside the Library
7445 Calling procedures outside your shared library has to be done by
7446 means of a \i\e{procedure linkage table}, or \i{PLT}. The PLT is
7447 placed at a known offset from where the library is loaded, so the
7448 library code can make calls to the PLT in a position-independent
7449 way. Within the PLT there is code to jump to offsets contained in
7450 the GOT, so function calls to other shared libraries or to routines
7451 in the main program can be transparently passed off to their real
7454 To call an external routine, you must use another special PIC
7455 relocation type, \I{PLT relocations}\i\c{WRT ..plt}. This is much
7456 easier than the GOT-based ones: you simply replace calls such as
7457 \c{CALL printf} with the PLT-relative version \c{CALL printf WRT
7461 \S{link} Generating the Library File
7463 Having written some code modules and assembled them to \c{.o} files,
7464 you then generate your shared library with a command such as
7466 \c ld -shared -o library.so module1.o module2.o # for ELF
7467 \c ld -Bshareable -o library.so module1.o module2.o # for BSD
7469 For ELF, if your shared library is going to reside in system
7470 directories such as \c{/usr/lib} or \c{/lib}, it is usually worth
7471 using the \i\c{-soname} flag to the linker, to store the final
7472 library file name, with a version number, into the library:
7474 \c ld -shared -soname library.so.1 -o library.so.1.2 *.o
7476 You would then copy \c{library.so.1.2} into the library directory,
7477 and create \c{library.so.1} as a symbolic link to it.
7480 \C{mixsize} Mixing 16 and 32 Bit Code
7482 This chapter tries to cover some of the issues, largely related to
7483 unusual forms of addressing and jump instructions, encountered when
7484 writing operating system code such as protected-mode initialisation
7485 routines, which require code that operates in mixed segment sizes,
7486 such as code in a 16-bit segment trying to modify data in a 32-bit
7487 one, or jumps between different-size segments.
7490 \H{mixjump} Mixed-Size Jumps\I{jumps, mixed-size}
7492 \I{operating system, writing}\I{writing operating systems}The most
7493 common form of \i{mixed-size instruction} is the one used when
7494 writing a 32-bit OS: having done your setup in 16-bit mode, such as
7495 loading the kernel, you then have to boot it by switching into
7496 protected mode and jumping to the 32-bit kernel start address. In a
7497 fully 32-bit OS, this tends to be the \e{only} mixed-size
7498 instruction you need, since everything before it can be done in pure
7499 16-bit code, and everything after it can be pure 32-bit.
7501 This jump must specify a 48-bit far address, since the target
7502 segment is a 32-bit one. However, it must be assembled in a 16-bit
7503 segment, so just coding, for example,
7505 \c jmp 0x1234:0x56789ABC ; wrong!
7507 will not work, since the offset part of the address will be
7508 truncated to \c{0x9ABC} and the jump will be an ordinary 16-bit far
7511 The Linux kernel setup code gets round the inability of \c{as86} to
7512 generate the required instruction by coding it manually, using
7513 \c{DB} instructions. NASM can go one better than that, by actually
7514 generating the right instruction itself. Here's how to do it right:
7516 \c jmp dword 0x1234:0x56789ABC ; right
7518 \I\c{JMP DWORD}The \c{DWORD} prefix (strictly speaking, it should
7519 come \e{after} the colon, since it is declaring the \e{offset} field
7520 to be a doubleword; but NASM will accept either form, since both are
7521 unambiguous) forces the offset part to be treated as far, in the
7522 assumption that you are deliberately writing a jump from a 16-bit
7523 segment to a 32-bit one.
7525 You can do the reverse operation, jumping from a 32-bit segment to a
7526 16-bit one, by means of the \c{WORD} prefix:
7528 \c jmp word 0x8765:0x4321 ; 32 to 16 bit
7530 If the \c{WORD} prefix is specified in 16-bit mode, or the \c{DWORD}
7531 prefix in 32-bit mode, they will be ignored, since each is
7532 explicitly forcing NASM into a mode it was in anyway.
7535 \H{mixaddr} Addressing Between Different-Size Segments\I{addressing,
7536 mixed-size}\I{mixed-size addressing}
7538 If your OS is mixed 16 and 32-bit, or if you are writing a DOS
7539 extender, you are likely to have to deal with some 16-bit segments
7540 and some 32-bit ones. At some point, you will probably end up
7541 writing code in a 16-bit segment which has to access data in a
7542 32-bit segment, or vice versa.
7544 If the data you are trying to access in a 32-bit segment lies within
7545 the first 64K of the segment, you may be able to get away with using
7546 an ordinary 16-bit addressing operation for the purpose; but sooner
7547 or later, you will want to do 32-bit addressing from 16-bit mode.
7549 The easiest way to do this is to make sure you use a register for
7550 the address, since any effective address containing a 32-bit
7551 register is forced to be a 32-bit address. So you can do
7553 \c mov eax,offset_into_32_bit_segment_specified_by_fs
7554 \c mov dword [fs:eax],0x11223344
7556 This is fine, but slightly cumbersome (since it wastes an
7557 instruction and a register) if you already know the precise offset
7558 you are aiming at. The x86 architecture does allow 32-bit effective
7559 addresses to specify nothing but a 4-byte offset, so why shouldn't
7560 NASM be able to generate the best instruction for the purpose?
7562 It can. As in \k{mixjump}, you need only prefix the address with the
7563 \c{DWORD} keyword, and it will be forced to be a 32-bit address:
7565 \c mov dword [fs:dword my_offset],0x11223344
7567 Also as in \k{mixjump}, NASM is not fussy about whether the
7568 \c{DWORD} prefix comes before or after the segment override, so
7569 arguably a nicer-looking way to code the above instruction is
7571 \c mov dword [dword fs:my_offset],0x11223344
7573 Don't confuse the \c{DWORD} prefix \e{outside} the square brackets,
7574 which controls the size of the data stored at the address, with the
7575 one \c{inside} the square brackets which controls the length of the
7576 address itself. The two can quite easily be different:
7578 \c mov word [dword 0x12345678],0x9ABC
7580 This moves 16 bits of data to an address specified by a 32-bit
7583 You can also specify \c{WORD} or \c{DWORD} prefixes along with the
7584 \c{FAR} prefix to indirect far jumps or calls. For example:
7586 \c call dword far [fs:word 0x4321]
7588 This instruction contains an address specified by a 16-bit offset;
7589 it loads a 48-bit far pointer from that (16-bit segment and 32-bit
7590 offset), and calls that address.
7593 \H{mixother} Other Mixed-Size Instructions
7595 The other way you might want to access data might be using the
7596 string instructions (\c{LODSx}, \c{STOSx} and so on) or the
7597 \c{XLATB} instruction. These instructions, since they take no
7598 parameters, might seem to have no easy way to make them perform
7599 32-bit addressing when assembled in a 16-bit segment.
7601 This is the purpose of NASM's \i\c{a16}, \i\c{a32} and \i\c{a64} prefixes. If
7602 you are coding \c{LODSB} in a 16-bit segment but it is supposed to
7603 be accessing a string in a 32-bit segment, you should load the
7604 desired address into \c{ESI} and then code
7608 The prefix forces the addressing size to 32 bits, meaning that
7609 \c{LODSB} loads from \c{[DS:ESI]} instead of \c{[DS:SI]}. To access
7610 a string in a 16-bit segment when coding in a 32-bit one, the
7611 corresponding \c{a16} prefix can be used.
7613 The \c{a16}, \c{a32} and \c{a64} prefixes can be applied to any instruction
7614 in NASM's instruction table, but most of them can generate all the
7615 useful forms without them. The prefixes are necessary only for
7616 instructions with implicit addressing:
7617 \# \c{CMPSx} (\k{insCMPSB}),
7618 \# \c{SCASx} (\k{insSCASB}), \c{LODSx} (\k{insLODSB}), \c{STOSx}
7619 \# (\k{insSTOSB}), \c{MOVSx} (\k{insMOVSB}), \c{INSx} (\k{insINSB}),
7620 \# \c{OUTSx} (\k{insOUTSB}), and \c{XLATB} (\k{insXLATB}).
7621 \c{CMPSx}, \c{SCASx}, \c{LODSx}, \c{STOSx}, \c{MOVSx}, \c{INSx},
7622 \c{OUTSx}, and \c{XLATB}.
7624 various push and pop instructions (\c{PUSHA} and \c{POPF} as well as
7625 the more usual \c{PUSH} and \c{POP}) can accept \c{a16}, \c{a32} or \c{a64}
7626 prefixes to force a particular one of \c{SP}, \c{ESP} or \c{RSP} to be used
7627 as a stack pointer, in case the stack segment in use is a different
7628 size from the code segment.
7630 \c{PUSH} and \c{POP}, when applied to segment registers in 32-bit
7631 mode, also have the slightly odd behaviour that they push and pop 4
7632 bytes at a time, of which the top two are ignored and the bottom two
7633 give the value of the segment register being manipulated. To force
7634 the 16-bit behaviour of segment-register push and pop instructions,
7635 you can use the operand-size prefix \i\c{o16}:
7640 This code saves a doubleword of stack space by fitting two segment
7641 registers into the space which would normally be consumed by pushing
7644 (You can also use the \i\c{o32} prefix to force the 32-bit behaviour
7645 when in 16-bit mode, but this seems less useful.)
7648 \C{64bit} Writing 64-bit Code (Unix, Win64)
7650 This chapter attempts to cover some of the common issues involved when
7651 writing 64-bit code, to run under \i{Win64} or Unix. It covers how to
7652 write assembly code to interface with 64-bit C routines, and how to
7653 write position-independent code for shared libraries.
7655 All 64-bit code uses a flat memory model, since segmentation is not
7656 available in 64-bit mode. The one exception is the \c{FS} and \c{GS}
7657 registers, which still add their bases.
7659 Position independence in 64-bit mode is significantly simpler, since
7660 the processor supports \c{RIP}-relative addressing directly; see the
7661 \c{REL} keyword (\k{effaddr}). On most 64-bit platforms, it is
7662 probably desirable to make that the default, using the directive
7663 \c{DEFAULT REL} (\k{default}).
7665 64-bit programming is relatively similar to 32-bit programming, but
7666 of course pointers are 64 bits long; additionally, all existing
7667 platforms pass arguments in registers rather than on the stack.
7668 Furthermore, 64-bit platforms use SSE2 by default for floating point.
7669 Please see the ABI documentation for your platform.
7671 64-bit platforms differ in the sizes of the fundamental datatypes, not
7672 just from 32-bit platforms but from each other. If a specific size
7673 data type is desired, it is probably best to use the types defined in
7674 the Standard C header \c{<inttypes.h>}.
7676 In 64-bit mode, the default instruction size is still 32 bits. When
7677 loading a value into a 32-bit register (but not an 8- or 16-bit
7678 register), the upper 32 bits of the corresponding 64-bit register are
7681 \H{reg64} Register Names in 64-bit Mode
7683 NASM uses the following names for general-purpose registers in 64-bit
7684 mode, for 8-, 16-, 32- and 64-bit references, respectively:
7686 \c AL/AH, CL/CH, DL/DH, BL/BH, SPL, BPL, SIL, DIL, R8B-R15B
7687 \c AX, CX, DX, BX, SP, BP, SI, DI, R8W-R15W
7688 \c EAX, ECX, EDX, EBX, ESP, EBP, ESI, EDI, R8D-R15D
7689 \c RAX, RCX, RDX, RBX, RSP, RBP, RSI, RDI, R8-R15
7691 This is consistent with the AMD documentation and most other
7692 assemblers. The Intel documentation, however, uses the names
7693 \c{R8L-R15L} for 8-bit references to the higher registers. It is
7694 possible to use those names by definiting them as macros; similarly,
7695 if one wants to use numeric names for the low 8 registers, define them
7696 as macros. The standard macro package \c{altreg} (see \k{pkg_altreg})
7697 can be used for this purpose.
7699 \H{id64} Immediates and Displacements in 64-bit Mode
7701 In 64-bit mode, immediates and displacements are generally only 32
7702 bits wide. NASM will therefore truncate most displacements and
7703 immediates to 32 bits.
7705 The only instruction which takes a full \i{64-bit immediate} is:
7709 NASM will produce this instruction whenever the programmer uses
7710 \c{MOV} with an immediate into a 64-bit register. If this is not
7711 desirable, simply specify the equivalent 32-bit register, which will
7712 be automatically zero-extended by the processor, or specify the
7713 immediate as \c{DWORD}:
7715 \c mov rax,foo ; 64-bit immediate
7716 \c mov rax,qword foo ; (identical)
7717 \c mov eax,foo ; 32-bit immediate, zero-extended
7718 \c mov rax,dword foo ; 32-bit immediate, sign-extended
7720 The length of these instructions are 10, 5 and 7 bytes, respectively.
7722 The only instructions which take a full \I{64-bit displacement}64-bit
7723 \e{displacement} is loading or storing, using \c{MOV}, \c{AL}, \c{AX},
7724 \c{EAX} or \c{RAX} (but no other registers) to an absolute 64-bit address.
7725 Since this is a relatively rarely used instruction (64-bit code generally uses
7726 relative addressing), the programmer has to explicitly declare the
7727 displacement size as \c{QWORD}:
7731 \c mov eax,[foo] ; 32-bit absolute disp, sign-extended
7732 \c mov eax,[a32 foo] ; 32-bit absolute disp, zero-extended
7733 \c mov eax,[qword foo] ; 64-bit absolute disp
7737 \c mov eax,[foo] ; 32-bit relative disp
7738 \c mov eax,[a32 foo] ; d:o, address truncated to 32 bits(!)
7739 \c mov eax,[qword foo] ; error
7740 \c mov eax,[abs qword foo] ; 64-bit absolute disp
7742 A sign-extended absolute displacement can access from -2 GB to +2 GB;
7743 a zero-extended absolute displacement can access from 0 to 4 GB.
7745 \H{unix64} Interfacing to 64-bit C Programs (Unix)
7747 On Unix, the 64-bit ABI is defined by the document:
7749 \W{http://www.nasm.us/links/unix64abi}\c{http://www.nasm.us/links/unix64abi}
7751 Although written for AT&T-syntax assembly, the concepts apply equally
7752 well for NASM-style assembly. What follows is a simplified summary.
7754 The first six integer arguments (from the left) are passed in \c{RDI},
7755 \c{RSI}, \c{RDX}, \c{RCX}, \c{R8}, and \c{R9}, in that order.
7756 Additional integer arguments are passed on the stack. These
7757 registers, plus \c{RAX}, \c{R10} and \c{R11} are destroyed by function
7758 calls, and thus are available for use by the function without saving.
7760 Integer return values are passed in \c{RAX} and \c{RDX}, in that order.
7762 Floating point is done using SSE registers, except for \c{long
7763 double}. Floating-point arguments are passed in \c{XMM0} to \c{XMM7};
7764 return is \c{XMM0} and \c{XMM1}. \c{long double} are passed on the
7765 stack, and returned in \c{ST0} and \c{ST1}.
7767 All SSE and x87 registers are destroyed by function calls.
7769 On 64-bit Unix, \c{long} is 64 bits.
7771 Integer and SSE register arguments are counted separately, so for the case of
7773 \c void foo(long a, double b, int c)
7775 \c{a} is passed in \c{RDI}, \c{b} in \c{XMM0}, and \c{c} in \c{ESI}.
7777 \H{win64} Interfacing to 64-bit C Programs (Win64)
7779 The Win64 ABI is described at:
7781 \W{http://www.nasm.us/links/win64abi}\c{http://www.nasm.us/links/win64abi}
7783 What follows is a simplified summary.
7785 The first four integer arguments are passed in \c{RCX}, \c{RDX},
7786 \c{R8} and \c{R9}, in that order. Additional integer arguments are
7787 passed on the stack. These registers, plus \c{RAX}, \c{R10} and
7788 \c{R11} are destroyed by function calls, and thus are available for
7789 use by the function without saving.
7791 Integer return values are passed in \c{RAX} only.
7793 Floating point is done using SSE registers, except for \c{long
7794 double}. Floating-point arguments are passed in \c{XMM0} to \c{XMM3};
7795 return is \c{XMM0} only.
7797 On Win64, \c{long} is 32 bits; \c{long long} or \c{_int64} is 64 bits.
7799 Integer and SSE register arguments are counted together, so for the case of
7801 \c void foo(long long a, double b, int c)
7803 \c{a} is passed in \c{RCX}, \c{b} in \c{XMM1}, and \c{c} in \c{R8D}.
7805 \C{trouble} Troubleshooting
7807 This chapter describes some of the common problems that users have
7808 been known to encounter with NASM, and answers them. It also gives
7809 instructions for reporting bugs in NASM if you find a difficulty
7810 that isn't listed here.
7813 \H{problems} Common Problems
7815 \S{inefficient} NASM Generates \i{Inefficient Code}
7817 We sometimes get `bug' reports about NASM generating inefficient, or
7818 even `wrong', code on instructions such as \c{ADD ESP,8}. This is a
7819 deliberate design feature, connected to predictability of output:
7820 NASM, on seeing \c{ADD ESP,8}, will generate the form of the
7821 instruction which leaves room for a 32-bit offset. You need to code
7822 \I\c{BYTE}\c{ADD ESP,BYTE 8} if you want the space-efficient form of
7823 the instruction. This isn't a bug, it's user error: if you prefer to
7824 have NASM produce the more efficient code automatically enable
7825 optimization with the \c{-O} option (see \k{opt-O}).
7828 \S{jmprange} My Jumps are Out of Range\I{out of range, jumps}
7830 Similarly, people complain that when they issue \i{conditional
7831 jumps} (which are \c{SHORT} by default) that try to jump too far,
7832 NASM reports `short jump out of range' instead of making the jumps
7835 This, again, is partly a predictability issue, but in fact has a
7836 more practical reason as well. NASM has no means of being told what
7837 type of processor the code it is generating will be run on; so it
7838 cannot decide for itself that it should generate \i\c{Jcc NEAR} type
7839 instructions, because it doesn't know that it's working for a 386 or
7840 above. Alternatively, it could replace the out-of-range short
7841 \c{JNE} instruction with a very short \c{JE} instruction that jumps
7842 over a \c{JMP NEAR}; this is a sensible solution for processors
7843 below a 386, but hardly efficient on processors which have good
7844 branch prediction \e{and} could have used \c{JNE NEAR} instead. So,
7845 once again, it's up to the user, not the assembler, to decide what
7846 instructions should be generated. See \k{opt-O}.
7849 \S{proborg} \i\c{ORG} Doesn't Work
7851 People writing \i{boot sector} programs in the \c{bin} format often
7852 complain that \c{ORG} doesn't work the way they'd like: in order to
7853 place the \c{0xAA55} signature word at the end of a 512-byte boot
7854 sector, people who are used to MASM tend to code
7858 \c ; some boot sector code
7863 This is not the intended use of the \c{ORG} directive in NASM, and
7864 will not work. The correct way to solve this problem in NASM is to
7865 use the \i\c{TIMES} directive, like this:
7869 \c ; some boot sector code
7871 \c TIMES 510-($-$$) DB 0
7874 The \c{TIMES} directive will insert exactly enough zero bytes into
7875 the output to move the assembly point up to 510. This method also
7876 has the advantage that if you accidentally fill your boot sector too
7877 full, NASM will catch the problem at assembly time and report it, so
7878 you won't end up with a boot sector that you have to disassemble to
7879 find out what's wrong with it.
7882 \S{probtimes} \i\c{TIMES} Doesn't Work
7884 The other common problem with the above code is people who write the
7889 by reasoning that \c{$} should be a pure number, just like 510, so
7890 the difference between them is also a pure number and can happily be
7893 NASM is a \e{modular} assembler: the various component parts are
7894 designed to be easily separable for re-use, so they don't exchange
7895 information unnecessarily. In consequence, the \c{bin} output
7896 format, even though it has been told by the \c{ORG} directive that
7897 the \c{.text} section should start at 0, does not pass that
7898 information back to the expression evaluator. So from the
7899 evaluator's point of view, \c{$} isn't a pure number: it's an offset
7900 from a section base. Therefore the difference between \c{$} and 510
7901 is also not a pure number, but involves a section base. Values
7902 involving section bases cannot be passed as arguments to \c{TIMES}.
7904 The solution, as in the previous section, is to code the \c{TIMES}
7907 \c TIMES 510-($-$$) DB 0
7909 in which \c{$} and \c{$$} are offsets from the same section base,
7910 and so their difference is a pure number. This will solve the
7911 problem and generate sensible code.
7914 \H{bugs} \i{Bugs}\I{reporting bugs}
7916 We have never yet released a version of NASM with any \e{known}
7917 bugs. That doesn't usually stop there being plenty we didn't know
7918 about, though. Any that you find should be reported firstly via the
7920 \W{http://www.nasm.us/}\c{http://www.nasm.us/}
7921 (click on "Bug Tracker"), or if that fails then through one of the
7922 contacts in \k{contact}.
7924 Please read \k{qstart} first, and don't report the bug if it's
7925 listed in there as a deliberate feature. (If you think the feature
7926 is badly thought out, feel free to send us reasons why you think it
7927 should be changed, but don't just send us mail saying `This is a
7928 bug' if the documentation says we did it on purpose.) Then read
7929 \k{problems}, and don't bother reporting the bug if it's listed
7932 If you do report a bug, \e{please} give us all of the following
7935 \b What operating system you're running NASM under. DOS, Linux,
7936 NetBSD, Win16, Win32, VMS (I'd be impressed), whatever.
7938 \b If you're running NASM under DOS or Win32, tell us whether you've
7939 compiled your own executable from the DOS source archive, or whether
7940 you were using the standard distribution binaries out of the
7941 archive. If you were using a locally built executable, try to
7942 reproduce the problem using one of the standard binaries, as this
7943 will make it easier for us to reproduce your problem prior to fixing
7946 \b Which version of NASM you're using, and exactly how you invoked
7947 it. Give us the precise command line, and the contents of the
7948 \c{NASMENV} environment variable if any.
7950 \b Which versions of any supplementary programs you're using, and
7951 how you invoked them. If the problem only becomes visible at link
7952 time, tell us what linker you're using, what version of it you've
7953 got, and the exact linker command line. If the problem involves
7954 linking against object files generated by a compiler, tell us what
7955 compiler, what version, and what command line or options you used.
7956 (If you're compiling in an IDE, please try to reproduce the problem
7957 with the command-line version of the compiler.)
7959 \b If at all possible, send us a NASM source file which exhibits the
7960 problem. If this causes copyright problems (e.g. you can only
7961 reproduce the bug in restricted-distribution code) then bear in mind
7962 the following two points: firstly, we guarantee that any source code
7963 sent to us for the purposes of debugging NASM will be used \e{only}
7964 for the purposes of debugging NASM, and that we will delete all our
7965 copies of it as soon as we have found and fixed the bug or bugs in
7966 question; and secondly, we would prefer \e{not} to be mailed large
7967 chunks of code anyway. The smaller the file, the better. A
7968 three-line sample file that does nothing useful \e{except}
7969 demonstrate the problem is much easier to work with than a
7970 fully fledged ten-thousand-line program. (Of course, some errors
7971 \e{do} only crop up in large files, so this may not be possible.)
7973 \b A description of what the problem actually \e{is}. `It doesn't
7974 work' is \e{not} a helpful description! Please describe exactly what
7975 is happening that shouldn't be, or what isn't happening that should.
7976 Examples might be: `NASM generates an error message saying Line 3
7977 for an error that's actually on Line 5'; `NASM generates an error
7978 message that I believe it shouldn't be generating at all'; `NASM
7979 fails to generate an error message that I believe it \e{should} be
7980 generating'; `the object file produced from this source code crashes
7981 my linker'; `the ninth byte of the output file is 66 and I think it
7982 should be 77 instead'.
7984 \b If you believe the output file from NASM to be faulty, send it to
7985 us. That allows us to determine whether our own copy of NASM
7986 generates the same file, or whether the problem is related to
7987 portability issues between our development platforms and yours. We
7988 can handle binary files mailed to us as MIME attachments, uuencoded,
7989 and even BinHex. Alternatively, we may be able to provide an FTP
7990 site you can upload the suspect files to; but mailing them is easier
7993 \b Any other information or data files that might be helpful. If,
7994 for example, the problem involves NASM failing to generate an object
7995 file while TASM can generate an equivalent file without trouble,
7996 then send us \e{both} object files, so we can see what TASM is doing
7997 differently from us.
8000 \A{ndisasm} \i{Ndisasm}
8002 The Netwide Disassembler, NDISASM
8004 \H{ndisintro} Introduction
8007 The Netwide Disassembler is a small companion program to the Netwide
8008 Assembler, NASM. It seemed a shame to have an x86 assembler,
8009 complete with a full instruction table, and not make as much use of
8010 it as possible, so here's a disassembler which shares the
8011 instruction table (and some other bits of code) with NASM.
8013 The Netwide Disassembler does nothing except to produce
8014 disassemblies of \e{binary} source files. NDISASM does not have any
8015 understanding of object file formats, like \c{objdump}, and it will
8016 not understand \c{DOS .EXE} files like \c{debug} will. It just
8020 \H{ndisstart} Getting Started: Installation
8022 See \k{install} for installation instructions. NDISASM, like NASM,
8023 has a \c{man page} which you may want to put somewhere useful, if you
8024 are on a Unix system.
8027 \H{ndisrun} Running NDISASM
8029 To disassemble a file, you will typically use a command of the form
8031 \c ndisasm -b {16|32|64} filename
8033 NDISASM can disassemble 16-, 32- or 64-bit code equally easily,
8034 provided of course that you remember to specify which it is to work
8035 with. If no \i\c{-b} switch is present, NDISASM works in 16-bit mode
8036 by default. The \i\c{-u} switch (for USE32) also invokes 32-bit mode.
8038 Two more command line options are \i\c{-r} which reports the version
8039 number of NDISASM you are running, and \i\c{-h} which gives a short
8040 summary of command line options.
8043 \S{ndiscom} COM Files: Specifying an Origin
8045 To disassemble a \c{DOS .COM} file correctly, a disassembler must assume
8046 that the first instruction in the file is loaded at address \c{0x100},
8047 rather than at zero. NDISASM, which assumes by default that any file
8048 you give it is loaded at zero, will therefore need to be informed of
8051 The \i\c{-o} option allows you to declare a different origin for the
8052 file you are disassembling. Its argument may be expressed in any of
8053 the NASM numeric formats: decimal by default, if it begins with `\c{$}'
8054 or `\c{0x}' or ends in `\c{H}' it's \c{hex}, if it ends in `\c{Q}' it's
8055 \c{octal}, and if it ends in `\c{B}' it's \c{binary}.
8057 Hence, to disassemble a \c{.COM} file:
8059 \c ndisasm -o100h filename.com
8064 \S{ndissync} Code Following Data: Synchronisation
8066 Suppose you are disassembling a file which contains some data which
8067 isn't machine code, and \e{then} contains some machine code. NDISASM
8068 will faithfully plough through the data section, producing machine
8069 instructions wherever it can (although most of them will look
8070 bizarre, and some may have unusual prefixes, e.g. `\c{FS OR AX,0x240A}'),
8071 and generating `DB' instructions ever so often if it's totally stumped.
8072 Then it will reach the code section.
8074 Supposing NDISASM has just finished generating a strange machine
8075 instruction from part of the data section, and its file position is
8076 now one byte \e{before} the beginning of the code section. It's
8077 entirely possible that another spurious instruction will get
8078 generated, starting with the final byte of the data section, and
8079 then the correct first instruction in the code section will not be
8080 seen because the starting point skipped over it. This isn't really
8083 To avoid this, you can specify a `\i\c{synchronisation}' point, or indeed
8084 as many synchronisation points as you like (although NDISASM can
8085 only handle 2147483647 sync points internally). The definition of a sync
8086 point is this: NDISASM guarantees to hit sync points exactly during
8087 disassembly. If it is thinking about generating an instruction which
8088 would cause it to jump over a sync point, it will discard that
8089 instruction and output a `\c{db}' instead. So it \e{will} start
8090 disassembly exactly from the sync point, and so you \e{will} see all
8091 the instructions in your code section.
8093 Sync points are specified using the \i\c{-s} option: they are measured
8094 in terms of the program origin, not the file position. So if you
8095 want to synchronize after 32 bytes of a \c{.COM} file, you would have to
8098 \c ndisasm -o100h -s120h file.com
8102 \c ndisasm -o100h -s20h file.com
8104 As stated above, you can specify multiple sync markers if you need
8105 to, just by repeating the \c{-s} option.
8108 \S{ndisisync} Mixed Code and Data: Automatic (Intelligent) Synchronisation
8111 Suppose you are disassembling the boot sector of a \c{DOS} floppy (maybe
8112 it has a virus, and you need to understand the virus so that you
8113 know what kinds of damage it might have done you). Typically, this
8114 will contain a \c{JMP} instruction, then some data, then the rest of the
8115 code. So there is a very good chance of NDISASM being \e{misaligned}
8116 when the data ends and the code begins. Hence a sync point is
8119 On the other hand, why should you have to specify the sync point
8120 manually? What you'd do in order to find where the sync point would
8121 be, surely, would be to read the \c{JMP} instruction, and then to use
8122 its target address as a sync point. So can NDISASM do that for you?
8124 The answer, of course, is yes: using either of the synonymous
8125 switches \i\c{-a} (for automatic sync) or \i\c{-i} (for intelligent
8126 sync) will enable \c{auto-sync} mode. Auto-sync mode automatically
8127 generates a sync point for any forward-referring PC-relative jump or
8128 call instruction that NDISASM encounters. (Since NDISASM is one-pass,
8129 if it encounters a PC-relative jump whose target has already been
8130 processed, there isn't much it can do about it...)
8132 Only PC-relative jumps are processed, since an absolute jump is
8133 either through a register (in which case NDISASM doesn't know what
8134 the register contains) or involves a segment address (in which case
8135 the target code isn't in the same segment that NDISASM is working
8136 in, and so the sync point can't be placed anywhere useful).
8138 For some kinds of file, this mechanism will automatically put sync
8139 points in all the right places, and save you from having to place
8140 any sync points manually. However, it should be stressed that
8141 auto-sync mode is \e{not} guaranteed to catch all the sync points, and
8142 you may still have to place some manually.
8144 Auto-sync mode doesn't prevent you from declaring manual sync
8145 points: it just adds automatically generated ones to the ones you
8146 provide. It's perfectly feasible to specify \c{-i} \e{and} some \c{-s}
8149 Another caveat with auto-sync mode is that if, by some unpleasant
8150 fluke, something in your data section should disassemble to a
8151 PC-relative call or jump instruction, NDISASM may obediently place a
8152 sync point in a totally random place, for example in the middle of
8153 one of the instructions in your code section. So you may end up with
8154 a wrong disassembly even if you use auto-sync. Again, there isn't
8155 much I can do about this. If you have problems, you'll have to use
8156 manual sync points, or use the \c{-k} option (documented below) to
8157 suppress disassembly of the data area.
8160 \S{ndisother} Other Options
8162 The \i\c{-e} option skips a header on the file, by ignoring the first N
8163 bytes. This means that the header is \e{not} counted towards the
8164 disassembly offset: if you give \c{-e10 -o10}, disassembly will start
8165 at byte 10 in the file, and this will be given offset 10, not 20.
8167 The \i\c{-k} option is provided with two comma-separated numeric
8168 arguments, the first of which is an assembly offset and the second
8169 is a number of bytes to skip. This \e{will} count the skipped bytes
8170 towards the assembly offset: its use is to suppress disassembly of a
8171 data section which wouldn't contain anything you wanted to see
8175 \H{ndisbugs} Bugs and Improvements
8177 There are no known bugs. However, any you find, with patches if
8178 possible, should be sent to
8179 \W{mailto:nasm-bugs@lists.sourceforge.net}\c{nasm-bugs@lists.sourceforge.net}, or to the
8181 \W{http://www.nasm.us/}\c{http://www.nasm.us/}
8182 and we'll try to fix them. Feel free to send contributions and
8183 new features as well.
8185 \A{inslist} \i{Instruction List}
8187 \H{inslistintro} Introduction
8189 The following sections show the instructions which NASM currently supports. For each
8190 instruction, there is a separate entry for each supported addressing mode. The third
8191 column shows the processor type in which the instruction was introduced and,
8192 when appropriate, one or more usage flags.
8196 \A{changelog} \i{NASM Version History}