IBM Personal Computer Assembly
Language Tutorial

Joshua Auerbach
Yale University
Yale Computer Center
175 Whitney Avenue
P. O. Box 2112
New Haven, Connecticut 06520

Installation Code YU

Integrated Personal Computers Project
Communications Group
Communications and Data Base Division

Session C316

This talk is for people who are just getting started with the PC MACRO
Assembler. Maybe you are just contemplating doing some coding in
assembler, maybe you have tried it with mixed success. If you are here to
get aimed in the right direction, to get off to a good start with the
assembler, then you have come for the right reason. I can't promise you'll
get what you want, but I'll do my best.

On the other hand, if you have already turned out some working assembler
code, then this talk is likely to be on the elementary side for you. If
you want to review a few basics and have no where else pressing to go, then
by all means stay.

Why Learn Assembler?

The reasons for LEARNING assembler are not the same as the reasons for
USING it in a particular application. But, we have to start with some of
the reasons for using it and then I think the reasons for learning it will
become clear.

First, let's dispose of a bad reason for using it. Don't use it just
because you think it is going to execute faster. A particular sequence of
ordinary bread-and-butter computations written in PASCAL, C, FORTRAN, or
compiled BASIC can do the job just about as fast as the same algorithm
coded in assembler. Of course, interpretive BASIC is slower, but if you
have a BASIC application which runs too slow you probably want to try com-
piling it before you think too much about translating parts of it to
another language.

On the other hand, high level languages do tend to isolate you from the
machine. That is both their strength and their weakness. Usually, when
implemented on a micro, a high level language provides an escape mechanism
to the underlying operating system or to the bare machine. So, for
example, BASIC has its PEEK and POKE. But, the route to the bare machine
is often a circuitous one, leading to tricky programming which is hard to

For those of us working on PC's connected to SHARE-class mainframes, we are
generally concerned with three interfaces: the keyboard, the screen, and
the communication line or lines. All three of these entities raise machine
dependent issues which are imperfectly addressed by the underlying operat-
ing system or by high level languages.

Sometimes, the system or the language does too little for you. For
example, with the asynch adapter, the system provides no interrupt handler,
no buffer, and no flow control. The application is stuck with the respon-
sibility for monitoring that port and not missing any characters, then
deciding what to do with all errors. BASIC does a reasonable job on some
of this, but that is only BASIC. Most other languages do less.

Sometimes, the system may do too much for you. System support for the key-
board is an example. At the hardware level, all 83 keys on the keyboard
send unique codes when they are pressed, held down, and released. But,
someone has decided that certain keys, like Num Lock and Scroll Lock are
going to do certain things before the application even sees them and can't
therefore be used as ordinary keys.

Sometimes, the system does about the right amount of stuff but does it less
efficiently then it should. System support for the screen is in this
class. If you use only the official interface to the screen you sometimes
slow your application down unacceptably. I said before, don't use assem-
bler just to speed things up, but there I was talking about mainline code,
which generally can't be speeded up much by assembler coding. A critical
system interface is a different matter: sometimes we may have to use
assembler to bypass a hopelessly inefficient implementation. We don't want
to do this if we can avoid it, but sometimes we can't.

Assembly language code can overcome these deficiencies. In some cases, you
can also overcome these deficiencies by judicious use of the escape valves
which your high level language provides. In BASIC, you can PEEK and POKE
and INP and OUT your way around a great many issues. In many other lan-
guages you can issue system calls and interrupts and usually manage, one
way or other, to modify system memory. Writing handlers to take real-time
hardware interrupts from the keyboard or asynch port, though, is still
going to be a problem in most languages. Some languages claim to let you
do it but I have yet to see an acceptably clean implementation done that

The real reason while assembler is better than "tricky POKEs" for writing
machine-dependent code, though, is the same reason why PASCAL is better
than assembler for writing a payroll package: it is easier to maintain.

Let the high level language do what it does best, but recognize that there
are some things which are best done in assembler code. The assembler,
unlike the tricky POKE, can make judicious use of equates, macros, labels,
and appropriately placed comments to show what is really going on in this
machine-dependent realm where it thrives.

So, there are times when it becomes appropriate to write in assembler; giv-
en that, if you are a responsible programmer or manager, you will want to
be "assembler-literate" so you can decide when assembler code should be

What do I mean by "assembler-literate?" I don't just mean understanding
the 8086 architecture; I think, even if you don't write much assembler code
yourself, you ought to understand the actual process of turning out assem-
bler code and the various ways to incorporate it into an application. You
ought to be able to tell good assembler code from bad, and appropriate
assembler code from inappropriate.

Steps to becoming ASSEMBLER-LITERATE

1. Learn the 8086 architecture and most of the instruction set. Learn
what you need to know and ignore what you don't. Reading: The 8086
Primer by Stephen Morse, published by Hayden. You need to read only
two chapters, the one on machine organization and the one on the
instruction set.

2. Learn about a few simple DOS function calls. Know what services the
operating system provides. If appropriate, learn a little about other
systems too. It will aid portability later on. Reading: appendices D
and E of the PC DOS manual.

3. Learn enough about the MACRO assembler and the LINKer to write some
simple things that really work. Here, too, the main thing is figuring
out what you don't need to know. Whatever you do, don't study the sam-
ple programs distributed with the assembler unless you have nothing

4. At the same time as you are learning the assembler itself, you will
need to learn a few tools and concepts to properly combine your assem-
bler code with the other things you do. If you plan to call assembler
subroutines from a high level language, you will need to study the
interface notes provided in your language manual. Usually, this forms
an appendix of some sort. If you plan to package your assembler rou-
tines as .COM programs you will need to learn to do this. You should
also learn to use DEBUG.

5. Read the Technical Reference, but very selectively. The most important
things to know are the header comments in the BIOS listing. Next, you
will want to learn about the RS 232 port and maybe about the video

Notice that the key thing in all five phases is being selective. It is
easy to conclude that there is too much to learn unless you can throw away
what you don't need. Most of the rest of this talk is going to deal with
this very important question of what you need and don't need to learn in
each phase. In some cases, I will have to leave you to do almost all of
the learning, in others, I will teach a few salient points, enough, I hope,
to get you started. I hope you understand that all I can do in an hour is
get you started on the way.

Phase 1: Learn the architecture and instruction set

The Morse book might seem like a lot of book to buy for just two really
important chapters; other books devote a lot more space to the instruction
set and give you a big beautiful reference page on each instruction. And,
some of the other things in the Morse book, although interesting, really
aren't very vital and are covered too sketchily to be of any real help.
The reason I like the Morse book is that you can just read it; it has a
very conversational style, it is very lucid, it tells you what you really
need to know, and a little bit more which is by way of background; because
nothing really gets belabored to much, you can gracefully forget the things
you don't use. And, I very much recommend READING Morse rather than study-
ing it. Get the big picture at this point.

Now, you want to concentrate on those things which are worth fixing in mem-
ory. After you read Morse, you should relate what you have learned to this

1. You want to fix in your mind the idea of the four segment registers
CODE, DATA, STACK, and EXTRA. This part is pretty easy to grasp. The
8086 and the 8088 use 20 bit addresses for memory, meaning that they
can address up to 1 megabyte of memory. But, the registers and the
address fields in all the instructions are no more that 16 bits long.
So, how to address all of that memory? Their solution is to put
together two 16 bit quantities like this:

calculation SSSS0 ---- value in the relevant segment register SHL 4
depicted in AAAA ---- apparent address from register or instruction
hexadecimal --------
RRRRR ---- real address placed on address bus

In other words, any time memory is accessed, your program will supply a
sixteen bit address. Another sixteen bit address is acquired from a
segment register, left shifted four bits (one nibble) and added to it
to form the real address. You can control the values in the segment
registers and thus access any part of memory you want. But the segment
registers are specialized: one for code, one for most data accesses,
one for the stack (which we'll mention again) and one "extra" one for
additional data accesses.

Most people, when they first learn about this addressing scheme become
obsessed with converting everything to real 20 bit addresses. After a
while, though, you get use to thinking in segment/offset form. You
tend to get your segment registers set up at the beginning of the pro-
gram, change them as little as possible, and think just in terms of
symbolic locations in your program, as with any assembly language.

MOV DS,AX ;Set value of Data segment
ASSUME DS:DATASEG ;Tell assembler DS is usable
MOV AX,PLACE ;Access storage symbolically by 16 bit address

In the above example, the assembler knows that no special issues are
involved because the machine generally uses the DS register to complete
a normal data reference.

If you had used ES instead of DS in the above example, the assembler
would have known what to do, also. In front of the MOV instruction
which accessed the location PLACE, it would have placed the ES segment
prefix. This would tell the machine that ES should be used, instead of
DS, to complete the address.

Some conventions make it especially easy to forget about segment regis-
ters. For example, any program of the COM type gets control with all
four segment registers containing the same value. This program exe-
cutes in a simplified 64K address space. You can go outside this
address space if you want but you don't have to.

2. You will want to learn what other registers are available and learn
their personalities:

AX and DX are general purpose registers. They become special only
when accessing machine and system interfaces.

CX is a general purpose register which is slightly specialized for

BX is a general purpose register which is slightly specialized for
forming base-displacement addresses.

AX-DX can be divided in half, forming AH, AL, BH, BL, CH, CL, DH,

SI and DI are strictly 16 bit. They can be used to form indexed
addresses (like BX) and they are also used to point to strings.

SP is hardly ever manipulated. It is there to provide a stack.

BP is a manipulable cousin to SP. Use it to access data which has
been pushed onto the stack.

Most sixteen bit operations are legal (even if unusual) when per-
formed in SI, DI, SP, or BP.

3. You will want to learn the classifications of operations available
WITHOUT getting hung up in the details of how 8086 opcodes are con-

8086 opcodes are complex. Fortunately, the assembler opcodes used to
assemble them are simple. When you read a book like Morse, you will
learn some things which are worth knowing but NOT worth dwelling on.

a. 8086 and 8088 instructions can be broken up into subfields and bits
with names like R/M, MOD, S and W. These parts of the instruction
modify the basic operation in such ways as whether it is 8 bit or
16 bit, if 16 bit, whether all 16 bits of the data are given,
whether the instruction is register to register, register to
memory, or memory to register, for operands which are registers,
which register, for operands which are memory, what base and index
registers should be used in finding the data.

b. Also, some instructions are actually represented by several differ-
ent machine opcodes depending on whether they deal with immediate
data or not, or on other issues, and there are some expedited forms
which assume that one of the arguments is the most commonly used
operand, like AX in the case of arithmetic.

There is no point in memorizing any of this detail; just distill the
bottom line, which is, what kinds of operand combinations EXIST in the
instruction set and what kinds don't. If you ask the assembler to ADD
two things and the two things are things for which there is a legal ADD
instruction somewhere in the instruction set, the assembler will find
the right instruction and fill in all the modifier fields for you.

I guess if you memorized all the opcode construction rules you might
have a crack at being able to disassemble hex dumps by eye, like you
may have learned to do somewhat with 370 assembler. I submit to you
that this feat, if ever mastered by anyone, would be in the same class
as playing the "Minute Waltz" in a minute; a curiosity only.

Here is the basic matrix you should remember:

Two operands: One operand:
R <-- M R
M <-- R M
R <-- R S *
R|M <-- I
R|M <-- S *
S <-- R|M *

* -- data moving instructions (MOV, PUSH, POP) only
S -- segment register (CS, DS, ES, SS)
R -- ordinary register (AX, BX, CX, DX, SI, DI, BP, SP,
M -- one of the following
pure address
any of the above indexed by SI
any of the first three indexed by DI

4. Of course, you want to learn the operations themselves. As I've sug-
gested, you want to learn the op codes as the assembler presents them,
not as the CPU machine language presents them. So, even though there
are many MOV op codes you don't need to learn them. Basically, here is
the instruction set:

a. Ordinary two operand instructions. These instructions perform an
operation and leave the result in place of one of the operands.
They are

1) ADD and ADC -- addition, with or without including a carry from
a previous addition

2) SUB and SBB -- subtraction, with or without including a borrow
from a previous subtraction

3) CMP -- compare. It is useful to think of this as a subtraction
with the answer being thrown away and neither operand actually

4) AND, OR, XOR -- typical boolean operations

5) TEST -- like an AND, except the answer is thrown away and nei-
ther operand is changed.

6) MOV -- move data from source to target

7) LDS, LES, LEA -- some specialized forms of MOV with side

b. Ordinary one operand instructions. These can take any of the oper-
and forms described above. Usually, the perform the operation and
leave the result in the stated place:

1) INC -- increment contents

2) DEC -- decrement contents

3) NEG -- twos complement

4) NOT -- ones complement

5) PUSH -- value goes on stack (operand location itself unchanged)

6) POP -- value taken from stack, replaces current value

c. Now you touch on some instructions which do not follow the general
operand rules but which require the use of certain registers. The
important ones are

1) The multiply and divide instructions

2) The "adjust" instructions which help in performing arithmetic
on ASCII or packed decimal data

3) The shift and rotate instructions. These have a restriction on
the second operand: it must either be the immediate value 1 or
the contents of the CL register.

4) IN and OUT which send or receive data from one of the 1024
hardware ports.

5) CBW and CWD -- convert byte to word or word to doubleword by
sign extension

d. Flow of control instructions. These deserve study in themselves
and we will discuss them a little more. They include

1) CALL, RET -- call and return

2) INT, IRET -- interrupt and return-from-interrupt

3) JMP -- jump or "branch"

4) LOOP, LOOPNZ, LOOPZ -- special (and useful) instructions which
implement a counted loop similar to the 370 BCT instruction

5) various conditional jump instructions

e. String instructions. These implement a limited storage-to-storage
instruction subset and are quite powerful. All of them have the
property that

1) The source of data is described by the combination DS and SI.

2) The destination of data is described by the combination ES and

3) As part of the operation, the SI and/or DI register(s) is(are)
incremented or decremented so the operation can be repeated.

They include

1) CMPSB/CMPSW -- compare byte or word

2) LODSB/LODSW -- load byte or word into AL or AX

3) STOSB/STOSW -- store byte or word from AL or AX

4) MOVSB/MOVSW -- move byte or word

5) SCASB/SCASW -- compare byte or word with contents of AL or AX

6) REP/REPE/REPNE -- a prefix which can be combined with any of
the above instructions to make them execute repeatedly across a
string of data whose length is held in CX.

f. Flag instructions: CLI, STI, CLD, STD, CLC, STC. These can set or
clear the interrupt (enabled) direction (for string operations) or
carry flags.

The addressing summary and the instruction summary given above masks a
lot of annoying little exceptions. For example, you can't POP CS, and
although the R <-- M form of LES is legal, the M <-- R form isn't etc.
etc. My advice is

a. Go for the general rules

b. Don't try to memorize the exceptions

c. Rely on common sense and the assembler to teach you about
exceptions over time. A lot of the exceptions cover things you
wouldn't want to do anyway.

5. A few instructions are rich enough and useful enough to warrent careful
study. Here are a few final study guidelines:

a. It is well worth the time learning to use the string instruction
set effectively. Among the most useful are

REP MOVSB ;moves a string
REP STOSB ;initializes memory
REPNE SCASB ;look up occurance of character in string
REPE CMPSB ;compare two strings

b. Similarly, if you have never written for a stack machine before,
you will need to exercise PUSH and POP and get very comfortable
with them because they are going to be good friends. If you are
used to the 370, with lots of general purpose registers, you may
find yourself feeling cramped at first, with many fewer registers
and many instructions having register restrictions. But, you have
a hidden ally: you need a register and you don't want to throw
away what's in it? Just PUSH it, and when you are done, POP it
back. This can lead to abuse. Never have more than two
"expedient" PUSHes in effect and never leave something PUSHed
across a major header comment or for more than 15 instructions or
so. An exception is the saving and restoring of registers at
entrance to and exit from a subroutine; here, if the subroutine is
long, you should probably PUSH everything which the caller may need
saved, whether you will use the register or not, and POP it in
reverse order at the end.

Be aware that CALL and INT push return address information on the
stack and RET and IRET pop it off. It is a good idea to become
familiar with the structure of the stack.

c. In practice, to invoke system services you will use the INT
instruction. It is quite possible to use this instruction effec-
tively in a cookbook fashion without knowing precisely how it

d. The transfer of control instructions (CALL, RET, JMP) deserve care-
ful study to avoid confusion. You will learn that these can be
classified as follows:

1) all three have the capability of being either NEAR (CS register
unchanged) or FAR (CS register changed)

2) JMPs and CALLs can be DIRECT (target is assembled into instruc-
tion) or INDIRECT (target fetched from memory or register)

3) if NEAR and DIRECT, a JMP can be SHORT (less than 128 bytes
away) or LONG

In general, the third issue is not worth worrying about. On a for-
ward jump which is clearly VERY short, you can tell the assembler
it is short and save one byte of code:


On a backward jump, the assembler can figure it out for you. On a
forward jump of dubious length, let the assembler default to a LONG
form; at worst you waste one byte.

Also leave the assembler to worry about how the target address is
to be represented, in absolute form or relative form.

e. The conditional jump set is rather confusing when studied apart
from the assembler, but you do need to get a feeling for it. The
interactions of the sign, carry, and overflow flags can get your
mind stuttering pretty fast if you worry about it too much. What
is boils down to, though, is

JZ means what it says
JNZ means what it says
JG reater this means "if the SIGNED difference is positive"
JA bove this means "if the UNSIGNED difference is positive"
JL ess this means "if the SIGNED difference is negative"
JB elow this means "if the UNSIGNED difference is negative"
JC arry assembles the same as JB; it's an aesthetic choice

You should understand that all conditional jumps are inherently
DIRECT, NEAR, and "short"; the "short" part means that they can't
go more than 128 bytes in either direction. Again, this is some-
thing you could easily imagine to be more of a problem than it is.
I follow this simple approach:

1) When taking an abnormal exit from a block of code, I always use
an unconditional jump. Who knows how far you are going to end
up jumping by the time the program is finished. For example, I
wouldn't code this:

TEST AL,IDIBIT ;Is the idiot bit on?
JNZ OYVEY ;Yes. Go to general cleanup

Rather, I would probably code this:

TEST AL,IDIBIT ;Is the idiot bit on?
JZ NOIDIOCY ;No. I am saved.
JMP OYVEY ;Yes. What can we say...

The latter, of course, is a jump around a jump. Some would say
it is evil, but I submit it is hard to avoid in this language.

2) Otherwise, within a block of code, I use conditional jumps
freely. If the block eventually grows so long that the assem-
bler starts complaining that my conditional jumps are too long

a) consider reorganizing the block but

b) also consider changing some conditional jumps to their
opposite and use the "jump around a jump" approach as shown

Enough about specific instructions!

6. Finally, in order to use the assembler effectively, you need to know
the default rules for which segment registers are used to complete
addresses in which situations.

a. CS is used to complete an address which is the target of a NEAR
DIRECT jump. On an NEAR INDIRECT jump, DS is used to fetch the
address from memory but then CS is used to complete the address
thus fetched. On FAR jumps, of course, CS is itself altered. The
instruction counter is always implicitly pointing in the code seg-

b. SS is used to complete an address if BP is used in its formation.
Otherwise, DS is always used to complete a data address.

c. On the string instructions, the target is always formed from ES and
DI. The source is normally formed from DS and SI. If there is a
segment prefix, it overrides the source not the target.

Learning about DOS

I think the best way to learn about DOS internals is to read the technical
appendices in the manual. These are not as complete as we might wish, but
they really aren't bad; I certainly have learned a lot from them. What you
don't learn from them you might eventually learn via judicious disassembly
of parts of DOS, but that shouldn't really be necessary.

From reading the technical appendices, you learn that interrupts 20H
through 27H are used to communicate with DOS. Mostly, you will use inter-
rupt 21H, the DOS function manager.

The function manager implements a great many services. You request the
individual services by means of a function code in the AH register. For
example, by putting a nine in the AH register and issuing interrupt 21H you
tell DOS to print a message on the console screen.

Usually, but by no means always, the DX register is used to pass data for
the service being requested. For example, on the print message service
just mentioned, you would put the 16 bit address of the message in the DX
register. The DS register is also implicitly part of this argument, in
keeping with the universal segmentation rules.

In understanding DOS functions, it is useful to understand some history and
also some of the philosophy of MS-DOS with regard to portability. General-
ly, you will find, once you read the technical information on DOS and also
the IBM technical reference, you will know more than one way to do almost
anything. Which is best? For example, to do asynch adapter I/O, you can
use the DOS calls (pretty incomplete), you can use BIOS, or you can go
directly to the hardware. The same thing is true for most of the other
primitive I/O (keyboard or screen) although DOS is more likely to give you
added value in these areas. When it comes to file I/O, DOS itself offers
more than one interface. For example, there are four calls which read data
from a file.

The way to decide rationally among these alternatives is by understanding
the tradeoffs of functionality versus portability. Three kinds of porta-
bility need to be considered: machine portability, operating system porta-
bility (for example, the ability to assemble and run code under CP/M 86)
and DOS version portability (the ability for a program to run under older
versions of DOS>.

Most of the functions originally offered in DOS 1.0 were direct descendents
of CP/M functions; there is even a compatibility interface so that programs
which have been translated instruction for instruction from 8080 assembler
to 8086 assembler might have a reasonable chance of running if they use
only the core CP/M function set. Among the most generally useful in this
original compatibility set are

09 -- print a full message on the screen
0A -- get a console input line with full DOS editing
0F -- open a file
10 -- close a file (really needed only when writing)
11 -- find first file matching a pattern
12 -- find next file matching a pattern
13 -- erase a file
16 -- create a file
17 -- rename a file
1A -- set disk transfer address

The next set provide no function above what you can get with BIOS calls or
more specialized DOS calls. However, they are preferable to BIOS calls
when portability is an issue.

00 -- terminate execution
01 -- read keyboard character
02 -- write screen character
03 -- read COM port character
04 -- write COM port character
05 -- print a character
06 -- read keyboard or write screen with no editing

The standard file I/O calls are inferior to the specialized DOS calls but
have the advantage of making the program easier to port to CP/M style sys-
tems. Thus they are worth mentioning:

14 -- sequential read from file
15 -- sequential write to file
21 -- random read from file
22 -- random write to file
23 -- determine file size
24 -- set random record

In addition to the CP/M compatible services, DOS also offers some special-
ized services which have been available in all releases of DOS. These

27 -- multi-record random read.
28 -- multi-record random write.
29 -- parse filename
2A-2D -- get and set date and time

All of the calls mentioned above which have anything to do with files make
use of a data area called the "FILE CONTROL BLOCK" (FCB). The FCB is any-
where from 33 to 37 bytes long depending on how it is used. You are
responsible for creating an FCB and filling in the first 12 bytes, which
contain a drive code, a file name, and an extension.

When you open the FCB, the system fills in the next 20 bytes, which
includes a logical record length. The initial lrecl is always 128 bytes,
to achieve CP/M compatibility. The system also provides other useful
information such as the file size.

After you have opened the FCB, you can change the logical record length.
If you do this, your program is no longer CP/M compatible, but that doesn't
make it a bad thing to do. DOS documentation suggests you use a logical
record length of one for maximum flexibility. This is usually a good

To perform actual I/O to a file, you eventually need to fill in byte 33 or
possibly bytes 34-37 of the FCB. Here you supply information about the
record you are interested in reading or writing. For the most part, this
part of the interface is compatible with CP/M.

In general, you do not need to (and should not) modify other parts of the

The FCB is pretty well described in appendix E of the DOS manual.

Beginning with DOS 2.0, there is a whole new system of calls for managing
files which don't require that you build an FCB at all. These calls are
quite incompatible with CP/M and also mean that your program cannot run
under older releases of DOS. However, these calls are very nice and easy
to use. They have these characteristics

1. To open, create, delete, or rename a file, you need only a character
string representing its name.

2. The open and create calls return a 16 bit value which is simply placed
in the BX register on subsequent calls to refer to the file.

3. There is not a separate call required to specify the data buffer.

4. Any number of bytes can be transfered on a single call; no data area
must be manipulated to do this.

The "new" DOS calls also include comprehensive functions to manipulate the
new chained directory structure and to allocate and free memory.

Learning the assembler

It is my feeling that many people can teach themselves to use the assembler
by reading the MACRO Assembler manual if

1. You have read and understood a book like Morse and thus have a feeling
for the instruction set

2. You know something about DOS services and so can communicate with the
keyboard and screen and do something marginally useful with files. In
the absence of this kind of knowledge, you can't write meaningful prac-
tice programs and so will not progress.

3. You have access to some good examples (the ones supplied with the
assembler are not good, in my opinion. I will try to supply you with
some more relevant ones.

4. You ignore the things which are most confusing and least useful. Some
of the most confusing aspects of the assembler include the facilities
combining segments. But, you can avoid using all but the simplest of
these facilities in many cases, even while writing quite substantial

5. The easiest kind of assembler program to write is a COM program. They
might seem harder, at first, then EXE programs because there is an
extra step involved in creating the executable file, but COM programs
are structurally very much simpler.

At this point, it is necessary to talk about COM programs and EXE programs.
As you probably know, DOS supports two kinds of executable files. EXE pro-
grams are much more general, can contain many segments, and are generally
built by compilers and sometimes by the assembler. If you follow the lead
given by the samples distributed with the assembler, you will end up with
EXE programs. A COM program, in contrast, always contains just one
segment, and receives control with all four segment registers containing
the same value. A COM program, thus, executes in a simplified environment,
a 64K address space. You can go outside this address space simply by tem-
porarily changing one segment register, but you don't have to, and that is
the thing which makes COM programs nice and simple. Let's look at a very
simple one.

The classic text on writing programs for the C language says that the first
thing you should write is a program which says


when invoked. What's sauce for C is sauce for assembler, so let's start
with a HELLO program of our own. My first presentation of this will be
bare bones, not stylistically complete, but just an illustration of what an
assembler program absolutely has to have:

HELLO SEGMENT ;Set up HELLO code and data section
ASSUME CS:HELLO,DS:HELLO ;Tell assembler about conditions at entry
ORG 100H ;A .COM program begins with 100H byte prefix
MAIN: JMP BEGIN ;Control must start here
MSG DB 'Hello, world.$' ;But it is generally useful to put data first
BEGIN: MOV DX,OFFSET MSG ;Let DX --> message.
MOV AH,9 ;Set DOS function code for printing a message
INT 21H ;Invoke DOS
RET ;Return to system
HELLO ENDS ;End of code and data section
END MAIN ;Terminate assembler and specify entry point

First, let's attend to some obvious points. The macro assembler uses the
general form

name opcode operands

Unlike the 370 assembler, though, comments are NOT set off from operands by
blanks. The syntax uses blanks as delimiters within the operand field (see
line 6 of the example) and so all comments must be set off by semi-colons.

Line comments are frequently set off with a semi-colon in column 1. I use
this approach for block comments too, although there is a COMMENT statement
which can be used to introduce a block comment.

Being an old 370 type, I like to see assembler code in upper case, although
my comments are mixed case. Actually, the assembler is quite happy with
mixed case anywhere.

As with any assembler, the core of the opcode set consists of opcodes which
generate machine instructions but there are also opcodes which generate
data and ones which function as instructions to the assembler itself, some-
times called pseudo-ops. In the example, there are five lines which gener-
ate machine code (JMP, MOV, MOV, INT, RET), one line which generates data
(DB) and five pseudo-ops (SEGMENT, ASSUME, ORG, ENDS, and END).

We will discuss all of them.

Now, about labels. You will see that some labels in the example end in a
colon and some don't. This is just a bit confusing at first, but no real
mystery. If a label is attached to a piece of code (as opposed to data),
then the assembler needs to know what to do when you JMP to or CALL that
label. By convention, if the label ends in a colon, the assembler will use
the NEAR form of JMP or CALL. If the label does not end in a colon, it
will use the FAR form. In practice, you will always use the colon on any
label you are jumping to inside your program because such jumps are always
NEAR; there is no reason to use a FAR jump within a single code section. I
mention this, though, because leaving off the colon isn't usually trapped
as a syntax error, it will generally cause something more abstruse to go

On the other hand, a label attached to a piece of data or a pseudo-op never
ends in a colon.

Machine instructions will generally take zero, one or two operands. Where
there are two operands, the one which receives the result goes on the left
as in 370 assembler.

I tried to explain this before, now maybe it will be even clearer: there
are many more 8086 machine opcodes then there are assembler opcodes to rep-
resent them. For example, there are five kinds of JMP, four kinds of CALL,
two kinds of RET, and at least five kinds of MOV depending on how you count
them. The macro assembler makes a lot of decisions for you based on the
form taken by the operands or on attributes assigned to symbols elsewhere
in your program. In the example above, the assembler will generate the
NEAR DIRECT form of JMP because the target label BEGIN labels a piece of
code instead of a piece of data (this makes the JMP DIRECT) and ends in a
colon (this makes the JMP NEAR). The assembler will generate the immediate
forms of MOV because the form OFFSET MSG refers to immediate data and
because 9 is a constant. The assembler will generate the NEAR form of RET
because that is the default and you have not told it otherwise.

The DB (define byte) pseudo-op is an easy one: it is used to put one or
more bytes of data into storage. There is also a DW (define word)
pseudo-op and a DD (define doubleword) pseudo-op; in the PC MACRO assem-
bler, the fact that a label refers to a byte of storage, a word of storage,
or a doubleword of storage can be very significant in ways which we will
see presently.

About that OFFSET operator, I guess this is the best way to make the point
about how the assembler decides what instruction to assemble: an analogy
with 370 assembler:

PLACE DC ......

In 370 assembler, the first instruction puts the address of label PLACE in
register 1, the second instruction puts the contents of storage at label
PLACE in register 1. Notice that two different opcodes are used. In the
PC assembler, the analogous instructions would be

PLACE DW ......

If PLACE is the label of a word of storage, then the second instruction
will be understood as a desire to fetch that data into DX. If X is a
label, then "OFFSET X" means "the ordinary number which represents X's off-
set from the start of the segment." And, if the assembler sees an ordinary
number, as opposed to a label, it uses the instruction which is equivalent
to LA.

If PLACE were the label of a DB pseudo-op, instead of a DW, then


would be illegal. The assembler worries about length attributes of its

Next, numbers and constants in general. The assembler's default radix is
decimal. You can change this, but I don't recommend it. If you want to
represent numbers in other forms of notation such as hex or bit, you gener-
ally use a trailing letter. For example,

is hexidecimal 21,
is the eight bit binary number pictured.

The next elements we should point to are the SEGMENT...ENDS pair and the
END instruction. Every assembler program has to have these elements.

SEGMENT tells the assembler you are starting a section of contiguous mate-
rial (code and/or data). The symmetrically named ENDS statement tells the
assembler you are finished with a section of contiguous material. I wish
they didn't use the word SEGMENT in this context. To me, a "segment" is a
hardware construct: it is the 64K of real storage which becomes address-
able by virtue of having a particular value in a segment register. Now, it
is true that the "segments" you make with the assembler often correspond to
real hardware "segments" at execution time. But, if you look at things
like the GROUP and CLASS options supported by the linker, you will discover
that this correspondence is by no means exact. So, at risk of maybe con-
fusing you even more, I am going to use the more informal term "section" to
refer to the area set off by means of the SEGMENT and ENDS instructions.

The sections delimited by SEGMENT...ENDS pairs are really a lot like CSECTs
and DSECTs in the 370 world.

I strongly recommend that you be selective in your study of the SEGMENT
pseudo-op as described in the manual. Let me just touch on it here.

name SEGMENT AT nnn

Basically, you can get away with just the three forms given above. The
first form is what you use when you are writing a single section of assem-
bler code which will not be combined with other pieces of code at link
time. The second form says that this assembly only contains part of the
section; other parts might be assembled separately and combined later by
the linker.

I have found that one can construct reasonably large modular applications
in assembler by simply making every assembly use the same segment name and
declaring the name to be PUBLIC each time. If you read the assembler and
linker documentation, you will also be bombarded by information about more
complex options such as the GROUP statement and the use of other "combine
types" and "classes." I don't recommend getting into any of that. I will
talk more about the linker and modular construction of programs a little
later. The assembler manual also implies that a STACK segment is required.
This is not really true. There are numerous ways to assure that you have a
valid stack at execution time.

Of course, if you plan to write applications in assembler which are more
than 64K in size, you will need more than what I have told you; but who is
really going to do that? Any application that large is likely to be coded
in a higher level language.

The third form of the SEGMENT statement makes the delineated section into
something like a "DSECT;" that is, it doesn't generate any code, it just
describes what is present somewhere already in the computer's memory.
Sometimes the AT value you give is meaningful. For example, the BIOS work
area is located at location 40 hex. So, you might see

EQUIP DB ? ;Location of equipment flags, first byte

in a program which was interested in mucking around in the BIOS work area.

At other times, the AT value you give may be arbitrary, as when you are
mapping a repeated control block:

PROGPREF SEGMENT AT 0 ;Really a DSECT mapping the program prefix
MEMSIZE DW ? ;Size of available memory

Really, no matter whether the AT value represents truth or fiction, it is
your responsibility, not the assembler's, to get set up a segment register
so that you can really reach the storage in question. So, you can't say


unless you first say something like

MOV AX,BIOSAREA ;BIOSAREA becomes a symbol with value 40H

Enough about SEGMENT. The END statement is simple. It goes at the end of
every assembly. When you are assembling a subroutine, you just say


but when you are assembling the main routine of a program you say

END label

where 'label' is the place where execution is to begin.

Another pseudo-op illustrated in the program is ASSUME. ASSUME is like the
USING statement in 370 assembler. However, ASSUME can ONLY refer to seg-
ment registers. The assembler uses ASSUME information to decide whether to
assemble segment override prefixes and to check that the data you are try-
ing to access is really accessible. In this case, we can reassure the
assembler that both the CS and DS registers will address the section called
HELLO at execution time. Actually, the SS and ES registers will too, but
the assembler never needs to make use of this information.

I guess I have explained everything in the program except that ORG
pseudo-op. ORG means the same thing as it does in many assembly languages.
It tells the assembler to move its location counter to some particular
address. In this case, we have asked the assembler to start assembling
code hex 100 bytes from the start of the section called HELLO instead of at
the very beginning. This simply reflects the way COM programs are loaded.
When a COM program is loaded by the system, the system sets up all four
segment registers to address the same 64K of storage. The first 100 hex
bytes of that storage contains what is called the program prefix; this area
is described in appendix E of the DOS manual. Your COM program physically
begins after this. Execution begins with the first physical byte of your
program; that is why the JMP instruction is there.

Wait a minute, you say, why the JMP instruction at all? Why not put the
data at the end? Well, in a simple program like this I probably could have
gotten away with that. However, I have the habit of putting data first and
would encourage you to do the same because of the way the assembler has of
assembling different instructions depending on the nature of the operand.

Unfortunately, sometimes the different choices of instruction which can
assemble from a single opcode have different lengths. If the assembler has
already seen the data when it gets to the instructions it has a good chance
of reserving the right number of bytes on the first pass. If the data is
at the end, the assembler may not have enough information on the first pass
to reserve the right number of bytes for the instruction. Sometimes the
assembler will complain about this, something like "Forward reference is
illegal" but at other times, it will make some default assumption. On the
second pass, if the assumption turned out to be wrong, it will report what
is called a "Phase error," a very nasty error to track down. So get in the
habit of putting data and equated symbols ahead of code.

OK. Maybe you understand the program now. Let's walk through the steps
involved in making it into a real COM file.

1. The file should be created with the name HELLO.ASM (actually the name
is arbitrary but the extension .ASM is conventional and useful)



(this is just one example of invoking the assembler; it uses the small
assembler ASM, it produces an object file and a listing file with the
same name as the source file. I am not going exhaustively into how to
invoke the assembler, which the manual goes into pretty well. I guess
this is the first time I mentioned that there are really two
assemblers; the small assembler ASM will run in a 64K machine and
doesn't support macros. I used to use it all the time; now that I have
a bigger machine and a lot of macro libraries I use the full function
assembler MASM. You get both when you buy the package).

3. If you issue DIR at this point, you will discover that you have
acquired HELLO.OBJ (the object code resulting from the assembly) and
HELLO.LST (a listing file). I guess I can digress for a second here
concerning the listing file. It contains TAB characters. I have found
there are two good ways to get it printed and one bad way. The bad way
is to use LPT1: as the direct target of the listing file or to try
copying the LST file to LPT1 without first setting the tabs on the
printer. The two good ways are to either

a. direct it to the console and activate the printer with CTRL-PRTSC.
In this case, DOS will expand the tabs for you.

b. direct to LPT1: but first send the right escape sequence to LPT1 to
set the tabs every eight columns. I have found that on some early
serial numbers of the IBM PC printer, tabs don't work quite right,
which forces you to the first option.



(again, there are lots of linker options but this is the simplest. It
takes HELLO.OBJ and makes HELLO.EXE). HELLO.EXE? I thought we were
making a COM program, not an EXE program. Right. HELLO.EXE isn't
really executable; its just that the linker doesn't know about COM pro-
grams. That requires another utility. You don't have this utility if
you are using DOS 1.0; you have it if you are using DOS 1.1 or DOS 2.0.
Oh, by the way, the linker will warn you that you have no stack
segment. Don't worry about it.



This is the final step. It produces the actual program you will exe-
cute. Note that you have to spell out HELLO.COM; for a nominally
rational but actually perverse reason, EXE2BIN uses the default exten-
sion BIN instead of COM for its output file. At this point, you might
want to erase HELLO.EXE; it looks a lot more useful than it is.
Chances are you won't need to recreate HELLO.COM unless you change the
source and then you are going to have to redo the whole thing.



You type hello, that invokes the program, it says


(oops, what did I do wrong....?)

What about subroutines?
What about subroutines?
What about subroutines?
What about subroutines?

I started with a simple COM program because I actually think they are easi-
er to create than subroutines to be called from high level languages, but
maybe its really the latter you are interested in. Here, I think you
should get comfortable with the assembler FIRST with little exercises like
the one above and also another one which I will finish up with.

Next you are ready to look at the interface information for your particular
language. You usually find this in some sort of an appendix. For example,
the BASIC manual has Appendix C on Machine Language Subroutines. The
PASCAL manual buries the information a little more deeply: the interface
to a separately compiled routine can be found in the Chapter on Procedures
and Functions, in a subsection called Internal Calling Conventions.

Each language is slightly different, but here are what I think are some
common issues in subroutine construction.

1. NEAR versus FAR? Most of the time, your language will probably call
your assembler routine as a FAR routine. In this case, you need to
make sure the assembler will generate the right kind of return. You do
this with a PROC...ENDP statement pair. The PROC statement is probably
a good idea for a NEAR routine too even though it is not strictly

FAR linkage: | NEAR linkage:
..... code and data | THENAME PROC NEAR
THENAME ENDP | ..... code and data ....

With FAR linkage, it doesn't really matter what you call the segment.
you must declare the name by which you will be called in a PUBLIC pseu-
do-op and also show that it is a FAR procedure. Only CS will be ini-
tialized to your segment when you are called. Generally, the other
segment registers will continue to point to the caller's segments.

With NEAR linkage, you are executing in the same segment as the caller.
Therefore, you must give the segment a specific name as instructed by
the language manual. However, you may be able to count on all segment
registers pointing to your own segment (sometimes the situation can be
more complicated but I cannot really go into all of the details). You
should be aware that the code you write will not be the only thing in
the segment and will be physically relocated within the segment by the
linker. However, all OFFSET references will be relocated and will be
correct at execution time.

2. Parameters passed on the stack. Usually, high level languages pass
parameters to subroutines by pushing words onto the stack prior to
calling you. What may differ from language to language is the nature
of what is pushed (OFFSET only or OFFSET and SEGMENT) and the order in
which it is pushed (left to right, right to left within the CALL state-
ment). However, you will need to study the examples to figure out how
to retrieve the parameters from the stack. A useful fact to exploit is
the fact that a reference involving the BP register defaults to a ref-
erence to the stack segment. So, the following strategy can work:

DW 3 DUP(?) ;Saved BP and return address
PUSH BP ;save BP register
MOV BP,SP ;Use BP to address stack
MOV ...,[BP].ARG2 ;retrieve second argument

This example uses something called a structure, which is only available
in the large assembler; furthermore, it uses it without allocating it,
which is not a well-documented option. However, I find the above
approach generally pleasing. The STRUC is like a DSECT in that it
establishes labels as being offset a certain distance from an arbitrary
point; these labels are then used in the body of code by beginning them
with a period; the construction ".ARG2" means, basically, " +

What you are doing here is using BP to address the stack, accounting
for the word where you saved the caller's BP and also for the two words
which were pushed by the CALL instruction.

3. How big is the stack? BASIC only gives you an eight word stack to play
with. On the other hand, it doesn't require you to save any registers
except the segment registers. Other languages give you a liberal
stack, which makes things a lot easier. If you have to create a new
stack segment for yourself, the easiest thing is to place the stack at
the end of your program and:

CLI ;suppress interrupts while changing the stack
MOV SSAVE,SS ;save old SS in local storage (old SP
; already saved in BP)
MOV SP,CS ;switch
MOV SS,SP ;the
STI ;(maybe)

Later, you can reverse these steps before returning to the caller. At
the end of your program, you place the stack itself:

DW 128 DUP(?) ;stack of 128 words (liberal)

4. Make sure you save and restore those registers required by the caller.

5. Be sure to get the right kind of addressibility. In the FAR call exam-
ple, only CS addresses your segment. If you are careful with your
ASSUME statements the assembler will keep track of this fact and gener-
ate CS prefixes when you make data references; however, you might want
to do something like

MOV AX,CS ;get current segment address

Be sure you keep your ASSUMEs in synch with reality.

Learning about BIOS and the hardware

You can't do everything with DOS calls. You may need to learn something
about the BIOS and about the hardware itself. In this, the Technical Ref-
erence is a very good thing to look at.

The first thing you look at in the Technical Reference, unless you are
really determined to master the whole ball of wax, is the BIOS listings
presented in Appendix A. Glory be: here is the whole 8K of ROM which deals
with low level hardware support layed out with comments and everything.

In fact, if you are just interested in learning what BIOS can do for you,
you just need to read the header comments at the beginning of each section
of the listing.

BIOS services are invoked by means of the INT instruction; the BIOS occu-
pies interrupts 10H through 1FH and also interrupt 5H; actually, of these
seventeen interrupts, five are used for user exit points or data pointers,
leaving twelve actual services.

In most cases, a service deals with a particular hardware interface; for
example, BIOS interrupt 10H deals with the screen. As with DOS function
calls, many BIOS services can be passed a function code in the AH register
and possible other arguments.

I am not going to summarize the most useful BIOS features here; you will
see some examples in the next sample program we will look at.

The other thing you might want to get into with the Tech reference is the
description of some hardware options, particularly the asynch adapter,
which are not well supported in the BIOS. The writeup on the asynch adapt-
er is pretty complete.

Actually, the Tech reference itself is pretty complete and very nice as far
as it goes. One thing which is missing from the Tech reference is informa-
tion on the programmable peripheral chips on the system board. These

the 8259 interrupt controller
the 8253 timer
the 8237 DMA controller and
the 8255 peripheral interface

To make your library absolutely complete, you should order the INTEL data
sheets for these beasts.

I should say, though, that the only I ever found I needed to know about was
the interrupt controller. If you happen to have the 8086 Family User's
Manual, the big book put out by INTEL, which is one of the things people
sometimes buy to learn about 8086 architecture, there is an appendix there
which gives an adequate description of the 8259.

A final example

I leave you with a more substantial example of code which illustrates some
good elementary techniques; I won't claim its style is perfect, but I think
it is adequate. I think this is a much more useful example than what you
will get with the assembler:

PAGE 61,132
TITLE SETSCRN -- Establish correct monitor use at boot time
; This program is a variation on many which toggle the equipment flags
; to support the use of either video option (monochrome or color).
; The thing about this one is it prompts the user in such a way that he
; can select the use of the monitor he is currently looking at (or which
; is currently connected or turned on) without really having to know
; which is which. SETSCRN is a good program to put first in an
; This program is highly dependent on the hardware and BIOS of the IBMPC
; and is hardly portable, except to very exact clones. For this reason,
; BIOS calls are used in lieu of DOS function calls where both provide
; equal function.

OK. That's the first page of the program. Notice the PAGE statement,
which you can use to tell the assembler how to format the listing. You
give it lines per page and characters per line. I have mine setup to print
on the host lineprinter; I routinely upload my listings at 9600 baud and
print them on the host; it is faster than using the PC printer.

There is also a TITLE statement. This simply provides a nice title for
each page of your listing. Now for the second page:

SUBTTL -- Provide .COM type environment and Data
; First, describe the one BIOS byte we are interested in
BIOSDATA SEGMENT AT 40H ;Describe where BIOS keeps his data
ORG 10H ;Skip parts we are not interested in
EQUIP DB ? ;Equipment flag location
MONO EQU 00110000B ;These bits on if monochrome
COLOR EQU 11101111B ;Mask to make BIOS think of the color board
BIOSDATA ENDS ;End of interesting part
; Next, describe some values for interrupts and functions
DOS EQU 21H ;DOS Function Handler INT code
PRTMSG EQU 09H ;Function code to print a message
KBD EQU 16H ;BIOS keyboard services INT code
GETKEY EQU 00H ;Function code to read a character
SCREEN EQU 10H ;BIOS Screen services INT code
MONOINIT EQU 02H ;Value to initialize monochrome screen

;COLORINIT EQU 03H ;Value to initialize color screen (80x25)
COLORINIT EQU 01H ;Value to initialize color screen (40X25)
; Now, describe our own segment
SETSCRN SEGMENT ;Set operating segment for CODE and DATA
ORG 100H ;Begin assembly at standard .COM offset
MAIN PROC NEAR ;COM files use NEAR linkage
JMP BEGIN ;And, it is helpful to put the data first, but
; ;then you must branch around it.
; Data used in SETSCRN
CHANGELOC DD EQUIP ;Location of the EQUIP, recorded as far pointer
MONOPROMPT DB 'Please press the plus ( + ) key.$' ;User sees on mono
COLORPROMPT DB 'Please press the minus ( - ) key.$' ;User sees on color

Several things are illustrated on this page. First, in addition to titles,
the assembler supports subtitles: hence the SUBTTL pseudo-op. Second, the
PAGE pseudo-op can be used to go to a new page in the listing. You see an
example here of the DSECT-style segment in the "SEGMENT AT 40H". Here, our
our interest is in correctly describing the location of some data in the
BIOS work area which really is located at segment 40H.

You will also see illustrated the EQU instruction, which just gives a sym-
bolic name to a number. I don't make a fetish of giving a name to every
single number in a program. I do feel strongly, though, that interrupts
and function codes, where the number is arbitrary and the function being
performed is the thing of interest, should always be given symbolic names.

One last new element in this section is the define doubleword (DD) instruc-
tion. A doubleword constant can refer, as in this case, to a location in
another segment. The assembler will be happy to use information at its
disposal to properly assemble it. In this case, the assembler knows that
EQUIP is offset 10 in the segment BIOSDATA which is at 40H.

SUBTTL -- Perform function
BEGIN: CALL MONOON ;Turn on mono display
CALL COLORON ;Turn on color display
MOV AH,GETKEY ;Obtain user response
CMP AL,'+' ;Does he want MONO?

CALL MONOON ;yes. give it to him

The main code section makes use of subroutines to keep the basic flow sim-
ple. About all that's new to you in this section is the use of the BIOS
interrupt KBD to read a character from the keyboard.

Now for the subroutines, MONOON and COLORON:

SUBTTL -- Routines to turn monitors on
MONOON PROC NEAR ;Turn mono on
LES DI,CHANGELOC ;Get location to change
MOV AX,MONOINIT ;Get screen initialization value
INT SCREEN ;Initialize screen
COLORON PROC NEAR ;Turn color on
LES DI,CHANGELOC ;Get location to change
MOV AX,COLORINIT ;Get screen initialization value
INT SCREEN ;Initialize screen
SETSCRN ENDS ;End of segment
END MAIN ;End of assembly; execution at MAIN

The instructions LES and LDS are useful ones for dealing with doubleword
addresses. The offset is loaded into the operand register and the segment
into ES (for LES) or DS (for LDS). By telling the assembler, with an
ASSUME, that ES now addresses the BIOSDATA segment, it is able to correctly
assemble the OR and AND instructions which refer to the EQUIP byte. An ES
segment prefix is added.

To understand the action here, you simply need to know that flags in that
particular byte control how the BIOS screen service initializes the adapt-
ers. BIOS will only work with one adapter at a time; by setting the equip-
ment flags to show one or the other as installed and calling BIOS screen
initialization, we achieve the desired effect.

The rest is up to you.