Turbo_Assembler_Version_2.0_Users_Guide_1990 Turbo Assembler Version 2.0 Users Guide 1990 Turbo_Assembler_Version_2.0_Users_Guide_1990 Turbo_Assembler_Version_2.0_Users_Guide_1990
User Manual: Turbo_Assembler_Version_2.0_Users_Guide_1990
Open the PDF directly: View PDF .Page Count: 518
T RB
A E BLER
BORLAND
,
Turbo AssemblefID
Version 2.0
User's Guide
BORLAND INTERNATIONAL. INC. 1800 GREEN HILLS ROAD
P.O. BOX 660001, scons VALLEY, CA 95066-0001
Copyright © 1988, 1990 by Borland International. All rights
reserved. All Borland products are trademarks or registered
trademarks of Borland International, Inc. Other brand and
product names are trademarks or registered trademarks of their
respective holders.
PRINTED IN THE USA.
Rl
10 9 8 7 6 5 4 3 2
c
o
N
T
Introduction
1
Hardware and software requirements .... 2
About the manuals .................... 2
The User's Guide .................. 2
Notational conventions ................. 3
How to contact Borland ................ 4
Chapter 1 Installing Turbo
Assembler
7
Chapter 2 Getting started with Turbo"
Assembler
9
Chapter 3 Command-line
reference
N
T
ll
11
13
14
14
15
17
18
19
21
Starting Turbo Assembler from DOS .... 21
Command-line options ................ 24
la .................................. 25
Ib .................................. 25
Ic .................................. 26
Id ..... ............................. 26
Ie ............... "................... 27
Ih or I? ................ '............. 27
s
Ii ...................... "............ 28
Ij ..•...•••••..•......... ."........... 29
Ikh ................... ~ .. ~ ..........
Iks ............ "....................
II ..................................
Ila .................................
1m .................................
Files on disk .......................... 7
Installing Turbo Assembler ............. 8
Writing your first Turbo Assembler
program .............................
Assembling your first program .......
L'm k'mg your fir st program ...........
Running your first program ..........
What happened? ...................
Modifying your first Turbo Assembler
program ........................... '..
Sending output to a printer ..........
Writing your second Turbo Assembler
program .. "...........................
Running REVERSE.ASM ............
E
Iml .................................
Imu ................................
Imv# ...............................
29
30
30
30
31
31
32
32
32
Irnx. ................................
In .................................. 33
Ip .................................. 33
Iq .................................. 34
Ir .................................. 34
Is .................................. 35
It ............................... ~ .. 35
Iv .................................. 35
Iw .................................
Ix ..................................
Iz ..................................
Izd .................................
Izi .................................
36
37
37
38
38
Indirect command files ................ 39
The configuration file ................. 39
Chapter 4 The nature of assembly
language
41
The architecture of a computer ......... 41
The making of assembly language .... 43
The 8088 and 8086 processors .......... 46
The capabilities of the 8086 ........... 47
Memory ........................... 47
Input and output ................... 50
Registers .......................... 51
The flags register ................. 52
The general-purpose registers ...... 54
The AX register ................ 54
The BX register ................. 55
The CX register ................. 55
The DX register ................ 56
The SI register .................. 56
. The DI register ................. 57
The BP register ................. 58
The SP register ................. 59
The instruction pointer ............ 61
The segment registers ............. 61
The CS register ................. 65
The DS register ................. 66
The ES register ................. 66
The SS register ................. 66
The 8086 instruction set ............. 67
The IBM PC and XT ................... 67
Input and output devices ............ 68
Systems software for the IBM PC ..... 68
DOS ............................ 69
Getting keystrokes .............. 70
Displaying characters on the
screen ......................... 71
Ending a program .............. 72
The BIOS ........................ 73
Selecting display modes ......... 73
Sometimes you absolutely need to go to
the hard ware ...................... 74
Other resources .................... 74
Memory-addressing modes ........ 93
Comments ........................ 101
Segment directives ................... 103
Simplified segment directives ....... 104
.STACK, .CODE,and .DATA ...... 104
DOSSEG ....................... 108
.MODEL ....................... 108
Other simplified segment
directives . . . . . . . . . . . . . . . . . . . . . . . 110
Standard segment directives ........ 111
The SEGMENT directive ......... 112
The ENDS directive .............. 112
The ASSUME directive ........... 112
Simplified versus standard segment
directives . . . . . . . . . . . . . . . . . . . . . . . .. 115
Alloca ting data . . . . . . . . . . . . . . . . . . . . . . 116
Bits, bytes, and bases ............... 116
Decimal, binary, octal, and hexadecimal
notation ........................ 119
Default radix selection ........... 123
Initialized data .................... 125
Initializing arrays ................ 125
Initializing character strings ....... 126
Initializing with expressions and
labels .......................... 128
Uninitialized data ................. 129
Named memory locations .......... 130
Moving data ........................ 132
Selecting data size ................. 133
Signed versus unsigned data ........ 135
Converting between data sizes ...... 136
Accessing segment registers ..... ~ ... 138
Moving data to and from the stack ... 139
Exchanging data ................... 139
I/O .............................. 140
Operations ......................... 141
Arithmetic operations . . . . . . . . . . . . . . 141
Addition and subtraction ......... 142
32-bit operands ................ 142
Incrementing and decrementing . 144
Multiplication and division ....... 145
Changing sign .................. 147
Logicaloperations ................. 148
Shifts and rotates .................. 150
Chapter 5 The elements of an
assembler program
77
The components and structure of an
assembler program ................... 78
Reserved words ...................... 79
The format of a line ................... 81
Labels ............................. 82
Instruction mnemonics and directives. 85
The END directive ................ 86
Operands .......................... 88
Register operands ................ 89
Constant operands ................ 90
Expressions ...................... 91
Label operands ................... 92
II
Loops and jumps ....................
Unconditional jumps ...............
Conditional jumps .................
Looping ..........................
Subroutines .........................
How subroutines work .............
Parameter passing .................
Returning values .... _..............
Preserving registers ................
An example assembly language
program ............................
153
154
156
158
161
162
166
166
167
%LIST and %NOLIST .......... 211
%CONDS and %NOCONDS .... 211
%INCL and %NOINCL ........ 212
%MACS and %NOMACS ....... 212
%CTLS and %NOCTLS ......... 212
&UREF and %NOUREF ........ 213
%SYMS and %NOSYMS ........ 213
The listing format control
directives ....................... 213
Field-width directives .......... 214
%PUSHLCTL and %POPLCTL .. 215
Other listing control directives .... 215
Displaying a message during assembly . 215
Assembling source code conditionally .. 216
Conditional assembly directives ..... 217
IF and IFE ...................... 217
IFDEF and IFNDEF .............. 218
Other conditional assembly
directives ....................... 220
ELSEIF family of directives ....... 222
Conditional error directives ......... 223
.ERR, .ERR1, and .ERR2 .......... 223
.ERRE and .ERRNZ .............. 224
.ERRDEF and .ERRNDEF ......... 224
Other conditional error directives .. 225
Pitfalls in assembler programming ..... 225
Forgetting to return to DOS ......... 226
Forgetting a RET instruction ......... 227
Generating the wrong type of return . 228
Reversing operands ................ 230
Forgetting the stack or reserving a too
small stack ........................ 230
Calling a subroutine that wipes out
needed registers ................... 231
Using the wrong sense for a conditional
jump ............................. 234
Pitfalls with string instructions ...... 235
Forgetting about REP string
overrun ........................ 235
Relying on a zero CX to cover a whole
segment ....................... ~. 237
Using incorrect direction flag
settings ........................ 239
168
Chapter 6 More about programming
173
in Turbo Assembler
Using equate substitutions ............ 174
The EQU directive ................. 174
The $ predefined symbol ......... 178
The = directive .................... 179
The string instructions ....... " ....... 180
Data movement string instructions ... 181
LODS .......................... 181
STOS .......................... 182
MOVS ......................... 183
Repeating a string instruction ..... 184
String pointer overrun ........... 185
Data scanning string instructions .... 185
SCAS .......................... 185
CMPS .......................... 188
Using operands with string
instructions ....................... 189
Multimodule programs ............... 190
The PUBLIC directive .............. 193
The EXTRN directive ............... 194
The GLOBAL directive ............. 197
Include files . . . . . . . . . . . . . . . . . . . . . . . .. 198
The listing file ....................... 200
Annotated source code ............. 201
Listing symbol tables ............... 205
The table of labels ............... 205
The table of groups and segments .. 206
The cross-reference table ............ 207
Controlling the listing contents and
format ........................... 210
The line-listing selection directives . 210
iii
An example of inline assembly ...... 270
Limitations of inline assembly ....... 274
Memory and address operand
limitations ...................... 274
Lack of default automatic variable
sizing in inline assembly .......... 276
The need to preserve registers ..... 277
Preserving calling functions and
register variables .............. 277
Suppressing internal register
variables ..................... 277
Disadvantages of inline assembly versus
pureC ........................... 277
Reduced portability and
maintainability .................. 278
Slower compilation .............. 278
Available with TCC only ......... 278
Optimization loss ................ 278
Error trace-back limitations ....... 279
Debugging limitations ........... 280
Develop in C and compile the final code
with inline assembly ............. 280
Calling Turbo Assembler functions from
Turbo C ............................ 281
The framework .................... 282
Memory models and segments .... 282
Simplified segment directives and
TurboC ...................... 283
Old-style segment directives and
TurboC ...................... 285
Segment defaults: When is it
necessary to load segments? ..... 287
Publics and externals ............. 290
Underscores .................. 290
The significance of uppercase and
lowercase ..................... 291
Label types ................... 292
Far externals .. : ............... 293
Linker command line ............ 294
Between Turbo Assembler and Turbo
C ................................ 295
Parameter-passing ............... 295
Preserving registers .............. 302
. Returning values ................ 302
Using the wrong sense for a t:epeated
string comparison ............... 239
Forgetting about string segment
defaults ........................ 240
Converting incorrectly from byte to
word operations ................. 241
Using multiple prefixes ........... 242
Relying on the operand(s) to a string
instruction ...................... 242
Forgetting about unusual side effects . 243
Wiping out a register with
multiplication ................... 244
Forgetting that string instructions alter
several registers ................. 245
Expecting certain instructions to alter
the carry flag .................... 245
Waiting too long to use flags ...... 246
Confusing memory and immediate
operands ......................... 246
Causing segment wraparound ....... 249
Failing to preserve everything in an
interrupt handler .................. 251
Forgetting group overrides in operands
and data tables .................... 252
Chapter 7 Interfacing Turbo
Assembler with Turbo
C
257
Using inline assembly in Turbo C ...... 258
How inline assembly works ......... 260
How Turbo C knows to use inline
assembly mode .................. 264
Invoking Turbo Assembler for inline
assembly ....................... 265
Where Turbo C assembles inline
assembly ....................... 266
Use the -1 switch for 80186/80286
instructions ..................... 267
The format of inline assembly
statements ........................ 268
Semicolons in inline assembly ..... 268
Comments in inline assembly ..... 268
Accessing structure/union
elements ....................... 269
iv
Stack maintenance ................. 328
Accessing parameters .............. 328
Using BP to address the stack ..... 328
The ARG directive ............. 329
.MODEL and Turbo Pascal ........ 330
Using another base or index
register ......................... 331
Function results in Turbo Pascal ....... 331
Scalar function results ............ 331
Real function results ............. 332
8087 function results ............. 332
String function restilts ............ 332
Pointer function results ......... 332
Allocating space for local data ......... 332
Allocating private static storage .. '... 332
Allocating volatile storage .......... 333
Assembly language routines for Turbo
Pascal .............................. 334
General-purpose hex conversion
routine . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
Exchanging two variables ........... 337
Scanning the DOS environment ..... 340
Calling an assembler function from
C ................................ 304
Pascal calling conventions .......... 307
Calling Turbo C from Turbo Assembler. 308
Link in the C startup code .......... 308
Make sure you've got the right segment
setup ............................ 309
Performing the call ................ 309
Calling a Turbo C function from Turbo
Assembler ........................ 311
Chapter 8 Interfacing Turbo
Assembler with Turbo
Pascal
315
The Turbo Pascal memory map ........ 315
The program segment prefix ........ 316
Code segments .................... 317
The global data segment ............ 317
The stack ......................... 317
The heap ......................... 318
Register use in Turbo Pascal .......... 318
Near or far? ......................... 319
Sharing information with Turbo Pascal . 319
The $1 compiler directive and external
subprograms ...................... 319
The PUBLIC directive .............. 320
The EXTRN directive ............... 321
Restrictions on using EXTRN
objects ......................... 323
Using segment fixups .............. 324
Dead code elimination ............. 325
Turbo Pascal parameter-passing
conventions ......................... 325
Value parameters .................. 325
Scalar types ..................... 326
Reals ........................... 326
Single, Double, Extended, and Comp:
The 8087 types .................. 326
Poin ters ........................ 326
Strings ......................... 327
Records and arrays .............. 327
Sets ............................ 327
Variable parameters ............... 327
Chapter 9 Advanced programming In
Turbo Assembler
345
Segment override prefixes ............ 345
An alternate form .................. 347
When segment override prefixes don't
work ................ 000 .......... 348
Accessing multiple segments ....... 0349
Locallabels ............ 0.... 0... 0. 00 349
Automatic jump-sizing .. 0.. 0. 0....... 353
Forward references to code and data 00. 358
Using repeat blocks and macros . 0. 0. 00362
Repeat blocks ................. 0.. 0 362
Repeat blocks and variable
parameters ...... 0. 0..... 00... 00 364
Macros 0..... 000.00.0000 ... 0.... 0.365
Nestingmacros 00000 ... 00.0 .. 0.0369
Macros and conditionals 0...... 00. 369
Stopping expansion with EXITM 00371
Defining labels within macros ..... 371
Fancy data structures .0.0 .. 00 .. 0.... 0373
The STRUC directive .. 0.......... 0. 374
v
Advantages and disadvantages of using
STRUC ......................... 377
Unique structure field names .... 378
Nesting structures ............. 378
Initializing structures .......... 379
The RECORD directive ............. 380
Accessing records ............... 382
The WIDTH operator .......... 383
The MASK operator ............ 384
.Why use records ................. 385
The UNION directive .............. 387
Segment directives ................... 389
The SEGMENT directive ........... 389
The name and align fields ......... 390
The combine field ................ 390
The use and class fields ........... 392
Segment size, type, name, and
nesting ......................... 392
Segment-ordering ................. 393
The GROUP directive .............. 395
The ASSUME directive ............. 397
The simplified segment directives .... 400
A multisegment program ........... 404
New segment types ................ 420
Simplified segment directives and
80386 segment types ............. 423
The FWORD 48-bit data type ...... 424
New registers ..................... 425
The 32-bit general-purpose
registers ........................ 426
The 32-bit flags register ........... 428
The 32-bit instruction pointer ..... 428
New segment registers ........... 429
New addressing modes ............ 431
New instructions .................. 434
Testing bits ..................... 435
Scanning bits .................... 436
Moving data with sign- or zeroextension ....................... 437
Converting to DWORD or QWORD
data ........................... 437
Shifting across multiple words .... 438
Setting bytes conditionally ........ 439
Loading SS, FS, and GS ........... 440
Extended instructions .............. 441
Special versions of MOV .......... 441
32-bit versions of 8086 instructions . 442
New versions of LOOP and
JCXZ .......................... 442
New versions of the string
instructions ................... 444
IRETD ....................... 444
PUSHFD and POPFD .......... 445
PUSHAD and POPAD ......... 445
New versions ofIMUL ........... 445
Mixing 16-bit and 32-bit instructions and
segments .......................... 446
An example 80386 function ......... 449
The 80287 .......................... 453
The 80387 .......................... 453
Chapter 10 The 80386 and other
processors
409
Switching processor types in assembler
code ............................... 410
The 80186 and 80188 ................. 411
New instructions .................. 411
PUSHAandPOPA .............. 411
ENTER and LEAVE .............. 412
BOUND ........................ 413
INSandOUTS .................. 414
Extended 8086 instructions .......... 415
Pushing immediate values ........ 415
Shifting and rotating by immediate
values .... ; ..................... 416
Multiplying by an immediate
value .......................... 416
The80286 .......................... 417
Enabling 80286 assembly ........... 418
The80386 .......................... 419
Selecting 80386 assembly mode ...... 419
Chapter 11 Turbo Assembler Ideal
Mode
455
What is Ideal mode? ................. 456
Why use Ideal mode? ................ 456
Entering and leaving Ideal mode ...... 457
MASM and Ideal mode differences .... 458
vi
Quoted strings as arguments to
directives ....................... 469
Segments and groups .............. 470
Accessing data in a segment belonging
to a group ...................... 470
Defining near or far code labels ...... 472
External, public, and global symbols .473
Miscellaneous differences ........... 474
Suppressed fixups ............... 474
Operand for BOUND instruction .. 474
Comments inside macros ......... 475
Local symbols ................... 475
A comparison of MASM and Ideal mode
programming ....................... 475
MASM mode sample program .. 476
Ideal mode sample program .... 477
An analysis of MASM And Ideal
modes ........................... 479
Ideal mode tokens ................. 459
Syr,nboltokens .................. 459
Duplicate member names ......... 460
Floating-point tokens ............ 460
EQU and =directives .............. 461
Expressions and operands .......... 461
Square brackets operator ......... 461
Example operands ............... 462
Operators ........................ 464
Periods in structure members ..... 464
Pointers to structures ............. 464
The SYM1YPE operator .......... 465
The HIGH and LOW operators .... 465
The Optional PlR operator ....... 466
The SIZE operator ............... 466
Directives ........................ 467
Listing controls .................. 467
Directives starting with a period (.) . 468
Reversed directive and symbol
name .......................... 469
vii
References
483
Index
485
T
A
L
B
5.1: TASM reserved words ............. 80
5.2: The operation of the 8086 AND, OR, and
XOR logical instructions .......... 148
5.3: Conditional jump instructions ..... 157
6.1: Source and destination for the MUL and
IMUL instructions ................ 244
7.1: Register settings when Turbo Centers
assembler ....................... 287
9.1: Default segments and types for tiny
memory model .................. 401
E
s
9.2: Default segments and types for small
memory model .................. 401
9.3: Default segments and types for medium
memory model .................. 401
9.4: Default segments and types for compact
memory model .................. 401
9.5: Default segments and types for large or
huge memory model ............. 402
9.6: Default segments and types for Turbo
Pascal (TPASCAL) memory model . 402
viii
F
G
u
2.1: The edit, assemble, link, and run
cycle ............................. 12
3.1: Turbo Assembler command line ..... 22
4.1: Five subsystems ................... 42
4.2: Memory address space of the 8086 ... 48
4.3: Separate memory and I/O address of
8086 ............................. 50
4.4: Registers of the 8086 ............... 52
4.5: Flags register of the 8086 ........... 53
4.6: AX, BX, SP, and the stack ........... 60
4.7: 20-bit memory addresses ........... 62
4.8: Calculation of memory address by
mov ............................. 63
4.9: DOS and BIOS systems software as a
control and interface layer .......... 69
5.1: The memory location of the character
string CharString .................. 94
5.2: Addressing the character string
CharString ....................... 95
5.3: Using BX to address CharString ..... 96
5.4: Storing WordVar and DwordVar.... 119
5.5: From binary 001100100 (decimal 1(0) to
octal 1440 ....................... 121
5.6: From binary 01100100 (decimal 100) to
hexadecimal64 .................. 121
5.7: Example of five-entry array ........ 126
5.8: Example of a shift left ............. 150
5.9: Example of SAR (arithmetic right
shift) ........................... 151
R
E
s
5.10: Example of ROR (rotate right) ..... 152
5.11: Example ofRCR (rotate right and
carry) .......................... 153
5.12: Operation of a subroutine ........ 162
6.1: Memory variables: offset vs. value .. 247
6.2: An example of segment
wraparound ..................... 250
6.3: Three segments grouped into one
segment group .................. 252
7.1: Turbo C compile and link cycle .... 260
7.2: Turbo C compile, assembly, and link
cycle ........................... 262
7.3: Compile, assemble, and link with Turbo
C, Turbo Assembler, and TLINK ... 281
7.4: State of the stack just before executing
Test's first instruction ............. 296
7.5: State of the stack after PUSH and
MOV ........................... 297
7.6: State of the stack after PUSH, MOV, and
SUB ............................ 298
7.7: State of the stack immediately after
MOV BP, SP ..................... 308
·8.1: Memory map of a Turbo Pascal 5.0
program ........................ 316
9.1: Locations and initial values of the fields
in 1Rec ......................... 381
10.1: The registers of the 80386 ......... 426
Ix
N
T
o
R
D
u
c
T
o
N
Welcome to Borland's Turbo Assembler, a multi-pass assembler
with forward-reference resolution, assembly speeds of up to
48,000 lines per minute (on an IBM PS/2 model 60), MASM
compatibility, and an optional Ideal mode extended syntax.
Whether you're a novice or an experienced programmer, you'll
appreciate these features along with a number of others we've
provided to make programming in assembly language easier.
We'll mention just a few of the highlights here and describe them
in detailla ter in the book:
•
•
•
•
•
•
full 80386 support
improved syntax type-checking
simplified segmentation directives
improved listing controls
PUSH, POP extensions
extended CALL statement with arguments and optional
language parameter
• local labels
!!! local stack symbols and calling arguments in procedures
• structures and unions
• nested directives
• Quirks mode to emulate MASM
• full source debugging output for Turbo Debugger
• built-in cross-reference utility (TCREF)
• configura tion and command files
Turbo Assembler is a powerful command-line assembler that
takes your source (.ASM) files and produces object (.OBJ)
modules. You then use TLINK.EXE, Borland's high-speed linker
program, to link your object modules and create executable (.EXE)
files.
Introduction
Turbo Assembler is set up to work with the 80x86 and 80x87
processor families. (For more information about the instruction
sets of the 8Ox86/80x87 families, consult the Intel data books.)
Hardware and software requirements
Turbo Assembler runs on the IBM PC family of computers,
including the XT, AT, and PS/2, along with all true compatibles.
Turbo Assembler requires MS-DOS 2.0 or later, and at least 256K
of memory.
Turbo Assembler generates instructions for the 8086, 80186, 80286,
and 80386 processors. It also generates floating-point instructions
for the 8087, 80287, and 80387 numeric coprocessors.
About the manuals
Turbo Assembler comes with two books and a quick-reference
guide: Turbo Assembler User's Guide (this book), Turbo Assembler
Reference Guide, and Turbo Assembler Quick-Reference Guide. The
User's Guide provides basic instructions for using Turbo
Assembler and a thorough examination of assembler
programming. The Reference Guide describes the operators,
predefined symbols, and directives Turbo Assembler uses. The
Quick Reference is a handy guide to processor and coprocessor
instructions.
Here's a more detailed look at what the User's Guide contains.
The User's Guide
Chapter 1: Installing Turbo Assembler tells you about the files on
your distribution disks and what you need to do to install Turbo
Assembler on your system.
Chapter 2: Getting started In Turbo Assembler provides you with
an introduction to programming in assembly language, and a few
sample programs to make you comfortable using the commandline switches.
Chapter 3: Command-line reference details all the command-line
options, plus tells you about using the configuration file and
command files.
2
Turbo Assembler User's Guide
Chapter 4: The nature of assembly language leads you through a
discussion of computers in general and the 8088 processor in
particular.
Chapter 5: The elements of an assembler program describes the
basic components of assembler, with some good solid information
about directives, instructions, accessing memory, segments, and
more.
Chapter 6: More about programming In Turbo Assembler goes one
step further than Chapter 5, discussing some advanced aspects of
Turbo Assembler-more about directives, string instructions, and
so on. This chapter also covers some common pitfalls you may
encounter as an assembly progralnmer.
Chapter 7: Interfacing Turbo Assembler with Turbo C describes
how to use Turbo C, a high-level language, with assembly
language. We detail how to link assembler modules to Turbo C
and how to call Turbo Assembler functions from Turbo C.
Chapter 8: Interfacing Turbo Assembler with Turbo Pascal tells
you how to interface your assembler code with your Turbo Pascal
code; sample programs are also provided.
Chapter 9: Advanced programming In Turbo Assembler provides
you with more details about everything we've touched on in
earlier chapters, such as segment override prefixes, macros,
segment directives, and so on.
Chapter 10: The 80386 and other processors covers programming
with the 80386.
Chapter 11: Turbo Assembler Ideal mode tells you all about Ideal
mode and why you'll want to use it.
References lists several useful books about assembly
programming.
Notational conventions
When we talk about IBM PCs or compatibles, we're referring to
any computer that uses the 8088, 8086, 80186, 80286, and 80386
chips (all of these chips are commonly referred to as 80x86). When
discussing PC-OOS, DOS, or MS-DOS, we're referring to version
2.0 or greater of the operating system.
Introduction
3
All typefaces were produced by Borland's Sprint: The Professional
Word Processor, output on a PostScript printer. The different
typefaces displayed are used for the following purposes:
Italics
In text, italics represent labels, placeholders,
variables, and arrays. In syntax expressions,
placeholders are set in italics to indicate that they
are user-defined.
Boldface
Boldface is used in text for directives, instructions,
symbols, and operators, as well as for commandline options.
CAPITALS
In text, capital letters are used to represent
instructions, directives, registers, and operators.
Monospace
Monospace type is used to display any sample
code, text or code that appears on your screen, and
any text that you must actually type to assemble,
link, and run a program.
Keycaps
In text, keycaps are used to indicate a key on your
keyboard. It is often used when describing a key
you must press to perform a particular function;
for example, "Press Enter after typing your program
name at the prompt."
How to contact Borland
If, after reading this manual and using Turbo Assembler, you
would like to contact Borland with comments, questions, or
suggestions, we suggest the following procedures:
• The best way is to log on to Borland's forum on CompuServe:
Type GO BPROGB at the main CompuServe menu and follow the
menus to Turbo Assembler. Leave your questions or comments
here for the support staff to process .
• If you prefer, write a letter detailing your problem and send it
to
Technical Support Department
Borland International
P.O. Box 660001
1700 Green Hills Drive
Scotts Valley, CA 95066 U.S.A.
4
Turbo Assembler User's Guide
• You can also telephone our Technical Support department at
(408) 438-5300. To help us handle your problem as quickly as
possible, have these items handy before you call:
•
•
•
•
product name and version number
product serial number
computer make and model number
operating system and version number
If you're not familiar with Borland's No-Nonsense License
statement, now's the time to read the agreement at the front of this
manual and mail in your completed product registration card.
Introduction
5
6
Turbo Assembler User's Guide
c
H
p
A
T
R
E
1
Installing Turbo Assembler
Before we get you up to speed on programming in assembler,
you'll need to get one thing out of the way. Take the Turbo
Assembler disks and make copies (via DOS) of each one to create
your "working" copies. Once you've done that, put the original
disks away. (There's a fee to replace disks that you damage, so
only use the originals to make backups and work copies.)
If you are going to use Turbo Assembler as a replacement for
MASM, read Appendix B in the Reference Manual to see in which
areas Turbo Assembler behaves differently from MASM.
Note: Be sure to read the README file before working with
Turbo Assembler. This file contains the latest information about
the program, as well as corrections and/or additions to the
manuals.
Files on disk
•
•
•
•
•
•
TASM.EXE: Turbo Assembler
TLINK.EXE: Turbo Linker
MAKE.EXE: Command-line MAKE utility
TLIB.EXE: Turbo Librarian
README.COM: Program to display README file
README: Any last minute information about the software and
documentation
Chapter 7, Installing Turbo Assembler
7
•
•
•
•
•
•
TCREF.EXE: A source file cross-reference utility
OBJXREF.COM: Object file cross-reference utility
GREP.COM: Grep utility
TOUCH.COM: A file-update utility
INSTALL.EXE: Installation program
MMACROS.ARC: An archived file of MASM mode macros
Installing Turbo Assembler
The INSTALL disk contains a program called INST ALL.EXE that
will assist you with the installation of Turbo Assembler 1.0. There
are two options for installation:
1. Hard Disk Users: This option allows you to pick the
subdirectories where the files will be loaded.
2. Floppy Disk Users: This option will install the files necessary
to use Turbo Assembler on a two-drive system. Be sure to
have four fonnatted disks ready before you start.
To start the installation, change your current drive to the one that
has the INSTALL program on it and type INSTALL. You will be
given instructions for each prompt in a box at the bottom of the
screen. For example, if you will be installing from drive A, type
INSTALL
Before you install Turbo Assembler, be sure to read the README
file to get further infonnation about this release.
Note: If you will be running INSTALL on a laptop or any other
system that uses an LCD display, you should set your system to
black-and-white mode before running INSTALL. You can do this
from DOS with the following command line:
mode bw80
You can also force INSTALL to come up in black-and-white mode
by using the Ib switch:
INSTALL /b
8
Turbo Assembler User's Guide
c
H
A
p
T
E
R
2
Getting staried with Turbo Assembler
If you've never programmed in assembly language before, this is
the place to begin. You might have heard that assembly language
programming is a black art suited only to hackers and wizards.
Don't believe it! Assembly language is nothing more than a
human form of the language of the computer itself and, as you'd
expect, the computer's own language is highly logical. As you
might also expect, assembly language is very powerful-in fact,
assembly language is the only way to tap the full power of the
Intel80x86 family, the processors at the heart of the IBM PC
family and compatibles.
You can write whole programs in nothing but assembly language
or you can, if you want, mix assembly language into programs
written in Turbo C, Turbo Pascal, Turbo Prolog, Turbo Basic, and
other languages. Either way, with assembly language, you can
write small and blindingly fast programs. As important as speed
is the assembly language code's ability to control every aspect of
your computer's operation, down to the last tick of the system
dock.
In this chapter, we'll introduce you to assembly language and
explore the unique qualities of assembly language programming.
You'll enter and run several working assembly language
programs, both to get a feel for the language and to get used to
working with the assembler.
Chapter 5, "The elements of an assembler program," picks up
where this chapter leaves off, covering the structure of ~n
Chapter 2, Getting started with Turbo Assembler
9
assembly language program and fundamental program elements
and summing up everything you've learned with a full-fledged
example program.
Chapter 6, "More about programming in Turbo Assembler,"
continues to explore assembly language programming, and
Chapter 9, "Advanced programming in Turbo Assembler,"
progresses to memory models, macros, and other advanced
topics.
Naturally, we can't make you expert assembly language
programmers in the course of a few chapters; we're simply
introducing you to assembly language and getting you started on
the road to writing your own programs. We strongly suggest that
you get one of the many excellent books devoted entirely to
assembly language programming and PC architecture (see the
references at the end of this book). In addition, you may find
IBM's DOS Technical Reference, BIOS Interface Technical Reference,
and Personal Computer XT Technical Reference manuals to be useful
reference material; these manuals document the assembly
language programming interface to the systems software and
hardware of IBM's personal computers.
Before you read further, you might want to read Chapter 3,
"Command-line reference," to familiarize yourself with the
command-line options. You should also install Turbo Assembler
(make working copies of your Turbo Assembler disks or copy the
files from your Turbo Assembler disks onto your hard disk) as
described in Chapter 1, ''Installing Turbo Assembler," if you
haven't already done so.
One final point: Assembly language is a complex topic, and there
are many things you will need to know in order to write even a
relatively simple assembly language program. Sometimes we'll
have to use features in our examples that we haven't discussed
yet, simply because we have to start somewhere. Bear with us; we'll
explain everything in due course. If, at any time, you're curious
about a specific feature, just look in Chapter 3, "Directives," in the
Reference Guide.
With that out of the way, and with Chapter 3 of the second
volume close at hand, it's time to create your first assembly
language program.
You can follow this tutorial step by step, typing in all the code
examples as you go, or you can unpack the example file on disk
(when you install Turbo Assembler) and have all the programs at
10
Turbo Assembler User's Guide
your fingertips. (Whatever your decision, the file names are
provided at the beginning of each example for your convenience.
Writing your first Turbo Assembler program
In the world of programming, the first program is traditionally a
program that displays the message, "Hello, world" and that's as
good a place as any for us to start.
Get into your text editor of choice (one that outputs ASCII files),
and type in the following lines that make up the program
HELLO.ASM:
.MODEL small
.STACK 100h
•DATA
HelloMessage DB 'Hello, world' ,13,10,'$'
.CODE
mov ax,@data
mov ds,ax
iset OS to point to the data segment
mov ah,9
iDOS print string function
mov dx,OFFSET HelloMessage ipoint to "Hello, world"
int 21h
idisplay "Hello, world"
mov ah,4ch
iDOS terminate program function
int 21h
iterminate the program
END
As soon as you've entered HELLO.ASM, save it to disk.
If you're familiar with C or Pascal, you might be thinking that the
assembler version of "Hello, world" seems a bit long. Well, yes,
assembler programs do tend to be long because each assembler
instruction by itself does less than a C or Pascal instruction. On
the other hand, you've got complete freedom in combining those
assembler instructions in any way you want. That means that,
unlike C and Pascal, assembler lets you program the computer to
do anything it's capable of-and that's often worth typing a few
extra lines.
Assembling your
first program
Now that you've saved HELLO.ASM, you'll want to run it. Before
you can run the program, though, you'll have to convert it into an
executable (able to be run or executed) form. This requires two
additional steps, assembling and linking, as shown in Figure 2.1,
Chapter 2, Getting started with Turbo Assembler
11
which depicts the complete edit, assemble, link, and run program
development cycle.
Figure 2.1
The edit, assemble, link, and
run cycle
Create a New Program
Assembler Source File
HELLO.ASM
Assemble
!
Object File
HELLO.OBJ
I
l
Link
Executable File
HELLO.EXE
I
Run
!
(If changes are needed)
~
The assembly step turns your source code into an intermediate
form called an object module, and the linking step combines one or
12
Turbo Assembler User's Guide
more object modules into an executable program. You can do
your assembling and linking from the command line.
To assemble HELLO.ASM, type
TASM hello
Unless you specify another file name, HELLO.ASM will be
assembled to HELLO.OBJ. (Note that you don't need to type in
the file extension name; Turbo Assembler assumes .ASM in this
case.) This is what you'll see onscreen:
Turbo Assembler Version 2.0 Copyright (c) 1988, 1990 by Borland
International, Inc.
Assembling file:
Error messages:
Warning messages:
Passes: 1
Remaining memory:
HELLO.ASM
None
None
266K
You won't receive any warnings or errors if you typed
HELLO.ASM exactly as shown. If you get warnings or errors,
they'll appear onscreen, along with the line numbers to indicate
where they occurred. If you get errors, check your code and make
sure it's precisely the same as the code we've shown you, then
assemble the program again.
Linking your first
program
After you've successfully assembled HELLO.ASM, you're only
one step away from running your first assembler program. Once
you've linked the just-assembled object code into an executable
form, you can run the program.
To link the program, you'll use TLINK, the linker accompanying
Turbo Assembler. At the command line, type
TLINK hello
Again, there's no need to enter the extension name; TLINK
assumes it's .OBJ. When linking completes (again, after a few
seconds at most), the linker automatically gives the .EXE file the
same name as your object file, unless you've specified otherwise.
When linking is successful, this message appears onscreen:
Turbo Link Version 3.0 Copyright (c) 1987, 1990 by Borland
International, Inc.
Chapter 2, Getting started with Turbo Assembler
13
Errors can occur during the linking process, although that's
unlikely with this example program. If you do receive any link
errors (they'll appear onscreen), modify your code to exactly
rna tch the code shown here, then assemble and link again.
Running your first
program
Now you're ready to run your program. Type hello at the DOS
prompt. The message
Hello, world
will be displayed onscreen. And that's all there is to it-you've
just created and run your first Turbo Assembler program!
What happened?
Now that you've gotten HELLO.ASM up and running, let's go
back and figure out exactly what happened along the path from
entering text to running the program.
When you first entered the assembler source code, the text was
stored by your text editor in memory. If the computer had been
turned off at this point, for whatever reason, the source code
would have been lost; consequently, we suggest you save your
source code early and often in order to avert possible tragedy.
When you saved the source code to disk, a permanent copy of the
text was stored in the file HELLO.ASM, where it would survive
even if you shut off your computer. (HELLO.ASM might not
survive a disk crash, however, so we also suggest that you back
up your disks regularly.) HELLO.ASM is a standard ASCII text
file; you can display it at the DOS prompt by typing
type hello.asm
and you can edit it with any text editor.
When you assembled HELLO.ASM, Turbo Assembler turned the
text instructions in HELLO.ASM into their binary equivalents in
the object file HELLO.OBI. HELLO.OBI is an intermediate file,
partway between source code and an executable file. HELLO.OBI
contains all the information needed to make executable code out
of the instructions that started out in HELLO.ASM, but it's in a
form that can readily be combined with other object files into a
single program. In Chapter 6, "More about programming in
Turbo Assembler," you'll see how useful this can be when you're
developing large programs.
14
Turbo Assembler User's Guide
Next, when you linked HELLO.OB], TLINK converted it into the
executable file HELLO.EXE. Finally, you ran HELLO.EXE when
you typed hello at the prompt.
Now type
dir hello.*
to list the various HELLO files on your disk. You'll find
HELLO.ASM, HELLO.OB], HELLO.EXE, and HELLO.MAP.
Modifying your first Turbo Assembler program
Now go back to your editor and modify your program to accept a
bit of input from the outside world. (The outside world is you,
typing at your keyboard.) Change the code to the following:
.MODEL small
.STACK 100h
•DATA
TimePrompt DB 'Is it after 12 noon (Y/N)?$'
GoodMorningMessage LABEL BYTE
DB 13,10,'Good morning, world!',13,10,'$'
GoodAfternoonMessage LABEL BYTE
DB 13,10,'Good afternoon, world!' ,13,10,'$'
•CODE
mov ax,@data
mov ds,ax
iset OS to point to data segment
mov dx,OFFSET TimePrompt
;point to the time prompt
mov ah,9
;DOS print string function #
int 21h
idisplay the time prompt
mov ah,1
;DOS get character function #
int 21h
;get a single-character response
cmp al,'y'
ityped lowercase y for after noon?
jz IsAfternoon
iyes, it's after noon
cmp al,'Y'
ityped uppercase Y for after noon?
jnz IsMorning
ino, it's before noon
IsAfternoon:
mov dx,OFFSET GoodAfternoonMessage ;point to the afternoon
; greeting
jmp DisplayGreeting
IsMorning:
mov dx,OFFSET GoodMorningMessage
;point to the before noon
i greeting
DisplayGreeting:
mov ah,9
iDOS print string function #
int 21h
;display the appropriate greeting
Chapter 2, Getting started with Turbo Assembler
15
mov ah,4ch
int 21h
;005 terminate program function t
;terminate the program
END
You've added two important new capabilities to your program:
input and decision-making. This program asks you whether it's
after noon, then accepts a single-character response from the
keyboard. If the character typed is an uppercase or lowercase Y,
the program displays a greeting appropriate for the afternoon;
otherwise, it gives a morning greeting. All the essential elements
of a useful program-input from the outside world, data
processing and decision-making, and output to the outside
world-are present in this code.
Save the modified program to disk. (This replaces your original
version of HELLO.ASM with the modified code, so the original
version will be lost.) Then reassemble and relink the program just
as you did in the previous examples. Run the program again by
typing hello at. the DOS prompt. The message
Is it after 12 noon (YIN)?
is displayed, with the cursor blinking after the question mark,
waiting for your response. Press Y. The program responds
Good afternoon, world!
HELLO.ASM is now an interactive, decision-making program.
In the course of your assembler programming, you will surely
make a wide variety of mistakes in typing and in program syntax.
Turbo Assembler ca tches many mistakes for you as it assembles
your code, reporting all such errors. The mistakes reported fall
into two categories: warnings and errors. Turbo Assembler
displays a warning message if it detects something suspicious, but
not necessarily wrong, in your code; sometimes warnings can be
ignored, but it's always best to check them out and make sure you
understand the problem. Turbo Assembler displays an error
message if it encounters something clearly wrong in your code
that makes it impossible to complete assembly and generate an
object file.
In other words, warnings are cautionary or nonfatal, while errors
must be fixed before you can run a program. The many error and
warning messages Turbo Assembler can generate are covered in
Appendix E in the Reference Guide.
16
Turbo Assembler User's Guide
As with any programming language, Turbo Assembler can't catch
logic errors for you. Turbo Assembler can tell you whether your
code can be assembled, but it can't tell you whether the assembled
code will perform as you intended it to-only you can be the
judge of that.
Don't worry if the example code doesn't make much sense to you
right now. Even programmers experienced in other languages
take some time to become fluent in 8086 assembly language;
there's really nothing quite like it under the sun. At this point,
you're just getting a feel for what assembler programs look like.
Later in this chapter, and in Chapter 5, "The elements of an
assembler program," we'll cover each of the elements of the
programs presented.
To list or send your program to a printer, consult your specific
text editor's manual. Turbo Assembler source files are normal
ASCII text files, so you can also print any assembler source file
from the DOS prompt with the PRINT command.
Sending output to
a printer
The printer is a handy output device; not only will you sometimes
want to send your program files to the printer, but you'll also
want your programs to send output to the printer on occasion.
The following is a version of the "Hello, world" program that
displays its output on the printer rather than on the screen:
.MODEL small
.STACK 100h
•DATA
HelloMessage DB 'Hello, world' ,13,10,12
HELLO_MESSAGE_LENGTH EQU $ - HelloMessage
•CODE
mov ax,@data
mov ds,ax
;set OS to point to the data segment
mov ah,40h
;DOS write to device function f
mov bx,4
;printer handle
;number of characters to print
mov cx,HELLO_MESSAGE_LENGTH
mov dx,OFFSET HelloMessage
i string to print
int 21h
iprint "Hello, world"
;DOS terminate program function f
mov ah,4ch
int 21h
;terminate the program
END
In this version of the "Hello, world" program, you've replaced the
OOS function to print a string on the screen with a OOS function
Chapter 2. Gefflng started with Turbo Assembler
17
that sends a string to a selected device or file-in this case, the
printer. Enter and run the program, and see that a sheet containing the familiar ''Hello, world" message is printed. (Don't
forget to save the program before running it. Again, this saves the
modified code in HELLO.ASM, and the previous version of the
program will be lost.)
You can modify this program to send output to the screen rather
than to the printer, displaying "Hello, world" onscreen again,
simply by changing
mov bx,4
;printer handle
mov bx,l
;standard output handle
to
Make this change, then reassemble and relink before running the
program again. You'll note that when the output is displayed on
the screen, the final character shown is the universal symbol for
"female" (~). This is actually a formfeed character, which the
program sent to the printer to force it to eject the sheet on which
you'd printed "Hello, world." Since the screen doesn't have
sheets, it doesn't know about formfeeds, so it simply displays the
corresponding member of the PC's character set when told to
print a formfeed character.
Writing your second Turbo Assembler program
Now you're ready to enter and run a program that actually does
something, REVERSE.ASM. Go back into your text editor and
enter the following:
.MODEL small
.STACK 100h
•DATA
MAXIMUM_STRING_LENGTH EQU 1000
StringToReverse DB MAXIMUM_STRING_LENGTH DUP(?)
ReverseString
DB MAXIMUM_STRING_LENGTH DUP(?)
•CODE
mov ax,@data
mov ds,ax
;set DS to point to the data segment
mov ah,3fh
;DOS read from handle function f
mov bx,O
;standard input handle
mov cx,MAXIMUM_STRING_LENGTH ;read up to maximum number of
; characters
18
Turbo Assembler User's Guide
mov
int
and
jz
mov
dx,OFFSET StringToReverse istore the string here
21h
iget the string
ax,ax
iwere any characters read?
Done
ino, so you're done
cx,ax
iPut string length in ex, where
i you can use it as a counter
push cx
isave the string length
mov bx,OFFSET StringToReverse
mov .si,OFFSET ReverseString
add si,cx
dec si
ipoint to the end of the
i reverse string buffer
ReverseLoop:
mov aI, [bx]
iget the next character
istore the characters in reverse order
mov [si],al
inc bx
ipoint to next character
ipoint to previous location
dec si
i in reverse buffer
loop ReverseLoop
imove next character, if any
pop cx
iget back the string length
iDOS write from handle function f
mov ah,40h
mov bx,l
istandard output handle
mov dx,OFFSET ReverseString iprint this string
iprint the reversed string
int 21h
Done:
iDOS terminate program function f
mov ah,4ch
int 21h
iterminate the program
END
You'll see what the program actually does in a moment; first, as
always, you should save your work.
Running
REVERSE.ASM
To run REVERSE.ASM, you must first assemble it; type
TASM reverse
then type
TLINK reverse
to create the executable file.
Type reverse at the prompt to run your program. If Turbo
Assembler reports any errors or warnings, carefully check your
code to see that it matches the code shown previously, then try
running the program again.
Chapter 2, Geffing started with Turbo Assembler
19
After you run your program, the cursor will sit blinking onscreen.
Apparently, the program is waiting for you to type something.
Try typing
ABCDEFG
then press Enter. The program displays
GFEDCBA
and ends. Type reverse agam at the command line. This time, type
0123456789
and press Enter. The program displays
9876543210
Now it's clear what REVERSE.ASM does: It reverses whatever
string of characters you type in. Speedy manipulation of
characters and strings is one of the areas in which assembly
language excels, as you'll see in the next few chapters.
Congratulations! You've entered, assembled, linked, and run
several assembler programs, and you've seen the fundamentals of
assembler programming-input, processing, and output-in
action.
If you don't want an object file but you do want a listing file, or if
you want a cross-reference file but don't want a listing file or
object file, you can specify the null device (NUL) as the file name.
For example,
TASM FILE1"NUL,
assembles file FILEl.ASM to object file FILE1.0BJ, doesn't
produce a listing file, and creates a cross-reference file FILEl.XRF.
Now you're ready to learn the basic elements of assembler
programming, covered in Chapter 5, liThe elements of an
assembler program."
20
Turbo Assembler User's Guide
c
H
p
A
T
E
R
3
Command-line reference
This chapter is dedicated to familiarizing you with Turbo
Assembler's command-line options. We'll describe each of the
command-line options you can use to alter the assembler's
behavior, then show how and when to use command files. Finally,
we describe the configuration file.
Starting Turbo Assembler from DOS
Turbo Assembler has a very powerful and flexible command-line
syntax. If you start Turbo Assembler without giving it any
arguments, like this,
TASM
you'll get a screenful of help describing many of the commandline options and the syntax for specifying the files you want to
assemble. Figure 3.1 shows you how this looks.
Chapter 3, Command-line reference
21
Figure 3.1
Turbo Assembler command
line
Turbo Assembler Version 2.0 Copyright (C) 1988, 1990 by Borland
International, Inc.
Syntax: TASM [options] source [,object] [,listing] [,xref]
/a,/s
/c
/dSYM[=VAL]
/e,/r
/h,/?
/iPATH
/jCMD
/khl,/ksl
/l,/la
/ml,/mx,/mu
/mvl
/ml
/n
/p
/q
/t
/wO,/wl,/w2
/w-xxx,/w+xxx
/x
/z
/zi,/zd
Alphabetic or Source-code segment ordering
Generate cross-reference in listing
Define symbol SYM = 0, or = value VAL
Emulated or Real floating-point instructions
Display this help screen
Search PATH for include files
Jam in assembler directive CMD (eg. /jIDEAL)
Hash table capacity I, String space capacity I
Generate listing: l=normal listing, la=expanded listing
Case sensitivity on symbols: ml=all, mx=globals, mu=none
Set maximum valid length for symbols
Allow I mUltiple passes to resolve forward references
Suppress symbol tables in listing
Check for code segment overrides in protected mode
Suppress OBJ records not needed for linking
Suppress messages if successful assembly
Set warning level: wO=none, w1=w2=warnings on
Disable (-) or enable (+) warning xxx
Include false conditionals in listing
Display source line with error message
Debug info: zi=full, zd=line numbers only
With the command-line options, you can specify the name of one
or more files that you want to assemble, as well as any options
that control how the files get assembled.
The general form of the command line looks like this:
TASM fileset [; fileset] ...
The semicolon (i) after the left bracket (D allows you to assemble
multiple groups of files on one command line by separating the
file groups. If you prefer, you can set different options for each set
of filesi for example,
TASM Ie FILE1; la FILE2
assembles FILE1.ASM with the Ie command-line option and
assembles file FILE2.ASM with the la command-line option.
In the general form of the command line, fileset can be
[option] ••• sourcefile [[+] sourcefile] ...
[, [objfile] [, [listfile] , [, [xreffile]]]]
This syntax shows that a group of files can start off with any
options you want to apply to those files, followed by the files you
22
Turbo Assembler User's Guide
want to assemble. A file name can be a single file name, or it can
use the normal OOS wildcard characters If- and ? to specify
multiple files to assemble. If your file name does not have an
extension, Turbo Assembler adds the .ASM extension. For
example, to assemble all the .ASM files in the current directory,
you would type
TASM
*
If you want to assemble multiple files, you can separate their
names with the plus sign (+):
TASM MYFILEI + MYFlLE2
You can follow the file name you want to assemble by an optional
object file name, listing file name, and a cross-reference file name.
If you do not specify an object file or listing file, Turbo Assembler
creates an object file with the same name as the source file and an
extension of .OBJ.
A listing file is not generated unless you explicitly request one. To
request one, place a comma after the object file name, followed by
a listing file name. If you don't explicitly provide a listing file
name, Turbo Assembler creates a listing file with the same name
as the source file and the extension .LST. If you supply a listing
file name without an extension, .LST is appended to it.
A cross-reference file is not generated unless you explicitly
request one. To request one, place a comma after the listing file
name, followed by a cross-reference file name. If you don't
explicitly provide a cross-reference file name, Turbo Assembler
creates a cross-reference file with the same name as the source file
and the extension .XRF. If you supply a cross-reference file name
without an extension,.XRF is appended to it. (TCREF, a crossreference utility, is described Oil disk.)
If you want to accept the default object file name and also request
a listing file, you must supply the comma that separates the object
file name from the listing file name:
TASM FILEl" TEST
This assembles FILEl.ASM to FILE1.0BJ and creates a listing file
named TEST.LST.
If you want to accept the default object and listing file names and
also request a cross-reference file, you must supply the commas
that separate the file names:
Chapter 3, Command-line reference
23
TASM MYFILE",MYXREF
This assembles file MYFILE.ASM to MYFILE.OBJ, with a listing in
file MYFILE.LST and a cross-reference in MYXREF.XRF.
If you use wildcards to specify the source files to assemble, you
can also use wildcards to indicate the object and listing file names.
For example, if your current directory contains XXI.ASM and
XX2.ASM, the command line
TASM XX*,YY*
assembles all the files that start with XX, generates object files that
start with W, and derives the remainder of the name from the
source file ·name. The resulting object files are therefore called
YYI.OBJ and YY2.0BJ.
If you don't want an object file but you do want a listing file, or if
you want a cross-reference file but don't want a listing file or
object file, you can specify the null device (NUL) as the file name.
For example,
TASM FlLEl"NUL,
assembles file FILE1.ASM to object file FILE1.0BJ, doesn't
produce a listing file, and creates a cross-reference file FILEI.XRF.
Command-line options
The command-line options let you control the behavior of the
assembler, and how it outputs information to the screen, listing,
and object file. Turbo Assembler provides you with some options
that produce no action, but are accepted for compatibility with the
current and previous versions of MASM:
Ib
Iv
Sets buffer size
Displays extra statistics
You can enter options using any combination of uppercase and
lowercase letters. You can also enter your options in any order
except where you have multiple II or Ij options; these are
processed in sequence. When using the Id option, you must also
be careful to define symbols before using them in subsequent Id
options.
Note: You can override command-line options by using
conflicting directives in your source code.
24
Turbo Assembler User's Guide
la
Figure 3.1 on page 22 summarizes the Turbo Assembler
command-line options; here's a detailed description of each
option.
/a
Function
Syntax
Remarks
Specifies alphabetical segment-ordering
fa
The la option tells Turbo Assembler to place segments in the object file in
alphabetical order. This is the same as using the .ALPHA directive in your
source file.
You usually only have to use this option if you want to assemble a source
file that was written for very early versions of the IBM or Microsoft
assemblers.
The Is option reverses the effect of this option by returning to the default
sequential segment-ordering.
If you specify sequential segment-ordering with the .SEQ directive in
your source file, it will override any la you provide on the command line.
Example
TASM / a TEST!
This command line creates an object file, TEST1.0BJ, that has its segments
in alphabetical order.
/b
Syntax
Remarks
fb
The Ib option is included for compatibility. It performs no action and has
no effect on the assembly.
Chapter 3, Command-line reference
25
Ie
Ie
Function
Syntax
Remarks
Enables cross-reference in listing file
Ie
The Ie option enables cross-reference information in the listing file. Turbo
Assembler adds the cross-reference information to the symbol table at the
end of the listing file. This means that, in order to see the cross-reference
information, you must either explicitly specify a listing file on the
command line or use the II option to enable the listing file.
For each symbol, the cross-reference shows the line on which it is defined
and all lines that refer to it.
Example
TASM /1 /e TEST!
This code creates a listing file that also has cross-reference information in
the symbol table.
Id
Function
Syntax
Remarks
Defines a symbol
Idsymbol [=value or expression]
The Id option defines a symbol for your source file, exactly as if it were
defined on the first line of your file with the directive. You can use this
option as many times as you want on the command line.
=
You can only define a symbol as being equal to another symbol or a
constant value. You can't use an expression with operators to the right of
the equal sign (=). For example, IdX=9 and IdX= Yare allowed, but
IdX= Y-4 is not allowed.
Example
TASM /dMAX=lO /dMIN=2 TESTl
This command line defines two symbols, MAX and MIN, that other
statements in the source file TESTl.ASM can refer to.
26
Turbo Assembler User's Guide
Ie
/e
Function
Syntax
Remarks
Generates floating-point emulator instructions
Ie
The Ie option tells Turbo Assembler to generate floating-point instructions
that will be executed by a software floating-point emulator. Use this
option if your program contains a floating-point emulation library that
mimics the functions of the 80x87 numeric coprocessor.
Normally, you would only use this option if your assembler module is
part of a program written in a high-level language that uses a floatingpoint emulation library. (Turbo C, Turbo Pascal, Turbo Basic, and Turbo
Prolog all support floating-point emulation.) You can't just link an
assembler program with the emulation library, since the library expects to
have been initialized by the compiler's startup code.
The Ir option reverses the effect of this option by enabling the assembly of
real floating-point instructions that can only be executed by a numeric
coprocessor.
If you use the NOEMUL directive in your source file, it will override the Ie
option on the command line.
The Ie command-line option has the same effect as using the EMUL
directive at the start of your source file, and is also the same as using the
IjEMUL command-line option.
Example
TASM / e SECANT
TCC -f TRIG.C SECANT.OBJ
The first command line assembles a module with emulated floating-point
instructions. The second command line compiles a C source module with
floating-point emulation and then links it with the object file from the
assembler.
/h or /?
Function
Syntax
Remarks
Displays a help screen
/h or /?
The Ih option tells Turbo Assembler to display a help screen that describes
the command-line syntax. This includes a list of the options, as well as the
various file names you can supply. The 11 option does the same thing.
Chapter 3, Command-line reference
27
Ih or I?
Example
TASM /h
Function
Sets an Include file path
Ii
Syntax
Remarks
/iPATH
The II option lets you tell Turbo Assembler where to look for files that are
included in your source file by using the INCLUDE directive. You can
place more than one II option on the command line (the number is only
limited by RAM).
When Turbo Assembler encounters an INCLUDE directive, the location
where it searches for the Include file is determined by whether the file
name in the INCLUDE directive has a directory path or is just a simple file
name.
If you supply a directory path as part of the file name, that path is tried
first, then Turbo Assembler searches the directories specified by II
command-line options in the order they appear on the command line. It
then looks in any directories specified by II options in a configuration file.
If you don't supply a directory path as part of the file name, Turbo
Assembler searches first in the directories specified by II command-line
options, then it looks in any directories specified by II options in a
configuration file, and finally it looks in the current directory.
Example
TASM / i \ INCLUDE / iD: \INCLUDE TEST!
If the source file contains the statement
INCLUDE MYMACS.INC
Turbo Assembler will first look for \ INCLUDE \ MYMACS.INC, then it
will look for D:\INCLUDE\MYMACS.INC. If it still hasn't found the file,
it will look for MYMACS.INC in the current directory. If the statement in
your source file had been
INCLUDE INCS\MYMACS.INC
Turbo Assembler would first look for INCS\MYMACS.INC and then it
would look for \INCLUDE\MYMACS.INC, and finally for D:\
INCLUDE\MYMACS.INC.
28
Turbo Assembler User's Guide
Ij
/j
Function
Syntax
Remarks
Defines an assembler startup directive
/jdirective
The Ij option lets you specify a directive that will be assembled before the
first line of the source file. directive can be any Turbo Assembler directive
that does not take any arguments, such as .286, IDEAL, %MACS,
NOJUMPS, and so on. See Chapter 3 in the Reference Guide for a complete
description of all Turbo Assembler directives.
I
You can put more than one Ij option on the command line; they are
processed from left to right across the command line.
Example
TASM /j .286 /jIDEAL TEST!
This code assembles the file TEST1.ASM with 80286 instructions enabled
and Ideal mode expression-parsing enabled.
/kh
Function
Syntax
Remarks
Sets the maximum number of symbols allowed
/khnsymbols
The Ikh option sets the maximum number of symbols that your program
can contain. If you don't use this option, your program can only have a
maximum of 8,192 symbols; using this option increases the number of
symbols to nsymbols, up to a maximum of 32,768.
Use this option if you get the Out of hash space message when assembling
your program.
You can also use this option to reduce the total number of symbols below
the default 8,192. This releases some memory that can be used when you
are trying to assemble a program but don't have enough available
memory.
Example
TASM /khlOOOO BIGFILE
This command tells Turbo Assembler to reserve space for 10,000 symbols
when assembling the file BIGFILE.
Chapter 3, Command-line reference
29
/ks
/ks
Function
Syntax
Sets the maximum size of Turbo Assembler's string space
/kskbytes
Remarks
Usually the string size is determined automatically and does not need to
be adjusted. However, if you have a source file that results in an Out of
string space message, you might want to increase the string space size by
using this option. Try starting with a value of 100, and increase it until
your program assembles without error. The maximum allowable value for
kbytes is 255.
Example
TASM Iksl50 SFILE
This tells Turbo Assembler to reserve 150K of string space.
/1
Function
Syntax
Generates a listing file
Ii
Remarks
The II option indicates that you want a listing file, even if you did not
explicitly ~pecify it on the command line. The listing file will have the
same name as the source file, with an extension of .LST.
Example
TASM 11 TEST!
This command line requests a listing file that will be named TEST1.LST.
/10
Function
Syntax
30
Shows high-level interface code in listing file
/la
Remarks
The Iia option tells Turbo Assembler to show all generated code in the
listing file, including the code that gets generated as a result of the highlevel language interface .MODEL directive.
Example
TASM Ila FILEl
Turbo Assembler User's Guide
1m
1m
Function
Syntax
Remarks
Sets the maximum number of assembly passes
Im[npasses]
Normally, Turbo Assembler functions as a single-pass assembler. The 1m
option allows you to specify the maximum number of passes that the
assembler should make during the assembly process. TASM automatically
decides whether it can perform less than the number of passes specified. If
you don't specify npasses, a default of five is used.
Some modules contain constructions that assemble properly only when
two passes are done. If multiple passes are not enabled, such a module
will produce at least one "Pass-dependent construction encountered"
warning. If the 1m option is enabled, Turbo Assembler will assemble this
module correctly but will not optimize the code by removing NOPs, no
matter how many passes are allowed. The warning IIModule is pass
dependent-<:ompatibility pass was done" is displayed if this occurs.
Example
TASM 1M2 TESTl
This tells Turbo Assembler to use up to two passes when assembling
TESTl.
Iml
Function
Syntax
Treats symbols as case-sensitive
Iml
Remarks
The Iml option tells Turbo Assembler to treat all symbol names as casesensitive. Normally, uppercase and lowercase letters are considered
equivalent so that the names ABCxyz, abcxyz, and ABCXYZ would all refer
to the same symbol. If you specify the Iml option, these three symbols will
be treated as distinct. Even when you specify Iml, you can still enter any
assembler keyword in uppercase or lowercase. Keywords are the symbols
built into the assembler that have special meanings, such as instruction
mnemonics, directives, and operators.
Example
TASM Iml TESTl
where TESTl.ASM contains the following statements:
abc OW 0
ABC OW 1
Chapter 3. Command-line reference
inot a duplicate symbol
31
/mu
Mov Ax,!Bp]
imixed case OK in keywords
/mu
Function
Syntax
Converts symbols to uppercase
/mu
Remarks
The Imu option tells Turbo Assembler to ignore the case of all symbols. By
default, Turbo Assembler specifies that any lowercase letters in symbols
will be converted to uppercase unless you change it by using the Iml
directive.
Example
TASM
/mu TEST!
makes sure that all symbols are converted to uppercase (which is the
default):
EXTRN myfunc:NEAR
call myfunc
idon't know if declared as
MYFUNC, Myfunc, ••.
i
/mv#
Function
Syntax
Sets the maximum length of symbols.
tmv'
Remarks
The Imv# option sets the maximum length of symbols that TASM will
distinguish between. For example, if you set Imv3, TASM will see ABCC
and ABCD as the same symbol, but not AB.
Function
Makes public and external symbols case-sensitive
/mx
Syntax
Remarks
32
/rnx
The Imx option tells Turbo Assembler to treat only external and public
symbols as case-sensitive. All other symbols used (within the source file)
are treated as uppercase.
Turbo Assembler User's Guide
/mx
You should use this directive when you call routines in other modules
that were compiled or assembled so that case-sensitivity is preserved; for
example, modules compiled by Turbo C.
Example
TASM Irnx TEST1;
where TESTl.ASM contains the following source lines:
EXTRN Cfunc:NEAR
rnyproc PROC NEAR
call Cfunc
In
Function
Syntax
Remarks
Suppresses symbol table in listing file
In
The In option indicates that you don't want the usual symbol table at the
end of the listing file. Normally, a complete symbol table listing appears at
the end of the file, showing all symbols, their types, and their values.
You must specify a listing file, either explicitly on the command line or by
using the II option; otherwise, In has no effect.
Example
TASM II In TEST!
This code generates a listing file showing the generated code only, and not
the value of your symbols.
Ip
Function
Syntax
Remarks
Checks for impure code in protected mode
Ip
The Ip option specifies that you want to be warned about any instructions
that generate "impure" code in protected mode. Instructions that move
data into memory by using a CS: override in protected mode are
considered impure because they might not work correctly unless you take
special measures.
You only need to use this option if you are writing a program that runs on
the 80286 or 80386 in protected mode.
Example
TASM Ip TEST!
Chapter 3. Command-line reference
33
/p
where TEST1.ASM contains the following statements:
.286P
CODE SEGMENT
temp OW ?
mov CS:temp,O
;impure in protected mode
/q
Function
Syntax
Suppresses .OBI records not needed for linking
/q
Remarks
The Iq option removes the copyright and file dependency records from
the resulting .OBI files, making it smaller. Don't use this option if you are
using MAKE or a similar program that relies on the dependency records.
Function
Generates real floating-point instructions
/r
Syntax
Remarks
Ir
The Ir option tells Turbo Assembler to generate real floating-point
instructions (instead of generating emulated floating-point instructions).
Use this option if your program is going to run on machines equipped
with an 8Ox87 numeric coprocessor.
The Ie option reverses the effect of this option in generating emulated
floating-point instructions.
If you use the EMUL directive in your source file, it will override the Ir
option on the command line.
The Ir command-line option has the same effect as using the NOEMUL
directive at the start of your source file, and is also the same as using the
IjNOEMUL command-line option.
Example
TASM /r SECANT
TPC /$N+ /$E- TRIG. PAS
The first command line assembles a module with real floating-point
instructions. The second compiles a Pascal source module with real
floating-point instructions that links in the object file from the assembler.
34
Turbo Assembler User's Guide
/s
Is
Function
Syntax
Remarks
Specifies sequential segment-ordering
Is
The Is option tells Turbo Assembler to place segments in the object file in
the order in which they were encountered in the source file. By default,
Turbo Assembler uses segment-ordering, unless you change it by placing
an Is option in the configuration file.
If you specify alphabetical segment-ordering in your source file with the
.ALPHA directive, it will override Is on the command line.
Example
TASM
Is
TEST!
This code creates an object file (TEST1.0BJ) that has its segments ordered
exactly as they were specified in the source file.
It
Function
Syntax
Remarks
Suppresses messages on successful assembly
It
The It option stops any display by Turbo Assembler unless warning or
error messages result from the assembly.
You can use this option when you are assembling many modules, and you
only want warning or error messages to be displayed onscreen.
Example
TASM
It TEST!
Iv
Syntax
Remarks
Iv
The Iv option is included for compatibility. It performs no action and has
no effect on the assembly.
Chapter 3, Command-line reference
35
/w
/w
Function
Syntax
Con troIs the generation of warning messages
/w
w- [warnclass]
w+ [warnclass]
Remarks
The Iw option controls which warning messages are emitted by Turbo
Assembler.
If you specify Iw by itself, "mild" warnings are enabled. Mild warnings
merely indicate that you can improve some aspect of your code's
efficiency.
If you specify Iw- without warnclass, all warnings are disabled. If you
follow Iw- with warnclass, only that warning is disabled. Each warning
message has a three-letter identifier:
ALN
ASS
BRK
ICG
LCO
OPI
OPP
OPS
OVF
POC
PQK
PRO
RES
TPI
Segment alignment
Assuming segment is 16-bit
Brackets needed
Inefficient code generation
Location counter overflow
Open IF conditional
Open procedure
Open segment
Arithmetic overflow
Pass-dependent construction
Assuming constant for [const] wa~ing
Write-to memory in protected mode needs CS override
Reserved word warning
Turbo Pascal illegal warning
If you specify Iw+ without warnclass, all warnings are enabled. If you
specify Iw+ with warnclass from the preceding list, only that warning will
be enabled.
By default, Turbo Assembler first starts assembling your file with all
warnings enabled except the inefficient code-generation (ICG) and the
write-to-memory in protected mode (PRO) warnings.
You can use the WARN and NOWARN directives within your source file to
control whether a particular warning is allowed for a certain range of
source lines. See Chapter 3 in the Reference Guide for more information on
these directives.
36
Turbo Assembler User's Guide
/w
Example
TASM Iw TEST!
The following statement in TEST1.ASM issues a warning message that
would not have appeared without the Iw option:
mov bx,ABC
ABC = 1
;inefficient code generation warning
With the command line
TASM Iw-OVF TEST2
no warnings are generated if TEST2.ASM contains
dw lOOOh
* 20h
Ix
Function
Syntax
Remarks
Includes false conditionals in listing
Ix
If a conditional IF, IFNDEF, IFDEF, and so forth evaluates to False, the Ix
option causes the statements inside the conditional block to appear in the
listing file. This option also causes the conditional directives themselves to
be listed; normally they are not.
You must specify a listing file on the command line or via the II option,
otherwise Ix has no effect.
You can use the .LFCOND, .SFCOND, and .TFCOND directives to override
the effects of the Ix option.
Ix TESTl
Example
TASM
Function
Displays source lines along with error messages
Iz
Syntax
Remarks
Iz
The Iz option tells Turbo Assembler to display the corresponding line
from the source file when an error message is generated. The line that
caused the error is displayed before the error message. With this option
disabled, Turbo Assembler just displays a message that describes the
error.
Chapter 3, Command-line reference
37
/z
Example
TASM / z TEST!
Function
Enables line-number information in object files
/zd
Syntax
Remarks
/zd
The Izd option causes Turbo Assembler to place line-number information
in the object file. This lets Borland's stand-alone debugger, Turbo
Debugger, display the current location in your source code, but does not
put the information in the object file that would allow the debugger to
access your data items.
If you run out of memory when trying to debug your program under
Turbo Debugger, you can use Izd for some modules and Izi for others.
Example
TASM / zd TEST!
Function
Enables debug information in object file
/zi
Syntax
Remarks
/zi
The Izi option tells Turbo Assembler to output complete debugging
information to the object file. This includes line-number records to
synchronize source code display and data type information to allow you
to examine and modify your program's data.
The Izi option lets you use all the features of Turbo Debugger to step
through your program and examine or change your data items. You can
use Izi on all your program's modules, or just on those you're interested in
debugging. Since the Izi switch adds information to the object and
executable programs, you might not want to use it on all your modules if
you run out of memory when running a program under Turbo Debugger.
Example
38
TASM /zi TEST!
Turbo Assembler User's Guide
Indirect command files
At any point when entering a command line, Turbo Assembler
lets you specify an indirect command file by preceding its name
with an "at" sign (@). For example,
TASM /dTESTMODE @MYPROJ.TA
causes the contents of the file MYPROJ.TA to become part of the
command line, exactly as if you had typed in its contents directly.
This useful feature lets you put your most frequently used
command lines and file lists in a separate file. And you don't have
to place your entire command line in one indirect file, since you
can use more than one indirect file on the command line and can
also mix indirect command files with normal arguments; for
example,
TASM @MYFILES @IOLIBS /dBUF=1024
This way you can keep long lists of standard files and options in
files, so that you can quickly and easily alter the behavior of an
individual assembly run.
You can either put all your file names and options on a single line
in the command file or you can split them across as many lines as
you want.
The configuration file
Turbo Assembler also lets you put your most frequently used
options into a configuration file in the current directory. If
running on DOS 3.x or later, it also looks in the directory that
TASM was loaded from. This way, when you run Turbo
Assembler, it looks for a file called TASM.CFG in your current
directory. If Turbo Assembler finds the file, it treats it as an
indirect file and processes it before anything else on the command
line.
This is helpful when you have all the source files for a project in a
single directory and you know that, for example, you always
want to assemble with emulated floating-point instructions (the Ie
option). You can place that option in the TASM.CFG file, so you
don't have to specify that option each time you start Turbo
Assembler.
Chapter 3, Command-line reference
39
The contents of the configuration file have exactly the same
format as an indirect file. The file can contain any valid
command-line options, on as many lines as you want. The options
are treated as if they all appeared on one line.
The contents of the configuration file are processed before any
arguments on the command line. This lets you override any
options set in the configuration file by simply placing an option
with the opposite effect on the command line. For example, if
your configuration file contains
la Ie
and you invoke Turbo Assembler with
TASM Is Ir MYFILE
where MYFILE is your program file, your file will be assembled
with sequential segment-ordering (Is) and real floating-point
instructions (lr), even though the configuration file contained the
la and Ie options that specified alphabetical segment-ordering and
emulated floating-point instructions.
40
Turbo Assembler User's Guide
c
H
A
p
T
E
R
4
The nature of assembly language
Earlier, we said that assembly language is the computer's own
language. In order to understand what that means, you first need
to understand exactly what a computer is. Then we'll teach you
just what it is that makes assembly language unique among the
many languages of the computer world.
In this chapter we'll cover the nature of computers in general, and
the 8086 processor in particular, to give you an understanding of
the special strengths of assembly language programming on the
8086. We'll also discuss issues of assembly language
programming specifically related to the IBM PC.
The architecture of a computer
Deep down, a computer is nothing more than a device that moves
data from one place to another, sometimes transforming the data
in various logical and arithmetical ways. For our purposes,
however, it's more useful to view a computer as a system
consisting of five functional subsystems-input, control,
arithmetic and logical processing, memory, and output-as
shown in Figure 4.1.
(For the moment, we're talking about computers in general; we'll
get to the 8088 shortly.)
Chapter 4, The nature of assembly language
41
Figure 4.1
Five subsystems
Arithmetic Subsystem
(Add, subtract,
multiply, divide,
and, or,
exclusive-or, etc.)
Input
Subsystem
(Keyboard,
Mouse,
Joystick,
Disk)
Control Subsystem
(Overall Coordination)
Output
Subsystem
(Display,
Printer,
Plotter,
Disk)
Memory Subsystem
(Up to 1 megabyte of
RAM and/or ROM)
The arithmetic subsystem of the computer is the aspect most
people think of when they think ofa computer. After all, what is a
computer if not a number-cruncher? As it turns out, though, most
computers spend very little time crunching numbers, and a great
deal of time working with character strings and performing input
and output; what need does a word processor have for
arithmetic? Nonetheless, the arithmetic subsystem is important,
for it is there that not only addition, subtraction, multiplication,
and division are performed, but logical operations (such as and,
or, and exclusive-or) as well.
It's all very well to perform arithmetic, but where do the source
values for, say, addition come from, and where does the result of
each operation go? The computer's memory subsystem comes into
play here, providing instantly accessible storage for many
thousands of characters or numbers. Computers also have floppy
and hard disk drives, which provide permanent (but relatively
slow) storage for enormous amounts of data, but these are
actually input/output (I/O) devices, not part of the memory
subsystem.
42
Turbo Assembler User's Guide
Programs without Input and
output tend to be rare, since
they can't accept new data
(rom the outside world and
can't do anything with whatever results they do
generate.
The input subsystem allows programs to manipulate data from
the outside world, ranging from single keystrokes to mouse
motions to whole databases stored in disk files. The output
subsystem lets programs display prompts and results on screens
and printers, and send data to disk files and tapes.
Finally, the control subsystem ties together the operation of the
other four subsystems and controls data movement.
The control and arithmetic subsystems together form what is
known as the processing unit, or processor. A processor forms the
core of any computer, providing data processing and controlling
the memory, input, and output subsystems. The processor sets the
tone for any computer, since it controls the operation of each of
the subsystems and coordinates them into a smoothly functioning
unit.
Nowadays, an entire processor is frequently built on a single chip.
For instance, the 8088 is a processor on a chip, complete with
arithmetic processing, control, and interfaces to input, output, and
memory.
It's with the processor that we make the connection between the
architecture of the computer and the unique nature of assembly
language.
The making of
assembly
language
We've said that the processor orchestrates the activities of the five
subsystems of a computer-adding values, moving them about
from memory to output, and so on-but that begs a fundamental
question: How does the processor know which operations to
perform? So far, the computer has all the capabilities we need, but
no script to follow.
The answer is surprisingly simple: The processor fetches data
from memory, and that data tells it what to do next. Data that tells
a processor what to do is usually called "instructions," but
instructions are simply values stored in memory, just like any
other data. The set of instructions that a processor can execute
(the instruction set) corresponds exactly to the actions that that
processor's hardware can perform. Put another way, a processor's
instructions comprise all the operations that any software can
ever ask the processor to do.
Chapter 4, The nature of assembly language
43
For example, if a processor lacks a multiplication instruction, then
there is no way the hardware of that computer can perform a
multiplication. Multiplication can instead be performed in
software by perfonning adds and shifts, but this tends to be much
slower. The key point here is that a processor's instruction set
reflects the actions that the computer's hardware is inherently
capable of perfonning. By the same token, each processor's
assembly language is unique to that processor because each
processor has unique capabilities and, therefore, a unique
instruction set.
Each instruction value has a specific, well-defined meaning to a
given processor. For example, the instruction value 4 tells the 8088
to add the value stored at the next memory address to the AL
register. (Don't worry about what the AL register is right nowwe'll get to that soon.) Consequently, a processor can be put
through a desired sequence of actions by an appropriate series of
instruction values; indeed, a program is nothing more than a
sequence of instruction values.
How does a processor know which instruction to execute next? By
maintaining an internal pointer that points to the place in
memory where the value of the next instruction to be performed
is stored. When that next instruction is read from memory and
executed, the pointer is advanced to the following instruction.
Some instructions can set the instruction pointer to a new value;
this gives a processor the ability to execute nonsequential series of
instructions, and even the ability to perform different series of
instructions depending on certain conditions.
Great, but what does that have to do with assembly language?
Just this: A processor's instruction set is that processor's assembly
language. Or, rather, assembly language is a human-oriented form
of a processor's instruction set (known as the processor's machine
language), which an assembler such as Turbo Assembler then
converts to machine language. While assembly language and
machine language are functionally equivalent, assembly language
is much easier for people to program in. After all, surely you'd
rather program with instructions like
add al,l
than with
4
1
44
Turbo Assembler User's Guide
wouldn't you? Both forms work equally well, but assembly
language lets you work with mnemonic names for machinelanguage instructions, with the assembler translating from
mnemonic instructions to their machine-language equivalents.
This is, of course, a tremendous advantage, since humans simply
don't think very well in purely numeric languages. Basically,
assembly language is a direct analog to machine language, but
implemented in a form with which humans can work efficiently.
The good news about assembly language is that it lets you control
the processor's actions one by one, for maximum efficiency. The
bad news is that each of the processor's actions, taken
individually, tend to do relatively little, reflecting the limited
repertoire of which the processor is actually capable. For example,
the process of adding two long integers and storing the result in a
third long integer takes only one line in C:
i
= j + k;
but requires six lines in 8088 assembler:
mov ax, [j]
mov dx, [j+2]
add
adc
mov
mov
ax, [k]
dx, [k+2]
[i],ax
[i+2] ,dx
Of course, the C code compiles to no less (and possibly more)
than the same six machine language instructions required by the
assembler code, but it is easier to write the one line of C code than
the six lines of assembler. (Remember, assembler instructions
reflect the basic ability of the computer, and programs written in
all languages must eventually be translated to machine language
before they can be run.)
Why use assembler at all if it's harder to program in than other
languages? For one thing, assembler lets you reach any part of
memory and control any input or output device directly, since
assembly language programs can do anything the processor is
capable of. For another, because assembler is the native language
of the computer, it stands to reason that well-written assembler
code must be the fastest code possible. The quality of the code
produced by any other language suffers from the need to translate
from that language to machine language, but assembler code
maps directly to machine language, with not one whit of
Chapter 4, The nature of assembly language
45
efficiency lost. In assembly language, you tell the computer what
to do, and it does it-no more and no less.
Of course, if you write an inefficient assembler program, it won't
run very rapidly, since the processor does exactly what assembly
language programs specify. Similarly, assembly language has
relatively little built-in support for data-type conversion, or fc;>r
guarding against mistakes, such as accidentally overwriting a
variable or running off the end of an array. What all this means is
that assembly language gives you the ability to write wonderfully
fast and clever programs, but those programs demand more care
and skill from you as a programmer than do programs written in
other languages.
Now that you understand how a processor and its assembly
language relate to one another, let's look specifically at assembly
language for the 8088.
The 8088 and 8086 processors
The 8088 is the processor used in the IBM PC and XT computers,
which form perhaps one of the most successful line of computers.
However, the 8088 is actually only one of a family of processors
known as the iAPx86 family. Other members of this family
include the 8086 processor used in the IBM Models 25 and 30; the
80286 processor used in the IBM AT, and the IBM PS/2 Models 50
and 60; and the 80386 processor used in the IBM PS/2 Model 80.
Each of these processors is, in some way, different from the 8088.
Chapter 10, liThe 80386 and other processors," provides a detailed
discussion of the various members of the iAPx86 family. The one
thing all iAPx86 family processors share is the ability to run code
written for the 8086 and 8088 processors.
The 8086 is actually the root of the iAPx86 family tree. The 8088 is
just an 8086 with a scaled-down external data bus; while the 8086
can transfer data to and from memory 16 bits at a time, the 8088
can transfer data only 8 bits at a time. The two processors have
exactly the same instruction sets. Consequently, the assembly
language used to program the IBM PC and its successors is
properly known as 8086 assembly language, not 8088 assembly
language. For the remainder of this chapter, understand that 8086
assembly language includes the 8088 as well.
46
Turbo Assembler User's Guide
The capabilities of
the 8086
The 8086 runs at 4.77 or 8 MHz
speeds; the 80286 can run at
6, 8, 10, 12, 16, and 20 MHz:
the 80386 can run at 16, 20,
25, and 33 MHz.
By today's standards, the 8086 is a processor of modest
capabilities. After all, the 8086 was designed ten years ago, and
ten years of technological evolution have brought major
innovations to the chip-design field. Nonetheless, the 8086
remains an important processor. One reason for this is the sheer
number of IBM PCs and compatibles; no one can afford to ignore
ten-million-plus computers. Another reason, however, is that the
8086 meets the needs, even today, of advanced software.
The 8086 can address a large amount of memory (over one million
characters or other byte-sized-8-bit-values), has a powerful
instruction set, and properly programmed can support highperformance programs. But the 8086 is not a super-fast processor,
not every language is capable of providing decent performance on
the 8086, and no other language can match assembly language
when it comes to writing excellent 8086 programs.
The resources the 8086 provides to the assembly language
programmer are memory, input and output (I/O) interfacing,
registers, and, of course, instructions. We'll explore those
resources next.
Memory
The 8086 is capable of addressing 1 megabyte (which is 2 to the
20th power or 1,048,576 storage locations, each of which is 8 bits
long) of memory at anyone time. The first byte of memory is at
address 0, and the last byte of memory "is at address OFFFFFh as
shown in Figure 4.2 on page 48.
(The last address, OFFFFFb, was given in hexadecimal, or base 16,
notation as denoted by the h suffix; it is equivalent to 1,048,575 in
the familiar decimal, or base 10, notation.) Fluency in hexadecimal
notation is essential in assembly language programming. We'll
touch on hexadecimal notation in Chapter 5, liThe elements of an
assembler program."
Chapter 4, The nature of assembly language
47
Figure 4.2
Memory address
space of the 8086
Hexadecimal
Address
00000
00001
00002
00003
00004
00005
00006
00007
00008
00009
OOOOA
OOOOB
OOOOC
00000
Decimal
Address
o
1
2
3
4
5
6
7
8
9
10
11
12
13
OOOOE
OOOOF
15
FFFEF
FFFFO
FFFF1
FFFF2
FFFF3
FFFF4
FFFF5
FFFF6
FFFF7
FFFF8
FFFF9
FFFFA
FFFFB
FFFFC
FFFFD
FFFFE
FFFFF
1048559
1048560
1048561
1048562
1048563
1048564
1048565
1048566
1048567
1048568
1048569
1048570
1048571
1048572
1048573
1048574
1048575
00010
14
16
One byte, 8 bits long, can hold one character, or one integer value
in the range 0 to 255. That doesn't mean that the 8086 can't handle
larger values. Two bytes taken together (known as a word) can
hold one integer value in the range 0 to 65,535; the 8086 can
manipulate word values as readily as byte values.
Four bytes taken together (known as a doubleword, or dword) can
hold one integer value in the range 0 to 4,294,967,295, or can hold
one single-precision floating-point value. Eight bytes together
48
Turbo Assembler User's Guide
(known as a quadword, or qword) can hold one double-precision
floating-point value. The 8086 doesn't handle these two data types
directly; however, the 8087 numeric coprocessor can work directly
with floating-point values and extended precision integer values,
and given the proper software, the 8086 can be made to handle
virtually any data type, albeit fairly slowly.
At any time, an 8086 program can read or change the contents of
any of the more than 1,000,000 bytes of memory. For example, the
code fragment
mov
mov
mov
mov
ax,O
ds,ax
bx,O
aI, [bx]
loads the contents of the byte at address 0 into the AL register.
Don't worry about the details here; the point is that the 8086's
memory address space provides for storage of slightly more than
1,000,000 working values that the 8086 can access quickly and
flexibly.
One megabyte (Mb) is a considerable amount of memory, far
more than the 64K (2 to the 16th power, or 65,536 bytes)
addressable by the processors that preceded the 8086. On the
other hand, the 8086's latest descendent, the 80386, can address
about 4,000 times as much memory as the 8086, so you can see
that the 8086 is, in fact, a little squeezed for memory space. Also,
in the IBM PC, only 640K of the 1 Mb address space is actually
available for use as general-purpose memory; the rest of the
address space is dedicated to use by system software and the
memory used for the display. Then, too, don't forget that
instructions, as well as data, are stored in memory, so both
program code and data must fit into no more than 640K of
memory on the PC.
While the 8086 is capable of addressing 1 Mb of memory, it does
not make it particularly easy to access more than 64K at anyone
time, due to a rather peculiar feature known as segmentation. We'll
look at segmentation in a later section, liThe segment registers,"
on page 61.
Chapter 4, The nature of assembly language
49
Input and output
The 8086 supports input and output devices in two ways: through
input/output (I/O) instructions and through memory addresses.
Some input and output devices are controlled through ports,
which are special I/O addresses in a 64K address space that's
separate from the 1 Mb memory address space, as shown in
Figure 4.3.
Figure 4.3
Separate memory
and I/O address of
8086
Memory
Address
I/O Address
.(fQr1l
00009
0000
0001
0002
0003
0004
0005
0006
0007
0008
0009
OOOOA
OOOA
00000
00001
00002
00003
00004
00005
00006
00007
00008
~
~
FFFF5
FFFF6
FFFF7
FFFF8
FFFF9
FFFFA
FFFFB
FFFFC
FFFFD
FFFFE
FFFFF
1---------1
J.--------1
1---------1
J.--------1
1---------1
1---------1
I . - - _ _ _ _- - - - l
FFF5
FFF6
FFF7
FFF8
FFF9
FFFA
FFFB
FFFC
FFFD
FFFE
FFFF
J.--------i
J.--------i
I---------i
J.--------i
J.--------i
I---------i
J.--------i
I---------i
I---------i
1 . - -_ _ _ _----1
There are far fewer I/O addresses on the 8086 than there are
memory addresses; while there are technically 64K I/O addresses
on the PC, practically speaking, only 4K I/O addresses are
available. Consequently, I/O addresses are not used for storing
values, but rather for providing control and data channels to
input and output devices. For example, serial devices such as
modems are controlled entirely through a few I/O addresses.
50
Turbo Assembler User's Guide
I/O addresses can be accessed only with two special instructions,
IN and OUT, which are used for nothing else. For example,
out dx,al
More on IN and our, and I/O
In general, In Chapter 5.
sends the contents of the AL register to the I/O port selected by
the DX register.
Some output devices are memory-mapped, meaning they are
controlled through normal memory addresses rather than I/O.
This is particularly true of display adapters, which can take up
16K, 32K, or even 256K of the 8086's memory address space with
their bit maps (the arrays of bytes describing the dots that the
.
adapters display on the screen).
A given device can be controlled through both I/O ports and
memory-mapped addresses. In fact, most display adapters
respond to I/O instructions for some functions and to memory
addresses for others.
Registers
The 8086 offers a few fast, on-chip storage elements known as
registers. You might think of registers as memory locations that
the 8086 can access faster than it can access regular memory, but
that's only part of what makes registers special. Each of the
registers has a unique nature, and provides certain capabilities
that no other register or memory location supports.
The registers fall into four categories: the flags register, the
general-purpose registers, the instruction pointer, and the
segment registers, as shown in Figure 4.4. Let's look at each in
turn.
Chapter 4, The nature of assembly language
51
Figure 4.4
Registers of the
8086
Bit Number
15
FLAGS
0
I
sub ex,l
jnz BeginningOfLoop
Don't worry about unfamiliar elements of this program; the
important point is that the instructions between the label
BeginningOfLoop and the JNZ instruction are executed repeatedly
until ex becomes o. Notice that two instructions-SUB eX,1 and
JNZ BeginningOfLoop-are required in order to count down CX
and jump back to BeginningOfLoop if ex is not yet o.
Chapter 4, The nature of assembly language
55
Counting down and looping is a frequently used program
element, so the 8086 provides a special instruction to make loops
faster and more compact. Not surprisingly, that instruction is
called LOOP. The LOOP instruction subtracts 1 from ex: and
jumps if CX isn't 0, all in one instruction. The following is
equivalent to the last example:
...
mov cx,lO
BeginningOfLoop:
loop BeginningOfLoop
We'll cover looping again in Chapter 5; for now, just remember
that the ex: register is especially useful for counting and looping.
The OX register
The DX register can be
treated as two 8-blt registers,
DHandDL.
The OX register is the only register that can be used as an I/O
address pointer with the IN and OUT instructions. In fact, there is
no way to address I/O ports 256 through 65,535 without using
OX. For example, the following code writes the value 62 to I/O
port 1000:
mov al,62
mov dx,lOOO
out dx,al
The other unique properties of OX relate to division and
multiplication. When you divide a 32-bit dividend by a 16-bit
divisor, the upper 16 bits of the dividend must be placed in OX;
after the division, the remainder of the division is stored in OX.
(The lower 16 bits of the dividend must be placed in AX, and the
quotient is stored in AX.) Similarly, when you multiply two 16-bit
factors, the upper 16 bits of the product are stored in OX (the
lower 16 bits of the product are stored in AX).
The SI register
Like the BX register, the SI register can be used as a memory
pointer. For example,
56
Turbo Assembler User's Guide
mov
mov
mov
mov
ax,O
ds,ax
si,20
aI, [si]
loads the 8-bit value stored at address 20 into AL. SI becomes an
unusually powerful memory pointer when used with the 8086's
string instructions. For example,
cld
mov ax,O
mov ds,ax
mov si,20
lodsb
not only loads AX with the value at the memory address pointed
to by SI, but also adds 1 to SI. This can be very effective when
accessing a sequential series of memory locations, such as a text
string. Better still, the string instructions can be made to
automatically repeat their actions any number of times, so a single
instruction can perform hundreds or even thousands of actions.
We'll discuss the string instructions in detail in Chapter 6.
The 01 register
The 01 register is much like the SI register in that it can be used as
a memory pointer and has special properties when used with the
powerful string instructions. For example,
mov
mov
mov
add
ax,O
ds,ax
di,1024
bl, [di]
adds the 8-bit value stored at address 1024 to BL. The 01 register
is a little different from SI when it comes to string instructions;
where SI always serves as a source memory pointer for string
instructions, 01 always serves as a destination memory pointer.
Moreover, with the string instructions, SI normally addresses
memory relative to the OS segment register, while 01 always
addresses memory relative to the ES segment register. (When SI
Chapter 4, The nature of assembly language
57
and 01 are used as memory pointers with non string instructions,
they always point relative to OS.) For example,
cld
mov dx,O
mov es,dx
mov di,2048
stosb
uses the SlOSB string instruction to both store the value in AL at
the memory address pointed to by 01 and add 1 to OI. But we're
getting ahead of ourselves here; you need to learn about segments
and segment registers before you can study the string instructions. Again, we'll look at the string instructions in Chapter 6,
''More about programming in Turbo Assembler."
The BP register
Like BX, 51, and 01, the BP register can be used as a memory
pointer, but with a difference. While the BX, 51, and 01 registers
normally act as memory pointers relative to the OS segment
register (or, in the case of OJ used with the string instructions, the
ES segment register), BP points relative to 55, the stack segment
register.
Chapter 7 explaIns how and
why C uses the stack to pass
parameters.
Once again, we're getting ahead of ourselves, since we haven't
covered segments yet, but the principle is as follows: One useful
way to pass parameters to a subroutine is by pushing the
parameters onto the stack. C and Pascal do this.
The stack resides in the segment pointed to by 55, or the stack
segment. Data, on the other hand, normally resides in the
segment pointed to by OS, or the data segment. Since BX, 51, and
OJ normally point to the data segment, there's no efficient way to
use BX, 51, or OJ to point to parameters passed on the stack
because the stack is usually in a different segment altogether.
BP solves this problem by providing addressing into the stack
segment. For example,
...
push bp
mov bp,sp
mov ax, [bp+4]
58
Turbo Assembler User's Guide
accesses the stack segment to load AX. with the first parameter
passed by a Turbo C call to an assembler subroutine.
In short, BP is designed to provide support for parameters, local
variables, and other stack-based memory-addressing needs.
The SP register
The SP register, also known as the stack pointer, is the least general
of the general-purpose registers, for it is almost always dedicated
to a specific purpose: maintaining the stack. The stack is an area of
memory into which values can be stored and from which they can
be retrieved on a last-in, first-out basis; that is, the last value
stored onto the stack is the first value you'll get when you read a
value from the stack. The classic analogy for the stack is that of a
stack of dishes. Since you can only add plates at the top of the
stack and remove them from the top of the stack, it stands to
reason that the first plate you put on the stack will be the last
plate you can remove.
The SP register points to the top of the stack at any given time; as
with the stack of dishes, the top of the stack is the location at
which the next value placed on the stack will be stored. The action
of placing a value on the stack is known as pushing a value on the
stack, and, indeed, the PUSH instruction is used to place values on
the stack. Similarly, the action of retrieving a value from the stack
is known as popping a value from the stack, and the POP
instruction is used to retrieve values from the stack.
For example, Figure 4.6 illustrates how SP, AX, and BX change as
the following code is executed, assuming that SP is initially set to
1,000:
mov
push
mov
push
pop
pop
ax,l
ax
bx,2
bx
ax
bx
While the 8086 allows you to store values in SP, and add to or
subtract from the value stored in SP, just as with the other
general-purpose registers, you should never do this unless you
know exactly what you're doing. If you change SP, you're
Chapter 4. The nature of assembly language
59
changing the location of the top of the stack, and that can quickly
lead to disaster.
Why? Well, pushes and pops aren't the only way the stack is used.
Whenever you call to or return from a subroutine (a procedure or
function), the stack is used. Also, some system resources, such as
the keyboard and the system clock, use the stack when they
interrupt the 8086 in order to perform their functions. What this
means is that the stack might be needed at any time. If you change
SP, even if only for a few instructions, then the correct stack might
not be available when some system resource needs it.
Figure 4.6
AX. BK SP. and the
stack
At start:
AX
?
996
BX
?
998
SP
1000
. - - - - - - ' r1000
After moy ax,1 I push ax:
AX
1
I
ex
?
SP
998
W
996
....--_r 998
1000
After moy bx,21 push bx:
AX
1
BX
2
SP
996
After pop ax:
U
....---- 996
AX
2
I
BX
2
SP
998
W
998
1000
996
....---- 998
1000
After pop bx:
60
AX
2
996·
?
BX
1
998
.1
SP
L..--_ _ _ _ _....J
~------~-~1000
?
1000
Turbo Assembler User's Guide
In short, leave SP alone unless you know just what you're doing.
Feel free to perform pushes, pops, calls, and returns, but don't
change the value of SP directly. Any of the other seven generalpurpose registers can be changed directly at any time.
The instruction pointer
The instruction pointer (lP) always contains the memory offset at
which the next instruction to be executed is stored. As one
instruction is executed, the instruction pointer is advanced to
point to the instruction at the next memory address. Normally,
the instruction at the next memory address is the next instruction
executed, but some instructions, such as calls and jumps, can
cause the instruction pointer to be loaded with a new value,
thereby branching to other code.
The instruction pointer can't be written to or read from directly;
only branching instructions such as those just described can load
the instruction pointer with a new value.
The instruction pointer does not, by itself, fully specify the
address at which the next instruction to be executed resides. Once
again, the segmented nature of 8086 memory addressing
complicates the picture. For instruction fetching, the CS segment
register provides a base address, and the instruction pointer then
provides an offset from that base address.
Each time we've talked about addressing memory, we've run into
segments, and each time we've postponed a full explanation until
the time came to talk about segments. That time has come.
The segment registers
Now we come to a most unusual aspect of the 8086-memory
segmentation. The basic premise of segmentation is this: The 8086
is capable of addressing 1 Mb of memory. Twenty-bit memory
addresses are required to address all locations in a 1 Mb memory
space. However, the 8086 only uses 16-bit pointers to memory; for
example, recall that the 16-bit BX register can be used to point to
memory. How, then, does the 8086 reconcile 16-bit pointers with a
20-bit address space?
The answer is that the 8086 uses a two-part memory-addressing
scheme. True, 16-bit memory pointers are used, but these form
only part of the full memory address. Each 16-bit memory pointer,
or memory offset, is combin~d with the contents of a 16-bit
segment register to form a full20-bit memory address.
Chapter 4, The nature of assembly language
61
Segments and offsets are combined as follows: The segment value
is shifted left by 4 bits (multiplied by 16) and then added to the
offset as shown in Figure 4.7.
Figure 4.7
20-blt memory
addresses
16-81t
Segment Register
16-81t
Offset
Segment Value
Times 16 Equals
A 20-81t Value
20-81t Memory Address
So, for example, consider the following code:
...
mov
mov
mov
mov
ax,lOOOh
ds,ax
si,201h
dl, lsi]
Here the OS segment register is set to 1000h, and SI is set to 201h,
which we can represent as the segment:offset pair 1000:201h.
(Segment:offset calculations can only be performed efficiently in
base 16-another good reason to familiarize yourself with
hexadecimal notation.) DL is loaded from the address
«DS ,. 16) + S1), or «1000h ,. 16) + 201h):
62
Turbo Assembler User's Guide
l000h
x
16
10000h
+ 201h
10201h
Figure 4.8 illustrates this example.
Figure 4.8
Calculation of
memory address by
mov
os
1000h
SI
201h
10000h
Memory
Address
10201h
Another way to look at this is to simply shift the segment value
left 4 bits, or one hexadecimal digit, which is the same as
multiplying by 16:
10000
+ 201
10201
You can now see that programs can only access the 8086's full
1 Mb memory space by using segment:offset pairs. In fact, you
must always use segment:offset pairs to access memory; all the
instructions and addressing modes of the 8086 default to
operating relative to one or another of the segment registers,
Chapter 4. The nature of assembly language
63
although some instructions can be explicitly told to use a different
segment register if desired.
Rarely will you actually load a number into a segment register.
Instead, you'll load segment registers with segment names, which
are turned into numbers in the course of assembling, linking, and
running a program. This is necessary because there's no way to
tell beforehand where in memory a given segment will reside; it
all depends on the version of DOS, the number and size of
memory-resident programs, and the memory needs of the rest of
the program. Using segment names lets Turbo Assembler and
DOS deal with all those complications.
The most common segment name is @data, which refers to the
default data segment when the simplified segment directives are
used. For example,
.MODEL small
•DATA
Varl DW 0
.CODE
mov ax,@data
mov ds,ax
END
loads OS to point to the default data segment, in which Varl
resides.
Once again, we're getting a bit ahead; in the next chapter, we'll
discuss the simplified segment directives and the loading of
segment registers.
The use of segments on the 8086 has a couple of interesting
implications. For one thing, only a 64K block of memory is
addressable relative to a segment register at anyone time because
64K is the maximum amount of memory that can be addressed
with a 16-bit offset. This means that it can be a real nuisance to
handle large (greater than 64K) blocks of data on the 8086, since
both a segment register and the offset value must be changed
frequently.
The addressing of large blocks of memory on the 8086 is made
still more difficult because, unlike the general-purpose registers,
the segment registers cannot serve as either source or destina tion
for arithmetic and logical instructions. In fact, the only operations
that can be performed on segment registers involve copying
64
Turbo Assembler User's Guide
values between segment registers and either general-purpose
registers or memory. For instance, adding 100 to the ES register
requires the following:
mov ax,es
add ax,lOO
mov es,ax
The upshot of all this is that the 8086 is best suited to handling
memory in chunks no larger than 64K.
A second implication of the use of segments is that any given
memory location is addressable with many possible
segment:offset combinations. For instance, the memory address
100h is addressable with segment:offset values of 0:100h, l:FUh,
2:EOh, and so on, since all those segment:offset pairs work out to
address 100h.
Like the general-purpose registers, each segment register plays a
specific role. The es register points to program code, the OS
register points to data, the SS register points to the stack, and the
ES segment is a wildcard ("extra") segment, free to point
wherever it's needed. Let's look at the segment registers in a bit
more detail.
The CS register
The es register points to the start of the 64K memory block, or
code segment, in which the next instruction to be executed resides.
The next instruction to be executed resides at the offset specified
by IP in the code segment; that is, at the segment:offset address
eS:IP. The 8086 can never fetch an instruction from a segment
other than that defined byeS.
The es register can be changed by a number of instructions,
including certain jumps, calls, and returns. The CS register cannot
be loaded directly under any circumstances.
No memory-addressing modes or memory pointers other than IP
normally operate relative to es.
Chapter 4, The nature of assembly language
6S
The OS register
Memory addressing Is
discussed further In Chapter
5.
The OS register points to the start of the data segment, which is
the 64K memory block where most memory operands reside.
Normally, memory offsets involving BX, SI, or DI operate relative
to OS, as do direct memory addresses. The data segment is,
basically, what its name implies: the segment in which the current
data set normally resides.
The ES register
The ES register points to the start of a 64K memory block known
as the extra segment. As the name implies, the extra segment isn't
dedicated to anyone purpose, but is available for whatever needs
arise. Sometimes, the extra segment is used to make an additional
64K block of memory available for data storage, but accessing
memory in the extra segment is normally less efficient than
accessing memory in the data segment, as discussed in Chapter 9,
"Advanced programming in Turbo Assembler."
Where the extra segment really shines is when the string
instructions are used. All string instructions that write to memory
use ES:DI as the memory address to write to. This means that ES
is extremely useful as the destination segment for block copies,
string comparisons, memory scanning, and clearing blocks of
memory. We'll look at the string instructions and the use of ES
registers in connection with them in Chapter 6, "More about
programming in Turbo Assembler."
The 55 register
The SS register points to the start of the stack segment, which is
the 64K memory block, where the stack resides. All instructions
that implicitly use the SP register-including pushes, pops, calls,
and returns-work in the stack segment because SP is only
capable of addressing memory in the stack segment.
As we discussed earlier, the BP register also operates relative to
the stack segment. This allows BP to be used for addressing
parameters and variables that are stored on the stack. (Again, we
discuss memory addressing in detail in the next chapter.)
66
Turbo Assembler User's Guide
The 8086
instruction set
To a programmer, the key resource of the 8086 is the instruction
set. As we discussed earlier, the instruction set includes all the
actions that a programmer can possibly tell the 8086 to perform.
(The complete instruction set of Turbo Assembler is in the Quick
Reference Guide.)
There are many instructions in the 8086 instruction set that
perform a wide variety of actions, ranging from doing nothing
(NaP) to copying as many as 65,535 bytes (REP MOVSB). We will
spend much of the rest of this chapter, and chapters 5, 6, and 9 as
well, covering the 8086's instruction set in detail.
The IBM PC and XT
We've focused on 8086 assembly language, but the truth of the
matter is that the 8086 processor is just part of a computer system,
and the hardware configuration and operating system of a
computer greatly affect assembly language programming.
The vast majority of programs written for the 8086 processor (and
perhaps the majority of programs written in the history of
computers) have been written for the IBM PC and XT and
compatible computers, running the MS-DOS operating system.
Turbo Assembler itself runs under the MS-DOS operating system
on IBM PCs, XTs, and compatibles (from now on referred to
simply as the IBM PC), so it's likely that you're planning to use
your copy of Turbo Assembler to develop assembler programs for
the IBM PC environment.
Without knowledge of the hardware configuration and the
operating system your assembler programs will run under, there's
no way for you to perform input or output, or even terminate
your programs. We haven't the space to cover nearly all the
capabilities of the IBM PC and its system software, but we'll show
you a few of the basic features of the PC. We suggest you read
more on your own in the books and manuals suggested at the
beginning of this chapter.
Chapter 4, The nature of assembly language
67
Input and output
devices All IBM PCs provide a keyboard, a display adapter and a monitor,
and a floppy disk drive. Modems, printers, mice, and hard disks
are frequently installed as well. Each of these devices is controlled
with a fairly complex series of accesses to I/O ports or memory
(or both). For example, selecting a new video mode on the Color
Graphics Adapter (CGA) requires over 30 OUT instructions;
keyboard, modem, and disk control sequences are more
complicated still.
Does this mean that you need to master endless control sequences
in order to write useful assembler programs on the IBM PC? Not
at all; your PC's systems software already does most of the work
for you.
Systems software
for the IBM PC Systems software is software that serves as a control and interface
layer between applications software, such as Turbo Assembler
and Quattro, and the hardware of your computer, as shown in
Figure 4.9.
In particular, systems software handles the complexities of
interfacing to individual devices. For example, several hundred
lines of assembly language code are required in order for your PC
to process a single keystroke, but your assembler programs can
get keystrokes by invoking just one system function. This is made
possible by the two main systems software components of the PC:
OOS and the BIOS (Basic Input/Output System).
In Figure 4.9, the OOS and BIOS systems software serves as a
control and interface layer between applications software and the
hardware of the IBM PC. Applications software always has the
option of controlling the hardware directly, but should use 005
or BIOS functions instead whenever possible.
68
Turbo Assembler User's Guide
Figure 4.9
DOS and BIOS
systems software as
a control and
interface layer
Applications Software
DOS
Accessed through Int 21 h DOS
functions and other Interrupts.
BIOS
Accessed through BIOS
functions by way of several
Interrupts.
IBM PC Hardware
-----------------------------------------------Display adapter, keyboard, printer, disk, mouse, joystick, and so
on. Accessed at UO ports and/or memory locations, depending
on the specific hardware Item.
DOS
DOS (short for Disk Operating System-also known as MS-DOS
and PC-DOS) is the program that controls your computer from
the moment it reads the disk at power-up until you turn the
power off. DOS takes up part of your precious 640K of available
memory, but there's no helping that, since without DOS your PC
is a very expensive paperweight. It's DOS that provides you with
the A> prompt (or C>, or whatever the prompt is on your
computer), and it's DOS that accepts and executes commands
such as DIR.
That's just the visible part of DOS. It also provides a broad array
of functions that are used heavily by just about every application.
Chapter 4, The nature of assembly language
69
It's through DOS functions that applications read from and write
to files, get keystrokes, allocate memory, run other programs, and
even set and get the time of day. For example, the assembler code
rnov ah,2
rnov dl,'A'
int 21h
;DOS function to display a character
;A is the character to display
;invoke DOS to execute the function
invokes the OOS "Display Output" function in order to display
the character A at the current cursor location on the screen.
IBM's DOS Technical
Reference manual Is the
primary reference for DOS
functions.
You should use OOS functions to perform operations such as
keyboard and file input, screen and file output, and printing
whenever possible. Since DOS itself is actually nothing but an
assembler program, it is certainly possible for you to do with your
own code everything that OOS functions do, but that's generally
not a good idea. Not all PC-compatible computers are alike, and
OOS frequently masks differences between makes of computers;
if you ignore the OOS functions and go straight to the hardware,
your programs might not run on other computers.
Then, too, programs that go around OOS might not coexist with
other programs, most notably memory-resident programs such as
SideKick and SuperKey. Besides, why spend time writing extra
code when OOS has already done the work for you? In short,
whenever a OOS function can do what you need done, use it!
In cases where OOS simply doesn't provide the functions you
need, it's time to use a BIOS function. We'll cover BIOS functions
shortly, but first let's take a look at some OOS functions that fulfill
essential needs: input, output, and program termination.
Getting keystrokes
Typing at the keyboard is the fundamental means of user
interaction with the PC. OOS provides a number of functions by
which an assembler program can obtain keystrokes; we're only
going to discuss one of those functions.
Perhaps the simplest means of getting keystrokes is with the
"Keyboard Input" function, DOS function number 1. OOS
functions are invoked by placing the function number in AH and
then executing an INT 21 h instruction. (The actual operation of the
INT instruction is a bit complex, but right now, all you need to
know is that you must execute an INT 21h instruction each time
70
Turbo Assembler User's GuIde
you want to invoke a DOS function.} The next character typed at
the keyboard is returned in AL.
For example, when this code is executed,
mov ah,l
int 21h
...
DOS places the next character typed at the keyboard into AL.
Note that if there is no keystroke waiting to be read, DOS waits
until a key is pressed, so this function can take an indefinitely
long period of time to complete.
Displaying characters on the screen
If keystrokes are the means of user interaction with software, the
screen is the complement. The PC is capable of all sorts of
displays, ranging from color text to high-resolution graphics, but
for the moment, we'll just go over displaying characters.
DOS function number 2 is a straightforward way to print a
character. Simply put 2 in AH and the character in DL, then
invoke DOS with INT 21 h. The following code echoes each
character typed to the screen:
mov
int
mov
mov
int
ah,l
21h
ah,2
dl,al
21h
;get next key pressed
;move character read from AL to DL
idisplay the character
Several other functions are available for reading and printing
characters and character strings, and you'll encounter some of
them in the example programs in this manual. Since a whole book
would be needed to cover all the DOS functions, we can't cover
them here. We strongly recommend, however, that you do get one
or more of the books and manuals listed at the end of this book
and learn more about the DOS functions-they're a key resource
in assembler programming.
There's one more point we'd like to make about keyboard, screen,
and file input and output in assembly language. Those of you
who are used to scant and prlntf in C and Readln and Writeln in
Pascal might be surprised to learn that DOS (and hence assembly
Chapter 4. The nature of assembly language
71
language) provides no support whatsoever for fonnatted input
and output; OOS only handles character and string input and
·output. In C, all you need to do to print an integer variable i is
this:
printf("%d\n",i);
C automatically converts the integer value, which is stored in a
16-bit memory location, into a string of ASCII characters and
prints the characters. In assembler, your code must explicitly
convert variables to character strings before displaying them.
Likewise, OOS only knows how to read characters and strings
from the keyboard, so you'll have to write code to convert
characters and strings entered by the user to other data types in
your assembler programs.
At the end of the next chapter, we'll show you an example
program that illustrates exactly what you have to do in an
assembler program to print out the value of a variable. For now,
bear in mind that DOS functions can print a character, or a string
of characters-and that's it. It's up to you to convert your data to
the character fonn that DOS can handle.
Ending a program
Now that you know a bit about reading and writing a program,
let's write a simple program that does nothing but echo one line of
keystrokes to the screen. You know all the OOS functions you'll
need, save one: You have no way to end the program once it's
finished executing.
Again, those of you familiar with C or Pascal might think that
assembler programs would simply end when they come to the
end of the main program, but that's not the case. You must
explicitly invoke a DOS function in order to terminate your
assembler programs.
There are several DOS functions for terminating programs, but
the preferred method is to execute a DOS function number 4Ch
(that's 76, for those of you who prefer decimal). With that
knowledge, here's the complete echo program
(ECHOCHAR.ASM):
.MODEL small
.STACK lOOh
•CODE
EchoLoop:
72
Turbo Assembler User's Guide
mov ah,l
int 21h
cmp al,13
jz EchoDone
mov dl,al
mov ah,2
int 21h
jmp EchoLoop
EchoDone:
mov ah,4ch
int 21h
END
iDOS keyboard input function f
iget the next key
iwas the key the Enter key?
iyes, so we're done echoing
iput the character into DL
iDOS display output function
idisplay the character
iecho the next character
iDOS terminate program function f
iterminate the program
Enter the program exactly as shown and run it. You'll see that
each character you type appears twice; once when it is echoed by
DOS as it's typed, and once as your program echoes it. The
important point about this program is that it reads keystrokes,
writes characters to the display, and terminates, all by way of
DOS functions.
The BIOS
IBM's BIOS Interface
Technical Reference manual
is the primary reference for
BIOS functions.
Sometimes DOS functions just don't meet your needs; then it's
time to turn to the PC's Basic Input/Output System, or BIOS.
Unlike DOS and applications software, the BIOS is not loaded
from disk and does not take up any of your 640K of available
memory; instead, the BIOS is stored in Read-Only Memory
(ROM) in the portion of the 8086's address space reserved for
system functions.
The BIOS is the lowest-level software in the PC; even DOS uses
BIOS functions to control the hardware. It's better to use BIOS
functions than to control hardware directly, since, like DOS, the
BIOS can mask differences between various computers and
devices. On the other hand, you should use DOS functions rather
than BIOS functions whenever you can, since programs that use
the BIOS can conflict with other programs, and tend to be less
portable across a variety of computer models.
Selecting display modes
The most pressing reason to use the BIOS is for controlling the
display, since DOS provides virtually no support for the rich
display capabilities of the PC. Only by invoking BIOS functions
can you set the screen mode, control colors, get display adapter
information, and so on. For example, the following code invokes
the BIOS to set the screen to four-color graphics mode on a CGA:
Chapter 4, The nature of assembly language
73
mov ah,O
mov al,4
int lOh
;BIOS set mode function f
;mode number for 320x200 4-color graphics
;execute BIOS video interrupt to set mode
If you recall that we said that over 30 OUT instructions are
required to set a video mode, you'll realize that the BIOS "Set
Mode" function saves you a great deal of work.
The BIOS provides a variety of functions other than those related
to display control, including keystroke-handling and disk control.
In general, however, you're better off performing these tasks
through DOS functions.
Sometimes you
absolutely need
to go to the
hardware
Now that you've heard all the reasons to use DOS functions (or, if
absolutely necessary, BIOS functions), it's time to admit that
sometimes you just flat-out have to access the hardware directly.
For instance, communications software has to control the PC;s
serial port directly with IN and OUT instructions, since neither
DOS nor the BIOS provides adequate support for serial
communications. Similarly, high-performance graphics must be
performed by accessing display memory directly, since DOS
doesn't support graphics, and the BIOS does so only in a painfully
slow manner.
The basic rule about going to the hardware is to make sure you
have no alternative. If there's a DOS or BIOS function you can use,
use it; if not, access the hardware directly. Mer all, the object-of
programming is to produce useful programs, not to follow rules.
On the other hand, the fewer rules you break, the fewer problems
you'll generally have.
Other resources
The PC provides a number of other hardware and software
resources for the assembly language programmer. We can't go
into those resources here, but we can list a few; for more
information, refer to the materials mentioned at the start of this
chapter.
• The ANSI.SYS driver provides enhanced display control
without the need for BIOS functions.
74
Turbo Assembler User's Guide
• The system timers support a time-of-day clock; they also
support sound-generation via the PC's speaker and precision
timing.
• The optional 8087 numeric coprocessor speeds up floating-point
calculations by orders of magnitude.
Chapter 4, The nature of assembly language
75
76
Turbo Assembler User's Guide
c
H
A
p
T
E
R
5
The elements of an assembler program
Now that you understand what it is that makes assembly
language unique, you're ready to tackle the nuts and bolts of
assembler programming.
You'll spend this chapter learning about the fundamental
components of an assembler program. First, we'll teach you about
the minimum requirements of a working assembler program.
Next, we'll discuss the various components of a line, and the
ways in which they can be combined. Along the way, you'llieam
a good bit about instructions, directives, and the ways in which
assembler programs can access memory. You'll find out how
segments are defined and used in Turbo Assembler, and you'll
. look at the allocation and initialization of memory variables.
Finally, we'll look at some commonly used instructions.
That's a lot of ground to cover, but when you're done with this
chapter, you'll know enough to start writing programs. You can,
put that knowledge to work with a word-counting program
provided at the end of the chapter.
Still, this chapter only begins to explore the many aspects of
assembly language, so Chapter 6, ''More about programming in
Turbo Assembler," and Chapter 9, Advanced programming in
Turbo Assembler," continue on to new assembly language topics.
1/
Chapter 5, The elements of an assembler program
77
The components and structure of an assembler
program
Now that you've developed an understanding of what 8086
assembly language is, you're ready to start writing assembler
programs. Let's start by looking at the minimum requirements of
a working assembler program. Even a simple assembler program
requires quite a few lines. For instance, consider the following
program:
•MODEL small
.STACK 200h
•DATA
DisplayString DB 13,10
ThreeChars
DB
3 DUP
DB
'$'
(?)
•CODE
Begin:
mov ax,@data
mov ds,ax
mov bx,OFFSET ThreeChars
mov
int
dec
mov
inc
ah,1
21h
al
[bx],al
bx
int 21h
dec al
mov [bx),al
inc bx
int 21h
dec al
mov [bx],al
mov dx,OFFSET DisplayString
mov
int
mov
int
78
ah,9
21h
ah,4ch
21h
inear code and data models
i512-byte stack
istart of the data segment
icarriage-return/linefeed pair
i to start a new line
istorage for three characters
i typed at the keyboard
ia trailing "$" to tell DOS when
i to stop printing DisplayString
i when function 9 is executed
istart of the code segment
ipoint DS to the data segment
ipoint to the storage location
i for first character
iDOS keyboard input function f
;get the next key pressed
;subtract 1 from the character
istore the modified character
;point to the storage location
i for the next character
;get the next key pressed
;subtract 1 from the character
istore the modified character
ipoint to the storage location
i for the next character
iget the next key pressed
isubtract 1 from the character
istore the modified character
ipoint to the string of
i modified characters
iDOS print string function f
iprint the modified characters
iDOS end program function f
iend the program
Turbo Assembler User's Guide
END Begin
;directive to mark the end of the source
; code and to indicate where to start
; execution when the progra~ is run
This program contains the simplified segment directives .MODEL,
.STACK, .DATA, and .CODE, as well as the END directive. Segment
directives, either simplified or standard, are required in every
assembler program in order to define and control segment usage,
and the END directive must always terminate assembler code.
We'll cover both segment directives and END in this chapter, and
some other directives as well.
Directives only provide the framework for an assembler program,
though; you also need lines in your source code that actually do
something, lines like
mov
(bx],al
and
inc dx
These are instruction mnemonics, corresponding to the
instruction set of the 8086 that you learned about in chapter 4.
Before you can use either instructions or directives, however, you
must first learn about the format of a line of assembler code,
which we'll get to right after a cursory look at Turbo Assembler's
reserved words.
In case you were wondering what the first example program
does, enter it, type in IBM, and the program will respond
HAL
The program reads three characters, subtracts the value 1 from
each of them, and prints the result.
Reserved words
Turbo Assembler reserved words, or keywords, are strictly for
use by the assembler; you can't use them for defining your own
equates, labels, or procedure names. Rather, you should think of
reserved words as the building blocks of assembly language. The
words listed in Table 5.1 include operators (+, *, -, +), directives
(.386, ASSUME, MASM, QUIRKS), and predefined symbols
(??tlme, ??verslon, @WordSlze), which are like predefined
equates, and aliases.
Chapter 5, The elements of an assembler program
79
Table 5.1: TASM reserved words
=
?
[]
I
()
+
*
.186
.286
.286C
.286P
.287
.386
.386C
.387
.8086 .
.8087
ALIGN
.ALPHA
AND
ARG
ASSUME
%BIN
BYTE
CATSTR
@Code
CODESEG
@CodeSize
COMM
COMMENT
%CONDS
.CONST
CONST
@Cpu
%CREF
.CREF
%CREFALL
%CREFREF
%CREFUREF
%CTLS
@Curseg
@data
•DATA
.DATA?
DATAPTR
DATASEG
80
@datasize
??date
DB
DO
%0 EPTH
OF
DISPLAY
DOSSEG
DP
DQ
DT
DUP
OW
DWORD
ELSE
ELSEIF
EMUL
END
ENDIF
ENDM
ENDP
ENDS
EQ
EQU
ERR
.ERR
.ERR1
.ERR2
.ERRB
.ERRDEF
ERRDIF
ERRDIFI
ERRE
ERRIDN
ERRIDNI
ERRIFNB
ERRIFNDEF
ERRNB
ERRNDEF
ERRNZ
EVEN
EVEN DATA
EXITM
EXTRN
FAR
FAR DATA
@fardata
.FARDATA
@fardata?
•FA RDATA?
@filename
??fllename
FWORD
GE
GLOBAL
GROUP
GT
HIGH
IDEAL
IF
IF1
IF2
IFB
IFDEF
IFDIF
IFDIFI
IFE
IRON
IRDNI
IFNB
IFNDEF
%lNCL
INCLUDE
INCLUDELIB
INSTR
IRP
IRPC
JUMPS
LABEL
.LALL
LARGE
LE
LENGTH
.LFCOND
%LlNUM
%LlST
.LlST
LOCAL
LOCALS
LOW
LT
MACRO
%MACS
MASK
MASM
MASM51
MOD
MODEL
.MODEL
MULTERRS
NAME
NE
NEAR
%NEWPAGE
%NOCONDS
%NOCREF
%NOCTLS
NOEMUL
%NOINCL
NOJUMPS
%NOLIST
NOLOCALS
%NOMACS
NOMASM51
NOMULTERRS
NOSMART
%NOSYMS
NOT
NOTHING
%NOTRUNC
NOWARN
OFFSET
OR
ORG
%OUT
P186
P286
P286N
P287
P386
P386N
P386P
P387
P8086
P8087
PAGE
%PAGESIZE
PARA
%PCNT
PN087
%POPLCTL
PROC
PTR
PUBLIC
PURGE
%PUSHLCTL
PWORD
QUIRKS
QWORD
RADIX
.RADIX
RECORD
REPT
.SALL
SEG
SEGMENT
.SEQ
.SFCOND
SHL
SHORT
SHR
SIZE
SIZESTR
SMALL
SMART
STACK
.STACK
STRUC
SUBSTR
SUBTTL
%SUBTTL
%SYMS
SYMTYPE
°kTABSIZE
TBYTE
%TEXT
.TFCOND
THIS
??time
TITLE
%TITLE
%TRUNC
TYPE
.TYPE
UDATASEG
UFARDATA
UNION
UNKNOWN
USES
??version
WARN
WIDTH
WORD
@WordSize
.XALL
.XCREF
.XLlST
XOR
Turbo Assembler User's Guide
The format of a line
Assembly language source code lines follow this format:
where is an optional symbolic name;
is either the mnemonic for an instruction or a directive;
contains a combination of zero, one, or two (or
sometimes more) constants, memory references, register
references, and text strings, as required by the particular
instruction or directive; <;comment> is an optional comment.
A backslash (\) can be placed almost anywhere as a linecontinuation character. It cannot be used to break up strings or
identifiers. The backslash means "read the next line in at this
point and continue processing." This way you can use it naturally
without losing the ability to comment each line the way you like.
For example,
foo mystructure
<0,
1,
2>
\
\
\
iStart of structure fill.
iZero value is first.
iOne value.
iTwo value and end of structure.
There are contexts where the line-continuation character is not
recognized. In general, it isn't recognized in any context where
characters are treated as text rather than identifiers, numbers, or
strings, or in MASM mode when the line continuation is used in
the first two symbols in the statement. For example,
ifdif <123\>,<456\>
does not recognize the two enclosed line-continuation characters.
comment \
begins a comment block, but does not define a near symbol called
COMMENT.
The line-continuation character is also not recognized inside of
macro definitions. It is recognized, however, when the macro is
expanded.
Let's look more closely at each of these elements of assembly
language code.
Chapter 5, The elements of an assembler program
81
Labels
Labels are nothing more than names used for referring to
numbers and character strings or memory locations within a
program. Labels let you give names to memory variables, values,
and the locations of particular instructions. For example, the
following code, which calculates five factorial (1 x 2 x 3 x 4 x 5 =
120), uses several labels:
. MODEL small
.STACK 200h
•DATA
FactorialValue ow ?
Factorial
OW?
•CODE:
FiveFactorial PROC
mov ax,@data
mov ds,ax
mov [FactorialValue],l
mov [Factorial],2
mov cx,4
FiveFactorialLoop:
mov ax, [Factorial Value]
mul [Factorial]
mov [FactorialValue],ax
inc [Factorial]
loop FiveFactorialLoop
ret
FiveFactorial ENDP
END
The labels FactorialValue and Factorial are equivalent to the
addresses of two 16-bit variables; they're used to refer to those
two variables later in the code. The label FiveFactorial is the name
of the subroutine (or function or procedure) containing the code,
allowing other parts of this program to call this code. Finally, the
label FiveFactorialLoop is equivalent to the address of the
instruction
mov
ax, [FactorialValue]
so that the LOOP at the end of the code can branch back to that
particular instruction.
Labels can consist of the following characters:
A-Z
82
a-z
@
$
?
0-9
Turbo Assembler User's Guide
A period (.) is also allowed in MASM mode (discussed in Chapter
11), as the first character only. The digits 0-9 cannot be used as the
first character of a label. A single $ or ? has a special meaning, so
neither can be used as a user symbol name.
Each label must be defined only once; that is, labels must be
unique. (There are exceptions to this rule; for example, special
labels defined with the directive and local labels in macros and
Ideal mode subroutines.) Labels can be used as operands any
number of times.
=
A label can appear on a line by itself, that is, on a line without an
instruction or directive. In this case, the value of the label is the
address of the instruction or directive on the next line in the
program. For instance, in the code
jrnp
DoAddition:
add
DoAddition
ax,dx
the next instruction executed after the JMP instruction, which
branches to the label DoAddition, is ADD AX,DX. The preceding
example is exactly the same as
ADD AX,DX Is forced to the
right because of the length
of DoAdditlon, making for
less readable code.
jmp
DoAddition:
DoAddition
add
ax,dx
There are two advantages to putting each label on its own line.
First, when you put each label on its own line, it's easier to use
long labels without messing up the format of your assembler
source code. Second, it's easier to add a new instruction right at a
label if the label's not on the same line as an instruction. To
convert the last example to
jrnp
DoAddition:
add
DoAddition
rnov dx, [MemVar]
ax,dx
you would have to split DoAddition from ADD AX,DX and then
add the new text. By contrast, if DoAddition were on a line by itself
Chapter 5. The elements of an assembler program
83
(as in the earlier example), you could simply add a new line after
DoAddition and be done with it.
Chapter 3 In the Reference
Guide lists aI/ directives. The
registers of the 8086 are listed
In Chapter 4.
A label cannot be the same as any of the built-in symbols used in
expressions. This includes the register names (AX, BX, and so on),
and the operators used in expressions (PTR, BYTE, WORD, and so
on). You also cannot use any of the IFxxx directives or .ERRxxx
directives as label names. A few other symbols reserved by Turbo
Assembler can only be used in certain contexts: These include
NAME, INCLUDE, and COMMENT, which can be used as structure
member names but not as general-purpose symbols. (Refer to
Chapter 9 for more about structures.)
A safe approach is to avoid using any of the built-in symbol
names for your labels. As an example, the labels
bx DW
PTR:
a
would be unacceptable, since BX is a register and PTR is an
expression operator. However, the label
Old BX DW
a
would be fine.
The following are examples of acceptable labels:
MainLoop
calc_long_sum
ErrorO
iterate
Draw$Dot
Delay_lOa_milliseconds
Both labels that appear on lines without directives or instruction
mnemonics and labels that appear on lines with instructions must
end with a colon. The colon merely ends the label, and is not part
of the label itself. For example, in
LoopTop:
Done:
84
mov
inc
and
jz
jmp
ret
aI, [sil
si
al,al
Done
LoopTop
Turbo Assembler User's Guide
the labels LoopTop and Done are defined with colons, but
references to those labels do not use colons.
Other labels generally should not have colons. The example code
at the start of this section provides several instances of labels
without colons.
Make your labels meaningful. Contrast
cmp al,'a'
jb
NotALowerCaseLetter
cmp aI,' z'
ja
NotALowerCaseLetter
sub al,20h
NotALowerCaseLetter:
;convert to uppercase
and
cmp
jb
cmp
ja
sub
al,'a'
Pl
al,'z'
Pl
al,20h
;convert to uppercase
Pl:
The version with descriptive labels is largely self-documenting,
while the second version is cryptic, to say the least. Labels can
also contain underscores; if you prefer, you can use labels like
noCa_lower_case_letter or Not_A_Lower_Case_Letter. It's purely a
matter of taste.
Instruction
mnemonics and The key field in a line of assembler code is the field. This field can contain either an instruction
mnemonic or a directive, two very different beasts.
You've encountered instruction mnemonics earlier in this chapter;
they're the human-readable names for the machine-language
instructions the 8086 executes directly. MOV, ADD, MUL, and JMP
are all instruction mnemonics, corresponding directly to the data
movement, addition, multiplication, and branching instructions of
the 8086.
Chapter 5, The elements of an assembler program
85
Turbo Assembler assembles each instruction mnemonic directly
to the corresponding machine-language instruction. Whenever
you insert one instruction mnemonic in an assembler program,
the result is one corresponding machine-language instruction in
the executable code.
Directives are quite the opposite of instruction mnemonics: They
generate no executable code at all, but rather control various
aspects of how Turbo Assembler operates, from the type of code
assembled (8086,80286,80386, and so on), to the segments used,
to the way in which listing files are generated. Although the
distinction blurs at times, you might think of instruction
mnemonics as generating the actual 8086 machine-language
program, while directives are responsible for providing high-level
features of Turbo Assembler that make assembly language
programming easier.
We will spend much of this manual teaching you about the
various instruction mnemonics and directives provided by Turbo
Assembler, all of which are discussed in Chapter 3 of the Reference
Guide as well. There are a few directives that you'll need in every
program you write, most notably the segment directives, which
we'll cover in a section later in this chapter called "Segment
directives" on page 103. Another directive you'll always need is
the END directive, which we'll look at next.
The END directive
Each and every program must contain an END directive to mark
the end of the program's source code. Any lines following an END
directive are ignored by Turbo Assembler. If you omit the END
directive, an error is generated; you might think that the end of
the file would mark the end of the program, but not so-an END
directive is always required.
END is typical of directives in general in that it generates no code.
For example,
.M)DEL small
•STACK 200h
•CODE
ProgramStart:
mov ah,4ch
int 21h
END ProgramStart
is perhaps the simplest possible assembler program, doing
nothing more than immediately returning to DOS. Note the use of
86
Turbo Assembler User's Guide
the END directive to terminate the bit of code this program
consists of.
You've no doubt noticed that ProgramStart appears on the same
line with END in the example. Besides terminating programs, END
optionally does double duty by indicating where execution
should begin when the program is run. For any of a number of
reasons, you may not want to start executing a program with the
first instruction in the .EXE file; END takes care of such cases. For
example, suppose you run the program assembled and linked
from this code (DELAY.ASM):
.MODEL small
.STACK 200h
.CODE
Delay:
mov cx,O
DelayLoop:
loop DelayLoop
ret
ProgramStart:
call D~lay
mov
int
END
ipause for the time required to
i execute 64K loops
ah,4ch
21h
ProgramStart
Execution does not start with the first instruction in the source
code, the MOV CX,O at label Delay. Instead, execution starts with
the CALL Delay instruction at label ProgramStart, as specified by
the END directive.
If you have two addresses In
your program, TLINK will use
the first one It finds and
Ignore the other.
In a program consisting of only one module (that is, one source
code file), the END directive should always specify the start
address for the program. In a program consisting of more than
one module, only the END directive in the module containing the
instruction at which the program is to start should specify the
start address; the END directives in all other modules should
appear as END, and nothing more. Think of it this way: Every
program needs a place to start-but it would make no sense to
have two or more places to start. Make sure you have one-and
only one-start address per program.
Chapter 5, The elements of an assembler program
87
Operands
Instruction mnemonics and directives tell Turbo Assembler what
to do. Operands, on the other hand, tell Turbo Assembler what
registers, parameters, memory locations, and so on to associate
with each instance of an instruction or directive. A MOV
instruction means nothing by itself; operands are necessary to tell
Turbo Assembler where to move the value from and where to
store it.
Zero, one, two, or more operands are required for various
instructions, and virtually any number of operands that will fit on
a single line can be accepted by various directives; the correct
number of operands depends on the specific instruction or
directive. (Occasionally, three operands are allowed.) Possible
operands include registers, constant values, labels, memory
variables, and text strings.
It's pretty obvious what an instruction with one operand does: It
operates on that one operand. For example,
push
ax
pushes AX onto the stack. Instructions with no operands are more
obvious still. However, what about the case of an instruction with
two operands, one of which is the source and the other the
destination? For instance, when the 8086 executes
mov
ax,bx
which register is it that gets read out, and which register is it that
receives that value?
You might think that the English equivalent of this instruction
would read, "Move the contents of AX into BX," but that's not the
case. Instead, the MOV instruction moves the contents of BX into
AX. With MOV instructions, mentally substitute an equal sign for
the comma between the two operands and then treat the line like
a C (or Pascal) assignment statement. With this approach, the
MOV example would translate into
ax
= bx;
Admittedly, it's a bit confusing having the rightmost operand as
the source, but at least 8086 assembly language is consistent in
. this respect. You'll soon get used to it.
88
Turbo Assembler User's Guide
Register operands
Registers are perhaps the most frequently used operands for
instructions. Registers can serve as either source or destination
and can even contain an address to jump to under certain
circumstances. There's very little that can be done with constants,
labels, or memory variables that can't be done with registers; on
the other hand, there are a number of instructions that can only
use register operands.
Here are some examples of instructions with register operands:
mov
push
xchg
ror
in
inc
di,ax
di
ah,dl
dx,cl
al,dx
si
Register operands can be mixed with other sorts of operands:
mov
add
cmp
al,l
[BaseCount],cx
51, [bx]
There's really very little to explain about the use of register
operands. To use a register as an operand, you specify that
register's name as an operand to an instruction, and the
instruction uses that register. If there are two operands, and the
register is the rightmost operand, it's the source register; if it's the
leftmost operand, it's the destination register and may also be one
of the source registers if the instruction requires two sources. For
instance, in
mov
mov
sub
cx,l
dx,2
dx,cx
ex is set to 1, OX is set to 2, and then ex is subtracted from OX
with the result, 1, stored back in OX. ex is the rightmost operand
to the SUB instruction, so it's one source register; OX is the
leftmost operand, so it's both the other source and the destination.
By the way, the action of the preceding SUB instruction is
expressed in English as "subtract ex from OX." Using the
approach of converting to e code to make sense of two-operand
instruction, the previous SUB instruction translates to this:
Chapter 5, The elements of an assembler program
89
dx
-= ex;
In Pascal, it translates to this: dx := dx-cx;
Constant operands
Registers are fine for storing variable values, but often you just
need a constant value for an operand. For example, suppose you
want to count 51 down by 4 in a loop, repeating the loop until 51
reaches zero. You could use
CountByFourLoop:
dec
dec
dec
dec
jnz
si
si
si
si
CountByFourLoop
but it's much easier to use
CountByFourLoop:
sub
jnz
si,4
CountByFourLoop
Characters can be used as constant operands as well, since a
character has a well-defined value. For example, since the
character A has the decimal value 65, these two instructions are
equivalent:
sub
sub
al,'A'
al,65
Constant values can be specified in binary, octal, or hexadecimal
notation, as well as in decimal. We'll discuss those notations in a
later section entitled "Bits, bytes, and bases" (page 116).
Constant operands can never be the leftmost of two operands,
since it's clearly not possible for a constant to be the destination
operand. Constant operands can, however, be used pretty much
anywhere that using a value for a source operand makes sense.
The 8086 does impose some limitations on the use of constants;
for example, you can't push a constant value {this is only a
90
Turbo Assembler User's Guide
restriction of 8086/8088). To push the value 5, you must execute
two instructions:
mov ax,S
push ax
.. .
You'll have to learn special cases where constants aren't allowed
on a case-by-case basis. Fortunately, there aren't many such
instructions, and, of course, Turbo Assembler lets you know right
away if you try to use a constant incorrectly.
Expressions
Constant expressions can be used wherever constant values are
accepted. Turbo Assembler supports full expression evaluation,
including nested parentheses, arithmetic, logical, and relational
operators, and a variety of operators for such purposes as
extracting the segment and offset components of labels and
determining the size of memory variables.
For example, the code
MemVar DB
NextVar DB
mov
mov
mov
mov
0
?
ax,SEG MemVar
ds,ax
bx,OFFSET MemVar+((3*2)-S)
BYTE PTR [bx],l
uses the SEG operator to load the constant value of the segment
Mem Var resides in into AX. and then copies that value from AX. to
OS. Next, this code uses a complex expression, involving the *, +,
-, and OFFSET operators, that resolves to the value OFFSET
MemVar+l, which is nothing more than the address of NextVar.
Finally, the BYTE PTR operator is used to select a byte-sized
operation when storing the constant value 1 to the location
pointed to by BX, which is NextVar.
An important point about expressions is that all expressions must
resolve to a constant value. OFFSET MemVar is a constant value--the offset of MemVar in its segment. After all, while the value
stored at MemVar may change, MemVar itself certainly isn't going
to move.
Chapter 5, The elements of an assembler program
91
Turbo Assembler can evaluate expressions consisting of constant
values as it assembles your code, precisely because constant
values are always known. To Turbo Assembler, OFFSET
MemVar+2 is just like 5 + 2; since all the component parts of this
expression are unchanging and well-defined at assembly time, the
expression can be resolved to a single constant value.
Here are the operators that can be used in expressions:
<>,
0, D, LENGTH, MASK, SIZE, WIDTH
• (structure member selector)
HIGH, LOW
+,-(unary)
: (segment override)
OFFSET, PTR, SEG, THIS, TYPE
*, /, MOD, SHL, SHR
+, - (binary)
EQ, GE, GT, LE, LT, NE
NOT
AND
OR,XOR
LARGE, SHORT, SMALL, .TYPE
Many operators are self-explanatory, doing just what you'd
expect them to do in any arithmetic expression. We'll explain
operators as we come to them in this chapter. In the meantime,
refer to Chapter 2 of the Reference Guide if you've any questions
about specific operators.
Label operands
Labels can serve as operands to many instructions. Given the
proper operators, labels can be used to generate constant values.
For example,
MemWord
92
ow
1
mov
al,SIZE MemWord
Turbo Assembler User's Guide
moves 2, the size in bytes of the memory variable Mem Word, into
AL. In this context, a label can become part of an expression, as
illustrated in the last section.
Labels can also be used as the destinations of CALL and JM~
instructions. For example, in
cmp
ja
ax,lOO
IsAbovelOO
IsAbovelOO:
the JA instruction jumps to the address specified by the operand
IsAbovel00 if AX. is above 100. Again, in this capacity labels are
used as constants, specifying memory addresses to be branched
to.
Finally, labels can be used as operands in much the same way as
registers are-as source or destination operands to data
manipulation instruc~ions. The code
TempVar DW
mov
sub
?
[TempVar],ax
ax, [TempVar]
invariably leaves AX. containing zero, since the first instruction
writes the value stored in AX. to the memory variable TempVar,
and the second instruction subtracts the value stored in TempVar
from AX.
The use of labels as operands is part of the larger topic of
memory-addressing modes, which we'll explore next.
Memory-addressing
modes
When you use a memory operand, exactly how do you specify
which memory location you want to work with? The obvious
answer is to give the name of the desired memory variable, as we
did in the last section. You can subtract the memory variable
Debts from the memory variable Assets with
Assets OW
Debts DW
?
mov ax, [Debts]
Chapter 5, The elements of an assembler program
93
sub [Assets],ax
There's more to memory-addressing than meets the eye, though.
Suppose you have a character string named CharString,
containing the letters ABCDEFGHIJKLM, which starts at offset
100 in the data segment, as shown in Figure 5.1.
FIgure 5.1
The memory
location of the
character strIng
Charstrlng
~
CharString
- - - - 1__
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
?
'A'
'B'
'C'
'0'
'E'
'F'
'G'
'H'
'I'
'J'
'K'
'L'
'M'
0
?
How can you read the ninth character, I, which is at address lOB?
In C, you can just use
C = CharString[8];
And in Pascal, you can use
C := CharString[9];
But how can you do the same in assembler? Certainly, referencing
CharString directly isn't going to do the trick, since the character
at CharString is A.
Actually, assembly language supports several different ways to
handle the addressing of character strings, arrays, and data
buffers. The simplest way to read the ninth character of Char String
is
•DATA
CharString DB
94
'ABCDEFGHIJKLM',O
Turbo Assembler User's Guide
•CODE
mov ax,@data
mov ds,ax
mov aI, [CharString+8]
In this case, this is the same as
mov aI, [100+8]
mov al,ds:[lOO+8]
(Ideal mode)
(MASM mode)
since CharString starts at offset 100. Turbo Assembler treats
everything between square brackets as an address, so the offset of
CharString and 8 are added together and used as a memory
address. The instruction effectively becomes
mov aI, [108]
mov al,ds: [108]
(Ideal mode)
(MASM mode)
as shown in Figure 5.2.
Figure 5.2
Addressing the
character string
Charstrlng
~
CharString
CharString + 8
•
~
99
100
?
'A'
'B'
'C'
'0'
104
'E'
101
102
103
105
106
107
108
109
110
111
112
113
114
I
ALI
'F'
'G'
'H'
'I'
'J'
'K'
'L'
'M'
0
?
This sort of addressing, where a memory location is specified
either by its name or by its name plus some constant, is known as
direct addressing. While direct addressing is straightforward to
use, it's not very flexible because it accesses the same memory
address every time. Let's look at another, more flexible way to
address memory.
Chapter 5, The elements of an assembler program
95
Consider the following, which also loads the ninth character of
CharString into AL:
mov bx,OFFSET CharString+8
mov aI, [bx]
This example uses BX to point to the ninth character. The first
instruction loads BX with the offset of CharString (remember that
the OFFSET operator returns the memory offset of a labeD, plus 8.
(This is an expression, with Turbo Assembler doing the OFFSET
calculation and the addition at assembly time.) The second
instruction specifies that AL should be loaded with the contents of
the memory offset pointed to by BX, as shown in Figure 5.3.
Figure 5.3
Using BX to address
Charstrlng
~
CharString
99
---ta_
100
101
102
103
104
105
106
BX I
108
107
I~ 108
109
110
111
112
113
114
?
'A'
'B'
'C'
'0'
'E'
'F'
'G'
'H'
'I'
'J'
'K'
'L'
'M'
0
?
ALI
It's the square brackets that indicate that the memory location
pointed to by BX, rather than BX itself, should be the source
operand. Don't forget the brackets when using BX as a memory
pointer; for example,
mov
ax, [bx]
iload AX from the memory offset
i pointed to by BX
ax,bx
iload AX with the contents of BX
and
mov
are two very different instructions.
96
Turbo Assembler User's Guide
Why bother to first load BX with the offset of a memory variable
and then access memory using BX as a pointer, when a single
instruction with a direct operand does the same thing? The special
quality of registers used as memory pointers is that, unlike
instructions that use direct operands, instructions that use
registers as pointers can point to different memory addresses at
different times in the execution of a program.
Suppose you want to find the last character of a null-terminated
CharString. In order to do this, you must start at the first character
of CharString, search for the zero byte that ends the string, and
then back up one character to read the last character. There's no
way to do this with direct addressing, since the string could be of
any length. Using BX as a pointer register, though, does the trick
nicely:
rnov bx,OFFSET CharString
;point to string start
FindLastCharLoop:
rnov aI, [bx]
;get next string char
;is this the zero byte?
crop
al,O
;yes, back to last char
je
FoundEndOfString
inc bx
;point to next char
;check the next char
jrnp FindLastCharLoop
FoundEndOfString:
dec bx
;point back to last char
rnov aI, [bx]
;get the last char in the string
If you're going to search through memory for characters or
words, if you're going to manipulate arrays, or if you're going to
copy blocks of data about, you'll find that pointer registers are
invaluable.
BX is not the only register that can be used as a memory pointer.
BP, 51, and DI can also be used, along with an optional constant
value or label. The general form of a memory operand looks like
this:
[base registertindex+register+displacement]
or
[base registertindex] [register+displacement]
where base register is BX or BP, index register is 51 or DI, and
displacement is any 16-bit constant value, including expressions
and labels. The three components are added together by the 8086
each time an instruction using a memory operand is executed.
Chapter 5, The elements of an assembler program
97
Each of the three parts of a memory operand is optional, although
. obviously you must use at least one of the three (or else you'd
have no memory address at a11!). Here's how the elements of a
memory operand look in another format:
BX
or
51
+
BP
(base)
or
+
displacement
D1
(index)
It works out that there are 16 ways to specify a memory address:
• [displacement]
•
•
•
•
•
•
•
[bx]
lsi]
[di]
[bx+si]
[bx+di]
[bp+si]
[bp+di]
•
•
•
•
•
•
•
•
[bp+displacement]
[bx+displacement]
[si+displacement]
[di+displacement]
[bx+si+displacement]
[bx+di+displacement]
[bp+si+displacement]
[bp+di+displacement]
where, again, displacement is anything that works out to a 16-bit
constant value.
Sixteen addressing modes certainly seem like a lot, but if you look
at the preceding list, you'll see that all those addressing modes are
built from nothing more than a few elements combined in a few
different ways. Here are some more ways you can load the ninth
character of Char5tring into AL, using the various addressing
modes:
•DATA
CharString DB 'ABCDEFGHIJKLM',O
•CODE
mov ax,@data
mov ds,ax
mov si,OFFSET CharString+8
mov aI, [si]
98
Turbo Assembler User's Guide
rnov bx,B
rnov aI, [CharString+bx]
rnov bx,OFFSET CharString
rnov aI, [bx+B]
rnov si,B
rnov aI, [CharString+si]
rnov bx,OFFSET CharString
rnov di,B
rnov aI, [bx+di]
rnov si,OFFSET CharString
rnov bx,B
rnov aI, [si+bx]
rnov bx,OFFSET CharString
rnov si,7
rnov aI, [bx+si+l]
rnov bx,3
rnov si,S
rnov aI, [bx+CharString+si]
Believe it or not, all these instructions reference exactly the same
memory location, [CharStringl+8.
There are several interesting points about this example. First, you
should understand that a plus (+) sign used inside square
brackets has a special meaning. At assembly time, Turbo
Assembler adds together all the constant values inside square
brackets, so that
rnov
[10+bx+l+si+l00],cl
effectively becomes
rnov
[bx+si+lll],cl
Then, when the instruction is actually executed (when the
program is run), the memory-addressing operands are added
together on the fly by the 8086. If BX contains 25 and 51 contains
52, then CL is stored to the memory address 25 + 52 + 111 = 188
when the MOV instruction is executed. The key here is that it's the
8086 that adds together the base register, the index register, and
the displacement when this instruction is executed. To put it
another way, Turbo Assembler adds the constants at assembly
Chapter 5, The elements of an assembler program
99
time, while the 8086 adds together the base and/or index and/or
displacement fields as the instruction is actually executed.
BP can be made to address
the data segment, and BX,
Sf, and DI can be made to
address the stack segment.
or the code segment, or the
extra segment, by use of
segment override prefixes.
Chapter 9 covers segment
override prefixes; most of the
time, though, you won't
need them, and for now
we 'I/Ignore theIr exIstence.
You might have noticed that we haven't used BP.in any of the
examples so far. That's because BP behaves a little differently from
BX. Recall that while BX is used as an offset into the data segment,
BP is used as an offset into the stack segment. That means that BP
can't normally be used to address CharString, which resides in the
data segment (more on segments shortly).
An explanation of the use of BP to address the stack segment is
given in Chapter 4. For now, it's enough to know that BP can be
used just as we've used BX in the examples, except that the data
addressed must reside in the stack segment when BP is used.
Finally, the square brackets around direct addresses are optional.
That is,
mov aI, [MemVar]
and
mov aI, MemVar
Pick a style for your own
code and stick with It.
do exactly the same thing. Nonetheless, we strongly recommend
placing square brackets around all memory references, in order to
red uce confusion and make your code as clear as possible. At
some point, you'll undoubtedly come across code that lacks
square brackets, since some people feel that the bracketless code is
more intuitive. As usual, it's a matter of taste, and you'll find that
your programming goes more smoothly if you choose a single
memory-addressing style and use it consistently.
You'll also run across memory-addressing forms like
mov al,CharString[bx]
and even
mov al,CharString[bx] [si]t1
All these forms are the same as putting the memory-addressing
elements inside a single pair of square brackets and separating
them with plus signs; the last example is the same as
mov aI, iCharStringtbxtsitl]
Square brackets around register pointers to memory are not
optional. Without square brackets, for instance, BX is treated as an
operand, not as a pointer to an operand.
100
Turbo Assembler User's Guide
Comments
Last, but surely not least, we come to the comment field.
Comments don't actually do anything, in the sense that they don't
affect the code of the executable file generated by Turbo
Assembler, but that doesn't mean they're not important.
Most likely, you already know how to program in some high-level
language-C, Pascal, Prolog, or whatever-since few people
begin their programming careers with assembly language. As you
learned that language, no doubt you were advised time and time
again to comment carefully. That's good advice, since both
complexity and passing time can make any program inscrutable
even to its author.
By comparison with assembly language, though, a Pascal
program is virtually self-documenting. Pascal code is full of
neatly delineated control structures, strongly typed variables,
arithmetic expressions, and procedure and function calls comple,te
with formal and actual parameters.
Assembly language, on the other hand, has no built-in control
structures, strong but erratically enforced data-typing, no
arithmetic expressions involving variables, and no inherent
parameter-passing mechanism. In short, assembler code is about
as far from structured, easily maintained code as you're ever
likely to see. This doesn't mean that assembler programs can't be
structured, or that they can't be maintained, but rather that you
must use comments (and subroutines and macros as well) to raise
assembler code above its natural cryptic level.
Comments make It easy for
you or someone else to look
over the code and quickly
understand It.
There are all sorts of ways to comment assembler code. One
useful approach is to put a comment at the right margin of each
instruction that might benefit from a bit of explanation. Fdr
instance, you've certainly got a better shot at understanding
mov [bx],al
istore the modified character
at a glance than
mov [bx],al
You don't have to comment every line; after a while, comments
like
mov
ah,l
iDOS keyboard input function t
Chapter 5, The elements of an assembler program
101
int
21h
iinvoke DOS to get the next key press
cease to serve any useful purpose. That doesn't mean you
shouldn't comment such lines, though; instead, make your
comments short and to the point:
mov
int
ah,l
21h
iget the next key press
Another good commenting technique is to use lines of only
comments to describe blocks of code. These comments can
describe code operation at a higher level than comments for
individual lines can. For example, consider the following:
i
Generate a checksum byte for the transfer buffer.
mov
mov
sub
Checksum:
add
inc
loop
bx,OFFSET TransferBuffer
cx,TRANSFER_BUFFER_LENGTH
al,al
iclear the checksum accumulator
al, [bx]
bx
Checksum
iadd in the current byte's value
ipoint to the next byte
Note that we didn't comment every line. In light of the comment
for this block of code, it's obvious that BX is loaded with the
address of the transfer buffer, and that ex: is loaded with the
length of the buffer. The key here is that the comment for this
block of seven lines neatly summarizes the operation of the code,
so the comments for the individual lines become less important.
Someone skimming through the code is likely to benefit more
from the block comments than from the line comments.
Another still higher-level commenting technique is that of
preceding each subroutine with a descriptive comment header.
Such a header can contain a description of the subroutine, a
summary of inputs and outputs, register preservation
information, and miscellaneous notes on the subroutine's
operation. For example,
i
102
Function to return the byte-sized checksum of a data buffer.
Turbo Assembler User's Guide
Input:
DS:BX - a pointer to the start of the buffer
CX - the length of the buffer
i
Output:
AL - the buffer checksum
Registers destroyed:
BX, CX
i
i
NOTE: The buffer must not exceed 64K in length, and must
not cross a segment boundary.
Checksum
sub
Checksum:
add
inc
loop
ret
Checksum
PROC
al,al
NEAR
iclear the checksum accumulator
aI, [bx]
bx
Checksum
iadd in the current byte's value
ipoint to the next byte
ENDP
If you think about it, you'll realize that once a subroutine is
written and working properly, there's rarely any reason to ever
look at the code of that subroutine again. What you will want to
know is exactly what happens when you call that subroutine; in
other words, you'll often want to know just how that subroutine
interacts with the code that's calling it. A descriptive header such
as the one we've written meets that need very well.
There are many other commenting techniques, and you'll no
doubt develop one suited to your programming style. The
important thing is to make it a point to comment your code
thoroughly from the start, so that commenting becomes an
integral part of your programming style.
Segment directives
In both this chapter and the last, we've spent considerable time
discussing what segments are and how they affect the code you
write. There's one thing we haven't dealt with yet, though, and
that's how Turbo Assembler knows exactly which segment or
segments data and code reside in.
Segment control is one of the more complex aspects of 8086
assembly language; accordingly, Turbo Assembler provides not
Chapter 5, The elements of an a~mbler program
103
one but two sets of segment control directives. The first set,
consisting of the simplified segment directives, makes segment
control relatively easy and is ideal for linking assembler modules
to high-level languages, but supports only some of the segmentrelated features of which Turbo Assembler is capable. The second
set, consisting of the standard segment directives, is more
complicated to use, but provides the complete segment control
required by demanding assembler applications.
See Chapter 9, -Advanced
programming In Turbo
Assembler- for a detailed
explanation of segment
directives.
Simplified
segment
directives
.STACK, .CODE, and
.DATA
Next, we'll look at both the simplified and the standard segment
directives. We'll just give you an overview of how to use the
segment directives so you'll know enough to write your own
programs.
The key simplified segment directives are .STACK, .CODE, .DATA,
.MODEL, and DOSSEG. We'll cover these in two groups in this
section, starting with .STACK, .CODE, and .DATA.
.STACK, .CODE, and .DATA define the stack, code, and data
segments, respectively..STACK controls the size of the stack. For
example,
. STACK 200h
For Information about
exceptions to using .STACK,
see the section -Forgetting
the stack or reserving a toosmall stack- on page 230 In
Chapter 6.
defines a stack 200h (512) bytes long. That's really all you have to
do as far as the stack is concerned; just make sure you've got a
.STACK directive in your program, and Turbo Assembler handles
the stack for you. 200h is a good stack size for normal programs,
although heavily stack-oriented programs-for instance,
programs using recursion-might require larger stacks.
.CODE marks the start of your program's code segment. You
might think it would be obvious to Turbo Assembler that all your
instructions belong in the code segment. Actually, though, Turbo
Assembler lets you have many code segments (by using the
standard segment directives), and .CODE tells Turbo Assembler
exactly which code segment to place your instructions in.
Defining your code segment is even simpler than defining your
stack segment, since there are no operands to .CODE. For
example,
.CODE
104
Turbo Assembler User's Guide
sub
mov
ax,ax
cx,lOO
;set the accumulator to zero
;. of loops to execute
.DATA is a bit more complex. As you'd expect, .DATA marks the
start of your data segment. You should place your memory
variables in this segment. For example,
. DATA
TopBoundary
Counter
ErrorMessage
DW
DW
DB
100
?
Odh,Oah,'***Error***' ,0dh,Oah,'$'
That's certainly straightforward. The complex part of .DATA (and
it's really not that complex) is that you must explicitly load the DS
segment register with the symbol @data before you can access
memory locations in the segment defined by .DATA. Since a
segment register can be loaded from either a general-purpose
register or a memory location, but can't be loaded with a constant,
the OS segment register is generally loaded with a two-instruction
sequence along the lines of
mov
mov
ax,@data
ds,ax
(Any general-purpose register could be used instead of AX.) The
preceding sequence sets OS to point to the data segment that
starts with the .DATA directive.
The following program displays the text stored at DataString on
the screen (DSLYSm.ASM on disk):
.r-KlDEL
•STACK
•DATA
DataString
. CODE
ProgramStart:
mov
mov
mov
mov
int
small
200h
DB 'This text is in the data segmentS'
bx,@data
ds,bx
;set DS to the .DATA segment
dx,OFFSET DataString ;point DX to the offset
; of DataString in
; the .DATA segment
ah,9
;DOS print string function f
2lh
;invoke DOS to print string
Chapter 5, The elements of an assembler program
105
mov
int
END
ah,4ch
21h
ProgramStart
iDOS terminate program function t
iinvoke DOS to end program
Without the two instructions that set the OS register to the
segment defined with .DATA, the print string function wouldn't
work properly. Data String resides in the .DATA segment and
cannot be accessed unless OS is set to that segment. You might
want to think of it this way: When you invoke DOS to print a
string, you pass the full segment:offset address of the string in
OS:OX. Only after you loaded OS with the .DATA segment and
OX with the offset of DataString did you have a full segment:offset
pointer to DataString.
You may well wonder at this point why it is that you have to load
OS, but not CS or 55. Then, too, what about ES?
Well, you never have to load CS explicitly because DOS does that
for you when you run a program. After all, if CS weren't already
set when the time came to execute the first instruction of a
program, the 8086 wouldn't know where to find the instruction,
and the program would never run. This may not be obvious to
you right now, but trust us-CS is automatically set when a
program begins, and you never need to load it explicitly.
Likewise, 55 is set by DOS before a program begins, and generally
stays the same for the duration of the program. While it is possible
to change 55, it's rarely desirable, and it's certainly not something
you'll want to attempt unless you know exactly what you're
doing. So, like CS, 55 is automatically set when a program begins,
and need not be touched thereafter.
OS is quite different. While CS points to instructions, and 55
points to the stack, OS points to data. Programs don't directly
manipulate instructions or stacks-but they do constantly
manipulate data directly. What's more, programs might want to
get at data in any of several different segments at any time;
remember that the 8086 allows you to access any memory location
in a 1 Mb range, but only in blocks of 64K (relative to a segment
register) at a time.
You may well want to load OS with one segment, access data in
that segment, and then load OS with another segment in order to
access a different block of data. In small- and medium-sized
programs, such as those we've presented here, you'll never need
more than one data segment, but larger programs often use
multiple data segments. Also, you'll need to load OS with
106
Turbo Assembler User's Guide
different values if you want to access system memory areas, such
as the memory locations used by the BIOS.
The upshot of all this is that Turbo Assembler lets you set DS to
any segment at any time. In return for this flexibility, you must
explicitly set DS to the segment you want-usually @data, which
is equivalent to the segment that starts with .DATA-before you
access memory locations in that segment.
The ES segment register is loaded just like DS. Often, you won't
need to bother with ES at all, but when you do need to access a
memory location in the segment pointed to by ES, you must first
load ES with that segment. For example, the following program
loads ES with the .DATA segment, then loads a character to print
from that segment via ES:
.MJDEL small
•STACK 200h
•DATA
Output Char
DB ' B'
•CODE
ProgramStart:
mov dx,@data
mov es,dx
;set ES to the .DATA segment
mov bx,OFFSET OutputChar ;point BX to offset of Output Char
mov al,es:[bx]
;get character to output from
; segment pointed to by ES
mov ah,2
;DOS display output function I
int 21h
;invoke DOS to print character
mov ah,4ch
;DOS terminate program function I
int 21h
;invoke DOS to end program
END ProgramStart
Note that ES is loaded with the two-instruction sequence
mov dx,@data
mov es,dx
just as DS was earlier.
Admittedly, there's no particular reason to use ES rather than DS
in this example, and, in fact, using ES meant that we had to use an
ES: segment override prefix (as discussed in Chapter 9). However,
there are many occasions when it's handy to have ES set to one
segment while DS is set to another, particularly when the string
instructions are used.
Chapter 5~ The elements of an assembler program
107
DOSSEG
See Chapter 3 In the
Reference Guide for more
on DOSSEG.
The DOSSEG directive causes the segments in an assembler
program to be grouped according to the Microsoft segmentordering conventions. For now, you don't need to worry about
what that means; all you need to know is that almost all standalone assembler programs will work just fine if you start them
with DOSSEG.
While it is not necessary to specify DOSSEG when linking
assembler modules to a high-level language, since the high-level
language automatically selects Microsoft segment-ordering,
DOSSEG doesn't hurt and is a useful reminder of the sort of
segment-ordering that is in effect.
All this means is that the simplest approach is to use DOSSEG as
the first line in all your programs (unless you have a specific
reason not to). That way, you'll be" able to rely on a consistent
segment order.
.MODEL
The .MODEL directive specifies the memory model for an
assembler module that uses the simplified segment directives.
Note that near code is branched to (jumped to) by loading the IP
register only, while far code is branched to by loading both CS
and IP. Similarly, near data is accessed with just an offset, while
far data must be accessed with a full segment:offset address. In
short, far means that full 32-bit segment:offset addresses are used,
while near means that 16-bit offsets can be used.
These are the available memory models:
108
tiny
Both program code and program data must fit within
the same 64K segment. Both code and data are near.
small
Program code must fit within a single 64K segment,
and program data must fit within a separate 64K
segment. Both code and data are near.
medium
Program code may be larger then 64K, but program
data must fit within a single 64K segment. Code is far,
while data is near.
compact
Program code must fit within a single 64K segment,
but program data may be larger than 64K. Code is
near, while data is far. No single data array may be
greater than 64K.
Turbo Assembler User's Guide
large
Both program code and program data may be larger
than 64K, but no single data array may be larger than
64K. Both code and data are far.
huge
Both program code and program data may be larger
than 64K, and data arrays may exceed 64K in size.
Both code and data are far. Pointers to elements
within an array are far.
Note that, from an assembler point of view, large and huge are
identical. Huge model does not automatically support data arrays
larger than 64K.
Few assembler programs require more than 64K of code or data,
so the small model serves well in most applications. You should
use the small model whenever possible, because far code
(medium, large, and huge models) makes program execution
slower; far data (compact, large, and huge models) is considerably
harder to manage in assembler.
The memory models described here correspond to the memory
models used by Turbo C (and many other compilers for the PC).
Whenever you link an assembler module to a high-level language,
be sure to use the correct .MODEL directive.. MODEL makes sure
that assembler segment names correspond to those used by highlevel languages, and that labels of type PROC, which are used to
name subroutines, procedures, and functions, default to the
type-near or far-used by high-level languages .
•MODEL is required if you're using the simplified segment
directives, since otherwise Turbo Assembler wouldn't know how
to set up the segments defined with .CODE and .DATA.. MODEL
must precede .CODE, .DATA, and .STACK.
Here's the framework of a program using simplified segment
directives:
.MJDEL small
. STACK 200h
•DATA
MemVar DW
0
•CODE
ProgramStart:
mov
mov
mov
ax,@data
ds,ax
ax, [MemVar]
Chapter 5, The elements of an assembler program
109
mov
int
END
Other simplified
segment directives
ah,4ch
21h
ProgramStart
There are several other less commonly used segment directives.
You'll need these only for large or sophisticated assembler
programs, so we'll just mention them now to let you know they
exist; refer to Chapter 9 for more information .
. DATA? is used just like .DATA except that it defines that portion
of the data segment containing uninitialized data. This is usually
used in an assembler module linked to a high-level language.
. FAR DATA lets you define a far data segment, that is, a data
segment other than the standard @data segment shared by all
modules .. FARDATA allows an assembler module to define its
own data segment of up to 64K in size. If a .FARDATA directiv~
has been given, @fardata is the name of the far data segment
Specified by that directive, just as @data is the name of the data
segment specified by .DATA.
. FARDATA? is much like .FARDATA except that it defines an
uninitialized far segment. As with .FARDATA and @fardata, if-a
.FARDATA? directive has been given, @fardata? is the name of
the far data segment specified by that directive.
.CONST defines that portion of the data segment containing
constant data. Once again, this only matters when linking
assembler code to a high-level language.
Some useful predefined labels are available when the simplified
segment directives are used:
• @FlleName is the name of the file being assembled.
• @Curseg is the name of the segment Turbo Assembler is
currently assembling into.
• @CodeSlze is 0 in memory models with near code segments
(tiny, small, and compact) and 1 in memory models with far
code segments (medium, large, and huge).
• Likewise, @DataSlzeis 0 in memory models with near data
segments (tiny, small, and medium), 1 in compact and large
memory models, and 2 in the huge model.
110
Turbo Assembler User's Guide
Standard
segment
directives
Next, we'll show the same sample program framework from the
last section, but this time we'll use the standard segment
directives SEGMENT, ENDS, and ASSUME:
DGROUP GROUP
ASSUME
STACK SEGMENT
DB
STACK ENDS
DATA SEGMENT
MemVar OW
_DATA, STACK
cs:_TEXT, ds:_DATA, ss:STACK
PARA STACK 'STACK'
200h DUP (?)
WORD PUBLIC 'DATA'
0
DATA ENDS
TEXT SEGMENT WORD PUBLIC 'CODE'
ProgramStart:
mov ax,_DATA
mov ds,ax
mov ax, [MemVar]
TEXT
mov ah,4ch
int 21h
ENDS
END ProgramStart
Now you know why the simplified segment directives are called
"simplified"! However, much of what the simplified segment
directives do is intended to make it easier to link assembler
modules to high-level languages and is unnecessary in standalone assembler programs. Here's the Hello, world program using
standard segment directives:
STACK
STACK
SEGMENT PARA STACK 'STACK'
DB
200h DUP (?)
ENDS
Data
SEGMENT WORD 'DATA'
HelloMessage
DB
'Hello, world' ,13,10,'$'
Data
ENDS
This example Isn't terribly
complicated, but It's clear
that the standard segment
directives are more complex
than the simplified segment
directives. Chapter 9
describes the standard ones
In detail.
Code
SEGMENT WORD 'CODE'
ASSUME cs:Code, ds:Data
ProgramStart:
mov ax, Data
mov ds,ax
iset OS to the Data segment
mov dx,OFFSET HelloMessage iDS:DX points to
i the hello message
Chapter 5, The elements of an assembler program
111
Code
mov
int
mov
int
ENDS
END
ah,9
2lh
ah,4ch
2lh
iDOS print string function t
iprint the hello string
iDOS terminate program function t
iend the program
ProgramStart
In this section, we're only going to give you an idea what each
standard segment directive does.
The SEGMENT directive
The SEGMENT directive defines the start ofa segment. The label
accompanying the SEGMENT directive is the name of the
segment; for example,
Cseg
SEGMENT
defines the start of a segment named Cseg. The SEGMENT
directive may optionally specify a number of segment attributes,
including alignment on a byte, word, doubleword, paragraph (16
byte), or page (256 byte) memory boundary. Other attributes
include the way in which the segment can be combined with
other segments with the same name and the class of the segment.
The ENDS directive
The ENDS directive defines the end of a segment. For example,
Cseg
ENDS
ends the segment named Cseg, which was started earlier with the
SEGMENT directive. When you use the standard segment
directives, you must explicitly end every segment.
The ASSUME directive
The ASSUME directive tells Turbo Assembler what segment a
given segment register is currently set to. An ASSUME CS:
directive is required in every program that uses the standard
segment directives, since Turbo Assembler needs to know about
the code segment in order to set up an executable program.
ASSUME OS: and ASSUME ES: are usually used as well so that
Turbo Assembler knows what memory locations you can address
at any given time.
ASSUME lets Turbo Assembler check that each access to a named
memory variable is valid, given the current segment register
settings. For example, consider the following:
112
Datal
Varl
Datal
SEGMENT WORD ' DATA'
OW
0
ENDS
Data2
Var2
SEGMENT WORD ' DATA'
0
OW
Turbo Assembler User's Guide
Data2
ENDS
Code
SEGMENT
ASSUME
ProgramStart:
mov
mov
ASSUME
mov
WORD 'CODE'
cs:Code
ax, Datal
ds,ax
ds:Datal
ax, [Var2]
.. .
Code
mov
int
ENDS
END
ah,4ch
2lh
;set OS to Datal
;try to load Var2 into AX--this will
; cause an error, since Var2 can't
; be reached in segment Datal
;DOS terminate program function I
;end the program
ProgramStart
Turbo Assembler flags an error in this code because the code tries
to access memory variable Var2 when OS is set to segment Datal,
and Var2 can't be addressed unless OS is set to segment Datal.
It's important to understand that Turbo Assembler doesn't
actually know that OS has been set to Datal; rather, by using the
ASSUME statement, you told Turbo Assembler to make that
,assumption. ASSUME is your way to tell Turbo Assembler what
the segment registers are set to at any given time, so that Turbo
Assembler can let you know when you've attempted the
impossible.
Turbo Assembler can't catch all such mistakes, however.
Whenever a memory reference involves a named memory
variable, such as previous Varl or Var2, Turbo Assembler can
check the validity of that reference, since each named memory
variable is explicitly associated with a segment. There's no way
Turbo Assembler can know what segment an instruction like
mov ai, [bx]
is intended to access, though. In such a case, Turbo Assembler
must assume that the segment OS is set to is the segment you
want to access.
If a segment register doesn't currently point to any named
segment, you can use NOTHING with ASSUME to convey that
information to Turbo Assembler. For example,
mov
mov
ax,Ob800h
ds,ax
Chapter 5, The elements of an assembler program
113
ASSUME ds:NOTHING
sets DS to point to the color text screen and then informs Turbo
Assembler that OS doesn't point to any named segment. Here's
another way to point to the color text screen:
ColorTextSeg
SEGMENT AT OB800h
ColorTextMemory LABEL BYTE
ColorTextSeg
ENDS
mov
ax,ColorTextSeg
mov
ds,ax
ASSUME ds:ColorTextSeg
Note that the AT directive that follows SEGMENT provides an
explicit starting address for the segment.
One final point about ASSUME: It may cause Turbo Assembler to
use a different segment register than you expect to access memory
in some cases. For example, consider the following code:
Datal
Varl
Datal
SEGMENT WORD ' DATA'
DW
0
ENDS
Data2
Var2
Data2
SEGMENT WORD ' DATA'
DW
0
ENDS
Code
SEGMENT
ASSUME
ProgramStart:
mov
mov
ASSUME
mov
mov
ASSUME
mov
Code
114
mov
int
ENDS
END
WORD 'CODE'
cs:Code
ax, Datal
ds,ax
ds:Datal
ax,Data2
es,ax
es:Data2
ax, [Var2]
ah,4ch
2lh
;set DS to Datal
;set ES to Data2
;load Var2 into AX--Turbo Assembler
; tells the 8086 to load
; relative to ES, since Var2
; can't be reached relative to DS
;DOS terminate program function •
;end the program
ProgramStart
Turbo Assembler User's Guide
This example should look familiar; it's a modified version of the
code we used earlier to show how ASSUME lets Turbo Assembler
tell you when you've attempted an impossible memory reference.
In this example, though, no error is reported, but that doesn't
mean Turbo Assembler is letting you make a mistake. Instead,
Turbo Assembler modifies
mov
ax, [Var2]
to access Var2 relative to the ES segment register rather than the
DS segment register.
Segment override prefixes,
and the standard segment
directives In general, are
discussed In Chapter 9.
What happens is this: The two ASSUME directives have informed
Turbo Assembler that DS is set to the Datal segment and that ES
is set to the Data2 segment. Then, when the MOV instruction
attempts to access Var2, which is in the Data2 segment, Turbo
Assembler correctly concludes that there's no way Var2 can be
accessed relative to OS; however, Var2 can be accessed relative to
ES. Consequently, Turbo Assembler inserts a special code known
as a segment override prefix before the MOV instruction in order to
tell the 8086 to use the ES rather than the OS segment register.
What does all this mean to you? It means that if you're careful to
use ASSUME directives to let Turbo Assembler know the current
OS and ES settings, Turbo Assembler can automatically help you
out by checking that accesses to named memory variables are
possible, and can even select the correct segment automatically in
some cases.
Simplified versus
standard
segment
directives
Now that you've seen both the simplified and standard segment
directives, the question remains: Which set of segment directives
should you use? The answer depends on the sort of assembler
programming you need to do.
If you're linking assembler modules to a high-level language,
you'll almost always want to use the simplified segment
directives. The simplified segment directives do a good job of
taking care of the segment-naming and memory-model details
associated with the interface to high-level languages.
If you're writing small- or medium-sized stand-alone assembler
programs, you'll generally want to use the simplified segment
directives, since they're easier to use and make programs more
readable.
Chapter 5, The elements of an assembler program
115
If you're writing large stand-alone assembler programs with
many segments and mixed -model programming (both near and
far code and/or near and far data in the same program), you'll
need to use the standard segment directives, since only with the
standard segment directives do you get full control over segment
type, alignment, naming, and the way in which segments are
combined.
The rule of thumb is this: Use the simplified segment directives
until you find you need the complete control over segment
definition that only the standard segment directives can provide.
Allocating data
Now that you know how to create segments, let's look at how to
fill those segments with meaningful data. The stack segment is no
problem; the stack resides there, and you can access the stack with
PUSH, POP, and addressing by way of the BP register. The code
segment is filled with the instructions generated by the instruction mnemonics in your programs, so that's no problem either.
That leaves the data segment. Turbo Assembler provides you
with a variety of ways to define variables in the data segment,
both initialized to some value and uninitialized. In order to
understand the sorts of data Turbo Assembler lets you define, we
must first teach you a bit about the fundamentals of assembler
data types.
Bits, bytes, and
bases The fundamental unit of storage in a computer is a hit. A bit can
store either the value 1 or the value O. A bit, by itself, is not very
useful. The 8086 doesn't deal directly with bits; in fact, it deals
with nothing smaller than a byte, which consists of 8 bits.
Since a bit is effectively a base 2 digit, a byte contains an 8-bit,
base 2 number. The largest possible 8-bit, base 2 number follows:
2
2
2
2
2
2
116
to
to
to
to
to
to
the
the
the
the
the
the
Oth
1st
2nd
3rd
4th
5th
power:
power:
power:
power:
power:
power:
1
2
4
8
16
32
Turbo Assembler User's Guide
2 to the 6th power:
2 to the 7th power:
64
+ 128
255
This means that a byte can store one value in the range a to 255.
Each of the 8086's 8-bit registers (AL, AH, BL, BH, CL, CH, DL,
and DH) stores exactly 1 byte. Each of the 8086'5 1,000,000-plus
addressable memory locations can also store exactly 1 byte.
The PC's character set (which includes uppercase and lowercase
letters, the digits 0 t09, special graphics, scientific, and foreign
characters, and assorted punctuation and other characters)
consists of precisely 256 characters in all. Does that number sound
familiar? It should, since the PC's character set was designed so
that 1 byte can store 1 character.
So now you know about the byte, which is the smallest unit that
the 8086 can address, and which can store one character, one
unsigned value between 0 and 255, or one signed value in the
range -128 to +127. A byte is clearly inadequate for many
assembler programming tasks, such as storing integer and
floating-point values and storing memory pointers.
The next larger storage unit of the 8086 is the 16-bit word. A word
is twice the size of a byte (16 bits). In fact, a word is stored in
memory at two consecutive byte locations; the 8086's memory
address space can be thought of as 500,OOO-plus words. Each of
the 8086's 16-bit registers (AX, BX, CX, DX, 51, Dr, BP, SP, CS, DS,
ES, SS, IP, and the flags register) stores one word. A word
contains a 16-bit, base 2 number. The largest possible 16-bit base 2
number follows:
2
2
2
2
2
2
2
2
2
2
2
2
2
to
to
to
to
to
to
to
to
to
to
to
to
to
the
the
the
the
the
the
the
the
the
the
the
the
the
Oth power:
1st power:
2nd power:
3rd power:
4th power:
5th power:
6th power:
7th power:
8th power:
9th power:
10th power:
11th power:
12th power:
Chapter 5, The elements of an assembler program
1
2
4
8
16
32
64
128
256
512
1024
2048
4096
117
2 to the 13th power:
8192
2 to the 14th power:
16384
2 to the 15th power: + 32768
65535
That's also the maximum size of an unsigned integer-which is no
coincidence, since integers are 16 bits long. Signed integers (which
can range from -32,768 to +32,767) are stored in words as well.
Since words are 16 bits in size, they can address any offset in a
given segment, so word-sized values are large enough to be used
as memory pointers. As you'll recall, the word-sized BX, BP, 51,
and DI registers are used as memory pointers.
The values stored in 32-bit (4-byte) units are known as
doublewords, or dwords. While the 8086 can't manipulate 32-bit
integer values directly, instructions such as AOC and SBB make it
possible to do 32-bit integer arithmetic with two successive 16-bit
operations. Doublewords support unsigned integers in the range
oto 4,294,967,295 and signed integers in the range -2,147,483,648
to +2,147,483,647.
The 8086 can load a segment:offset pointer from a doubleword
into both a segment register and a general-purpose register with
an LOS or LES instruction, but that's as far as direct support for
doublewords goes. Single-precision floating-point numbers are
also stored in doublewords. (Single-precision numbers require 4
bytes and can handle values from 10-38 to 1038.)
Each double-precision floating-point value requires a full 8 bytes.
Such 64-bit values are known as quadwords. The 8086 has no
built-in support for quad words. However, the 8087 numeric
coprocessor uses quadwords as its basic data type. (Doubleprecision numbers handle values that range from 10-308 to 10308
and have an accuracy up to 16 digits.)
Turbo Assembler supports one more data size for temporary
(intermediate) floating-point values, a data element 10 bytes in
length. This 10-byte data element can also be used to store packed
binary-coded decimal (BCD) values, in which each byte stores
two decimal digits.
It's worth noting that the 8086 stores word and doubleword
values in memory, low byte first. That is, if a word value is stored
at address 0, then bits 7 to 0 of the value are stored at address 0,
and bits 15 to 8 are stored at address I, as illustrated by Figure
118
Turbo Assembler User's Guide
5.4.(WordVar contains the value 199Fh; DwordVar contains the
value 12345678h.)
Figure 5.4
storing WordVar
and DwordVar.
WordVar
----1__
0
1
2
3
4
DWordVar - - - - 5
6
7
8
9
9Fh
19h
?
?
?
78h
56h
34h
12h
?
~
Similarly, if a doubleword value is stored at address 5, bits 7-0 are
stored at address 5, bits 15-8 are stored at address 6, bits 23-16 are
stored at address 7, and bits 31-24 are stored at address 8. This
may seem a bit odd, but it's the way that every processor in the
iAPx86 family works.
Decimal, binary, octal,
and hexadecimal
notation
Now that you know the assembly language data types, the next
question is, IIHow do you represent values?" Decimal (base 10)
values are easy, since we've been using decimal notation all our
lives. It's certainly easy enough to type
mov
cx,100 ;set loop counter to 100
and, indeed, Turbo Assembler assumes that values are expressed
in decimal unless you indicate otherwise. Unfortunately, decimal
is not particularly well suited for many aspects of assembly
language because computers are binary (base 2) devices.
Well, then, it seems logical to use binary notation in assembler
programs. You can indicate to Turbo Assembler that a number is
expressed in binary notation simply by putting a b at the end of
the number. (Of course, the number must consist only of Os and 1s
because those are the only two digits in binary notation.) For
instance, decimalS is expressed in binary as 101b.
The problem with binary notation is that base 2 numbers are so
long that they're hard to use. This occurs because each base 2 digit
can store only two possible values, 0 and 1, as shown in Table 5.1.
For instance, here's the last example in binary notation:
mov
cx,1100100b
Chapter 5, The elements of an assembler program
;set loop counter to 100 decimal
119
Word and doublew~rd binary values are even harder to read and
use.
If you're not already familiar with these notations, we strongly
suggest that you get a good book on the topic, since fluency with
binary, octal, and hexadecimal notation is a key element in
assembly language programming.
Decimal
Binary
Octal
Hexadecimal
18
19
20
21
22
23
24
25
26
0
1
10
11
100
101
110
111
1000
1001
1010
1011
1100
1101
1110
1111
10000
10001
10010
10011
10100
10101
10110
10111
11000
11001
11010
20
21
22
23
24
25
26
27
30
31
32
10
11
12
13
256
100000000
400
100
4096
1000000000000
10000
1000
65536
10000000000000000
200000
10000
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
0
1
2
3
4
5
6
7
10
11
12
0
1
2
3
4
5
6
7
8
9
A
13
B
14
C
15
16
D
E
F
17
14
15
16
17
18
19
1A
There are two notations, octal and hexadecimal, that are not only
well matched to the underlying binary nature of the computer,
but are also reasonably compact.
120
Turbo Assembler User's Guide
The suffix 0 Indicates octal
notation: you can also use
the suffix q, which Isn't so
easily confused with zero.
Figure 5.5
From binary
001100100 (decimal
100) to octal 1440
Octal, or base 8, notation uses the digits 0 to 7, displayed in a 3bit-per-digit form. Figure 5.5 shows how the bits of the binary
value 001100100b (100 decimal) can be collected in groups of three
bits to form the octal value 1440.
Binary
144
Octal
Consequently, octal numbers are only one third as long as their
binary equivalents. In octal, the last example becomes
mov
cx,1440
iset loop counter to 100 decimal
Octal notation works perfectly well and is widely used in some
parts of the computer world. By and large, however, IBM PC
programmers almost always use hexadecimal (base 16) notation
rather than octal.
Each hexadecimal digit can take on any of 16 values. Here's how
you count from zero in hexadecimal:
a
1 2 3 4 5 6 7 8 9 ABC D E F 10 ..•
The letters after 9 are the six additional hexadecimal digits A to F.
(Lowercase a to f can also be used.) While it might seem strange
to use letters as digits, you've got no choice, since you need 16
digits and there are only 10 traditional decimal digits. Figure 5.6
shows how the bits of the binary value 01100100b (100 decimal)
can be collected in groups of 4 bits to form the hexadecimal value
64h. (Hexadecimal numbers are denoted with an h suffix.)
Figure 5.6
From binary
01100100 (decimal
100) to
hexadecimal 64
Binary
Hexadecimal
~r-=r
Chapter 5, The elements of an assembler program
6 4
121
Hexadecimal notation essentially displays values in 4-bits-perdigit form, as shown in Figure 5.6. Consequently, hexadecimal
numbers are only one-fourth as long as their binary equivalents.
In fact, any offset or other word value can be expressed in just
four hexadecimal digits. In hexadecimal, the last example
becomes
mov
cx,64h ;set loop counter to 100 decimal
Hexadecimal numbers must begin with one of the digits 0 to 9,
since a hexadecimal number like BAD4 could be mistaken for the
label BAD4h. Here's an example where the hexadecimal value
OBAD4h and the label BAD4h coexist:
•DATA
BAD4h
...°
DW
•CODE
mov ax,OBAD4h
mov
ax,BAD4h
;label BAD4h
;loads AX with a hexadecimal
; constant (the leading dictates
; that this is a constant)
°
;loads AX from the memory
; variable BAD4h (the lack of a
; leading dictates that this
; is a label)
°
In general, only an operand starting with a digit 0 to 9 can be a
constant numeric value.
Floating-point numbers can be denoted in one of two ways. First,
you can specify a floating-point value in the familiar mantissa/
exponent form; for example,
1.1
-12.45
1. OE12
252.123E-6
Turbo Assembler converts the mantissa/ exponent form to binary
form following floating-point format. If you wish, you may
specify floating-point values directly in IEEE or Microsoft binary
form by specifying the number in hexadecimal and placing an r
suffix at the end of the value.
Real numbers can only be used with the DD, DQ, and DT
directives, which we discuss later. If you choose to use the r suffix,
122
Turbo Assembler User's Guide
you must specify exactly the maximum number of hexadecimal
digits for the data type you're initializing (plus a leading zero, if
necessary); for example,
DD
DO
40000000r
OC014CCCCCCCCCCCCr
DT
4037D529AE9E86000000r
;2.0 (exactly 8 long)
;-5.2 (16 long plus
; a leading zero)
;1.2EI7 (exactly 20 long)
In general, it's much simpler to use the mantissa/exponent form.
The letter d can be used as a suffix to indicate that a number is
decimal. Why would you ever need to use the d suffix when
Turbo Assembler assumes that all numbers are decimal? As you
might have guessed, the answer is that you can tell Turbo
Assembler to assume that numbers are in some notation other
than decimal. This is done with the .RADIX directive, which we'll
cover in the next section.
Finally, character constant values can be used with the characters
enclosed in single or double quotes. The value of a character is its
ASCII value. For instance, all the following lines load the
character A into AL:
mov
mov
mov
mov
al,65
al,41h
al,'A'
al,"A"
Where can values in the various notations we've described be
used? Binary, octal, decimal, hexadecimal, and character values
can be used anywhere a constant can be used; for example,
mov
add
sub
and
mov
ax,1001b
cx,5bh
[Count],1770
al,1
aI,' A'
Floating-point values can only be used with DO, OQ, and DT; BCD
(Binary Coded Decimal) values can only be used with CT.
Default radix selection
Most of the time, you'll probably want to use decimal values by
default, simply because that's the most familiar notation.
Occasionally, however, it's convenient to have numbers without
suffixes default to another notation-that's when the .RADIX
directive is needed. (Radix means "base of a numbering system,"
by the way.)
Chapter 5, The elements of an assembler program
123
Incidentally. the operand to
.RADIX Is always decimal, no
matter what default notation
Is selected: In other words,
one .RADIX directive doesn't
affect the notation of the
next .RADIX directive ~
operand.
.RADIX selects the base in which numbers without suffixes are
assumed to be specified. For example,
.RADIX 16
selects base 16, or hexadecimal, as the default notation. The
following code illustrates the effect of the .RADIX directive:
...
•RADIX
mov
•RADIX
sub
16
ax,100
10
ax,100
•RADIX 2
add
ax,100
...
;select base 16, hexadecimal, as default
;= 100h, or 256 decimal
;select base 10, decimal, as default
;-100 decimal, result is
; 256 - 100 = 156 decimal
;select base 2, binary, as default
;+100b, or 4 decimal result is
; 156 + 4 = 160 decimal
.RADIX can select base 2,8, 10, or 16 as the default.
There is a potential problem to consider when you use the .RADIX
directive. No matter what default notation is selected, values
specified with DO, DQ, and DT are assumed to be decimal values
unless a suffix is used. This means that in
.RADIX 16
DD
1E?
1E7 is taken to be 1 times 10 to the seventh power, not 1E711. In
fact, you're best advised to place the h suffix on all hexadecimal
values even after a .RADIX 16 directive. Why? Remember that b
and d are valid suffixes, specifying binary and decimal notation,
respectively. Unfortunately, band d are also valid hexadecimal
digits. If .RADIX 16 is in effect, what is Turbo Assembler to make
of numbers like 129D and 101B?
As it happens, Turbo Assembler always pays attention to valid
suffixes, so 129D is 129 decimal and 101B is 101 binary, or 5
decimal. What this means is that even when .RADIX 16 is in effect,
any hexadecimal number ending in b or d must have an h suffix.
Given that, it's simplest just to put h suffixes on all hexadecimal
numbers, and given that, it becomes clear that, in general, it's not
particularly useful to use .RADIX 16.
124
Turbo Assembler User's Guide
Initialized data
Now we're ready to look at the ways in which Turbo Assembler
lets you define memory variables. First, let's look at the definition
of initialized data.
The data definition directives, DB, DW, DD, DF, DP, DQ, and DT,
let you define memory variables of varying data sizes as follows:
1 byte
DB
DW
DD
DF, DP
DQ
DT
2 bytes =one word
4 bytes = one doubleword
6 bytes = one far pointer word (386)
8 bytes = one quad word
10 bytes
For example, this code defines five initialized memory variables
and illustrates how some of those variables might be used .
•DATA
ByteVar
WordVar
DwordVar
OwordVar
TwordVar
Initializing arrays
' z'
DB
OW
DO
DO
DT
101b
2BFh
3070
100
i1 byte
i2 bytes (1 word)
i4 bytes (1 doubleword)
i8 bytes (1 quadword)
i10 bytes
mov
mov
int
ah,2
dl, [ByteVar]
21h
add
ax, [WordVar]
add
adc
WORD PTR [DwordVar],ax
WORD PTR [DwordVar+2],dx
iDOS display output function f
icharacter to display
iinvoke DOS to display the character
Multiple values may appear with a single data definition
directive. For instance,
SampleArray
ow
0, 1, 2, 3, 4
creates the five-entry array SampleArray, made up of word-sized
elements, as shown in Figure 5.7. Any number of values that will
fit on a line may be used with the data definition directives.
Chapter 5, The elements of an assembler program
125
Figure 5.7
Example of f1veentry array
•
SampleArray
?
o
1
2
3
4
?
What if you want to define an array that's too large to fit on a
single line? Just add more lines; it's not required that a label be
used with the data definition directives. For instance, this code
creates an array of doubleword-sized elements named
SquaresArray, consisting of the squares of the first 15 integers:
. ..
SquaresArray
00
00
00
0, 1, 4, 9, 16
25, 36, 49, 64, 81
100, 121, 144, 169, 196
Turbo Assembler lets you define blocks of memory initialized to a
given value with the DUP operator. For example,
BlankArray
ow
100h oUP (0)
creates an array BlankArray, consisting of 256 (decimal) words
initialized to zero. Likewise, this creates an array of 92 bytes, each
initialized to the character A:
ArrayOfA
Initializing character
strings
DB
92 OUP (' A')
What about creating character strings? Characters are valid
operands to the data definition directives, so you could define a
character string as follows:
String DB
'A', 'B', 'C', '0'
You don't have to do all that typing, though, since Turbo
Assembler provides a handy shortcut:
String DB
'ABCO'
!fyou want to use a C-style string, which is terminated with a
zero byte, you have to explicitly put the zero byte at the end.
126
Turbo Assembler User's Guide
Likewise, uyou want carriage-return or linefeed characters, you
have to insert them as well. The following defines a string of text
followed by a carriage-return character, a linefeed character, and a
tennina ting zero byte:
HelloString
DB
'Hello, world' ,Odh,Oah,O
You must print carriage-return/linefeed pairs in order to advance
to the left margin of the next line. For example, the program
.OODEL small
.STACK 200h
•DATA
Stringl DB
'Linel' ,'$'
String2 DB
'Line2' ,'$'
String3 DB
'Line3' ,'$'
.CODE
ProgramStart:
mov ax,@data
mov ds,ax
mov ah,9
mov dx,OFFSET Stringl
int 2lh
mov dx,OFFSET String2
int 2lh
mov dx,OFFSET String3
int 2lh
mov ah,4ch
int 2lh
END ProgramStart
;DOS print string function f
;string to print
;invoke DOS to print string
;string to print
;invoke DOS to print string
;string to print
;invoke DOS to print string
;DOS terminate program function
prints the following output:
LinelLine2Line3
If, however, you add a carriage-return/linefeed pair at the end of
each string,
Stringl DB
String2 DB
String3 DB
'Linel' ,Odh,Oah,'$'
'Line2',Odh,Oah,'$'
'Line3',Odh,Oah,'$'
the output becomes
Linel
Line2
Line3
Chapter 5, The elements of an assembler program
127
Initializing with
expressions and labels
The initial value of an initialized variable must be a constant, but
it doesn't necessarily have to be a number. Expressions are fine:
TestVar DW
((924/2)+1)
as are labels:
. DATA
Buffer
BufferPointer
DW
DW
16 DUP (0)
Buffer
Whenever a label is used as an operand to a data definition
directive, it's the value of the label itself that's used, not the value
stored at that label. In the last example, the initial value of
BufferPointer is the offset in the .DATA segment of Buffer, not the
value zero that's stored at Buffer, much as if OFFSET Buffer had
been used to initialize BufferPointer. In other words, given the
previous initialization of BufferPointer, both
mov ax, OFFSET Buffer
and
mov ax, [BufferPointer]
load AX with the same value, the offset of Buffer.
Labels can be used in data definition expressions. For example,
the following code initializes the variable WordArrayLengfh to the
length in bytes of WordArray:
•DATA
WordArray
DW
WordArrayEnd
LABEL
WordArrayLength DW
50 DUP (0)
WORD
(WordArrayEnd - WordArray)
If you wanted to calculate the length of WordArray in words rather
than bytes, you could do it simply by dividing the length in bytes
by two:
WordArrayLengthlnWords DW
128
(WordArrayEnd - WordArray) / 2
Turbo Assembler User's Guide
Uninitialized data
Sometimes it doesn't make sense to assign an mitial value to a
memory variable. For instance, suppose your program reads the
next ten characters typed at the keyboard into an array named
KeyBuffer as follows:
mov
mov
GetKeyLoop:
mov
int
mov
inc
loop
cx,10
bx,OFFSET KeyBuffer
ah,l
21h
[bx),al
bx
GetKeyLoop
if of characters to read
ithe characters will be
i stored in Key Buffer
iDOS keyboard input function f
;get the next key pressed
isave the character
ipoint to storage location for next key
You could define KeyBuffer to be initialized with
Key Buffer
DB 10 DUP (0)
but that really doesn't make much sense, since the initial values in
KeyBuffer are immediately overwritten in GetKeyLoop. What you
really need is a way to define a memory variable as uninitialized,
and Turbo Assembler provides that capability with the question
mark (?).
The question mark tells Turbo Assembler you are reserving a
storage location, but not initializing it. For example, the proper
way to define KeyBuffer in the last example is like this:
KeyBuffer
DB 10 DUP (?)
This line reserves 10 bytes starting at the label KeyBuffer, but does
not set those bytes to any specific value.
Of course, whenever you use an uninitialized memory variable,
you must be sure to initialize it in your program before using it.
For instance, it would be a mistake to use the contents of KeyBuffer
in the last example before filling it, since the initial values stored
in KeyBuffer are not defined.
Chapter 5, The elements of an assembler program
129
Named memory
locations
So far, we've seen how to name memory locations by preceding a
data definition directive such as DB with a label. The LABEL
directive is another handy way to name a memory location,
without allocating any storage.
LABEL lets you specify both a label's name and its type without
having to define any data. For example, the following is another
way to define the array KeyBuffer used in the last example:
KeyBuffer
DB
LABEL BYTE
10 DUP (?)
The label types that can be defined with LABEL include
BYTE
WORD
DWORD
FWORD
PWORD
aWORD
TBYTE
NEAR
FAR
PROC
UNKNOWN
BYTE, WORD, DWORD, FWORD, PWORD, aWORD, and TBYTE
are self-explanatory, labeling 1-,2-,4-,6-,8-, and lO-byte data
items, respectively. Here's an example of initializing a memory
variable as a pair of bytes but accessing it as a word:
. DATA
WordVar LABEL
DB
WORD
1,2
.CODE
mov
ax, [WordVar]
When this code is executed, AL is loaded with 1 (the first byte of
WJrdVar), and AH is loaded with 2.
NEAR and FAR are used in code to select the type of call or jump
needed to reach a certain label. For example, here the first JMP is
a far jump (loading both CS and IP) because it is to a FAR label,
while the second jump is a near jump (loading only IP) because it
is to a NEAR label.
130
Turbo Assembler User's Guide
FarLabel and NearLabel
both describe the same
address, that of the MOV
Instruction, but allow you to
branch to that location In
two different ways.
•CODE
FarLabel
NearLabel
rnov
LABEL
LABEL
ax,l
FAR
NEAR
jrnp
FarLabel
jrnp
NearLabel
When you are using the simplified segment directives, PRoe is a
handy way to define a label in the appropriate size, near or far, for
the current code model. When the memory model is tiny, small,
or compact, LABEL PROe is the same as LABEL NEAR; when the
memory model is medium, large, or huge, LABEL PROe is the
same as LABEL FAR. This means that if you change the memory
model, you can change certain labels automatically as well.
For example, in
.IDDEL small
.CODE
EntryPoint
LABEL
PROC
EntryPoint is near, but if you change the memory model to large,
EntryPoint will become far. Normally, you will use the PRoe
directive (discussed in the section "Subroutines" on page 161),
rather than LABEL, to define the sort of entry points that you
would want to have change as the memory model changes;
however, sometimes you'll need more than one entry point into a
subroutine and then you'll need LABEL, as well as PROe.
Finally, we come to LABEL UNKNOWN. UNKNOWN is simply a
way of saying that you're not sure what data type a label is going
to be used as. If you're familiar with C, UNKNOWN is similar to
C's void type. As an example of UNKNOWN, suppose you have a
memory variable, TempVar, that's sometimes accessed as a byte
and sometimes accessed as a word. The following code does the
job by using LABEL UNKNOWN:
•DATA
TernpVar LABEL
UNKNOWN
Chapter 5, The elements of an assembler program
131
DB
?,?
• CODE
mov
[TempVar],ax
add
dl, [TempVar]
Moving data
Up to this point, you've learned a lot about the nature of assembly
language, fundamental assembler concepts, and the structure of
assembler programs. Now that you've got that solid foundation,
it's time to focus on assembler instructions, which form the part of
any assembler program that actually puts the 8086 through its
paces. Let's start with the most basic of assembler operationsmoving data.
MOV is the instruction that moves data on the 8086. Actually, MOV
is something of a misnomer; COpy might be more like it, since
MOV actually stores a copy of the source operand in the
destination operand, without affecting the source. For example,
mov
mov
mov
. ..
ax,O
bx,9
ax,bx
first stores the constant 0 in AX, then stores the constant 9 in BX,
and finally copies the contents of BX to AX as shown in these next
few diagrams.
After rnov ax,O:
AX
o
ex
?
After rnov bx, 9:
132
Turbo Assembler User's Guide
AX
o
BX
9
Mer mav ax, bx:
AX
9
BX
9
Note that the value 9 is not moved from BX to AX, but is rather
copied from BX to AX.
MOV accepts almost any pair of operands that makes sense except
when a segment register is an operand. (We'll discuss this
situation in the section "Accessing Segment Registers" on page
138.) Any of the following can be used for the source (right-hand)
operand to MOV:
•
•
•
•
a constant
an expression that resolves to a constant
a general-purpose register
a memory location accessed with any of the addressing modes
discussed in the section "Memory-addressing modes" on page
93
Either a general-purpose register or a memory location can be
used for the destination (left-hand) operand to MOV.
Selecting data
size
In assembly language, it's possible to copy byte or word values
with the MOV instruction. Let's look at how Turbo Assembler
determines what data size to work with.
In many cases, the operands to MOV tell Turbo Assembler exactly
what the data size should be. If a register is involved, then the
data size must be the size of that register. For example, the data
sizes of the following instructions are clear:
mov
mov
al,!
dx,si
Chapter 5, The elements of an assembler program
;byte-sized
;word-sized
133
mov
mov
bx, [di)
[bp+si+2),al
iword-sized
ibyte-sized
Likewise, named memory locations have inherent sizes, so the
data sizes of the following instructions are known to Turbo
Assembler:
•DATA
TestChar
TempPointer
DB
OW
?
TestChar
. CODE
mov
mov
[TestChar),'A'
[TempPointer),O
Sometimes, though, you'll have a MOV instruction that has no
defined size whatsoever. For example, there's no way Turbo
Assembler can be sure whether the following instruction should
store a byte- or word-sized value:
mov
[bx),l
and, in fact, Turbo Assembler will complain that it doesn't know
how to assemble such an instruction. It would also be handy to be
able to handle the case where you want to temporarily access a
word-sized variable as a byte, or vice versa.
Turbo Assembler gives you a means to flexibly define data size in
the form of the WORD PTR and BYTE PTR operators. WORD PTR
tells Turbo Assembler to treat a given memory operand as wordsized, and BYTE PTR tells Turbo Assembler to treat a given
memory operand as byte-sized, regardless of its predefined size.
For example, the last example could be made to store a wordsized value 1 to the word pointed to by BX with
mov
WORD PTR [bx),l
or could be made to store a byte-sized value 1 to the byte pointed
to by BX with
mov
BYTE PTR [bx),l
Note that WORD PTR and BYTE PTR make no sense when
applied to registers, since registers are always a fixed size; in this
case, WORD PTR and BYTE PTR are ignored. Similarly, WORD
134
Turbo Assembler User's Guide
PTR and BYTE PTR are ignored when applied to a constant,
which always takes on the same size as the destination operand.
WORD PTR and BYTE PTR have another use, which is to
temporarily select a different data size for a named memory
variable. Why would that be useful? Consider the following:
• DATA
Sourcel DO
Source2 DO
Sum
DD
l2345h
5432lh
?
. CODE
mov
mov
add
adc
mov
mov
ax, WORD PTR [Sourcel]
;get low word of
; Sourcel
dx,WORD PTR [Sourcel+2] ;get high word of
; Sourcel
ax, WORD PTR [Source2]
;add to Source2
; low word
dx,WORD PTR [Source2+2] ;add to Source2
; high word
WORD PTR [Sum],ax
;store low word of sum
WORD PTR [Sumt2],dx ;store high word of sum
The variables this example works with are all long integers or
doublewords. However, the 8086 can't perform doubleword
addition directly, so you have to break up the addition into a
series of word-sized operations. WORD PTR lets you access parts
of Sourcel, Source2, and Sum as words, even though the variables
themselves are doublewords.
While the FAR PTR and NEAR PTR operators don't strictly affect
data size, they are similar to WORD PTR and BYTE PTR. FAR PTR
forces a label that is the target of a jump or call to be treated as a
far label, causing the jump or call to load both CS and IP. NEAR
PTR, on the other hand, forces a label to be treated as a near label,
which is branched to by loading only IP.
Signed versus
unsigned data
Both signed and unsigned numbers are made up of a series of
binary digits. The distinction between the two is made by you, the
assembler programmer, not by the 8086 itself. For example, the
value OFFFFh can be either 65,535 or -1, depending on how your
Chapter 5, The elements of an assembler program
135
program chooses to interpret it. How do you know that OFFFFh is
-1? Add 1 to it,
mov
add
·..
ax,Offffh
ax,l
and you'll find that the result is 0, which is just what you'd expect
to get from adding -1 and 1 together.
The same ADD instruction works just fine whether you're
considering the operands to be signed or unsigned. For example,
suppose you were to subtract 1 from OFFFFh as follows:
· ..
·..
mov
sub
ax,Offffh
ax,l
The result would be OFFFEh, which is either 65,534 (as an
unsigned number) or -2 (as a signed number).
If this seems confusing, you should read one of the books
recommended at the end of this book in order to learn more about
two's complement arithmetic, the means by which the 8086 handles
signed numbers. Unfortunately, we haven't the space to cover
signed arithmetic here, although it's a useful subject for an
assembler programmer to understand. Right now, you just need
to know that ADD, SUB, ADC, and SBB work equally well with
signed and unsigned numbers, so no special instructions are
needed for signed addition and subtraction. Sign does matter for
multiplication and division, as you'll see later; it also matte~
when you're converting between data sizes and when you're
executing conditional jumps.
Converting
between data
sizes
Sometimes it's necessary to convert words to bytes, or vice versa.
This is one area where it matters whether the values are signed or
unsigned.
First, let's look at converting a word to a byte. That's simple; just
toss away the high byte of the word. For example,
mov
mov
136
ax,S
bl,al
Turbo Assembler User's Guide
converts the word value 5 in AX to the byte value 5 in BL. Of
course, you must be sure that the value you're converting will fit
in a byte; trying to convert 100h to a byte with
mov
mov
dx,lOOh
al,dl
would be fruitless, since only the lower byte, which is 0, would be
stored in AL.
Converting an unsigned byte to a word is simply a matter of
zeroing the upper byte of the word. For example,
mov
mov
mov
e!,12
al,e!
ah,O
converts the unsigned byte value 12 in CL to the unsigned word
value 12 in AX.
Converting a signed byte to a word is a bit more complex, so the
8086 provides you with a special instruction to handle that task:
CBW. CBW converts a signed byte in AL to a signed word in AX.
The following code converts the signed byte value -1 in OH to the
signed word value -1 in OX:
mov
mov
cbw
mov
dh,-l
al,dh
dx,ax
The 8086 also provides a special instruction, CWO, for converting
a signed word in AX to a signed doubleword in OX:AX (the high
word is in OX). The following converts the signed word value
+10,000 in AX to the signed doubleword value +10,000 in OX:AX:
mov
cwd
ax,lOOOO
Unsigned word values can be converted to unsigned doubleword
values by zeroing the high word of the value.
Chapter 5, The elements of an assembler program
137
Accessing
segment registers
Although the MOV instruction can be used to move values to and
from segment registers, this is a special case, more limited than
other uses of MOV. If a segment register is one operand to MOV,
the other operand must be a general-purpose register or a
memory location. It's not possible to load a constant directly into a
segment register, and one segment register may not be copied
directly to another segment register.
Since segment names are constants, it's necessary to load segment
registers by way of a general-purpose register or a memory
variable. For example, here are two ways to set ES to the .DATA
segment:
•DATA
DataSeg
ow
@data
•CODE
mov
mov
ax,@data
es,ax
mov
es, [DataSeg]
What you'd like to do, but can't, is this:
mov
es,@data
ithis won't work!
In order to copy the contents of one segment register to another
segment register, you have to pass the value through a generalpurpose register or memory. For example, both
mov
mov
· ..
and
· ..
· ..
ax,cs
ds,ax
push cs
pop ds
138
Turbo Assembler User's Guide
copy the contents of CS to OS. The first method executes faster,
but the second is smaller in code size.
It's worth noting that it's not only the MOV instruction that limits
you when it comes to the use of segment registers; most
instructions can't use segment registers as operands at all.
Segment registers can be pushed to and popped from the stack,
but that's about it; they can't be used in addition, subtraction,
logical opera tions, or comparisons.
Moving data to
and from the
stack
You've already encountered the stack, the last-in, first-out storage
area in the stack segment. The top of the stack is always pointed
to by SP. The MOV instruction can be used to access data on the
stack via memory-addressing modes that use BP as a base pointer;
for example,
mov
ax, [bp+4J
loads AX. with the contents of the word at offset BP+4 in the stack
segment. (See Chapter 4 for a discussion of accessing the stack via
BP.)
Most often, the stack is accessed with PUSH and POP. PUSH
stores the operand on top of the stack, and POP retrieves the
value on the top of the stack and stores it in the operand. For
example,
...
mov
push
pop
ax,l
ax
bx
pushes the value in AX. (which is 1) on top of the stack, then pops
1 from the top of the stack and stores it in BX.
Exchanging data
The XCHG instruction lets you swap the contents of two
operands. This is a convenient way to perform an operation that
would otherwise require three instructions. For example,
xchg
ax,dx
swaps the contents of AX. and DX, an operation that is equivalent
to
Chapter 51 The elements of an assembler program
139
. ..
push
mov
pop
ax
ax,dx
dx
I/O
So far, we've discussed moving data between constant values,
registers, and the memory address space of the 8086. As you'll
recall, the 8086 has a second, independent address space, known
as the input/output, or I/O, address space. The 65,536 I/O
addresses, or ports, are generally used as control-and-data
channels to devices such as disk drives, display adapters,
keyboards, and printers.
Most of the 8086's instructions, including MOV, can only access
operands in the memory address space. Only two instructions, IN
and OUT, can access I/O ports.
IN copies a value from a selected I/O port into AL or AX. The I/O
port address that serves as the source can be selected in one of
two ways. If the I/O port address is less than 256 (lOOh), you can
specify the address as part of the instruction; for example,
in
al,41h
copies a byte from I/O port 41h to AL.
Alternatively, you can use DX to point to the I/O port to be read:
mov
in
dx,41h
al,dx
Why bother using DX as an I/O pointer? For one thing, if the I/O
port address is greater than 255, you must use ox. For another,
the use of OX gives you more flexibility in addressing I/O ports;
for instance, a subroutine can use a passed I/O port pointer by
loading it into DX.
Don't be fooled by the syntax of the IN instruction; AL and AX are
the only possible destination operands. Likewise, OX and a
constant value less than 256 are the only possible source
operands. Much as you might like to, you can never use an
instruction like
in
140
bh,si
;this won't work!
Turbo Assembler User's Guide
OUT is exactly like IN, except that AL or AX is the source operand,
and an I/O port pointed to by DX or a constant value less than
256 is the destination operand. The following code sets I/O port
3B4h to OFh:
mov
mov
out
dx,3b4h
al,Ofh
dx,al
Operations
Data movement is certainly important, since a computer spends
much of its time moving data about from here to there. Still, it's
equally important to be able to manipulate the data by
performing arithmetic and logical operations on it. Next, we'll
take a look at the arithmetic and logical operations supported by
the 8086.
Arithmetic
operations
Even if your PC doesn't spend all its time crunching numbers,
you know that it could if you needed it to. After all, spreadsheets,
database programs, and engineering packages all run on the PC.
Given that, it's pretty obvious that the 8086 must be a powerful
math engine, right?
Well, yes and no. While it's certainly true that software that runs
on the 8086 can do wonderful math, the 8086 itself provides
surprisingly rudimentary arithmetic capabilities. For starters, the
8086 has no instructions to support any sort of floating-point
arithmetic (arithmetic with numbers such as 5.2 and 1.03E17, as
opposed to arithmetic with integers), let alone transcendental
functions; that's the job of the 8087 numeric coprocessor. This
doesn't mean that 8086 programs can't do floating-point
arithmetic; certainly, spreadsheets run on PCs without 8087s.
However, 8086 programs must perform floating-point arithmetic
by a slow, involved series of shift, add, and test instructions,
rather than with a single speedy instruction, as can be done with
the 8087.
Also, the 8086 provides no arithmetic or logical instructions that
can directly handle operands larger than 16 bits.
Chapter 5, The elements of an assembler program
141
What arithmetic operations does the 8086 have built-in support for,
then? Well, the 8086 can perform 8- and 16-bit signed and
unsigned addition, subtraction, multiplication, and division, and
has special, fast instructions for incrementing and decrementing
operands. The 8086 also provides support for addition and
subtraction of values larger than 16 bits, although operations on
such values require multiple instructions.
Addition and
subtraction
We've already encountered the ADD and SUB instructions in
many of our example programs. They operate much as you'd
expect. ADD adds the contents of the source (right-hand) operand
to the contents of the destination operand, and stores the result
back in the destination operand. SUB is the same except that it
subtracts the source operand from the destination.
So, for example, this code first loads the value 99 stored at BaseVal
into DX, then adds the constant 11 to it, resulting in the value 110
in OX, and finally subtracts the value 10 stored at Adjust from OX.
•DATA
BaseVal DW
99
Adjust DW
10
•CODE
mov
add
sub
dx, [BaseVal]
dx,ll
dx,[Adjust]
The result: 100 is stored in DX.
32-bit operands
ADD and SUB work with either 8- or 16-bit operands. If you want
to add or subtract, say, 32-bit operands, you must break the
operation into a series of word-sized operations and use ADC or
SBB.
When you add two operands, the 8086 stores a status in the carry
flag (the C bit in the flags register) that indicates whether there
was a carry out of the destination; that is, whether the result of the
addition was too large to fit in the destination. You're familiar
with the concept of carry-in decimal arithmetic; if you add 90 and
10, you get a carry-out to the third digit:
142
Turbo Assembler User's Guide
90
+ 10
100
Now consider this addition of two hexadecimal values:
FFFF
+ 1
10000
The lower word of the result is zero, and the carry is 1, since the
result, 10000h, doesn't fit into 16 bits.
ADe is just like ADD except that it takes the carry flag (which was
presumably set by a previous addition) into account. Whenever
you add two values that are larger than a word, add the lower
(least significant) words of the values together first with ADD,
then add the remaining words of the values together with one or
more ADC instructions, adding the most-significant words last.
For example, the following code adds a doubleword value stored
in CX:BX to a doubleword value stored in DX:AX:
add
adc
ax,bx
dx,cx
and the following adds the quad word value at DoubleLongl to the
quadword value at DoubleLong2:
mov
add
mov
adc
mov
adc
mov
adc
ax, [DoubleLong1]
[DoubleLong2],ax
ax, [DoubleLong1+2]
[DoubleLong2+2],ax
ax, [DoubleLong1+4]
[DoubleLong2+4],ax
ax, [DoubleLong1+6]
[DoubleLong2+6],ax
see operates along much the same lines as ADe. As see
performs a subtraction, it takes into account any borrow that
occurred during the previous subtraction. For example, the
following code subtracts a doubleword value stored in CX:BX
from a doubleword value stored in DX:AX:
Chapter 5, The elements of an assembler program
143
sub
sbb
ax,bx
dx,cx
With both ADC and SBB, you must make sure that the carry flag
hasn't changed since the last addition or subtraction, or else the
carry /borrow,status stored in the carry flag would be lost. For
instance, the following will not add CX:BX to DX:AX correctly:
add
sub
adc
ax,bx
si,si
dx,cx
. ..
iadd the lower words
iset S1 to 0 (clears the carry flag)
iadd the upper words •••
i this won't work properly, since the
i carry flag from the add has been destroyed!
Incrementing and decrementing
When an assembler program needs to perform an addition, odds
are good that it will be adding the value 1. This is known as
incrementing. Likewise, the value 1 is often subtracted from
registers and memory variables. This is known as decrementing.
For operations such as counting down or counting up, and for
advancing pointer registers through memory, incrementing and
decrementing are all the addition and subtraction that's needed.
In recognition of the frequent need for incrementing and
decrementing, the 8086 provides the instructions INC and DEC. As
you might expect, INC adds 1 to a register or memory variable,
and DEC subtracts 1 from a register or memory variable.
For example, the following code fills the 10-byte array T empArray
with the numbers 0, 1,2,3,4,5, 6, 7,8,9:
•DATA
TempArray
DB
FillCount
DW
10 DUP (?)
?
•CODE
mov al,O
mov bx,OFFSET TempArray
mov [FillCount),10
FillTempArrayLoop:
mov [bx),al
144
ifirst value to store
i in TempArray
ipoint BX to TempArray
if of elements to fill
iset the current element
Turbo Assembler User's Guide
inc bx
inc al
dec [FillCount]
jnz FillTempArrayLoop
; of TempArray
;point to next element of
; TempArray
;next value to store
;count down t of elements
; to fill
;do another element if we
; haven't filled all elements
Why would you want to use, say,
inc
bx
instead of
add
bx,l
since they do the same thing? Well, where the ADD is 3 bytes long,
the INC is only 1 byte long, and executes faster as well. In fact, it's
more compact to perform two INC instructions than to add 2 to a
word-sized register. (Increments and decrements of byte-sized
register and
INC instructions than to add 2 to a
word-sized register. (Increments and decrements of byte-sized
register and memory variables are 2 bytes long-still shorter than
adding or subtracting.)
In short, INC and DEC are the most efficient instructions available
for incrementing and decrementing registers and memory
variables. Use them whenever you can.
Multiplication and
division
The 8086 can perform certain types of integer multiplication and
division. This is one of the strong points of the 8086, since many
microprocessors provide no direct support at all for
multiplication and division, and it's fairly complex to perform
those operations in software.
The MUL instruction multiplies two 8- or 16-bit unsigned factors
together, generating a 16- or 32-bit product. Let's look at the 8-bitby-8-bit multiply first.
One of the factors to an 8-bit-by-8-bit MUL must be stored in AL;
the other may be in any 8-bit general-purpose register or memory
operand. MUL always stores the 16-bit product in AX. For
example,
mov
mov
mul
al,25
dh,40
dh
Chapter 5, The elements of an assembler program
145
multiplies AL times DH, placing the result, 1000, in AX. Note that
MUL only requires one operand; the other factor is always AL (or
AX, in the case of a 16-bit-by-16-bit multiply).
A 16-bit-by-16-bit MUL is similar; one factor must be stored in AX,
while the other may be in any 16-bit, general-purpose register or
memory operand. MUL puts the 32-bit product in DX:AX, with the
lower (least significant) 16 bits of the product in AX and the upper
(most significant) 16 bits of the product in DX. For instance,
rnov
rnul
ax,lOOO
ax
loads AX with 1000 and then squares AX, placing the result,
1,000,000, in DX:AX.
Unlike addition and subtraction, multiplication does care whether
the operands are signed or unsigned, so there's a second
multiplication instruction, IMUL, for multiplying 8- or 16-bit
signed factors. Apart from handling signed values, IMUL is just
like MUL. The code
rnov al,-2
rnov ah,lO
irnul ah
stores the value -20 in AX.
The 8086 lets you divide a 32-bit value by a 16-bit value, or a 16bit value by an 8-bit value, with certain restrictions. Let's look at
16-bit-by-8-bit division first.
In 16-bit-by-8-bit unsigned division, the dividend must be stored
in AX. The 8-bit divisor may be in any 8-bit, general-purpose
register or memory variable. DIV always puts the 8-bit quotient in
AL, and the 8-bit remainder in AH. For example,
rnov
rnov
div
ax,5l
dl,lO
dl
results in 5 (51 divided by 10) in AL and 1 (the remainder of 51
divided by 10) in AH.
146
Turbo Assembler User's Guide
Note that the quotient is an 8-bit value. This means that the result
of a l6-bit-by-8-bit division must be no larger than 255. If the
quotient is too large, an interrupt 0 (the divide-by-zero interrupt)
is generated. The code
mov
mov
div
ax,Offffh
bl,l
bl
generates a divide-by-zero interrupt. (As you might expect, a
divide-by-zero interrupt is also generated if zero is used as a
divisor.)
For 32-bit-by-16-bit division, the dividend must be stored in
OX:AX. The l6-bit divisor may be in any l6-bit, general-purpose
register or memory variable. DIV always puts the l6-bit quotient
in AX, and the l6-bit remainder in OX. For example,
mov
mov
mov
div
ax,2
dx,l
bx,10h
bx
;load DX:AX with 10002h
results in lOOOh (l0002h divided by lOh) in AX and 2 (the
remainder of lOO02h divided by 10h) in OX.
Again, the quotient is only a 16-bit value, so the result of a 32-bitby-16-bit division must be no larger than OFFFFh, or 65,535, else a
divide-by-zero interrupt is generated.
Like multiplication, division cares whether signed or unsigned
operands are used. DIV is used for unsigned operands, and IDIV is
used for signed operands. For example, this stores -6 in AX and
-67 in OX:
. DATA
TestDivisor
DW
100
•CODE
mov ax,-667
cwd
idiv [TestDivisor]
Chapter 5, The elements of an assembler program
;set DX:AX to -667
147
Changing sign
Finally, we come to the NEG instruction, with which you can
reverse the sign of the contents of a general-purpose register or
memory variable. For example, the code
mov
neg
mov
neg
. ..
ax,l
ax
bx,ax
bx
;set AX to 1
;negate AX, which becomes -1
;copy AX to BX
;negate BX, which becomes 1
ends up with -1 in AX and 1 in BX.
Logical
operations
Turbo Assembler supports a full set of instructions that perform
logical operations, including AND, OR, XOR, and NOT. These
instructions are very useful for manipulating individual bits
within a byte or word, and for performing Boolean algebra.
Given two source bits, the logical instructions produce the results
shown in Table 5.2. The logical instructions perform these bit-wise
operations on corresponding bits of the source operands; for
example,
and ax,dx
performs a logical AND with bit 0 of AX and bit 0 of OX as the
source bits and bit 0 of AX as the destination, and does the same
for bit 1, bit 2, and so on, up to bit 15.
Table 5.2
The operation of
the 8086 AND, OR.
and XOR logical
Instructions
Source Bit A
Source Bit B
0
0
1
1
1
0
0
0
0
1
1
0
AANDB
AORB
AXORB
0
0
1
1
1
1
1
0
The AND instruction combines two operands according to the
rules shown in Table 5.2, setting each bit in the destination to 1
only if both corresponding source bits are 1. AND lets you isolate a
specific bit, or force specific bits to O. For example,
. ..
mov
in
and
148
dx,3dah
al,dx
al,l
Turbo Assembler User's Guide
isolates bit 0 of the status byte of the Color /Graphics Adapter
(CGA). This code leaves AL set to 1 if display memory on the
CGA can be updated without causing snow, and set to 0
otherwise.
The OR instruction combines two operands according to the rules
shown in Table 5.2, setting each bit in the destination to 1 if either
of the corresponding source bits is 1. OR lets you force a specific
bit(s) to 1. For example,
mov
mov
mov
or
ax,40h
ds,ax
bx,lOh
WORD PTR [bx],0030h
forces both bit 5 and bit 4 of the BIOS equipment flag word to 1,
causing the BIOS to support the monochrome display adapter.
The XOR instruction combines two operands according to the
rules shown in Table 5.2 (page 148), setting each bit in the
destination to 1 only if one of the corresponding source bits is 0,
and the other is 1. This lets you flip the value of selected bits
within a byte. For example,
mov
xor
al,OlOlOlOlb
al,llllOOOOb
sets AL to 10100101b, or ASh. The key here is that when AL is
exc1usive-ORed with 11110000b, or OFOh, the 1 bits in OFOh flip the
value of the corresponding bits in AL, while the 0 bits in OFOh
leave the corresponding bits in AL unchanged. The result is that
all bits in the upper nibble of AL are changed, while all bits in the
lower nibble of AL remain the same.
By the way, XOR is a handy way to zero a register. For instance,
this code sets AX to 0:
xor ,ax,ax
Finally, NOT simply flips each bit in the operand to the opposite
state, just as if an XOR with a source operand of OFFh had been
executed. For instance, consider
mov
bl,lOllOOOlb
Chapter 5, The elements of an assembler program
149
not
xor
iflip BL to OlOOlllOb
iflip BL back to l0110001b
bi
bl,Offh
Shifts and rotates
The 8086 provides a variety of means by which to move bits left
or right in a register or memory variable. The simplest of these is
the logical shift.
SHL (shift left, also known as SAL) moves each bit in the
destination one place to the left, or toward the most-significant
bit. Figure 5.8 shows how the value 10010110b (96h or 150
decimal) stored in AL is shifted left with SHL AL,t. The result is
the value 00101100b (2Ch or 44 decimal), which is stored back in
AL. The carry flag is set to 1.
Figure 5.8
Example of a shift
left
AL
Carry
CH
Bit
~o
7
6
5
4
3
2
1
0
The most-significant bit is shifted out.of the operand altogether
and into the carry flag, and a 0 is shifted into the least-significant
bit.
Of what use is a left shift? The most common use of SHL is to
perform fast multiplies by powers of two, since each SHL
multiplies the operand by 2. For example, the following code
multiplies OX by 16:
shl
shl
shl
shl
dx,l
dx,l
dx,1
dx,1
iDX
iDX
iDX
iDX
*2
*4
*8
* 16
Multiplying by shifts is much faster than using the MUL
instruction.
You'll notice that there's a second operand to SHL in the previous
example, the value 1. This indicates that OX should be shifted left
150
Turbo Assembler User's Guide
by 1 bit. Unfortunately, the 8086 doesn't support 2, 3, or any
constant value other than 1 for a shift amount. However, CL can
be used to supply a shift count; for instance,
rnov
shl
cl,4
dx,cl
multiplies DX times 16, just as the last example did.
If there's a left shift, it seems logical that there must also be a right
shift, and there is-in fact, there are two right shifts.
SHR (shift right) is much like SHL: It shifts the bits in the operand
to the right, either by 1 or CL bits, then shifts the least-significant
bit into the carry flag and shifts 0 into the most-significant bit.
SHR is a quick way to do unsigned division by powers of two.
SAR (arithmetic shift right) is just like SHR, except that with SAR,
the most-significant bit of the operand is shifted right to the next
bit, and then back to itself. Figure 5.9 shows how the value
10010110b (96h or-106 in signed decimal) stored in AL is shifted
right with SAR AL,1. The result is the value 11001011b (OCBh or
-53 in signed decimal), which is stored back in AL. The carry flag
is set to O.
Figure 5.9
Example of SAR
(arithmetic right
shift)
Carry
~----------------------~~
Bit
7
6
5
4
3
2
1
o
This has the effect of preserving the sign of the operand, so SAR is
useful for performing signed division by powers of two. For
example,
rnov
sar
bx,-4
bx,l
leaves -2 stored in BX.
Chapter 5, The elements of an assembler program
151
There are also four rotate instructions: ROR, ROL, RCR, and RCL.
ROR is like SHR, except that the least-significant bit is shifted back
into the most-significant bit, as well as to the carry flag. Figure
5.10 shows how.the value 10010110b (96h or 150 decimal) stored
in AL is rotated right with ROR AL, 1. The result is the value
01001011b (04Bh or 75 in decimal), which is stored back in AL.
The carry flag is set to O.
Figure 5.10
Example of ROR
AL
(rotate right)
Bit
7
6
5
4
3
2
1
o
ROL reverses the action of ROR, shifting the operand in a circular
fashion, but to the left, with the most-significant bit shifting back
into the least-significant bit. ROR and ROL are useful for
realigning the bits in a byte or word. For example,
mov
mov
ror
si,49Flh
e!,4
si,e!
leaves 149Fh in 51, moving bits 3-0 to bits 15-12, bits 7-4 to bits 3-0,
and so on.
RCR and RCL are a bit different. RCR is like a right shift in which
the most-significant bit is shifted in from the carry flag. Figure
5.11 shows how the value 100101106 (96h or 150 decimal) stored in
AL is rotated right through the carry flag, which initially contains
the value 1, with ROR AL, 1. The result is the value 11001011b
(OCBh or 203 in decimal), which is stored back in AL. The carry
flag is set to O.
152
Turbo Assembler User's Guide
Figure 5.11
Example of RCR
(rotate right and
carry)
Bit
7
6
5
4
3
2
1
o
Likewise, RCL is like a left shift in which the least-significant bit is
shifted in from the carry flag. RCR and RCL are useful for shifts
involving multiple-word operands. For instance, the following
multip~ies the doubleword value in DX:AX by 4:
shl
rci
shl
rci
ax,l
dx,l
ax,l
dx,l
ibit 15 of AX is shifted into carry
icarry is shifted into bit 0 of DX
ibit 15 of AX is shifted into carry
icarry is shifted into bit 0 of DX
The rotate instructions, like the shift instructions, can shift an
operand either by 1 bit or by the number of bits specified by CL.
Loops and jumps
Up until now, you've seen th~ 8086 execute instructions in strict
sequence, with each instruction executing immediately after the
instruction at the preceding address. Given the code
mov
add
ax, [BaseCountj
ax,4
push ax
you could be very sure that the ADD would execute immediately
after the MOV, and the PUSH some time after that.
If that were all the 8086 could do, it would be a dull computer
indeed. A fundamental feature of any useful computer is the
presence of an instruction that can jump, or branch, to an
instruction other than the one following it in memory. Equally
Chapter 5, The elements of an assembler program
153
important is the ability to branch conditionally, depending on a
status or on the result of an operation. Naturally, the 8086 has
instructions for both sorts of branching; in addition, the 8086
provides special branching instructions to facilitate repeated
processing of a block of instructions.
Unconditional
jumps
The fundamental branching instruction of the 8086 is the JMP
instruction. JMP instructs the 8086 to execute the instruction at the
target label as the next instruction after the JMP. For example,
when this code is finished
rnov
jrnp
AddOneToAX:
inc
jrnp
AddTwoToAX:
inc
AXIsSet:
ax,l
AddTwoToAX
ax
AXIsSet
ax
AX contains 3, and the ADD and JMP instructions following the
label AddOneToAX are never executed. Here, the instruction
jrnp
AddTwoToAX
instructs the 8086 to set IP, the instruction pointer, to the offset of
the label AddTwoToAX, so the next instruction executed is
add
ax,2
An operator sometimes used with JMP is SHORT. JMP usually
uses a 16-bit displacement to point to the destination label;
SHORT instructs Turbo Assembler to use an 8-bit displacement
instead, thereby saving 1 byte per JMP. For instance, the last
example is 2 bytes shorter as
rnov
jrnp
AddOneToAX:
inc
jrnp
AddTwoToAX:
inc
154
ax,l
SHORT AddTwoToAX
ax
SHORT AXIs Set
ax
Turbo Assembler User's Guide
AXIsSet:
The drawback to using SHORT is that short jumps can only reach
labels within 128 bytes of the JMP instruction, so in some cases
Turbo Assembler can inform you that it can't reach a given label
with a short jump. It only makes sense to use SHORT on forward
jumps, since Turbo Assembler automatically makes backward
jumps short if a short jump will reach the destination, and long
otherwise.
JMP can also be used to jump to another code segment, loading
both CS and IP with a single instruction. For example,
CSegl
SEGMENT
ASSUME cs:Csegl
FarTarget
LABEL
Csegl
ENDS
Cseg2
SEGMENT
ASSUME cs:Cseg2
jmp
Cseg2
FAR
FarTarget
ithis is a far jump
ENDS
performs a far jump.
If you wish, you can use the FAR PTR operator to force a label to
be treated as far; for instance,
jmp
nop
NearLabel:
FAR PTR NearLabel
performs a far jump to NearLabel, even though NearLAbel is in the
same code segment as the JMP instruction.
Finally, you can jump to an address stored in a register or
memory variable. For example,
mov
jmp
ax,OFFSET TestLabel
ax
Chapter 5, The elements of an assembler program
155
TestLabel:
branches to TestLabel, as does
•DATA
JumpTarget
DW
TestLabel
•CODE
jmp
[JumpTarget]
TestLabel:
Conditional jumps
Jumps such as those described in the last section are only part of
what you need to write useful programs. You really need to be
able to write code that's capable of making decisions, and that's
what the conditional jumps give you.
A conditional jump instruction can either branch or not to a
destination label, depending on the state of the flags register. For
example, consider the following:
mov
int
cmp
je
mov
AWasTyped:
push
ah,l
21h
aI, 'A'
AWasTyped
[TempByte],al
iDOS keyboard input function
iget the next key press
iwas capital "A" pressed?
iyes, handle it specially
ino, store the character
ax
isave the char on the stack
First, this code gets a key press by way of a DOS function. Then it
uses the CMP instruction to compare the character typed to the
character A. The CMP instruction is like a SUB that doesn't affect
anything; the whole purpose of CMP is to let you compare two
operands without changing them. CMP does, however, set the
flags just as SUB would. So, in the preceding code the zero flag is
set to 1 only if AL con tains the character A.
Now we come to the crux of the example. JE is a conditional jump
instruction that branches only if the zero flag is 1. Otherwise, the
156
Turbo Assembler User's Guide
instruction immediately following JE, in this case a MOV
instruction, is executed. The zero flag will be set in the previous
example only if the A key is pressed, and only then will the 8086
branch to the PUSH instruction at the label A Was Typed.
The 8086 provides a remarkable variety of conditional jumps,
giving you the ability to branch on just about any flag or
combination of flags you could imagine (and several more
besides). You can jump conditionally on the state of the zero,
carry, sign, parity, and overflow flags, and on the combination of
flags that indicate the results of operations with signed numbers.
Table 5.3 summarizes the conditional jump instructions.
Table 5.3
Conditional jump
instructions
Name
Meaning
Rags Checked
JB/JNAE
Jump if below
Jump if not above or equal to
CF=l
JAElJNB
Jump if above or equal to
Jump if not below
CF=O
JBElJNA
Jump if below or equal to
Jump if not above
CF=lorZF=l
JAlJNBE
Jump if above
Jump if not below or equal to
CF=O and ZF=O
JElJZ
Jump if equal to
ZF=l
JNElJNZ
Jump if not equal to
ZF=O
JUJNGE
Jump if less than
Jump if not greater than or equal to
SF~F
JGElJNL
Jump if greater than or equal to
Jump if not less than
SF=OF
JLElJNG
Jump if less than or equal to
Jump if not greater than
ZF=l
JG/JNLE
Jump if greater than
Jump if not less than or equal to
ZF=O or SF=OF
JP/JPE
Jump if parity
Jump if parity even to
PF=l
JNP/JPO
Jump if no parity
Jump if parity odd
PF=O
JS
Jump if sign
SF=l
JNS
Jump if not sign
SF=O
JC
Jump if carry
CF=l
JNC
Jump if not carry
CF=O
JO
Jump if overflow
OF=l
Chapter 5, The elements of an assembler program
orSF~F
157
Table 5.3: Conditional Jump Instructions (continued)
Jump if not overflow
JNO
OF=O
CF =carry flag; SF =sign flag; OF =overflow flag; ZF =zero flag; PF =parity flag
For more information about synonyms and the conditional jump
instructions in general, consult Chapter 6, which also provides
detailed information about the ways in which each 8086
instruction can modify the flags register.
Flexible as they are, the conditional jump instructions have a
serious limitation: They are always short jumps. In other words,
the destination label for a conditional jump instruction must be
within 128 bytes of the instruction.
For example, Turbo Assembler can't assemble
.. .
JumpTarget:
DB
1000 DUP (?)
dec
jnz
ax
JumpTarget
since /umpTarget is over 1000 bytes away from the JNZ
instruction. This is what's needed in a case like this:
JumpTarget:
DB
dec
jz
jmp
SkipJump:
1000 DUP (?)
ax
SkipJump
JumpTarget
Here, a conditional jump is used to make the decision about
whether to make a long unconditional jump.
Looping
One sort of programming construct that can be built with
conditional jumps is the loop. A loop is nothing more than a block
of code that ends with a conditional jump, so that the code can be
158
Turbo Assembler User's Guide
executed repeatedly until a tennination condition is reached. You
might be familiar with looping constructs such as for and while in
C, while and repeat in Pascal, and FOR in BASIC.
What are loops used for? They're used to manipulate arrays, test
the status of I/O ports until a certain state is reached, clear blocks
of memory, read strings from the keyboard, display strings on the
screen, and more. Loops are the basic means of handling anything
that requires repeating a given action. As such, they're used
frequently; so frequently, in fact, that the 8086 provides several
special instructions just for looping: LOOP, LOOPE, LOOPNE, and
JCXZ.
Let's look at LOOP first. Suppose you wanted to print the 17
characters in the string TestString. You could do it with this code:
•DATA
TestString
DB
'This is a test ••• '
. CODE
mov
cx,1?
mov
bx,OFFSET TestString
PrintStringLoop:
mov
dl, [bx]
;get the next character
;point to the following char
bx
inc
;DOS display output fn t
mov
ah,2
;invoke DOS to print character
int
21h
;count down the string length
dec
cx
jnz
PrintStringLoop ;do the next character,
; if any remain
However, there's a better way. You may remember that earlier we
noted that CX is especially useful for code that loops. Here's how
loop PrintStringLoop
does just what
dec
jnz
cx
PrintStringLoop
does, and does it faster and in one less byte. Whenever you have a
loop that repeats until a counter reaches 0, just keep the count in
CX and use the LOOP instruction.
.
Chapter 5, The elements of an assembler program
159
What about loops that have more complex termination conditions
than a simple counter counting down? LOOPE and LOOPNE
provide for two such cases.
LOOPE does the same thing LOOP does, except LooPE will end
the loop (fail to branch) if either ex counts down to 0 or the zero
flag is set to 1. (Remember that the zero flag is set to 1 if the last
arithmetic result was 0, or if the two operands to the last
comparison were not equal.) Similarly, LOOPNE will end the loop
if either ex counts down to 0 or the zero flag is cleared to O.
Imagine you want to repeat a loop, saving key presses, until
either the Enter key has been pressed or 128 characters have been
read and stored. The following code uses LOOPNE to do the job:
•DATA
Key Buffer
DB
12B DUP (?)
•CODE
mov
mov
cx,12B
bx,OFFSET KeyBuffer
mov
int
mov
inc
cmp
loopne
ah,1
21h
[bx),al
bx
al,Odh
KeyLoop
KeyLoop:
;DOS keyboard input function t
;read the next key
;store the key
;set pointer for next key
;was it the enter key?
;if not, get another key, unless we've
; already read the maximum t of keys
LOOPE is also known as LooPZ, and LOOPNE is also known as
LOOPNZ, just as JE is also known as JZ.
There's one more loop-related instruction, and that's JCXZ. JCXZ
branches only if ex is 0; this is a useful way to test ex before
beginning a loop. For example, the following code, which is called
with BX pointing to a block of bytes to be set to 0 and ex
containing the length of the block, uses JCXZ to skip the entire
loop if ex is 0:
jcxz
ClearLoop:
mov
inc
160
SkipLoop
BYTE PTR [si),O
si
;if CX is 0, there's nothing to do
°
;set the next byte to
;point to the next byte to clear
Turbo Assembler User's Guide
loop
SkipLoop:
ClearLoop
iclear the next character, if any
Why is it desirable to skip the loop if ex is O? Well, if you execute
LOOP with CX equal to 0, CX is decremented to OFFFFh, and the
LOOP instruction branches to the destination label. Then the loop
is executed 65,535 more times! What you want here is a ex setting
of 0 to mean that no bytes are to be zeroed, not 65,536 bytes. JCXZ
lets you test for that case quickly and efficiently.
There are a couple of interesting notes about the looping
instructions. First, be aware that a looping instruction, like a
conditional jump, can only branch to a label within a range of
about 128 bytes before or after the looping instruction. Loops
larger than about 128 bytes require use of the "conditional jump
around an unconditional jump" technique described in the
previous section, "Conditional Jumps" (page 156). Second, it's
important to realize that none of the looping instructions affect
the flags in any way. This means that
loop LoopTop
isn't exactly the same as
dec
jnz
cx
LoopTop
since DEC alters the overflow, sign, zero, auxiliary carry, and
parity flags, while LOOP alters no flags at all. By the same token,
the DEC instruction isn't exactly the same as
sub
jnz
cx,l
LoopTop
since SUB affects the carry flag, while the DEC instruction does
not. True, these are small differences, but it's important to
understand the instruction set thoroughly when programming in
assembly language.
Subroutines
So far, we've only looked at programs consisting of a single long
chunk of code. Each program has started at the top of the code,
executed each instruction in turn (with an occasional detour for
looping or decision-making), and then ended at the bottom of the
Chapter 5. The elements of an assembler program
161
code. That's fine for small programs, but larger programs require
a programming construct known as a subroutine.
You're probably familiar with subroutines from a high-level
language. In C, subroutines are known as functions, and in Pascal
and Basic, they're known as procedures and functions.
Subroutines, procedures, and functions all amount to the same
thing-a separate section of code that accepts well-defined inputs,
performs a certain action, and optionally returns a specific result
value.
Subroutines let you build programs in a modular fashion, with
the subroutines hiding the details of specific tasks so you can
focus on the overall flow of the program. Subroutines can also
make programs far more compact, since a single subroutine can
be called from many places in a program, and can even perform
different functions when passed different values. In a large
program (whether written in assembler, C, Pascal, or some other
language), subroutines are essential to creating orderly,
maintainable code.
How subroutines
work The fundamental operation of a subroutine is illustrated by Figure
5.12.
Figure 5.12
Operation of a
subroutine
(IP is loaded with 1110,
and 1007 is pushed on
the stack)
1000
moval,1
DoCalc:1110
1002
movbl,3
1112
add al,bl
1004
call DoCalc
1114
and al,7
1116
add al,'O'
1007
1009
movah,2
Int 21h
..
L
(The value
1118
on top of the
stack, 1007, is popped
into IP)
shl ai, 1
ret
The code that calls the subroutine executes a CALL instruction,
. which pushes the address of the next instruction onto the stack
and then loads IP with the address of the desired subroutine,
thereby branching to the subroutine. The subroutine then
executes just as any other code would. Subroutines can-and
often do-contain calls to other subroutines; in fact, properly
162
Turbo Assembler User~ Guide
designed subroutines can even call themselves, a practice known
as recursion.
When the subroutine has finished its task, it executes a RET
instruction, which pops into IP the address pushed by the original
CALL instruction. This causes execution of the calling routine to
resume at the instruction following the CALL instruction.
For example, the following program prints the three strings:
Hello, world!
Hello, solar system!
Hello, universe!
by using the subroutine PrintString:
.IDDEL small
•STACK 200h
•DATA
WorldMessage
DB
'Hello, world!' ,Odh,Oah,O
SolarMessage
DB
'Hello, solar system!' ,Odh,Oah,O
UniverseMessage DB
'Hello, universe!' ,Odh,Oah,O
•CODE
ProgramStart
PRoe
NEAR
mov ax,@data
mov ds,ax
mov bx,OFFSET WorldMessage
call PrintString
iprint Hello, world!
mov bx,OFFSET SolarMessage
call PrintString
iprint Hello, solar system!
mov bx,OFFSET UniverseMessage
call PrintString
iprint Hello, universe!
mov ah,4ch
iDOS terminate program fn f
int 21h
i ••• and done
ProgramStart ENDP
Subroutine to print a null-terminated string on the screen.
Input:
DS:BX - pointer to string to print.
Registers destroyed: AX, BX
PrintString
PROC
NEAR
PrintStringLoop:
mov dl, [bx]
iget the next char of the string
and dl,dl
;is the character's value zero?
jz
EndPrintString iif so, then we're done with the
i string
inc bx
ipoint to the next character
Chapter 5, The elements of an assembler program
163
mov ah,2
int 21h
jmp PrintStringLoop
EndPrintString:
ret
PrintString
ENDP
END ProgramStart
iDOS display output function
iinvoke DOS to print the char
iprint the next char, if any
;return to calling program
There are two things to note here. First, PrintString is not hardwired to print a specific string, but rather prints whatever string
the calling program points to by way of BX. Second, two new
directives, PROC and ENDP, are used to bracket PrintString.
PROC is used to start a procedure. The label associated with
PROC, in this case PrintString, is the name of the procedure, just
as if
Print String
LABEL PROC
had been used. PROC does something more, though: It specifies
whether near or far RET instructions should be used within that
procedure.
Let's take a moment to examine the implications of that last
statement. Recall that when a near label is branched to, IP is
loaded with a new value. When a far label is branched to, both CS
and IP are loaded. If a CALL instruction references a far label,
both CS and IP are loaded, just as with a jump.
It stands to reason, then, that both CS and IP must be pushed
when a far call occurs; otherwise, how could a RET instruction
have enough information to return to the calling code? Think of it
this way: If a far call loaded CS and IP, but pushed only IP, then a
return could only load IP from the top of the stack. The result of
the RET would be a CS:IP consisting of the CS of the called
routine paired with the IP of the calling routine, which clearly
makes no sense.
Instead, what happens is that both CS and IP are pushed by a call
to a far label. How, though, will Turbo Assembler know what
type of returns, far or near, to generate in a given subroutine? One
way is for you to specify the type of each return explicitly, with
the RETN (near return) and RETF (far return) instructions.
However, a better answer lies with the PROC and ENDP
directives.
164
Turbo Assembler User's Guide
The ENDP directive marks the end of subroutines that start with
PROC directives. A given ENDP marks the end of the subroutine
that started with PROC and the same label. For example,
TestSub
PROC NEAR
TestSub
ENDP
marks the beginning and end of the subroutine TestSub.
PROC and ENDP don't actually generate any code; they're
directives, not instructions. What they do do is control the type of
RET instructions used in a given subroutine.
If the operand to a PROC directive is NEAR, then all RET
instructions between that PROC directive and the corresponding
ENDP directive are assembled as near returns. If, on the other
hand, the operand to a PROC directive is FAR, then all RET
instructions within that procedure are assembled as far returns.
So, for example, to change the type of all RET instructions in the
TestSub example to far, change the PROC directive to
TestSub
PROC
FAR
In general, it's best to use near subroutines whenever possible,
since far calls are larger and slower than near calls, and far returns
are slower than near returns. However, far subroutines become
necessary when you need more than 64K of program code.
If you're using the simplified segment directives, it's better still to
use the PROC directive without any operand at all, as in
TestSub
Small is the memory model
default.
PROC
When Turbo Assembler encounters such a directive, it
automatically makes the procedure near or far according to the
memory model selected with the .MODEL directive. Tiny-, small-,
and compact-model programs have near calls, while medium-,
large-, and huge-model programs have far calls. For example, in
.MJDEL small
TestSub PROC
TestSub is near-callable, while in
Chapter 5, The elements of an assembler program
165
.IDDEL large
TestSub PROC
TestSub is far-callable.
Parameter
passing
Information is often passed to subroutines by the code that calls
them (referred to as the "calling code"). For instance, the example
program in the last section used the BX register to pass a pointer
to the PrintString subroutine. This action is known as parameterpassing, where the parameters tell the subroutine exactly what to
do.
There are two commonly used ways to pass parameters: in the
registers and on the stack. Register-passing is often used by pure
assembler code, while stack-passing is used by most high-level
languages, including Pascal and C, and by assembler subroutines
called by those languages.
When register-passing,
carefully comment each
subroutine as to which
parameters It expects to
receive and In which
registers they should be
placed.
Chapters 7 and 8 provide
details about the
parameter-passing
conventions of Turbo C and
Turbo Pascal and provide
sample assembler code.
Passing parameters in registers is as simple as it sounds-just put
the parameter values in the appropriate registers and call the
subroutine. Each subroutine can have its own parameter
requirements, although you'll find it easiest to establish some
conventions and stick with them in order to avoid confusion. For
example, you might want to make it a rule that the first pointer
parameter is always passed in BX, the second in SI, and so on.
Passing parameters on the stack is a bit more complex. If you use
stack-passing, you'll probably want to use the convention
established by your favorite high-level language to easily link
your assembler subroutines to code written in that language.
Returning values
Chapters 7 and 8 provide
the details of the return-value
conventions of Turbo C and
Turbo Pascal.
Subroutines often return values to the calling code. In assembler
subroutines that are going to be called from a high-level language,
you must follow that language's conventions for returning values.
For example, C-callable functions must return 8- and 16-bit values
(chars, Ints, and near pointers) in AX, and 32-bit values (longs and
far pointers) in DX:AX.
In pure assembler code, you have complete freedom about how to
return values; you can put them in any register you wish. In fact,
166
Turbo Assembler User's Guide
subroutines can even return status information in the flags
register, in the form of carry or zero flag settings. However, once
again, it's best to establish and follow some conventions. One
useful convention is to return 8-bit values in AL and 16-bit values
in AX; that way, you'll get in the habit of not expecting valuable
information in AX to remain unchanged by calls.
The major problem with using subroutine return values in
assembler is that, in the course of returning information,
subroutines may destroy information that's important to the
calling routine. In assembler, it's easy to code a call to a subroutine
without remembering that the subroutine returns a value in, say,
51 (or that the subroutine simply alters 51); then you've got a
program bug that might be hard to find.
For this reason, it's best to keep the number of values a subroutine
returns in the registers to a minimum-preferably no more than
one-and to return additional values by storing them at memory
locations indicated by passed pointers, as both C and Pascal do.
Preserving
registers
Preserving registers properly during subroutine calls is, in
general, a major problem of assembler programming. In modem
high-level languages, a subroutine normally can't modify the
calling code's variables unless the calling code explicitly makes
that possible. Not so in assembler, where the calling code's
variables are often stored in the same registers that the subroutine
uses. For example, if a subroutine modifies a register that the
calling code sets before the call but uses after the call, you've got a
bug.
One solution to this problem is that as each subroutine is entered,
it always pushes all the registers that it uses, and then restores the
registers by popping them before returning to the calling code.
Unfortunately, this is time-consuming and requires a considerable
amount of code. Another option is to make it a rule that calling
code should never expect subroutines to preserve registers, and
so should always preserve any registers it cares about. This is
unattractive because a large part of the reason to use assembler is
the freedom to use registers efficiently.
In short, there's a conflict between speed and ease of coding in
assembly language. If you're going to use assembler, you might as
well write fast, compact code, and that means being intelligent
about register preservation and playing an active part in making
Chapter 5. The elements of an assembler program
167
sure each subroutine call produces no register conflicts. Your best
approach is to comment each subroutine carefully as to the
registers it destroys, and then refer to those comments each time
you use a CALL instruction.
The sort of attention to detail involved both in keeping an eye on
register preservation and in using registers as effectively as
possible is an important part of good assembly language
programming. High-level languages do those things for you-but
then again, high-level languages can't create programs as fast and
compact as those you're going to write in assembler.
An example assembly language program
Let's put together what you've learned so far. This example
program, WCOUNT.ASM, counts the number of words in a file
and displays the count on the screen.
Program to count the number of words in a file. Words are
delimited by whitespace, which consists of spaces, tabs,
carriage returns, and linefeeds.
Usage: wc
assigns the text string "TABLE_OFFSET+2" to INDEX_START,
while
TABLE OFFSET
INDEX START
EQU 1
EQU TABLE_OFFSET+2
assigns the value 3 (the result of 1 + 2) to INDEX_START. In
general, it's a good practice to put angle brackets around text
string operands to EQU to make sure those operands aren't
evaluated as expressions by accident.
Once a given label is equated to a value or text string with EQU in
a given source module, it can never be redefined in that module.
The following is guaranteed to produce an error:
X
EQU 1
X
EQU 101
If you need to redefine equated labels (and there are, on occasion,
some very good reasons to do so), you'll need to use the
directive, which we'll discuss shortly.
=
The $ predefined
symbol
178
Recall that Turbo Assembler offers several predefined symbols,
such as @data. Another simple but surprisingly useful predefined
symbol is $, which is always set to the current value of the
location counter. In other words, $ is always equal to the current
Turbo Assembler User's Guide
$ can be used In expressions,
or anywhere else a constant
maybe used.
offset in the segment that Turbo Assembler is currently
assembling into. $ is a constant offset value, just as OFFSET
MemVaris.
$ is particularly handy for calculating data and code lengths. For
example, suppose you want to equate the symbol
STRING_LENGTH to the length in bytes of a string. Without $,
you'd have to do the following:
StringStart
LABEL
BYTE
db Odh,Oah,'Hello, world' ,0dh,Oah
String End
LABEL
BYTE
STRING_LENGTH EQU (StringEnd-StringStart)
with $, though, all you need is
StringStart
LABEL
BYTE
db Odh,Oah,'Hello, world' ,0dh,Oah
STRING_LENGTH EQU (S-StringStart)
Here's how you'd calculate the length in words of an array of
words:
WordArray DW 90h, 25h, 0, 1Gh, 23h _
WORD ARRAY LENGTH EQU (($-WordArray)/2)
Of course, you could count the individual elements by hand, but
with longer arrays and strings, that would quickly become
tedious.
Incidentally, three other useful predefined variables are ??date,
??tlme, and ??filename. ??date contains the date of assembly, as
a quoted text string in the form 01/02/87. The ??tlme variable
contains the time of assembly in the form 13:45:06, and
??filename contains the name of the file being assembled in the
form of an 8-character quoted text string such as IITEST.ASM".
The = directive
=
The directive is like the EQU directive in all respects save one:
Where labels defined with EQU can never be redefined (an error
occurs if they are), labels defined with = can be redefined freely.
Chapter 6, More about programming In Turbo Assembler
179
This is very useful for labels that need to be changed on the fly, or
that are reused within a single source module.
For example, the following code uses
for the first 100 multiples of 10:
•DATA
MultiplesOf10 LABEL
TEMP = 0
REPT 100
DW TEMP
TEMP = TEMP+10
ENDM
=to generate a lookup table
WORD
iBX is I to multiply by 10.
i Shift left to mUltiply * 2
i for lookup in word-sized table
mov ax, [MultiplesOf10+bx] iget the number * 10
shl bx,l
=
All operands to must resolve to a numeric value; unlike EaU,
cannot be used to assign text strings to labels.
=
The string instructions
We've come to the most unusual and powerful instructions of the
8086-the string instructions. String instructions are like no other
8086 instructions in that they can both access memory and
increment or decrement a pointer register in a single instruction.
A single string instruction can access memory as many as 130,000
times!
As their name implies, string instructions are particularly useful
for manipulating text strings. String instructions are equally adept
at handling arrays, data buffers, and any sort of string of bytes or
words. You should strive to use the string instructions whenever
possible, since they are, as a rule, shorter and faster than
equivalent combinations of normal 8086 instructions such as MOV,
INC, and LOOP.
We'll examine the string instructions in two functional groups: the
string instructions used for data movement (LODS, STOS, and
MOVS), and the string instructions used for data scanning and
comparison (SCAS and CMPS).
180
Turbo Assembler User's Guide
Data movement
string instructions
LODS
The data movement string instructions are much like the MOV
instruction, but do more than MOV and operate faster. We'll look
at LOOS first. Note that the direction flag controls the direction in
which pointer registers are changed for all string instructions.
LOOS, which loads a byte or word from memory into the
accumulator, comes in two flavors, LOOSB and LOOSW. LOOSB
loads the byte addressed by DS:SI into AL, and either increments
or decrements SI, depending on the state of the direction flag. If
the direction flag is 0 (set with CLO), then SI is incremented, and if
the direction flag is 1 (set with STO), then SI is decremented. This
is not true only of LOOSB; the direction flag controls the direction
in which pointer registers are changed for all string instructions.
For example, the LOOSB in ~he following code,
cld
mov si,O
lodsb
loads AL with contents of the byte at offset 0 in the data segment
and increments SI to 1. That's equivalent to
mov si,O
mov aI, lsi]
inc si
However,
lodsb
is considerably faster (and 2 bytes smaller) than
mov aI, lsi]
inc si
LOOSW is just like LOOSB, save that the word addressed by DS:SI
is loaded into AX, and SI is either incremented or decremented by
2, rather than 1. For example,
std
Chapter 6, More about programming In Turbo Assembler
181
mov si,10
lodsw
. ..
loads the word at offset 10 in the data segment into AX, then
decrements 51 by 2 to 8.
STOS
STOS is the complement to LODS, writing a byte or word value in
the accumulator to the memory location pointed to by E5:DI, and
incrementing or decrementing DI. STOSB writes the byte iI:t AL to
the memory location E5:DI, then increments or decrements DI,
depending on the direction flag. For example,
std
mov di,Offffh
mov al,55h
stosb
writes the value 55h to the byte at offset OFFFFh in the segment
pointed to by E5, then decrements DI to OFFFEh.
STOSW does much the same, writing a word value in AX to
address ES:DI, then incrementing or decrementing DI by 2. For
instance,
cld
mov di,Offeh
mov ax,102h
stosw
...
writes the word value 102h in AX to offset OFFEh in the segment
pointed to by E5, then increments DI to 10OOh.
LODS and STOS work nicely together for copying buffers. For
example, the following subroutine copies the zero-terminated
string at DS:5I to the string at ES:DI:
; Subroutine to copy one zero-terminated string to another.
; Input:
DS:SI - string to copy from
ES:DI - string to copy to
; Output: None
182
Turbo Assembler User's Guide
Registers destroyed: AL, SI, DI
CopyString
cld
PROC
CopyStringLoop:
lodsb
stosb
cmp al,O
jnz CopyStringLoop
ret
CopyString
ENDP
imake SI and DI increment with string
i instructions
iget source string character
istore char in destination string
iwas the char zero to end the string?
ino, do next character
iyes, done
You could equally well use LODS and STOS to copy blocks of
bytes that aren't zero-terminated with a loop like
mov cx,ARRAY_LENGTH_IN_WORDS
mov si,OFFSET SourceArray
mov ax,SEG SourceArray
mov ds,ax
mov di,OFFSET DestArray
mov ax,SEG DestArray
mov es,ax
cld
CopyLoop:
lodsw
stosw
loop Copy Loop
However, there's an even better way to move a byte or word from
one memory location to another, and that's with the MOVS
instruction.
MOVS
MOVS is like LODS and STOS rolled into one. MOVS reads the
byte or word stored at DS:SI, then writes that value to the address
ES:DI. The byte or word never passes through a register at all, so
AX isn't modified. MOVSB is as short as any instruction can be, at
only 1 byte long, and is even faster than the LODS/STOS
combination. With MOVS, the last example becomes still faster:
mov
mov
mov
mov
cx,ARRAY_LENGTH_IN_WORDS
si,OFFSET SourceArray
ax,SEG SourceArray
ds,ax
Chapter 6. More about programming In Turbo Assembler
183
mov di,OFFSET DestArray
mov ax,SEG DestArray
mov es,ax
cld
CopyLoop:
movsw
loop CopyLoop
Repeating a string
instruction
While the code in the last example looks pretty efficient, you may
well be thinking that what you'd really like to do is get rid of that
LOOP instruction and move the whole array with a single
instruction. You're in luck-the 8086 gives you that option with
the string instructions in the form of the REP prefix.
REP isn't an instruction; instead, it's an instruction prefix.
Instruction prefixes modify the operation of the following
instruction. What REP does is tell the following string instruction
to execute repeatedly until the ex register reaches zero. (If ex is
zero when the repeated instruction begins, the instruction
executes zero times-in other words, it doesn't do anything at all.)
Using REP, you can replace
CopyLoop:
movsw
loop Copy Loop
in the last example with
rep movsw
That single instruction will move a block of as many as 65,535
words (OFFFFh) from memory starting at DS:SI to memory
starting at ES:DI.
Of course, a string instruction repeated 65,535 times doesn't
execute anywhere near as quickly as an instruction executed once;
all those memory accesses take time. However, each repetition of
a repeated string instruction executes more quickly than would a
single instance of that string instruction, making repeated string
instructions a very fast way to read from, write to, or copy
memory.
REP can be used with LODS and STOS as well as with MOVS (and
also with the SCAS and CMPS instructions, which we'll discuss
next). It's useful to repeat STOS to clear or fill blocks of memory;
for example,
184
Turbo Assembler User's Guide
cld
mov
mov
mov
sub
mov
rep
ax,SEG WordArray
es,ax
di,OFFSET WordArray
aX,ax
cx,WORD_ARRAY_LENGTH
stosw
fills l'\brdArray with zeros. There's no correspondingly useful
application for repeating LODS.
REP can only cause string instructions to repeat. An instruction
like
rep mov aI, [bx]
which doesn't make a whole lot of sense anyhow, ignores the REP
prefix and executes as a plain old
mov aI, [bx]
String pOinter overrun
Data scanning
string instructions
seAS
Note that when a string instruction is executed, it increments or
decrements 51, 01, or both after memory is accessed. This means
that after the instruction the pointer registers don't point to the
memory location just accessed; instead, they point to the next
memory location to be accessed. This is actually very convenient,
since it allows you to build efficient loops such as those in the
examples in the last section. It can, however, occasionally cause
confusion, especially with the data scanning string instructions,
which we'll discuss next.
Now we'll look at the data scanning string instructions, SCAS and
CMPS, which are used for scanning and comparing blocks of
memory.
SCAS is used to scan memory for a match or non-match of a
particular byte or word value. As with all string instructions,
SCAS comes in two forms, SCAse and
SCA~W.
SCASe compares AL to the byte value at address ES:OI, setting
the flags to reflect the comparison, just as if a CMP instruction had
been executed. As with STOSB, 01 is incremented or decremented
Chapter 6, More about programming In Turbo Assembler
185
by seAse. For example, the following finds the first lowercase t
in the string TextString:
•DATA
TextString
DB 'Test text',O
TEXT_STR1NG_LENGTH EQU ($-TextString)
.CODE
mov ax,@data
mov es,ax
mov di,OFFSET TextString
mov al,'t'
mov cx,TEXT_STR1NG_LENGTH
cld
ScanJor_t_Loop:
scasb
je Found t
loop Scan_For_t_Loop
iNo "t" found
i"t" found
Found t:
dec di
iES:D1 points to the start of
i TextString
icharacter to scan for
ilength of string to scan
iscan with D1 incrementing
idoes ES:DI match AL?
iyes, we found "t"
ino, scan next character
ipoint back to offset of "t"
Note that D1 is decremented after t is found in this example,
which reflects the string pointer overrun we discussed earlier.
When this code performs the final, successful SeASe, D1 is
incremented after the comparison, since the last thing a string
instruction does is increment or decrement its pointer{s). As a
result, D1 points to the byte after the t that was found and must be
adjusted to compensate for the overrun and point to the t.
You might get a better feel for what seAse does by comparing its
use in the last example to similar code without string instructions:
ScanJor_t_Loop:
cmp es: [di], al
je Found_t
inc dl
loop Scan_For_t_Loop
idoes ES:D1 match AL
iyes, we found "t"
ino, scan next character
The last example isn't exactly the same as the seAse example
preceding it, however, since seAse increments D1 immediately
186
Turbo Assembler User's Guide
i~ements
and the last example
it after the JE instruction in order
to a void altering the flags set by CM P.
This brings up an important point about string instructions in
general. String instructions never set the flags to reflect the
chang~s they make to 51, 01, and/or ex. LOOS, STOS, and MOVS
don't change any flags, and SCAS and CMPS only change flags
according to the results of the comparisons they make.
It certainly would be handy to be able to reduce the loop in the
previous example to a single instruction, and, as you've probably
guessed, REP lets you do just that. However, you might want to
stop the loop on either a match or a non-match. Here are two
forms of REP to use with SCAS (and CMPS as well)-REPE and
REPNE.
REPE (also known as REPZ) tells the 8086 to repeat SCAS (or
CMPS) until either ex becomes zero or a non-match occurs. You
might think of REPE as being the "repeat while equal" prefix.
Likewise, REPNE (also known as REPNZ) tells the 8086 to repeat
SCAS (or CMPS) until either ex becomes zero or a match occurs.
Think of REPNE as being the "repeat while not equal" prefix.
Here's code that uses a single repeated SCAse instruction to scan
TextString for the character t:
mov ax,@data
mov es,ax
mov di,OFFSET TextString
mov al,'t'
mov cx,TEXT_STR1NG_LENGTH
cld
repne scasb
je
Found t
iES:D1 points to the start of
i TextString
icharacter to scan for
ilength of string to scan
iscan with D1 incrementing
iscan the whole string to see
i if there's at least one "t"
i yes, we found lit II
i No lit" found
i"t" found
Found t:
dec di
ipoint back to offset of "til
Like all string instructions, SCAS increments its pointer register,
01, if the direction flag is 0 (cleared with CLO), and decrements OJ
if the direction flag is 1 (set with STO).
Chapter 6, More about programming In Turbo Assembler
187
SCASW is a word-sized form of SCASB, comparing AX to E5:0I,
and incrementing or decrementing 01 by two rather than one at
the end of each execution. The following code uses REPE SCASW
to find the last nonzero entry in an array of word-sized integers:
mov ax,SEG ShortIntArray
mov es,ax
mov di,OFFSET ShortIntArray+((ARRAY_LEN_IN_WORDS-l) *2)
;ES:DI points to the end of
; ShortIntArray
mov cx,ARRAY_LEN_IN_WORDS
;search for non-match with zero
sub ax,ax
std
;search backward from end,
; decrementing DI
;search until we come to a
repe scasw
; nonzero word or run out of
; array
jne FoundNonZero
;The whole array is filled with zeros.
;We found a nonzero element--adjust DI for overrun to point to it.
FoundNonZero:
inc di
inc di
CMPS
The CMPS string instruction is designed to let you compare two
strings of bytes or words. A single repetition of CMPS compares
two memory locations, then increments both 51 and 01. You
might think of CMPS as being like a MOVS that compares two
memory locations instead of copying one memory location to
another.
CMPSB compares the byte at DS:5I to the byte at E5:0I, sets the
flags accordingly, and increments or decrements 51 and 01,
depending on the direction flag. AX is not modified in any way.
Like the other string instructions, CMPS comes in both byte and
word sizes, can either increment or decrement 51 and 01, and will
repeat if preceded by a REP prefix. Here's the code to check
whether the first 50 elements in two word-sized arrays are
identical, using REP CMPSW:
mov si,OFFSET Arrayl
mov ax,SEG Arrayl
mov ds,ax
188
Turbo Assembler User's Guide
mov
mov
mov
mov
cld
repe
jne
iFirst 50
di,OFFSET Array2
ax,SEG Array2
es,ax
icompare the first 50 elements, at most
cx,50
cmpsw
ArraysAreDifferent
elements are identical.
iAt least one element differs between the two arrays.
ArraysAreDifferent:
dec si
dec si
ipoint back to the element that differed
;both arrays
dec di
dec di
Using operands
with string
instructions
We've only looked at the explicit byte and word forms of the
string instructions so far; in other words, we've looked at LOOSB
and LODSW, but haven't used LOOS. It's acceptable to use the
nonexplicit versions of the string instructions, as long as you
provide operands so that Turbo Assembler knows whether you
want byte- or word-sized operations.
For example, the following is acceptable and is equivalent to
MOVSB:
•DATA
Stringl LABEL
BYTE
db ' abcdefghi'
STRINGl_LENGTH EQU ($-Stringl)
String2 DB 50 DUP (?)
.CODE
mov ax,@data
mov ds,ax
mov es,ax
mov si,OFFSET Stringl
mov di,OFFSET String2
mov cx,STRINGl_LENGTH
cld
rep movs es:[String2], [Stringl]
Chapter 6, More about programming In Turbo Assembler
189
Since you specified Stringl and String2 as operands to MOVS,
Turbo Assembler makes the data size of the MOVS instruction the
data size of the operands, which is byte in this case.
There's a catch to using operands with string instructions,
however. String instruction operands aren't real operands, in the
sense that they're built into the instruction; a string instruction
just uses whatever SI and/or DI happen to be when that
instruction is executed. The operands are only used to set data
size, not to actually load pointers. Look at it this way: When you
use an instruction like
mov aI, [Stringl]
the offset of Stringl is built right into the machine-language
instruction for MOV. However, when you use
lods [Stringl]
the machine-language instruction assembled is just the single byte
for LOOSB; Stringl is not built into the instruction. It's your
responsibility to make sure that DS:SI points to the start of Stringl
in this case.
Operands to string instructions are sort of like using the ASSUME
directive for segments. ASSUME doesn't actually set a segment
register; it just tells Turbo Assembler how you have set a segment
register so Turbo Assembler can do error-checking for you.
Similarly, operands to string instructions don't set any registers;
they just tell Turbo Assembler what you've set SI and/or DI to so
Turbo Assembler can determine operand size and do errorchecking. Refer to the section "Relying on the operand(s) to a
string instruction" on page 242 for further discussion of operands
to string instructions.
In the section "Pitfalls with string instructions" on page 235, we
discuss several points to look out for when using the string
instructions.
Multimodule programs
Sooner or later, you're going to outgrow keeping each program's
source code in a single file. Single-file source code is fine for short
programs, such as the examples in this manual, but even
medium-sized programs must be broken into several files, or
190
Turbo Assembler User's Guide
modules, that are assembled separately and linked together. The
primary advantage of multimodule programs is that after you
edit the source code, you only need to reassemble the modules
you've changed, rather than every line of the program. Also, it's
much easier to find your way around several short files than one
massive file.
It's surprisingly easy to create multimodule programs. Turbo
Assembler provides three directives to support such programs:
PUBLIC, EXTRN, and GLOBAL. We'll look at each in tum, but
before we do, we'll look at a sample program consisting of two
modules, so that you'll understand the context in which we're
discussing the multimodule directives. Here's the main program,
MAIN.ASM:
•MODEL
small
•STACK
200h
•DATA
Stringl
DB 'Hello,',O
String2
DB 'world' ,0dh,Oah,'S',0
GLOBAL
FinalString:BYTE
FinalString
DB- 50 DUP (?)
.CODE
ConcatenateStrings:PROC
EXTRN
ProgramStart:
mov ax,@data
mov ds,ax
mov ax, OFFSET Stringl
mov bx,OFFSET String2
call ConcatenateStrings
icombine the two strings
i into a single string
mov ah,9
mov dx,OFFSET FinalString
iprint the resulting string
int 21h
mov ah,4ch
int 21h
iand done
END ProgramStart
And here's the other module of the program, SUB1.ASM:
. MODEL
•DATA
GLOBAL
.CODE
small
FinalString:BYTE
Subroutine copies first one string, and then another
to FinalString.
Chapter 6, More about programming in Turbo Assembler
191
Input:
DS:AX = pointer to first string to copy
DS:BX = pointer to second string to copy
Output: None
Registers destroyed: AL, SI, DI, ES
PUBLIC
ConcatenateStrings
ConcatenateStrings PROC
cld
istrings count up
mov di,SEG FinalString
mov es,di
mov di,OFFSET FinalString iES:DI points to destination
mov si,ax
ifirst string to copy StringlLoop:
lodsb
iget string 1 character
and aI, al
i is it O?
jz DoString2
iyes, done with string 1
stosb
isave string 1 character
jmp StringlLoop
DoString2:
mov si,bx
isecond string to copy String2Loop:
lodsb
iget string 2 character
isave string 2 character
stosb
i (including 0 when we find it)
and al,al
iis it O?
jnz String2Loop
ino, do next character
ret
idone
ConcatenateString ENDP
END
These two modules would be assembled separately with
TASM main
and
TASM subl
and would then be linked into the program MAIN.EXE with
tlink main+subl
When run with the command
main
MAIN.EXE displays the output (you guessed it)
Hello, world
192
Turbo Assembler User's Guide
Now that you've seen a multimodule program in action, let's
examine the three directives that make multimodule
programming possible.
The PUBLIC
directive What the PUBLIC directive does is simple enough: It instructs
Turbo Assembler to make the associated label or labels available
to other modules. Labels of almost any sort, including procedure
names, memory variable names, and equated labels, may be made
available to other modules by way of PUBLIC. For example,
•DATA
PUBLIC
ARRAY LENGTH
MemVar
Array1
MemVar, Array1, ARRAY_LENGTH
EQU 100
DW 10
DB ARRAY LENGTH DUP (?l
.CODE
PUBLIC
NearProc, FarProc
NearProc PROC NEAR
NearProc ENDP
FarProc
LABEL
PROC
END
Here the names of an equated label, a word variable, an array, a
near procedure, and a far procedure are made available to any
other module that is linked to this module.
There is one sort of label that cannot be made public, and that's an
equated label that is not equal to a 1- or 2-byte constant value. For
example, the following labels couldn't be made public:
LONG VALUE
TEXT SYMBOL
EQU 10000h
EQU
Turbo Assembler normally ignores case when assembling, so all
public labels are normally converted to uppercase. If you want
case-sensitivity for public labels, you must use either the Iml or
Imx command-line switch to Turbo Assembler in all modules that
contain or reference public labels.
For example, without Iml or Imx, other modules won't be able to
distinguish between the following two labels:
Chapter 6, More about programming In Turbo Assembler
193
PUBLIC
Symbol 1, SYMBOL1
When you use the Imx command-line switch to allow case
sensitivity for public and external symbols, you must be careful to
use the proper case for the symbol name in the PUBLIC or EXTRN
directive. Turbo Assembler makes the symbol available to other
modules with the name that appears in the EXTRN or PUBUC
directive, not how it appears where defined or referred to
elsewhere in the module. For example,
PUBLIC Abc
abC Dw
causes the name Abc to become public, not abC.
You don't need to have a
.MODEL directive In effect to
use this feature.
You can also specify a language for each symbol in a PUBLIC
directive. Valid languages are C, PASCAL, BASIC, FORTRAN,
PROLOG, and NOLANGUAGE. This causes any language-specific
rules to be applied to a symbol name automatically before it is
published in the object file. For instance, if you declare
PUBLIC C myproc
then the symbol myproc in the source file will actually be
published to the outside world as _myproc, since the convention
for the C language is to precede symbol names with an
underscore character. Using a language specifier in a PUBLIC
directive temporarily overrides the current language setting
(default or one established with the .MODEL directive).
The EXTRN
directive
In the last section, we used PUBUC to make the labels MemVar,
Arrayl, ArrayLength, NearProc, and FarProc available to other
modules. The next question is, "How do other modules reference
those labels?"
The EXTRN directive is used to make public labels from other
modules available in a given module. Once EXTRN has been used
to make a public label from another module available, that label
can be used just as if it were defined in the current module. Here's
how another module would use EXTRN to reference the public
labels we defined in the last section:
...
•DATA
EXTRN
194
MemVar:WORD,Arrayl:BYTE,ARRAY_LENGTH:ABS
Turbo Assembler User's Guide
•CODE
EXTRN
NearProc:NEAR,FarProc:FAR
mov ax, [MemVar]
mov bx,OFFSET Arrayl
mov cx,ARRAY_LENGTH
call NearProc
call FarProc
Note that all five labels are used as you'd normally use labels;
only the EXTRN directives differ from single-module assembler
source code.
Each label declared with EXTRN is followed by a colon and a type.
The type is necessary because Turbo Assembler has no way of
knowing what sort of label you've declared with EXTRN unless
you tell it. With one exception, the types that can be used with
external labels are the same as those that can be used with the
LABEL directive. Available types are
ABS
BYTE
DATAPTR
DWORD
FAR
FWORD
NEAR
PRoe
QWORD
Structure Name
TBYTE
UNKNOWN
WORD
An absolute value
A byte-sized data variable
A near or far data pointer, depending on the
current memory model
A doubleword-sized (4 byte) data variable
A far code label (branched to by loading CS:IP)
A 6-byte data variable
A near code label (branched to by loading IP
only)
A procedure code label, near or far according
to.MODEL
A quadword-sized (8 byte) data variable
Name of a user-defined STRUe type
A lO-byte data variable
An unknown type
A word-sized (2 byte) data variable
The only unfamiliar external data type is ABS, which is used to
declare a label that's defined in its original module with EQU or =;
in other words, a label that is simply a name for a constant value
and is not associated with a code or data address.
It's important that you specify the correct data type for external
labels, since Turbo Assembler has to generate code on the basis of
the data types you specify, and has no way of knowing if you've
Chapter 6, More about programming In Turbo Assembler
195
made an incorrect specification. For instance, if you accidentally
typed
.CODE
EXTRN
FarProc:NEAR
call FarProc
given
PUBLIC FarProc
FarProc PROC FAR
ret
FarProc ENDP
in another module, Turbo Assembler would generate a near call
to FarProc, in accordance with the data type you specified with
EXTRN. This code surely wouldn't work properly, since FarProc is
actually a far procedure and ends with a far RET instruction.
As described in the last section, Turbo Assembler is normally
case-insensitive, and public la];>els are normally converted to
uppercase. This means that external labels are normally expected
to be uppercase. Use the Iml or Imx command-line switch if you
want case-sensitive external labels.
You don't need to have a
.MODEL directive In effect to
use this feature.
You can also specify a language for each symbol in an EXTRN
directive. Valid languages are C, PASCAL, BASIC, FORTRAN,
PROLOG, and NOLANGUAGE. This ,causes any language-specific
rules to be applied to a symbol name automatically before it is
published in the object file. For instance, if you declare
EXTRN C myproc:NEAR
then the symbol myproc in the source file will actually refer to the
.external symbol_myproc. Using a language specifier in an EXTRN
directive temporarily overrides the current language setting
(default or one established with the .MODEL directive).
196
Turbo Assembler User's Guide
The GLOBAL
directive At this point, you may well wonder why it takes two directives,
PUBLIC and EXTRN, to do a single job-sharing labels between
modules. Actually, the only reason two directives are required is
for compatibility with other assemblers; Turbo Assembler gives
you the GLOBAL directive, which does everything both PUBLIC
and EXTRN do.
If you declare a label global and then define it (with DB, DW,
PROC, LABEL, or the like), then that label is made available to
other modules, just as if you'd used PUBLIC instead of GLOBAL.
If, on the other hand, you declare a label global and then use it
without defining it, then that label is treated as an extemallabel,
just as if you'd used EXTRN.
For example, consider the following:
•DATA
GLOBAL
FinalCount
FinalCount:WORD,PrornptString:BYTE
DW?
.CODE
GLOBAL
DoReport:NEAR,TallyUp:FAR
TallyUp PROC FAR
call DoReport
Here FinalCount and TallyUp are defined, so they're made public
labels, available to other modules. PromptString and DoReport
aren't defined in this module, so they're made external labels and
are assumed to have been made public in some other module.
One particularly handy place to use GLOBAL is in an Include file.
(We'll discuss include files in the next section.) Suppose you have
a set of labels that you want to make available to all the modules
in a multimodule program. It would be nice to be able to declare
all those labels in an include file, and then include that file in each
module. Unfortunately, that's impossible using PUBLIC and
EXTRN because EXTRN won't work in the module a given label is
defined in, and PUBLIC will only work in the module a given label
is defined in. However, GLOBAL will work in all modules, so you
can make up an include file that declares all the labels of interest
to be global, and include that file in all your modules.
Chapter 6, More about programming In Turbo Assembler
197
You don't need to have a
.MODEL directive In effect to
use this feature.
As with the PUBUC and EXTRN directives, you can specify a
language for each symbol in a GLOBAL directive. Valid languages
are C, PASCAL, BASIC, FORTRAN, PROLOG, and NOLANGUAGE.
This causes any language-specific rules to be applied to a symbol
name automatically before it is published in the object file. For
instance, if you declare
GLOBAL C myproc
then the symbol myproc in the source file will actually be
published to the outside world as _myproc. Using a language
specifier in a GLOBAL directive temporarily overrides the current
language setting (default or one established with the .MODEL
directive).
Include files
Include files are rarely used
for code, since you can
readily link separate code
modules together, but It Is
perfectly acceptable to put
code Into an Include file,
should you so desire.
You'll often find that you'd like to insert the same block of
assembler source code in several source modules. You may want
to share equates or macros among different parts of a program, or
you may simply want to reuse equates or macros in several
programs. Then, too, you may have a long program that you
don't want to break into several linkable modules (a program that
will be stored in ROM, for example), but which is too big to
conveniently keep in a single file. The INCLUDE directive meets
all these needs.
When Turbo Assembler encounters an INCLUDE directive, it
marks its place in the current assembler module, goes to disk and
finds the specified include file, and starts assembling the include
file, just as if the lines in the include file were right in the current
module. When the end of the include file is reached, Turbo
Assembler returns to the line after the INCLUDE directive in the
current module, and resumes assembly there. The key point is
this: The text of the include file is literally inserted into the
assembly of the current assembler module at the location of the
INCLUDE directive.
For instance, if MAINPROG.ASM contains
•CODE
mov ax,1
INCLUDE INCPROG.ASM
198
Turbo Assembler User's Guide
push ax
and INCPROG.ASM contains
mov bx,5
add ax,bx
then the result of assembling MAINPROC.ASM is exactly
equivalent to
•CODE
mov
mov
add
push
ax,l
bx,5
ax,bx
ax
Include files can be nested
arbitrarily deep.
Include files can be nested (can include another file). You can
easily tell included lines in a listing file because Turbo Assembler
places a number at the left end of included lines, which indicates
how deeply the module files are nested.
For compatibility with MASM,
you can use backward
slashes (\) In INCLUDE path
specifications.
How does Turbo Assembler know where to find Include files?
Well, if you specify a drive or path as part of the file name
operand to INCLUDE, Turbo Assembler looks exactly where you
specify, and nowhere else. If you specify only a file name, with no
drive or path, Turbo Assembler first searches the current
directory for the specified file. If Turbo Assembler can't find the
file in the current directory, it searches the directories specified
. with the -I command-line switch, if any. For example, given the
Turbo Assembler command line
TASM -ic:\include testprog
and given the line
INCLUDE
MYMACROS.ASM
in TESTPROG.ASM, Turbo Assembler will first search the current
directory for MYMACROS.ASM, and, failing that, will search the
directory C: \INCLUDE. If MYMACROS.ASM isn't in either of
those places, Turbo Assembler will report an error.
Chapter 6, More about programming in Turbo Assembler
199
The listing file
The object and/or listing file
names don't have to match
the source file name, but
there srarely a reason for
your source file to have one
name and your object or
listing files to have another.
Normally, Turbo Assembler produces only one file as the result of
assembly: an object (.OBJ) file with the same name as the source
(.ASM) file. You can, if you wish, ask Turbo Assembler to produce
a listing file with the extension .LST as well, simply by typing two
additional commas (or two additional file names) on the
command line. For example, where
TASM hello
assembles HELLO.ASM and produces the object file HELLO.OBI,
the command line
TASM hello"
generates the listing file HELLO.LST, as do both
TASM hello,hello,hello
and
TASM /1 hello
The listing file is basically the source file annotated with a variety
of information about the results of the assembly. Turbo Assembler
lists the actual machine code for each instruction, along with the
offset in the current segment of the machine code for each line.
What's more, Turbo Assembler provides tables of information
about the labels and segments used in the program, including the
value and type of each label, and the attributes of each segment.
Turbo Assembler can also, on demand, generate a cross-reference
table for all labels used in a source file, showing you where each
label was defined and where it was referenced. (See the Ie
command-line option in Chapter 3.)
We'll look at the basics of the listing file first-the assembled
machine code and offset for each instruction.
200
Turbo Assembler User's Guide
Annotated
source code
Here's the listing file for the original example program,
HELLO.ASM:
Turbo Assembler Version 2.0
01-18-90 14:31:58
Page 1
HELLO.ASM
DOSSEG
.MODEL small
0000
.STACK 100h
0000
0100
•DATA
0000 48 65 6C 6C 6F 2C 20 + HelloMessage DB 'Hello, world',13,10,12
77 6F 72 6C 64 OD OA +
DC
HELLO_MESSAGE_LENGTH EQU $ - HelloMessage
8
= OOOF
9 OOOF
.CODE
10 0000 B8 DODOs
mov ax,@data
mov ds,ax
;set DS to point to data seg
11 0003 8E D8
mov ah,40h
;DOS write to device function t
12 0005 B4 40
mov bx,l
;standard output handle
13 0007 BB 0001
mov cx,HELLO_MESSAGE_LENGTH ;number of characters to print
14 OOOA B9 OOOF
mov dx,OFFSET HelloMessage ;string to print
15 DODD BA OOOOr
int 21h
;print "Hello, world"
16 0010 CD 21
mov ah,4ch
;DOS terminate program function t
17 0012 B4 4C
int 21h
;terminate the program
18 0014 CD 21
19
END
1
2
3
4
5
Turbo Assembler Version 2.0
Symbol Table
01-18-90 14:31:58
Symbol Name
Type
Value
??DATE
??FILENAME
??TIME
??VERSION
@CODE
@CODESIZE
@CPU
@CURSEG
@DATA
@DATASIZE
@FILENAME
@WORDSIZE
HELLOMESSAGE
HELLO- MESSAGE- LENGTH
Text
Text
Text
Number
Text
Text
Text
Text
Text
Text
Text
Text
Byte
Number
"06-29-88"
"HELLO
"16:21:26"
004A
TEXT
0
0101h
TEXT
DGROUP
0
HELLO
2
DGROUP:OOOO
OOOF
Groups & Segments
Bit Size Align Combine Class
Chapter 6. More about programming In Turbo Assembler
Page 2
201
DGROUP
STACK
DATA
TEXT
Group
16 0100 Para Stack STACK
16 OOOF Word Public DATA
16 0016 Word Public CODE
The top of each page of the listing file displays a header consisting
of the version of Turbo Assembler that assembled the file, the date
and time of assembly, and the page number within the listing.
There are two parts to the listing file: the annotated source code
listing and the symbol tables. The original assembler code is
displayed first, with a header containing the name of the file
where the source code resides. The assembler source code is
annotated with information about the machine code Turbo
Assembler assembled from it. Any errors or warnings
encountered during assembly are inserted immediately following
the line they occurred on.
The code lines in the listing file follow this format:
• indicates the level of nesting of Include files and
macros within your listing file.
• is the number of the line in the listing file (not
including header and title lines). Line numbers are particularly
useful when the cross-reference feature of Turbo Assembler,
which refers to lines by line number, is used. In HELLO.LST,
the DOSSEG directive is line 1 of the listing file, the .MODEL
directive is line 2, and so on.
Be aware that the line numbers in the field are
not the source module line numbers. For example, if a macro is
expanded or a file is included, the line-number field will
continue to advance, even though the current line in the source
module stays the same. In order to translate a line number (for
example, one produced by the cross-referencer) back to the
source file, you must look up the line number in the listing file,
then find that same line (by eye, not by number) in the source
file.
B is the offset in the current segment of the start of the
machine code generated by the associated assembler source
line. For instance, HelloMessage starts at offset a in the data
segment.
• is the actual sequence of hexadecimal byte and
word values that is assembled from the associated assembler
202
Turbo Assembler User's Guide
source line. For example, MOV AX,@data starts at offset 0 in the
code segment. The information just to the right of the offset
field for a given instruction is the machine code assembled from
that instruction, so the machine code assembled for MOV
AX,@data is B8 OOOOs (all in hexadecimal). OBBh is the machine
language instruction to load AX with a constant value, while
OOOOs is the constant value of @data, which is loaded into AX.
(Actually, OOOOs is just a placeholder for the value of @data;
we'll get to that in a minute.) Altogether, the instruction MOV
AX,@data assembles to 3 bytes of machine code.
Note that the listing file indicates that the instruction following
MOV AX,@data, which is MOV DS,AX, starts at offset 3 in the
code segment. This makes perfect sense, given that MOV
AX,@data starts at offset 0 and is 3 bytes long. The machine
code assembled from MOV DS,AX-8e DB-is 2 bytes long, so
the next instruction should start at offset 5; looking at the listing
file, we see that that is the case.
• Finally, is simply the original assembler line,
comments and all. Some assembler lines, such as those that
contain only comments, don't generate any machine code; these
lines have no or fields, but do have a
line number.
Recall that we said that the OOOOs value for @data was only a
placeholder for the real value in the instruction
rnov ax,@data
This is because segment values are assigned by the linker, not by
Turbo Assembler, so Turbo Assembler can't fill in the correct
value. What Turbo Assembler can do, however, is let you know
that a given value is a segment value that will be resolved by the
linker. That's done by appending the letter s to the end of the
machine code generated for
rnov ax,@data
Likewise, the offset in the machine code assembled from
rnov dx,OFFSET HelloMessage
ends with r, indicating that the offset might have to be relocated
when its segment is combined with other segments by the linker.
Here's the full list of notations used by Turbo Assembler to
indicate assembly characteristics (such as relocatability):
Chapter 6, More about programming in Turbo Assembler
203
Notation
r
s
sr
e
se
so
+
Meaning
Indicates offset fixup type for symbols within the
.module
Indicates segment fixup type for symbols within the
module
Indicates segment and offset fixup type within the
module
Indicates offset fixup on an external symbol
Indicates pointer fixup on an external symbol
Indicates segment-only fixup
Indicates object code that has been truncated or
wrapped to the next line
In the object code listing, r, s, and sr are used to indicate offset,
segment, and pointer (segment and offset) fixup types for symbols
within" the module. e indicates an offset fixup on an external
symbol, and se indicates a pointer fixup on an external symbol.
Segment fixups on external symbols appear as s, just like for local
symbols. The object code field can also contain a + symbol in the
last column, indicating that there is more object code to display,
but it has been truncated.
The leftmost field of the listing is the level counter, which is blank
when assembling from the main file. Include files cause this field
to contain a 1 that becomes a 2,3, and so on, for each nested
include level. Likewise, macro expansions put a level counter in
this field.
You may have noticed that the listing file shows some of the
machine code entries as byte values (two hexadecimal digits) and
others as word values. There's a logical pattern here: Whenever
Turbo Assembler assembles machine code that represents a word
value, such as OFFSET HelloMessage, which is a 16-bit offset, that
value is shown as a word value. This is useful because, otherwise,
the low-byte-first approach the 8086 uses for storing words would
cause words to appear with the bytes reversed.
For example, the instruction
mov ax,1234h
assembles to 3 bytes of machine code: OB8h, 034h, and 012h, in
tha t order. If Turbo Assembler listed this machine code as 3 bytes,
it would appear as
B8 34 12
204
Turbo Assembler User's Guide
with the bytes of the word value swapped. Instead, Turbo
Assembler lists this machine code as
B8 1234
which is certainly easier to read.
When we discussed the field, we talked about the offset in
the current segment of the labels and lines in a program. How do
you know what segment a given label or line is in? That's the job
of the listing tables, which we'll cover next.
Listing symbol
tables
The second part of the listing file begins with the header "Symbol
Table" and consists of two tables: one describing the labels used
in the source code and the other describing the segments used.
By the way, if you have no use for the symbol table portion of the
listing file, you can instruct Turbo Assembler to generate only the
annotated source code portion of the listing with the In
command-line switch.
The table of labels
The first table, which we'll call the table of labels, lists all the
labels in the source code in alphabetical order, "along with their
types and the values to which they were set. For example, the
listing file HELLO.LST contains the following entry:
HELLOMESSAGE
HelloMessage In the last
section was marked with an
r, meaning HelloMessage
may be relocated to another
offset by the linker as the
other segments In DGROUP
are linked into the program.
The map file produced by
the linker Is the place to look
for information about
segment relocation.
BYTE DGROUP:OOOO
HELLOMESSAGE is the name of the label, or symbol; it's in
uppercase because Turbo Assembler converts all symbols to
uppercase unless you use the Imx or Iml command-line switch.
BYTE represents the data size of the data element referred to by
the name HelloMessage. DGROUP:OOOO is the value of the label
HelloMessage, meaning that the string pointed to by the label
HelloMessage starts at offset 0 in the segment group DGROUP.
Similarly, ProgramS tart is listed as a label of type near, with the
value _TEXT:OOOO; _TEXT is the name of the segment defined
with .CODE, so ProgramStart is at the first address in the code
segment. As you can see, we've answered an earlier question
about how to find out what segment a given label is in, since the
value field of the table of labels reports the segmentin which the
label resides.
Chapter 6, More about programming In Turbo Assembler
205
The other labels listed in the HELLO.LST listing file are the labels
that are predefined by Turbo Assembler when the simplified
segment directives are used. These labels are all set to text strings,
and contain values such as _TEXT and DGROUP.
Labels can be any of the following data types:
ABS
ALIAS
BYTE
DWORD
FAR
NEAR
NUMBER
aWORD
STRUCT
TBYTE
TEXT
WORD
As we discussed at the beginning of this chapter, equated labels
can be set to any constant value or to a text string; the value field
of the table of labels reports the values of such labels exactly as
you set them. For a label associated with memory addresses, such
as HelioMessage, it's the address of the label that is reported in the
value field.
The table of labels is the place to look for type and value
information about any label used anywhere in your source code.
The table of groups
and segments
The other table in the symbol table portion of the listing is the
table of groups and segments. Segment groups such as DGROUP
are simply reported as groups here, since segment groups have no
attributes of their own, but rather consist of one or more
segments. The segments making up a group in a given module
appear directly under that group's name in the table of groups
and segments, indented two columns to show they belong to the
group. In HELLO.LST, the segments STACK and _DATA are
members of the DGROUP segment group.
Segments do have attributes, and the table of groups and
segments lists five attributes for each segment. Reading from the
left, the table of groups and segments reports the data size, overall
size, alignment, combine type, and class for each segment. We'll
discuss each of these separately.
Refer to Chapter 10 for
Information on USE32
segments.
The data size is always 16 except for USE32 segments in code
assembled for the 80386 processor.
The segment size is given as four hexadecimal digits. For
example, the STACK segment is 0200h (512 decimal) bytes long.
The alignment type describes what sort of memory boundaries a
segment can start on. These are the possible alignment types:
206
Turbo Assembler User's Guide
• BYTE: Segment can start at any address
• DWORD: Segment can start at any address that is a multiple of 4
• PAGE: Segment can start at any address that is a multiple of 256
• PARA: Segment can start at any address that is a multiple of 16
• WORD: Segment can start at any even address
Chapter 9 provides more
information about alignment
and combine types and
segment classes.
In HELLO.LST, the STACK segment starts on a paragraph
boundary, while the _DATA and _TEXT segments are wordaligned.
The combine type dictates how segments of the same name are
combined with a given segment. For example, identically named
segments with combine-type PUBLIC are concatenated into a
larger segment, while those with combine-type COMMON are
overlaid into a single common segment.
Finally, the segment class specifies the overall class in which a
segment belongs, such as CODE, DATA, and STACK. The linker
uses this information to order segments when it links the
segments into a program.
The crossreference table
Symbol tables don't crossreference your labels,
groups, and segments.
The symbol table portion of the listing file normally tells you a
great deal about labels, groups, and segments, but there are two
things it doesn't tell you: where labels, groups, and segments are
defined and where they're used. Cross-referenced symbol
information makes it easier to find labels and follow program
execution when debugging a program.
There are two ways to instruct Turbo Assembler to produce
cross-reference information in the listing file. The Ie commandline switch is one way to ask Turbo Assembler to place crossreference information in the listing file; for example,
TASM
Ie hello"
generates cross-reference information in the listing file
HELLO.LST. Note, however, that Ie by itself is not enough to
generate cross-reference information; you must also instruct
Turbo Assembler to generate a listing file in which the crossreference information can be placed.
You can also ask Turbo Assembler to generate a listing file
containing cross-reference information by adding a fourth field to
the command line, as in
Chapter 6, More about programming In Turbo Assembler
207
TASM hello,hello,hello,hello
or
TASM hello",
Suppose we assemble REVERSE.ASM, the second example
program you looked at in Chapter 4, with the Ie command-line
switch:
TASM Ie reverse"
Turbo Assembler creates the following listing file, REVERSE.LST:
REVERSE.ASM
1
2
3
4
5
6 0000
7 03EB
8
9
10 0000
11 0003
13 0005
14 0007
15 OOOA
16
17 0000
18
19 0010
20 0012
21 0014
22 0016
23
24 0018
25 0019
26 001C
27 001F
28 0021
29
30
31 0022
32 0024
33 0026
34 0027
35
36 0028
37 002A
38 002B
39 0020
40 0030
41 0033
208
= 03E8
03EB* (??)
03E8* (??)
B8
8E
B4
BB
B9
OOOOs
08
3F
0000
03E8
BA OOOOr
CD
23
74
8B
21
CO
1F
C8
51
BB OOOOr
BE 03EBr
03 F1
4E
8A 07
88 04
43
4E
E2
59
B4
BB
BA
CD
F8
40
0001
03E8r
21
DOSSEG
•MODEL small
•STACK 200h
•DATA
MAXIMUM STRING LENGTH EQU 1000
MAXIMUM STRING LENGTH DUP(?)
StringToReverse DB
ReverseString DB
MAXIMUM=STRING=LENGTH DUP(?)
•CODE
ProgramStart:
mov ax,@data
mov ds,ax iset OS to point to the data segment
mov ah,3fh iDOS read from handle function t
mov bx,O
istandard input handle
mov cx,MAXIMUM STRING LENGTH
iread up to maximum t of characters
mov dx,OFFSET StringToReverse
istore the string here
int 21h
iget the string
and aX,ax iwere any characters read?
jz
Done
ino, so we're done
mov cX,ax iPut string length in CX, where
i can use it as a count
push cx
isave the string length
mov bx,OFFSET StringToReverse
mov si,OFFSET ReverseString
add si,cx
dec si
ipoint to the end of the reverse
i string buffer
ReverseLoop:
mov aI, [bx] iget the next character
mov lsi] ,al istore the characters in reverse order
ipoint to next character
inc bx
ipoint to previous location in
dec si
i reverse buffer
loop ReverseLoop imove next character, if any
pop cx
iget back the string length
mov ah,40h iDOS write from handle function t
mov bx,l
istandard output handle
mov dx,OFFSET ReverseString iprint this string
iprint the reversed string
int 21h
Turbo Assembler User's Guide
42
43 0035 B4 4C
44 0037 CD 21
Done:
45
mov
int
ah,4ch ;DOS terminate program function I
21h
;terminate the program
END
ProgramStart
Symbol Table
Symbol Name
Type
Value
Cref defined at I
@code
@curseg
DONE
MAXIMUM- STRING- LENGTH
PROGRAMSTART
REVERSELOOP
REVERSESTRING
STRINGTOREVERSE
Text
Text
Near
Number
Near
Near
Byte
Byte
TEXT
TEXT
TEXT:0035
03E8
- TEXT:OOOO
TEXT:0022
DGROUP:03E8
DGROUP:OOOO
12 18
12 13 14
Groups & Segments
Bit Size Align Combine Class
Cref defined at I
DGROUP
STACK
DATA
TEXT
Group
16 0200 Para
16 07DO Word
16 0039 Word
12 2 10
Stack STACK
Public DATA
Public CODE
18
21 142
15 6 7 15
19 45
130 36
t7 26 40
16 17 25
13
12 14
12
2 18
8
Once again, the listing file contains annotated source code and the
symbol tables. There's something new in the symbol tables,
however, and that's the cross-reference field.
For each symbol (label, group, or segment), the cross-reference
field lists the line numbers of all the lines in the program on
which that symbol was referenced. Lines on which a symbol was
defined are prefixed with a #.
The value of
MAXIMUM_SmING_LENGTH Is
a number, 03EBh or decimal
1000.
For example, let's find out where the MAXIMUM_STRING_
LENGTH label is defined and used. The listing file informs you
that it was defined on line 5; if you look at the first part of the
listing file, you'll see that this is the case.
The cross-reference field for MAXIMUM_STRING_LENGTH also
tells you that the label is referenced (but not defined) on lines 6, 7,
and 15. A glance at the first part of the listing file shows that this
is correct.
The Ie switch allows you to enable cross-referencing for an entire
file. You certainly won't always want a cross-reference listing for
every symbol-such a listing could be huge for a long source
module. Turbo Assembler provides you with directives that let
you enable and disable cross-referencing in selected portions of
your listings.
Chapter 6, More about programming In Turbo Assembler
209
The O/oCREF directive enables cross-referencing for succeeding
lines. The %NOCREF directive disables cross-referencing for
succeeding lines. Either of these directives overrides the
command-line Ie switch. If cross-referencing is enabled anywhere
in a source module, then the symbol table section reports the lines
on which all labels, groups, and segments were defined. However,
only those lines on which the labels, groups, and segments were
referenced (and for which cross-referencing was enabled) are
listed as cross-reference entries.
For example, consider
%NOCREF
ProgramStart
PROC
jmp LoopTop
%CREF
LoopTop:
loop LoopTop
%NOCREF
mov ax,OFFSET ProgramStart
;line 1
; line 2
;line 3
;line 4
;line 5
Line 1 will be listed as the definition line (with a #) for
ProgramStart, even though it was in an area in which crossreferencing is turned off because the definition lines for all labels
are listed if cross-referencing is turned on anywhere in a module.
Similarly, line 3 will be listed as the definition line for LoopTop.
For compatibility with other
assemblers, .CREF and .XCREF
are provided, controlling
cross-referencing in the same
way as %CREF and %NOCREF,
respectively.
Controlling the
listing contents
and format
210
Line 4 will appear as a cross-reference line for LoopTop because it
occurs after %CREF and before %NOCREF. However, line 2 will
not appear as a cross-reference line for LoopTop, because it occurs
when cross-referencing is disabled. Likewise, line 5 will not
appear as a cross-reference for ProgramStart.
Turbo Assembler gives you a remarkable degree of control over
which lines of source code should be listed, and over the format
of the listing file as a whole. The listing control directives fall into
two categories: the line-listing selection directives, which select
the information to be included in the listing file, and the listing
format control directives, which determine the actual format of
the listing file.
Turbo Assembler User's Guide
The line-listing selection
directives
The line-listing selection directives enable or disable inclusion of
certain lines in the listing file. In general, these directives are
useful for suppressing from the listing file information that you
don't care about at the moment, in order to keep the listing file to
a manageable size.
%L1ST and %NOUST
%UST and %NOLIST are the most basic of the line-listing selection
directives, enabling and disabling inclusion of succeeding lines in
the listing file. For example, given
%NOLIST
mov ax,l
%LIST
mov bx,2
%NOLIST
add ax,bx
only the middle line, mov bx, 2, will be included in the listing file.
By default, %LIST is selected.
%CONDS and %NOCONDS
%CONDS and %NOCONDS allow you to enable and disable the
listing of false conditionals and conditional statements. The listing
of such conditionals is normally disabled. For example, given the
code
%CONDS .
IFE IS8086
shl ax,?
ELSE
mov el,?
shl ax,el
ENDIF
both of the conditional sections, along with the conditional
assembly directives, will be placed in the listing file, rather than
just the conditional section that's true at the time of assembly.
Chapter 6. More about programming In Turbo Assembler
211
%INCL and %NOINCL
O/OINCL and O/ONOINCL allow you to enable and disable the listing
of lines included from other files by way of the INCLUDE
directive. The listing of included text is normally enabled. For
example, given the code
%NOINCL
INCLUDE
%INCL
INCLUDE
HEADER.ASM
INIT.ASM
the lines included from HEADER.ASM won't be placed in the
listing file, while the lines included from INIT.ASM will appear in
the listing file. (However, both INCLUDE directives will appear in
the listing file.)
%MACS and %NOMACS
O/OMACS and O/ONOMACS allow you to enable and disable the
listing of the text of macro expansions. The listing of macro
expansions is normally disabled. For example, given the code
MAKE BYTE MACRO VALUE
DB VALUE
ENDM
%NOMACS
MAKE BYTE 1
%MACS
MAKE BYTE 2
the text generated by the first expansion of the MAKE_BYTE
macro, DB 1, won't appear in the listing file, while the text
generated by the second expansion of MAKE_BYTE, DB 2, will
appear in the listing file. (However, both MACRO directives
appear in the listing file.)
%CTLS and %NOCTLS
O/OCTLS and O/ONOCTLS allow you to enable and disable the listing
of listing control directives themselves. The listing of listing
control directives is normally disabled. For example, given the
code
212
Turbo Assembler User's Guide
%NOCTLS
%NOINCL
%CTLS
%NOMACS
the listing control directive O/ONOINCL won't appear in the listing
file, while the listing control directive O/ONOMACS will.
&UREF and %NOUREF
O/OUREF and O/ONOUREF allow you to enable and disable the listing
of unreferenced symbols-in other words, symbols that are
defined but never used-in the symbol tables. The listing of
unreferenced symbols is normally enabled. You must specify a
cross-reference listing in order for those two options to have any
effect.
%SYMS and %NOSYMS
O/OSYMS and O/ONOSYMS allow you to enable and disable the
inclusion of the symbol tables in the listing file. The inclusion of
the symbol tables in the listing file is normally enabled.
The listing format
control directives
The listing format control directives alter the format of the listing
file. You can use these directives to tailor the appearance of the
listing file to your tastes and needs.
The O/OTITLE directive selects a title to be printed at the top of each
page of the annotated source code portion of the listing file. Only
one title can be specified in a given program. The O/OSUBTTL
directive selects a subtitle to be printed below the title on each
page of the listing. Any number of subtitles can be specified in a
program. For example, if the source module SPACEWAR.ASM
contained the directives
%TITLE
%SUBTTL
'Space Wars Game Program'
'Gravitational Effects Subroutines'
each page of the annotated source code would start with the lines
Turbo Assembler Version 2.0 1-18-90 21:53:35 Page 1 SPACEWAR.ASM
Space Wars Game Program
Gravitational Effects Subroutines
Chapter 6. More about programming in Turbo Assembler
213
%NEWPAGE forces Turbo Assembler to start a new page in the
listing file.
%TRUNC instructs Turbo Assembler to truncate fields that exceed
their maximum width, while %NOTRUNC instructs Turbo
Assembler to wrap fields that exceed their maximum width to the
next line. Normally, fields that overflow are not truncated. Note
that %NOTRUNC is on by default.
%PAGESIZE specifies the height in rows and width in columns of
the listing pages Turbo Assembler generates. For example,
%PAGESIZE 66,132
instructs Turbo Assembler to generate pages 132 columns wide by
66 rows high. Note that O/oPAGESIZE does not send page size
commands to the printer; rather, you should set up the printer
before printing the listing file, then use %PAGESIZE to instruct
Turbo Assembler to generate pages that match the way you've set
up your printer.
Field-width directives
Five directives control the width of the five fields of the annotated
source code portion of the listing file. The full format of a line in
this section of the listing file is
Earlier we described four of the five fields; the fifth field is the
field, which indicates how many macro or include levels
deep the current line is nested. For example, if the current line is
produced by a macro that itself is called from within a macro,
then the depth field will read 2.
The %DEPTH directive specifies the width in characters of the
field. The %LlNUM directive specifies the width in
characters of the field. The %PCNT directive
specifies the width of the field. (If you think of this field as
the "program counter" field, %PCNT is easier to remember.) The
%BIN directive specifies the width of the field.
Finally, the %TEXT directive specifies the width of the
field.
214
Turbo Assembler User's Guide
%PUSHLCTL and %POPLCTL
You might, at times, want to briefly change the current listing
control state and then restore it. Perhaps, in order to list every
byte of a data table, you need to enable wrapping and adjust the
width of the fields, or perhaps you want to enable listing of all
types of lines for debugging purposes. After you modify the
listing control state, it can be a real nuisance to restore the listing
controls to their previous state, especially since some of the listing
controls may have been set in an Include file or in some fardistant part of the source module.
Turbo Assembler provides the %PUSHLCTL and %POPLCTL
directives to handle this situation. %PUSHLCTL pushes the
current listing control state onto an internal stack, and
%POPLCTL pops the current listing control state from that stack.
(Both directives have a maximum of 16 levels.) These two
directives only save and restore the listing controls that can be
enabled and disabled (like %TRUNC and %NOTRUNC), and not
those that take a numeric argument (like %BIN). For example, in
the following code, the listing control state is exactly the same
after %POPLCTL as it was before %PUSHLCTL:
%LIST
%TRUNC
%PUSHLCTL
%NOLIST
%NOTRUNC
%NEWPAGE
%POPLCTL
Other listing control
directives
Turbo Assembler provides several other listing control directives
for compatibility with other assemblers. These directives include
TITLE, SUBTTL, PAGE, .UST, .XLlST, .LFCOND, .SFCOND,
.TFCOND, .LALL, .SALL, and .XALL. (Refer to Chapter 2 of the
Reference Guide for details on these directives.)
Displaying a message during assembly
Turbo Assembler provides two directives that allow you to
display a string on the console during assembly: DISPLAY and
Chapter 6, More about programming in Turbo Assembler
215
%OUT. These directives can be used to report on the progress of
an assembly, either to let you know how far the assembly has
progressed or to let you know that a certain part of the code has
been reached.
The two directives are essentially the same except that DISPLAY
displays a quoted string onscreen and %OUT displays a
nonquoted string onscreen. For instance, the following code
DISPLAY
%OUT
'This message produced by DISPLAY'
This message produced by %OUT
displays the following lines onscreen:
This message produced by DISPLAY
This message produced by %OUT
Assembling source code conditionally
You'll find there are times when it would be very useful to be able
to have a single assembler source module assemble to any of
several different versions of a program. For example, you might
want two versions of a given program: one version that uses
standard 8086 instructions and one version that takes advantage
of the powerful instructions of the 80186 and 80286.
You could maintain two separate source modules, one for each
version, but then you'd have a hard time keeping both modules
up to date. The simplest sol?tion would be to build both versions
into a single source module, with a single equated label that
selects which version gets assembled at any given time.
Turbo Assembler's conditional assembly directives give you this
capability and more. Consider the following code:
IF IS8086
mov
push
ELSE
push
ENDIF
call
216
ax,3dah
ax
3dah
GetAdapterStatus
Turbo Assembler User's Guide
If the value of the labellS8086 is nonzero, then the parameter
value 3dah is pushed on the stack with the two-step process
required by the 8086. If, however, 1S8086 is zero, then the
parameter value is pushed directly, using a special form of PUSH
that's available on the 80186 and 80286, but not the 8086. The code
in this example uses conditional assembly to support two versions
of the same program, one for the 8086 and one for the 80186 and
80286.
Turbo Assembler supports a variety of conditional assembly
directives, and also gives you the ability to generate assembly
errors in a variety of ways. We'll look at the conditional assembly
directives first.
Conditional
assembly
directives
IF and IFE
The simplest and most useful conditional assembly directives are
IF and IFE, which are used in conjunction with ENOIF and,
optionally, ELSE. IFDEF and IFNDEF are also frequently used,
while IFB, IFNB, IRON, IFOIF, IF1, and IF2 are useful in certain
situations.
IF causes the following block of code (up to the matching ELSE or
ENDIF) to be assembled only if the value of the operand is
nonzero. The operand may be a constant value or an expression
that evaluates to a constant value. For example,
IF REPORT- ASSEMBLY- STATUS
DISPLAY 'Reached assembly checkpoint l'
ENDIF
displays
Reached assembly checkpoint 1
when the IF is reached only if REPORT_ASSEMBLY_STATUS is nonzero.
An IF conditional can be terminated with either ENDIF or ELSE. If
an IF conditional is terminated with ELSE, then the code
following ELSE is assembled only if the operand to the associated
IF was zero. The block of code following the ELSE must be
terminated with an ENOIF.
IF conditionals can also be nested. For instance, this code
Chapter 6, More about programming In Turbo Assembler
217
;See whether arrays are to be defined (otherwise, they're
; allocated dynamically)
IF DEFINE ARRAY
;Make sure the array isn't too long
IF (ARRAY_LENGTH GT MAX_ARRAY_LENGTH)
ARRAY- LENGTH
MAX- ARRAY LENGTH
ENDIF
;Set the array to an initial value if that's indicated
IF INITIALIZE ARRAY
Array
DB ARRAY LENGTH DUP (INITIAL_ARRAY_VALUE)
ELSE
Array
DB ARRAY LENGTH DUP (?)
ENDIF
ENDIF
nests an IF and an IF .... ELSE inside another IF.
IFE is exactly like IF except that the following code is assembled
only if the operand is zero. The code associated with the following
IFE directive always assembles:
IFE 0
ENDIF
Like IF, IFE can have an associated ELSE directive.
Understand that the conditional assembly directives operate at
assembly time only, not when the program is run. These are not
like If statements in C, executing different code depending on
some run-time condition; instead, they assemble different code
depending on some assembly-time condition.
IFDEF and IFNDEF
IF and IFE are your primary tools for building programs that can
assemble into more than one version. Two other directives that
are useful in this connection are IFDEF and IFNDEF.
The block of code between an IFDEF directive and its associated
ENDIF is assembled only if the label that's the operand to IFDEF
exists (that is, if the label has already been defined when the
IFDEF directive is executed). For example, given the code
DEFINED_LABEL EQU 0
IFDEF DEFINED LABEL
218
Turbo Assembler User's Guide
DB
0
ENDIF
the DB directive will assemble; if, however, you were to delete the
equate that sets DEFINED_LABEL (and assuming DEFINED_LABEL isn't
set anywhere else in the program), then the DB directive would not
be assembled. Note that the value of DEFINED LABEL doesn't matter
to IFDEF.
IFNDEF is the opposite of IFDEF, assembling its associated code
only if the label that's the operand is not defined.
You may well wonder what IFDEF and IFNDEF are used for. One
use is guarding against attempts to define the same label twice
with EQU in a complex program; if the label's already defined,
you can use IFDEF to avoid defining it again and causing an error.
Another use is selecting the version of a program to be assembled,
much like what was done with IF previously; instead of checking
to see whether, say, INITIAUZE_ARRAYS is zero or nonzero, you
could simply check to see whether it is defined at all.
One handy way to select program version is by way of the Id
command-line switch. Id defines the associated label, and
optionally assigns that label a value. So, for example, you could
use a command line like
TASM /dINITIALIZE ARRAYS=l test
to assemble the program TEST.ASM with the label
INITIALIZE ARRAYS set to 1.
While that's undeniably useful, there's a potential problem here.
What if you're relying on INITIALIZE_ARRAYS being set on the
command line, but forget to type the appropriate Id switch? Also,
suppose you want to initialize arrays as a special case, and don't
want to be bothered with typing /dINITIALIZE _ARRAYS at other
times?
IFNDEF comes to your rescue in this case. You can use IFNDEF to
test whether INITIALIZE_ARRAYS is already defined (from the
command line), and then initialize it only if it's not already set.
That way, the command-line definition takes precedence, but
there's a default state for the label if no command-line definition
was specified. Here's the code to define INITIALIZE_ARRAYS only if
it's not already defined:
Chapter 6, More about programming In Turbo Assembler
219
IFNDEF
INITIALIZE ARRAYS
INITIALIZE ARRAYS EQU 0
ENDIF
idefault to not initializing
When you use IFNDEF this way to define an undefined symbol,
you'll get a warning message indicating that you are using a
pass-dependent construction. You can ignore this message if all
you are doing is defining a symbol inside the IFNDEF conditional
block. The message happens because Turbo Assembler can't tell if
you are going to put instructions or directives inside the block. If
you do more in the block than just define a symbol, you will
probably want to enable multi-pass processing with the 1m switch.
If you are only defining a symbol, enabling multi-pass processing
will cause the warning to not be given.
Other conditional
assembly directives
The IFB, IFNB, IFIDN, and IFDIF directives are used for testing
parameters passed to macros. (Macros are discussed in Chapter 9,
Advanced programming in Turbo Assembler.") IFB causes its
associated code to be assembled If the macro parameter that is its
operand is blank, while IFNB does the same if its operand is not
blank. IFB and IFNB are sort of the equivalent of IFNDEF and
IFDEF for macro parameters.
/I
For example, consider the macro TEST, defined as
i
Macro to define a byte or a word.
Input:
VALUE = value of byte or word
DEFINE_WORD = 1 to define a word, 0 to define a byte
i
Note: If PARM2 is not specified, a byte is defined.
TEST MACRO
VALUE, DEFINE_WORD
IFB
DB VALUE
idefine a byte if PARM2 is blank
ELSE
IF DEFINE WORD
DW VALUE
idefine a word if PARM2 is nonzero
ELSE
DB VALUE
idefine a byte if PARM2 is zero
ENDIF
ENDIF
ENDM
220
Turbo Assembler User's Guide
If TEST is invoked with
TEST 19
then a byte with the value 19 is defined, while if TEST is invoked
with
TEST 19,1
then a word with the value 19 is defined.
IFIDN causes its associated code to be assembled if the two macro
parameters that are its operands are identical, while IFDIF does
the same if its pair of operands are different. For example, the
following macro, which converts a signed byte to a signed word
in AX, doesn't bother to copy the source operand to AL if the
source operand is AL:
; Macro to convert a signed byte in an a-bit register or
; named memory location to a signed word in AX.
Input:
SIGNED_BYTE - the name of the register or memory location
containing the signed byte to convert to a signed word
MAKE SIGNED- WORD
MACRO
SIGNED BYTE
,
;make sure the operand isn't AL
IFDIFI
mov al,SIGNED_BYTE
ENDIF
cbw
ENDM
IFIDN and IFDIF are sensitive to the case of their arguments. Their
companion directives IFIDNI and IFDIR treat as equivalent
uppercase and lowercase letters in their arguments.
Note that angle brackets are required around all operands to IFB,
IFNB, IFIDN, and IFDIF.
If you don't use the 1m command-line switch to enable multiple
passes, then IF1 is always considered true, and IF2 is always
considered false because there is never a second pass. A "Passdependent construction encountered" warning is displayed in
this circumstance if Turbo Assembler encounters either IF1 or IF2
in a module.
If you use the 1m command-line switch, two passes are done
automatically if your module contains either IF1 or IF2. In this
case, IF1 is true on the first pass, IF2 is true on the second pass,
Chapter 6, More about programming In Turbo Assembler
221
and a "Module is pass-dependent--compatibility pass was done"
warning is also displayed.
ELSEIF family of
directives
Each of the IF family of directives (IF, IFB, IFIDN, and so on) has a
related member in the ELSEIF family (for example, ELSEIF,
ELSEIFB, ELSEIFIDN). They act like a combination of the ELSE
directive with one of the IF directives. You can use them to make
your code more readable when you want to test against multiple
conditions or values and only assemble a single block of code.
Consider the following code fragment:
IF BUFLENGTH GT 1000
CALL DOBIGBUF
ELSE
IF BUFLENGTH GT 100
CALL MEDIUMBUF
ELSE
IF BUFLENGTH GT 10
CALL SMALLBUF
ELSE
CALL TINYBUFP
ENDIF
ENDIF
ENDIF
You can use the ELSEIF directive to improve the readability of
this code:
IF BUFLENGTH GT 1000
CALL DOBIGBUF
ELSEIF BUFLENGTH GT 100
CALL MEDIUMBUF
ELSEIF BUFLENGTH GT 10
CALL SMALLBUF
ELSE
CALL TINYBUF
ENDIF
This roughly corresponds to the case or switch statements in
Pascal and C. However, this capability is actually far more
general, since you don't have to use the same kind of ELSEIF test
throughout the conditional code block. For example, the
following is perfectly valid:
PUSHREG MACRO ARG
IFIDN ,
push si
push di
222
Turbo Assembler User's Guide
ELSEIFB
push ax
ENDIF
ENDM
Conditional error
directives
Turbo Assembler allows you to unconditionally or conditionally
generate assembly errors with the conditional error directives:
.ERR
.ERRB
.ERRDIFI
.ERRIDNI
.ERR1
.ERRDEF
.ERRE
.ERRNB
.ERR2
.ERRDIF
.ERRIDN
.ERRNDEF
.ERRNZ
Why on earth would you intentionally generate an assembly
error? Well, the conditional error directives allow you to catch a
variety of mistakes in your programs, such as equated labels that
are too large or too small, labels that aren't defined, and missing
macro parameters.
Take another look at the list of conditional error directives. You'll
note that the conditional error directives are very similar to the
conditional assembler directives, and that's no coincidence, since
most of the conditional error directives test the same conditions.
For example, .ERRNDEF generates an error if the operand label is
not defined, just as IFNDEF assembles the associated code if the
operand label is not defined .
.ERR, .ERR1, and .ERR2
Whenever Turbo Assembler encounters the .ERR directive, an
error is generated. By itself, that's not a particularly useful
function; however, .ERR is useful when combined with a
conditional assembly directive.
For example, suppose you want to generate an error if the equate
for the length of a given array is set to too large a number. The
following code would do the job:
IF (ARRAY_LENGTH GT MAX_ARRAY_LENGTHl
.ERR
ENDIF
If the array isn't too long, Turbo Assembler won't assemble the
code within the IF block, so the .ERR directive will never be
assembled, and no error will be generated .
. ERR1 and .ERR2 do just what .ERR does, but only on pass 1 or
pass 2, respectively. Ify,ou don't use the 1m command-line switch
Chapter 6, More about programming In Turbo Assembler
223
to enable multiple passes, then .ERR1 will always display an
error; .ERR2 will never display an error, because there is never a
second pass. A "Pass-dependent construction encountered"
warning is displayed in this circumstance if Turbo Assembler
encounters either .ERR1 or .ERR2 in a module.
If you use the 1m command-line switch, two passes are done
automatically if your module contains either .ERR1 or .ERR2. In
this case, .ERR1 displays an error on the first pass, .ERR2 displays
an error on the second pass, and a "Module is pass-dependentcompatibility pass was done" warning is also displayed.
.ERRE and .ERRNZ
The .ERRE directive generates an error if its operand, which must
evaluate to a constant expression, is equal to zero . .ERRE is
equivalent to performing .lFE combined with .ERR. For example,
.ERRE
TEST LABEL-l
is equivalent to
IFE TEST LABEL-l
.ERRE
ENDIF
.ERRE can be used to generate an error when a relational
expression returns false, since the value of a false expression is O.
Similarly, the .ERRNZ directive generates an error if its operand is
not equal to zero; this is equivalent to IF followed by .ERR .
. ERRNZ can be used to generate an error when a relational
expression returns true, since the value of a true expression is
nonzero. For example,
.ERRNZ
ARRAY- LENGTH GT MAXARRAY
-LENGTH
performs the same action as do the IF and .ERR directives in the
example in the last section .
.ERRDEF and .ERRNDEF
.ERRDEF generates an error if the label that is its operand is
defined, while .ERRNDEF generates an error if the label that is its
operand is undefined. These directives let you perform the
equivalent of IFDEF or IFNDEF and .ERR in a single line. For
example,
.ERRNDEF MAX PATH LENGTH
is equivalent to
IFNDEF
224
MAX PATH LENGTH
Turbo Assembler User's Guide
.ERR
ENDIF
Other conditional error
directives
The four remaining conditional error directives are intended for
use in macros only, and are directly analogous to the four
conditional assembly directives intended for use in macros that
we discussed in the previous section, "Other Conditional
Assembly Directives," on page 220.
. ERRB generates an error if the macro parameter that is its
operand is blank, and .ERRNB generates an error if the macro
parameter that is its operand is not blank. .ERRIDN generates an
error if the two macro parameters that are its operands are
identical, and .ERRDIF generates an error if the two macro
parameters that are its operands are different.
For example, the following macro generates an error if it's
invoked with any number of parameters other than two. This is
accomplished by using .ERRB and .ERRNB to make sure that
PARM2 isn't blank and PARM3 is blank.
The macro also uses .ERRIDN
to make sure that PARM21sn 't
DX, In which case it would be
wiped out when PARM 1 Is
loaded.
i
Macro to add two constants, registers, or named memory
locations and store the result in DX.
Input:
PARMl - one operand to add
PARM2 - the other operand to add
ADD TWO OPERANDS
MACRO PARMl,PARM2,PARM3
.ERRB
ithere must be two parameters
.ERRNB
; ••• but not three
.ERRIDN , isecond parameter can't be DX
mov dx,PARMl
add dx, PARM2
ENDM
Pitfalls in assembler programming
Each computer language has its own set of oft-encountered
programming problems, and assembly language is certainly no
exception. Here are some of the common pitfalls of assemblylanguage programming, along with tips on how to avoid them.
Chapter 6, More about programming In Turbo Assembler
225
Forgetting to
return to DOS
In Pascal, C, and other languages, a program ends automatically
and returns to DOS when there is no more code to execute, even if
no explicit termination command was written into the program.
Not so in assembly language, where only those actions that you
explicitly request are performed. When you run a program that
has no command to return to OOS, execution simply continues
right past the end of the program's code and into whatever code
happens to be in the adjacent memory.
For example, consider the following program:
. MODEL
small
.CODE
DoNothing PROC NEAR
nop
DoNothing ENDP
END DoNothing
Past experience might lead you to think that either the ENDP
directive or the END directive properly terminates this program,
just as } and end. do in C and Pascal, but that's not the case. The
executable code generated by assembling and linking this
program consists only of a single NOP instruction. In assembler,
the ENDP directive-like all directives-generates no code; it's
simply a note to the assembler that the code for the DoNo thing
procedure has ended. Similarly, the END DoNothing directive
merely tells the assembler that the code for this module has
ended, and that the program should start execution at DoNo thing.
Nowhere in the source code are instructions generated to transfer
control back to 005 when the program is finished; as a result,
when the program is run, whatever random instructions happen
to be lying in memory at the address following the NOP will be
executed immediately following the NOP. At this point, all bets
are off, with a hung computer and a soft or hard reboot far more
likely than the desired return to 005.
While there are several means by which an assembler program
can return to 005, the recommended technique is to execute 005
function 4Ch. The following version of the preceding program
termina tes properly:
226
Turbo Assembler User's Guide
•MODEL
small
.CODE
DoNothing PROC NEAR
nop
mov ah,4Ch
int 21h
DoNothing ENDP
END DoNothing
;DOS terminate process function
;invoke DOS to end program
Always remember that directives don't generate code, and that
Turbo Assembler generates programs that do exactly what your
source code tells them to do, no more and no less.
Forgetting a RET
instruction
Recall that the proper invocation of a subroutine consists of a call
to the subroutine from another section of code, execution of the
subroutine, and a return from the subroutine to the calling code.
Remember to insert a RET instruction in each subroutine, so that
the RETurn to the calling code occurs. When typing a program,
it's easy to skip a RET and end up with code like this:
; Subroutine to mUltiply a value by 80.
; Input: AX - value to multiply by 80
; Output: DX:AX - product
MultiplyBy80 PROC NEAR
mov dx,80
mul dx
MultiplyBy80 ENDP
; Subroutine to get the next key press.
; Output: AL - next key pressed
; AH destroyed
Get Key
mov
, int
ret
Get Key
PROC NEAR
ah,l
21h
PROC NEAR
The MultiplyByBO ENDP,directive can fool you into thinking that
MultiplyByBO has been terminated properly, when in fact the call
to MultiplyByBO not only multiplies AX by 80 but also continues
on into GetKey and returns the next key typed in AL. The proper
code for MultiplyByBO is
Subroutine to mUltiply a value by 80.
; Input: AX - value to multiply by 80
Chapter 6, More about programming in Turbo Assembler
227
; Output: DX:AX - product
MultiplyBy80 PROC NEAR
rnov dx,80
rnul dx
ret
MultiplyBy80 ENDP
Generating the
wrong type of The PROC directive has two effec~s. First, it defines a name by
return which a procedure can be called. Second, it controls whether the
procedure is a near or far procedure.
The type of a procedure-near or far-is used by the assembler to
determine what type of calls to generate when that procedure is
called from within the same source file. The type of a procedure is
also used to determine the type of RET performed when the
procedure returns control to the calling code. Consider the
following code:
; Near subroutine to shift DX:AX right 2 bits.
LongShiftRight2
shr dx,+
rer ax,l
shr dx,l
rer ax,l
ret
LongShiftRight2
PROC NEAR
;shift DX:AX right 1 bit
;shift DX:AX right another bit
ENDP
Turbo Assembler makes the RET in this code near, since
LongShiftRight2 is a near procedure. If the PROC directive is
changed to read
LongShiftRight2
PROC FAR
however, a far RET is generated.
So far, everything makes sense. After all, the RET instructions in a
procedure should match the type of the procedure, shouldn't
they?
Yes and no. The problem is that it's possible and often desirable to
group several subroutines in the same procedure. Since these
subroutines lack an associated PROC directive, their RET
instructions take on the type of the overall procedure, which is
not necessarily the correct type for the individual subroutines. For
example,
228
Turbo Assembler User's Guide
; Far subroutine to shift DX:AX right 2 bits.
LongShiftRight2 PROC FAR
call LongShiftRight
call LongShiftRight
ret
LongShiftRight:
shr dx,l
rcr ax,l
ret
LongShiftRight2 ENDP
;shift DX:AX right 1 bit
;shift DX:AX 'right another bit
;shift DX:AX right 1 bit
does not work properly. LongShiftRight2 makes near calls to
LongShiftRight, since they are both in the same code segment.
However, since LongShiftRight is embedded in the LongShiftRight2
procedure, the return at the end of LongShiftRight subroutine
becomes a far RET, and matching far calls with near returns is
likely to lead to a crash.
One good solution is to make sure that each subroutine has an
associated PROC directive. Nested PROC directives work well:
; Far subroutine to shift DX:AX right 2 bits.
LongShiftRight2 PROC FAR
call LongShiftRight
call LongShiftRight
ret
LongShiftRight
PROC NEAR
shr dx,l
rcr ax,l
ret
LongShiftRight
ENDP
LongShiftRight2 ENDP
;shift DX:AX right 1 bit
;shift DX:AX right another bit
;shift DX:AX right 1 bit
as do sequential PROC directives:
; Far subroutine to shift DX:AX right 2 bits.
LongShiftRight2 PROC FAR
call LongShiftRight
call LongShiftRight
ret
LongShiftRight2 ENDP
LongShiftRight
PROC NEAR
shr dx,l
rcr ax,l
ret
LongShiftRight
ENDP
Chapter 6, More about programming in Turbo Assembler
;shift DX:AX right 1 bit
;shift DX:AX right another bit
;shift DX:AX right 1 bit
229
You can also use RETN and RETF to explicitly generate a near or
far return, respectively. You can use these outside of a procedure
defined with the PRoe directive and rest assured that the correct
return will always be generated.
Reversing
operands
To many people, the order of instruction operands in 8086
assembly language seems backward, and there is certainly some
justification for this viewpoint. If the line
mov ax,bx
meant "move AX. to BX," the line would scan smoothly from left
to right, and this is the way many microprocessor manufacturers
have designed their assembly languages. However, Intel took a
different approach with 8086 assembly language; for us the line
means "move BX to AX," and that can sometimes cause
confusion.
The thinking behind the ordering of Intel's operands is that the
operands appear in the same order as they would in C or Pascal
code, with the destination on the left. Consequently, one way to
think of operand-ordering in 8086 assembly language is to
mentally insert an equal sign in place of the comma between
operands and reword the line to form an assignment. For
example, think of
mov ax,bx
as
ax
=
bx
Constant operands, such as
add bx, (OFFSET BaseTable
* 4)
t
2
which can be thought of as
bx t= (OFFSET BaseTable
* 4)
t 2
also lend themselves to this approach.
Forgetting the
stack or reserving
a too small stack
230
In most cases, you are treading on thin ice if you don't explicitly
alloca te space for a stack. Programs without an allocated stack
will sometimes run, since the default stack may happen to fall in
Turbo Assembler User's Guide
an unused area of memory. But there is no assurance that these
programs will run under all circumstances, since not a single byte
is guaranteed to be available for the stack.
Most programs should have a .STACK directive to reserve space
for the stack, and for each program that directive should reserve
more than enough space for the deepest stack you can conceive of
the program using.
Why more than enough space rather than just enough space? In
general, it's difficult to be sure just how much stack space a given
program needs, and the sort of bugs that occur when the stack
grows in to other parts of the program and overwrites them are
often very difficult to reproduce and track down. Then, too, many
debuggers use a little extra space on the stack when getting
control back from a program. So be generous when allocating
stack space, and save yourself future headaches. A minimum
stack size of 512 bytes is a good rule of thumb.
Writing .EXE rather than
.COM programs and
reserving ample stack space
Is a simple way to avoid
these potential problems.
Calling a
subroutine that
wipes out
needed registers
The only assembler programs that should not have a stack
allocated are programs that are going to be made into .COM or
.BIN files ..BIN files contain code hard-wired to run at a specific
address, and since .BIN files are generally used as interpreted
BASIC subroutines, they use BASIC's stack. .COM programs run
with the stack at the very top of the program's segment (which is a
maximum of 64K long, or less if there's less than 64K available), so
the maximum size of the stack is simply the amount of memory
left in the program's segment. Beware if any of the .COM
programs you write approach 64K in size, since the stack shrinks
accordingly. Also be aware that large .COM programs may
encounter stack problems when run on computers with little
available memory or when run from a 005 shell under another
program.
When writing assembler code, it's easy to think of the registers as
local variables, dedicated to the use of the procedure you're
working on at the moment. In particular, there's a tendency to
assume that registers are unchanged by calls to other procedures.
It just isn't so, though-the registers are global variables, and each
procedure can preserve or destroy any or all registers.
As an example, consider the following:
Chapter 6. More about programming in Turbo Assembler
231
mov
mov
call
add
bx, [TableBase]
ax, [Element]
DivideBy10
bx,ax
ipoint BX to base of table
iget element t
idivide element t by 10
ipoint to appropriate entry
Subroutine to divide a value by 10.
Input: AX - value to divide by 10
i Output: AX - value divided by 10
DX - remainder of value divided by 10
i BX destroyed.
DivideBy10
PROC NEAR
mov dx,O
mov bx,10
div bx
ret
DivideBy10
ENDP
iprepare DX:AX as 32-bit dividend
iBX is the 16-bit divisor
The calling routine assumes that BX is preserved by DivideByl0,
when in fact DivideByl0 sets BX to 10. There are a number of
possible solutions in this particular case. BX could be pushed and
popped either at the start or end of DivideByl0:
mov
mov
call
add
bx, [TableBase]
ax, [Element]
DivideBy10
bx,ax
;point BX to base of table
iget element t
idivide element t by 10
ipoint to appropriate entry
Subroutine to divide a value by 10.
Input: AX - value to divide by 10
i Output: AX - value divided by 10
DX - remainder of value divided by 10
DivideBy10
PROC NEAR
push bx
mov dx,O
mov bx,10
div bx
pop bx
ret
DivideBy10
ENDP
ipreserve BX
iprepare DX:AX as 32-bit dividend
iBX is the 16-bit divisor
irestore original BX
or in the calling routine around the call to DivideByl0:
mov bx, [TableBase]
mov ax, [Element]
push bx
232
ipoint BX to base of table
;get element t
ipreserve table base
Turbo Assembler User's Guide
call DivideBy10
pop bx
add bx,ax
idivide element I by 10
irestore table base
ipoint to appropriate entry
Subroutine to divide a value by 10.
Input: AX - value to divide by 10
i Output: AX - value divided by 10
DX - remainder of value divided by 10
DivideBy10
PROe NEAR
mov dx,O
mov bx,10
div bx
ret
DivideBy10
ENDP
iprepare DX:AX as 32-bit dividend
iBX is the 16-bit divisor
or BX could simply be loaded after, rather than before, the call
mov
call
mov
add
ax, [Element]
DivideBy10
bx, [TableBase]
bx,ax
iget element I
idivide element I by 10
ipoint BX to base of table
ipoint to appropriate entry
Subroutine to divide a value by 10.
Input: AX - value to divide by 10
Output: AX - value divided by 10
DX - remainder of value divided by 10
DivideBy10
PROe NEAR
mov dx,O
mov bx,10
div bx
ret
DivideBy10
ENDP
iprepare DX:AX as 32-bit dividend
iBX is the 16-bit divisor
An obvious solution to the general problem of subroutines that'
accidentally clobber registers is for all subroutines to preserve all
registers as a matter of course. Unfortunately, pushing and
popping registers takes time and code space, negating some of the
advantages of programming in assembler. Another approach is to
preface each subroutine with a comment indicating which
registers are preserved and which are destroyed. Then carefully
check that there are no problems in each case where you must ,(,
assume a register is preserved across a subroutine call. Yet
,'",
another approach is to explicitly preserve needed registers in
calling,routines.
Chapter 6, More about programming In Turbo Assembler
233
Using the wrong
sense for a
conditional jump
The profusion of conditional jumps in assembly language (JE,
JNE, JC, JNC, JA, JB, JG, and so on) allows tremendous flexibility
in writing code-and also makes it easy to select the wrong jump
for a given purpose. Moreover, since condition-handling in
assembly language requires at least two separate lines, one for the
comparison and one for the conditional jump (and many more
lines for complex conditions), assembly language conditionhandling is less intuitive and more prone to errors than
condition-handling in C and Pascal.
• One common error is the use of JA, JB, JAE, or JBE for
comparing signed values or, similarly, the use of JG, JL, JGE, or
JLE for comparing unsigned values.
• Another common error is the use of, say, JA when JAE was
intended. Remember that without the e on the end of JAE, JBE,
JLE, or JGE, the comparison does not include the case where
the two operands are equal.
• And yet another common error is the use of inverted logic, such
as JS when JNS was intended.
One approach that can help minimize errors when using
conditional jumps is to comment the tests and conditional jumps
in C-like notation. For example,
if ( Length > MaxLength )
mov ax, [Length]
cmp ax, [MaxLength]
jng LengthIsLessThanMax
jmp EndMaxLengthTest
} else {
LengthIsLessThanMax:
EndMaxLengthTest:
234
Turbo Assembler User's Guide
Pitfalls with string
instructions String instructions are uniquely powerful among 8086
instructions, and with that power come some unique problems,
which are described next.
Forgetting about REP
string overrun
String instructions have a curious property: After they're
executed, the pointers they use wind up pointing to an address 1
byte away (or 2 bytes if a word instruction) from the last address
processed. For example, after this code executes
cld
mov si,O
lodsb
irnake string instructions count up
ipoint to offset a
iread the byte at offset a
51 will contain 1, not O. This makes sense, since the next LOOSe is
likely to want to access address 1, and the LOOSe after that to
access address 2, but it can cause some confusion with repeated
string instructions, especially REP SCAS and REP CMPS.
Consider the code
cld
les di, [bp+ScanString]
mov cx,MAX_STRING_LEN
mov al,O
repne scasb
imake string instructions count up
ipoint ES:DI to the string to scan
icheck up to the longest string
isearch for the terminating null
iperform search
Suppose ES is 2000h, DI is 0, and the memory starting at 2000:0000
contains
41h 61h 72h 64h OOh
After this code executes, DI will contain 5, the offset of the byte
after the 0 byte that was found. In order to return a pointer to the
last character in the string, the preceding code would have to read
cld
les di, [bp+ScanString]
mov cx,MAX_STRING_LEN
mov al,O
repne scasb
jne NoMatch
Chapter 6, More about programming in Turbo Assembler
imake string instructions count up
ipoint ES:DI to the string to scan
icheck up to the longest string
isearch for the terminating zero
iperform search
ierror-terminating zero not found
235
dec
dec
ret
NoMatch:
mov
mov
ret
di
di
;point back to the zero
;point back to last character
di,O
es,di
;return a null pointer
Remember also that when the direction flag is set, causing string
instructions to count down, DI will point to the byte before, not
after, the last character scanned.
Similar confusion can arise because ex is decremented during
REP SCAS and REP CMPS one more time than might be
expected. ex is not only decremented once for each byte that
matches the "repeat while" condition (equal or not equal), but
also once for the byte that fails to match the "repeat while"
condition and thereby causes the instruction to terminate. For
instance, if in the last example the byte at 2000:0000 contained
zero, after execution ex would contain MAX_STRING_LEN -1,
even though not a single nonzero character was found. A
subroutine to count the number of characters in a string must
account for this:
; Returns the length of a zero-terminated string in bytes.
; Input: ES:DI - start of string
; Output: AX - length of string, not including terminating 0
ES:DI - points to last byte of string, or
0000:0000 if terminating 0 not found
StringLength PROe NEAR
cld
;search counts up
push cx
;preserve ex
mov cx,OFFFFh
;maximum length to search
mov al,O
;terminating byte to search for
;search for the terminating 0
repne scasb
;error if end of string not found
jne StringLengthError
mov ax,OFFFFh
;maximum length searched
;see how many bytes were counted
sub ax,cx
dec ax
;don't count the terminating zero
dec di
;point back to terminating zero
dec di
;point back to last character
jmp short StringLengthEnd
StringLengthError:
mov di,O
;return a null pointer
mov es,di
StringLengthEnd:
236
Turbo Assembler User's Guide
pop cx
ret
StringLength
irestore the original CX
ENDP
Another potential problem arising from ex counting on the byte
that terminates a REP SCAS or REP CMPS is that ex might be
zero at the end of the comparison even though the termination
condition was found. This code does not correctly evaluate
whether two arrays are the same, since ex will count down to
zero when comparing two non-equal arrays that differ only at the
last byte:
repz cmpsb
jcxz ArraysAreTheSame
The correct code for testing array equality is
repz cmpsb
jz ArraysAreTheSame
In short, ex should be used only as a count of the bytes scanned
by REP SCAS and REP CMPS, not as an indicator of whether the
data scanned or compared was equal or non-equal.
If you find yourself having trouble figuring out just what repeated
string instructions will do in your programs, one good approach
is to use either pencil and paper or a debugger to trace, step-bystep, through the workings of your repeated string code.
Relying on a zero ex to
cover a whole
segment
Any repeated string instruction executed with ex equal to zero
will do nothing. Period. This can be convenient in that there's no
need to check for the zero case before executing a repeated string
instruction; on the other hand, there's no way to access every byte
in a segment with a byte-sized string instruction. For example, the
following code 'scans the segment at ES for the first occurrence of
the letter A:
cld
sub di,di
mov al,'A'
mov cx,OFFFFh
repne SCASb
je AFound
isearches count up
istart at offset zero
isearch for letter 'A'
ifirst scan the first 64 Kb-l bytes
iscan the first 64 Kb-l bytes
i found it
Chapter 6. More about programming in Turbo Assembler
237
scasb
je AFound
AFound:
ididn't find it yet-scan the last byte
ifound it at the last byte
i there's no letter 'A' in this segment
iDI - 1 points to the letter 'A'
There's an asymmetry in the 8086 instruction set concerning the
use of zero ex values when counting. While repeated string
instructions don't do anything if ex is 0, the LOOP instruction
does execute if ex is 0, decrementing ex to OFFFFh and jumping
to the loop address. This means that a full64K can be processed in
a single loop. The preceding example of scanning the segment at
ES for the letter A can be implemented with LOOP as
isearches count up
istart at offset zero
cld
sub di,di
mov al,'A'
sub cx,cx
ASearchLoop:
scasb
je AFound
loop ASearchLoop
icheck the next byte
iit's a letter 'A'
ithere's no letter 'A' in this segment
AFound:
iDI - 1 points to the letter 'A'
isearch 64 Kb bytes
On the other hand, the case of ex equal to zero does have to be
specially checked for when using LOOP in those cases where ex
equal to zero really does mean, "Don't do anything"; otherwise,
64K loops instead of zero loops will be executed with potentially
disastrous results. The JCXZ instruction helps you handle such
cases:
Subroutine to fill up to 64K -1 byte with a given byte value.
Input: AL - fill value
ex - number of bytes to fill
DS:BX - first address to fill
BX, ex altered.
FillBytes PRoe NEAR
jcxz FillBytesEnd
FillBytesLoop:
mov [bx],al
inc bx
loop FillBytesLoop
FillBytesEnd:
ret
238
iif the I of bytes to fill is 0, done
i fill
a byte
ipoint to the next byte
ido for the number of bytes specified
Turbo Assembler User's Guide
FillBytes ENDP
Without JCXZ, FillBytes would fill the entire segment pointed to
by E5 with AL when ex was zero, instead of leaving memory
unchanged.
Using incorrect
direction flag settings
When a string instruction is executed, its associated pointer or
pointers-51 or DI or both-increment. Or decrement. It all
depends on the state of the direction flag.
The direction flag can be cleared with CLD to cause string
instructions to increment (count up) and can be set with STD to
cause string instructions to decrement (count down). Once cleared
or set, the direction flag stays in the same state until either
another CLD or STD is executed or the flags are popped from the
stack with POPF or IRET. While it's handy to be able to program
the direction flag once and then execute a series of string
instructions that all operate in the same direction, the direction
flag can also be responsible for intermittent and hard-to-find bugs
by causing string instructions to behave differently, depending on
code that executed much earlier.
Why is this? In most programs, the direction flag is almost always
cleared, since counting up is intuitively easier than counting
down and works fine in most cases. There are, however, certain
cases where only counting down will do. You can get in the habit
of assuming that the direction flag will always be cleared, but
forget to clear the flag after one of the few procedures that sets the
direction flag. The result will be that parts of your program that
require counting up will work perfectly-except after executing
that one procedure that leaves the direction flag set.
The remedy is obvious. Always program the direction flag to the
desired state before using string instructions if there is any chance
that the direction flag is not already programmed correctly. In
general, it's a good idea to program the direction flag correctly at
the beginning of any procedure that uses string instructions.
Using the wrong sense
for a repeated string
comparison
The CMPS instruction compares two areas of memory, while the
SCAS instruction compares the accumulator to an area of
memory. When prefixed by REPE, either of these instructions can
perform a comparison until either ex becomes zero or a not-equal
comparison occurs. When prefixed by REPNE, either instruction
can perform a comparison until either ex becomes zero or an
Chapter 6, More about programming In Turbo Assembler
239
equal comparison occurs. Unforhmately, it's easy to become
confused about which of the REP prefixes does what.
A good way to remember the function of a given REP prefix is to
mentally insert a "while" after the "rep" portion of the prefix.
Then REPE becomes "rep while e," or "repeat while equal," and
REPNE becomes "rep while ne," or "repeat while not equa1."
Forgetting about string
segment defaults
Each string instruction defaults to using a source segment (if any)
of OS, and a destination segment (if any) of ES. It's easy to forget
this and try to perform, say, a STOSB to the data segment, since
that's where all the data you're processing with non string
instructions normally resides. Similarly, it's common to
accidentally write code such as
cld
mov
mov
repe
jz
dec
mov
icount up while searching
al,O
cx,80
scasb
AllZero
di
aI, [di]
ilength of buffer
ifind first nonzero character, if any
ino nonzero character
ipoint back to first nonzero character
iget first nonzero character
AllZero:
Refer to Chapter 9 for an
explanation of segment
prefixes.
The problem with this code is that unless DS and ES are the same,
the last MOV won't load the correct byte into AL, since STOSB
operates relative to ES and MOV operates relative to DS. The
correct code would use a segment override prefix on the move.
cld
mov
mov
repe
jz
dec
mov
icount up while searching
al,O
cx,80
scasb
AllZero
di
al,es:[di]
ilength of buffer
ifind first nonzero character, if any
ino nonzero character
ipoint back to first nonzero character
iget first nonzero character (from ES!)
AllZero:
Also, remember that while it is possible to override OS as the
string source segment, as, for example, in
240
Turbo Assembler User's Guide
lods es:[SourceArray]
it is not possible to override ES as the string destination segment,
so this code won't work:
stos ds:[DestArray]
In fact, Turbo Assembler catches this as an error during assembly.
Converting incorrectly
from byte to word
operations
In general, it's desirable to use the largest possible data size
(usually word, but dword on an 80386) for a string instruction,
since string instructions with larger data sizes often run faster. For
example,
mov cx,200
inumber of bytes to move
shr cx,l
rep movsw
iconvert from t of bytes to t of words
imove the block a word at a time
runs about 50% faster on an 8088 than
mov cx,200
inumber of bytes to move
rep movsb
imove the block a byte at a time
There are a couple of potential pitfalls here, though. First, the
conversion from a byte count to a word count by a simple
shr cx,l
loses a byte if CX is odd, since the least-significant bit is shifted
out. Cases where CX might be odd can be handled with the
following conditional code:
shr cx,l
jnc MoveWord
movsb
MoveWord:
rep movsw
iconvert to word count
iodd byte count?
iyes, odd byte count, so move odd byte
imove even t of bytes a word at a time
Chapter 6, More about programming in Turbo Assembler
241
Second, make sure you remember SHR divides the byte count by
two. Using, say, STOSW with a byte rather than a word count can
wipe out other data and cause all sorts of problems. For example,
mov cx,200
;number of bytes to move
rep movsw
;move the block a word at a time
will wipe out the 200 bytes (100 words) immediately following the
destination block.
USing multiple prefixes
"String instructions with multiple prefixes do not work reliably
and should generally be avoided. An example is this code
rep movs es:[DestArray],ss:[SourceArray]
which has both a REP prefix and an SS segment override prefix.
Multiple prefixes are a problem because string instructions can be
interrupted in the middle of repeated execution by a hardware
interrupt. On some Intel processors, including the 8086 and 8088,
when a string instruction with multiple prefixes resumes after an
interrupt has been serviced, all prefixes other than the last are
ignored. As a result, the instruction might not be repeated the
correct number of times or the wrong segment might be accessed.
If you absolutely must use a string instruction with multiple
prefixes, disable interrupts for the duration of the instruction, as
follows:
cli
rep movs es:[DestArray],ss:[SourceArray]
sti
Relying on the
operand(s) to a string
instruction
The optional operand or operands to a string instruction are used
for data sizing and segment overrides, only, and do not guarantee
that the memory location referenced will actually be accessed. For
example,
DestArray dw
cld
242
256 dup (?)
;count up during fill
Turbo Assembler User's Guide
mov
mov
mov
rep
al,'*'
cx,256
di,O
stos es:[DestArray]
ibyte to fill with
inumber of words to fill
istart address for fill
ido the fill
sets the 256 bytes starting at offset 0 in segment ES to the asterisk
character, regardless of where DestArray is located. All that
ES:[DestArray) does is tell the assembler to use a STOSW, since
DestArray is an array of words. It is the contents of SI and/or DI,
not the operands, thafdetermine what offsets are accessed by
string instructions. Nonetheless, using the optional operand or
operands with string instructions can be a useful way of ensuring
that you're not accidentally performing, say, word-sized accesses
to a byte array.
Similarly, the optional operand to the XLAT instruction is used for
type-checking and segment overrides only. The code
LookUpTable
LABEL BYTE
ASCIITable
LABEL BYTE
mov bx,OFFSET ASCIITable
mov aI, [CharacterToTranslate]
xlat [LookUpTable]
ipoint to look-up table
iget the byte looked up
ilook the byte up
looks up the byte at location AL in ASCIITable, not LookUpTable,
but assembles just fine because all XLAT does with its one
operand is make sure that it is byte-sized and looks for a segment
override. The XLAT instruction always looks up the contents of
offset BX+AL, regardless of any operand used.
Forgetting about
unusual side
effects
Since assembler programs are written in the 8086's native
language, any changes in the 'states of the registers and flags of
the 8086 are of keen interest to the assembly language programmer. Most of the ways in which assembler programs can alter the
state of the processor are obvious and straightforward. For
example,
add bx, [Grade]
adds the 16-bit value at location Grade to BX and updates the
overflow, sign, zero, auxiliary carry, parity, and carry flags to
Chapter 6, More about programming In Turbo Assembler
243
reflect the outcome of the addition. Some instructions produce
less obvious changes in the state of the processor, though. Here's a
quick look at some such instructions.
Wiping out a register
with multiplication
Table 6.1
Source and
destination for the
MULand IMUL
Instructions
Multiplication-whether it be 8 bit by 8 bit, 16 bit by 16 bit, or 32
bit by 32 bit-always destroys the contents of at least one register
other than the portion of the accumulator used as a source
operand. This is inevitable given that the result of an 8 bit by 8 bit
multiplication can be as large as 16 bits in size, the result of a
16 bit by 16 bit multiplication can be 32 bits in size, and the result
of a 32 bit by 32 bit multiplication can be 64 bits in size. Multiplication source and destination operands are shown in Table 6.1.
Source
operand
size in bits
Source
Implied
Explicit
operand
operand
16x16
reg8"
reg16 ....
32x32t
reg32:t:
8x8
Destination
High
Low
Example
AL
AH
AL
mul dl
AX
EAX
DX
EDX
AX
EAX
imul bx
mul esi
.. regS can be any of AH, AL, BH, BL, CH, CL, OH, or OL.
.... reg16 can be any of AX, BX, CX, OX, 51, 01, BP, or SP.
t 32 x 32 multiples are not supported by the 8086, 8088, 80186, 80188, or 80286.
t reg32 can be any of EAX, EBX, ECX, EOX, ESI, EDI, EBP, or ESP.
While this seems simple enough, there's a glaring lack of detail in
the syntax of the MUL and IMUL instructions, since only one of the
two source operands and the size of the operation are explicitly
stated; both the portion of the accumulator used as a source
operand and the registers used as the destination are merely
implied. This lack of detail makes it easy to overlook the extra
register that's destroyed. For instance, there are many cases in
which the result of, say, a given 16-bit by 16-bit multiplication is
known by the programmer to be guaranteed to fit in AX, and in
such cases, there's a tendency to forget that DX gets wiped out
too. Just remember that every use of MUL and IMUL wipes out not
only AL, AX, or EAX, but also AH, DX, or EDX as well.
244
Turbo Assembler User's Guide
Forgetting that string
instructions alter
several registers
The string instructions (MOVS, STOS, LODS, CMPS, and SCAS)
can affect several of the flags and as many as three registers
during execution of a single instruction. As with the MUL
instruction, the many effects of the string instructions are not
explicitly expressed in the operands to those instructions. When
you use string instructions, remember that either SI or DI or both
either increment or decrement (depending on the state of the
direction flag) on each execution of a string instruction. ex is also
decremented at least once and possibly as far as zero each time a
string instruction with a REP prefix is used.
Expecting certain
instructions to alter the
carry flag
While some instructions affect registers or flags unexpectedly,
other instructions 'don't affectall the flags you might expect them
to. For example,
inc ah
seems logically equivalent to
add ah,l
and so it is-with a single exception. Where ADD sets the carry
flag if the result is too large for the destination, INC does not affect
the carry flag in any way. As a result,
add ax,l
adc dx,O
is a valid way to increment a 32-bit value stored in DX:AX, while
inc ax
adc dx,O
is not. The same is true of DEC, while LOOP, LOOPZ, and
LOOPNZ don't affect any flags at all. Actually, this can sometimes
be used to your advantage, since under certain circumstances it
can be handy to execute one of these instructions without
destroying the current carry flag setting. The important thing is to
know exactly what each instruction you use does.
Chapter 6, More about programming in Turbo Assembler
245
Waiting too long to use
flags
Flags last only until the next instruction that alters them, which is
not very long, by and large. It's a good practice to act on flags as
soon as possible after they are set, thereby avoiding all sorts of
potential bugs. For example, it's often tempting to test a condition,
set a register or two, and only then branch according to the result
. of the test. The code
cmp ax,l
mov ax,O
jg HandlePositive
is a perfectly valid way to test the status of AX, then force it to
zero before jumping to the code that handles the status. On the
other hand, the code
cmp ax,l
sub ax,ax
jg HandlePositive
which seems appealing because it is both shorter and faster than
the first case, does not work because the subtraction wipes out all
the flag settings generated by the compare. This is typical of the
sort of problem that can result from delaying the use of a flag
status.
Confusing
memory and
immediate
operands
An assembler program can refer either to the offset of a memory
variable or to the value stored in that memory variable.
Unfortunately, assembly language is neither strict nor intuitive
about the ways in which these two types of references can be .
made, and as a result, offset and value references to a memory
variable are often confused.
Figure 6.1 illustrates the distinction between the offset and the
value of a memory variable. The offset of the word-sized variable
MemLoc is 5002h, while the value of MemLoc is 1234h.
246
Turbo Assembler User's Guide
Figure 6.1
Memory variables:
offset vs. value
The offset
of MemLoc
3000:4FFE
0001
3000:5000
205F
3000:5002
(1234 '\
3000:5004
9145
3000:5006
0000
=r-
The value
of MemLoc
In Figure 6.1, the offset of the word-sized variable MemLoc is the
constant value SOO2h, obtained with the OFFSET operator. For
example,
,
rnov bx,OFFSET MernLoc
loads S002h into BX. The value SO02h is an immediate operand; in
other words, it is built right into the instruction and never
changes.
The value of MemLoc is 1234h, read from the memory at offset
S002h in the data segment. One way to read this value is by
loading BX, 51, DI, or BP with the offset of MemLoc and using that
register to address memory. The code
rnov bx,OFFSET MernLoc
rnov ax, [bx]
loads the value of MemLoc, 1234h, into AX. Alternatively, the
value of MemLoc can be loaded directly into AX with either
rnov ax,MernLoc
or
rnov ax, [MernLoc]
Here the value 1234h is obtained as a direct, rather than an
immediate, operand; the MOV instruction has the offset S002h
built into it, and loads AX with the value at S002h, which in this
case happens to be 1234h. Consequently, the value 1234h is not
permanently associated with MemLoc. For instance,
rnov
[MernLoc],5555h
Chapter 6, More about programming in Turbo Assembler
247
mov ax, [MemLoc]
loads the value 5555h, not 1234h, into AX.
The key point is that while the offset of MemLoc is a constant
value that describes a fixed address in the data segment, the value
of MemLoc is the changeable number stored at that memory
address. The instructions
mov
add
[MemLoc],l
[MernLoc],2
make the value of MemLoc 3, but the instruction
add OFFSET MernLoc,2
is equivalent to
add 5002h,2
which is nonsensical, since it's impossible to add one constant to
another.~
A surprisingly common problem is that in the heat of coding a
program, OFFSET is sometimes forgotten, leaving, for example,
mov si,MernLoc
when the offset of MemLoc is desired. At first glance, this line
doesn't look wrong, and since MemLoc is a word-sized variable,
this line will not cause an assembly-time error. However, at runtime 5I will be loaded with the data at MemLoc (1234h in Figure
6.1 on page 247), rather than the offset of MemLoc (5002h in Figure
6.1}-with unpredictable results.
There is no sure-fire way to avoid this prqblem, but you might
want to make it a rule to enclose all references to memory in
square brackets. Then references to address constants will be
prefixed with OFFSET and references to memory will be enclosed
in square brackets, thus eliminating the ambiguous use of
memory variable names. This convention makes the functions of
mov si,OFFSET MernLoc
and
mov si, [MemLoc]
instantly clear, while
mov si,MemLoc
should set off mental alarms.
248
Turbo Assembler User's Guide
Causing segment
wraparound One of the most difficult aspects of programming the 8086 is that
memory isn't accessible as one long array of bytes, but is rather
made available in chunks of 64K relative to segment registers.
Segments can introduce subtle bugs, since if ~ program attempts
to access an address past the end of a segment, it actually ends up
wrapping back to access the start of that segment instead.
As an example, suppose that the memory starting at 10000h
contains the data shown in Figure 6.2. When DS is set to 1000h,
code that accesses the string "Testing" at 1000:FFF9 wraps back to
address the byte at 1000:0000 as the next byte addressed after the
gat 1000:FFFF because offsets cannot exceed OFFFFh, the
maximum 16-bit value.
Now suppose that the following subroutine is called with DS:SI
equal to 1000:FFF9 in order to convert the string "Testing" at
10OO:FFF9 to uppercase:
Subroutine to convert a zero-terminated string to uppercase.
i Input: DS:SI - pointer to string.
ToUpper PROC NEAR
mov aI, lsi]
cmp al,O
jz ToUpperDone
cmp aI,' a'
jb ToUpperNext
cmp aI,' z'
ja ToUpperNext
and al,NOT 20h
mov [si], al
ToUpperNext:
inc si
jmp ToUpper
ToUpperDone:
ret
ToUpper ENDP
iget the next character
i if zero •••
i ••• done with string
iis it a lowercase letter?
inot lowercase
inot lowercase
iit's lowercase, so make it uppercase
isave the uppercase version
ipoint to the next character
Chapter 6, More about programming In Turbo Assembler
249
Figure 6.2
An example of
segment
wraparound
10000
21
10001
90
10002
29
10003
52
10004
7F
1FFF9
54 ('T')
1FFFA
65 ('e')
1FFFB
73 ('s')
1FFFC
74 ('t')
1FFFO
69 ('i')
1FFFE
6E ('n')
1FFFF
67 ('g')
20000
00 (NULL)
~
-
First byte addressable
relative to OS 1000h
(Address 1000:0000)
=
Last byte addressable
relative to OS 1000h
(Address 1000:FFFF)
=
After ToUpper processes the first seven characters of the string, SI
will increment from OFFFFh to O. (Recall that SI is only a 16-bit
register and so can't count higher than OFFFFh.) The zero byte
stored at address 20000h that terminates the string is never
reached; instead ToUpper starts to convert the unrelated bytes at
10000h to uppercase, and doesn't stop until it happens to
encounter a 0 byte. At some later point, these altered bytes may
cause this program to perform incorrectly. Often, it is very
difficult to trace bugs caused by such accidentally altered bytes
250
Turbo Assembler User's Guide
back to the routine that wrapped off the end of a segment, since
the cause can be far distant from the symptom in time and may be
in a totally unrelated portion of the source code.
There's no simple rule of thumb here, other than always making
sure your code doesn't unwittingly try to run off the end of a
segment. It is also very dangerous (to your sanity, at least) to try
to access a word at offset OFFFFh; the machine will hang.
Failing to
preserve
everything in an
interrupt handler
An interrupt handler is a routine that is jumped to whenever a
given hardware interrupt, such as the keyboard interrupt, occurs.
Interrupt handlers perform a variety of actions, such as buffering
keys or updating the system clock. An interrupt might occur at
any time, in the middle of any code, so an interrupt handler must
leave the registers and flags of the processor in exactly the same
state on exit from the handler as they were in on entry to the
handler. Were this not done, the code executing when an interrupt
occurs might suddenly find that the state of the processor has
changed unpredictably.
For instance, if the code
mov ax, [ReturnValuel
ret
were executing, an interrupt could occur between the two
instructions. If the interrupt handler fails to preserve the contents
of AX, the value returned to the calling program would be based
on what the interrupt handler did rather than on the contents of
the Return Value variable.
Consequently, every interrupt handler should explicitly preserve
the contents of all registers. While it is valid to explicitly preserve
only those registers that the handler modifies, it's good insurance
to just push all registers on entry to an interrupt handler and pop
all registers on exit. After all, you might go back someday and
change the code of the interrupt handler-so that it modifies
additional registers-but forget to add instructions to preserve
those registers.
It is not necessary to save the flags in an interrupt handler. When
an interrupt occurs, the flags are automatically pushed on the
stack, and when the interrupt handler executes an IRET to return
Chapter 6, More about programming in Turbo Assembler
251
to the interrupted program, the flags are automatically restored
from the stack.
A corollary to the absolute necessity of preserving all registers in
an interrupt handler is this: Make no assumptions about the state of
the registers or flags when an interr~pt handler is entered. A classic
example of this is an interrupt handler that executes string
instructions without first explicitly setting the direction flag.
Remember, any sort of code can be executing when an interrupt
occurs, so after you save the interrupted code's registers, you
must immediately set up the registers (including segment
registers) and flags as needed by your code before doing anything
else.
Forgetting group
overrides in
operands and
data tables
Figure 6.3
Three segments
grouped into one
segment group
The concept of a segment group is simple and useful: You specify
that several segments belong in the same group, and the linker
combines those segments into a single segment, with all the data
in all the grouped segments addressable relative to the same
segment register. Figure 6.3 illustrates three segments, Segl, Seg2,
and Seg3, grouped into GroupSeg.
------
Offset 0 In GroupSeg
= offset 0 In Seg1
-.
_.
_.
Seg1
----
_.
_.
_.
(SK long)
_.
_.
-
Offset 2000h In GroupSeg
= offset 0 In Seg2
-
Offset 5000h In GroupSeg
= offset 0 In Seg3
_.
----
_.
_.
_.
Seg2
---- (12K long)
-.
_.
_.
_.
_.
_.
_.
_.
252
Seg3
(SK long)
Turbo Assembler User's Guide
All three segments are addressable simultaneously, relative to a
single segment register loaded with the base address of GroupSeg.
Segment groups allow you to logically partition data into a
number of areas without having to load a segment register every
time you want to switch from one of those logical data areas to
another.
Unfortunately, there are a few problems with the way the
Microsoft Macro Assembler (MASM) handles segment groups, so
until Turbo Assembler came along, segment groups were quite a
nuisance in assembler. They were, however, an unavoidable
nuisance, for they are required in order to link assembler code to
high-level languages such as C.
Turbo Assembler Ideal mode has none of the problems with
group overrides described in this section. This is yet another good
reason to make the switch from MASM-style coding to Ideal
mode.
One problem MASM has with segment groups is that MASM
treats all offsets obtained with the OFFSET operator in a given
grouped segment as offsets into that segment, rather than as
offsets into the segment group. For example, given the segment
grouping shown in Figure 6.3, the assembler would assemble
mov ax,OFFSET VaIl
into
mov ax,O
since Varl is at offset 0 in Seg2, even though Var1 is at offset 2000h
in GroupSeg. Since data in segment groups is always intended to
be addressed relative to the segment group rather than the
individual segments, this creates quite a problem.
There is a solution to this problem, and that's using a group
override prefix. The line
mov ax,OFFSET GIoupSeg:Varl
does assemble the offset of Varl correctly, calculating it relative to
the segment group, GroupSeg.
MASM has another, similar problem concerning data tables used
with segment groups. Just as with the OFFSET operator, offsets
assembled into data tables are generated relative to segments, not
Chapter 6, More about programming in Turbo Assembler
253
segment groups. The following code shows an example of this
problem.
Stack
DB
Stack
SEGMENT WORD STACK 'STACK'
512 DUP(?)
ireserve space for a 1/2K stack
ENDS
i Define data segment group DGROUP, consisting of Datal & Data2.
DGROUP
GROUP
Datal, Data2
i The first segment in DGROUP.
Datal
Scratch
Datal
SEGMENT WORD PUBLIC 'DATA'
DB 100h DUP(O)
ia 256-byte scratch buffer
ENDS
; The second segment in DGROUP.
SEGMENT woRn PUBLIC 'DATA'
DB 100hDUP('@')
ia 256-byte buffer,
i set to @-signs
ia pointer to Buffer
BufferPtr DW Buffer
Data2
ENDS
Data2
Buffer
Code SEGMENT PARA PUBLIC 'CODE'
ASSUME
CS:Code, DS:DGROUP
Start
PROC NEAR
mov ax,DGROUP
mov ds,ax
mov bx,OFFSET DGROUP:BufferPtr
ipoint DS to DGROUP
ipoint to buffer pointer
Note: The DGROUP: group override is required to get the
correct offset.
mov bx, [bx]
ipoint to the buffer itself
(Code to handle the buffer would go here.)
mov ah,4Ch
int 21h
Start
ENDP
Code ENDS
END Start
iDOS terminate function
iterminate & return to DOS
In this code, the offset of BufferPtr in
254
Turbo Assembler User's Guide
rnov bx,OFFSET OGROUP:BufferPtr
assembles correctly, since the DGROUP: group override prefix is
used. However, the other reference to an offset, in
BufferPtr
ow
Buffer
which should cause the value of BufferPtr to be initialized to the
offset of Buffer, does not assemble correctly, since the offset of
Buffer is taken relative to the Data2 segment rather than relative to
the DGROUP segment group. The solution is again a DGROUP
override prefix; change
BufferPtr
ow
Buffer
BufferPtr
ow
OGROUP:Buffer
to
i
i
ia
pointer to Buffer
Note: The OGROUP: group override is required to get the
correct offset.
Omission of group override prefixes when using segment groups
in MASM/Quirks mode can lead to some nasty bugs, since your
programs might read, modify, or jump to the wrong area of
memory. As a general rule, don't use groups in assembler with
MASM/Quirks mode unless you have to. When you have to use
groups in MASM/Quirks mode, as when interfacing to high-level
languages, constantly remind yourself to prefix group overrides
when specifying the offsets of all grouped data. The group
overrides are easy enough to use-the trick is remembering to use
them.
A useful technique for dealing with grouped segments in
MASM/Quirks mode is using LEA instead of MOV OFFSET. For
example,
lea ax,Varl
has the same effect as
mov ax, OFFSET GroupSeg:Varl
without requiring a group override prefix. However, LEA is a byte
larger and a little slower than MOV OFFSET.
By the way, segment group problems occur only with offsets, not
with memory accesses. Lines such as
mov ax, [Varl]
Chapter 6, More about programming in Turbo Assembler
255
do not require group override prefixes.
256
Turbo Assembler User's Guide
c
H
A
p
T
E
R
7
Interfacing Turbo Assembler with Turbo
C
While many programmers can-and do-develop entire
programs in assembly language, many others prefer to do the
bulk of their programming in a high-level language, dipping into
assembly language only when low-level control or very highperformance code is required. Still others prefer to program
primarily in assembler, taking occasional advantage of high-level
language libraries and constructs.
_ Turbo C lends itself particularly well to supporting mixed C and
assembler code on an as-needed basis, providing not one but two
mechanisms for integrating assembler and C code. The inline
assembly feature of Turbo C provides a quick and simple way to
put assembler code directly into a C function. For those who
prefer to do their assembler programming in separate modules
written entirely in assembly language, Turbo Assembler modules
can be assembled separately and linked to Turbo C code.
First, we'll cover the use of inline assembly in Turbo C. Next,
we'll discuss the details of linking separately assembled Turbo
Assembler modules to Turbo C, and explore the process of calling
Turbo Assembler functions from Turbo C code. Finally, we'll
cover calling Turbo C functions from Turbo Assembler code.
(Note: When we refer to Turbo C, we mean versions 1.5 and
greater.) Let's begin.
Chapter 7, Interfacing Turbo Assembler with Turbo C
257
Using inline assembly in Turbo C
If you were to think of an ideal way to use assembler to fine-tune
a C program, you would probably ask for the ability to insert
assembler instructions at just those critical places in C code where
the speed and low-level control of assembler would result in a
dramatic improvement in performance. While you're at it, you
might as well wish away the traditional complexities of
interfacing assembler with C. Better still, you'd like to be able to
do all this without changing any other C code one bit, so that
already-working C code won't have to be altered.
The high-performance code
in Turbo CS libraries Is written
In inline assembly.
Turbo C fulfills every item on your wish list with inline assembly.
Inline assembly is nothing less than the ability to place virtually
any assembler code anywhere in your C programs, with full
access to C constants, variables, and even functions. In truth,
inline assembly is good for more than just fine-tuning, since it's
very nearly as powerful as programming strictly in assembler.
Inline assembly lets you use just as much or as little assembler in
your C programs as you'd like, without having to worry about
the details of mixing the two.
Consider the following C code, which is an example of inline
assembly:
i = 0;
asm dec WORD PTR i;
itt;
°
1* set i to (in C) *1
1* decrement i (in assembler) */
1* increment i (in C) *1
The first and last lines look normal enough, but what is that
middle line? As you've probably guessed, the line starting with
a5m is inline assembly code. If you were to use a debugger to look
at the executable code this C source compiles to, you would find
mov WORD PTR [bp-02J,OOOO
dec WORD PTR [bp-02J
inc WORD PTR [bp-02J
with the inline assembly DEC instruction nestled between the
compiled code for
i
=
0;
and
258
Turbo Assembler User's Guide
itt;
There are a few limitations on
what Inline assembler code Is
allowed to do; see the
section ·Umitat/ons of inline
assembly" on page
274.
Basically, each time the Turbo C compiler encounters the 8sm
keyword that indicates inline assembly, it drops the associated
assembler line directly into the compiled code with only one
change: References to C variables are transformed into the
appropriate assembler equivalent, just as the reference to i in the
preceding example was changed to WORD PTR [BP-02j. In short, the
asm keyword lets you insert virtually any assembler code
anywhere in your C code.
The ability to drop assembler code directly into the code Turbo C
generates might sound a bit dangerous, and, in truth, inline
assembly does have its risks. While Turbo C takes care to compile
its code so as to avoid many potentially hazardous interactions
with inline assembly, there's no doubt that ill-behaved inline
assembly code can cause serious bugs.
On the other hand, any poorly written assembler code, whether
it's inline or in a separate module, has the potential to run amuck;
that's the price to be paid for the speed and low-level control of
assembly language. Besides, bugs are far less common in inline
assembly code than in pure assembler code, since Turbo C attends
to many programming details, such as entering and exiting
functions, passing parameters, and allocating variables. All in all,
the ability to easily fine-tune and turbo-charge portions of your C
code with inline assembly is well worth the trouble of having to
iron out the occasional assembler bug.
Here are some Important notes about Inline assembly:
1. You must invoke TCC.EXE, the command-line version of
Turbo C, in order to use inline assembly. TC.EXE, the userinterface version of Turbo C, does not support inline assembly.
2. It's very possible that the version ofTLINK that came with
your copy of Turbo Assembler is not the same version that
came with your copy of Turbo C. Since important
enhancements were made to TLINK in order to support Turbo
Assembler, and since further enhancements will no doubt be
made, it is important that you link Turbo C modules
containing inline assembly with the most recent version of
TLINK that you have. The safest way to accomplish this is to
make sure that there's only one TLINK.EXE file on the disk
you use to run the linker; that TLINK.EXE file should have the
latest version number of all the TLINK.EXE files you've
received with other Borland products.
Chapter 7, Interfacing Turbo Assembler with 'Turbo C
259
How inline
assembly works
Normally, Turbo C compiles each file of C source code directly to
an object file, then invokes TLINK to tie the object files together
into an executable program. Figure 7.1 shows such a compileand -link cycle. To start this cycle, you enter the command line
tee filename
which instructs Turbo C to first compile FILENAME.C to
FILENAME.OB] and then invoke TLINK to link FILENAME.OB]
into FILENAME.EXE.
Figure 7.1
Turbo C compile
and link cycle
C Source File
FILENAME.C
Compile
Object File
FILENAME.OBJ
Executable File
FILENAME.EXE
When inline assembly is used, however, Turbo C automatically
adds one extra step to the compile-and-link sequence.
260
Turbo Assembler User's Guide
Turbo C handles each module containing inline assembly code by
first compiling the whole module to an assembly language source
file, then invoking Turbo Assembler to assemble the resulting
assembler code to an object file, and finally invoking TLINK to
link the object files together. Figure 7.2 illustrates this process,
showing how Turbo C produces an executable file from a C
source file containing inline assembly code. You start this cycle
with the command line
tee -B filename
which instructs Turbo C to first compile FILENAME.ASM, then
invoke Turbo Assembler to assemble FILENAME.ASM to
FILENAME.OBJ, and finally invoke TLINK to link
FILENAME.OBJ into FILENAME.EXE.
Inline assembly code is simply passed along by Turbo C to the
assembly language file. The beauty of this system is that Turbo C
need not understand anything about assembling the inline code;
instead, Turbo C compiles C code to the same level-assembler
code-as the inline assembly code and lets Turbo Assembler do
the assembling.
To see exactly how Turbo C handles inline assembly, enter the
following program under the name PLUSONE.C (or load it from
the example disk):
'include
int main(void)
int TestValue;
scanf("%d",&TestValue);
asm inc WORD PTR TestValue;
printf("%d",TestValue);
/* get the value to increment */
/* increment it (in assembler) */
/* print the incremented value */
and compile it with the command line
tcc -5 plusone
Chapter 7, Interfacing Turbo Assembler with Turbo C
261
Figure 7.2
Turbo C compile.
assembly. and link
cycle
C Source File
FILENAME.C
Compile
Assembler Source File
FILENAME.ASM
Assemble
Executable File
FILENAME.EXE
The -S option instructs Turbo C to compile to assembler code and
then stop, so the file PLUSONE.ASM should now be on your disk.
In PLUSONE.ASM you should find
This code should give you a
strong appreciation for all
the work Turbo C saves you
by supporting inline
assembly.
262
ifndef
?debug macro
ENDM
ENDIF
name
TEXT SEGMENT
DGROUP GROUP
ASSUME
TEXT ENDS
DATA SEGMENT
d@
LABEL
??version
Plus one
BYTE PUBLIC 'CODE'
_DATA,_BSS
cs:_TEXT,ds:DGROUP,ss:DGROUP
WORD PUBLIC 'DATA'
BYTE
Turbo Assembler User's Guide
d@w
DATA
BSS
b@
b@w
-
BSS
TEXT
main
Here:S the assembler code
for the scant call, followed by
the inline assembler
instruction to increment
TestValue, followed by the
assembler code for the prlnff
code.
LABEL WORD
ENDS
SEGMENT WORD PUBLIC 'BSS'
LABEL BYTE
LABEL WORD
?debug C E90156E11009706C75736F6E652E63
?debug C E90009B9100F696E636C7564655C737464696F2E68
?debug C E90009B91010696E636C7564655C7374646172672E68
ENDS
SEGMENT BYTE PUBLIC 'CODE'
?debug L 3
PROC
NEAR
push
bp
bp,sp
mov
sp
dec
sp
dec
?debug L 8
ax,WORD PTR [bp-2]
lea
push
ax
ax,OFFSET DGROUP:_s@
mov
push
ax
call
NEAR PTR - scanf
pop
ex
pop
ex
?debug L 9
inc
WORD PTR [bp-2]
?debug L 10
push
WORD PTR [bp-2]
ax,OFFSET DGROUP: - s@+3
mov
push
ax
NEAR PTR _printf
call
pop
ex
pop
ex
@1:
main
TEXT
DATA
- s@
Turbo C automatically
translates the C variable
TestValue to the equivalent
assembler addressing of that
variable, (BP-2).
DATA
?debug
mov
pop
ret
ENDP
ENDS
SEGMENT
LABEL
DB
DB
DB
DB
DB
DB
ENDS
L 12
sp,bp
bp
WORD PUBLIC 'DATA'
BYTE
37
100
0
37
100
0
Chapter 7, Interfacing Turbo Assembler with Turbo C
263
TEXT
TEXT
SEGMENT
EXTRN
EXTRN
ENDS
PUBLIC
END
BYTE PUBLIC 'CODE'
_printf:NEAR
scanf:NEAR
main
Turbo C compiled the scant call to assembly language, dropped
the inline assembly code directly into the assembler output file,
and then compiled the prlntf call to assembler. The resulting file is
a valid assembler source file, ready to be assembled with Turbo
Assembler.
Had you not used the -S option, Turbo C would have proceeded
to invoke Turbo Assembler to assemble PLUSONE.ASM and
would then have invoked TLINK to link the resultant object file,
PLUSONE.OBJ, into the executable file PLUSONE.EXE. This is the
normal mode of operation of Turbo C with inline assembler; we
used -S for explanatory purposes only, so that we could examine
the intermediate assembly language step Turbo C uses when
supporting inline assembly. The -S option is not particularly
useful when compiling code to be linked into executable
programs, but provides a handy means by which to examine both
the instructions surrounding your inline assembly code and the
code generated by Turbo C in general. If you're ever uncertain
about exactly what code you're generating with inline assembly,
just examine the .ASM file produced with the -S option.
How Turbo C knows to
use inline assembly
mode
Normally, Turbo C compiles C code directly to object code. There
are several ways to tell Turbo C to support inline assembly by
compiling to assembly language and then invoking Turbo
Assembler.
The -8 command-line option instructs Turbo C to generate object
files by way of compiling to assembler code, then invoking Turbo
Assembler to assemble that code.
The -S command-line option instructs Turbo C to compile to
assembler code, and then stop. The .ASM file generated by Turbo
C when the -S option is specified can then be separately
assembled and linked to other C and assembler modules. Except
when debugging or simply exploring, there's generally no reason
to use -S in preference to -8.
The #pragma directive
Ipragma inline
264
Turbo Assembler User's Guide
has the same effect as the -B command-line option, instructing
Turbo C to compile to assembly and then invoke Turbo
Assembler to assemble the result. When Turbo C encounters
#pragma Inllne, compilation restarts in assembler output mode.
Consequently, it's best to place the #pragma Inllne directive as
close to the start of the C source code as possible, since any C
source code preceding #pragma Inllne will be compiled twice,
once in normal C-to-object mode and again in C-to-assembler
mode. While this doesn't hurt anything, it does waste time.
Finally, if Turbo C encounters inline assembly code in the absence
of -B, -S, and #pragma Inllne, the compi}er issues a warning like
Warning test.c 6: Restarting compile using assembly in function main
and then restarts compilation in assembler-output mode, just as if
a #pragma inline directive had been encountered at that point.
Make it a point to avoid this warning by using the -B option or
#pragma Inllne, since restarting compilation on encountering
inline assembly makes for relatively slow compiles.
Invoking Turbo
Assembler for inline
assembly
In order for Turbo C to be able to invoke Turbo Assembler, Turbo
C must first be able to find Turbo Assembler. Exactly how this
happens varies with different versions of Turbo C.
Versions of Turbo C later than 1.5 expect to find Turbo Assembler
under the file name TASM.EXE in either the current directory or
one of the directories pointed to by the DOS PATH environment
variable. Basically, Turbo C can invoke Turbo Assembler under
the same circumstances in which you could type the command
TASM
and run Turbo Assembler from the command-line prompt. So, if
you have Turbo Assembler in the current directory or anywhere
in your command search path, Turbo C will automatically find it
and run it to perform inline assembly.
See the README file on the
distribution disk for
information about how to
patch those versions of Tee.
Versions 1.0 and 1.5 of Turbo C behave a little differently. Since
these versions of Turbo C were written before Turbo Assembler
existed, they invoke MASM, the Microsoft Macro Assembler, to
perform inline assembly. Consequently, these versions of Turbo C
search the current directory and the command search path for the
file MASM.EXE, rather than the file TASM.EXE, and so do not
automatically use Turbo Assembler.
Chapter 7, Interfacing Turbo Assembler with Turbo C
265
Where Turbo C
assembles inline
assembly
Inline assembly code can end up in either Turbo C's code segment
or Turbo C's data segment. Inline assembly code located within a
function is assembled into Turbo C's code segment, while inline
assembly code located outside a function is assembled into Turbo
C's data segment.
For example, the C code
1* Table of square values *1
asm SquareLookUpTable label word;
asm dw 0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100;
1* Function to look up the square of a value between a and 10 *1
int LookUpSquare(int Value)
(
asm mov bx,Value;
asm shl bx,l;
1* get the value to square *1
1* mUltiply it by 2 to look up in
a table of word-sized elements *1
asm mov ax, [SquareLookUpTable+bx);
1* look up the square *1
return(_AX);
1* return the result *1
puts the data for SquareLookUpTable in Turbo C's data segment
and the inline assembly code inside LookUpSquare in Turbo C's
code segment. The data could equally well be placed in the code
segment; consider the following version of LookUpSquare, where
SquareLookUpTable is in Turbo C's code segment:
1* Function to look up the square of a value between a and 10 *1
int LookUpSquare(int Value)
(
asm jmp SkipAroundData
1* jump past the data table *1
1* Table of square values *1
asm SquareLookUpTable label word;
asm dw 0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100;
SkipAroundData:
asm mov bx,Value;
asm shl bx,l;
1* get the value to square *1
1* mUltiply it by 2 to look up
in a table of word-sized elements *1
asm mov ax, [SquareLookUpTable+bx);
1* look up the square *1
return(_AX);
1* return the result *1
Since SquareLookUpTable is in Turbo C's code segment, it would
seem that a CS: segment override prefix should be required in
order to read from it. In fact, this code automatically assembles
266
Turbo Assembler User's Guide
with a CS: prefix on the access to SquareLookUpTable; Turbo C
generates the correct assembler code to let Turbo Assembler know
which segment SquareLookUpTable is in, and Turbo Assembler
then generates segment override prefixes as needed.
Use the -1 switch for
80186/80286
instructions
If you want to use assembler instructions unique to the 80186
processor, such as
shr ax,3
and
push 1
it's easiest to use the -1 command-line option to Turbo C, as in
this example,
tee -1 -B heapmgr
.where HEAPMGR.C is a program that contains inline assembly
instructions unique to the 80186.
The primary purpose of the -1 option is to instruct Turbo C to
take advantage of the full 80186 instruction set when compiling,
but the -1 option also causes Turbo C to insert the .186 directive at
the start of the output assembler file; this instructs Turbo
Assembler to assemble the full 80186 instruction set. Without the
.186 directive, Turbo Assembler will flag inline assembly
instructions unique to the 80186 as errors. If you want to assemble
80186 instructions without having Turbo C use the full 80186
instruction set, just insert the line
asm .186;
at the start of each Turbo C module containing inline 80186
instructions. This line will be passed through to the assembler file,
where it will instruct Turbo Assembler to assemble 80186
instructions.
While Turbo C provides no built-in support for 80386, 80287, and
80387 processors, inline assembly that supports the 80286, 80287,
80386, and 80387 can be enabled in a similar manner, with the
a5m keyword and the .286, .286C, .286P, .386, .386C, .386C, .287,
and .387 Turbo Assembler directives.
The line
asm .186;
Chapter 7, Interfacing Turbo Assembler with Turbo C
267
illustrates an important point about inline assembly: Any valid
assembler line can be passed to the assembler file by use of the
8sm prefix, including segment directives, equates, macros, and so
on.
The format of
inline assembly
statements
Inline assembly statements are much like normal assembler lines,
but there are a few differences. The format of an inline assembly
statement is
asrn
[] <; or newline>
where
See "Memory and address
operand limitations" on
page 274 for Important
Information regarding label.
• The 8sm keyword must start every inline assembly statement.
• [] is a valid assembler label. The square brackets indicate
that label is optional, just as it is in assembler.
• is any valid assembler instruction or
directive.
• contains the operand{s) acceptable to the instruction
or directive; it can also reference C constants, variables, and
labels within the limitations described in the section
''Limitations of inline assembly" on page 274.
• <; or newline> is a semicolon or a newline, either of which
signals the end of the 8sm statement.
Semicolons in inline
assembly
One aspect of inline assembly that no C purist could miss is that,
alone among C statements, inline assembly statements do not
require a terminating semicolon. A semicolon can be used to
terminate each statement, but the end of the line will do just as
well. So, unless you're planning to put multiple inline assembly
statements on each line (which is not a good practice from the
perspective of clarity), semicolons are purely optional. While this
may not seem to be in the spirit of C, it is in keeping with the
convention adopted by several UNIX-based compilers.
Comments in in line
assembly
The previous description of the format of an inline assembly
statement lacks one key element-a comment field. While
semicolons can be placed at the end of inline assembly statements,
semicolons do not begin comment fields in inline assembly code.
How, then, are you to comment your inline assembly code?
Strangely enough, with C comments. Actually, that's not strange
268
Turbo Assembler User's Guide
a t all, for the C preprocessor processes inline assembly code along
with the rest of your C code. This has the advantage of allowing
you to use a uniform commenting style throughout your C
programs containing inline assembly, and also makes it possible
to use C-defined symbolic names in both C and inline assembly
code. For example, in
idefine CONSTANT 51
int i;
i = CONSTANT;
/* set i to constant value */
asm sub WORD PTR i,CONSTANT; /* subtract const value from i */
both C and inline assembly code use the C-defined symbol
CONSTANT, and i winds up equal to O.
The last example illustrates one wonderful feature of inline
assembly, which is that the operand field might contain direct
references not only to C-defined symbolic names but also to C
variables. As you will see later in this chapter, accessing C
variables in assembler is normally a messy task, and convenient
reference to C variables is a primary reason why inline assembler
is the preferred way to integrate assembler and C for most
applications.
Accessing structure/
union elements
Inline assembly code can directly reference structure elements.
For example,
struct Student
char Teacher[30];
int Grade;
JohnQPublic;
asm mov ax, JohnQPublic.Grade;
loads AX with the contents of member Grade of the Student type
structure JohnQPublic.
Inline assembly code can also access structure elements addressed
relative to a base or index register. For instance,
asm mov bx,OFFSET JohnQPublic;
asm mov ax, [bx] .Grade;
Chapter 7, Interfacing Turbo Assembler with Turbo C
269
also loads AX. with member Grade of /ohnQPublic. Since Grade is at
offset 30 in the Student structure, the last example actually
becomes
asm mov bx,OFFSET JohnQPublic;
asm mov ax, [bx]+30
The ability to access structure elements relative to a pointer
register is very powerful, since it allows inline assembly code to
handle arrays of structures and passed pointers to structures.
If, however, two or more structures that you're accessing with
inline assembly code have the same member name, you must
insert the following:
asm mov bx,[di]. (struct tm) tm hour> alt
For example,
struct Student
char Teacher[30];
int Grade;
JohnQPublic;
struct Teacher
int Grade;
long Income;
);
asm mov ax, JohnQPublic. (struct Student) Grade
An example of
inline assembly
270
So far, you've seen a variety of code fragments that use inline
assembly, but no real working inline assembly programs. This
section remedies that situation by presenting a program that
employs inline assembly to greatly speed the process of
converting text to uppercase. The code presented in this section
serves both as an example of what inline assembly can do and as a
template to which you can refer to as you develop your own
inline assembly code.
Turbo Assembler User's Guide
Take a moment to examine the programming problem to be
solved by the sample program. We'd like to develop a function,
named StringToUpper, that copies one string to another string,
converting all lowercase characters to uppercase in the process.
We'd also like to have this function work equally well with all
strings in all memory models. One good way to do this is to have
far string pointers passed to the function, since pohi.ters to near
strings can always be cast to pointers to far strings, but the reverse
is not always true.
Unfortunately, we run into a performance issue here. ,While Turbo
C handles far pointers perfectly well, far pointer-handling in
Turbo C is much slower than near pointer-handling. This isn't a
shortcoming of Turbo C, but rather an unavoidable effect when
programming the 8086 in a high-level language.
On the other hand, string and far pointer-handling is one area in
which assembler excels. The logical solution, then, is to use inline
assembly to handle the far pointers and string copying, while
letting Turbo C take care of everything else. The following
program, STRING UP .C, does exactly that:
/* Program to demonstrate the use of StringToUpper(). It calls
StringToUpper to convert TestString to uppercase in UpperCaseString, then prints UpperCaseString and its length. */
Ipragma inl1ne
linclude
/* Function prototype for StringToUpper() */
extern unsigned int StringToUpper(
unsigned char far * DestFarString,
unsigned char far * SourceFarString);
Idefine MAX- STRING- LENGTH 100
char *TestString = "This Started Out As Lowercase!";
char UpperCaseString[MAx_STRING_LENGTH];
main ()
(
unsigned int StringLength;
/* Copy an uppercase version of TestString
to UpperCaseString */
StringLength = StringToUpper{UpperCaseString, TestString);
/* Display the results of the conversion */
printf("Original string:\n%s\n\n", TestString);
printf ("Uppercase string: \n%'s \n \n", UpperCaseString);
printf("Number of characters: %d\n\n", StringLength);
Chapter 7, Interfacing Turbo Assembler with Turbo C
271
/* Function to perform high-speed translation to uppercase from
one far string to another
Input:
DestFarString
- array in which to store uppercased
string (will be zero-terminated)
SourceFarString - string containing characters to be
converted to all uppercase (must be
zero-terminated)
Returns:
The length of the source string in characters, not
counting the terminating zero. */
unsigned int StringToUpper(unsigned char far * DestFarString,
unsigned char far * SourceFarString)
unsigned int CharacterCount;
'define LOWER CASE A 'a'
'define LOWER CASE Z 'z'
asm ADJUST VALUE EQU 20h;
asm cld;
asm push ds;
asm Ids si,SourceFarString;
asm les di,DestFarString;
CharacterCount = 0;
StringToUpperLoop:
asm lodsb;
asm cmp aI, LOWER_CASE_A;
asm jb SaveCharacter;
asm cmp aI, LOWER_CASE_Z;
asm ja SaveCharacter;
asm sub aI, ADJUST_VALUE;
SaveCharacter:
asm stosb;
CharacterCount++;
asm and al,al;
asm jnz StringToUpperLoop;
CharacterCount--;
asm pop ds;
return(CharacterCount);
272
/* amount to subtract from
lowercase letters to make
them uppercase */
/* save C's data segment */
/* load far pointer to
source string */
/* load far pointer to
destination string */
/* count of characters */
/* get the next character */
/* if < a then it's not a
lowercase letter */
/* if > z then it's not a
lowercase letter */
/* it's lowercase; make it
uppercase */
/* save the character
/* count this character
/* is this the ending O?
/* no, process the next,
char, if any
/* don't count the terminating 0
/* restore C's data segment
*/
*/
*/
*/
*/
*/
Turbo Assembler User's Guide
When run, STRINGUP.C displays the output
Original string:
This Started Out As Lowercase!
Uppercase string:
THIS STARTED OUT AS LOWERCASE!
Number of characters: 30
demonstrating that it does indeed convert all lowercase letters to
uppercase.
The heart of S1RINGUP.C is the function StringToUpper, which
performs the entire process of string copying and conversion to
uppercase. StringToUpper is written in both C and inline assembly,
and accepts two far pointers as parameters. One far pointer points
to a string containing text; the other far pointer points to another
string, to which the text in the first string is to be copied with all
lowercase letters converted to uppercase. The function declaration
and parameter definition are all handled in C, and, indeed, a
function prototype for StringToUpper appears at the start of the
program. The main program calls StringToUpper just as if it were
written in pure C. In short, all the ad vantages of programming in
Turbo C are available, even though StringToUpper contains inline
assembly code.
The body of StringToUpper is written in a mixture of C and inline
assembly. Assembler is used to read each character from the
source string, to check and, if need be, translate the character to
upperca~e, and to write the character to the destination string.
Inline assembly allows StringToUpper to use the powerful LOOSB
and STOSB string instructions to read and write the characters.
In writing StringToUpper, we knew that we wouldn't need to
access any data in Turbo C's data segment, so we simply pushed
DS at the start of the function, then set DS to point to the source
string and left it there for the rest of the function. One great
advantage that inline assembly has over a pure C implementation
is this ability to load the far pointers once at the start of the
function and then never reload them until the function is done. By
contrast, Turbo C and other high-level languages generally reload
far pointers every time they are used. The ability to load far
pointers just once means that StringToUpper processes far strings
as rapidly as if they were near strings.
Chapter 7, Interfacing Turbo Assembler with Turbo C
273
One other interesting point about StringToUpper is the way in
which C and assembler statements are mixed. #define is used to
set LOWER_CASE_A and LOWER_CASE_Z, while the assembler
EQU directive is used to set ADJUST_VALUE, but all three
symbols are used in the same fashion by the inline assembly code.
Substitution for the C-defined_symbols is done by the Turbo C
preproces~or, while substitution for ADJUST_VALUE is done by
Turbo Assembler, but both can be used by inline assembly code.
C statements to manipulate CharacterCount are sprinkled
throughout StringToUpper. This was done only to illustrate that C
code and inline assembly code can be intermixed. CharacterCount
could just as easily have been maintained directly by inline
assembly code in a free register, such as CX or DX; StringToUpper
would then have run faster.
Freely intermixing C code and inline assembly code carries risks if
you don't understand exactly what code Turbo C generates in
between your inline assembly statements. Using the Turbo C's-S
compiler option is the best way to explore what happens when
you mix inline assembly and C code. For instance, you can learn
exactly how the C and inline assembly code in StringToUpper fit
together by compiling STRINGUP.C with the -S option and
examining the output file STRINGUP.ASM.
STRINGUP.C vividly demonstrates the excellent payback that
judicious use ofinline assembly provides. In StringToUpper, the
insertion of just 15 inline assembly statements approximately
doubles string-handling speed over equivalent C code.
Limitations of
inline assembly
Memory and address
operand limitations
274
There are very few limitations as to how inline assembly might be
used; by and large, inline assembly statements are simply passed
through to Turbo Assembler unchanged. There are, however,
notable limitations involving certain memory and address
operands, and a few other restrictions concerning register usage
rules and the lack of default sizing of automatic C variables used
in inline assembly.
The only alterations Turbo C makes to inline assembly statements
is to convert memory and memory address references, such as
variable names and jump destinations, from their C
representations to the assembler equivalents. These alterations
introduce two limitations: Inline assembly jump instructions can
Turbo Assembler User's Guide
only reference C labels, while inline assembly non-jump
instructions can reference anything but C labels. For example,
...
asm jz NoDec;
asm dec cx;
NoDec:
is fine, but
...
asm jnz NoDec;
asm dec cx;
asm NoDec:
will not compile properly. Similarly, inline assembly jumps
cannot have ~nction names as operands. Inline assembly
instructions other than jumps can have any operands except C
labels. For example,
asm BaseValue DB '0';
asm mov al,BYTE PTR BaseValue;
compiles, but
BaseValue:
asm DB '0';
asm mov al,BYTE PTR BaseValue;
does not compile. Note that a call is not considered a jump, so
valid operands to inline assembly calls include C function names
and assembler labels, but not C labels. If a C function name is
referenced in inline assembly code, it must be prefixed with an
underscore; see the section "Underscores" on page 290 for details.
Chapter 7. Interfacing Turbo Assembler with Turbo C
275
Lack of default
automatic variable
sizing in inline assembly
When Turbo C replaces a reference to an automatic variable in an
inline assembly statement with an operand like [BP-02], it does
not place a size operator, such as WORD PTR or BYTE PTR, into
the altered statement. This means that
int i;
asm mov ax,i;
is output to the assembler file as
mov ax, [bp-02]
In this case, there's no problem, since the use of AX tells Turbo
Assembler that this is a 16-bit memory reference. Moreover, the
lack of a size operator gives you complete flexibility in controlling
operand size in inline assembly. However, consider
int i;
asm mov i,O;
asm inc i;
which becomes
mov
inc
[bp-02],O
[bp-02]
Neither of these instructions has an inherent size, so Turbo
Assembler can't assemble them. Consequently, when you refer to
an automatic variable in Turbo Assembler without a register as
either the source or the destination, be sure to use a size operator.
The last example works just fine as
int i;
asm mov WORD PTR i,O;
asm inc BYTE PTR i;
276
Turbo Assembler User's Guide
The need to preseNe
registers
At the end of any inline assembly code you write, the following
registers must contain the same values as they did at the start of
the inline code: BP, SP, es, OS, and 55. Failure to observe this rule
can result in frequent program crashes and system reboots. AX,
BX, ex, DX, 51, DI, ES, and the flags may be freely altered by
inline code.
Preserving calling functions and register variables
Turbo e requires that 51 and DI, which are used as register
variables, not be destroyed by function calls. Happily, you don't
have to worry about explicitly preserving 51 or DI if you use them
in inline assembly code. If Turbo e detects any use of those
registers in inline assembly, it preserves them at the start of the
function and restores them at the end-yet another of the
conveniences of using inline assembly.
Suppressing internal register variables
Since register variables are stored in 51 and DI, there would seem
to be the potential for conflict between register variables in a
given module and 4tline assembly code that uses 51 or DI in that
same module. Again, though, Turbo e anticipates this problem;
any use of 51 or DI in inline code will disable the use of that
register to store register variables.
Turbo eversion 1.0 did not guarantee avoidance of conflict
between register variables and inline assembly code. If you are
using version 1.0, you should either explicitly preserve 51 and DI
before using them in inline code or update to the latest version of
the compiler.
Disadvantages of
inline assembly
versus pure C
We've spent a good bit of time exploring how inline assembly
works and learning about the potential benefits of inline
assembly. Wh~e inline assembly is a splendid feature for many
applications, it does have certain disadvantages. Let's review
those disadvantages, so you can make informed decisions about
when to use inline assembly in your programs.
Chapter 7, Interfacing Turbo Assembler with Turbo C
277
Reduced portability
and maintainability
The very thing that makes inline assembly code so effective-the
ability to program the 8086 processor directly-also detracts from
a primary strength ofC, portability. If you use inline assembly, it's
a pretty safe bet that you won't be able to port your code to
another processor or C compiler without changes.
Similarly, inline assembly code lacks the clear and concise
formatting C provides, and is often unstructured as well.
Consequently, inline assembly code is generally more difficult to
read and maintain than C code.
When you use inline assembly code, it's a good practice to isolate
the inline code in self-contained modules, and to structure the
inline code carefully with plenty of comments. That way, it's easy
to maintain the code, and it's a relatively simple matter to find the
inline assembly code and rewrite it in C if you need to port the
program to a different environment.
Slower compilation
Available with Tee only
Optimization loss
Compilation of C modules containing inline assembly code is
considerably slower than compilation of pure C code, primarily
because inline assembly code must effectively be compiled twice,
first by Turbo C and then again by Turbo Assembler. If Turbo C
has to restart compilation because neither the -B option, the-S
option, nor #pragma Inllne was used, compilation time for inline
assembly becomes longer still. Fortunately, slow compilation of
modules containing inline assembly is less of a problem now than
it was in the past, since Turbo Assembler is so much faster than
earlier assemblers.
As we mentioned earlier, the inline assembly feature is unique to
TCC.EXE, the command-line version of Turbo C. TC.EXE, the
integrated development environment version of Turbo C, does
not support inline assembly.
When inline assembly is used, Turbo C loses some control over
the code of your programs, since you can directly insert any
assembler statements into any C code. To some extent, you, as the
inline assembly programmer, must compensate for this, by
avoiding certain disruptive actions, such as failing to preserve the
DS register or writing to the wrong area of memory.
On the other hand, Turbo C doesn't require you to follow all its
internal rules when you program in inline assembler; if it did,
278
Turbo Assembler User's Guide
you'd scarcely be better off using inline assembly than if you
programmed in C and let Turbo C generate the code. What Turbo
C does do is turn off some of its optimizations in functions
containing inline assembly statements, thereby allowing you a
relatively free hand in coding inline assembly. For example, some
portions of the jump optimizer are turned off when inline
assembly is used, and register variables are disabled if the inline
code uses SI and DI. This partial loss of optimization is worth
considering, given that you are presumably using inline assembly
in order to boost code quality to its maximum.
If you are greatly concerned about producing the fastest or most
compact code with inline assembly, you might want to write your
functions that contain inline assembly code entirely in inline
assembly-that is, don't mix C and inline assembly code within
the same function. That way, you have control of the code in the
inline assembly functions, Turbo C has control of the code in the
C functions, and both you and Turbo C are free to generate the
best possible code without restrictions.
Error trace-back
limitations
Since Turbo C does little error-checking of inline assembly
statements, errors in inline assembly code are often detected by
Turbo Assembler, not Turbo C. Unfortunately, it can sometimes
be difficult to relate the error messages produced by Turbo
Assembler back to the original C source code, since the error
messages and the line numbers they display are based on the
.ASM file output by Turbo C and not the C code itself.
For example, in the course of compiling TEST.C, a C program
containing inline assembly code, Turbo Assembler might
complain about an incorrectly sized operand on line 23;
unfortunately, "23" refers to the number of the error-producing
line in TEST.ASM, the intermediate assembler file Turbo C
generated for Turbo Assembler to assemble. You're on your own
when it comes to figuring out what line in TEST.C is ultimately
responsible for the error.
Your best bet in a case like this is to first locate the line causing the
error in the intermediate .ASM file, which is left on the disk by
Turbo C whenever Turbo Assembler reports assembly errors. The
.ASM file contains special comments that identify the line in the C
source file from which each block of assembler statements was
generated; for example, the assembler lines following
; ?debug L 15
Chapter 7, Interfacing Turbo Assembler with Turbo C
279
were generated from line 15 of the C source file. Once you've
located the line that caused the error in the .ASM file, you can
then use the line-number comments to map the error-generating
line back to the C source file.
Debugging limitations
Versions of Turbo C up to and including version 1.5 can't
generate source-level debugging infonnation (infonnation
required to let you see C source code as you debug) for modules
containing inline assembly code. When inline assembly is used,
Turbo C versions 1.5 and earlier generate plain assembler code
with no embedded debugging infonnation. Source-level
debugging capabilities are lost, and only assembler-level
debugging of C modules containing inline code is possible.
Later versions of Turbo C take advantage of special Turbo
Assembler features to provide state-of-the-art, source-level
debugging when used with Turbo Debugger to debug modules
containing inline assembly code (and pure C modules too, of
course).
Develop in C and
compile the final code
with inline assembly
In light of the disadvantages of inline assembly we've just
discussed, it may seem that in line assembly should be used as
sparingly as possible. Not so. The trick is to use inline assembly at
the right point in the development cycle-at the end.
Most of the disadvantages of inline assembly boil down to a
single problem: Inline assembly can slow down the edit/ compile/
debug cycle considerably. Slower compilation, inability to use the
integrated environment, and difficulty in finding compilation
errors all mean that development of code containing inline
assembly statements will probably be slower than development of
pure C code. Still, the proper use of inline assembly can result in
dramatic improvements in code quality. What to do?
The answer is simple. Initially, develop each program entirely in
C, taking full advantage of the excellent development
environment provided by TC.EXE. When a program reaches full
functionality, with the code debugged and running smoothly,
switch to TCC.EXE and begin to convert critical portions of the
program to inline assembly code. This approach allows you to
develop and debug your overall program efficiently, then isolate
and enhance selected sections of the code when it comes time to
fine-tune the program.
280
Turbo Assembler User's Guide
Calling Turbo Assembler functions from Turbo C
C and assembler have traditionally been mixed by writing
separate modules entirely in C or assembler, compiling the C
modules and assembling the assembler modules, and then linking
the separately compiled modules together. Turbo C modules can
readily be linked with Turbo Assembler modules in this fashion.
Figure 7.3 shows how to do this.
Figure 7.3
Compile, assemble,
and link with Turbo
C, Turbo Assembler,
and TLiNK
C Source File
FILENAM1.C
Assembler Source File
FILENAM2.ASM
Object File
FILENAM1.0BJ
Object File
FILENAM2.0BJ
Executable File
FILENAM1.EXE
The executable file is produced from mixed C and assembler
source files. You start this cycle with
tee filenaml filenam2.asm
This instructs Turbo C to first compile FILENAM1.C to
FILENAM1.0BJ, then invoke Turbo Assembler to assemble
FILENAM2.ASM to FILENAM2.0BJ, and finally invoke TLINK
Chapter 7, Interfacing Turbo Assembler with Turbo C
281
to link FILENAM1.0B] and FILENAM2.0B] into
FILENAM1.EXE.
Separate compilation is very useful for programs that have sizable
amounts of assembler code, since it makes the full power of Turbo
Assembler available and allows you to do your assembly
language programming in a pure assembler environment,
without the 8sm keywords, extra compilation time, and C-related
overhead of inline assembly.
There is a price to be paid for separate compilation: The assembler
programmer must attend to all the details of interfacing C and
assembler code. Where Turbo C handles segment specification,
parameter-passing, reference to C variables, register variable
preservation, and the like for inline assembly, separately compiled
assembler functions must explicitly do all that and more.
There are two major aspects to interfacing Turbo C and Turbo
Assembler. First, the various parts of the C and assembler code
must be linked together properly, and functions and variables in
each part of the code must be made available to the rest of the
code as needed. Second, the assembler code must properly handle
C-style function calls. This includes accessing passed parameters,
returning values, and following the register preservation rules
required of C functions.
Let's start by examining the rules for linking together Turbo C and
Turbo Assembler code.
The framework
In order to link Turbo C and Turbo Assembler modules together,
three things must happen:
• The Turbo Assembler modules must use a Turbo C-compatible
segment-naming scheme.
• The Turbo C and Turbo Assembler modules must share
appropriate function and variable names in a form acceptable
to Turbo C.
• TLINK must be used to combine the modules into an
executable program.
This says nothing about what the Turbo Assembler modules
actually do; at this point, we're only concerned with creating a
framework within which C-compatible Turbo Assembler
functions can be written.
282
Turbo Assembler User's Guide
Memory models and
segments
See -standard segment
directives· in Chapter 5,
page 111, for an introduction
to the simplified segment
directives.
For a given assembler function to be callable from C, that function
must use the same memory model as the C program and must use
a C-compatible code segment. Likewise, in order for data defined
in an assembler module to be accessed by C code (or for C data to
be accessed by assembler code), the assembler code must follow C
data segment-naming conventions.
Memory models and segment-handling can be quite complex to
implement in assembler. Fortunately, Turbo Assembler does
virtually all the work of implementing Turbo C-compatible
memory models and segments for you in the form of the
simplified segment directives.
Simplified segment directives and Turbo C
The DOSSEG directive instructs Turbo Assembler to order
segments according to the Intel segment-ordering conventions,
the same conventions followed by Turbo C (and many other
popular language products, including those from Microsoft).
The .MODEL directive tells Turbo Assembler that segments
created with the simplified segment directives should be
compatible with the selected memory model (tiny, small, compact,
medium, large, or huge), and controls the default type (near or
far) of procedures created with the PROC directive. Memory
models defined with the .MODEL directive are compatible with
the equivalently named Turbo C models.
Finally, the .CODE, .DATA, .DATA, .FARDATA, .FARDATA, and
.CONST simplified segment directives generate Turbo Ccompatible segments.
For example, consider the following Turbo Assembler module,
named OOTOTAL.ASM:
Underscores U prefix many
of the labels in DoTota/
because they are normally
required by Turbo C. For
more detail, see the section
-Underscores· on page 290.
i select Intel-convention segment ordering
•MODEL
. DATA
EXTRN
PUBLIC
_StartingValue
•DATA?
RunningTotal
•CODE
PUBLIC
small
iselect small model (near code and data)
iTC-compatible initialized data segment
_Repetitions:WORD
iexternally defined
_StartingValue
iavailable to other modules
DW 0
iTC-compatible uninitialized data segment
DW?
iTC-compatible code segment
DoTotal
Chapter 7, Interfacing Turbo Assembler with Turbo C
283
DoTotal
mov
mov
mov
TotalLoop:
inc
loop
mov
ret
DoTotal
END
PROC
;function (near-callable in small model)
cx,[_Repetitions]
;f of counts to do
ax, [_StartingValue]
;set initial value
[RunningTotal],ax
[RunningTotal]
TotalLoop
ax, [RunningTotal]
;RunningTotal++
;return final total
ENDP
The assembler procedure _DoTotal is readily callable from a
small-model Turbo C program with the statement
DoTotalO ;
Note that _DoTotal expects some other part of the program to
define the external variable Repetitions. Similarly, the variable
StartingValue is made public, so other portions of the program can
access it. The following Turbo C module, SHOWTOT.C, accesses
public data in DOTOTAL.ASM and provides external data to
OOTOTAL.ASM:
extern int StartingValue;
extern int DoTotal(void);
int Repetitions;
mainO
{
int i;
Repetitions = 10;
StartingValue = 2;
printf("%d\n", DoTotal());
To create the executable program SHOWTOT.EXE from
SHOWTOT.C and OOTOTAL.ASM, enter the command line
tcc showtot dototal.asm
If you wanted to link _DoTotal to a compact-model C program,
you would simply change the .MODEL directive to .MODEL
COMPACT. If you wanted to use a far segment in
OOTOTAL.ASM, you could use the .FARDATA directive.
In short, generating the correct segment ordering, memory model,
and segment names for linking with Turbo C is a snap with the
simplified segment directives.
284
Turbo Assembler User's Guide
Old-style segment directives and Turbo C
Simply put, it's a nuisance interfacing Turbo Assembler code to C
code using the old-style segment directives. For example, if you
replace the simplified segment directives in OOTOTAL.ASM with
old-style segment directives, you get
DGROUP GROUP
DATA SEGMENT
EXTRN
PUBLIC
_StartingValue
DATA ENDS
BSS
SEGMENT
RunningTotal
BSS
ENDS
TEXT SEGMENT
ASSUME
PUBLIC
DoTotal
mov
mov
mov
Total1oop:
inc
loop
mov
ret
DoTotal ENDP
TEXT ENDS
END
For an overview of Turbo C
segment usage, refer to
Chapter 4 of the Turbo C
Programmer's Guide.
_DATA,_BSS
WORD PUBLIC 'DATA'
_Repetitions:WORD
_StartingValue
DW 0
iexternally defined
iavailable to other modules
WORD PUBLIC 'BSS'
OW?
BYTE PUBLIC 'CODE'
cs:_TEXT,ds:DGROUP,ss:DGROUP
DoTotal
PROC
ifunction (near-callable
i in small model)
cx, [_Repetitions]
if of counts to do
ax, [_StartingValue]
[RunningTotal),ax iset initial value
[RunningTotal]
TotalLoop
ax, [RunningTotal]
iRunningTotal++
ireturn final total
The version with old-style segment directives is not only longer,
but also much harder to read and harder to change to match a
different C memory model. When you're interfacing to Turbo C,
there's generally no advantage to using the old-style segment
directives. If you still want to use the old-style segment directives
when interfacing to Turbo C, you'll have to identify the correct
segments for the memory model your C code uses.
The easiest way to determine the appropriate old-style segment
directives for linking with a given Turbo C program is to compile
the main module of the Turbo C program in the desired memory
model with the -S option, which causes Turbo C to generate an
assembler version of the C code. In that C code, you'll find all the
old-style segment directives used by Turbo C; just copy them into
your assembler code. For example, if you enter the command
Chapter 7, Interfacing Turbo Assembler with Turbo C
285
tcc -S showtot.c
the file SHOWTOT.ASM is generated:
ifndef ??version
?debug macro
ENDM
ENDIF
NAME
showtot
TEXT SEGMENT BYTE PUBLIC 'CODE'
DGROUP GROUP _DATA,_BSS
ASSUME cs:_TEXT,ds:DGROUP,ss:DGROUP
TEXT ENDS
DATA SEGMENT WORD PUBLIC 'DATA'
LABEL BYTE
- d@
LABEL WORD
- d@w
DATA ENDS
SEGMENT WORD PUBLIC 'BSS'
- BSS
_b@
LABEL BYTE
_b@w
LABEL WORD
?debug C E91481D5100973686F77746F742E63
ENDS
BSS
TEXT SEGMENT BYTE PUBLIC 'CODE'
?debug L 3
NEAR
main PROC
?debug L 6
mov
WORD PTR DGROUP:_Repetitions,lO
?debug L 7
mov
WORD PTR DGROUP:_StartingValue,2
?debug L 8
NEAR PTR DoTotal
call
push
ax
ax,offset DGROUP:_s@
mov
push
ax
call
NEAR PTR _printf
pop
cx
pop
cx
@1:
?debug L 9
ret
main ENDP
TEXT ENDS
SEGMENT WORD PUBLIC 'BSS'
- BSS
_Repetitions
LABEL WORD
DB
2 dup (?)
?debug C E9
BSS
ENDS
DATA SEGMENT WORD PUBLIC 'DATA'
_s@
LABEL BYTE
286
Turbo Assembler User's Guide
DATA
TEXT
- TEXT
Chapter 9 covers segment
directives in detail.
DB
DB
DB
DB
ENDS
EXTRN
SEGMENT
EXTRN
EXTRN
ENDS
PUBLIC
PUBLIC
END
37
100
10
a
_StartingValue:WORD
BYTE PUBLIC 'CODE'
DoTotal:NEAR
_printf:NEAR
_Repetitions
main
The segment directives for _DATA (the initialized data segment),
_TEXT (the code segment), and _BSS (the uninitialized data
segment), along with the GROUP and ASSUME directives, are in
read y-to-assemble form, so you can use them as is.
Segment defaults: When Is it necessary to load segments?
Under some circumstances, your C-callable assembler functions
might have to load D5 and/or E5 in order to access data. It's also
useful to know the relationships between the settings of the
segment registers on a call from Turbo C, since sometimes
assembler code can take advantage of the equivalence of two
segment registers. Let's take a moment to examine the settings of
the segment registers when an assembler function is called from
Turbo C, the relationships between the segment registers, and the
cases in which an assembler function might need to load one or
more segment registers.
On entry to an assembler function from Turbo C, the C5 and D5
registers have the following settings, depending on the memory
model in use (55 is always used for the stack segment, and E5 is
always used as a scratch segment register):
Table 7.1
Register settings
when Turbo C
enters assembler
Model
cs
os
Tiny
Small
Compact
Medium
Large
Huge
_TEXT
TEXT
-TEXT
fllename_TEXT
filename_TEXT
filename_TEXT
DGROUP
DGROUP
DGROUP
DGROUP
DGROUP
Chapter 7, Interfacing Turbo Assembler with Turbo C
calling_filename_DATA
287
filename is the name of the assembler module, and callingJilename
is the name of the module calling the assembler module.
In the tiny model, _TEXT and DGROUP are the same, so CS equals
OS on entry to functions. Also in the tiny, small, and medium
models, SS equals OS on entry to functions.
So, when is it necessary to load a segment register in a C-callable
assembler function? For starters, you should.never have to (or
want to) directly load the CS or SS registers. CS is automatically
set as needed on far calls, jumps, and returns, and can't be
tampered with otherwise. SS always points to the stack segment,
which should never change during the course of a program
(unless you're writing code that switches stacks, in which case
you had best know exactly what you're doing!).
ES is always available for you to use as you wish. You can use ES
to point at far data, or you can load ES with the destination
segment for a string instruction.
That leaves the OS register. In all Turbo C models other than the
huge model, OS points to the static data segment (DGROUP) on
entry to functions, and that's generally where you'll want to leave
it. You can always use ES to access far data, although you may
find it desirable to instead temporarily point OS to far data that
you're going to access intensively, thereby saving many segment
override instructions in your code. For example, you could access
a far segment in either of the following ways:
.FARDATA
Counter DW
0
•CODE
PUBLIC AsmFunction
PROC
AsmFunction
mov
mov
inc
AsmFunction
ax,@fardata
es,ax
es: [Counter]
ipoint ES to far data segment
iincrement counter variable
ENDP
or
288
Turbo Assembler User's Guide
.FARDATA
Counter DW
0
.CODE
PUBLIC AsmFunction
AsmFunction
PROC
ASSUME ds:@fardata
mov
ax,@fardata
mov
ds,ax
[Counter]
inc
ASSUME ds:@data
mov
ax,@data
mov
ds,ax
AsmFunction
ipoint DS to far data segment
iincrement counter variable
ipoint DS back to DGROUP
ENDP
The second version has the advantage of not requiring an ES:
override on each memory access to the far data segment. If you do
load OS to point to a far segment, be sure to restore it as in the
preceding example before attempting to access any variables in
DGROUP. Even if you don't access DGROUP in a given assembler
function, be sure to restore OS before exiting, since Turbo C
assumes that functions leave OS unchanged.
Handling OS in C-callable huge model functions is a bit different.
In the huge model, Turbo C doesn't use DGROUP at all. Instead,
each module has its own data segment, which is a far segment
relative to all the other modules in the program; there is no
commonly shared near data segment. On entry to a function in
the huge model, OS should be set to point to that module's far
segment and left there for the remainder of the function, as
follows:
.FARDATA
. CODE
PUBLIC
AsmFunction
push
mov
mov
pop
ret
AsmFunction
PROC
ds
ax,@fardata
ds,ax
ds
Chapter 7, Interfacing Turbo Assembler with Turbo C
289
AsmFunction
ENDP
Note that the original state of DS is preserved with a PUSH on
entry to AsmFunction and restored with a POP before exiting; even
in the huge model, Turbo C requires all functions to preserve DS.
Publics and externals
Turbo Assembler code can call C functions and reference external
C variables, and Turbo C code can likewise call public Turbo
Assembler functions and reference public Turbo Assembler
variables. Once Turbo C-compatible segments are set up in Turbo
Assembler, as described in the preceding sections, only the
following few simple rules need be observed in order to share
functions and variables between Turbo C and Turbo Assembler.
Underscores
Normally, Turbo C expects all external labels to start with an
underscore character C). Turbo C automatically prefixes an
underscore to all function and external variable names when
they're used in C code, so you only need to attend to underscores
in your assembler code. You must be sure that all assembler
references to Turbo C functions and variables begin with
underscores, and you must begin all assembler functions and
variables that are made public and referenced by Turbo C code
with underscores.
For example, the following C code,
extern int ToggleFlag();
int Flag;
main()
{
ToggleFlag () ;
links properly with the following assembler program:
.M:lDEL
•DATA
EXTRN
.CODE
PUBLIC
_ToggleFlag
cmp
jz
mov
290
small
Jlag:WORD
_ToggleFlag
PROC
[Jlag],O
SetFlag
[Jlag],O
iis the flag reset?
iyes, set it
ino, reset it
Turbo Assembler User's Guide
jmp
Labels not referenced by C
code, such'as SetRag, don't
need leading underscores.
short EndToggleFlag
;done
SetFlag:
;set flag
EndToggleFlag:
ret
_ToggleFlag
ENDP
END
When you use the C language specifier in your EXTRN and
PUBLIC directives,
DOSSEG
•MODEL
•DATA
EXTRN
.CODE
PUBLIC
ToggleFlag
cmp
jz
mov
jmp
SetFlag:
mov
EndToggleFlag:
ret
ToggleFlag
END
SMALL
C Flag:word
C ToggleFlag
PROC
[Flag],O
Set Flag
[Flag],O
short EndToggleFlag
[Flag],l
ENDP
Turbo Assembler causes the underscores to be prefixed automatically when Flag and ToggleFlag are published in the object
module.
By the way, it is possible to tell Turbo C not to use underscores by
using the -u- command-line option. But you have to purchase the
run-time library source from Borland and recompile the libraries
with underscores disabled in order to use the -u- option. (See
"Pascal calling conventions" on page 307 for information on the
-p option, which disables the use of underscores and casesensitivity.)
The significance of uppercase and lowercase
Turbo Assembler is normally insensitive to case when handling
symbolic names, making no distinction between uppercase and
lowercase letters. Since C is case-sensitive, it's desirable to have
Turbo Assembler be case-sensitive, at least for those symbols that
Chapter 7, Interfacing Turbo Assembler with Turbo C
291
are shared between assembler and C. Iml and Imx make this
possible.
The Iml command-line switch causes Turbo Assembler to become
case-sensitive for all symbols. The Imx command-line switch
causes Turbo Assembler to become case-sensitive for public
(PUBLIC), external (EXTRN), global (GLOBAL), and communal
(COMM) symbols only.
Label types
While assembler programs are free to access any variable as data
of any size (8 bit, 16 bit, 32 bit, and so on), it is generally a good
idea to access variables in their native size. For instance, it usually
causes problems if you write a word to a byte variable:
SrnallCount DB
°
rnov WORD PTR [SrnallCount],Offffh
Consequently, it's important that your assembler EXTRN
statements that declare external C variables specify the right size
for those variables, since Turbo Assembler has only your
declaratio?1 to go by when deciding what size access to generate to
a C variable. Given the statement
char c
in a C program, the assembler code
EXTRN
c: WORD
inc
[c]
could lead to nasty problems, since every 256th time the
assembler code incremented c, c would turn over. And, since c is
erroneously declared as a word variable, the byte at OFFSET c + 1
would incorrectly be incremented, with unpredictable results.
Correspondence between C and assembler data types is as
follows:
292
Turbo Assembler User's Guide
C Data Type
Assembler Data Type
unsigned char
char
enurn
unsigned short
short
unsigned int
int
unsigned long
long
float
double
long double
near'"
far '"
byte
byte
word
word
word
word
word
dword
dword
dword
qword
tbyte
word
dword
Far externals
If you're using the simplified segment directives, EXTRN
declarations of symbols in far segments must not be placed within
any segment, since Turbo Assembler considers symbols declared
within a given segment to be associated with that segment. This
has its drawbacks: Turbo Assembler cannot check the addressability of symbols declared EXTRN outside any segment, and so
can neither generate segment overrides as needed nor inform you
when you attempt to access that variable when the correct
segment is not loaded. Turbo Assembler still assembles the
correct code for references to such external symbols, but can no
longer provide the normal degree of segment addressability
checking.
If you want to (though we discourage it), you can use the old-
style segment directives to explicitly declare the segment each
external symbol is in and then place the EXTRN directive for that
symbol inside the segment declaration. However, this is a good
bit of work; if you don't mind taking responsibility for making
sure that the correct segment is loaded when you access far data,
it's easiest to just put EXTRN declarations of far symbols outside
all segments. For example, suppose that FILE1.ASM contains
.FARDATA
FilelVariable
DB
0
Then ifFILE1.ASM is linked to FILE2.ASM, which contains
Chapter 7, Interfacing Turbo Assembler with Turbo C
293
•DATA
EXTRN FilelVariable:BYTE
•CODE
Start PROC
mov
ax,SEG FilelVariable
mov
ds,ax
SEG Filel Variable will not return the correct segment. The EXTRN
directive is placed within the scope of the DATA directive of
FILE2.ASM, so Turbo Assembler considers FilelVariable to be in
the near DATA segment of FILE2.ASM, rather than in the
FARDATA segment.
The following code for FILE2.ASM allows SEG Filel Variable to
return the correct segment:
•DATA
@eurseg ENDS
EXTRN
.CODE
Start
PROC
mov
mov
FilelVariable:BYTE
ax,SEG FilelVariable
ds,ax
The trick here is that the @curseg ENDS directive ends the .DATA
segment, so no segment directive is in effect when Filel Variable is
declared external.
Linker .command line
The simplest way to link Turbo C modules with Turbo Assembler
modules is to enter a single Turbo C command line and let Turbo
C do all the work. Given the proper command line, Turbo C will
compile the C code, invoke Turbo Assembler to do the
assembling, and invoke TLINK to link the object files into an
executable file. Suppose, for example, that you have a program
consisting of the C files MAIN.C and STAT.C and the assembler
files SUMM.ASM and DISPLAY.ASM. The command line
tee main stat summ.asm display.asm
compiles MAIN.C and STAT.C, assembles SUMM.ASM and
DISPLAY.ASM, and links all four object files, along with the C
start-up code and any required library functions, into MAIN.EXE.
You only need remember the .ASM extensions when typing your
assembler file names.
294
Turbo Assembler User's Guide
If you use TLINK in stand-alone mode, the object files generated
by Turbo Assembler are standard object modules and are treated
just like C object modules.
Between Turbo
Assembler and
Turbo C
Parameter-passing
Now that you understand how to build and link C-compatible
assembler modules, you need to learn what sort of code you can
put into C-callable assembler functions. There are three areas to
examine here: receiving passed parameters, using registers, and
returning values to the calling code.
Turbo C passes parameters to functions on the stack. Before
calling a function, Turbo C first pushes the parameters to that
function onto the stack, starting with the rightmost parameter and
ending with the leftmost parameter. The C function call
Test (i, j, 1);
compiles to
mov
push
push
push
call
add
ax,l
ax
WORD PTR DGROUP:_j
WORD PTR DGROUP: i
NEAR PTR Test
sp,6
in which you can clearly see the rightmost parameter, 1, being
pushed first, then j, and finally i.
Read about Pascal calling
conventIons on page 307.
Upon return from a function, the parameters that were pushed on
the stack are still there, but are no longer of any use. Consequently, immediately following each function call, Turbo C adjusts the
stack pointer back to the value it contained before the parameters
were pushed, thereby discarding the parameters. In the previous
example, the three parameters of 2 bytes each take up 6 bytes of
stack space altogether, so Turbo C adds 6 to the stack pointer to
discard the parameters after the call to Test. The important point
here is that under C calling conventions, the calling code is
responsible for discarding the parameters from the stack.
Assembler functions can access parameters passed on the stack
relative to the BP register. For example, suppose the function Test
in the previous example is the following assembler function:
Chapter 7, Interfacing Turbo Assembler with Turbo C
295
. MODEL small
•CODE
PUBLIC Test
Test PROC
push
mov
mov
add
sub
pop
ret
bp
bp,sp
ax, [bp+4]
ax, [bp+6]
ax, [bp+8]
bp
jget parameter 1
jadd parameter 2 to parameter 1
;subtract parameter 3 from sum
Test ENDP
END
You can see that Test is getting the parameters passed by the C
code from the stack, relative to BP. (Remember that BP addresses
the stack segment.> But just how are you to know where to find the
parameters relative to BP?
Figure 7.4 shows what the stack looks like just before the first
instruction in Test is executed:
i
j
= 25;
= 4;
Test(i, j,
1) i
Figure 7.4
state of the stack
Just before
executing Test's first
Instruction
SP
.,
Return Address
SP+ 2
25 ( I )
SP+ 4
4 (j)
SP+ 6
1
The parameters to Test are at fixed locations relative to SP,
starting at the stack location 2 bytes higher than the location of the
return address that was pushed by the call. After loading BP with
SP, you can access the parameters relative to BP. However, you
must first preserve BP, since the calling C code expects you to
return with BP unchanged. Pushing BP changes all the offsets on
296
Turbo Assembler User's Guide
the stack. Figure 7.5 shows the stack after these lines of code are
executed:
push bp
mov bp,sp
Figure 7.5
state of the stack
after PUSH and
MOV
SP
SP+ 2
•
Caller's BP
Return Address
•
BP
BP+ 2
SP+ 4
25 ( I)
BP+ 4
SP+ 6
4 (j)
BP+ 6
SP+ 8
1
BP+ 8
This is the standard C stack frame, the organization of a function's
parameters and automatic variables on the stack. As you can see,
no matter how many parameters a C program might have, the
leftmost parameter is always stored at the stack address
immediately above the pushed return address, the next parameter
to the right is stored just above the leftmost parameter, and so on.
As long as you know the order and type of the passed parameters,
you always know where to find them on the stack.
Space for automatic variables can be reserved by subtracting the
required number of bytes from SP. For example, room for a 100byte automatic array could be reserved by starting Test with
push bp
mov bp,sp
sub sp,lOO
as shown in Figure 7.6.
Chapter 7, Interfacing Turbo Assembler with Turbo C
297
Figure 7.6
state of the stack
after PUSH. MOV.
and SUB
--I----BP - 100
SP
SP + 100
SP + 102
~
Caller's BP
Return Address
•
BP
BP+ 2
SP + 104
25 (I)
BP+ 4
SP+ 106
4 (j)
BP+ 6
SP + 108
1
BP+ 8
Since the portion of the stack holding automatic variables is at a
lower address than BP, negative offsets from BP are used to
address automatic variables. For example,
mov BYTE PTR [bp-100],O
would set the first byte of the IOO-byte array you reserved earlier
to zero. Passed parameters, on the other hand, are always
addressed at positive offsets from BP.
While you can, if you wish, allocate space for automatic variables
as shown previously, Turbo Assembler provides a special version
of the LOCAL directive that makes allocation and naming of
automatic variables a snap. When LOCAL is encountered within a
procedure, it is assumed to define automatic variables for that
procedure. For example,
LOCAL
LocalArray:BYTE:l00,LocalCount:WORD = AUTO_SIZE
defines the automatic variables LocalArray and LocalCount.
LocalArray is actually a label equated to [BP-I00], and LocalCount
is actually a label equated to [BP-I02], but you can use them as
variable names without ever needing to know their values.
AUTO_SIZE is the total number of bytes of automatic storage
298
Turbo Assembler User's Guide
required; you must subtract this value from SP in order to allocate
space for the automatic variables.
Here's how you might use LOCAL:
TestSub PROC
LOCAL
LocaIArray:BYTE:100,LocaICount:WORD=AUTO_SIZE
push bp
ipreserve caller's stack frame pointer
mov bp,sp
iset up our own stack frame pointer
sub sp,AUTO_SIZE
;allocate room for automatic variables
mov [LocaICount],10 iset local count variable to 10
i (LocalCount is actually [BP-102])
mov cx, [LocalCount]
mov al,'A'
lea bx, [LocalArray]
FillLoop:
mov
inc
loop
mov
[bx],al
bx
FillLoop
sp,bp
pop bp
ret
TestSub ENDP
;get count from local variable
;we'll fill with character IIAII
ipoint to local array
; (LocalArray is actually [BP-100])
; fill next byte
ipoint to following byte
ido next byte, if any
;deallocate storage for automatic
; variables (add sp,AUTO_SIZE would
; also have worked)
;restore caller's stack frame pointer
In this example, note that the first field after the definition of a
given automatic variable is the data type of the variable: BYTE,
WORD, DWORD, NEAR, and so on. The second field after the
definition of a given automatic variable is the number of elements
of that variable's type to reserve for that variable. This field is
optional and defines an automatic array if used; if it is omitted,
one element of the specified type is reserved. Consequently,
LocalArray consists of 100 byte-sized elements, while LocalCount
consists of 1 word-sized element.
Also note that the LOCAL line in the preceding example ends with
=AUTO_SIZE. This field, beginning with an equal sign, is
optional; if present, it sets the label following the equal sign to the
number of bytes of automatic storage required. You must then use
that label to allocate and deallocate storage for automatic
variables, since the LCCAL directive only generates labels, and
doesn't actually generate any code or data storage. To put this
another way: LOCAL doesn't allocate automatic variables, but
Chaoter 7, Interfacing Turbo Assembler with Turbo C
299
simply generates labels that you can readily use to both allocate
storage for and access automatic variables.
A very handy feature of LOCAL is that the labels for both the
automatic variables and the total automatic variable size are
limited in scope to the procedure they're used in, so you're free to
reuse an automatic variable name in another procedure.
Refer to Chapter 3 in the
Reference Guide for
additional Information about
both forms of the LOCAL
directive.
As you can see, LOCAL makes it much easier to define and use
automatic variables. Note that the LOCAL directive has a
completely different meaning when used in macros, as discussed
in Chapter 9.
By the way, Turbo C handles stack frames in just the way we've
described here. You may well find it instructive to compile a few
Turbo C modules with the -S option and look at the assembler
code Turbo C generates to see how Turbo C creates and uses stack
frames.
So far, so good, but there are further complications. First of all,
this business of accessing parameters at constant offsets from BP
is a nuisance; not only is it easy to make mistakes, but if you add
another parameter, all the other stack frame offsets in the function
must be changed. For example, suppose you change Test to accept
four parameters:
Test (Flag, i, j, 1);
Suddenly i is at offset 6, not offset 4, j is at offset 8, not offset 6,
and so on. You can use equates for the parameter offsets:
Flag
AddParm1
AddParm2
SubParm1
EQU
EQU 6
EQU 8
EQU 10
mov ax, [bp+AddParm1]
add ax, [bp+AddParm2]
sub ax, [bp+SubParm1]
but it's still a nuisance to calculate the offsets and maintain them.
There's a more serious problem, too: The size of the pushed return
address grows by 2 bytes in far code models, as do the sizes of
passed code pointers and data pointer in far code and far data
models, respectively. Writing a function that can be easily
assembled to access the stack frame properly in any memory
model would thus seem to be a difficult task.
300
Turbo Assembler User's Guide
Fear not. Turbo Assembler provides you with the ARG directive,
which makes it easy to handle passed parameters in your
assembler routines.
The ARG directive automatically generates the correct stack
offsets for the variables you specify. For example,
arg FiIIArray:WORD,Count:WORD,FiIIValue:BYTE
specifies three parameters: FillArray, a word-sized parameter;
Count, a word-sized parameter, and FillValue, a byte-sized'
parameter. ARG actually sets the label FillArray to [BP+4]
(assuming the example code resides in a near procedure), the
label Count to [BP+6], and the label FillValue to [BP+8]. However,
ARG is valuable precisely because you can use ARG-defined labels
without ever knowing the values they're set to.
For example, suppose you've got a function FillSub, called from C
as follows:
main ()
{
jdefine ARRAY LENGTH 100
char TestArray[ARRAY_LENGTH]i
FiIISub(TestArray,ARRAY_LENGTH,'*')i
You could use ARG in FillSub to handle the parameters as follows:
FillSub
ARG
push
mov
mov
mov
mov
FillLoop:
mov
inc
loop
pop
ret
FillSub
Look at Chapter 3 in the
Reference Guide for
additional information about
the ARG directive.
PROC NEAR
FiIIArray:WORD,Count:WORD,FiIIValue:BYTE
bp
ipreserve caller's stack frame
bp,sp
iset our own stack frame
bx, [FiIIArray]
iget pointer to array to fill
cx, [Count]
iget length to fill
aI, [FillValue]
iget value to fill with
[bx],al
bx
Fi llLoop
bp
ifill a character
;point to next character
;do next character
;restore caller's stack frame
ENDP
That's really all it takes to handle passed parameters with ARG.
Better yet, ARG automatically accounts for the different sizes of
near and far returns. Another convenience is that the labels
defined with ARG are limited in scope to the procedure they're
used in when you declare them using the local label prefix (see
Chapter 7, Interfacing Turbo Assembler with Turbo C
301
LOCALS in the Reference Guide}. So you need never worry about
conflict between parameter names in different procedures.
Preserving registers
As far as Turbo C is concerned, C-callable assembler functions can
do anything they please, as long as they preserve the following
registers: BP, SP, CS, OS, and 55. While these registers can be
altered during the course of an assembler function, when the
calling code is returned, they must be exactly as they were when
the assembler function was called. AX, BX, CX, OX, ES, and the
flags can be changed in any way.
51 and 01 are special cases, since they're used by Turbo C as
register variables. If register variables are enabled in the C
module calling your assembler function, you must preserve 51
and 01; but if register variables are not enabled, 51 and 01 need
not be preserved.
It's good practice to always preserve 51 and 01 in your C-callable
assembler functions, regardless of whether register variables are
enabled. You never know when you might link a given assembler
module to a different C module, or recompile your C code with
register variables enabled, without remembering that your
assembler code needs to be changed as well.
Returning values
A C-callable assembler function can return a value, just like a C
function. Function values are returned as follows:
Return Value Type
Return Value Location
unsigned char
char
enum
unsigned short
short
unsigned int
int
unsigned long
long
float
double
long double
near"
far"
AX
AX
AX
AX
AX
AX
AX
DX:AX
DX:AX
8087 top-of-stack (TOS) register (ST(O»
8087 top-of-stack (TOS) register (ST(O»
8087 top-of-stack (TOS) register (ST(O»
AX
DX:AX
In general, 8- and 16-bit values are returned in AX, and 32-bit
values are returned in DX:AX, with the high 16 bits of the value in
302
Turbo Assembler User's Guide
DX. Floating-point values are returned in ST(O), which is the
8087's top-of-stack (TOS) register, or in the 8087 emulator's TOS
register if the floating-point emulator is being used.
Structures are a bit more complex. Structures that are 1 or 2 bytes
in length are returned in AX, and structures that are 4 bytes in
length are returned in DX:AX. Three-byte structures and
structures larger than 4 bytes must be stored in a static data area,
and a pointer to that static data must then be returned. As with all
pointers, near pointers to structures are returned in AX, and far
pointers to structures are returned in DX:AX.
Let's look at a small model C-callable assembler function,
FindLastChar, that returns a pointer to the last character of a
passed string. The C prototype for this function would be
extern char
* FindLastChar(char * StringToScan)i
where StringToScan is the nonempty string for which a pointer to
the last character is to be returned.
Here's FindLastChar:
.r-KlDEL
. CODE
PUBLIC
FindLastChar
push
mov
cld
mov
mov
mov
mov
mov
repnz
dec
dec
mov
pop
ret
FindLastChar
small
FindLastChar
PROC
bp
bp,sp
iwe need string instructions to count up
ax,ds
es,ax iset ES to point to the near data segment
di,
ipoint ES:DI to start of passed string
al,O
;search for the null that ends the string
cx,Offffh isearch up to 64K-l bytes
scasb
ilook for the null
di
ipoint back to the null
di
ipoint back to the last character
ax,di
ireturn the near pointer in AX
bp
ENDP
END
The final result, the near pointer to the last character in the passed
string, is returned in AX.
Chapter 7, Interfacing Turbo Assembler with Turbo C
303
Calling an
assembler
function from C
Now look at an example of Turbo C code calling a Turbo
Assembler function. The following Turbo Assembler module,
COUNT.ASM, contains the function LineCount, which returns
counts of the number of lines and characters in a passed string:
Small model C-callable assembler function to count the number
of lines and characters in a zero-terminated string.
Function prototype:
extern unsigned int LineCount(char * near StringToCount,
unsigned int near * CharacterCountPtr);
Input:
char near * StringToCount: pointer to the string on which
a line count is to be performed
unsigned int near * CharacterCountPtr: pointer to the
int variable in which the character count is
to be stored
NEWLINE EQU
DOSSEG
. MODEL
. CODE
PUBLIC
LineCount
push
mov
push
Dah
;the linefeed character is C's
; newline character
small
LineCount
PROC
bp
bp,sp
si
ipreserve calling program's
i register variable, if any
mov
sub
mov
LineCountLoop:
lodsb
and
jz
inc
cmp
jnz
inc
jmp
EndLineCount:
inc
304
si, [bp+4]
cX,cx
dx,cx
al,al
EndLineCount
cx
al,NEWLINE
LineCountLoop
dx
LineCountLoop
dx
ipoint SI to the string
;set character count to D
;set line count to D
;get the next character
iis it null, to end the string?
iyes, we're done
;no, count another character
;is it a newline?
;no, check the next character
;yes, count another line
;count the line that ends with the
; null character
Turbo Assembler User's Guide
mov
bx, [bp+6)
ipoint to the location at which to
i return the character count
mov
mov
pop
pop
ret
LineCount
END
[bx),cx
ax,dx
si
iset the character count variable
ireturn line count as function value
irestore calling program's register
i variable, if any
bp
ENDP
The following C module, CALLCT.C, is a sample invocation of
the LineCount function:
char * TestString="Line 1\nline 2\nline 3" i
extern unsigned int LineCount(char * StringToCount,
unsigned int * CharacterCountPtr)i
main ()
{
unsigned int LCounti
unsigned int CCounti
LCount = LineCount(TestString, &CCount)i
printf("Lines: %d\nCharacters: %d\n", LCount, CCount)i
The two modules are compiled and linked together with the·
command line
tcc -ms callct count.asm
As shown here, LineCount will only work when linked to smallmodel C programs, since pointer sizes and locations on the stack
frame change in other models. Here's a version of LineCount,
COUNTLG.ASM, that will work with large-model C programs
(but not small-model ones, unless far pointers are passed, and
LineCount is declared far):
i Large model C-callable assembler function to count the number
of lines and characters in a zero-terminated string.
Function prototype:
extern unsigned int LineCount(char * far StringToCount,
unsigned int * far CharacterCountPtr)i
char far * StringToCount: pointer to the string on which
a line count is to be performed
unsigned int far * CharacterCountPtr: pointer to the
int variable in which the character count
is to be stored
Chapter 7, Interfacing Turbo Assembler with Turbo C
305
NEWLINE EQU
Oah
ithe linefeed character is C's newline
character
i
.MODEL
•CODE
PUBLIC
LineCount
push
mov
push
push
Ids
sub
mov
LineCountLoop:
lodsb
and
jz
inc
cmp
jnz
inc
jmp
EndLineCount:
inc
large
LineCount
PROC
bp
bp,sp
si
ds
si, [bpt6]
cx,cx
dx,cx
al,al
EndLineCount
cx
al,NEWLINE
LineCountLoop
dx
LineCountLoop
dx
les
bx, [bp+10]
mov
mov
es:[bx],cx
ax,dx
pop
pop
ds
si
pop
ret
LineCount
END
bp
ipreserve calling program's
i register variable, if any
ipreserve C's standard data seg
ipoint DS:SI to the string
iset character count to 0
iset line count to 0
iget the next character
iis it null, to end the string?
iyes, we're done
ino, count another character
iis it a newline?
ino, check the next character
iyes, count another line
icount line ending with null
i character
ipoint ES:BX to the location at
i which to return char count
iset the char count variable
ireturn the line count as
i the function value
irestore C's standard data seg
irestore calling program's
i register variable, if any
ENDP
COUNTLG.ASM can be linked to CALLCT.C with the following
command line:
tcc -ml callct countlg.asm
306
Turbo Assembler User's Guide
Pascal calling
conventions
See Chapter 8 for more
Information about Pascal
calling conventions.
So far, you've seen how C normally passes parameters to functions by having the calling code push parameters right to left, call
the function, and discard the parameters from the stack after the
call. Turbo C is also capable of following the conventions used by
Pascal programs in which parameters are passed from left to right
and the called program discards the parameters from the stack. In
Turbo C, Pascal conventions are enabled with the -p commandline option or the pascal keyword.
Here's an example of an assembler function that uses Pascal
conventions:
; Called as: TEST(i, j, k);
i
j
k
TEST
TEST
equ
equ
equ
8
;leftmost parameter
4
;rightmost parameter
.MJDEL
. CODE
PUBLIC
PROC
push
mov
mov
add
sub
pop
ret
ENDP
END
small
TEST
bp
bp,sp
ax, [bp+i]
ax, [bp+j]
ax, [bp+k]
bp
6
;get i
;add j to i
;subtract k from the sum
;return, discarding 6 parameter bytes
Figure 7.7 shows the stack frame after MOV BP,5P has been
executed.
Note that RET 6 is used by the called function to clear the passed
parameters from the stack.
Pascal calling conventions also require all external and public
symbols to be in uppercase, with no leading underscores. Why
would you ever want to use Pascal calling conventions in a C
program? Code that uses Pascal conventions tends to be
somewhat smaller and faster than normal C code, since there's no
Chapter 7, Interfacing Turbo Assembler with Turbo C
307
need to execute an ADD SP n instruction to discard the
parameters after each call.
Figure 7.7
state of the stack
Immediately after
MOVBP,SP
SP
•
Caller's BP
-----BP
SP+ 2
Return Address
BP+ 2
SP+ 4
k
BP+ 4
SP+ 6
BP+ 6
SP+ 8
BP+ 8
Calling Turbo C from Turbo Assembler
Although it's most common to call assembler functions from C to
perform specialized tasks, you may on occasion want to call C
functions from assembler. As it turns out, it's actually easier to call
a Turbo C function from a Turbo Assembler function than the
reverse, since no stack-frame handling on the part of the
assembler code is required. Let's take a quick look at the
requirements for calling Turbo C functions from assembler.
Link in the C
startup code
As a general rule, it's a good idea to only call Turbo C library
functions from assembler code in programs that link in the C
startup module as the first module linked. This "safe" class
includes all programs that are linked from TC.EXE or with a
TCC.EXE command line, and programs that are linked directly
with TLINK that have COT, COS, CDC, COM, COL, or COH as the
first file to link.
You should generally not call Turbo C library functions from
programs that don't link in the C startup module, since some
Turbo C library functions will not operate properly if the startup
code is not linked in. !fyou really want to call Turbo C library
functions from such programs, we suggest you look at the startup
308
Turbo Assembler User's Guide
source code (the file CO.ASM on the Turbo C distribution disks)
and purchase the C library source code from Borland, so you can
be sure to provide the proper initializa tion for the library
functions you need. Another possible approach is to simply link
each desired library function to an assembler program, called
X.ASM for instance, which does nothing but call each function,
linking them together with a command line like this:
tlink x,x"cm.lib
where m is the first letter of the desired memory model (t for tiny,
for small, and so on). If TLINK reports any undefined symbols,
then that library function can't be called unless the C startup code
is linked into the program.
5
Note: Calling user-defined C functions that in turn call C library
functions falls into the same category as calling library functions
directly; lack of the C startup can potentially cause problems for
any assembler program that calls C library functions, directly or
indirectly.
Make sure you've
got the right
segment setup
Performing the
call
As we learned earlier, you must make sure that Turbo C and
Turbo Assembler are using the same memory model and that the
segments you use in Turbo Assembler match those used by Turbo
C. Refer to the previous section, "The framework," (page 282) if
you need a refresher on matching memory models and segments.
Also, remember to put EXTRN directives for far symbols either
outside all segments or inside the correct segment.
You've already learned how Turbo C prepares for and executes
fu.TIction calls in the section "Calling Turbo Assembler functions
from Turbo C" on page 281. We'll briefly review the mechanics of
C function calls, this time from the perspective of calling Turbo C
functions from Turbo Assembler.
All you need to. do when passing parameters to a Turbo C
function is push the rightmost parameter first, then the next
rightmost parameter, and so on, until the leftmost parameter has
been pushed. Then just call the function. For example, when
programming in Turbo C, to call the Turbo C library function
strcpy to copy SourceString to DestString, you would enter
Chapter 7, Interfacing Turbo Assembler with Turbo C
309
strcpy(DestString, SourceString);
To perform the same call in assembler, you would use
lea
lea
push
push
call
add
ax,SourceString
bx,DestString
ax
bx
_strcpy
sp,4
;rightmost parameter
;leftmost parameter
;push rightmost first
;push leftmost next
;copy the string
;discard the parameters
Don't forget to discard the parameters by adjusting SP after the
call.
You can simplify your code and make it language independent at
the same time by taking advantage of Turbo Assembler's CALL
instruction extension:
call destination [language [,argl] ••• ]
where language is C, PASCAL, BASIC, FORTRAN, PROLOC or
NOLANGUAGE, and arg is any valid argument to the routine
that can be directly pushed onto the processor stack.
Using this feature, the preceding code can be reduced to
lea ax,SourceString
lea bx,DestString
call strcpy c,bx,ax
Turbo Assembler automatically inserts instructions to push the
arguments in the correct order for C (AX first, then BX), performs
the call to _strcpy (Turbo Assembler automatically inserts an
underscore in front of the name for C), and cleans up the stack
after the call.
If you're calling a C function that uses Pascal calling conventions,
you have to push the parameters left to right and not adjust SP
afterward:
lea
lea
push
push
call
bx,DestString
ax,SourceString
bx
ax
STRCPY
;leftmost parameter
;rightmost parameter
;push leftmost first
;push rightmost next
;copy the string
;leave the stack alone
Again, you can use Turbo Assembler's CALL instruction extension
to simplify your code:
lea
310
bx,DestString
;leftmost parameter
Turbo Assembler User's Guide
lea ax,SourceString
irightmost parameter
call strcpy pascal,bx,ax
Turbo Assembler automatically inserts instructions to push the
arguments in the correct order for Pascal (BX first, then AX) and
performs the call to STRCPY (converting the name to all
uppercase, as is the Pascal convention).
Of course, the last example assumes that you've recompiled
strcpy with the -p switch, since the standard library version of
strcpy uses C rather than Pascal calling conventions. C functions
return values as described in the section "Returning values" (page
302); 8- and 16-bit values in AX, 32-bit values in OX:AX, floatingpoint values in the 8087 TOS register, and structures in various
ways according to size.
Rely on C functions to preserve the following registers and only
the following registers: 51, 01, BP, OS, 55, SP, and CS. Registers
AX, BX, CX, OX, ES, and the flags may be changed arbitrarily.
Calling a Turbo C
function from
Turbo Assembler
One case in which you might wish to call a Turbo C function from
Turbo Assembler is when you need to perform complex
calculations. This is especially true when mixed integer and
floating-point calculations are involved; while it's certainly
possible to perform such operations in assembler, it's simpler to
let C handle the details of type conversion and floating-point
arithmetic.
Let's look at an example of assembler code that calls a Turbo C
function in order to get a floating-point calculation performed. In
fact, let's look at an example in which a Turbo C function passes a
series of integer numbers to a Turbo Assembler function, which
sums the numbers and in turn calls another Turbo C function to
perform the floating-point calculation of the average value of the
series.
The C portion of the program in CALCAVG.C is
extern float Average(int far * ValuePtr, int NumberOfValues)i
'define NUMBER_OF_TEST_VALUES 10
1, 2, 3, 4, 5, 6, 7, 8, 9, 10
}i
main ()
{
Chapter 7. Interfacing Turbo Assembler with Turbo C
311
printf("The average value is: %f\n",
Average(TestValues, NUMBER_OF_TEST_VALUES))i
float IntDivide(int Dividend, int Divisor)
(
return ( (float) Dividend / (float) Divisor )i
and the assembler portion of the program in AVERAGE.ASM is
Turbo C-callable small-model function that returns the average
of a set of integer values. Calls the Turbo C function
IntDivide() to perform the final division.
Function prototype:
extern float Average(int far
* ValuePtr, int NumberOfValues)i
Input:
int far * ValuePtr:
int NumberOfValues:
ithe array of values to average
ithe number of values to average
•MODEL small
EXTRN
IntDivide:PROC
•CODE
PUBLIC _Average
_Average
PROC
push
bp
mov
bp,sp
les
bx, [bp+4]
mov
cx, [bp+8]
mov
ax,O
AverageLoop:
add
ax,es: [bx]
bx,2
add
loop
AverageLoop
push
WORD PTR [bp+8]
ipoint ES:BX to array of values
it of values to average
iclear the running total
iadd the current value
ipoint to the next value
iget back the number of values
passed to IntDivide as the
rightmost parameter
ax
ipass the total as the leftmost parameter
_IntDivide ;calculate the floating-point average
sp,4
;discard the parameters
bp
;average is in 8087's TOS register
ENDP
i
i
push
call
add
pop
ret
_Average
END
The C main function passes a pointer to the array of integers
TestValues and the length of the array to the assembler function
Average. Average sums the integers, then passes the sum and the
312
Turbo Assembler User's Guide
number of values to the C function IntDivide. IntDivide casts the
sum and number of values to floating-point numbers and
calculates the average value, doing in a single line of C code what
would have taken several assembler lines. IntDivide returns the
average to Average in the 8087 TOS register, and Average just leaves
the average in the TOS register and returns to main.
CALCAVG.C and AVERAGE.ASM could be compiled and linked
into the executable program CALCAVG.EXE with the command
tee ealeavg average.asm
Note that Average will handle both small and large data models
without the need for any code change, since a far pointer is
passed in all models. All that would be needed to support large
code models (huge, large, and medium) would be use of the
appropriate .MODEL directive.
Taking full advantage of Turbo Assembler's languageindependent extensions, the assembly code in the previous
example could be written more concisely as
DOSSEG
. MODEL
EXTRN
.CODE
PUBLIC
Average
les
mov
mov
AverageLoop:
add
add
loop
call
ret
Average
END
small,C
C IntDivide:PROC
C Average
PROC C ValuePtr:DWORD,NumberOfValues:WORD
bx,ValuePtr
cx,NumberOfValues
ax,O
ax,es:[bx]
bx,2
ipoint to the next value
AverageLoop
IntDivide C,ax,NumberOfValues
ENDP
Chapter 7, Interfacing Turbo Assembler with Turbo C
313
314
Turbo Assembler User's Guide
c
H
A
p
T
E
R
8
Interfacing Turbo Assembler with Turbo
Pascal
Turbo Assembler provides extensive and powerful facilities to let
you add assembly language code to your Turbo Pascal programs.
In this chapter, we'll tell you everything you need to know to
make full use of these facilities, including lots of examples and
"inside" information.
Unless a version number is
stated specifically. when
referring to Turbo Pascal, we
mean versions 4.0 and
greater.
Why use Turbo Assembler with Turbo Pascal? Most of the
programs you're likely to write can be written entirely in Turbo
Pascal. Unlike most Pascals, Turbo Pascal lets you access virtually
all of your machine's resources directly through the Port[], Mem[],
MemW[], and MemL[] arrays, and you can call the BIOS and
operating system with the IntrO and MsDosO procedures.
Why, then, would you want to use assembly language with Turbo
Pascal? The two most likely reasons: to perform the relatively few
operations that are not directly available from Turbo Pascal and to
take advantage of the raw speed that only assembly language can
provide. (Turbo Pascal itself is so quick because it is written in
assembly language.) This chapter shows you how and when to
harness the power of assembly language with Turbo Pascal.
The Turbo Pascal memory map
Before you can begin writing assembly language code to work
with Turbo Pascal programs, it's important to understand how the
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
315
compiler lays out information in memory. The Turbo Pascal
memory model embodies aspects of both the medium and the
large models, which are described in Chapter 5. There is a single
global data segment, allowing fast access to global variables and
typed constants through DS. However, each unit has its own code
segment, and the heap can grow to use all available memory.
Addresses in Turbo Pascal are always passed as far (32-bit)
pointers so that they can reference objects anywhere in memory.
The memory map of a Turbo Pascal program looks like this:
Figure 8.1
Memory map of a
Turbo Pascal 5.0
program
Low Memory
Program Segment Prefix
(256 Bytes)
Maximum
code segment
si.ze: 64K
Main Program Code Segment
Last Unit Code Segment
~
-
.
---
First Unit Code Segment
os
Run Time Library Code Segment
Typed Constants
-------------------------------Global Variables
SS
t
~t9c~ Sii:~
Minimum: 1K
Default:
16K
Maximum: 64K
Stack (Grows downward)
Heap (Grows upward)
Heap
Ptr
t
t
___ End of . EXE
file
No size limit
Heap "free list" (Grows downward)
Maximum free
list size: 64K
High Memory
The program
segment prefix
316
The program segment prefix (PSP) is a 256-byte area created by MSDOS when the program is loaded. Among other things, it contains
Turbo Assembler User's Guide
infonnation about command-line parameters used to invoke the
program, the amount of available memory, and the DOS
environment (a list of string variables used by DOS).
In Turbo Pascal 3.0, the segment address of the PSP was the same
as that of all the rest of the code. This is no longer the case. In
Turbo Pascal versions 4.0 and later, the main program, t~e units it
uses, and the run-time library all occupy different segments.
Turbo Pascal therefore stores the segment address of the PSP in a
predeclared global variable called PrefixSeg, so that you can gain
access to PSP infonnation.
Code segments
Every Turbo Pascal program has at least two code segments: one
for the code of the main program and one for the run-time library .
In addition, each unit's code occupies a separate code segment.
Since each code segment can be up to 64K in size, your program
can occupy as much memory as you want (subject, of course, to
what is available on the machine). Programmers who fonnerly
used overlays to generate programs larger than 64K can now keep
all the code in memory for faster execution. Viewed from Turbo
Assembler, the code segment into which an assembly language
module is linked has the name CODE, or CSEG.
The global data
segment
Turbo Pascal's global data segment follows the run-time library
code segment. It contains up to 64K of initialized and uninitialized data: typed constants and global variables. As in Turbo Pascal
3.0, typed constants are really not constants at all, but variables
that start with a pre-initialized value when the program is loaded.
But unlike Turbo Pascal 3.0, Turbo Pascal 4.0 does not place typed
constants in the code segment. Instead, Turbo Pascal 4.0 places
typed constants in the global data segment, where it can access
them even more quickly than Turbo 3.0 could. The global data
segment has the name DATA, or DSEG, when it's referenced from
a Turbo Assembler module.
The stack
In Turbo Pascal 4.0 and later, the global data segment is above the
stack. Note that this arrangement is different from the one used in
Turbo Pascal 3.0. The stack and heap do not grow toward each
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
317
other. Instead, a fixed amount of memory is alloca ted for the
stack. The default size, 16K, is more than enough for the vast
majority of programs; however, you can specify a stack size as
small as 1K (for short programs) or as large as 64K (for programs
with a lot of recursion). Stack and heap sizes can be selected with
the $m compiler directive.
As in most 80x86 programs, the stack pointer starts at the top of
the stack segment and grows downward. Whenever a procedure
or function is called, Turbo Pascal normally checks to make sure
that the stack is not exhausted. This check can be turned off with
the {$s-} compiler directive.
The heap
At the top of the Turbo Pascal memory map is the heap. By
default, the heap takes up all memory not allocated for the code,
data, and stack segments, but the $m directive can be used to limit
the maximum size of the heap. (It can also be used to prevent the
program from running if a minimum amount of heap space is not
available.)
Storage is allocated dynamically on the heap, beginning from the
bottom, each time you do a NewO or GetMemO. Space is freed
when you do a Dispose, Release, or FreeMem. When Dispose and
FreeMem are used, Turbo Pascal 4.0 keeps track of free areas in the
middle of the heap using a data structure called a free list. The free
list, which can be up to 64K in size, grows downward from the
very top of the heap area.
Register use in Turbo Pascal
Like Turbo Pascal 3.0, Turbo Pascal 4.0 imposes a minimum of
restrictions on register use. When a call is made to a function or
procedure, the values of only three registers must be preserved:
stack segment (S5), data segment (DS), and base pointer (BP). DS
points to the global data segment (called DATA), and SS points to
the stack segment. BP is used by each procedure or function to
reference its activation record-the stack space it uses for
parameters, local variables, and temporary storage. All subprograms must also adjust the stack pointer (SP) before exiting, so
that the parameters no longer remain on the stack.
318
Turbo Assembler User's Guide
Near or for?
Because a Turbo Pascal program contains multiple code segments,
it uses a mixture of near and far calls to access procedures and
functions. What's the difference? Well, a near call can only be used
to access a subprogram that resides in the same code segment
where the call. is made, while a far call can access a subprogram
anywhere in memory. This flexibility incurs a small penalty,
however: A far call takes a bit more time and space than a near
call.
Any subprogram can be
forced to be for by the {Sf+J
compiler directive.
Each subprogram in your Turbo Pascal program must be written
(either by the compiler or by you) to be called in only one of these
two ways. Which should you choose? Subprograms declared in
the Interface section of a unit must always be far so that they can
be called from other units. But subprograms declared in the main
program, or declared only in the Implementation section of a unit,
are usually near.
When you write assembly language routines to interface with
Turbo Pascal, you must check to make sure that your routine has
the correct "distance." Turbo Pascal does not report an error if
you declare a PROC as near in assembly language when the
corresponding external procedure declaration is positioned in
such a way that it needs to be far.
Sharing information with Turbo Pascal
The $1 compiler
directive and
external
subprograms
The two keys to using Turbo Assembler with Turbo Pascal are the
{$l} compiler directive and the external subprogram declaration.
The directive {$l MYFILE.OBJ} causes Turbo Pascal to look for
MYFILE.OBJ, a file in standard MS-DOS linkable object format, and
link it into your Turbo Pascal program. If the file name given in
the {$l} directive does not have an extension, .OBI is assumed.
Each Turbo Assembler procedure or function that you want to be
visible within the Turbo Pascal program must be declared as a
PUBLIC symbol, and must have a corresponding external
declaration within that program. The syntax of an external
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
319
procedure or function declaration in Turbo Pascal is very similar
to that of a forward declaration:
procedure AsmProc(a : Integer; b : Real); eztetnal;
function AsmFunc(c : Word; d : Byte); ezteInll;
These declarations might correspond to the following declarations
within your Turbo Assembler program:
CODE
SEGMENT BYTE PUBLIC
AsmProc PROC NEAR
PUBLIC AsmProc
AsmProc ENDP
AsmFunc PROC FAR
PUBLIC Bar
AsmFunc ENDP
CODE ENDS
A Turbo Pascal external procedure declaration must be at the
outermost level of the program or unit; that is, it may not be
nested within another procedure declaration. An attempt to
declare an external procedure at any other level will cause a
compile-time error.
1111"
Turbo Pascal does not check to make sure that PROCs declared
with the near and far attributes correspond to near and far
subprograms in your Turbo Pascal program. In fact, it does not
even check to see whether the public labels AsmProc and AsmFunc
are the names of PROCs. It is up to you to make sure that the
assembly language and Pascal declarations are consistent.
The PUBLIC
directive Only labels that are declared PUBLIC in an assembly language
module are visible to Turbo Pascal. Labels are the only objects
that can be exported from assembly language to Turbo Pascal.
Further, every label that is made PUBLIC must have a
corresponding procedure or function declaration in the Turbo
Pascal program, or the compiler will report an error. A public
label need not be part of a PROC declaration. As far as Turbo
Pascal is concerned,
AsmLabel
320
PROC FAR
PUBLIC Bar
Turbo Assembler User's Guide
and
AsmLabel:
PUBLIC Bar
are equivalent.
The EXTRN
directive
This Includes variables
declared after the {$I}
complier directive and the
extemal declarat/on(s)
associated with the module.
A Turbo Assembler module can access any Turbo Pascal
procedure, function, variable, or typed constant that is declared at
the outermost level of the program or unit to which it is linked.
Turbo Pascal labels and ordinary constants are not visible to the
assembly language.
Suppose your Turbo Pascal program declares the following global
variables:
var
a
b
c
d
e
f
g
h
i
: Byte;
: Word;
Shortint;
: Integer;
: Real;
: Single;
: Double;
: Extended;
: Comp;
: Pointer;
You can access any of these variables inside your assembly
language program with EXTRN declarations, as follows:
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
EXTRN
A:
B:
C:
D:
E:
F:
G:
H:
I :
BYTE
WORD
BYTE
WORD
FWORD
DWORD
QWORD
TBYTE
QWORD
J : DWORD
;1 byte
;2 bytes
;Assembly language treats signed & unsigned alike
;Ditto
;6-byte software real
;4-byte IEEE floating point
;8-byte IEEE double-precision floating point
;10-byte IEEE temporary floating point
;8087 8-byte signed integer
;Turbo Pascal pointer
You can access Turbo Pascal procedures and functions-including
library routines-in a similar manner. Suppose you have a Turbo
Pascal unit that looks like this:
Sample;
{ Sample unit that defines several pascal procedures that are
unit
Chapter 8. Interfacing Turbo Assembler with Turbo Pascal
321
called from an assembly language procedure.
interface
procedure TestSample;
procedure PublicProc;
{Must be far since it's visible outside I
ilIpleaentation
var
A : word;
procedure AsmProc;
exte~l;
{$L ASMPROC.OBJI
procedure PublicProc;
begin ( PublicProc I
Writeln('In PublicProc');
{PublicProc I
end;
procedure NearProc;
begin ( NearProc I
{ Must be near
Writeln('In NearProc');
{NearProc I
end;
{$F+I
procedure FarProc;
begin ( FarProc I
{ Must be far due to compiler directive
Writeln('In FarProc');
{FarProc I
end;
{$F-I
procedure TestSample;
begin { TestSample }
Writeln('In TestSample');
A := 10;
Writeln('Value of A before ASMPROC = , ,A);
AsmProc;
Writeln('Value of A after ASMPROC = , ,A);
end { TestSample I;
end.
The procedure AsmProc can call procedures PublicProc, NearProc,
or FarProc by using EXTRN directives as follows:
DATA
DATA
CODE
322
SEGMENT WORD PUBLIC
ASSUME DS:DATA
EXTRN A:WORD
ENDS
;variable from the unit
SEGMENT BYTE PUBLIC
ASSUME CS:CODE
EXTRN PublicProc: FAR ;far procedure
Turbo Assembler User's Guide
EXTRN NearProc: NEAR
EXTRN FarProc : FAR
; (exported by the unit)
;near procedure (local to unit)
ifar procedure
; (local but forced far)
AsmProc PROC NEAR
PUBLIC AsmProc
call FAR PTR PublicProc
call NearProc
call FAR PTR FarProc
mov
cx,ds:A
;pull in var A from the unit
sub
;do something to change it
cx,2
mov
;store it back
ds:A,cx
ret
AsmProc ENDP
CODE
ENDS
END
The main program that tests this Pascal unit and assembler code
follows:
proqraa TSample;
u...
Sample;
begin
TestSample;
end.
To build the sample program with the command-line compiler
and the assembler, use the following batch file commands:
TASM ASMPROC
TPC /B TSAMPLE
TSAMPLE
Since an external subprogram must be declared at the outermost
procedural level of your Turbo Pascal program, you can't use
EXTRN declarations to access objects that are local to a procedure
or function. However, your Turbo Assembler subprogram can
receive these objects as value or var parameters when it's called
from Turbo Pascal.
Restrictions on using
EXTRN objects
Turbo Pascal's qualified identifier syntax, which uses a unit name
followed by a period to access an object in a specific unit, is not
compatible with Turbo Assembler's syntax rules and will
therefore be rejected. The declaration
EXTRN SYSTEM.Assign : FAR
produces a Turbo Assembler error message.
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
323
There are two other minor restrictions on the use of EXTRN
objects with Turbo Pascal. The first is that references to
procedures and functions cannot use address arithmetic. Thus, if
you declare
EXTRN PublicProc : FAR
you can't write a statement such as
call 'PublicProc + 42
The second restriction is that the Turbo Pascal linker will not
recognize operators that chop words into bytes, so you cannot
apply these operators to EXTRN objects. For instance, if you
declare
EXTRN i : WORD
you can't use the expressions LOW i or HIGH i in your Turbo
Assembler module.
Using segment
fixups Turbo Pascal generates .EXE files, which can be loaded at any
available address in your PC's memory. Since the program cannot
know in advance where a given segment of your program will be
loaded, the linker tells the DOS .EXE loader to fix up all references
to segments in your program when it is loaded. After the fixups
are done, all references to segments (such as CODE and DATA)
contain the correct values.
Your Turbo Assembler code can use this facility to obtain the
segment addresses of objects at run time. For instance, suppose
your program needs to change the value of DS, but you don't
want to spend the cycles required to save the original contents on
the stack or move them to a temporary location. Instead, you can
use the Turbo Assembler SEG operator as follows:
mov ax,SEG DATA
mov ds,ax
iget actual address of Turbo Pascal's global DS
iput it in DS for Turbo Pascal to use
When your Turbo program is loaded, DOS will plug the correct
value for SEG DATA right into the immediate operand field of the
MOV instruction. This is the fastest way to reload the segment
register.
324
Turbo Assembler User's Guide
This technique is also necessary to allow interrupt service routines
to save information in Turbo Pascal's global data segment. DS will
not necessarily contain Turbo Pascal's DS at interrupt time, but
the preceding sequence can be used to gain access to Turbo Pascal
variables and typed constants.
Dead code
elimination
Turbo Pascal features dead code elimination, which means that it
does not include code for routines that are never executed when it
writes the final.EXE file. But, because it does not have complete
information about the contents of your Turbo Assembler
modules, Turbo Pascal can only perform limited optimization on
them.
Turbo Pascal will eliminate the code of an .OB} module if and only
if no calls are made to any visible procedure or function in that
module. Conversely, if any routine in the module is referenced,
the entire module stays.
1111"
To make the most efficient use of Turbo's dead code elimination
feature, it's a good idea to break up your assembly language into
small modules with only a few routines each. Doing so will allow
Turbo to "trim the fat" from your finished program, if it can.
Turbo Pascal parameter-passing conventions
Turbo Pascal passes parameters using the CPU's stack (or, in the
case of Single, Double, Extended, or Comp value parameters, the
numeric processor's stack). Parameters are always evaluated and
pushed on the stack in the order they appear in the declaration of
the subprogram, from left to right. In this section, we'll explain
how these parameters are represented.
Value parameters
A value parameter is a parameter whose value cannot be changed
by the subprogram to which it is passed. Unlike many compilers,
Turbo Pascal does not blindly copy every value parameter onto
the CPU stack; the method used depends on the type, as we
explain in this and the ne~t few pages.
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
325
Scalar types
Value parameters of all the scalar types (Boolean, Char, Shortint,
Byte, Integer, Word, Longint, subrange types, and enumerated
types) are passed as values on the CPU stack. If an object is 1 byte
in size, it is pushed as a full 16-bit word; however, the mostsignificant byte of that word contains no useful information. (This
byte cannot be relied on to be 0, as it could in Turbo Pascal
versions 3.0 and earlier.) If the object is 2 bytes in size, it is simply
pushed as is. If the object is 4 bytes long (a Longint), it is pushed
as two 16-bit words. As is standard on the 8088 family of
processors, the most-significant word is pushed first and occupies
the higher address on the stack.
Note that the Comp type, while it is an Integer type, is not
considered to be a scalar type for the purposes of parameterpassing. Thus, in Turbo Pascal 4.0, value parameters of this type
are passed on the 8087 stack, not the CPU stack. In Turbo Pascal
5.0, values of the Comp type are passed on the main CPU stack.
Reals
Value parameters of the type Real (Turbo Pascal's 6-byte software
floating-point type) are passed as 6 bytes on the stack. This is the
only type larger than 4 bytes that is ever passed on the stack.
Single, Double,
Extended, and Comp:
The 8087 types
In Turbo Pascal 4.0, value parameters of the 8087 types are passed
on the coprocessor stack, not the CPU stack. Since the 8087 stack
is only eight levels deep, a Turbo Pascal 4.0 subprogram cannot
have more than eight 8087-type value parameters. Al18087-type
parameters must be popped from the numeric processor stack
before the subprogram returns.
Turbo Pascal 5.0 uses the same parameter-passing conventions for
8087 values as Turbo C does: They are passed on the main CPU
stack with the other parameters.
Pointers
326
Value parameters of all pointer types are pushed directly on the
stack as far pointers-first a word containing the segment, then
another containing the offset. The segment occupies the higher
address, in accordance with Intel conventions. Your Turbo
Assembler program can use the LOS or LES instruction to retrieve
a pointer parameter.
Turbo Assembler User's Guide
Strings
For more Information, refer to
Chapter 13, "Overlays," In
the Turbo Pascal Reference
Guide (5.0).
String parameters, regardless of size, are usually not pushed on
the stack. Instead, Turbo Pascal pushes a far pointer to the string.
It's the responsibility of the called subprogram not to change the
string referenced by the pointer; the subprogram must make and
work on a copy of the string, if necessary.
The only exception to this rule is when a routine in overlaid unit
A passes a string constant as a value parameter to a routine in
overlaid unit B. In this context, an overlaid unit means any unit
compiled with ($o+} (Overlays Allowed). In this case, temporary
storage is reserved on the stack for the string constant before the
call is made and the stack address is passed to the routine in unit
B.
Records and arrays
Records and arrays that are exactly I, 2, or 4 bytes long are
duplicated directly onto the stack when they are passed as value
parameters. If an array or record object is any other size
(including 3 bytes), a pointer to it is pushed instead. In the case of
records and arrays that aren't 1,2, or 4 bytes long, the subprogram must make a local copy of the structure if it modifies it.
Sets
Sets, like strings, are usually not pushed verbatim on the stack.
Instead, a pointer to the set is pushed. The pointer received by the
subprogram will point to a"normalized" 32 byte representation
of the set. The first bit of the lowest byte of this set will always
correspond to the element of the base type (or its parent type)
with the ordinal value O.
.
For more information, refer to
Chapter 13, "Overlays," in
the Turbo Pascal Reference
Guide (5.0).
The only exception to this rule is when a routine in overlaid unit
A passes a set constant as a value parameter to a routine in
overlaid unit B. In this context, an overlaid unit means any unit
compiled with ($O+} (Overlays Allowed). In this case, temporary
storage is reserved on the stack for the set constant before the call
is made and the stack address is passed to the routine in unit B.
Variable
parameters All var parameters are passed exactly the same way: as far
pointers to their actual locations in memory.
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
327
Stack
maintenance
If you use the .MODEL, PROC,
and ARG directives, the
assembler automatically
adds the number of
parameter bytes to be
popped to all RET Instructions.
Accessing
oarameters
When cdmputing parmater
locations, take Into account
any registers whose contents
you might have pushed.
Turbo Pascal expects that all parameters on the main CPU stack
will be removed before a subprogram returns.
There are two ways to adjust the stack. You can use the RETN
instruction (where N is the number of bytes of parameters
pushed), or you can save the return address in registers (or in
memory) and pop the parameters off one by one. The popping
technique is useful when you're optimizing for speed on the 8086
and 8088 (the slowest processors in the family), where base-plusoffset addressing costs eight cycles (minimum) per access. It can
also save space, since a POP instruction takes only a single byte.
When your Turbo Assembler routine receives control, the top of
the stack contains a return address (two or four words, depending
on whether the routine is near or far) and, above it, any
parameters being passed.
There are three basic techniques for accessing the parameters
passed to your Turbo Assembler routine by Turbo Pascal. You can
• use the BP register to address the stack
• use another base or index register to get the parameters
• pop the return address, then pop the parameters
The first and second techniques are somewhat complicated, and
we cover them in the next two sections. The third technique
involves popping the return address into a safe place and then
popping the parameters into registers. This technique works best
when your routine does not require any local variable space.
Using BP to address the
stack
The first (and most often used) technique for accessing the
parameters passed from Turbo Pascal to Turbo Assembler is to
use the BP register to address the stack, like this:
CODE
MyProc
328
SEGMENT
ASSUME cs:CODE
PROC FAR
PUBLIC MyProc
EQU WORD PTR [bp+6]
EQU WORD PTR [bp+8]
push bp
iprocedure MyProc(i,j : integer)i
ij above saved BP & return address
ii just above j
iffiust preserve caller's BP
Turbo Assembler User's Guide
mov bp,sp
mov ax,i
imake BP point to top of stack
iaddress i via BP
In computing the stack offsets of parameters to be accessed in this
way, remember to allow 2 bytes for the saved BP register.
1111.
Note the use of text equates for the parameters in this example.
These help to make the code more mnemonic. They have only one
minor drawback: Because only the eau directive can be used to
do this kind of equate (not the =directive), you will not be able to
redefine the symbols i and j again in the same Turbo Assembler
source file. One way to get around this is to use more descriptive
parameter names so that they do not repeat; another is to
assemble each routine separately.
The ARG directive
When you access your parameters via the BP register, however,
Turbo Assembler provides an alternative to calculating stack
offsets and performing text equates-the ARG directive. Used
inside a PROC, the ARG directive automatically determines the
offsets of the parameters relative to BP. It also calculates the size
of the parameter block for use in the RET instruction. Because the
symbols created by the ARG directive are defined only within the
surrounding PROC, you do not need unique parameter names for
each procedure or function.
Here's how the preceding example looks rewritten with the ARG
directive:
CODE
SEGMENT
ASSUME cs:CODE
MyProc PROC FAR
;procedure MyProc(i,j : integer); external;
PUBLIC MyProc
ARG j : WORD, i: WORD = RetBytes
push bp
;must preserve caller's BP
mov bp,sp
;make BP point to the top of the stack
mov ax,i
iaddress i via BP
Turbo Assembler's ARG directive creates local symbols for the
parameters i and j. The line
ARG j: WORD, i : WORD
=
RetBytes
automatically equates the symbol i to [WORD PTR BP+6], the
symbol j to [WORD PTR BP+8], and the symbol RetBytes to the
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
329
number 4 (the size in bytes of the parameter block) for the
duration of the procedure. The values take into account both the
pushed BP and the size of the return address; if MyProc were a
NEAR PROC, i would have been equated to [BP+4], j to [BP+6],
and RetBytes would still have contained the value 4 (so that, in
either case, MyProc could end with the instruction RET RetBytes).
1111"
When you use the ARG directive, remember to list the parameters
in reverse order. You would place the last parameter in the Turbo
Pascal procedure (or function) header first in the ARG directive,
and vice versa.
Another precaution is in order when you use the ARG directive
with Turbo Pascal. Unlike some other languages, Turbo Pascal
always pushes a byte-sized value parameter as a full 16-bit
word-and you are responsible for telling Turbo Assembler about
the extra byte. For instance, suppose you wrote a function whose
Pascal declaration looked like this:
function MyProc(i,j : Char) : string; exter.oali
The ARG directive for this procedure would have to look
something like this:
ARG j : BYTE: 2, i : BYTE : 2 = RetBytes RETURNS result : DWORD
The: 2 after each argument is necessary to tell Turbo Assembler
that each character is pushed as an array of 2 bytes (where, in this
case, the upper byte of each pair holds no useful information).
See Chapter 3 In the
Reference Guide for
complete Information on the
ARG directive.
In a function that returns a string (like the previous one), the
RETURNS option in the ARG directive lets you define a variable
that equates to a place on the stack that points to the temporary
function result (discussed shortly). The variable in the RETURNS
portion of ARG doesn't affect the size (in bytes) of the parameter
block.
.MODEL and Turbo
Pascal
The .MODEL directive with a parameter of TPASCAL sets up
simplified segmentation, memory model, and language support.
Previously, you've seen how to set up an assembler program for
Pascal procedures and functions. Here's the same example
recoded to use the .MODEL and PROC directives:
.MODEL
•CODE
MyProc PROC
PUBLIC
mov
330
TPASCAL
FAR i:BYTE,j:BYTE RETURNS result:DWORD
MyProc
ax,i
Turbo Assembler User's Guide
ret
Notice that now you don't specify the parameters in reverse order
and a lot of other statements are not required. Using TPASCAL
with the .MODEL directive sets up Pascal calling conventions,
defines the segment names, does the PUSH BP and MOV BP,SP,
and also sets up the return with POP BP and RETN (where N is
the number of parameter bytes).
USing another base or
index register
The second way to access parameters is to use another base or
index register-BX, 51, or Ol-to get them from the stack.
Remember, however, that the default segment for these registers
is OS, not 55; you will have to use a segment override or change a
segment register to use them.
Here's how to use BX to get at your parameters:
CODE
SEGMENT
ASSUME cs:CODE
MyProc PROC FAR
iprocedure MyProc(i,j : integer)i
PUBLIC MyProc
EQU WORD PTR ss:[bx+4]
;j above return address
EQU WORD PTR ss: [bx+6]
;i just above j
mov bx,sp
;make BX point to top of stack
mov ax,i
;address i via BX
In routines where a small number of references are made to
parameters, this technique saves time and space. Why? Because
BX, unlike BP, need not be restored at the end of the routine.
Function results in Turbo Pascal
Turbo Pascal functions return their results in different ways
depending on the result type.
Scalar function results
Function results of scalar types are returned in CPU registers.
Values of 1 byte are returned in AL, 2-byte values in AX, and
4-byte values in OX:AX (most-significant word in OX).
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
331
Real function results
Function results of Turbo Pascal's 6-byte software real type are
returned in three CPU registers. The most-significant word goes
in DX, the middle word in BX, and the least-significant word in
AX.
8087 function results
String function results
Don't remove the function
result pointer from the stack:
Turbo Pascal expects it to be
available after the call.
Function results of 8087 types are returned in the 8087's "top-ofstack" register, ST(O) (or just ST).
Function results of a string type are returned in a temporary area
allocated by Turbo Pascal before the call. A far pointer to this area
is pushed on the stack before the first parameter is pushed. Note
that this pointer is not part of the parameter list.
Pointer function results
Pointer function results are returned in DX:AX (segment:offset).
Allocating space for local data
Your Turbo Assembler routines can allocate space for their own
variables-both static (remaining between calls) and volatile
(disappearing after a call). We'll discuss how to do both in the
next two sections.
Allocating private
static storage
Turbo Pascal allows your Turbo Assembler program to reserve
space for static variables in the global data segment (DATA, or
DSEG). To allocate the space, simply use directives such as DB,
DW, and so on, like this:
DATA
SEGMENT PUBLIC
MyInt DW?
MyByte DB?
DATA
iReserve a word
iReserve a byte
ENDS
Two important restrictions apply to variables allocated by Turbo
Assembler in the global data segment. First, these variables are
private-they cannot be made visible to your Turbo Pascal
332
Turbo Assembler User's Guide
program (though you can pass pointers to them). Second, they
can't be pre-initialized, as typed constants are. The statement
MyInt DW 42
ithis will NOT initialize Mylnt to 42
will not cause an error when the module is linked into your Turbo
program, but Mylnt will not actually start with the value 42 when
the program is run.
You can get around these restrictions by declaring Turbo Pascal
variables or typed constants and using the EXTRN directive to
make them visible to Turbo Assembler.
Allocating volatile
storage
Your Turbo Assembler routines can also allocate volatile storage
(local variables) on the stack for the duration of each call. This
storage must be reclaimed and the BP register restored before the
routine returns. In the following example, the procedure MyProc
reserves space for two integer variables, a and b:
CODE
The LOCAL directive Is used to
create symbols and allocate
space for local variables.
SEGMENT
ASSUME cs:CODE
MyProc PROC FAR
iprocedure MyProc(i Integer)i
PUBLIC MyProc
LOCAL a : WORD, b : WORD = LocalSpace
ia at [bp-2], b at [bp-4]
EQU WORD PTR [bp+6]
iparameter i above saved BP
i
i and return address
push bp
imust preserve caller's BP
mov bp,sp
imake BP point to top of stack
sub sp,LocalSpace
imake room for the two words
iload A's initial value into AX
mov ax,42
mov a,ax
iand thence into A
xor ax,ax
iclear AX
iand initialize B to 0
mov b,ax
ido whatever needs to be done
ithis restores the original SP
mov sp,bp
ithis restores the original BP
pop bp
ret 2
ithis pops the word parameter
MyProc ENDP
CODE ENDS
END
The statement
LOCAL a : WORD, b : WORD = LocalSpace
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
333
equates the symbol a to [BP-2], the symbol b to [BP-4], and the
symbol LocalSpace to the number 4 (the size of the local variable
area) for the duration of the procedure. There is no corresponding
statement to create symbols that reference parameters, so you
must still equate i to [BP+6].
A more clever way to initialize local variables is to push their
values instead of decrementing SP. Thus, you might replace the
SUB SP, LocalSpace with
mov
push
xor
push
1111"
ax,42
ax
ax,ax
ax
iget the initial value for A
iPut it in A
izero AX
iand move the zero into B
If you use this method, be sure to keep careful track of the stack!
The symbols a and b should not be referenced before the pushes
are performed.
Other optimizations include using the PUSH CONST instructions
to initialize local variables (available on the 80186, 80286, and
80386), or saving BP in a register instead of pushing it (if there is a
register to spare).
Assembly language routines for Turbo Pascal
In this section, we've provided some examples of assembly
language routines that you can call from a Turbo Pascal program.
General-purpose
hex conversion The bytes at num are converted to a string of hex digits of length
routine (byteCount * 2). Since each byte produces two characters, the
maximum value of byteCount is 127 (not checked). For speed, we
use an add-daa-adc-daa sequence to convert each nibble to a hex
digit (1 nibble equals.4 bits).
HexStr is written to be called with a far call. This means that it
should be declared either in the Interface section of a Turbo Pascal
unit or with the $/+ compiler directive active.
CODE
SEGMENT
ASSUME cs:CODE,ds:NOTHING
i Parameters (+2 because of push bp)
byteCount EQU BYTE PTR ss:[bp+6]
334
Turbo Assembler User's Guide
num
EQU DWORD PTR ss:[bp+8]
; Function result address (+2 because of push bpI
resultPtr EQU DWORD PTR ss:[bp+l2]
HexStr
PROC FAR
PUBLIC HexStr
push bp
mov bp,sp
les di,resultPtr
mov dx,ds
Ids si,num
mov al,byteCount
xor ah,ah
mov cx,ax
add si,ax
dec si
shl ax,l
cld
stosb
;get pointer into stack
;get address of function result
;save Turbo's DS in DX
;get number address
;how many bytes?
imake a word
ikeep track of bytes in CX
;start from MS byte of number
;how many digits? (2/byte)
;store t digits (going forward)
;in destination string's length byte
HexLoop:
HexStr
CODE
std
lodsb
mov ah,al
shr al,l
shr al,l
shr al,l
shr al,l
add al,90h
daa
adc al,40h
daa
cld
stosb
mov al,ah
and al,OFh
add al,90h
daa
adc al,40h
daa
stosb
loop HexLoop
mov ds,dx
pop bp
ret 6
ENDP
ENDS
END
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
;scan number from MSB to LSB
;get next byte
;save it
;extract high nibble
ispecial hex conversion sequence
;using ADDs and DAA's
;nibble now converted to ASCII
;store ASCII going up
;repeat conversion for low nibble
;keep going until done
;restore Turbo's DS
iparameters take 6 bytes
335
The sample Pascal program that uses HexSfr follows:
prograa HexTest;
var
num : Word;
{$F+}
function HexStr (var num; byteCount
Byte)
.tr1ng; exter.nal;
{$L HEXSTR.OBJ}
{$F-}
be¢n
num := $face;
Writeln('The Converted Hex String is
"' , HexStr (num, sizeof (num) ) , "" ) ;
end.
Use the following batch file commands to build and run the
example Pascal and assembly program:
TASM HEXSTR
TPC HEXTEST
HEXTEST
If you use the .MODEL directive, the program HexStr could be
written like this:
HexStr
.MODEL TPASCAL
•CODE
PROC FAR num:DWORD,byteCount:BYTE RETURNS resultPtr:DWORD
PUBLIC HexStr
les di,resultPtr
;get address of function result
mov dx,ds
;save Turbo's DS in DX
Ids si,num
;get number address
mov al,byteCount
;how many bytes?
;make a word
xor ah,ah
mov cx,ax
;keep track of bytes in CX
add si,ax
;start from MS byte of number
dec si
;how many digits? (2/byte)
shl ax,l
;store # digits (going forward)
cld
;in destination string's length byte
stosb
HexLoop:
std
lodsb
mov ah,al
shr al,l
shr al,l
shr al,l
shr al,l
336
;scan number from MSB to LSB
; get next byte
;save it
;extract high nibble
Turbo Assembler User's Guide
HexStr
CODE
add al,90h
daa
adc al,40h
daa
cld
stosb
mov al,ah
and al,OFh
add al,90h
daa
adc al,40h
daa
stosb
loop HexLoop
mov ds,dx
ret
ENDP
ENDS
END
ispecial hex conversion sequence
iusing ADDs and DAA's
inibble now converted to ASCII
istore ASCII going up
irepeat conversion for low nibble
ikeep going until done
irestore Turbo's DS
You can use the same sample Pascal program and just assemble
the alternative HexStr, recompiling the sample program with the
same batch file commands.
Exchanging two
variables
With this procedure, you can exchange two variables of size count.
If count is 0, the processor will attempt to exchange 64K.
CODE
i
SEGMENT
ASSUME cs:CODE,ds:NOTHING
Parameters (note that offset are +2 because of push bp)
varl
var2
count
EQU
EQU
EQU
DWORD PTR ss:[bp+12]
DWORD PTR ss:[bp+8]
WORD PTR ss:[bp+6]
Exchange PROC FAR
PUBLIC Exchange
cld
mov
dx,ds
push
bp
bp,sp
mov
Ids
si,varl
les
di,var2
mov
cx,count
cx,l
shr
carry)
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
;exchange goes upward
isave DS
i9~t
iget
iget
iget
iget
stack base
first address
second address
number of bytes to move
word count (low bit ->
337
jnc
mov
movsb
mov
jz
ExchangeWords:
mov
ExchangeLoop:
mov
movsw
mov
loop
Finis:
mov
pop
ret
Exchange ENDP
CODE
ENDS
END
ExchangeWords ;if no odd byte, enter loop
aI, es: [di]
;read odd byte from var2
;move a byte from varl to var2
;write var2 byte to varl
[si-l],al
Finis
;done if only 1 byte to exchange
bx,-2
;BX is a handy place to keep -2
ax,es: [di]
[bx] lsi] ,ax
ExchangeLoop
ds,dx
bp
10
iread a word from var2
;do a move from varl to var2
;write var2 word to varl
;repeat "count div 2" times
;get back Turbo's DS
The sample Pascal program that uses Exchange follows:
program TextExchange;
type
EmployeeRecord
=
record
Name
Address
City
State
Zip
.trlng[30] ;
.trlng[ 30];
.tring[15] ;
.tring[2] ;
.tring[lO] ;
end;
var
OldEmployee, NewEmployee
EmployeeRecord;
{$F+}
procedure Exchange(var Varl,Var2; Count
{$L XCHANGE.OBJ}
{$F-}
begin
with OldEmployee do
begin
Name := 'John Smith';
Address := '123 F Street';
City := 'Scotts Valley':
State := 'CA';
Zip := '90000-0000';
Word); external;
end;
with NewEmployee do
begin
338
Turbo Assembler User's Guide
Name := 'Mary Jones'i
Address := '9471 41st Avenue'i
City := 'New York';
State := 'NY';
Zip := '10000-1111'i
endi
Writeln('Before: ' ,OldEmployee.Name,' , ,NewEmployee.Name);
Exchange(OldEmployee,NewEmployee,sizeof(OldEmployee));
Writeln('After: ',OldEmployee.Name,' , ,NewEmployee.Name)i
Exchange(OldEmployee,NewEmployee,sizeof(OldEmployee))i
Writeln('After: ' ,OldEmployee.Name,' , ,NewEmployee.Name)i
end.
To build and run the example Pascal and assembler program, use
the following batch file commands:
TASM XCHANGE
TPC XCHANGE
XCHANGE
Using the .MODEL directive, the Exchange assembly language
program would be written as
.MODEL TPASCAL
.CODE
Exchange PROC FAR var1:DWORD,var2:DWORD,count:WORD
PUBLIC Exchangei
cld
iexchange goes upward
mov
dx,ds
isave DS
Ids
si,var1
iget first address
les
di,var2
iget second address
mov
cx,count
iget number of bytes to move
shr
cx,l
iget word count (low bit -) carry)
jnc
ExchangeWords iif no odd byte, enter loop
mov
al,es:[di]
;read odd byte from var2
movsb
;move a byte from var1 to var2
mov
[si-1],al
iwrite var2 byte to var1
idone if only 1 byte to exchange
jz
Finis
ExchangeWords:
mov
iBX is a handy place to keep -2
bx,-2
ExchangeLoop:
iread a word from var2
mov
ax,es: [di]
movsw
ido a move from var1 to var2
mov
[bx] [si],ax
iwrite var2 word to var1
loop
ExchangeLoop irepeat "count div 2" times
Finis:
mov
ds,dx
iget back Turbo's DS
ret
Exchange ENDP
CODE
ENDS
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
339
END
You can use the same sample Pascal program and just assemble
the alternative Exchange, recompiling the sample program with
the same batch file commands.
Scanning the
DOS environment
With the EnvString function, you can scan the DOS environment
for a string of the form s=SOMESTRING and return
SOMESTRING if it is found.
DATA
DATA
CODE
SEGMENT PUBLIC
EXTRN prefixSeg WORD igives location of PSP
ENDS
SEGMENT PUBLIC'
ASSUME cs:CODE,ds:DATA
EnvString PROC FAR
PUBLIC EnvString
push
bp
cld
mov
es, [prefixSeg]
mov
es,es: [2Ch]
di,di
xor
mov
bp,sp
Ids
si,ss: [bp+6]
;work upward
ilook at PSP
iES:DI points at environment
;which is paragraph-aligned
;find the parameter address
;which is right above the
; return address
ASSUME
lodsb
or
jz
mov
mov
ds:NOTHING
al,al
RetNul
ah,al
dx,si
xor
al,al
mov
;we want ch=O for next count,
; if any
;get back pointer to
si,dx
; string sought
;get length
cl,ah
si,dx
;get pointer to string sought
cmpsb
;compare bytes
Skip
;if fails, try next string
byte ptr es:[di],'='
;compare succeeded; is next
; char '='?
NoEqual
; if not, still no match
;look at length
; is it zero?
;if so, return
;otherwise, save in AH
;DS:DX contains pointer
; to first parm char
;make a zero
Compare:
mov
mov
mov
repe
jne
cmp
jne
340
ch,al
Turbo Assembler User's Guide
Found:
mov
aX,es
mov
mov
inc
les
mov
inc
mov
ds,ax
si,pi
si
bx, ss: [bp+10]
di,bx
di
cl,255
imake DS:SI point to string
i we found
iget
iget
iPut
iget
iset
past the equal (=) sign
address of function result
it in ES:DI
past the length byte
up a maximum length
CopyLoop:
Done:
lodsb
or
jz
stosb
loop
not
al,al
Done
CopyLoop
cl
iget a byte
izero test
iif zero, we're done
iPut it in the result
imove up to 255 bytes
iwe've been decrementing CL
i from 255 during save
isave the length
mov
mov
mov
ASSUME
pop
ret
ASSUME
es: [bx],cl
ax,SEG DATA
ds,ax
ds:DATA
bp
dec
di
mov
sub
jbe
repne
jcxz
cmp
jne
cx, 7FFFh
isearch a long way if necessary
cx,di
ienvironment never >32K
RetNul
iif we're past end, leave
scasb
ilook for the next null
RetNul
iexit if not found
byte ptr es:[di],al isecond null in a row?
Compare
iif not, try again
irestore DS
4
ds:NOTHING
Skip:
icheck for null from this
i character on
NoEqual:
RetNul:
les
stosb
mov
mov
ASSUME
pop
ret
EnvString ENDP
CODE
ENDS
END
di,ss: [bp+l0]
ax,SEG DATA
ds,ax
ds:DATA
bp
iget address of result
istore a zero there
irestore DS
4
The sample Pascal program that uses EnvString follows:
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
341
prograa EnvTest;
( program looks for environment strings
var
EnvVariable : stringi
EnvValue : stringi
($F+)
function
EnvString{s:.t~)
.tringi exteInal;
($L ENVSTR.OBJ)
($F-)
begin
EnvVariable := 'PROMPT'i
EnvValue := EnvString(EnvVariable)i
if EnvValue=', then EnvValue := '*** not found ***'i
Writeln('Environment Variable:' ,EnvVariable, 'Value:' ,EnvValue);
end.
To build and run the example Pascal and assembler program, use
the following batch file commands:
TASM ENVSTR
TPC ENVTEST
ENVTEST
If you used the .MODEL directive, the EnvString assembly
language program would be written like this:
.MODEL TPASCAL
•DATA
EXTRN prefixSeg : WORD
.CODE
EnvString PROC FAR EnvVar:DWORD
PUBLIC EnvString
cld
mov
es, [prefixSeg]
mov
es, es: [2Ch]
xor
di,di
mov
bp,sp
Ids
si,EnvVar
;gives location of PSP
RETURNS EnvVal:DWORD
;work upward
ilook at PSP
;ES:DI points at environment
;which is paragraph-aligned
;find the parameter address
;which is right above the
; return address
ds:NOTHING
ASSUME
lodsb
or
jz
mov
mov
al,al
RetNul
ah,al
dx,si
xor
al,al
ilook at length
iis it zero?
;if so, return
iotherwise, save in AH
iDS:DX contains pointer to
i first parm character
imake a zero
Compare:
342
Turbo Assembler User's Guide
mov
mov
mov
mov
repe
jne
cmp
jne
ch,al
iwe want ch=O for next count, if any
si,dx
iget back pointer to string sought
cl,ah
iget length
si,dx
iget pointer to string sought
cmpsb
icompare bytes
Skip
iif compare fails, try next string
byte ptr es:[di),'='
icompare succeededi is next char '='
NoEqual
iif not, still no match
mov
mov
mov
inc
les
mov
inc
mov
aX,es
imake DS:SI point to string we found
ds,ax
si,di
si
iget past the equal (=) sign
bx,EnvVal iget address of function result
di,bx
iput it in ES:DI
iget past the length byte
di
cl,255
iset up a maximum length
Found:
CopyLoop:
Done:
lodsb
or
jz
stosb
loop
not
iget a byte
izero test
iif zero, we're done
iput it in the result
Copy Loop imove up to 255 bytes
cl
;we've been decrementing CL from
255 during save
mov
es:[bx),cl
isave the length
mov
ax,SEG DATA
mov
ds,ax
irestore DS
ASSUME ds:DATA
ret
ASSUME ds:NOTHING
al,al
Done
Skip:
dec
di
mov
sub
jbe
repne
jcxz
cmp
jne
cx,7FFFh isearch a long way if necessary
ienvironment never >32K
cx,di
RetNul
;if we're past end, leave
;look for the next null
scasb
RetNul
iexit if not found
byte ptr es: [di),al ;second null in a row?
Compare
iif not, try again
les
stosb
mov
mov
ASSUME
ret
di,EnvVal
;check for null from this char on
NoEqual:
RetNul:
ax,SEG DATA
. ds,ax
ds:DATA
Chapter 8, Interfacing Turbo Assembler with Turbo Pascal
;get address of result
istore a zero there
irestore DS
343
E~tring
CODE
ENDP
ENDS
END
You can use the same sample Pascal program and just assemble
the alternative EnvString, recompiling the sample program with
the same batch file commands.
344
Turbo Assembler User's Guide
c
H
A
p
T
E
R
9
Advanced programming in Turbo
Assembler
Over the course of the beginning chapters of this manual, we've
covered the essentials of assembler programming, and then some.
Now we're ready to get into several advanced features of Turbo
Assembler.
In this chapter, we'll explore several aspects of assembler
programming that we've only touched on so far, such as segment
override prefixes, macros, the segment directives, and writing
programs that contain multiple code and data segments. We'll
also look at some useful features that you haven't seen before,
including local labels, automatic jump-sizing, forward references,
and the data structure directives.
Segment override prefixes
Most of the time, memory operands specify memory locations in
the segment pointed to by the DS segment register. For example,
the instruction sequence
mov bx,lOh
mov si,S
mov ax, [bx+si+l]
Chapter 9, Advanced programming in Turbo Assembler
345
loads the word stored at offset 16h in the segment pointed to by
DS into AX. Another way to put this is to say that AX is loaded
from the memory address DS:0016.
One exception to the rule of loading from the segment pointed to
by DS is that the STOS and MOVS string instructions write to the
segment pointed to by ES, and the SCAS and CMPS string
instructions take source operands from the segment pointed to by
E5. (One of the source operands to CMPS is in the data segment,
and one is in the extra segment.>
Another exception is that any memory operand involving BP
accesses the segment pointed to by SS. For example,
mov bp,lOOOh
mov aI, [bp+6]
.. ,
loads AL with the contents of memory location S5:1006.
Suppose, however, you'd like to access a location in the CS
segment as a memory operand; that's useful for jump tables,
especially in multisegment programs. Or suppose you'd like to
access a location on the stack with BX, or a location in DS with BP,
or a location in ES with a nonstring instruction. Can you do that?
The answer is yes. You can use segment override prefixes to make
many instructions access the segment of your choice. For
example,
mov bx,lOOh
mov cl,ss:[bx+lOh]
loads CL with the contents of offset 110h in the stack segment,
and
mov bp,200h
mov si,cs:[bp+l]
loads S1 with the contents of offset 201h in the code segment.
Basically, all you need to do to cause a given instruction to access
a segment other than its default segment is put a segment
override prefix-CS:, DS:, ES:, or SS:-in front of the memory
operand for that instru~tion.
346
Turbo Assembler User's GuIde
Incidentally, segment override prefixes aren't called "prefixes"
because they prefix memory operands in the instruction line.
Rather, a segment override prefix is actually an instruction prefix
byte, which modifies the operation of the instruction that follows
it, just as the REP prefix that we discu~sed in Chapter 6 is an
instruction prefix byte. So, for example, when the 8086 encounters
the instruction bytes
AO 00 00
which form the instruction
mov al, [0]
it loads AL with the contents of offset 0 in the data segment.
However, since the value of the ES: segment override prefix is
26h, when the 8086 encounters
26 AO 00 00
which forms the instruction
mov al,es: [0]
it loads AL with the contents of offset 0 in the extra segment, not
the data segment.
An alternate form
Turbo Assembler supports an alternate segment override prefix
form, where you put the segment override prefix on a separate
line. The separate line-segment overrides are SEGCS for a CS:
segment override, SEGOS for a OS: segment override, SEGES for
an ES: segment override, and SEGSS for an 5S: segment override.
Each of these will override the next line of code only, not all
subsequent lines. For example, the following stores OX to offset
999h in the extra segment:
mov si,999h
seges
mov [si],dx
This alternate form is useful for putting segment override prefixes
on instructions that have no operands, such as LOOSB. The
following loads AL from SS:SI:
segss
Chapter 9, Advanced programming In Turbo Assembler
347
lodsb
When segment
override prefixes Segment override prefixes don't work with all instructions. For
don t work . example, string instruction accesses to the extra segment can't be
I
overridden. That is,
lods es:[ByteVar]
is fine, loading AL from ES:SI, but
stos ds:[ByteVar]
can't work. If you do try to override a string instruction access to
the extra segment as shown above, Turbo Assembler will let you
know that's not allowed. However, if you use SEGCS or the like to
create a segment override, Turbo Assembler doesn't know what
instruction you're going to override and so can't generate an error
in such cases. For example,
segds
stosb
won't generate an assembly error, but STOSS will still write to the
extra segment, not the data segment.
Along the same lines, be aware that segment override prefixes can
never affect accesses to the stack. Pushes to the stack always go to
the stack segment, and pops from the stack always come from the
stack segment. For instance, an instruction such as
segcs
push [bx]
uses the segment override prefix to select the segment from which
the value to be pushed should be fetched; that value is written to
offset SP-2 in the stack segment, as always. Likewise, instructions
are always fetched from the segment pointed to byeS.
See Chapfer6 fordefails.
348
You should generally avoid mixing segment override prefixes
with REP prefixes, since problems can result if an instruction
using both overrides is interrupted.
Turbo Assembler User's' GuIde
Accessing
multiple segments
Segment override prefixes are useful whenever you need to access
multiple segments. This necessity can arise, for example, if you
need to access data stored both on the stack and in the data
segment, which commonly occurs when the stack is used for
dynamically allocated variables and the data segment is used for
static variables. Another possibility is that a program simply has
more than 64K of data, so accesses to any of several segments may
be needed at any time.
One particularly useful application for segment override prefixes
occurs when you mix string and nonstring instructions. For
example, suppose that for a given string you want to convert all
characters with values less than 20h to spaces. The following code
uses a segment override prefix to perform that task efficiently:
mov ax,SEG StringToConvert
mov es,ax
mov di,OFFSET StringToConvert
cld
ConvertLoop:
mov al,es:[di]
and al,al
jz ConvertLoopOone
cmp al,20h
jnb SaveChar
mov al,' ,
SaveChar:
stosb
jmp Convert Loop
ConvertLoopOone:
stosb
iES:OI points to the
i string to convert
;make STOSB increment OI
iget the next character
iis it the end of string?
iyes, done
ido we need to convert it?
ino, save it
imake it a space
isave this character and
i point to the next
icheck the next character
iend the string with a zero
Local labels
Local labels-labels with limited scope-are one of the pleasures of
using Turbo Assembler. Let's look at why you might need them.
Chapter 9, Advanced programming in Turbo Assembler
349
Suppose you have several sections of code in a source module
that perform similar functions. For example, consider the
following:
Sub! PROC
sub ax,ax
IntCountLoop:
add ax, [bx]
inc bx
inc bx
loop IntCountLoop
ret
Sub! ENDP
Sub2 PROC
sub ax,ax
mov dx,ax
LongCountLoop:
add ax, [bx]
adc dx, [bx+2]
add bx,4
loop LongCountLoop
ret
Sub2 ENDP
When two sections of code perform similar functions, it often
, follows that they'll contain similar labels. For example, Subl and
Sub2 each contain a label that marks the top of a counting loop.
When there are only a few labels in a whole program, you can
easily make sure that all the labels are different. In large
programs, however, it can become a nuisance. Then, too, it's
common practice to take a subroutine that works, block-copy it
and rename it, and modify it into a new subroutine. The problem
with this is that it's easy to forget to change a label here or there,
causing the new subroutine to jump to a label in the old
subroutine. For example, if you copied and modified Subl to
make Sub2, you could inadvertently end up with
Sub2 PROC
sub ax,ax
mov dx,ax
LongCountLoop:
add ax, [bx]
adc dx,[bx+2]
350
Turbo Assembler User's Guide
add bx,4
loop IntCountLoop
ret
Sub2 ENDP
which would jump to the middle of Subl-with potentially
disastrous results.
What you really need, then, is a type of label that is limited in
scope to a single subroutine, so it won't conflict with labels in
other subroutines.
That's just what local labels are. Local labels, which by default
usually start with two at-signs (@@), are limited in scope to the
range of instructions between two non-local labels. (Non-local
labels are those defined with PROC and labels ending with colons
that don't start with two at-signs.) As far as Turbo Assembler is
concerned, local labels don't even exist outside the range
delimited by the nearest non-local labels.
Symbols that you define with the LABEL directive do not cause a
new local symbol block to start.
For example, you can use local labels to change the code at the
beginning of this section with
LOCALS
Subl PROC
sub ax,ax
@@CountLoop:
add ax, [bx]
inc bx
inc bx
loop @@CountLoop
ret
Subl ENDP
Sub2 PROC
sub ax,ax
mov dx,ax
@@CountLoop:
add ax, [bx]
adc dx, [bx+2]
add bx,4
loop @@CountLoop
ret
Sub2 ENDP
Chapter 9, Advanced programming in Turbo Assembler
351
Here you need not worry about the loop label in one subroutine
conflicting with the label in the other subroutine, and there's no
chance that one subroutine will accidentally jump to a label in the
other subroutine.
You'll note that we used the LOCALS directive before we used
any local labels. In MASM mode, local labels are disabled by
default, and must be enabled with LOCALS before you can use
them. In Ideal mode, local labels are normally enabled, although
you can disable them with NOLOCALS if you want.
Local labels are also useful when you've got several short
conditional jumps in a subroutine, and you don't want to have to
spend time thinking of unique names for them. For example, you
might want to use local labels when you're testing for any of
several values:
LOCALS
crnp aI,' A'
jnz @@Pl
jrnp HandleA
@@Pl:
crnp al,'B'
jnz @@P2
jrnp HandleB
@@P2:
crnp al,'C'
jnz @@P3
jrnp HandleC
@@P3:
With local labels, you don't have to worry about whether labels
like Pl are used elsewhere in the program.
Remember, any non-local label delimits the scope of a local label.
For instance, the following wouldn't assemble:
Subl PROC NEAR
LOCALS
@@CountLoop:
add ax, [bx]
jnz NotZero
inc dx
352
Turbo Assembler User's Guide
NotZero:
inc bx
inc bx
loop @@CountLoop
The problem here is that the non-local label NotZero lies between
the LOOP instruction's reference to the local label @@CountLoop
and the definition of @@CountLoop. The scope of a local variable
extends only to the nearest non-local label, so when Turbo
Assembler assembles the LOOP instruction, the local label
@@CountLoop is nowhere to be found.
You can change the local symbol prefix from the normal two atsigns (00) to any other two characters that can be used at the start
of a symbol name. You do this by putting the new prefix
characters as an argument to the LOCALS directive:
LOCALS
This sets the local symbol prefix to two underscore characters.
This can be useful if you want to start using local symbols in a
module that already has symbols that start with the default local
symbol prefix.
When you change the local symbol prefix in this manner, local
symbols are automatically enabled at the same time, exactly as if
you had used the LOCALS directive without any argument. If you
subsequently use the NOLOCALS directive to disable local
symbols, Turbo Assembler also remembers the prefix characters
that you specified. This lets you simply use LOCALS with no
arguments to restore local symbols with the prefix you previously
specified.
Automatic jump-sizing
Many years ago, the designers of the 8086 decided that the
conditional jump instructions would only support I-byte jump
displacements. This meant that each conditional jump would only
be capable of jumping to a destination within about 128 bytes of
the conditional jump instruction itself. .
Today, of course, those conditional jumps are with us still, and
they're both a blessing and a curse. While the 8086's conditional
jump instructions sometimes make for compact code (since the
conditional jump instructions are only 2 bytes long), they also
Chapter 9, Advanced programming in Turbo Assembler
353
often make for awkward, inefficient code, since 5-byte instruction
sequences like this
jnz NotZero
jrnp IsZero
NotZero:
are required when conditional jump destinations are too far away
to reach with a I-byte displacement.
Worse, there's no way to know beforehand whether a given
conditional jump will reach a given label, so you're put in the
position of trying to jump to the label directly, thereby risking an
assembly error, or coding a conditional jump around an
unconditional jump, thereby possibly wasting 3 bytes and
slowing execution. Stin more annoying is the all-too-common
occurrence of a "Relative jump out of range" error when you add
an instruction or two inside a loop.
While Turbo Assembler can't solve all the conditional-jump
problems of the 8086, it comes close by way of the JUMPS
directive. Once you've specified JUMPS, Turbo Assembler
automatically turns normal conditional jumps into conditional
jumps around unconditional jumps whenever that's what it takes
to reach the destina tion label.
How does automatic jump-sizing work? Consider the following
code:
JUMPS
RepeatLoop:
jrnp SkipOverData
DB lOOh DUP (?)
SkipOverData:
dec
dx
jnz RepeatLoop
Clearly, the JNZ at the bottom of the loop can't reach RepeatLoop,
. since over 256 bytes lie between the two. Since JUMPS was
specified, however, no assembly-time error will result. Instead,
Turbo Assembler actually assembles this code into the equivalent
of
354
Turbo Assembler User's Guide
RepeatLoop:
jrnp SkipOverData
DB lOOh DUP (?)
SkipOverData:
iternporary data storage in CS
dec dx
jz $+5
jrnp RepeatLoop
automatically using a JZ and a JMP in place of the JNZ at the
bottom of the loop.
Turbo Assembler doesn't always generate a conditional/
unconditional jump pair when JUMPS is active; the conditional
jump you specify is always used if it will reach the destination.
For instance, the following assembles with JNZ at the bottom of
the loop, since here the destination label is near enough to reach
with a I-byte displacement:
JUMPS
RepeatLoop:
add BYTE PTR [bx],l
inc bx
dec dx
jnz RepeatLoop
As we mentioned earlier, Turbo Assembler's automatic jumpsizing doesn't solve all the 8086's problems with conditional
jumps. Turbo Assembler always handles automatic sizing of
backward jumps (jumps to labels earlier in the code than a given
jump instruction) perfectly, with nary a wasted byte or
instruction.
Since Turbo Assembler normally functions as a single-pass
assembler, a compromise is required when automatically sizing
forward jumps. The good news is that forward conditional jumps
to near labels always assemble if automatic jump-sizing is
enabled; the bad news is that several extra NOP instructions are
inserted if it turns out that a conditional jump could have reached
the destination label after all. You can avoid this problem by
using Turbo Assembler's multiple-pass capability (invoked with
the 1m command-line switch), although this does slow assembly
speed slightly.
Chapter 9, Advanced programming In Turbo Assembler
355
A moment's thought will make it clear why automatic sizing of
forward jumps with a single pass can't always generate optimal
code. When Turbo Assembler reaches a conditional jump
instruction that makes a forward reference, there's no way to
know how far away that label is; after all, Turbo Assembler hasn't
even encountered that label yet. With automatic jump-sizing
enabled, Turbo Assembler would like to generate a conditional
jump (a 2-byte instruction) if the destination is near enough to
read directly, and a conditional jump around an unconditional
jump (a 2-byte instruction followed by a 3-byte instruction)
otherwise. Unfortunately, Turbo Assembler doesn't yet know
whether a 2-byte instruction br a 5...byie pair of instructions is
necessary when it encounters a conditiorial forward jump.
Still~
Turbo Assembler has to pick some siZe right away, in order to
know where to assemble the following instructions. Consequently, Turbo Assembler ha.s no alternative but to make the safe choice
and reserve 5 bytes for a coriditional/unconditional jump pair.
Then, if Turbo Assembler later reaches the destination label and
decides that a 2-byte instruction will do the trick, it will assemble
a conditional jump, followed by three NOP instructions that fill
out the 5 reserved bytes.
Suppose Turbo Assembler is assembling the following:
JUMPS
jzDestLabel
inc ax
If JZ can't reach DestLabel directly, Turbo Assembler assembles
the equivalent of the following:
jnz $+5
jrnp DestLabel
inc ax
;2 bytes long
;3 bytes long
If, on the other hand, JZ can reach DestLabel directly, Turbo
Assembler assembles the following:
jz DestLabel
nop
nop
nop
356
;2 bytes long
;each nop is 1 byte long
Turbo Assembler User's Guide
inc ax
The key here is that Turbo Assembler must take up 5 bytes for
each automatically sized forward conditional jump, so three NOP
instructions are inserted in automatically sized forward
conditional jumps that can reach their destinations. Those three
NOP instructions take up space and take time to execute (3 cycles
each on an 8086). Consequently, you're best advised to use
automatically sized forward conditional jumps sparingly, or else
enable Turbo Assembler's multi-pass capability, whenever you're
particularly sensitive to code size and performance issues.
NOJUMPS is always selected
at the start of assembly,' if
you want to use automatic
jump-sizing, you must
explicitly enable it with the
JUMPS directive.
If you're writing a program containing high-performance code,
you might want to enable automatic jump-sizing for noncritical
sections of your program, but disable automatic jump-sizing in
the key code sections. Alternatively, you might want to enable
automatic jump-sizing for backward jumps but disable it for
forward jumps. You can do this by pairing the JUMPS instruction
with the NOJUMPS instruction, which turns off automatic jumpsizing.
For example, the following uses automatic jump-sizing for the
backward jump, but not for the forward jump:
LoopTop:
lodsb
cmp al,80h
NOJUMPS
jb SaveByteValue
neg al
SaveByteValue:
stosb
dec dx
JUMPS
jnz LoopTop
Here, we've directly specified a 2-byte conditional jump for the
forward jump to SaveByte Value, but let Turbo Assembler select the
best code for the backward jump to LoopTop.
Chapter 9, Advanced programming in Turbo Assembler
357
Forward references to code and data
1111.,
In the last section, you saw an example of how forward
conditional jumps can make Turbo Assembler generate less
efficient code when automatic jump-sizing is enabled without
performing multiple passes. The truth of the matter is that all
sorts of forward references can cause problems for Turbo
Assembler, so you should avoid forward references-that is,
references to labels farther on in the code-whenever possible.
Why? Well, as Turbo Assembler assembles a source module, it
makes a single pass through the code, progressing steadily from
the first line in the source module to the last. This means that
Turbo Assembler assembles the first line in a module, then the
second line, then the third line, and so on. While that may seem
obvious, the implication of the order in which Turbo Assembler
assembles lines may be less obvious: Turbo Assembler doesn't
know anything about a line until it reaches it, and so forward
references force Turbo Assembler to make assumptions, which
might turn out to be incorrect. If those assumptions are indeed
incorrect, Turbo Assembler might generate less than maximally
efficient code. Even if Turbo Assembler can generate efficient
code, it might be necessary to go back to earlier lines and make
corrections, and so assembly might take more time than it
otherwise would.
Consider the following:
jrnp DestLabel
DestLabel:
When Turbo Assembler encounters the line
jrnp DestLabel
it hasn't reached the definition of the label DestLabel yet;
consequently, Turbo Assembler has no idea whether DestLabel is
near or far, and, if it's near, whether it can be reached with a I-byte
displacement or whether a full2-byte displacement is needed.
358
Turbo Assembler User's Guide
Consequently, Turbo Assembler needs to make an assumption
about the nature of DestLabel in order to continue assembling.
Turbo Assembler could assume that DestLabel is far and reserve 5
bytes for a far JMP instruction; however, most jumps are 3-byte
near jumps, and it would be a shame to waste 2 bytes on every
forward-referenced near jump. At the opposite end of the
spectrum, Turbo Assembler could assume DestLabel can be
reached with a single-byte displacement and reserve just 2 bytes
for a JMP SHORT instruction; the problem here is that many
jumps are not short, and if Turbo Assembler reserved only 2
bytes, an error would occur if the jump proved to be either near
or far.
As a compromise, Turbo Assembler assumes that all forward
jumps are near, unless you specify otherwise with either the
SHORT or the FAR PTR operator. Three bytes are always
reserved for forward jumps. If a forward jump turns out to be far,
an error results; you must always use FAR PTR to allow forward
jumps to far labels to assemble. That's a bit of a nuisance, but if
you forget the FAR PTR, Turbo Assembler will simply inform
you that a data type override is required, and you can insert the
required FA~ PTR operator and reassemble.
If, on the other hand, a forward jump proves to be short, Turbo
Assembler assembles a short jump, but inserts a NOP instruction
to pad out the 3 bytes that were reserved for the jump, thereby
wasting a byte. For example, Turbo Assembler assembles this:
jrnp DestLabe'l
DestLabel:
into this:
jrnp SHORT DestLabel
nop
DestLabel:
Chapter 9, Advanced programming in Turbo Assembler
359
While the jump works perfectly well, and executes quickly, it is
larger than it needs to be. Of course, you can use the SHORT
operator to turn any forward-referenced jump into a true 2-byte
instruction, but that's not as convenient as if Turbo Assembler
:were able to ge~erate the appropriate jump automatically.
It's important to understand that it's the forward reference that's
the culprit here. If Turbo Assembler knew the distance to the
destination label, the most efficient jump could be assembled. But
with forward references, Turbo Assembler can't know the
distance to the destination until it reaches it, and it can't reach the
destination until it assembles the forward-referenced jump. Turbo
Assembler resolves this dilemma by making a simplifying
assumption that allows assembly to proceed, but at the possible
cost of larger code than is necessary.
1111"
Whenever Turbo Assembler does know the type of a jumpSHORT, NEAR, or FAR-the most efficient possible code can be
generated. Consequently, it's a good idea to use the SHORT
operator on short forward jumps (and, of course, FAR PTR is
required for far forward jumps).
Jumps aren't the only instructions that you should avoid using
with forward referencesj forward references to data can easily
generate inefficient code as well. Consider the following:
.CODE
. ..
mov bl,Value
Value
EQU 1
When Turbo Assembler reaches the MOV instruction, there's no
way to know whether Value is an equated label or a memory
variable. If Value is a memory variable, a 4-byte instruction will be
required, while if Value is an equated label (one that's used as a
constant), a 2-byte instruction will do the job.
As usual, Turbo Assembler must assume the worst in order to
continue assembling, so 4 bytes are reserved for the MOV
360
Turbo Assembler User's Guide
instruction. Then, when Value is reached and discovered to be an
equated label rather than a memory variable, Turbo Assembler
must go back to the MOV instruction and make it a 2-byte
instruction with a constant operand, and must insert two NOP
instructions to fill out the third and fourth bytes that were
reserved. Note that none of this would have happened if Value
had been defined before the MOV instruction, since Turbo
Assembler would have known that Value wasn't a memory
variable.
In fact, backward references present none of the problems of
forward references, since Turbo Assembler always knows
everything there is to know about backward-referenced labels. As
a result, Turbo Assembler always automatically assembles the
most efficient possible code for instructions that involve only
backward-referenced operands. This makes it highly desirable to
avoid forward references whenever possible.
You might wonder if the forward-referencing problems with calls
are as severe as they are for jumps. The answer is no. Forwardreferenced far calls must have FAR PTR type overrides, since
Turbo Assembler assumes forward calls are near. Since there is no
such thing as a short call; inefficient code for calls is never
generated.
Many forward references result in an assembly error rather than
inefficient code. For example, forward references to equated labels
can't be assembled, and forward references to far labels can't be
assembled without a type override.
This is true even if you use a
type override on forward
references to that label.
Even when Turbo Assembler can generate efficient code for
forward references, assembly is slower than for backward
references. This happens because whenever it encounters a label
that has previously been forward-referenced, Turbo Assembler·
must return to each instruction that performed a forward
reference to that label and assemble it properly, now that the
value and type of that label are known.
The conclusion is clear: Avoid forward references in your code
whenever possible, to let Turbo Assembler generate the best
possible code as quickly as possible. If you force multiple passes
to be performed by using the 1m command-line switch, optimal
code will be genera ted, but the assembly process will take longer
than it would have with a single pass. Put data definitions at the
beginning of your source modules before the code that references
them. When you can't avoid forward references, always use a
Chapter 9, Advanced programming in Turbo Assembler
361
type override operator to let Turbo Assembler know exactly what
type of label you're working with.
Using repeat blocks and macros
One of the things a computer does well is repetitive work. You
might get bored with typing dozens of values for DB directives, or
with entering slight variations on the same code over and over,
but your computer will never tire of such work. Turbo Assembler
provides repeat blocks and macros to free you from just that sort
of monotonous work.
Repeat blocks
A repeat block starts with the REPT directive and ends with the
ENDM directive. The code within the repeat block is assembled
the number of times specified by the operand to the REPT
directive. For example,
REPT 10
OW
0
ENOM
generates the same code as
ow 0
ow 0
ow 0
ow 0
ow 0
ow 0
ow 0
ow 0
ow 0
ow . 0
That doesn't seem earthshaking, particularly given that
ow
10 OUP (0)
does the same thing, but now let's combine repeat blocks and the
= directive to make a table of the first ten integers:
362
Turbo Assembler User's Guide
IntVal
0
REPT 10
DW IntVal
IntVa1+1
IntVal
ENDM
This generates the equivalent of
DW
DW
DW
DW
DW
DW
DW
DW
DW
DW
0
1
2
3
5
6
8
9
Try doing that with CUP! Better yet, if you want the first 100
integers, all you need do is change the operand to REPT to 100;
that's certainly a lot easier than typing 100 lines.
One excellent application for REPT is in the generation of tables
used for fast multiplication and division. For example, the
following multiplies a number between 0 and 99 (stored in BX) by
10-very rapidly-and places the result in AX.
•DATA
TableOfMultiplesOf10
LABEL
BaseVal
0
REPT 100
DW BaseVal
BaseVal
BaseVal+10
ENDM
WORD
•CODE
shl bx,l iprepare for look up in
; table of word-sized entries
mov ax, [TableOfMultiplesOf10+bx] ;look up the result of
; mUltiplication times 10
Keep in mind that the text in a repeat block is simply assembled
as many times as the operand to REPT dictates. There's no
Chapter 9, Advanced programming in Turbo Assembler
363
difference between executing a repeat block 10 times and making
9 additional copies of the code in a repeat block and then
assembling all 10 instances of the code.
This means that any valid assembler code, including instructions,
can be placed within a repeat block. For example, the following
generates code to divide the 32-bit unsigned value in DX:AX by
16:
REPT 4
shr dx,1
rer ax,1
ENDM
Repeat blocks can be nested. For instance, the following generates
10 NOP instructions:
REPT 5
REPT 2
nop
ENDM
ENDM
Repeat blocks and
variable parameters
IRP and IRPC provide two means by which to provide a variable
parameter to each pass of a repeat block.
IRP substitutes the first entry in a list for a parameter on the first
repetition of a repeat block, the second entry on the second
repetition, and so on until the list is used up. For example,
IRP PARM,
DB PARM
ENDM
generates
DB
DB
DB
DB
DB
DB
364
0
1
9
16
25
Turbo Assembler User's Guide
IRPC is similar, save that it substitutes one character from a string
on each repetition of a repeat block. The following code sets the
zero flag if AL is equal to any of the characters in the string that's
the second argument to IRPC:
IRPC TEST_CHAR,azklg
cmp al,'&TEST_CHAR&'
jz
EndCompare
ENDM
EndCompare:
The last example uses the ampersand (&) to force evaluation of
the repeat block parameter TEST_CHAR, even within quotes. The
ampersand is a macro operator that works in a repeat block
because repeat blocks are actually a type of macro. Other macro
features, such as the LOCAL and EXITM directives, also work in
repeat blocks. We'll discuss macros next.
Macros
The basic operation of a macro is quite simple: You assign a name
to a block of text, or a macro; then, when Turbo Assembler
encounters that macro name later in your source code, the block
of text associated with the name is assembled. You might think of
the macro name being expanded into the full text of the macro;
hence the term macro expansion is often used to describe the
substitution of macro text for a macro name.
A useful analogy is an include file. When Turbo Assembler
encounters an INCLUDE directive, the text in the specified file is
immediately assembled, just as if it were in the source module
containing the I~CLUDE. If a second INCLUDE of the same file is
encountered, Turbo Assembler assembles that text again.
Macros are similar to include files in that the text, or body, of the
macro is assembled each time the macro name is encountered.
Macros are actually a great deal more flexible than include files,
however, since they can optionally be passed parameters and can
contain local labels. They are much faster than include files, since
the text of a macro does not have to be read from disk. Let's take
a look at basic macro operation.
The following code uses the macro MULTIPLY_BY_4 to multiply
the value in AX by 4, storing the result in DX:AX:
Chapter 9, Advanced programming in Turbo Assembler
365
MULTIPLY BY 4 MACRO
sub dx,dx
shl ax,l
rcl dx,l
shl ax,l
rcl dx, 1
ENDM
mov ax, [MemVar]
MULTIPLY BY 4
mov WORD PTR [Result],ax
mov WORD PTR [Resultt2],dx
When Turbo Assembler encounters MULTIPLY_BY_4, it
assembles the four instructions that make up the body of that
macro on the spot. It's almost as if a new instruction has been
defined, MULTIPLY_BY_4, which you can use just as you use
MOV and MUL. Of course, that new macro instruction consists of
five 8086 instructions, but it's certainly easier to read the previous
code with the macro than without.
You could just as well have used a subroutine named MultiplyBy4
instead of a macro in this example, as follows:
MultiplyBy4
PROC
sub dx,dx
shl ax,l
rcl dx,l
shl ax,l
rcl dx,l
ret
MultiplyBy4
ENDP
mov
call
mov
mov
In general, you'l/ want to use
subroutines for minimum
code size, and macros for
speed and flexibility.
366
ax, [MemVar]
MultiplyBy4
WORD PTR [Result],ax
WORD PTR [Resultt2],dx
How do you choose between subroutines and macros? Well,
you'll generally produce smaller code by using a subroutine, since
with subroutines the code for a specific task is assembled only
once, with calls to that code sprinkled throughout the program.
However, you'll produce faster code with macros, since macros
avoid the overhead of CALL and RET instructions. Moreover, a
Turbo Assembler User's Guide
single macro can be tailored to generate slightly different code for
a number of similar tasks, while a subroutine can't.
What sort of flexibility does a macro provide? Macro flexibility is
limited only by your imagination, since macros can accept
parameters and can contain conditional assembly directives.
Macro parameters appear as operands to the MACRO directive.
For example, V ALUE and LENGTH are parameters to the macro
FILL_ARRAY, defined as follows:
FILL ARRAY
MACRO
REPT LENGTH
DB VALUE
ENDM
ENDM
VALUE, LENGTH
When a macro is invoked, parameters to the macro can be placed
as operands to the macro invocation. For example, FILL_ARRAY
could be invoked as
ByteArray LABEL
FILL ARRAY
BYTE
2,9
The parameters that appear in the macro invocation (2 and 9 in
the previous code) are known as actual parameters. The parameters
that appear in the macro definition (VALUE and LENGTH in the
preceding code) are known as formal parameters. Each time a
macro is· invoked, the formal parameters are set to the values of
the corresponding actual parameters before the macro is
expanded, so
ByteArray LABEL
FILL ARRAY
BYTE
2,9
causes the following code to assemble:
ByteArray LABEL
REPT 9
DB 2
ENDM
BYTE
Chapter 9, Advanced programming in Turbo Assembler
367
The values of the actual parameters to a macro invocation are
substituted for the formal parameters in the macro definition, so
you can generate different macro code simply by changing the
actual parameters used in a macro invocation. For instance, if you
wanted to initialize ByteArray to be 8 bytes in length, initialized to
OFFh, and ByteArray2 to be lOOh bytes long, initialized to 0, all
you'd need would be
ByteArray LABEL
BYTE
FILL_ARRAY Offh,S
ByteArray2 LABEL
BYTE
FILL_ARRAY O,lOOh
Formal parameters can be used anywhere in a macro. However,
there's a problem when formal parameters are mixed with other
text. For example, in the macro
PUSH WORD REG MACRO
push RLETTERx
ENDM
RLETTER
Turbo Assembler can't know whether the string RLETTER
embedded in RLETTERx is the name of the formal parameter or
part of the operand to PUSH, so it assumes it's part of the operand.
Alas, pushing RLETTERx isn't likely to succeed unless you
happen to have memory variable of that name, and the desired
result of pushing a register wouldn't be achieved in any case.
If such text isn't the name of
a formal parameter, Turbo
Assembler Ignores the
ampersands.
The solution is to enclose the formal parameter name in a pair of
ampersands (&&). When Turbo Assembler encounters macro text
enclosed in ampersands, it checks first to see whether that text is
the name of a formal parameter; if so, it substitutes the value of
tha t parameter.
For example, the following expansion of PUSH_WORD_REG,
PUSH WORD REG
MACRO
push &RLETTER&x
ENDM
RLETTER
PUSH WORD REG b
assembles to
368
Turbo Assembler User's 'Guide
push bx
Ampersands are required only when there might be a question
about a reference to a formal parameter; for example, they're not
needed in
PUSH WORD REG
MACRO
push REGISTER
ENDM
REGISTER
However, it never hurts to use ampersands, so use them
whenever you're in doubt about whether they're needed.
Nesting macros
You've already seen that macros can contain repeat blocks.
Macros can invoke other macros as well; this is known as nesting
macros. For example, in
PUSH WORD REG
MACRO
push REGISTER
ENDM
REGISTER
PUSH ALL REGS
MACRO
IRP REG,
PUSH WORD REG REG
ENDM
ENDM
the macro PUSH_ALL_REGS contains a repeat block, which in
turn contains an invocation of the macro PUSH_WORD_REG.
Macros and
conditionals
Perhaps the most powerful feature of macros is their ability to
contain conditional assembly directives. This allows a single
macro to assemble different sorts of code depending on the state
of equated labels and parameters to each macro invocation.
For example, we'll return to the earlier example of a macro that
performs multiplication. In this case, however, if the factor passed
'. as a parameter to the new MULTIPLY macro is any power of two,
we'll multiply by using the faster shift and rotate instructions;
otherwise, we'll use the MUL instruction. Here's the macro:
MULTIPLY MACRO
i
FACTOR
Check FACTOR against each of the 16 possible powers of two.
Chapter 9, Advanced programming In Turbo Assembler
369
IS- POWER-OF-TWO = 0
COUNT
15
POWER OF TWO
8000h
REPT 16
IF POWER_OF_TWO EO FACTOR
;FACTOR is a power of two
IS POWER OF TWO = 1
EXITM
ENDIF
COUNT
COUNT-1
POWER OF TWO
POWER OF TWO SHR 1
ENDM
IF IS POWER OF TWO
sub dx,dx
REPT COUNT
shl ax,l
re! dx,l
ENDM
ELSE
mov dx,FACTOR
mul dx
ENDIF
ENDM
MULTIPLY actually checks on the fly whether the multiplication
is by a power of two and assembles the appropriate code. So the
code
MULTIPLY 10
assembles to
mov dx,10
mul dx
but the code,
MULTIPLY
8
assembles to
sub
shl
re!
shl
reI
shl
370
dx,dx
ax,l
dx,l
ax,l
dx,l
ax,l
Turbo Assembler User's Guide
reI dx,l
Don't confuse macros with
subroutines, and don't
confuse conditional
assembly with If statements
and the like in high-level
languages.
Stopping expansion
with EXITM
Bear in mind that macros are expanded at assembly time, not at
run-time. MULTIPLY assembles new code each time it is invoked;
the IF directive in MULTIPLY determines which instructions get
assembled.
The next example contains a directive you haven't seen before:
EXITM. The EXITM directive instructs Turbo Assembler to stop
expanding the current macro or repeat block. If, however, the
current macro or repeat block is nested inside another macro or
repeat block, expansion of the nesting macro or repeat block
continues.
Shiftn MACRO OP,N
Count = 0
REPT N
shl OP,N
Count = Count + 1
IF Count GE 8
EXITM
ENDIF
ENDM
Defining labels within
macros
ino more than 8 allowed
One potential problem with macros arises when you define a label
within a macro. For example, the following causes an error due to
the redefinition of SlcipLabel, since each expansion of the macro
DO_DEC defines SkipLabel:
DO DEC
MACRO
jexz SkipLabel
dee ex
SkipLabel:
ENDM
DO DEC
DO DEC
Fortunately, Turbo Assembler provides a simple solution in the
form of the LOCAL directive. A LOCAL directive in a given macro
causes the scope of the specified label or labels to be restricted to
that macro. For example, LOCAL can be used as follows to allow
the last example to assemble:
Chapter 9, Advanced programming in Turbo Assembler
371
DO DEC
MACRO
LOCAL
SkipLabel
jexz SkipLabel
dec ex
SkipLabel:
ENDM
If LOCAL is used in a macro, it must be used immediately
following the MACRO directive. Multiple labels can be declared
local with a single LOCAL directive, and multiple LOCAL
directives can be used:
TEST MACRO
LOCAL
LOCAL
MACRO
LoopTop,LoopEnd,Skiplne
NoEvent,MaeroDone
ENDM
The names actually assigned to local labels are of the form
??XXXX
where XXXX is a hexadecimal number between 0 and OFFFFh.
Consequently, you should not assign your own labels names that
start with ??, since these might conflict with the local labels Turbo
Assembler generates.
Forward references to macros are not allowed; macros must be
defined before they're invoked. This makes good sense, in light of
our earlier discussion of forward references, since Turbo
Assembler has no idea how many bytes it would have to reserve
for a forward-referenced macro. Otherwise, though, macros can
be defined anywhere in a source module.
Any valid assembler line can appear in a macro. This includes
data definition directives, as well as code, and even includes
segment directives, labels of all sorts, and listing control
directives.
Refer to Chapter 6 in this
manual and Chapter 31n the
Reference Guide for
Information about these
directives.
372
There are several conditional assembly directives that are
designed specifically for use in macros; these include IFDIF, IFIDN,
IFDIFI, IFIDNI, IFB, and IFNB. There are also several conditional
error directives for use in macros, including ERRDIF, ERRIDN,
ERRDIFI, ERRIDNI, ERRB, and ERRNB.
Turbo Assembler User's Guide
The special operators are all
defined more fully In Chapter
2 of the Reference Guide.
There are a number of special operators that you can use within
macros:
&
<>
%
"
Substitute operator
Literal text string operator
Quoted character opera tor
Expression evaluate opera tor
Suppressed comment
The & substitution operator has been discussed in the previous
section on macros.
Fancy data structures
Turbo Assembler provides three directives to ease the task of
managing complex data structures: STRUC, RECORD, and UNION.
You've probably noticed that the directive names are similar to
those used by high-level languages, and, indeed, there are some
similarities between Turbo Assembler's data structure directives
and those of high-level languages.
Don't be misled, however; as you will see, assembly language
data structure directives, while helpful, are less sophisticated than
those of high-level languages. For example, assembly language
doesn't limit the scope of the name of a structure element to that
structure, so every structure element name must be unique in its
source module.
Also, unlike C and Pascal, assembly language data structure
directives are conveniences, not necessities; there are ways to
handle data structures, records, and unions in assembler without
using the data structure directives. Nonetheless, the data
structure directives are convenient and well worth knowing
about.
Consult Chapter 11 to learn
more about the enhanced
features of Ideal mode.
The following discussion applies to Turbo Assembler operating in
MASM mode. In Ideal mode, Turbo Assembler supports
considerably more powerful forms of the data structure
directives.
One point about Turbo Assembler's fancy data structures before
we begin: Structures, records, and unions can appear anywhere in
a source module, as long as they are never forward-referenced by
instructions or directives.
Chapter 9, Advanced programming In Turbo Assembler
373
The STRUC
directive
The STRUC directive, which lets you define a data structure, is
useful whenever you have to deal with data that's partitioned into
logical groups. For those of you who are familiar with C, STRUC
is similar to C's struct statement.
For example, suppose you want to define a data structure
containing a name, age, and income for one client. Here's such a
structure:
CLIENT
NAME
AGE
INCOME
CLIENT
STRUC
DB 'Name goes here •••• '
DW
DD
ENDS
The CUENT structure contains three fields: The NAME field,
which contains a name up to 20 characters in length; the AGE
field, which contains an age stored as a 16-bit value; and the
INCOME field, which contains an income stored as a 32-bit value.
You could use the CUENT structure as follows:
CLIENT
NAME
AGE
INCOME
CLIENT
STRUC
DB 'Name goes here ..•• '
DW
DD
ENDS
•DATA
MisterBark
CLIENT
<'John Q. Bark' ,32,10000>
.CODE
mov
mov
mov
mov
ax, [MisterBark.Age]
bx,OFFSET MisterBark
ax, WORD PTR [bx.INCOME]
dx,WORD PTR [bx.INCOME+2]
There's much to examine in this example. First, notice that
structure definitions end with the ENDS directive. This is the
same directive that ends segment definitions. It's all right to nest
structure definitions inside segment definitions. For example, the
folloWing defines a structure inside a data segment:
374
Turbo Assembler User's Guide
Data
SEGMENT
Test
STRUC
Test
ENDS
Data
WORD PUBLIC 'DATA'
ENDS
Second, note that the variable MisterBark of structure type CUENT
is created as if there were a new data type named CUENT, and in
fact that's exactly what you've done by defining the CUENT
structure. In fact, if you use the SIZE operator on a CUENT
structure, you'll get the value 26, which is the size of the structure.
When MisterBark is created, three parameters to the declaration
are provided within angle brackets. These parameters become the
initial values for the corresponding fields of MisterBark; the string
'John Q. Bark' is the initial value of the NAME field, 32 is the
initial value of the AGE field, and 10,000 is the initial value of the
INCOME field.
You need not specify the initial value of any or all of the fields of a
structured variable when you create it. For example,
MisterBark
CLIENT
<>
doesn't initialize any of the fields of MisterBark, and
MisterBark
CLIENT
<,,19757>
initializes only the INCOME field. However, the angle brackets
are required even if no fields are initialized.
If you don't specify an initial value when you create a memory
variable, there are two possible ways in which the initial value of
each field can be set. If you specified a value for a given field
when you defined the structure type, that's the default value
assigned to that field. If you specified a question mark for a given
field when you defined the structure type, the default value for
that field is O.
For example, in the following code, an initial value is specified for
only one field of MisterBark-the NAME field-when MisterBark is
created. However, an initial value is specified for the AGE field
when the CUENT structure is defined, so that's the value assigned
to the AGE field of MisterBark. No value is specified in either place
Chapter 9, Advanced programming in Turbo Assembler
375
for the INCOME field, so the INCOME field is initialized to O.
Here's the example:
CLIENT
NAME
AGE
INCOME
CLIENT
STRUC
DB 'Name goes here •••• '
DW 21
DD
ENDS
•DATA
MisterBark
CLIENT
<'John Q. Bark'>
The result of this code is that the NAME field is initialized to 'John
Q. Bark', the AGE field is initialized to 21, and the INCOME field
is initialized to O. Note that the initial value for the NAME field
specified when MisterBark is created overrides the initial value
specified when the CUENT structure was defined.
You can initialize arrays of structures with the DUP operator. For
example,
Clients
CLIENT
52 DUP
«»
creates the array Clients,·consisting of 52 structures of type
CUENT, each initialized to the default values.
If you look back to the original structure example, you'll see a
new operator there-the period (.) operator. The period operator
is actually just another form of the plus operator for memoryaddressing; that is, all the following lines do exactly the same
thing:
mov
mov
mov
mov
ax, [bx.AGE]
ax, [bx].AGE
ax, [bx+AGE]
ax, [bx]+AtE
The period operator is often used with structure references for
consistency with C notation, which also uses the period operator,
and to make it clear that a structure field is being accessed; you
can use whichever operator you prefer-period or plus.
The structure fields defined with the STRUC directive are actually
labels equated to the offset of the field in the structure. Given the
376
Turbo Assembler User's Guide
earlier definition for CUENT and MisterBark, the following two
lines are equivalent:
mov [MisterBark.AGE] ,ax
mov [MisterBark+20] ,ax
and this would work as well:
AGE_FIELD EQU 20
mov [MisterBark+AGE_FIELD],ax.
Advantages and
disadvantages of using
STRUC
Why use STRUC, then? For one thing, structure fields provide
data-typing; Turbo Assembler knows MisterBark.AGE in the first
example is a word-sized variable, since there AGE is a structure
element, but MisterBark+AGE in the second example has no
inherent size.
For another, it's much easier to change a structure definition than
to change constant offsets, or even a set of equates. For example, if
you decided that the NAME field should be 30 characters long, all
you'd have to do is change the entry for NAME in the CLIENT
definition. !fyou were using equates, you'd have to manually
calculate and change the offsets of both the AGE and INCOME
fields; in a larger structure, you'd have quite a bit of work to do.
Finally, STRUC makes it easy to create and initialize data
structures.
In short, STRUC is a convenient and maintainable way to create
and access data structures. On the other hand, assembler data
structures are by no means as error-proof as C data structures. For
example, when you use a register to point to a data structure,
there's no way for Turbo Assembler to tell whether the register
contains a pointer to a valid data structure of that type. In the
following code, BX is loaded with 0, but there's no way for Turbo
Assembler to know whether or not there's a valid CUENT data
structure at offset 0:
mov bx,O
mov dx, [bx.AGE]
Chapter 9, Advanced programming in Turbo Assembler
377
This is not a problem with assembly language; rather, it reflects
the nature of assembly language. When there's a choice between
letting you have complete freedom in programming and
protecting you from yourself, assembly language gives you the
freedom. The important thing to keep in mind is that Turbo
Assembler can perform only limited error-checking on your
structure references; it's up to you to make sure you've got the
right pointers loaded.
Unique structure field names
One somewhat annoying result of the fact that structure field
names are actually just labels is that structure field names must be
unique in their source module. For example, if you defined the
CUENTstructure in a given source module, you couldn't have a
label named INCOME anywhere else in that module, not even in
another structure. INCOME is just a label with the value 22, and
of course, you can't have two labels with the same name in a
single source module. The following will produce an error, due to
the attempted redefinition of AGE:
CLIENT
NAME
AGE
INCOME
CLIENT
STRUC
DB 'Name goes here .... '
DW?
DD ?
ENDS
AGE EQU 21
Nesting structures
Structures can be nested; for example,
. DATA
AGE STRUC
YEARS
MONTHS
AGE STRUC
CLIENT
NAME
AGE
378
STRUC
DW?
DW ?
ENDS
'STRUC
DB 'Name goes here .... '
AGE STRUC <>
Turbo Assembler User's Guide
INCOME
CLIENT
DW
ENDS
MisterBark
CLIENT
<>
•CODE
mov dx, [MisterBark.AGE.MONTHS]
mov si,OFFSET MisterBark
mov ex, [si.AGE.YEARS]
nests an AGE_STRUC structure named AGE in the CUENT
structure, then references the MONTHS and YEARS fields of AGE
in the CUENT structure MisterBark.
Initializing structures
There are a few cautions regarding the initialization of structures.
First, if you attempt to initialize a string field of a structure with a
string tha t is longer than the field, an assembly error will be
generated.
Second, the only kind of field that can be initialized with a string
value is a string field. The following would not assemble:
TEST STRUC
TEXT DB 30 DUP {' 'l
TEST ENDS
TStrue
TEST <'Test string'>
even though TEXT was initialized to spaces, because Turbo
Assembler considers TEXT to be an array of 30 spaces, not a string
of 30 characters. The following would assemble:
TEST STRUC
TEXT DB 'String goes here •.••••..•••.. '
TEST ENDS
TStrue
TEST <'Test string'>
Third, while you can define more than one data element as
belonging to a single structure field, you can only initialize, at
most, one element per field when you create an instance of that
Chapter 9, Advanced programming in Turbo Assembler
379
structure. For example, in the following code, when TestStruc is
created, the first byte of field A is initialized to 1, and the first byte
of field B is initialized to 2, while the second byte of each field is
initialized to 20h (a space):
T
A
B
T
STRUC
DB Offh,Offh
DB Offh,Offh
ENDS
TestStruc T
Refer to Chapter 11 for
Information about Ideal
mode.
<1,2>
In this section, we've discussed the MASM mode version of the
STRUC directive. In Ideal mode, the STRUC directive is
considerably more powerful, providing more of the features
available to structures in high-level languages.
The RECORD
directive The RECORD directive provides you with a means to define bit
fields within a byte or word. The bit field definitions can then be
used to generate masks to isolate one or more of the bit fields, as
well as shift counts to right-justify any bit field. The RECORD
directive bears no relation to the Pascal record statement.
Suppose you want to define a data structure that contains three
1-bit flags and a 12-bit value. You could do this with the RECORD
directive as follows:
TEST REC RECORD
FLAG1:1,FLAG2:1,FLAG3:1,TVAL:12
This example defines three flags, named FLAG1, FLAG2, and
FLAG3, and a data field named TVAL. The value after the colon
for each field specifies that field's size in bits; each of the flags is
one bit in size, and TVAL is 12 bits in size.
How are the fields stored within the record? That's a bit complex.
The first field, FLAG1, is the leftmost (most significant) bit of the
record. The second field, FLAG2, is the next most significant bit of
the record, and so on, until you reach TVAL, which ends at the
least significant bit of the record. However, the record is only 15
bits in size, leaving one bit in the word unaccounted for. (Records
are always exactly 8 or 16 bits long.) The rule is that records as a
whole are always right-justified in a byte or word.
380
Turbo Assembler User's Guide
As we said, it's a bit complex. Here's an example to clarify things.
A record of type TEST_REC is defined with a line like
TRee TEST_REC <1,O,O,52h>
Here we've created the variable TRee of record type TEST_REC.
The values in the angle brackets are made the initial values of the
corresponding fields, so the FLAGl field of TRee is initialized to 1,
the FLAG2 and FLAG3 fields are initialized to 0, and the TVAL
field is initialized to S2h. Figure 9.1 shows the locations and initial
values of the four fields of TRee.
Figure 9.1
Locations and initial
values of the fields
in TRec
TVAL
l
TRee
52h
15
Bit
14
13
12
11
o
If the overall size of a record (the sum total of all the fields) is 8
bits or less, the record is stored in a byte; otherwise, the record is
stored in a word. Records longer than 16 bits are not supported
except when 80386 assembly is enabled; in which case, records up
to 32 bits in size are allowed.
Initializing a record variable is much like initializing a structure
variable. If you specify an initial value for a given record field
when you create the record variable, the field is initialized to that
value, as illustrated by the last example.
If you don't specify an initial value for a given record field when
you create a record variable, there are two possible default values.
When you create a record type, you can optionally specify a
default value for any or all fields. For example,
TEST REC RECORD
FLAG1:l=1,FLAG2:1=O,FLAG3:1,TVAL:12=Offfh
specifies default values of 1 for FLAG1, 0 for FLAG2, and OFFFh
for TVAL, with no explicit default value for FLAG3. The default
value for any field lacking an explicit default value is 0, so the
default value for FLAG3 is O.
So, given the following definition of TEST_REC and creation of
TRee
. DATA
Chapter 9, Advanced programming in Turbo Assembler
381
TEST REC RECORD
FLAG1:1=1,FLAG2:1=0, FLAG3: 1, TVAL: 12=Offfh
TRee TEST_REC <,1,,2>
the fields are initialized as follows:
•
•
•
•
FLAGI initialized to 1
FLAG2 initialized to 1
FLAG3 initialized to 0
TVAL initialized to 2
The overall value of the record variable TRee is 6002h. Note that
initial values specified when a record variable is created override
initial values specified when the record type is defined.
Once defined, a record type is much like any other data type. You
can, for example, use a record type with the SIZE operator, and
you can define arrays of records with the DUP operator. For
example, the following declares an array of 90 records of type
TEST_REe:
TReeArray TEST_REC 90 DUP «1,1,1,0»
As with STRUC field names, record field names are labels. Since
labels can only be defined once in a source module, this means
that all record field names must be unique within their source
module.
Accessing records
Now that you know how to create a record and how the various
fields in a record are stored, you're ready to learn how to access
records. You might reasonably think that you could access record
fields the way you access structure fields, as in
mov aI, [TRee.FLAG2]
;this doesn't work!!!
but that's not the case. The 8086 can only work with 8- or 16-bit
wide memory operands, so there's no way to load a 1-bit field, for
instance, into a register. What you can do with record fields is
determine their size in bytes, determine how many bits they need
to be shifted to be right-justified, and generate masks to isolate
them. In other words, even though the 8086 doesn't let you work
directly with record fields, Turbo Assembler supports
manipulating record fields with instructions such as AND and
SHR.
382
Turbo Assembler User's Guide
The value of a given record field is the number of bits by which
you'd have to shift the record in order to right-justify that field
(that is, place bit of the field at bit of the record). For instance,
°
°
mov al,FLAG1
mov ah,TVAL
loads AL with 14 and AH with 0, so
mov ax, [TRee]
mov el,FLAG1
shr ax,el
right-justifies the FLAGl field of TRee in AX.
The value of a given record type itself is the byte or word value
that would be generated by creating a record with given initial
values. For example,
mov ax,TEST_REC <1,1,1,Offfh>
loads AX with 7FFFh, the value you'd get if you created a
TEST_REC type record with the initial values . Bear
iIi mind the distinction between loading AX with the record type
TEST_REC, as in the last example, and loading AX with the
record variable TRee, as in
TEST REC RECORD
FLAG1:1=1,FLAG2:1=O,FLAG3:1,TVAL:12=Offfh
TRee TEST REC <,1,,2>
.CODE
mov ax, [TRee]
which loads AX with 6002h, the value of the variable TRee.
The WIDTH operator
The WIDTH operator returns the size of a record or record field in
bits. For example, the following line stores 15, the number of bits
in a TEST_REC record, in AL:
mov aI, WIDTH TEST_REC
;size of a TEST_REC record in bits
Chapter 9, Advanced programming in Turbo Assembler
383
and the following stores 1, the width of each of the flag fields, in
AL, AH, and BL, and 12, the width of the TVAL field, in BH:
mov
mov
rnov
rnov
aI, WIDTH
ah,WIDTH
bI,WIDTH
bh,WIDTH
FLAG!
FLAG2
FLAG3
TVAL
The MASK operator
Finally, the MASK operator returns a mask sUitable for isolating a
record or record field with the AND instruction. For example,
rnov ax,MASK TEST_REC
stores 7FFFh in AX, and
rnov ax,MASK TEST_REC
rnov dx, [TRee]
and dx,ax
stores the value of the record TRee in DX, masking off bit 15,
which isn't part of the TEST_REC record.
MASK is more useful for isolating an individual record field. The
following detects whether the FLAG3 field of TRee is set:
rnov ax, [TRee]
and ax, MASK FLAG3
jz FIag3NotSet
Note that the TEST instruction can be used non-destructively in
place of AND; the following performs the same test as the
previous example without modifying any registers or memory
locations:
jz
FIag3NotSet
The MASK operator is also useful for manipulating record fields
in conjunction with the shift instructions, as you'll see shortly.
384
Turbo Assembler User's Guide
Why use records
Now you've seen what records are and how they're used. When
would you really want to use records? Well, records aren't used
all that often, but they are handy when you've got multiple data
fields encoded in a single byte or word. Some variables used by
the BIOS are structured as records. For example, the low byte of
the BIOS equipment flag variable, which stores equipment-related
information (such as what video adapter is active and the number
of floppy drives present) is a record of the structure
EQ_FLAG RECORD NUMDISKS:2,VIDEO:2,RSRVD:2,MATHCHIP:l,AREDISKS:l
where NUMDISKS is the number of floppy disk drives installed
minus 1; VIDEO indicates what sort of display adapter is
currently active; RSRVD is a field reserved for different uses in
different IBM microcomputers; MATHCHIP is 1 if a numeric
coprocessor such as an 8087 is installed; ARE DISKS is 1 if any
floppy disk drives are installed.
Here's a function that uses the EQ_FLAG record and the record
operators to return the setting of the display adapter field of the
BIOS equipment flag variable:
i
i
Returns current setting of the display adapter field of
the BIOS equipment flag variable.
i Input: None
i Output:
AL = 0
1
2
3
if
if
if
if
no display adapter is currently selected
40x25 color display is currently selected
80x25 color display is currently selected
80x25 monochrome display is currently selected
i Registers destroyed: AX,CL,ES
EQ_FLAG RECORD NUMDISKS:2,VIDEO:2,RSRVD:2,MATHCHIP:l,AREDISKS:l
GetBIOSEquipmentFlag
PROC
mov ax,40h
mov es,ax
ipoint ES to the BIOS data segment
mov al,es:[lOh]
iget the low bit of the equipment flag
and al,MASK VIDEO iisolate the display adapter field
mov cl, VIDEO
iget the number of bits to shift
i the display adapter field right to
i right-justify it
Chapter 9, Advanced programming in Turbo Assembler
385
shr al,cl
ret
GetBIOSEquipmentFlag
iright-justify display adapter field
ENDP
Here's a complementary function that sets the display adapter
field of the BIOS equipment flag to a specified value:
Sets the display adapter field of the BIOS equipment flag
variable.
Input:
AL = 0
1
2
3
if
if
if
if
no display adapter is currently selected
40x25 color display is currently selected
80x25 color display is currently selected
80x25 monochrome display is currently selected
Output: None
Registers destroyed: AX,CX,ES
EQ_FLAG RECORD NUMDISKS:2,VIDEO:2,RSRVD:2,MATHCHIP:l,AREDISKS:l
SetBIOSEquipmentFlag
mov cx,40h
mov es,cx
mov el, VIDEO
PROC
shl
mov
and
and
al,e!
ah,es: [lOh]
ah,NOT MASK VIDEO
aI, MASK VIDEO
or
al,ah
rnov es:[lOh],al
ret
SetBIOSEquipmentFlag ENDP
See Chapter 11 for
information about Ideal
mode.
386
ipoint ES to the BIOS data segment
iget the number of bits to shift
i the passed value left to align it
i with the display adapter field
ialign the value
iget the low bit of equipment flag
iclear the display adapter field
imake sure the new display adapter
i field setting is valid
iinsert the new display adapter
i field setting in the equipment
i flag value
iset the new equipment flag
In this section, we've discussed the MASM mode version of the
RECORD directive. The Ideal mode version of the RECORD
directive differs slightly from the MASM mode version.
Turbo Assembler User's Guide
The UNION
directive
The UNION directive provides a way to reference a given memory
location as more than one data type. UNION is similar to C's union
statement.
Suppose you have a counter that you use sometimes as an 8-bit
counter and sometimes as a 16-bit counter. You could declare it to
be a union of the two with
FLEX COUNT
COUNT8
COUNT16
FLEX COUNT
UNION
DB?
DW
ENDS
Note that, as with STRUC, UNION definitions must end with
ENDS.
Given the previous definition of the FLEX_COUNT union, you
could create and use a dual-purpose counter as follows:
•DATA
Counter FLEX COUNT
.CODE
mov
LoopTop:
[Counter.COUNT16],Offffh
dec [Counter.COUNT16]
jnz ShortLoopTop
mov [Counter.COUNT8],255
ShortLoopTop:
dec [Counter.COUNT8]
jnz ShortLoopTop
As with STRUC, the period operator is used to reference union
fields; the plus operator could be used as well. Referencing a
variable by way of its union fields is equivalent to using type
overrides. The preceding example is equivalent to
Chapter 9, Advanced programming in Turbo Assembler
387
•DATA
Counter DW
?
.CODE
mov WORD PTR [Counter],Offffh
LoopTop:
dec WORD PTR [Counter]
jnz LoopTop
mov BYTE PTR [Counter],255
ShortLoopTop:
dec BYTE PTR [Counter]
jnz ShortLoopTop
The advantage of using a union over type overrides is that you're
much more likely to use the correct union element name than you
are to remember the type override in every instance. Also, the
multiple-mode operation of a union variable is instantly apparent
when you look at the variable's definition, so code containing
unions is easier to understand and maintain.
You can nest both unions and structures within unions. For
example, the following union allows a 4-byte memory variable to
be accessed as either a doubleword-sized segment:offset pointer
or as a word-sized offset variable and a word-sized segment
variable:
SEG OFF
POFF
PSEG
SEG OFF
STRUC
OW ?
DW ?
ENDS
PUN ION
DPTR
XPTR
PUNION
UNION
DO
SEG OFF
ENDS
<>
.CODE
mov
mov
[bx.XPTR.POFF],si
[bx.XPTR.PSEG],ds
les di, [bx.DPTR]
388
Turbo Assembler User's Guide
As with STRUC and RECORD, the field names defined with
UNION are normal labels, with no scope limitations.
Consequently, union field names must be unique in their source
module.
Refer to Chapter 11 for
information about Ideal
mode.
In this section, we've discussed the MASM mode version of the
UNION directive. In Ideal mode, the UNION directive is
considerably more powerful, providing more of the features
available to structures in high-level languages.
Segment directives
In Chapter 5, you learned how to use the simplified segment
directives, and you learned enough about the standard segment
directives to be able to make a working program. Now we're
going to discuss each of the standard segment directives in detail,
and provide you with more information about what the
simplified segment directives do. We're also going to show you a
sample program that uses several code and data segments, to give
you a feel for how multisegment programs operate.
Recall that the simplified segment directives are easier to use but
less powerful than the standard segment directives. The standard
segment directives we cover in the next sections are SEGMENT,
GROUP, and ASSUME.
The SEGMENT
directive The SEGMENT directive is used to start a segment. Each
SEGMENT directive must have a matching ENDS to terminate that
segment. Unlike the simplified segment directives, SEGMENT
gives you complete control over the attributes of each segment.
The complete form of the SEGMENT directive is
name SEGMENT align combine use 'class'
where align, combine, use, and class are all optional. We'll discuss
each of these fields in turn.
Chapter 9, Advanced programming in Turbo Assembler
389
The name and align
fields
name is the name of the segment. Segment names are labels, so
they must be unique in their source modules. The same name
must be used with ENDS when the segment is ended.
align specifies the memory boundary on which the segment
should start. The following are valid alignments:
• BYTE uses the next available byte address .
.. DWORD uses the next doubleword-aligned address.
• PAGE uses the next page address (256-byte aligned).
• PARA uses the next paragraph address (16-byte aligned).
• WORD uses the next word-aligned address.
If no alignment is explicitly specified, paragraph-alignment is
used.
Byte-alignment makes for the most compact programs. Wordalignment is preferable on 16-bit computers such as the AT, since
16-bit processors operate more efficiently on word-aligned data;
doubleword-alignment is preferable on 32-bit computers for
much the same reason. Paragraph-alignment is necessary for
segments that will be a full64K long.
The combine field
combine controls the manner in which segments of the same name
in other modules will be combined with this segment when the
modules that make up the program are linked together. combine
can be anyone of the following types:
AT
COMMON
MEMORY
PUBLIC
STACK
VIRTUAL
PRIVATE
You might find it useful to refer to the later section (on page 400),
"The simplified segment directives," which shows the combine
types used by high-level languages.
A combine type of AT causes the start of the segment to be placed
at a specific address in memory. No code is actually generated;
instead, AT segments are used as templates for accessing memory
areas such as the ROM BIOS data segment and display memory.
For example,
VGA- GRAPHICS- MEMORY SEGMENT AT OAOOOh
390
Turbo Assembler User's Guide
BitMapStart
LABEL
BYTE
VGA- GRAPHICS- MEMORY ENDS
mov ax,VGA_GRAPHICS_MEMORY
mov eS,ax
ASSUME
es:VGA- GRAPHICS- MEMORY
mov di,OFFSET BitMapStart
mov cx,08000h
sub aX,ax
cld
rep stosw
clears the VGA graphics screen.
The combine type COMMON specifies that the beginning of this
segment and the beginning of all other segments of the same
name should be aligned, so that the segments overlay each other.
The total segment size is only the size of the largest segment of
this name. One way in which the COMMON combine type can be
used is by including a file that defines a COMMON segment in
each module referencing that segment, so that all modules
effectively share exactly the same segment.
The combine type PUBLIC instructs the linker to concatenate this
segment with other segments of the same name, so the segments
are effectively pieced together to make a larger segment. The total
size of the segment is the sum of the size of all segments of this
name. As with all segments, the total size of PUBLIC segments
can't exceed 64K. PUBLIC is used when multiple modules share
the same segment, but each defines its own variables. Variables in
PUBLIC segments are often shared between modules by way of
GLOBAL directives.
The MEMORY combine type is the same as PUBLIC.
The combine type STACK instructs the linker to concatenate all
segments of this name into one segment, and to build the EXE file
so that SS:SP is set to point to the end of this segment when the
program is run. This is a specialized combine type to be used for
the stack and nothing else.
A combine type of VIRTUAL defines a special kind of segment,
which will be treated as a common area and attached to another
segment at link time. The VIRTUAL segment is assumed to be
attached to the enclosing segment. The VIRTUAL segment also
inherits its attributes from the enclosing segment. The ASSUME
directive considers a VIRTUAL segment to be a part of its parent
Chapter 9, Advanced programming in Turbo Assembler
391
segment; in all other ways, a VIRTUAL segment is treated just like
a normal segment. The linker treats virtual segments as a common
area that will be combined across modules. This permits static
data that comes into many modules from Include files to be
shared.
Finally, the combine type PRIVATE instructs the linker not to
combine this segment with any other segments. This allows you
to define segments that are local to a given module, without
having to worry about possible conflicts if segments of the same
name are used in other modules. Segments default to combine
type PRIVATE if no combine type is specified.
The use and class fields
The next section has more
information about segment
ordering.
Segment size, type,
name, and nesting
The use field of the SEGMENT directive is for 80386 assembly
only; Chapter 10 offers more information on the use field.
The class field is used to control the order in which the linker
places segments. All segments of a given class are placed in a
contiguous block of memory, no matter what their order is in the
source code. The section liThe simplified segment directives"
shows the classes used by high-level languages; for simplicity,
you might want to follow these conventions.
The cumulative size of the segments in a class is limited only by
the availability of memory at run-time; however, no individual
segment can exceed 64K.
Note that the class type, if present, must be enclosed in quotes.
Also, class types must be unique in their source modules; that is,
no label used in a given module may have the same name as a
class type used in that module.
You can define the same segment name multiple times in the
same source module; all instances are considered to refer to a
single segment. However, you must make sure that all definitions
of a given segment in a source module have the same attributes;
otherwise, Turbo Assembler will generate an error.
One handy way to avoid such errors is to specify attributes only
the first time you define a segment in a given source module.
When a redefined segment with no attributes is encountered,
Turbo Assembler automatically uses the attributes specified when
the segment was first defined.
Finally, segments can be nested, which means you can define a
segment before you end an earlier segment, as follows:
392
Turbo Assembler User's Guide
DataSeg
SEGMENT PARA PUBLIC 'DATA'
DataSeg2 SEGMENT PARA PRIVATE 'FAR_DATA'
DataSeg2. ENDS
DataSeg
ENDS
Nesting is not generally useful, but there is at least one case where
it's handy, and that's in a macro. In order to define a segment in a
macro, you'd normally have to end and then restart the current
segment, and to do that you'd need to know the current segment's
name, which is not necessarily obvious in the context of a macro.
By contrast, segment-nesting allows you to define a segment
without ever knowing what the name of the current segment is, as
follows:
TEST MACRO
TestSeg
SEGMENT WORD PRIVATE 'FAR DATA'
TestSeg
ENDS
ENDM
After a nested segment ends, Turbo Assembler simply resumes
assembling into the segment that was active when the nested
segment began.
Segment-ordering
By and large, you don't need to worry about the order in which
the segments end up in the .EXE files you create. First of all, the
order in which segments appear in .EXE files doesn't often matter.
Second, most of the cases in which you might care about segment
order are easily handled by a high-level language compiler or the
DOSSEG directive. If you're linking to a high-level language, that
language's compiler will usually control the segment order. If
you're writing a pure assembler program and have specified the
DOSSEG directive, your segments will end up in Microsoftstandard segment order, as follows:
• Segments of class CODE
Chapter 9, Advanced programming in Turbo Assembler
393
• Segments of class other than CODE that are not part of
DGROUP
• Segments that are part of DGROUP, in the following order:
• Segments of classes other than STACK and BSS
• Segments of class BSS
• Segments of class STACK
If you're curious about the order in which the linker is placing
your segments, just use the Is command-line switch to instruct
TLINK to generate a detailed segment map file and take a look at
the map file.
A question remains: How are segments ordered if you aren't
linking to a high-level language and you don't use the DOSSEG
directive? Most of the time, you'll have no need to know the
answer to that question, but in case it does matter to you, here's
the answer. (It's a bit more complex than you might think.)
When no explicit segment-ordering, such as that forced by
DOSSEG, is in effect, the linker groups all segments of a given
class together, where the class of a segment is specified by the
class field of the SEGMENT directive. The groups of segments
themselves are placed in the .EXE file simply in the order in
which the linker encounters them; the first segment class the
linker encounters in loading the .OB] files is placed first in the
.EXE file, the second segment class encountered comes next, and
so on. This means that the order in which .OB] files are linked
affects the final order of the segments in the .EXE file.
Now you've got the segments loosely ordered by class. How, then,
are the segments within each class ordered? Once again, they're
placed in the .EXE file in the order in which the linker
encountered them. One factor here is the order in which the .OB]
files are linked; another factor is the order in which the segments
are placed in each .OB] file. Turbo Assembler gives you two
choices regarding the order in which segments appear in .OB]
files.
The .SEQ directive instructs Turbo Assembler to place segments
in the .OB] file in the order in which they appear in the source file.
With sequential-ordering, the order of the segments in a given
source module can affect the order of the segments in the .EXE
file. This is the default mode of operation of Turbo Assembler, so
sequential segment-ordering will occur even if you omit the .SEQ
directive, as long as the .ALPHA directive is not used.
394
Turbo Assembler User's Guide
The .ALPHA directive instructs Turbo Assembler to place
segments in the .OBI file in alphabetic order. With alphabeticordering, the order of the segments in a given source module does
not affect the order of the segments in the .EXE file. This is the
default mode of operation of some older assemblers, so you may,
on occasion, have to use .ALPHA in order to get assembler
programs to run properly.
So, now you've got segments loosely ordered by class, and
ordered within the class by the order of appearance of the
segments. You can control the order of appearance of segments
within the class both by the order in which .OBI files are linked
and by the .SEQ and .ALPHA directives. If .SEQ is selected, the
order of appearance of segments in a given source module can
affect the order of the segments in the .EXE file.
You can see that segment-ordering is no simple matter. Odds are,
though, that you'll never have to worry about segment order; it
doesn't usually make any difference anyway, and when it does, a
high-level compiler or the DOSSEG directive generally takes care
of segment-ordering for you.
The GROUP
directive The GROUP directive is used to combine two or more segments
into one logical entity, so that all the segments can be addressed
relative to a single segment register.
Suppose you have a program that accesses data in two segments.
Normally, you'd have to load a segment register and perform a
new ASSUME each time you wanted to access first one segment
and then the other; that's both time-consuming and a nuisance. It's
far easier to combine the segments into a single group named
DataGroup, load DS with the start of DataGroup, ASSUME DS to
DataGroup, and then access either segment at any time. Here's the
code:
DataGroup GROUP
DataSegl,DataSeg2
DataSegl SEGMENT
MemVarl DW 0
DataSegl ENDS
PARA PUBLIC 'DATA'
DataSeg2 SEGMENT
MemVar2 DW 0
DataSeg2 ENDS
PARA PUBLIC 'DATA'
. Chapter 9, Advanced programming in Turbo Assembler
395
mov ax,DataGroup
mov ds,ax
ASSUME
ds:DataGroup
mov ax, IMemVarl]
mov IMemVar2],ax
Why would you want to use groups, when using a single segment
name and the combine type PUBLIC produces the same result
more easily? Actually, in pure assembler programs, there's not
that much need for groups, although you can certainly use them if
you want. Groups are primarily used when interfadng assembler
code to high-level languages. In particular, the group DGROUP is
used by high-level languages to allow the stack, initialized near
data, uninitialized near data, and constant segments to be
accessed relative to a single segment register.
1111"
The one key rule with groups is that all the segments in a group
must lie within a single 64K segment, since they must all be
accessed relative to a single segment register. Bear in mind that
segment-ordering is dependent on many factors, as discussed in
the last section, so segments might lie some distance apart if
you're not careful. The safest approach is to declare all segments
in a group to be of the same class, and to define them one after the
other at the start of all modules they're defined in.
However, when you are either linking to a high-level language or
have used the DOSSEG directive anywhere in your program,
there's no need to worry about making sure that the segments in
DGROUP are kept together; in both these cases, the linker
automatically makes all segments in DGROUP adjacent.
While the segments in a group must fit within a 64K segment,
they do not have to be contiguous once they're linked. Nongrouped segments can lie between the segments that make up a
group.
1111"
If you do use a group, you must be careful always to use the
group name with ASSUME when you load a segment to point to
the group. Otherwise, Turbo Assembler will generate offsets
relative to the segment start, not the group start, even though the
segment register is pointing to the group start. For example, the
following would cause errors given the previous definition of
DGROUP:
396
Turbo Assembler User's Guide
mov ax,DGROUP
mov ds,ax
ASSUME
ds:Stack
;will produce incorrect offsets!
Instead, use
mov ax,DGROUP
mov ds,ax
ASSUME
ds:DGROUP
In short, if you load a segment register to point to a group, be sure
to ASSUME to that group, not to any of its component segments.
See •Forgetting group
overrides in operands and
data tables, • on page 252 in
Chapter 6 for more
information.
MASM, the Microsoft Macro Assembler, has a bug regarding the
OFFSET operator with groups. This bug also surfaces when
initializing data to the address of labels in a group. In the interests
of compatibility, Turbo Assembler reproduces this bug. The
workaround for this bug is always to place group override
prefixes on labels when you use them with the OFFSET operator
or use them to initialize data.
The ASSUME
directive The ASSUME directive lets you tell Turbo Assembler what
segment or group a given segment register is pointing to. Note
that this is not the same as actually loading a segment register to
point to that segment; you must do that separately with the MOV
instruction. The purpose of ASSUME is to allow Turbo Assembler
to check the validity of your memory references and to insert
segment override prefixes automatically on your memory
accesses as needed.
An ASSUME for CS must appear before any code in each source
module, so that Turbo Assembler knows what segment to assume
the instructions are in, for purposes of jumps, calls, and setting
the starting address of the program.
Other ASSUME directives for the various segment registers can be
inserted as often as needed in any source module. The assumed
segment for any segment register can be changed whenever you
wish. Any or all segment assumptions can be changed with a
single ASSUME directive.
Chapter 9, Advanced programming in Turbo Assembler
397
You can specify an assumption for a segment register with either
a segment name, a group name, or a segment extracted from a
label with the SEG operator. Additionally, you can use the
NOTHING keyword to instruct Turbo Assembler to assume that
any or all segment registers aren't pointing to any segment.
Here's an example of using ASSUME:
Stack
SEGMENT
512 DUP
ENDS
Stack
TGROUP
GROUP
DataSeg1 SEGMENT
DB
DataSeg1 ENDS
DataSeg2 SEGMENT
PARA STACK 'STACK'
(0)
DataSeg1,DataSeg2
PARA PUBLIC 'DATA'
PARA PUBLIC 'DATA'
DataSeg2 ENDS
DataSeg3 SEGMENT
DW 0
MemVar
PARA PUBLIC 'DATA'
DataSeg3 ENDS
CodeSeg SEGMENT PARA PUBLIC 'CODE'
ASSUME
cs:CodeSeg,ds:TGROUP,ss:Stack,es:NOTHING
ProgramStart:
mov ax,TGROUP
mov ds,ax
ASSUME
ds:TGROUP
mov ax,SEG MemVar
mov es,ax
ASSUME
es:SEG MemVar
;same as DataSeg3
push ds
pop es
mov ax,CodeSeg
mov ds,ax
ASSUME
ds:CodeSeg,es:TGROUP
CodeSeg ENDS
END ProgramStart
If an ASSUME directive refers to a group, the specified segment
register is assumed to point to the start of that group. However, if
an ASSUME directive refers to a segment that's part of a group,
the segment register is assumed to point to the start of the
398
Turbo Assembler User's Guide
segment, not the group. This can cause problems, since segment
registers are generally set to point to the start of groups, not
segments that make groups. For example, the following would
load AX. from the wrong memory location, since OS points to the
start of TGROUP, but the ASSUME statement incorrectly indicates
that OS points to the start of DataSeg2:
TGROUP
GROUP
DataSegl SEGMENT
DataSegl
DataSeg2
MemVar
DataSeg2
ENDS
SEGMENT
DW 0
ENDS
DataSegl,DataSeg2
PARA PUBLIC 'DATA'
PARA PUBLIC 'DATA'
CodeSeg SEGMENT PARA PUBLIC 'CODE'
ASSUME
cs:CodeSeg
mov ax,TGROUP
mov ds,ax
ASSUME
ds:DataSeg2
mov ax, [MemVarl
;not correct!!! (should be TGROUP)
;will load from the wrong offset,
relative to DataSeg2 rather than
; TGROUP
When you use the simplified segment directives, it's generally not
necessary to use ASSUME, since Turbo Assembler automatically
generates the appropriate segment assumptions. However, if you
change any segment registers while you're using the simplified
segment directives, you will have to perform the appropriate
ASSUME directives. For example, the following sets OS to point to
the .DATA segment, the .CODE segment, the .FARDATA segment,
and finally back to the .DATA segment:
. DATA
.FARDATA
. CODE
mov ax,@data
mov ds,ax
ASSUME
ds:@data
mov ax,@code
mov ds,ax
Chapter 9, Advanced programming In Turbo Assembler
399
ASSUME
ds:@code
mov ax,@fardata
mov ds,ax
ASSUME
ds:@fardata
mov ax,@data
mov ds,ax
ASSUME
ds:@data
You should exercise care
that your ASSUME directives
correspond to the actual
settings of the segment
registers at all times.
As we've pointed out before, the ASSUME directive can cause
Turbo Assembler to insert segment override prefixes on memory
accesses whenever Turbo Assembler (operating on the basis of the
ASSUME directives you've issued) thinks that's necessary to access
a given memory variable. For example, Turbo Assembler will put
an ES: override on the instruction that accesses MemVar in the
following code, since the ASSUME directive incorrectly indicates
that DS can't reach the segment where MemVar resides:
DataSeg
MemVar
SEGMENT
DB ?
DataSeg
ENDS
PARA PUBLIC 'DATA'
CodeSeg SEGMENT PARA PUBLIC 'CODE'
ASSUME
cs:CodeSeg,ds:NOTHING,es:DataSeg
mov ax,DataSeg
mov ds,ax
mov es,ax
mov [MemVarl,l
The simplified
segment
directives
Memory models are discussed In Chapter 5.
400
We discussed the simplified segment directives in some detail in
Chapter 5. However, the main aspect of simplified segment
directives that we haven't covered yet is exactly what segments
the various simplified segment directives create. That's not
something you'll normally have to know, but if you're mixing
simplified and standard segment directives, you might need that
information.
The segments and segment groups created by the .CODE, .DATA,
.DATA?, .STACK, .CONST, .FARDATA, and .FARDATA? directives
Turbo Assembler User's Guide
depend on the memory model selected by the .MODEL directive.
The following tables show the correspondence of memory models
and the segments created by the simplified segment directives:
Table 9.1
Default segments
and types for tiny
memory model
Table 9.2
Default segments
and types for small
memory model
Table 9.3
Default segments
and types for
medium memory
model
Table 9.4
Default segments
and types for
compact memory
model
Directive
Name
Align
Combine
Class
Group
'CODE'
TEXT
WORD
PUBLIC
FAR_DATA PARA
private
'FAR DATA'
'FAR-BSS'
PARA
private
FAR BSS
'DATA'
DATA
WORD
PUBLIC
CaNST
WORD
PUBLIC
'CONsr
'BSS'
BSS
WORD
PUBLIC
'STACK'
STACK
STACK
PARA
- STACK not assumed to be in DGROUP if FARSTACK specified.
DGROUP
Directive
Group
.CODE
.FARDATA
.FARDATA?
.DATA
.CONST
.DATA?
.STACK-
Name
Align
Combine
Class
'CODE'
TEXT
WORD
PUBLIC
'FAR DATA'
FAR_DATA PARA
private
'FAR-BSS'
PARA
private
FAR BSS
_DATA
WORD
PUBLIC
'DATA'
CaNST
WORD
PUBLIC
'CaNST'
WORD
PUBLIC
'BSS'
BSS
'STACK'
STACK
PARA
STACK
- STACK not assumed to be in DGROUP if FARSTACK specified.
.CODE
.FARDATA
.FARDATA?
.DATA
.CONST
.DATA?
.STACK-
Directive
Name
Align
Combine
Class
'CODE'
name_TEXT
WORD
PUBLIC
private
'FAR DATA'
FAR DATA PARA
'FAR-BSS'
PARA
private
FAR=BSS
WORD
PUBLIC
'DATA'
DATA
CaNST
WORD
PUBLIC
'CONsr
'BSS'
BSS
WORD
PUBLIC
'STACK'
STACK
STACK
PARA
- STACK not assumed to be in DGROUP if FARSTACK specified.
.CODE
.FARDATA
.FARDATA?
.DATA
.CONST
.DATA?
.STACK-
Directive
Name
Align
Combine
Class
'CODE'
_TEXT
WORD
PUBLIC
'FAR DATA'
FAR DATA PARA
private
PARA
private
'FAR=BSS'
FAR=BSS
WORD
PUBLIC
'DATA'
DATA
'CONsr
CaNST
WORD
PUBLIC
'BSS'
BSS
WORD
PUBLIC
'STACK'
STACK
STACK
PARA
- STACK not assumed to be in DGROUP if FARSTACK specified.
.CODE
.FARDATA
.FARDATA?
.DATA
.CONST
.DATA?
.STACK-
Chapter 9, Advanced programming in Turbo Assembler
DGROUP
DGROUP
DGROUP
DGROUP
DGROUP
DGROUP
DGROUP
DGROUP
Group
DGROUP
DGROUP
DGROUP
DGROUP
Group
DGROUP
DGROUP
DGROUP
DGROUP
401
Table 9.5
Default segments
and types for large
or huge memory
model
Table 9.6
Default segments
and types for Turbo
Pascal (TPASCAL)
memory model
Directive
Name
Align
Combine
Class
'CODE'
.CODE
name TEXT WORD PUBLIC
FAR-DATA PARA
'FAR_DATA'
.FARDATA
private
'FAR BSS'
.FARDATA? FAR-BSS
PARA
private
_DATA
PUBLIC
'DATA'
.DATA
WORD
'CONST'
.CONST
CONST
WORD
PUBLIC
'BSS'
.DATA?
WORD
PUBLIC
BSS
.STACK"
STACK
'STACK'
STACK
PARA
"STACK not assumed to be in DGROUP if FARSTACK specified.
Directive
Name
Align
Combine
.CODE
.DATA
CODE
DATA
BYTE
WORD
PUBLIC
PUBLIC
Group
DGROUP
DGROUP
DGROUP
DGROUP
In past chapters, you've probably noticed that programs using the
simplified segment directives don't need ASSUME, GROUP, or
ENDS directives. The .MODEL directive automatically performs
the appropriate ASSUME directives for the selected memory
mode, assuming the segments shown in the preceding tables .
.MODEL also performs the group definition for DGROUP, as
shown in the previous tables.
As for ENDS, the start of a new segment with a simplified
segment directive-for example, .CODE or .DATA-automatically
ends the current segment, if there is one.
Take a look now at the more esoteric simplified segment
directives: .DATA?, .CONST, .FARDATA, and .FARDATA?
.FARDATA is really the only one of these you'll ever use in a pure
assembler program; the others are strictly for matching the
segment usage of high-level languages .
. DATA? starts the segment that is to contain uninitialized near
data in DGROUP. Since both the .DATA and .DATA? segments are
in the same group, there's really no reason not to simply skip
using .DATA? altogether and use question marks 'to define
uninitialized data in the .DATA segment, except when you're
following the conventions of a high-level language .
.CONST, which starts the segment that is to contain constant near
data in DGROUP, falls into the same category as .DATA? You
might as well put your constant data in .DATA and skip .CONST,
except when you're following the conventions of a high-level
language.
402
Turbo Assembler User's Guide
.FARDATA is used to create a far data segment unique to a given
source module; that is, a segment that's not shared with any other
module. That segment is named FAR_DATA but is of combine
type PRIVATE, so it's not combined with any other segment.
.FARDATA allows you to define up to 64K of local data storage in
each module. Of course, if you use .FARDATA, you must set a
segment register to point to that segment, as follows:
.MODEL
small
•DATA
InitValue DW 0
.FARDATA
MemArray DW 100 DUP (?)
.CODE
mov ax,@data
mov ds,ax
mov ax,@fardata
mov es,ax
mov ax, [InitValue]
ASSUME
es:@fardatai
mov di,OFFSET MemArray
mov cx,100
cld
rep stosw
Note that the predefined label @fardata contains the name of the
segment defined with the .FARDATA directive.
While a segment defined with .FARDATA isn't shared with any
other module (as, for example, the segment defined with .DATA
is), you can use GLOBAL to share specific variables in the
.FARDATA segment with other modules. For example, the
following makes MemVar available to other modules:
. MODEL
small
.FARDATA
GLOBAL
MemVar:WORD
MemVar
DW 0
Another module could then reference MemVar as follows:
. MODEL
GLOBAL
•DATA
small
MemVar:WORD
. CODE
Chapter 9, Advanced programming in Turbo Assembler
403
mov ax,SEG MemVar
mov ds,ax
ASSUME
ds:SEG MemVar
mov ax, [MemVar]
Note that the declaration of MemVar as GLOBAL comes before
any segment is declared. This is necessary because a global
declaration of a given variable must be performed either inside
the variable's segment or outside all segments. Since, by
definition, no module can share another module's .FARDATA
segment, the declaration of MemVar must be performed outside
all segments.
.
.FARDATA? is much like .FARDATA, except that it creates a
private segment named FAR_BSS. FAR_BSS segments are used
by high-level languages for uninitialized far data. If you're not
interfacing to a high-level language, there's no reason you
shouldn't define your uninitialized far data in the segment
defined with .FARDATA and forget about .FARDATA? True, the
.FARDATA segment gives you an additional64K of far storage,
but if you really need more than 64K of far storage that's unique
to a given module, you should probably be using the standard
segment directives anyway.
If you do use .FARDATA?, the predefined label @fardata?
contains the name of the segment defined by .FARDATA, suitable
for use in ASSUME directives and in loading segment registers.
A multisegment
program
The next program has two code segments and two data segments.
This is hardly a comprehensive example of multisegment
programming, but we don't have the space for a program running
to hundreds or thousands of lines; this one will serve to give you
a feel for switching data segments, loading full segment:offset
pointers, and calling code in other segments.
Here's the example:
; Program to demonstrate use of multiple code and data segments.
; Reads a string from the console, stores it in one data
segment, copies the string to another data segment, converting
it to lowercase in the process, then prints the string to the
404
Turbo Assembler User's Guide
console. Uses functions in another code segment to read,
print, and copy the string.
Stack
DB
Stack
SEGMENT PARA STACK 'STACK'
512 DUP (?)
ENDS
. MAX_STRING_LENGTH
EQU 1000
SourceDataSeg SEGMENT PARA PRIVATE 'DATA'
InputBuffer
DB MAX_STRING_LENGTH DUP (?)
SourceDataSeg ENDS
DestDataSeg
OutputBuffer
DestDataSeg
SEGMENT PARA PRIVATE 'DATA'
DB MAX_STRING_LENGTH DUP (?)
ENDS
SubCode SEGMENT PARA PRIVATE 'CODE'
ASSUME
cs:SubCode
Subroutine to read a string from the console. String end is
marked by a carriage-return, which is converted to a
carriage-returnflinefeed pair so it will advance to the next
line when printed. A a is added to terminate the string.
Input:
ES:DI - location to store string at
Output: None
Registers destroyed: AX,DI
GetString PROC FAR
GetStringLoop:
mov ah,l
int 21h
stosb
cmp aI, 13
jnz GetStringLoop
mov BYTE PTR es: [di],10
mov BYTB PTR es: [di+1],0
;get the next character
isave it
iis it a carriage-return?
ino-not done yet
iend the string with a linefeed
; and with a zero
ret
GetString ENDP
Subroutine to copy a string, converting it to lowercase.
Input:
DS:SI - string to copy
ES;DI - place to put string
Chapter 9, Advanced programming in Turbo Assembler
405
Output: None
Registers destroyed: AL, SI, DI
CopyLowercase PROC FAR
CopyLoop:
lodsb
cmp aI,' A'
jb NotUpper
cmp al,'Z'
ja NotUpper
add al,20h
NotUpper:
stosb
and al,al
jnz CopyLoop
ret
CopyLowercase ENDP
iconvert to lowercase if it's uppercase
iwas that the 0 that ends the string?
ino, copy another character
Subroutine to display a string to the console.
Input:
DS:SI - string to display
Output: None
Registers destroyed: AH,DL,SI
DisplayString PROC FAR
DisplayStringLoop:
mov dl, [si)
and dl,dl
jz DisplayStringDone
inc si
mov ah,2
int 21h
jmp DisplayStringLoop
DisplayStringDone:
ret
DisplayString ENDP
Sub Code ENDS
Code SEGMENT
ASSUME
ProgramStart:
cld
iget the next character
iis this the 0 that ends the string?
iyes, we're done
ipoint to the following character
idisplay the character
PARA PRIVATE 'CODE'
cs:Code,ds:NOTHING,es:NOTHING,ss:Stack
imake string instructions increment
i their pointer registers
Read a string from the console into InputBuffer.
406
Turbo Assembler User's Guide
mov ax,SourceDataSeg
mov es,ax
ASSUME
es:SourceDataSeg
mov di,OFFSET InputBuffer
call GetString
iread string from the console and
i store it at ES:DI
Print a linefeed to advance to the next line.
mov ah,2
mov dl,lO
int 21h
Copy the string from InputBuffer to OutputBuffer, converting
it to lowercase in the process.
push es
pop ds
ASSUME
ds:SourceDataSeg
mov ax,DestDataSeg
mov es,ax
ASSUME
es:DestDataSeg
mov si,OFFSET InputBuffer
mov di,OFFSET Output Buffer
call CopyLowercase
iCOPY from DS:SI •..
i ... to ES:DI. ..
i ••. making it lowercase
Display the lowercase string.
push es
pop ds
ASSUME
ds:DestDataSeg
mov si,OFFSET OutputBuffer
call DisplayString
idisplay string at DS:SI
i to the console
Done.
mov ah,4ch
int 21h
Code ENDS
END ProgramStart
1111.
Note that, in this example, the subroutines come before the main
program. This is done in order to avoid forward references, since
the subroutines and the main program reside in different code
segments. If the main program came first, you'd have to put FAR
PTR overrides on each subroutine call because Turbo Assembler
Chapter 9, Advanced programming In Turbo Assembler
407
can't automatically assemble far forward-referenced jumps. Given
the way the program is organized, however, all the subroutine
calls are backward references, so Turbo Assembler can
automatically generate far calls to the subroutines.
Otherwise, the program is quite straightforward. The subroutines
use full segment:offset pointers to data, and the main program
sets DS and ES to different data segments as needed. Note the use
of the string instructions when copying the string and converting
it to lowercase; since LODS defaults to using OS and STOS uses
ES, these instructions are ideally suited for use in code that must
access two segments at once.
408
Turbo Assembler User's Guide
c
H
A
p
T
E
R
10
The 80386 and other processors
So far, we've focused on assembly language programming for the
8086 processor. (We've also implicitly covered the 8088, which is
used in the IBM PC and XT, since the 8088 is basically an 8086
with an 8-bit external data bus.)
The 8086 is not the only processor Turbo Assembler supports,
however; there is a whole family of 8086-superset processors,
known as the iAPx86 family, and a family of math coprocessors
that are supersets of the 8087, as well.
The most exciting member of the iAPx86 family is, without a
doubt, the 80386, which brings minicomputer power to personal
computers. Nonetheless, each of the members of the iAPx86
family has interesting enhancements over the basic 8086.
First, we'll look at the ways in which the 80186 and 80286
processors extend the capabilities of the 8086. Next, we'll look at
80386 programming to see how to enable Turbo Assembler's
80386 features, examine the new segment types used in 80386
programming, and look at the new registers, addressing modes,
and instructions of the 80386. After that, we'll examine Turbo
Assembler's powerful ability to mix 16- and 32-bit instructions
and segments, and we'll look at some sample 80386 code. Finally,
we'll take a brief look at the ways in which the 80287 and 80387
math coprocessors extend the capabilities of the 8087.
Chapter 70, The 80386 and other processors
409
Switching processor types in assembler code
Turbo Assembler defaults to supporting the assembly of 8086
code only. In order for Turbo Assembler to support another
iAPx86-family processor or coprocessor, you must issue the
appropriate directive. The following directives tell Turbo
Assembler what type of processor to support when it's assembling
code:
.186
.286
.286C
.286P
.287
.386
.386C
.386P
.387
.8086
.8087
These directives can be inserted anywhere in assembler source
files, and take effect immediately. Multiple processor-type
directives can be placed in a single source file; at any given point
in a source file, the last processor type specified is the processor
type currently selected.
For the remainder of this
chapter, all references made
to the 8086 apply to the 8088
oswell.
The .8086 directive can be used anytime to instruct Turbo
Assembler to return to supporting 8086 assembly only. For
example, the following function adds two 32-bit values by using
8086 code, then 80386 code, and finally 8086 code again:
.MODEL
small
•CODE
Add32
PROC
mov ax, [bp+4]
mov dx, [bp+6]
mov bx, [bp+8]
mov ex, [bpi 10]
.386
shl eax,16
mov ax,dx
rol eax,16
mov dx,cx
shl edx,16
mov dx,bx
add eax,edx
rol eax,16
mov dx,ax
shr eax,16
.8086
ret
Add32
ENDP
END
410
;get
;get
;get
;get
;use
low half of source 1
high half of source 1
low half of source 2
high half of source 2
80386 registers for actual addition
;put 32 bits of source 1 in EAX
;put 32 bits of source 2 in EDX
;add source 1 and source 2
;put high half of result in DX
;low half of result is in AX
Turbo Assembler User's Guide
The 80186 and 80188
The 80186 is the iAPx86-family processor most like the 8086. The
80186 supports all the instructions of the 8086 and adds a few new
instructions, along with extended forms of some 8086
instructions. In addition, the 80186 is considerably faster than the
8086 at many operations, especially memory address calculations,
so the 80186 runs code written for the 8086 at a significantly
higher speed than does the 8086.
The 80188 is program-compatible with the 80186; the only
difference between the two is that the 80186 has a 16-bit external
data bus, and the 80188 has an 8-bit external data bus.
Turbo Assembler support for assembly of 80186 code is enabled
with the .186 directive.
For information about 80 186
instructions, see Chapter 3 of
the Reference Guide.
Next, let's take a look at the new and extended instructions of the
80186.
New instructions
Warningl
Before we begin, take note that the 8086 does not recognize any of
the instructions we're about to discuss. Consequently, any
program that uses even one of the new or extended instructions of
the 80186 won't run on an 8086.
Here are the new 80186 instructions:
PUSHA and POPA
BOUND
INS
ENTER
LEAVE
OUTS
POPA
PUSHA
PUSHA and POPA provide an efficient means by which to push
and pop all eight general-purpose registers. PUSHA pushes the
eight general-purpose registers onto the stack in the order AX,
CX, DX, BX, SP, BP, 51, DI. POPA pops DI, 51, BP, BX, DX, CX, and
AX from the stack, reversing the action of PUSHA. SP is not
popped by POPA; instead, SP is incremented by 16, the length of
the block of registers pushed on the stack by PUSHA, and the
value of SP pushed by PUSHA is cleared from the stack by POPA
and thrown away. The segment registers, the flags, and the
instruction pointer are not affected by PUSHA or POPA.
For example, the code
Chapter 70, The 80386 and other processors
411
Don't forget to use the .186
directive to enable 80 186
assembly before using
80 186-specific Instructions
such as PUSHA and POPA.
.186
SampleFunction PROC
pusha
popa
ret
SampleFunction ENDP
preserves all 8 general-purpose registers with just two instructions, rather than the 16 instructions required to push and pop
each register separately.
Be aware that while PUSHA is faster than eight separate PUSH
instructions, it is slower than three or four pushes; if you only
need to preserve a few registers, it's best to save just those
registers with PUSH. The same is true of POPA and POP.
ENTER and LEAVE
ENTER and LEAVE are used to set up and discard stack frames, in
which passed parameters and local (automatic) variables can be
accessed relative to BP. ENTER and LEAVE are particularly useful
when interfacing assembler functions to stack-oriented languages
such as C. (See Chapters 7 and 8 for information on interfacing
assembler functions to Turbo C and Turbo Pascal.)
ENTER preserves the calling routine's BP, sets BP to point to the
start of the passed parameters (if any) in a new stack frame,
adjusts SP as needed to allocate room for local variables, and even
copies a block of pointers to higher-level stack frames into the
new stack frame if necessary.
LEAVE undoes everything ENTER does, restoring both BP and SP
to the state they were in before the corresponding ENTER was
executed.
For example, the following function uses ENTER to set up a Ccompatible stack frame with 20 bytes reserved for local variables,
and uses LEAVE to discard that stack frame and restore the calling
code's stack frame:
412
Turbo Assembler User's Guide
SampleFunction PROC
enter 20,0
leave
ret
SampleFunction ENDP
The first operand to ENTER is a 16-bit immediate value specifying
the number of bytes to reserve for local variables in the new stack
frame. The second operand to ENTER is an 8-bit immediate value
specifying the nesting level of the function for which the new
stack frame is being created; basically, this operand specifies the
number of stack frame pointers to be copied from the calling
code's stack frame into the new stack frame.
1111"
A RET instruction is required after LEAVE in order to return to
the calling code; LEAVE discards the current stack frame, but does
not perform a return.
Warningl
ENTER and LEAVE do not preserve any of the calling code's
registers; PUSH and POP or PUSHA and POPA should be used for
this purpose.
BOUND
BOUND checks that a 16-bit value is within a signed range
specified by two adjacent words of memory, with the upper
bound stored at the address immediately above the lower bound.
Both bounds are treated as signed values, so a maXimum range of
-32,768 to +32,767, inclusive, can be specified. Values matching
the upper and lower bounds are considered to fall within the
specified range.
BOUND is generally used to guard against attempts to access
before the beginning or past the end of an array. For example, this
code checks whether BX is in the range a to 99, inclusive, before
using it as an index into the laO-byte array TestArray .
•DATA
TestArrayBounds LABEL DWORD
DW
0
DW
99
TestArray DB
100 DUP (?)
;lower array bound (inclusive)
;upper array bound (inclusive)
.CODE
mov
ax,@data
Chapter 70, The 80386 and other processors
413
mov
ds,ax
bound bx, [TestArrayBounds]
mov aI, [TestArray+bx]
If BX is not in the range, an INT 5 is generated. An interrupthandler for INT 5 must, of course, be set up before BOUND can be
used.
The first operand to BOUND is the 16-bit general-purpose register
containing the value to be range-checked. The second operand to
BOUND is the doubleword containing the range. This doubleword
contains the signed 16-bit lower bound as its lower word and the
signed 16-bit upper bound as its upper word.
Warningl
INS and OUTS
One tricky point about BOUND is that the instruction pointer
pushed when INT 5 is generated by a failed bounds test points to
the BOUND instruction that caused the INT 5, not the following
instruction. If the failing condition is not corrected by the INT 5
handler before it executes an IRET, the same BOUND instruction
will generate another INT 5, and so on, indefinitely. Consequently,
INT 5 handlers for BOUND instructions should either issue a
message and terminate the program without executing an IRET or
correct the out-of-range condition before executing an IRET to
continue.
INS and OUTS support efficient data transfer between I/O ports
and memory.
INS moves one or more bytes (or words) from an I/O port
pointed to by DX to a memory array pointed to by E5:DI,
incrementing DI by 1 (or 2) after each byte (or word) is transferred
(or decrementing 51 if the direction flag is set). DX is not affected
by INS. As with all string instructions that write to memory, the
use of E5 as the destination segment cannot be overridden.
OUTS moves one or more bytes (or words) from a memory array
pointed to by D5:51 to an I/O port pointed to by DX,
incrementing 51 by 1 (or 2) after each byte (or word) is transferred
(or decrementing 51 if the direction flag is set). DX is not affected
by OUTS. A segment register other than DS can be selected with a
segment override prefix. The following code uses INSB to copy a
block of 300h bytes to memory from I/O port 3000h, then uses
OUTSB to copy that block of bytes to I/O port 3001h:
414
Turbo Assembler User's Guide
cld
mov
mov
mov
mov
mov
mov
rep
mov
mov
mov
rep
Extended 8086
instructions
ax,@data
ds,ax
es,ax
dx,3000h
di,OFFSET Buffer
cx,300h
insb
dx,3001h
si,OFFSET Buffer
cx,300h
outsb
;copy 300h bytes from buffer to port
The 80186 offers extended versions of several 8086 instructions as
well:
ROL
ROR
SAL
IMUL
PUSH
RCL
RCR
Pushing immediate
values
;copy 300h bytes to buffer from port
SAR
SHL
SHR
While the 8086 can push register or memory operands only, the
80186 can push an immediate value as well:
push
19
Pushing an immediate value is useful for passing constant
parameters to functions on the stack. For example, the 8086 code
for this C call,
Average(S, 2);
is this:
mov
push
mov
push
call
add
ax,2
ax
ax,S
ax
_Average
sp,4
And it can be reduced to this on the 80186:
push
push
call
add
2
S
_Average
sp,4
Chapter 70, The 80386 and other processors
415
Note that while the 8086 processor does not have a PUSH
immediate value instruction, Turbo Assembler 2.0's syntax allows
you to specify such an instruction in your source file. When the
PUSH instruction is encountered, it's replaced in the object code
by a 10-byte sequence, which simulates this operation while
preserving all registers and flags.
Shifting and rotating by
immediate values
While the 8086 can only rotate or shift by either 1 bit or the
number of bits specified by the contents of CL, the 80186 can
rotate or shift by a constant value:
ror
shl
ax,3
dl,7
This is convenient for performing multi-bit shifts without having
to load CL with the shift count. For example, the following 8086
code to multiply AX by 256,
mov
shl
el,S
ax,el
becomes this on the 80186:
shl
Multiplying by an
immediate value
ax,S
The 8086 can only multiply an 8- or 16-bit register or memory
operand by AL or AX, placing the result in AX or DX:AX. The
80186 provides two new forms of multiplication for use when the
product of a 16-bit multiplication will fit in 16 bits.
One new form of multiplication multiplies a 16-bit register by a
16-bit immediate value and stores the result back in the 16-bit
register. For example, this code multiplies DX by 4 and places the
product in DX:
imul dx,4
The first operand, which can be any 16-bit general-purpose
register, is both the source of one of the factors and the destination
for the product. The second operand, which must be a 16-bit
immediate value, is the other factor.
The other new form of multiplication multiplies a 16-bit register
or memory location by a 16-bit immediate value and stores the
416
Turbo Assembler User's Guide
result in a specified 16-bit register. For example, this code
multiplies DX by 600h and places the product in CX:
imul cx,dx,600h
Similarly, this code multiplies the 16-bit value at [BX+SI+1] by 3
and places the product in AX.
imul ax, [bxtsitl],3
The first operand to this form of IMUL is the destination for the
product. This operand can be any 16-bit general-purpose register.
The second operand, which can be any 16-bit general-purpose
register or memory location, is the source of one of the factors.
The third operand, which must be a 16-bit immediate value, is the
other factor.
A bit of thought will show that the first of the new forms of
multiplication is actually just a subset of the second new form. For
example, this following code,
imul si,lO
is just a shorthand form of
imul si,si,lO
The underlying hex code is the same for both new forms of the
IMUL instruction. Nonetheless, it's convenient to be able to use the
simpler two-operand IMUL when the same register serves as both
source and destination.
1111.
With either of the new forms of multiplication, any portion of the
result that does not fit in 16 bits is lost; if significant bits are lost
(when the result is a signed value), the carry and overflow flags
are set. The new forms of multiplication make no distinction
between signed and unsigned multiplication, since the result is
only 16 bits long, and the lower 16 bits of the product of both
signed and unsigned 16-bit by 16-bit multiplies are always the
same. Consequently, only the IMUL instruction can be used to
denote the new forms of multiplication.
The 80286
The 80286 was the first iAPx86-family processor to eliminate the
1-MB memory limitation and the first to support memory
protection and virtual memory. The 80286 provides all the
Chapter 70, The 80386 and other processors
417
instructions of the 8086 and 80186, and adds a number of
instructions that support management of a sophisticated memory
architecture.
The 80286 has two modes of operation: real mode and protected
mode. An 80286 operating in real mode is much like an 80186,
providing exactly the same instruction set and nothing more. This
is the mode in which 80286-based computers, such as the IBM AT,
run PC-OOS and applications such as Quattro and Turbo Pascal.
The memory management features of the 80286 are available only
in protected mode. And it's only in this mode that multiple
programs can be run at once without interfering with each other,
and more than 1 MB of memory can be addressed. This is the
mode in which 80286-based computers run OS/2.
Here are the protected-mode instructions of the 80286:
CLTS
LGDT
LlDT
LLDT
LMSW
LTR
These 80286 instructions are intended for operating system usage
only; applications should never need to (or be able to) use
protected-mode instructions. The use of these instructions and the
protected mode of the 80286 in general are specialized and
complex topics that we won't go into in this manual.
The 80286 adds two new status states to the flags register: the
nested task bit and the I/O privilege-level field. Like the
protected-mode instructions, both bits are intended for use by
systems software only and are of no concern to the applications
programmer. The 80286 also contains several new registers that
can be manipulated only with protected-mode instructions, such
as the Task register, the Machine Status Word register, and the
Global Descriptor Table register; again, these registers are not
used by applications, so we will not cover them in this manual.
Enabling 80286
assembly Turbo Assembler support for assembly of nonprotected-mode
80286 code is enabled with the .286 directive. (The .286C directive
also enables Turbo Assembler support for 80286 instructions, for
compatibility with earlier assemblers.)
Note that the .286 directive implicitly enables support for a118086
and 80186 instructions, since the 80286 supports the full
instruction set of earlier iAPx86-family processors.
418
Turbo Assembler User's Guide
Support for protected-mode 80286 instructions is enabled with the
.286P directive. Nonprotected-mode 80286 instructions are
enabled by the .286P directive as well, just as if a .286 directive
had been executed.
For detailed information
about 80286 Instructions,
refer to Chapter 3 of the
Reference Guide.
One important point about protected-mode 80286 instructions is
that the 8086 and 80186 do not recognize any of these instructions.
Consequently, any program that uses protected-mode instructions
won't run on an 8086 or 80186. However, the 80386 does support
both the protected-mode and nonprotected-mode instructions of
the 80286.
The 80386
The 80386 processor is a landmark in the evolution of the
microcomputer, providing new and extended instructions, an
expanded set of 32-bit registers, linear segments up to 4 gigabytes
long, the ability to emulate multiple 8086 processors
simultaneously, a barrel shifter for fast shifts and rotates, paged
memory, higher clock speeds than any previous iAPx86-family
processor (resulting in faster execution), and more. As you might
expect, extensions to 8086/80186/80286 assembly language are
needed to support the full power of the 80386. Turbo Assembler
provides a full set of 80386 extensions, supporting all modes and
features of the 80386.
The 80386 is a remarkably sophisticated processor-orders of
magnitude more complex than the 8086-so we can't cover the
many aspects of programming the 80386. We can, however, take a
look at the 80386 support built into Turbo Assembler.
Selecting 80386
assembly. mode
As with the 80286, there are two sorts of 80386 instructions,
privileged and non-privileged. Any program can execute nonprivileged instructions, while only programs executing at a
current privilege level of 0 (the most-privileged level) can execute
privileged instructions. The privileged instructions of the 80386
are a superset of the 80286's privileged instructions and, like 80286
privileged instructions, are intended for operating system use
only.
Support for non-privileged 80386 instructions is enabled with the
.386 directive. (The .386C directive enables Turbo Assembler
Chapter 70, The 80386 and other processors
419
support for 80386 instructions for compatibility with earlier
assemblers.)
1111"
The .386 directive implicitly enables support for all 8086 and
80186 instructions and all 80286 non-privileged instructions, since
the 80386 supports the full instruction set of earlier iAPx86-family
processors.
Support for privileged 80386 instructions is enabled with the
.386P directive. Non-privileged 80386 instructions are enabled by
the .386P directive as well, just as if a .386 directive had been
executed. Since the 80386 supports all privileged instructions of
the 80286, the .386P directive implicitly enables support for all
80286 privileged inshuctions.
New segment
types
The ability of the 80386 to support either 80286-style 64K
segments or linear segments up to 4 gigabytes (GB) in length
requires two new segment types, USE16 and USE32.
A 16-bit offset, either stored in a base or index register (BX, 51, DI,
or BP) or used as a direct addressing offset, is all that's needed in
order to point to any location in a 64K segment. This is the mode
of operation of the 80286 (and the 8086). 80386 segments that have
a maximum length of 64K are given a use type of USE16, as
follows:
.386
DataSeg
Varl DW
Ptrl DW
DataSeg
SEGMENT
USEl6
?
Varl
ENDS
CodeSeg SEGMENT USEl6
ASSUME
cs:CodeSeg
mov ax,DataSeg
mov fs, ax
ASSUME
fs:DataSeg
mov [Varl],O
mov bx, [Ptrl)
inc WORD PTR fs:[bx]
CodeSeg
420
;set Varl to zero
;load a l6-bit pointer to Varl
;increment Varl
ENDS
Turbo Assembler User's Guide
Note the use of FS, one of the two new extra segments (along with
GS) available on the 80386.
Note also that an offset stored in any of the 80386's eight generalpurpose 32-bit registers can be used to address a USE16 segment,
as long as the magnitude of the offset doesn't exceed OFFFFh
(65535).
A 32-bit offset, stored in any of the eight general-purpose 32-bit
registers or used as a direct addressing offset, is needed to point
to any given location in a 4 GB segment. 80386 segments that have
a maximum length of 4 GB are given a use type of USE32, as
follows:
.386
BigDataSeg
SEGMENT
Varl DW ?
Ptrl DD Varl
BigDataSeg
ENDS
USE32
CodeSeg SEGMENT USEl6
ASSUME
cs:CodeSeg
mov ax,BigDataSeg
mov fs,ax
ASSUME
fs:BigDataSeg
mov [Varl],O
mov eax, [Ptrl]
inc WORD PTR fs: [eax]
CodeSeg
iset Varl to zero
iload 32-bit pointer to Varl
iincrement Varl
ENDS
Note the use of EAX as a pointer register; the 80386 allows all
eight general-purpose 32-bit registers (EAX, EBX, ECX, EDX, ES1,
ED1, EBP, and ESP) to be used as either base or index registers, as
discussed in 'New addressing modes" on page 431.
The SMALL and LARGE operators can be used to override the
default offset size of a given operand. SMALL forces the use of a
16-bit offset, and LARGE forces the use of a 32-bit offset. For
example,
.386
CodeSeg SEGMENT USEl6
ASSUME
cs:CodeSeg
mov ax,DataSeg
mov ds,ax
Chapter 10, The 80386 and other processors
421
ASSUME
ds:DataSeg
mov ax, [LARGE TestLoc]
CodeSeg
ENDS
DataSeg
TestLoc
DataSeg
SEGMENT USE32
DW 0
ENDS
successfully makes a forward reference to TestLoc (even though
TestLoc is in a USE32 segment> by using LARGE to force the
reference to TestLoc to be performed with a 32-bit offset. Without
the LARGE override, an error would be generated, since the
assembler assumes 16-bit offsets for forward references made
within the USE16 segment CodeSeg.
See •New addressing
modes· on page 431.
The action of SMALL and LARGE is actually a bit more subtle than
a simple selection between 16- and 32-bit offset size. SMALL
instructs Turbo Assembler to assemble a given instruction for use
with the 8086's 16-bit addressing modes, which are inherently
capable of addressing only 64K of memory. LARGE, on the other
hand, instructs Turbo Assembler to assemble a given instruction
to use the 80386's new 32-bit addressing modes, which are capable
of addressing 4 GB of memory.
For example, the code
.386
CodeSeg
SEGMENT
USE16
mov ax, [SMALL ebxtesitl]
CodeSeg
ENDS
assembles to
mov
ax, [bxtsitl]
Here, SMALL told Turbo Assembler to use an 8086-style 16-bit
addressing mode, so instead of EBX and E5I, the assembled code
uses BX and 51. However, the code
.386
422
Turbo Assembler User's Guide
CodeSeg
SEGMENT
USE16
mov ax, [SMALL eax+ecx+lj
CodeSeg
ENDS
will not assemble, since EAX+ECX+1 is not a valid 16-bit memory
addressing mode. (On the other hand, EAX+ECX+ 1 is a valid 32bit memory addressing mode, as you will see in the section ''New
addressing modes.")
Take a look at the section, "Mixing 16-bit and 32-bit instructions
and segments," on page 446 for more information about SMALL
and LARGE and for information regarding the interaction of small
and large operators with USE16 and USE32 segments. The issue
of selection between USE32 and USE16 segments is also covered
in that section.
One important implication of the selection of USE16 or USE3~
segments concerns the size of indirect jumps. You'll find out
about this in the section entitled "The 32-bit instruction pointer"
(page 428).
If neither USE32 nor USE16 is specified in a segment definition,
USE32 is always assumed when assembling for the 80386.
Simplified segment
directives and 80386
segment types
If you use both .386 and the simplified segment directives,
segments default to DWORD alignment. This makes sense, given
that 80386-based computers run fastest with doubleword-aligned
data.
When you use the simplified segment directives, Turbo
Assembler generates USE32 segments if .386 is given before the
.MODEL directive, and USE16 segments if .386 is given after the
.MODEL directive. For example, this code creates 32-bit code and
data segments:
.386
. MODEL
. DATA
large
. CODE
while this code creates 16-bit code and segments:
. MODEL
large
Chapter 10, The 80386 and other processors
423
.386
. DATA
•CODE
The FWORD 48-bit data
type
An interesting point about USE32 segments is that the size of a far
pointer (that is, a full segment:offset pointer) to a location in a
USE32 segment is 6 bytes rather than the customary 4 bytes
because offsets in USE32 segments are 32 bits in size. For
example, with a USE16 segment, a far pointer to an 8000h-byte
buffer Buffer is stored in 4 bytes and loaded as follows:
.386
DataSeg
SEGMENT USE16
Buffer
DB 8000h DUP (?)
BufferPtr LABEL
miORO
DW OFFSET Buffer
DW SEG Buffer
DataSeg
ENDS
CodeSeg
SEGMENT USE16
ASSUME
cs:CodeSeg
mov ax,DataSeg
mov ds,ax
ASSUME
ds:DataSeg
les bx,[BufferPtr]
;load ES:BX with 16-bit segment
; and 16-bit offset of Buffer
CodeSeg
ENDS
With a USE32 segment, on the other hand, a far pointer to Buffer is
stored in 6 bytes and loaded as follows:
.386
DataSeg
SEGMENT USE32
Buffer
DB 8000h DUP (?)
BufferPtr LABEL
FWORO
DD OFFSET Buffer
DW SEG Buffer
DataSeg
ENDS
CodeSeg
SEGMENT USE32
ASSUME
cs:CodeSeg
mov ax,DataSeg
424
Turbo Assembler User's Guide
mov ds,ax
ASSUME
ds:DataSeg
les ebx, [BufferPtr]
CodeSeg
1111"
;load ES:EBX with 16-bit segment
; and 32-bit offset of Buffer
ENDS
Note the use of the new FWORD data type. FWORD values are 6
bytes long. FWORD PTR operators can be used just like BYTE
PTR, WORD PTR, and DWORD PTR operators.
19s
esi,FWORD PTR [BufferPtr]
There is also a new directive, OF, for defining 6-byte variables:
.386
DataSeg
FPtr OF
DataSeg
SEGMENT USE32
?
ENDS
CodeSeg SEGMENT USE32
ASSUME
cs:CodeSeg
mov ax, DataSeg
mov ds,ax
ASSUME
ds:DataSeg
mov eax,OFFSET DestinationFunction
mov DWORD PTR [FPtr],eax
mov ax,SEG DestinationFunction
mov WORD PTR [FPtr+4],ax
jmp [FPtr]
CodeSeg
ENDS
New registers
The 80386 extends the general-purpose registers, flags register,
and instruction pointer of the 8086 to 32 bits in size, and adds two
new segment registers. Figure 10.1 shows the register set of the
80386; the 80386 extensions to the basic 8086 register set are
shaded.
In addition, the 80386 contains several special registers, some new
and some compatible with the 80286, that can be manipulated
only with privileged instructions. As with the 80286, these
registers are used only by systems software, so we won't cover
them in this manual.
Chapter 70, The 80386 and other processors
425
Figure 10.1
The registers of the
80386
31
16 15
EAX
EBX
ECX
EOX
ESI
EOI
EBP
ESP
EIP
EFLAGS
=I~ !~!~.=!~!~!~!~ =~!~!~!~ ~=!~!~!~ ~: !~!~!~ ~ =~ ~!~ ~!~=!~!~!~!~: !~!~!~!~!=~!~j~ ~ ~=!~!~!~!~=~ ~ j ji~= =I=P====~I <~:I~~e~tion
<
t;:; ;:lj ~ j~j=~ ~j~ ~ =~ ~ ~ ~ ~=~ ~ ~j~ =~j~ ~ ~ ~=~ ~ ~j ~ =~ ~ ~ ~ :=~ ~ ~ ~ ~=j~j~j~j~j=~j ~j~j ~=t:!j~ ~ 1L. - _F_LA_G_S_- -I ~:~I~t er
31
16 15
0
15
o
CS
OS
ES
I
I
GS
Ij j ~j~1~1~ 1 ~1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1!~!~!~!~!~!~!j!j!j!j!j!j!j!1!j!j!j!1!j!j!~!j!j!j!j!~!j!~!j!1!1!1!~1~!~1~!~!~j~!~!j!j!~!~ ~1
SS
I
FS
Segment
Registers
I
o
15
Let's examine the new registers of the 80386.
The 32-bit generalpurpose registers
426
The 32-bit versions of the general-purpose registers are called
EAX, EBX, ECX, EDX, ESI, EDI, EBP, and ESP. The lower 16 bits
of these registers form the 8086's set of 16-bit registers we've come
to know so well; for example, the lower 16 bits of EAX are register
Turbo Assembler User's Guide
AX. Similarly, the lower 8 bits of EAX are register AL.
Consequently, portions of register EAX may now be referred to by
four different names: the 32-bit EAX register, the 16-bit AX
register, and the 8-bit AH and AL registers. The same is true of
EBX, ECX, and EDX.
The 32-bit general-purpose registers of the 80386 are used in the
same way as the 16- and 8-bit registers. For example, this code
stores 1 in EAX, sets EBX to 0, and adds EAX to EBX:
rnov eax,l
sub ebx,ebx
add ebx,eax
The 32-bit general-purpose registers can be used wherever the
familiar 16-bit registers can be used.
There is one slight shortcoming in accessing 32-bit registers:
There's no way to use the upper 16 bits of a 32-bit register diiectly
as a 16-bit register. If you want to use the upper 8 bits of AX as a
register, you can just refer to AH; and if you want to use the lower
16 bits of ESI as a register, you can just refer to SI. But there's no
equivalent way to refer to the upper 16 bits of, say, EAX. This can
be a nuisance when you're working with a mixture of word- and
doubleword-sized values, but there is a reasonable workaround.
To access the upper 16 bits of a 32-bit register, just rotate the
register 16 bits in either direction, access the lower 16 bits of the
register, and rotate the register 16 bits again. For instance, the
following code loads a 16-bit value into AX, rotates EDX 16 bits to
swap the high and low words of EDX, moves AX into DX, and
swaps the high and low words of EDX again.
rnov
ror
rnov
ror
ax, [Sarnple16BitValue]
edx,16
dx,ax
edx,16
.The net effect: The value initially loaded into AX is ultimately
moved into the high word of EDX. While this procedure is
awkward, it is not as slow as it might seem; thanks to the 80386's
barrel shifter, each ROR instruction takes only three cycles to
execute.
Chapter 70, The 80386 and other processors
427
The 32-bit flags register
The lower word of the 80386's flags register is identical to the
80286's flags register. The upper 16 bits of the 80386's flags register
contains two new flags. One of the new flags indicates whether
the 80386 is currently executing as a virtual 8086, and the other
new flag is intended for use in writing debugging tools. These
flags are generally not used by applications software.
The 32-bit instruction
pointer
The 80386's instruction pointer is 32 bits in size, in contrast to the
8086's 16-bit instruction pointer. This extended instruction pointer
supports code segments up to 4 GB in length.
The 80386's extended instruction pointer creates some
complications in specifying indirect jumps via memory. For
example, the following code clearly specifies a far indirect jump
with a 16-bit segment and a 32-bit offset:
jmp
[FWORD PTR JumpVector]
Consider the following, however:
jmp
[OWORD PTR JumpVector]
Is this a near 32-bit indirect jump or a far indirect jump with a 16bit segment and a 16-bit offset? Either type of jump may
legitimately be specified with a DWORD operand.
Here's where the LARGE and SMALL operators come in handy.
The construct
jmp
SMALL [OWORD PTR JumpVector]
assembles as a far indirect jump to the address specified by the
16-bit segment and 16-bit offset stored at /umpVector, and
jmp
LARGE [OWORD PTR JumpVector]
assembles as a near indirect jump to the address specified by the
current CS and the 32-bit offset stored at /umpVector. In the first
case, the SMALL operator instructs Turbo Assembler to treat the
jump as if it were occurring from a USE16 segment; in USE16
segments, 32-bit indirect jump operands consist of a 16-bit
segment and a 16-bit offset. In the second case, the LARGE
operator instructs Turbo Assembler to treat the jump as if it were
occurring in a USE32 segment; in USE32 segments, 32-bit indirect
jump operands consist of 32-bit offsets only.
428
Turbo Assembler User's Guide
1111.,
Note that SMALL and LARGE appear outside the brackets in the
preceding examples; the positioning of SMALL and LARGE is
significant. When SMALL and LARGE appear outside the
brackets, they affect the operand size, in this case, the size of the
jump. When SMALL and LARGE appear inside the brackets, they
affect the address size. For example, this code instructs Turbo
Assembler to use a near 32-bit offset to point to lumpVector, but
does not tell Turbo Assembler whether to treat the value stored at
lump Vector as a near 32-bit offset or a far 16-bit segment and 16-bit
offset combination:
jrnp
[LARGE DWORD PTR JurnpVector]
50 this does not resolve the original problem of determining the
type of the jump.
1111.,
LARGE and SMALL can be used both inside and outside the
brackets in a single expression. For instance, this code specifies a
far indirect jump to the 16-bit segment and 16-bit offset address
stored at the doubleword variable lumpVector, which is itself
addressed with a near 32-bit offset:
jrnp
New segment registers
SMALL [LARGE DWORD PTR JurnpVector]
The 80386 adds two new segment registers, FS and G5 to the four
segment registers supported by the 8086. The two new segment
registers are not dedicated to any particular function, and no
instructions or addressing modes access FS or G5 by default.
Consequently, the use of FS or G5 is never required, but they can
be handy for code that accesses data in several segments at once.
FS and GS are used just as E5 is used for nonstring instructions,
by means of a segment override prefix. The override prefix may
be explicit:
.386
TestSeg SEGMENT USEl6
SCRATCH_LEN EQU
lOOOh
Scratch DB SCRATCH LEN DUP (?)
TestSeg ENDS
CodeSeg SEGMENT USEl6
ASSUME
cs:CodeSeg
rnov ax,TestSeg
rnov fs,ax
rnov bx,OFFSET Scratch
Chapter 70, The 80386 and other processors
429
rnov cx,SCRATCH_LEN
rnov al,O
Clear Scratch:
rnov fs: [bx] , al
inc bx
loop ClearScratch
CodeSeg
ENDS
or implicit, by way of an ASSUME directive:
.386
TestSeg SEGMENT USE16
SCRATCH_LEN EQU
lOOOh
Scratch DB SCRATCH LEN DUP (?)
TestSeg ENDS
CodeSeg SEGMENT USE16
ASSUME
cs:CodeSeg
rnov ax,TestSeg
rnov gs,ax
ASSUME gs:TestSeg
sub bx,bx
rnov cx,SCRATCH_LEN
rnov al,O
Clear Scratch:
rnov [Scratch+bx],al
inc bx
loop Clear Scratch
CodeSeg
ENDS
In the last example, the directive ASSUME GS:TestSeg told Turbo
Assembler to insert an override prefix automatically on each
access by name (as opposed to access by pointer register) to
variables in TestSeg, so you didn't have to type the override prefix
explicitly. The override prefix is, however, still there in the
executable code, adding an extra byte to the size of each
instruction that accesses memory by way of the FS or GS register.
Consequently, whenever possible, it's preferable to use the DS
register (or the ES register as the destination of a string
instruction) instead of the FS or GS register.
430
Turbo Assembler User's Guide
New addressing
modes
The 80386 supports all the memory addressing modes of the 8086,
80186, and 80286, and adds a set of powerful new addressing
modes as well. Any of the eight 32-bit, general-purpose registers
of the 80386 may be used as a base register, and any 32-bit,
general-purpose register other than 5P may be used as an index
register. By contrast, the 8086 allows only BX and BP to be used as
base registers, and only 51 and 01 to be used as index registers.
For example, suppose that EDI contains 1oo00h and EAX contains
4. Then the following code is a perfectly legal instruction on the
80386, incrementing the byte at offset 10006h (lOooOh + 4 + 2) in
the segment pointed to by os:
inc
BYTE PTR [editeaxt2]
Here's another example of the 80386's new addressing capabilities:
mov ecx, [espt4]
mov ebx, [espt8]
mov WORD PTR [ecxtebx],O
The 80386 can do still more in the new addressing modes,
however. The index register can be multiplied by 2, 4, or 8 as part
of the calculation of the memory address, simply by placing *2, *4,
or *8 after the index register, a feature known as index scaling. For
instance, the ninth doubleword-sized entry in the table
DwordTable can be loaded into EAX with this code:
mov ebx,8
mov eax, [DwordTabletebx*4]
which is equivalent to
mov
shl
mov
shr
ebx,8
ebx,2
eax, [DwordTabletebx]
ebx,2
Index scaling can be extremely useful for accessing elements in
word, doubleword, and quad word arrays. For example, consider
Chapter 70, The 80386 and other processors
431
the following code, which sorts the elements in a word array in
ascending order:
.386
CodeSeg SEGMENT USE32
ASSUME
cs:CodeSeg
Sorts a word array in ascending order.
Input:
DS:EBX - pointer to start of word array to sort
EDX - 1ength of array in word elements
Registers destroyed:
AX, ECX, EDX, ESI, EDI
SortWordArray PROC
and edx,edx
jz EndSortWordArray
mov esi,O
SortOnNextWord:
dec edx
jz EndSortWordArray
mov ecx,edx
mov edi,esi
CompareToAIIRemainingWords:
inc edi
mov ax, [ebx+esi*2]
cmp ax, [ebx+edi*2]
jbe
xchg
mov
NoSwap:
loop
inc
NoSwap
ax, [ebx+edi*2]
[ebx+esi*2],ax
CodeSeg
ENDS
icompare element 0 to all other
; elements first
;count down number to compare
;number of elements to compare
; this element against
;compare this element to all
; remaining elements
;index of next element to compare
;is the current element less
; than the compared element?
;yes, no need to swap them
;swap the current and
; compared elements
CompareToAllRemainingWords
esi
;point to next element to compare
; to all remaining elements
jmp SortOnNextWord
EndSortWordArray:
ret
SortWordArray ENDP
432
Turbo Assembler User's Guide
Sortlt\brdArray keeps the element numbers, or indexes, of the
current and compared elements in ES1 and ED!. These values are
not pointers, or counts by two, even though the array is a word
array; rather, they are simple scalar array indexes, just as n is an
array index in the C statement
i
= Array[n]i
The key in SortWordArray is that the index scaling feature of the
80386 allows you to multiply the indexes by two as part of the
memory addressing field, thereby converting the indexes to
offsets into a word array.
1111"
If only one register is used to address memory, that register is
always considered to be the base register. If two registers are used
to address memory, the leftmost register inside the brackets is
considered the base register, and the rightmost register is
considered the index register. If, however, scaling is used with <:>ne
of two registers inside the brackets, the scaled register is always
considered to be the index register.
The question of which register is the base register is important
because by default the base register controls the segment to which
a given memory access refers. Memory accesses made with EBP
or ESP as the base register refer to the segment pointed to by 55,
while memory accesses made with EAX, EBX, ECX, EOX, ES1, or
E01 as the base register refer to the segment pointed to by OS. For
example, the following instructions refer to OS:
mov
xehg
shr
mov
sub
aI, [eax]
edx,[ebx+ebp]
BYTE PTR [esi+esp+2],1
[ebp*2+edx],ah
ex, [esi+esi*2]
and the following instructions refer to 55:
rol WORD PTR [ebp],l
dee DWORD PTR [esp+4]
add ax, [eax*2+esp]
mov [ebp*2],edi
The default segment selected by the base register can be
overridden with either an explicit segment override prefix or as
the result of an ASSUME directive. For example,
.386
TestSeg
SEGMENT
Chapter 10, The 80386 and other processors
USE32
433
Arrayl
TestSeg
DW 100h DUP (0)
ENDS
CodeSeg SEGMENT USEl6
ASSUME
cs:CodeSeg
mov ax,TestSeg
mov fs,ax
ASSUME
fs:TestSeg
mov dx, [ebx+Arrayl]
;implicit override as a result of
; ASSUME
mov
mov
IncLoop:
inc
inc
inc
loop
esi,OFFSET Arrayl
cx,lOOh
WORD PTR fs:[esi]
esi
esi
IncLoop
CodeSeg
ENDS
;explicit override
The new addressing modes of the 80386 work with 32-bit .
memory-addressing registers only; 16-bit registers can only be
used for memory addressing in the same limited way that they
are on the 8086. For example, the following MOV instruction is
illegal, even on an 80386:
mov
ax, [cx+dx+lOh]
Index scaling of 16-bit registers is also not allowed. And 16- and
32-bit registers can't be combined for memory-addressing
purposes; so, for example, this code cannot be used:
add dx, [bx+eax]
New instructions
For detailed information
about 80386 instructions, see
Chapter 3 In the Reference
Guide.
Next, we're going to take a look at the new and extended
instructions of the 80386.
Keep in mind that the 8086,80186, and 80286 do not recognize
any of the new and extended instructions we're about to discuss.
Consequently, any program that uses the new or extended
instructions of the 80386 won't run on earlier processors.
Here are the new instructions of the 80386:
BSF
BSR
434
BTR
BTS
LFS
LGS
MOVZX
SETxx
Turbo Assembler User's Guide
BT
BTC
Testing bits
CDa
CWDE
LSS
MOVSX
SHLD
SHRD
The bit-test instructions of the 80386 are BT, BTC, BTR, and BTS.
BT is the basic bit-test operation, copying the value of a specified
bit into the carry flag. For example, the following code jumps to
Bit3Isl only if bit 3 of EAX is nonzero:
bt
jc
eax,3
Bit3Isl
Bit3Isl:
If EAX contains 0OO00008h, this code will jump to Bit3Isl; if EAX
contains OFFFFFFF7h, the preceding code will not jump. The first
operand to BT is the 16- or 32-bit, general-purpose register or
memory location containing the bit to test. The second operand is
the bit number to test, specified by either an 8-bit immediate
value or the contents of a 16- or 32-bit, general-purpose register. If
a register is used as the second operand, its size must match the
size of the first operand.
Note that the number of the bit to test can be specified by a
register as well as an immediate value, and the field to be bittested can be in memory as well as in a register. Here's a valid way
to set the carry flag to the state of bit 5 of the word at the address
Table+ebx+esi .. 2:
mov ax,S
bt WORD PTR [Tabletebxtesi*2],ax
Remember that bit numbers are counted from zero at the leastsignificant bit up to the most-significant bit. If AL contains 80h,
then bit 7 of AL is set.
BTC is exactly like BT except that the value copied to the carry
flag is the complement of the specified bit. That is, the carry flag is
1 if the specified bit is 0, and the carry flag is 0 if the specified bit
is 1. BTC saves the need for a CMC instruction whenever a carry
status is required that is the inverse of the bit under test. '
BTR is also just like BT except that the specified bit is set to 0 after
its value is copied to the carry flag. Similarly, BTS sets the
specified bit to 1 after its value is copied to the carry flag. These
Chapter 70, The 80386 and other processors
435
bit-test instructions are useful for both testing and setting the
status of a flag in a single indivisible instruction. (By indivisible,
we mean that it is impossible for an interrupt to occur between
the testing of the flag and the setting of the flag to the new value.)
Scanning bits
The BSF and BSR instructions of the 80386 are useful for finding
the first or last bit that is nonzero in a word or dword operand.
BSF scans the source operand, starting with bit 0 (the leastsignificant bit), for the first bit that is nonzero. If all bits in the
source operand are zero, the zero flag is cleared; otherwise, the
zero flag is set and the bit number of the first nonzero bit found is
loaded irtto the destination register.
As an example, this code uses BSF to locate the first (leastsignificant) nonzero bit in DX; since the first nonzero bit in DX is
located at bit 2, a 2 is loaded into CX.
mov
bsf
jnz
shr
dx,OOOllOlOlOlOllOOb
cx,dx
AIIBitsAreZero
dx,cl
AllBit sAreZero:
CL is then used as the value to shift DX by, with the result that DX
is shifted to the right by exactly the amount needed to move the
least-significant nonzero bit to bit O.
The second operand to BSF is the 16- or 32-bit, general-purpose
register or memory location to scan, and the first operand is the
16- or 32-bit, general-purpose register in which to store the
number of the first nonzero bit in the scanned data. Both
operands must be the same size.
BSR is similar to BSF except that BSR scans from the mostsignificant bit of the source operand toward the least-significant
bit. In this example, the index of the most-significant nonzero bit
in TestVar,27, is placed in EAX:
TestVar
DD
OFFFFFOOh
bsr eax, [TestVar]
436
Turbo Assembler User's Guide
Moving data with sign- MOVZX and MOVSX allow you to copy an 8- or 16-bit value into a
or zero-extension . 16- or 32-bit, general-purpose register without wasting
instructions on extending the value to the destination size.
MOVZX pads out the most-significant bits of the destination with
zeros, while MOVSX sign-extends the value to the destination's
size. Both instructions are used just like a standard MOV.
For example, with 8086 instructions, the following is required to
copy an unsigned value in DL to BX:
mov bl,dl
sub bh,bh
while on the 80386, the single instruction
movzx
bx,dl
does the job. Sign-extension is even tougher with 8086
instructions. To copy the signed byte-memory variable TestByte to
DX without MOVSX, the following is required:
mov aI, [TestByte]
cbw
mov dx,ax
but MOVSX does the job with just one instruction:
movsx
dx, [TestByte]
MOVZX and MOVSX can also move 8-bit values to 32-bit registers:
movsx
Converting to DWORD
or QWORD data
eax,al
The 8086 provides the CBW and CWO instructions for converting
signed byte values in AL to signed words, and signed word
values in AX to signed doublewords, respectively. The 80386 adds
two more signed conversion instructions, CWDE and COQ, which
make good use of the 80386's 32-bit registers.
CWDE converts a signed word value stored in AX into a signed
doubleword value, just as CWO does. The difference between the
two is that while CWO places the 32-bit result in DX:AX, CWOE
places the 32-bit result in EAX, where it can readily be
manipulated by the 80386's 32-bit instructions.
Chapter 70, The 80386 and other processors
437
For example, the end result of
mov ax,-l
ewde
is the 32-bit value -1 in EAX.
CDa converts a signed doubleword value in EAX into a signed
quad word (8-byte) value in EDX:EAX. The code
mov eax,-7
edq
stores the value -7 in the 64-bit register pair EDX:EAX, with the
high doubleword of the result, OFFFFFFFFh, stored in EDX, and
the low doubleword of the result, OFFFFFFF9h (-7), stored in EAX.
Shifting across multiple
words
Multiple-word shifts-for example, shifting a 32-bit value 4 bits to
the left-are a nuisance on the 8086, since each word must be
shifted one bit at a time, with bits flowing one by one from one
register to the next through the carry flag. The SHRD and SHLD
instructions of the 80386 remedy this situation by supporting
multiple-bit shifts across two registers, or between a register and a
memory location.
For example, suppose a 32-bit value is stored in DX:AX on an
8086. The following is required to shift that 32-bit value left
(toward the most-significant bit) by four bit positions: .
shl
reI
shl
rc1
shl
rcl
shl
rc1
ax,l
dx,l
ax,l
dx,l
ax,l
dx,l
ax,l
dx,l
On an 80386, the same result can be accomplished with just two
instructions:
shId dx,ax,4
438
Turbo Assembler User's Guide
shl ax,4
(Of course, the whole 32-bit value could simply have been stored
in EAX and shifted with
shl eax,4
but the example code was intended to illustrate the advantage of
using SHLD rather than 8086 instructions.)
The first operand to SHLD is the 16- or 32-bit, general-purpose
register or memory location to shift; the second operand is the 16or 32-bit, general-purpose register to shift bits in from; and the
third operand is the number of bits to shift by. The sizes of the
first and second operands must match. The third operand may be
either an immediate value or CL; in the latter case, the destination
is shifted the number of bits specified by CL.
SHRD is much like SHLD, but shifts from the most-significant bit
toward the least-significant bit. In this example, the 64-bit value
stored in TestQV\brd is shifted right by 7 bits:
mov
mov
shrd
shr
mov
Setting bytes
conditionally
el,7
eax,DWORD PTR [TestQword+4]
DWORD PTR [TestQword],eax,cl
eax,el
DWORD PTR [TestQword+4],eax
A common application for conditional tests and jumps is to set a
memory location to reflect a certain status. For instance, you may
want to set flags to indicate whether two variables are equal,
whether a pointer is null, or whether the carry flag was set by a
previous operation. The 8086 is less than ideal for such
operations, since multiple instructions (including time-wasting
jumps) are required to set a flag to reflect the results of a
conditional test. The 80386 provides the powerful group of SET
instructions to speed such test-and-set cases.
For example, imagine that you want to set the memory variable
TestFlag only if the most-significant bit of AX is set. On the 8086,
you would have to do the following:
mov [TestFlag],O
test ah,80h
Chapter 70, The 80386 and other processors
iassume the MSB isn't set
439
jz MSBNotSet
mov [TestFlag],l
MSBNotSet:
On the 80386, all you need do is this:
test ah,80h
setnz [TestFlag]
and TestFlag will be set to 1 if bit 7 of AH is 1, and to 0 if bit 7 of
AHisO.
You can test any of the familiar jump conditions with a SET
instruction: SETNC sets the destination to 1 if the carry flag is 0
and resets the destination to 0 if the carry flag is 1; SETS sets the
destination if the sign flag is 1 and resets it if the sign flag is 0; and
so on. The operand to a SET instruction may be an 8-bit, generalpurpose register or an 8-bit memory variable; 16- and 32-bit
operands are not permitted.
Loading SS, FS, and GS
The 8086 instruction LOS allows you to load both DS and one of
the general-purpose registers from memory with a single
instruction, thereby setting up a far pointer very efficiently. LES
provides a similar capability, but loads ES instead of D5. The
80386 adds three new instructions for loading far pointers: LSS,
LFS, and LGS, which load far pointers based on the 55, FS, and
G5 segment registers, respectively.
For example, this loads a far pointer to the video bit map at
AOOO:OOOO into GS:BX:
DataSeg
SEGMENT
ScreenPointer LABEL
dw 0
dw OAOOOh
DataSeg ENDS
USE16
DWORD
CodeSeg
SEGMENT USE16
ASSUME
cs:CodeSeg, ds:DataSeg
mov ax, DataSeg
mov ds,ax
19s bx, [ScreenPointer]
440
Turbo Assembler User's Guide
CodeSeg
ENDS
As with LOS and LES, either small or large far pointers may be
loaded with LSS, LFS, and LGSi see the section entitled "The
FWORD 48-bit data type" on page 424 for information about
small and large far po in ters.
Extended
instructions
The 80386 not only adds a number of powerful new instructions
to the 8086/80186/80286 instruction set, but extends a number of
existing instructions as well. The extended instructions follow:
CMPS
IMUL
INS
IRET
J,A
JAE
JB
JBE
JC
JCXZ
JE
JG
JGE
JL
JLE
JNA
JNAE
JNB
JNBE
JNC
JNE
JNG
JNGE
JNL
JNLE
JNO
JNP
JNS
JNZ
JO
JP
JPE
JPO
JS
JZ
LOOS
LOOP
MOV
MOVS
OUTS
POPA
POPF
PUSHA
PUSHF
SCAS
STOS
In addition, many instructions can handle 32-bit operands on the
80386, even though their mnemonics haven't explicitly changed.
Special versions of
MOV
The 80386 supports special forms of the MOV instruction that
allow code running at privilege level 0 (the most-privileged level)
to move data between the 32-bit, general-purpose registers and
special 80386 registers. Here are the 80386 registers that can be
accessed in this way:
CRO
CR2
CR3
DRO
DR1
DR2
DR3
DR6
DR7
TR6
TR7
For example, debug register DRO could be loaded with a linear
address to be trapped on with
.386P
mov eax,OFFSET FunctionEntry
mov drO,eax
and the system control flags could be loaded from control register
CRO into EDX with
Chapter 70, The 80386 and other processors
441
.386P
rnov edx,crO
Note that the .386P directive must be in effect in order for Turbo
Assembler to assemble the special forms of MOV, since they are
pri vileged instructions.
.
In general, the special 80386 registers that can be accessed by the
new forms of the MOV instruction are used by systems software
only, and are not used by applications.
32-bit versions of 8086
instructions
Many 8086 instructions are extended to take on new 32-bit
addressing and operand capabilities on the 80386. The following
code performs a 32-bit subtraction of the 32-bit EBX register from
the 32-bit variable at address EBP+EAX ,. 8+10h, with 32-bit
registers used to point to the destination memory location:
sub DWORD PTR [ebp+eax*8+10h],ebx
The 32-bit capabilities added to most 8086 instructions don't
require a new instruction mnemonic; the 32-bit nature of the
operation is generally indicated by the operands or by the
segment type the operation occurs in. Several 8086 instructions
do, however, require new mnemonics in order to support their
extended 32-bit, 80386 capabilities. We'll look at these instructions
next.
New versions of LOOP and JCXZ
The LOOP, LOOPE, LOOPNE, and JCXZ instructions normally
operate on the 16-bit CX register. The 80386 provides both 16-bit
and 32-bit versions of these instructions; the 32-bit versions'
operate on ECX rather than CX.
The LOOP, LOOPE, and LOOPNE instructions use either CX or
ECX as the loop counter, depending on whether the segment they
are in is a 16-bit or a 32-bit segment. If you want to make sure that
CX is always used as the loop control register, even in a 32-bit
segment, use the word form of these instructions: LOOPW,
LOOPWE, and LOOPWNE. Likewise, if you want to make sure
that ECX is always used as the loop control register, use the
double-word form of these instructions: LOOPD, LOOPDE, and
LOOPDNE.
442
Turbo Assembler User's Guide
LOOPD decrements ECX and jumps to the destination offset if the
resulting value is not zero. For example, the following loop is
executed 80000000h times:
mov ecx,80000000h
LoopTop:
loopd LoopTop
LOOPDE decrements ECX and jumps to the destination offset
while the zero flag is 1 and ECX is not zero. (LOOPDZ is another
form of the same instruction.) Similarly, LOOPDNE decrements
ECX and jumps to the destination offset while the zero flag is 0
and ECX is not zero. (LOOPDNZ is equivalent.) For instance, the
following loop repeats until either the value read from the I/O
port at DX becomes 09h or the port has been checked 10000000h
times, whichever comes first:
mov
LoopTop:
in
cmp
loopdne
jnz
ecx,lOOOOOOOh
al,dx
al,09h
LoopTop
TimedOut
TimedOut:
Note that the action of JNZ in this example reflects the result of
the comparison, not of LOOPDNE, since loop instructions don't
affect the status flags. The 80386 also provides a version of JCXZ
suited to 32-bit operations. Where JCXZ jumps if CX is zero,
JECXZ jumps if ECX is zero. For example, the following loop is
capable of handling 32-bit counts:
LoopTop:
jecxz LoopEnd
jmp
LoopEnd:
LoopTop
Chapter 70, The 80386 and other processors
443
New versions of the string InstrucHons
On the 80386, all string instructions may operate on byte, word, or
doubleword values. The doubleword versions of the string
instructions simply end with d rather than the usual w or b. The
new instructions follow:
CMPSD
INSD
LODSD
MOVSD
OUTSD
SCASD
STOSD
Each of these instructions works with 32 bits of data at a time, and
increments or decrements its associated pointer registers by four
on each repetition. For example, the following code fragment uses
MOVSD to copy the two doublewords starting at the offset
DwordTable to the two doublewords starting at the offset Buffer:
eld
mov
mov
mov
rep
si,OFFSET DwordTable
di,OFFSET Buffer
ex,2
movsd
This produces the same result as the following code, which uses
MOVSB:
eld
mov
mov
mov
rep
1111"
si,OFFSET DwordTable
di,OFFSET Buffer
ex,S
movsb
One way to think of the doubleword string instructions is that
their relationship to the word string instructions is similar to that
of the word string instructions to the byte string instructions.
IRETD
IRETD is similar to IRET. It pops EIP, then CS as a doubleword
(discarding the higher word), then EFLAGS as a doubleword.
444
Turbo Assembler User's Guide
PUSHFD and POPFD
PUSHFD pushes the fu1l32-bit flags register of the 80386 onto the
stack. POPFD pops the fu1l32-bit flags register from the stack.
By contrast, PUSHF and POPF push and pop only the lower 16
bits of the flags register.
PUSHAD and POPAD
PUSHAD pushes the eight 32-bit general-purpose registers onto
the stack in the following order: EAX, ECX, EOX, EBX, ESP, EBP,
ESI, ED!. The value pushed for ESP is the value of ESP at the start
of the PUSHAD instruction. POPAD pops seven of the eight 32-bit,
general-purpose registers from the stack, reversing the order of
PUSHAD so that EOI, ESI, EBP, EBX, EOX, ECX, and EAX can be
saved with PUSHAD and then restored with POPAD. ESP is not
restored by POPAD, but instead is incremented by 32 to discard
the block of the eight 32-bit, general-purpose registers previously
pushed by PUSHAD from the stack. The previously pushed value
of ESP is ignored.
By contrast, PUSHA and POPA push and pop only the lower 16
bits of the eight general-purpose registers.
New versions of IMUL
In addition to the 8086/80186/80286 forms of IMUL, the 80386
provides what is perhaps the most convenient form of IMUL yet:
Any general-purpose register or memory location can be
multiplied by any general-purpose register with the result stored
back in one of the source registers. Gone is the need to have one
of the operands be a constant, or for the accumulator to be the
destination. For example,
.'
irnul ebx, [edi*4+4]
multiplies EBX by the doubleword value stored at memory
address edi *4+4, and stores the result back into EBX.
As you can see, the first operand to this form of IMUL is the
destination register; this operand may be any 16- or 32-bit,
general-purpose register. The second operand may be any 16- or
32-bit, general-purpose register or memory location. The sizes of
the two operands must match. The overflow and carry flags are
set to 1 if the result, considered a signed value, is too large for the
destination.
Chapter, 70, The 80386 and other processors
445
As you might expect, the 80386 also extends the 8086/80186/
80286 forms of IMUL to support 32-bit operands. For example, this
code multiplies ECX times 10000000h and stores the result in EBP:
imul ebp,ecx,lOOOOOOOh
and this multiplies EAX times EBX and stores the result in
EDX:EAX:
imul ebx
Mixing 16-bit and
32-bit instructions
and segments
Normally, you'll want to have only 16-bit (USE16) segments. Even
in this case, you can still use the 32-bit registers for arithmetic and
logical operations.
You can also use any combination of 16-bit and 32-bit data and
code segments. Unless you are writing operating system software
and know exactly what you are doing, there is absolutely no
reason for you to use 32-bit code segments. Unless you take
special measures to switch the processor into a mode suitable for
executing 32-bit code segments, there is no way they'll work
under DOS. Future operating systems may give you ways to
meaningfully use 32-bit code segments, but for now, you
shouldn't use them.
However, there is no reason why you can't use 32-bit data
segments in your programs and take advantage of the "flat"
addressing provided 1;>y the 32-bit registers of the 80386.
USE32 code segments only
work In protected mode.
Let's review the key aspects of USE16 and USE32 segments.
USE16 segments can be a maximum of 64K in length, so any
location in a USE16 segment can be pointed to with a 16-bit
address. USE32 segments, on the other hand, can be as long as 4
GB in length, so a 32-bit address is required to point to an
arbitrary location in a USE32 segment.
Clearly, if you need segments longer than 64K, you must use
USE32. By contrast, there's no case in which you must use USE16
segments. This may well lead you to wonder why we don't just
simplify things and use 32-bit segments all the time. The answer
lies in the way in which the 80386 supports word and
doubleword operands and 16- and 32-bit offsets.
The 80386 evolved from the 8086, which uses a single bit to
distinguish between its only two operand sizes, 8- and 16-bits.
The 8086 has a single set of memory-addressing modes-the
446
Turbo Assembler User's Guide
familiar modes involving BX, 51, 01, and BP-supporting 16-bit
offsets only. This code fragment has an 8-bit operand size and
uses an 8086-style 16-bit addressing mode to address memory:
mov
aI, [bx+l000h]
In USE16 code segments, the 80386 normally still uses the same
bit as does the 8086 to select between 8- and 16-bit operands and
still uses 16-bit offsets. However, any given instruction in a USE16
segment may be converted to support 32-bit operands by placing
an operand-size prefix (066h) before the instruction; in this case,
the size bit of the instruction selects between 8- and 32-bit
operands instead of 8- and 16-bit operands.
Similarly, any given instruction in a USE16 segment may be
converted to use the 80386's 32-bit addressing modes (a large
address, as described in the earlier section ''Newadd.ressing
modes" on page 431) by placing an address-size prefix (067h)
before the instruction.
For example, the code assembled from
.386
DataSeg
TestLoc
DataSeg
SEGMENT
DD
ENDS
USE16
CodeSeg SEGMENT USE16
mov ax,DataSeg
mov ds,ax
ASSUME ds:DataSeg
db 66h
mov ax, WORD PTR [TestLoc]
CodeSeg
ENDS
loads the 4 bytes at TestLoc into EAX, rather than the 2 bytes at
TestLoc into AX because the operand-size prefix transforms the
operand size of the instruction to 32 bits.
Along the same lines, instructions in USE32 code segments
normally access 8- or 32-bit operands and nonnally use the 32-bit
addressing modes of the 80386; however, operand-size and
address-size prefixes can be used to cause individual instructions
to operate in 16-bit mode (that is, 8086 mode, with word operands
and/or small addresses), just as if they were in a USE16 segment.
Chapter 70, The 80386 and other processors
447
In short, the operand-size and address-size prefixes can cause an
instruction executing in a USE16 code segment to act as if it were
in a USE32 segment, and can cause an instruction executing in a
USE32 code segment to act as if it were in a USE16 segment.
Don't worry about learning to use operand-size and address-size
prefixes in your 80386 code; the generation of the prefixes
necessary to use 16-bit features in USE32 segments or 32-bit
features in USE16 segments is handled by Turbo Assembler
transparently to the programmer. For example, if you use the
following instruction in a USE32 code segment,
mov
[bx],ax
Turbo Assembler automatically prefixes the instruction with ali
operand-Size prefix and an address-size prefix. We've explained
the workings of the size prefixes here only so you'll understand
the key element in selecting between 16- and 32-bit segment sizes:
the need to minimize the number of size prefixes generated.
Suppose, for example, that you selected a USE16 segment and
then only referred to doubleword-sized operands, addressed with
32-bit addressing modes, such as
mov
eax, [edxtecx*2tl]
Turbo Assembler would have to generate operand-size and/or
address-size prefixes for virtually every instruction in your code
causing the size of your code to balloon and performance to
suffer. Given a USE32 segment, however, the same code would
require no size prefixes at all.
You can now see that the segment-size selection process is a bit
more complex than it seemed. If you need a segment larger than
64K, you must select a USE32 segment. If you need a segment
smaller than 64K, you should select a USE32 segment if you use
more 32- than 16-bit operands and addressing modes. And you
should select a USE16 segment if the reverse is true. It's not
always easy to tell which segment type would be more efficient,
but you can always assemble your code both ways and see which
is more compact.
Now you can also see why the LARGE and SMALL operators are
sometimes necessary to allow forward references to assemble.
Since the USE type of the code segment determines the default
size of address references, forward references are assumed to be
of the same size as the code segment USE type. LARGE must be
used for forward references from USE16 code segments to USE32
448
Turbo Assembler User's Guide
data segments, and you may want to use SMALL in order to force
use of 16-bit addressing for forward references from USE32 code
segments to USE16 data segments.
An example
80386 function
Let's look at some sample 80386 code. Desirable as it would be to
examine a complete 80386 program, that's just not possible right
now, since there's no widely used 80386-based operating system,
and therefore no standard way to request memory, accept
keystrokes, display output, or even terminate a program. Instead,
let's look at a complete function written in 80386 assembler.
Our sample function, named CalcPrimes, takes advantage of the
tremendous length of a USE32 segment to calculate all primes in a
given range in a very straightforward way; the function simply
calculates all multiples of all numbers in the range 2 to the
maximum prime desired, marking every multiple in a single huge
table as being non-prime. On an 8086, this approach would work
well only for arrays shorter than 64K, the maximum segment size,
and would break down entirely at 1 MB, the maximum amount of
memory the 8086 processor can address.
By contrast, USE32 segments and 32-bit registers make it possible
for the 80386 to easily handle a table up to nearly 4 GB in length;
in fact, the 80386 can, with help from paged memory, even handle
memory requirements in the terabyte (1000 GB) range! Of course,
the calculation times for checking such enormous primes would
be unacceptably long, but that's the point; unlike the 8086 and
80286, the 80386's memory-addressing architecture is not a
limiting factor for programs requiring tremendous amounts of
memory.
Here's CalcPrimes:
; Sample 80386 code to calculate all primes between
; 0 and MAX_PRIME (inclusive).
; Input: None
; Output:
ES:EAX - a pointer to PrimeFlags, which contains a 1 at
the offset of each number that is a prime and a 0 at
the offset of each number that is not a prime.
; Registers destroyed:
Chapter· 70, The 80386 and other processors
449
EAX, EBX
Based on an algorithm presented in "Environments,"
by Charles Petzold, PC Magazine, Vol. 7, No.2 .
• 386
MAX_PRIME EQU 1000000
DataSeg
PrimeFlags
DataSeg
;highest I to check for being prime
SEGMENT USE32
DB (MAX_PRIME + 1) DUP (?)
ENDS
CodeSeg SEGMENT USE32
ASSUME
cs:CodeSeg
CalcPrimes
PROC
push ds
mov ax,DataSeg
mov ds,ax
ASSUME
ds:DataSeg
mov es,ax
ASSUME
es:DataSeg
;save caller's DS
Assume all numbers in the specified range are primes.
mov
mov
mov
cld
rep
al,1
edi,OFFSET PrimeFlags
ecx,MAX_PRIME+l
stosb
Now eliminate all numbers that aren't primes by calculating all
mUltiples (other than times 1) less than or equal to MAX_PRIMES
of all numbers up to MAX_PRIME.
mov eax,2
PrimeLoop:
mov ebx,eax
;start with 2, since 0 & 1 are primes,
and can't be used for elimination
of mUltiples
;base value to calculate
all multiples of
MultipleLoop:
add ebx,eax
cmp ebx, MAX_PRIME
;calculate next mUltiple
;have we checked all
; mUltiples of this number?
ja CheckNextBaseValue ;yes, go to next number
mov [PrimeFlags+ebx],0 ;this number is not prime, since
; it's a multiple of something
jmp MultipleLoop
;eliminate the next mUltiple
CheckNextBaseValue:
450
Turbo Assembler User's Guide
inc eax
cmp eax,MAX_PRIME
jb PrimeLoop
ipoint to next base value (the
i next value to calculate all
i mUltiples of)
ihave we eliminated all multiples?
ino, check the next set
i Return a pointer to the table of prime and non-prime statuses
in ES:EAX.
mov eax,OFFSET PrimeFlags
pop ds
irestore caller's DS
ret
CalcPrimes
ENDP
CodeSeg
ENDS
END
Notice how easily the 80386 allows you to handle 32-bit integers
and an array 1,000,000 bytes in length; in fact, the whole function
is, remarkably, only 20 bytes in length. CalcPrimes returns, as its
result, a large far pointer to the table PrimeFlags, in which the
offset corresponding to each number contains a 1 if that number is
prime and a 0 if that number is not prime. For example,
PrimeFlags+3 would be I, since 3 is a prime number, and
PrimeFlags+4 would be 0, since 4 is not.
The length of PrimeFlags, and the largest number to be checked as
to whether it is a prime, are defined by the equated symbol
MAX_PRIME. It would actually be more practical to have the
calling routine pass the address of a table of arbitrary size to
CalcPrimes, along with the largest number to be checked (which
would presumably also be the length of the table minus 1).
CalcPrimes could then meet the prime-calculation needs of any
calling code on the fly, rather than having to be reassembled to
handle new table sizes. The preceding example uses a local
PrimeFlags primarily to illustrate the use of USE32.
A version of CalcPrimes that works with passed table and table
length parameters follows:
i Sample 80386 code to calculate all primes between
i
0 and a specified value (inclusive).
i Input (assumes a large far call, with 6 bytes of return address
; pushed on the stack):
ESP+06h on entry (last parameter pushed) - the
doubleword value of the maximum number to be checked as
Chapter 70, The 80386 and other processors
451
to whether it is a prime.
ESP+OAh on entry (first parameter pushed) - a large far
(6 byte offset) pointer to the table in which to store a
1 at the offset of each number that is a prime and a 0 at
the offset of each number that is not a prime. The table
must be at least [ESP+06h]+1 bytes in length, where
[ESP+06h] is the other parameter.
Output: None
Registers destroyed:
EAX, EBX, EDX, ED!
Based on an algorithm presented in "Environments,"
by Charles Petzold, PC Magazine, Vol. 7, No.2 .
• 386
CodeSeg SEGMENT USE32
ASSUME
cs:CodeSeg
CalcPrimes
PROC FAR
push es
push fs
isave caller's ES
isave caller's FS
Get parameters.
mov ecx, [esp+4+06h]
lfs edx, [esp+4+0ah]
Assume all numbers in the specified range are primes.
push
pop
mov
mov
cld
push
inc
rep
pop
fs
es
al,1
edi,edx
ecx
ecx
stosb
ecx
ipoint ES to table's segment
isave maximum number to check
iset up to maximum number, inclusive
iget back maximum number to check
Now eliminate all numbers that aren't primes by calculating all
mUltiples (other than times 1) less than or equal to the
maximum number to check of all numbers up to the maximum number
to check
mov eax,2
452
;start with 2, since 0 & 1 are primes, and
; can't be used for elimination of mUltiples
Turbo Assembler User's Guide
PrimeLoop:
mov ebx,eax ;base value to calculate all mUltiples of
MultipleLoop:
add ebx,eax ;calculate next mUltiple
cmp ebx,ecx ;have we checked all multiples of number?
ja CheckNextBaseValue
;yes, go to next number
mov BYTE PTR fs: [edx+ebxJ,O ;this number is not prime,
; since it's a mUltiple of
; something
jmp MultipleLoop
;eliminate the next multiple
CheckNextBaseValue:
;point to next base value (the next value
inc eax
; to calculate all mUltiples ofl
;have we eliminated all multiples?
cmp eax,ecx
;no, check the next set of mUltiples
jb PrimeLoop
pop fs
;restore caller's FS
pop es
;restore caller's ES
ret
CalcPrimes
ENDP
CodeSeg
ENDS
END
The 80287
The instruction set of the 80287 math coprocessor is exactly the
same as the instruction set of the 8087, with one exception. The
exception is the FSETPM instruction of the 80287, which places
the 80287 in protected mode. 80287 protected mode corresponds
to the protected mode of the 80286 processor, with which the
80287 is normally coupled (although the 80287 is sometimes used
with the 80386 as well). Of course, any program that uses FSETPM
will not run on an 8087, since the 8087 doesn't support that
instruction.
For detailed information
about 80287 instructions, see
Chapter 3 in the Reference
Guide.
Turbo Assembler support for 80287 assembly is enabled with the
.287 directive.
The 80387
The instruction set of the 80387 math coprocessor is a superset of
the 8087 /80287 instruction set. The new instructions of the 80387
follow:
FCOS
Chapter 10, The 80386 and other processors
FSINCOS
FUCOMP
453
FPREM1
FSIN
FUCOM
FUCOMPP
FUCOM performs an unordered compare between ST(O) and
another 80387 register. This instruction is just like FCOM except
that the result status is set to unordered if one of the operands is a
NAN, rather than generating an invalid-operation exception as
FCOM does in that case. FUCOMP performs an unordered
compare and pops the 80387's stack, and FUCOMPP performs an
unordered compare and pops the stack twice.
FCOS calculates the cosine of the ST(O) register, FSIN calculates
the sine of the ST(O) register, and FSINCOS calculates the sine and
cosine of the ST(O) register.
FPREM1 calculates an IEEE-compatible remainder of ST(O)
divided by ST(1).
1111"
Don't forget that any program that uses any of these instructions
will not run on an 8087 or 80287. Also, because the 80387 handles
real-mode and protected-mode operations in the same way, it
ignores the FSETPM instruction on the 80287.
For detailed Information
about 80387 Instructions, see
Chapter 3 of the Reference
Guide.
Turbo Assembler support for 80387 assembly is enabled with the
.387 directive.
454
Turbo Assembler User's Guide
c
H
A
p
T
E
R
11
Turbo Assembler Ideal Mode
For those of you who are struggling to.make MASM do your
bidding, this may be the most important chapter in the manual. In
addition to near-perfect compatibility with MASM syntax, Turbo
Assembler smooths the bumps and grinds of assembly language
programming with a MASM derivative we call Ideal mode.
Among other things, Ideal mode lets you know solely by looking
at the source text exactly how an expression or instruction
operand will behave. There's no need to memorize a storehouse of
knowledge for all MASM's many quirks and tricks. Instead, with
Ideal mode, you write clear, concise expressions that do exactly
what you want.
Ideal mode uses nearly all MASM's same keywords, operators,
and statement constructions. This means you can explore Ideal
mode's features one at a time without having to learn a large
number of new rules or keywords. All Ideal mode features are
extensions or reorganizations of existing MASM capabilities.
This chapter describes the features of Ideal mode and explains
how using Ideal mode's new syntax rules can save you time and
effort. We'll also discuss in detail all the new capabilities of Ideal
mode and explain the differences between Ideal and MASM
syntaxes.
Chapter 77, Turbo Assembler Ideal Mode
455
What is Ideal mode?
Turbo Assembler's Ideal mode introduces a new syntax for
expressions and instruction operands. The new syntax isn't
radically different from existing MASM syntax; rather, Ideal mode
is a simpler and cleaner implementation of MASM operators and
keywords, using forms that make better sense, both to you and to
Turbo Assembler.
Ideal mode adds strict type-checking to expressions. Strict typechecking helps reduce errors caused by assigning values of the
wrong types to registers and variables, and by using constructions
that appear correct in the source text but are assembled differently
than you expect. Instead of playing guessing games with values
and expressions, as Ideal mode lets you write code that makes
logical and aesthetic sense .
. Because of strict type-checking, Ideal mode expressions are both
easier to understand and less prone to producing unexpected
results. And, as a result, many of the MASM problems we warn
you about in other chapters disappear under Ideal mode's
watchful eye.
Ideal mode also has a number of features that make programming
easier for novices and experts alike. Some of these features
include the following:
• duplicate member names among multiple structures
complex HIGH and LOW expressions
• predictable EQU processing
• correct handling of grouped data segments
• improved consistency among directives
II sensible bracketed expressions
II
Why use Ideal mode?
There are many good reasons why you should use Turbo
Assembler's Ideal mode. If you are just learning assembly
language, you can easily construct Ideal mode expressions and
statements that have the effects you desire. You don't have to
fiddle around tryi~g different things until you get an instruction
that does what you want. If you are an experienced assembly
language programmer, you can use Ideal mode features to write
456
Turbo Assembler User's Guide
complex programs using language extensions such as nestable
structures and unions.
As a direct benefit of a cleaner syntax, Ideal mode assembles files
30% faster than MASM mode. The larger your projects and files,
the more savings in assembly time you'll gain by switching to
Ideal mode.
Strong type-checking rules, enforced by Ideal mode, let Turbo
Assembler catch errors that you would otherwise have to find at
run-time or by debugging your code. This is similar to the way
high-level language compilers assist you by pointing out
questionable constructions and mismatched data sizes.
Although Ideal mode uses a different syntax for some
expressions, you can still write programs that assemble equally
well in both MASM and Ideal modes. You can also switch
between MASM and Ideal modes as often as necessary within the
same source file. This is especially helpful when you're
experimenting with Ideal mode features, or when you're
converting existing programs written in the MASM syntax. You
can switch to Ideal mode for new code that you add to your
source files, while you maintain full MASM compatibility for
other portions of your program.
Entering and leaving Ideal mode
Use the IDEAL and MASM directives to switch between Ideal and
MASM modes. Turbo Assembler always starts assembling a
source file in MASM mode. To switch to Ideal mode, include the
IDEAL directive in your source file before using any Ideal mode
capabilities. From then on, or until the next MASM directive, all
statements behave as described in this chapter. You can switch
back and forth between MASM and Ideal modes in a source file as
many times as you wish and at any place. Here's a sample:
DATA
abc
xyz
DATA
SEGMENT
LABEL BYTE
DW 0
ENDS
istart in MASM mode
iabc addresses xyz as a byte
idefine a word at label xyz
iend of data segment
IDEAL
iswitch to Ideal mode
SEGMENT CODE
PROC MyProc
Chapter 77, Turbo Assembler Ideal Mode
isegment keyword now comes first
iproc keyword comes first, too
457
iIdeal mode programming goes here
ENDP MyProc
ENDS
MASM
CODE SEGMENT
Func2 PROC
irepeating MyProc label is optional
irepeating segment name not required
iswitch back to MASM mode
iname now required before segment keyword
iname now comes before proc keyword, too
iMASM-mode programming goes here
IDEAL
iswitch to Ideal mode again!
ido some programming in Ideal mode
MASM
Func2 ENDP
CODE ENDS
iback to MASM mode. Getting dizzy?
iname again required before keyword
iname again required here
As you can see, in Ideal mode, directive keywords such as PRoe
and SEGMENT appear before the identifying symbol names, the
reverse of MASM's order. Also, you have the option of repeating a
segment or procedure name after the ENDP and ENDS directives.
Adding the name can help clarify the program by identifying the
segment or procedure that is ending. This is a good idea,
especially in programs that nest multiple segments and
procedures. You don't have to include the symbol name after
ENDP and ENDS, however.
MASM and Ideal mode differences
This section describes the main differences between Ideal and
MASM modes. If you know MASM, you may want to experiment
with individual features by converting small sections of your
existing programs to Ideal mode. Just remember to surround the
new code with the IDEAL and MASM keywords. By following this
scheme, a kind of learn-as-you-go approach to Ideal mode
proficiency, you can assemble your current programs without
having to revise every instruction to use Ideal mode's special
features. Eventually, of course, you may decide to program
exclusively in Ideal mode. Or you may choose to mix and match
MASM and Ideal mode modules. The choice is yours to make.
458
Turbo Assembler User's Guide
Ideal mode
tokens
Turbo Assembler reads and understands your program by
dividing the text into individual words or symbols called tokens.
Examples of tokens include labels such as V ALUE, NAME, or
AGE, and other symbols, numbers, parts of expressions, and
arithmetic operators such as +, -, '" and /.
Two types of tokens, symbols and floating-point numbers, have
slightly different forms in Ideal mode. As described next, these
changes clarify several ambiguities in the MASM syntax.
Symbol tokens
In Ideal mode, a period (.) is not permitted as part of a symbol
name. You can use a period only as a structure member operator
or in a floating-point number.
Structure and union members (some people call them fields) are
not defined as global symbols, accessible from every place in yourprogram. Structure and union members exist only within the
structure to which they belong. This lets you have multiple
structures that contain members with the same names. You can
also duplicate member names outside of a structure for other
purposes, as in this sample:
Pennies DW 0
STRUC Heaven
Dimes
DW ?
Nickels DW?
Pennies DW?
ENDS
Take Heaven <>
ino conflict
They say you can't take it with you but, just in case they're wrong,
this example shows how to create a variable with three fields,
storing your net worth in dimes, nickels, and pennies in a
structure named Heaven. The fields Dimes and Nickels are unique
to the structure. Pennies, though, occurs twice. First, there's
Pennies outside the structure's pearly gates, and then there's
Pennies from Heaven.
Seriously, this example demonstrates that the same name, Pennies,
can occur both inside and outside of a structure with no conflict,
something that you can't do in MASM to save your soul.
The variable Pennies outside of Heaven is distinct from the
member Pennies used inside the structure. Consequently, to
Chapter 77, Turbo Assembler Ideal Mode
459
reference a duplicated name inside of a structure requires three
elements: the structure name, a period, and the member name. In
this example, Take.Pennies equals the offset of the Pennies field
inside Heaven. Pennies alone, however, equals the offset to the
variable outside of the structure.
Duplicate member
names
Ideal mode also lets you duplicate member names in different
structures. The members can be of the same or of different types,
as in the following two structures, both of which have Size fields
of the same type and in the same postion, plus Amount fields of
different types in different positions:
STRUC SomeStuff
Size
DW ?
Flag
DB ?
Amount
DW ?
ENDS
STRUC OtherStuff
DW
Size
Amount
DB ?
ENDS
Floating-point tokens
ino conflict here
inor here
In Ideal mode, floating-point decimal numbers must always
include a period (.):
FP
DT
l.Oe? ildeal mode floating-point value
This defines a lO-byte floating-point value, named FP, equal to
1.0e7. In MASM mode, you can use the acceptable, though less
clear, form:
FP
DT
lE?
iMASM mode floating-point value
This may not seem so bad until you consider what happens if, in
an earlier section of the program, you issue a .RADIX 16 command
that changes the default number base from decimal to hexadecimal. In this case, disaster strikes as MASM now assembles
your floating-point value as the hexadecimal number OlE7! By
requiring you to use a decimal point, Ideal mode never
accidentally confuses floating-point and hexadecimal numbers
this way.
460
Turbo Assembler User's Guide
EQU and =
directives
EQU definitions, also called equates, are always treated as text in
Ideal mode. In MASM mode, equates are sometimes treated as
text and, at other times, as numbers. Consider these-examples:,
;Declare a few equates
A
4
B
5
C
EQU B + A
B
6
;Declare a variable
V DW C
;9 in MASM mode, 10 in Ideal mode
MASM evaluates B + A when processing the EQU expression. At
this time, A equals 4 and B equals 5; therefore, C equals 9. Ideal
mode processes the same expression differently, storing in string
form everything that follows EQU, in this case, B + A. Later, Ideal
mode substitutes this string where C appears. In this example,
because the expression evaluation is delayed until the declaration
of variable V and because B was previously redefined to 6,
variable V equals 10 (6+4) in Ideal mode.
1111"
Expressions and
operands
Square brackets
operator
In Ideal mode, EQU always defines a string. An equal sign (=)
always defines a calculated expression. It might help you to
remember this rule if you visualize an equal sign (=) evaluating
expressions immediately and EQU delaying expression evaluation
until the place where the constant name appears. By the way,
some people refer to this as "early" and "late" binding.
The biggest difference between Ideal and MASM mode
expressions is the way square brackets function. In Ideal mode,
square brackets always refer to the contents of the enclosed
quantity. Brackets never cause implied additions to occur. Many
standard MASM constructions, therefore, are not permitted by
Ideal mode.
In Ideal mode, square brackets must be used in order to get the
contents of an item. For example,
mov ax,wordptr
Chapter 77, Turbo Assembler Ideal Mode
461
displays a warning message. You are are trying to load a pointer
(wordptr) into a register (AX). The correct form is
mov ax, [wordptr]
Plainly, you are loading the contents of the location addressed by
wordptr (in the current data segment at DS) into AX.
If you wish to refer to the offset of a symbol within a segment,
you must explicitly use the OFFSET operator, as in this example:
mov ax, OFFSET wordptr
Example operands
Let's examine a few confusing, though typical, bracketed
operands that MASM mode accepts, and then compare the
examples with the correct and easier-to-understand forms that
Ideal mode requires. As you'll see, Ideal mode's unambiguous use
of brackets helps make your intentions perfectly clear:
mov
ax, [bx] lsi]
iMASM mode
This causes a syntax error in Ideal mode. If brackets specify the
contents of memory, then this instruction appears to be loading
both the value addressed by BX and the value addressed by 51
into AX at the same time. Of course, you can do no such thing.
What you probably mean, and what Ideal mode requires, is this:
mov
ax, [bx+si]
ildeal mode
Now, the instruction is clear. The contents of the memory location
at the OFFSET BX+SI, relative to the current data segment
addressed by OS, is loaded into AX. (The size of the memory
location is a 16-bit word because AX is a 16-bit register. If you
replace AX with AL, or another 8-bit register, then the size of the
memory location is a byte.) Here's a similar example:
mov
ax,es: [bx] lsi]
iMASM mode
This also causes an Ideal mode error. The instruction seems to be
saying, "apply an ES: segment override to the value addressed by
BX, and add the whole shebang to the contents of the memory
location addressed by 51, loading the result (whatever that is) into
AX." This is senseless, of course, and you probably mean this:
mov
ax, [es:bx+si]
ildeal mode
Good! This adds the BX and SIregisters together, giving an offset
value relative to segment register ES, overridden from the default
462
Turbo Assembler User's Guide
data segment DS. The 16-bit contents of this location is loaded
into AX. Here's another MASM example that you'll often see:
mov
ax,6[bx]
iMASM mode
A mathematician might think you are multiplying 6 times the
value of the location addressed by BX. Or, is this some kind of
undocumented array indexing technique, or just a typing error?
Actually, it's none of the above, as the Ideal mode form shows
mov
ax, [bx+6]
ildeal mode
Of course! You want to load into AX the contents of the location
in the current data segment 6 bytes away from the offset specified
by BX. More clear than that, you cannot get. Expressions in
MASM mode, though, are not always so understandable:
mov
ax,es:[bp+8] [si+6]
iMASM mode
Let's see, you take the value 8 bytes away from BP, apply a
segment override ES:, and ... no, the override must go with the
value 6 bytes from SI. But no, that's not right, maybe you take the
value at BP+8, add to the contents of [SI+6], apply an override
and ... Oh, forget it! Ideal mode makes this and other complex
operands easy to read and easy to write:
mov
ax, [es:bp+si+14]
ildeal mode
Obviously, the value located at offset BP+SI+14 in segment ES is
loaded into AX, plain and simple. Believe it or not, there's more:
mov
aI, BYTE PTR [bx]
iMASM mode
MASM apparently allows you to specify the contents of memory
locations as byte pointers, at least that's what this instruction
appears to be doing. You can, of course, point to bytes or words
only with pointers (registers and labels) as Ideal mode makes
perfectly evident:
mov
aI, [BYTE PTR bx]
ildeal mode
Obviously, you are telling Turbo Assembler that BX is a byte
pointer, loading into register AL the byte located BX bytes from
the start of the current data segment. One more example and then
we're done:
rep
movs BYTE PTR [di], lsi]
iMASM mode
MASM appears to allow you to convert characters addressed by
DI (and maybe SI?) into byte pointers. Of course, you can't do
Chapter 77 Turbo Assembler Ideal Mode
I
463
that. What you no doubt mean, and what Ideal mode wants to
see, is this:
rep
movs [BYTE PTR di], [BYTE PTR si]
ildeal mode
Although this is longer, registers DI and 51 are clearly byte
pointers for the MOVS instruction.
These examples, are by no means complete, and you probably will
encounter many other confusing MASM operands with brackets.
When this happens, try switching to Ideal mode, even if just for
that one instruction. Then, use the foregoing samples as guides to
rewriting the instruction in a form that you can understand. By
doing this, you can use Ideal mode not only to help you write
better and more readable programs, but also to help you
understand bracketed constructions that, in MA5M, are
frequently about as clear as mud on a foggy day.
Operators
Chapter 2 In the reference
manuallisfs operator
precedence and completely
describes all operators in
MASM and Ideal modes.
The changes made to the expression operators in Ideal mode
increase the power and flexibility of some operators while leaving
unchanged the overall behavior of expressions. The precedence
levels of some operators have been changed to facilitate common
operator combinations.
Periods in structure
members
For specifing accurately the structure members to which you ;re
referring, the period (.) structure member operator is far more
strict in Ideal mode. The expression to the left of a period must be
a structure pointer. The expression to the right must be a member
name in that structure. Using the earlier SomeStuff and OtherS tuff
structure examples, here's how to load registers with the values of
specific structure members:
iDeclare variables using the structure types
S Stuff SomeStuff <>
Stuff OtherStuff <>
mov ax, [S_Stuff.Amount]
iload word value
mov bl, [O_Stuff.Amount]
iload byte value
°
Pointers to structures
464
Often, you'll want to use a register containing the address of a
structure, in other words, the offset to the first byte of a structure
stored in memory. Or you might have a memory variable that
addresses a structure. In these cases, to reference a specific
Turbo Assembler User's Guide
structure member by name, you must tell Turbo Assembler which
structure you are referring to:
mov ex, [(SorneStuff PTR bx) .Amount]
This lets Turbo Assembler know that BX is a pointer to a SomeS tuff
structure and that you want to load the contents of the Amount
field from that structure into register CX. The parentheses are
required because the period (.) operator has higher precedence
than PTR. Without parentheses, Ideal mode tries to bind Amount
to BX, which is impossible, of course, because registers do have
field names. Only structures have field names and, therefore, you
must convert pointers to structures before referring to fields in
structures that the registers address.
The SYMTYPE operator
Because an Ideal mode symbol cannot start with a period, the
.TYPE operator in MASM mode is named SYMTYPE in Ideal
mode (see Chapter 1 in the Reference Guide). Despite the name
change, the directive works identically in both modes with one
exception: SYMTYPE will not return a value for an undefined
identifier. Otherwise, this operator returns the types of various
symbols.
Abyte
Aword
Array
Btype
Wtype
Atype
The HIGH and LOW
operators
DB 0
DW 0
DD 20 DUP (8)
SYMTYPE Abyte
SYMTYPE Aword
SYMTYPE Array
;1
;2
;4
In Ideal mode, the HIGH and LOW operators have two meanings.
Usually, HIGH specifies the high (most-significant) byte of a
constant and LOW specifies the LOW (least-significant) byte as in
MaxVal
1234h
mov ah, HIGH MaxVal
rnov aI, LOW MaxVal
;loads 12h into AH
;loads 34h into AL
In Ideal mode, HIGH and LOW can be used also to select the high
or low part of a memory-referencing expression:
WordVal
DblVal
QVal
mov
mov
mov
Chapter 77, Turbo Assembler Ideal Mode
DW 0
DD 0
DQ 0
bl, [BYTE LOW WordVal]
ax, [WORD HIGH DblVal]
ax, [WORD LOW QVal]
465
The first MOV instruction loads BL with the low byte of the 2-byte
word labeled by W1rdVal. The second MOV loads AX with the
high word of the 4-byte value stored at DblVal. The third MOV
loads AX with the lowest word of the 8-byte (quadword) value at
QVal. Notice that the syntax is the same as for the PTR operator,
with BYTE or WORD keywords before the LOW or HIGH
operators, followed by a memory-referencing expression.
You can also use HIGH and LOW together to extract just the
information you need from a multiple-byte value:
OVal 00 12345678h
rnov aI, [BYTE LOW WORD HIGH OVal]
;loads 34h into AL
In combination with BYTE and WORD, the LOW and HIGH
keywords extract bytes and words from any position in a variable.
Here, DVal is a doubleword, 4-byte quantity. To better
understand complex combinations such as this, read the
expression from left to right. In this case, the move instruction
loads AL with lithe low byte (BYTE LOW) of the high word
(WORD HIGH) of Dval."
The Optional PTR
operator
You can use shorthand pointer overrides in expressions. To do
this, omit the PTR operator. For example,
[BYTE PTR OverTheRainbow]
in Ideal mode shorthand is the same as
[BYTE OverTheRainbow]
The SIZE operator
The SIZE operator in Ideal mode reports the actual number of
bytes occupied by a data item. This makes it easy to determine the
'
lengths of strings:
theTitle
theAuthor
titleSize
authorSize
OB
OB
"The Sun Also Rises"
"Ernest Hemingway", 0
SIZE theTitle
SIZE theAuthor
; Ideal--18, MASM--l
; Ideal--16, MASM--l
In this example, theTitle and theAuthor are strings. In MASM
mode, the SIZE operator equals the LENGTH of a name multiplied
by its TYPE. The LENGTH equals the number of items allocated, in
this case 1. (Even though a string has multiple characters,
LENGTH considers strings to be single-byte items by virtue of the
DB directive.) The TYPE value for DB is also 1. Consequently, in
466
Turbo Assembler User's Guide
MASM mode, both titleSize and authorSize equal 1, which is not
much help in trying to calculate the string lengths.
In Ideal mode, SIZE returns the number of bytes occupied by the
first item after storage-allocation directives like DB or DW.
Because of this, titleSize equals the number of characters in
theTitle. Likewise, author Size equals the number of characters in
the string, theAuthor. Notice, however, that theAuthor ends in a 0
byte, marking the string end. SIZE does not take this byte into
account, returning only the number of characters in the preceding
string. In fact, SIZE returns the length of only the first item in any
list of multiple values. For example,
CountDown
TwoLines
eDsize
TLsize
DB
DB
9,8,7,6,5,4,3,2,1,"Blast off"
"First line", 13, 10, "Second line"
SIZE CountDown
SIZE TwoLines
;1
;10
Here, CountDown addresses 9-byte values followed by the string,
"Blast off." Even so, SIZE of CountDown (CDSize) in both Ideal
and MASM modes equals 1, the size of the first element in the list.
The same is not true of the second example, TwoLines, which is a
typical way to store two strings separated with an ASCII carriage
return (13) and linefeed (10). But the two strings are labeled in the
program under one name, TwoLines. SIZE again returns the size of
the first item in this series, in this case, the string "First line." In
Ideal mode, TLSize equals 10, the number of characters in the
string. In MASM mode, TLSize equals 1, the size of the first DB
element, a single byte (character).
Directives
Directives in Ideal mode function identically and, in most cases,
have the same names as their MASM-mode equivalents.
However, there are a few important differences among similar
directives in both modes, as this section explains.
. Listing controls
Because a symbol cannot start with a period (.) in Ideal mode, all
MASM mode listing controls begin with percent signs (%). Also,
several names have been changed to more accurately describe the
operations controlled by the directives. The following table shows
the listing control directives in both modes:
Chapter 77 Turbo Assembler Ideal Mode
I
467
MASMmode
Ideal mode
.CREF
.LALL
.LFCOND
.LIST
.SFCOND
.xALL
.xCREF
.XLIST
%CREF
%MACS
%CONDS
%LIST
%NOCONDS
%NOMACS
%NOCREF
%NOLISf
Because the percent sign (%) starts all listing control directives in
Ideal mode, the o/oOUT directive in MASM mode becomes
DISPLAY in Ideal mode:
DISPLAY "Starting to Assemble I/O Driver"
Directives starting with
a period (.)
Other MASM directives that start with periods (.) are renamed for
clarity. For instance, all processor control directives such as .286,
which look more like a number than a directive, now start with P,
as in P286N. All forced error directives of the form .ERRxxx have
been renamed ERRIFxxx. Several other directives have the same
names minus the leading periods.
The following table lists the directives that start with a period in
MASM mode and the Ideal mode equivalents:
468
MASM mode
Ideal mode
MASM mode
Ideal mode
.186
.286
.286C
.286P
.287
.386
.386C
.386P
.387
.8086
.8087
.CODE
.CONST
. DATA
•DATA?
.ERR
.ERR1
P186
P286N
P286N
P286
P287
P386N
P386N
P386
P387
P8086
P8087
CODESEG
CONST
DATASEG
UDATASEG
ERR
ERRIF1
.ERR2
.ERRB
.ERRDEF
.ERRDIF
.ERRDIFI
.ERRE
.ERRIDN
.ERRIDNI
.ERRNB
.ERRNDEF
.ERRNZ
.FARDATA
.FARDATA?
.MODEL
.RADIX
.STACK
ERRIF2
ERRIFB
ERRIFDEF
ERRIFDIF
ERRIFDIFI
ERRIFE
ERRIFlDN
ERRIFIDNI
ERRIFNB
ERRIFNDEF
ERRIF
FAR DATA
UFARDATA
MODEL
RADIX
STACK
Turbo Assembler User's Guide
Reversed directive and
symbol name
Ideal mode's parsing order is simpler than MASM's. If the first
token is a keyword, it determines the operation to be performed
by the directive. If the first token is not a keyword, then the
second token determines the operation.
Because of this change, some operations have reversed directive
and symbol name orders, as the next table details:
MASMmode
Ideal mode
name ENDP
name ENDS
name GROUP segs
name LABEL type
name MACRO args
name PROC type
name RECORD args
name SEGMENT args
ENDP [name]
ENDS [name]
GROUP name segs
LABEL name type
MACRO name args
PROC name type
RECORD name args
SEGMENT name args
STRUCname
UNION name
nameSTRUC
name UNION
Notice that ENDS and ENDP do not require matching names to
close the definitions. If you include a name, spell it the same as
you did in the preceding SEGMENT or PRoe directive. Some
programmers always include the name to add extra readability to
their programs. This is especially useful when you're using nested
procedures or segments, but it isn't required.
Some directives are identical in both MASM and Ideal modes. For
example, the following directives define symbols as part of the
language syntax and, therefore, are the same in both modes:
=
DD
DB
DP
OF
DQ
DT
DW
EQU
Quoted strings as
arguments to directives
The INCLUDE directive takes a quoted file name in Ideal mode:
INCLUDE "MYDEFS.INC"
In MASM mode you don't have to use quotes:
INCLUDE MYDEFS.INC
0/0TITLE and O/OSUBTTL also require their title strings to be
surrounded by quotes:
Chapter 77 Turbo Assembler Ideal Mode
I
469
%TITLE
%SUBTTL
"Macro Definitions"
"Block Structuring Macros"
;comment ignored
;comment ignored
As these two examples demonstrate, requiring quotes around
titles and subtitles lets you add comments at the ends of these
lines. The comments are not included in the listing file. In MASM
mode, everything after .TITLE and .SUBTTL becomes part of the
title string, including any comments.
Segments and
groups
The way Turbo Assembler handles segments and groups in Ideal
mode can make a difference in getting a program up and running.
If you're like most people, you probably shudder at the thought
of dealing with a bug that has anything to do with the interaction
of segments and groups.
Much of the difficulty in this process stems from the arbitrary
way that MASM and, therefore, Turbo Assembler's MASM mode,
make assumptions about references to data or code within a
group. Fortunately, Ideal mode alleviates some of the more
nagging problems caused by MASM segment and group
directives as you'll see in the information that follows.
Accessing data in a
segment belonging to
a group
In Ideal mode, any data item in a segment that is part of a group
is considered to be principally a member of the group, not of the
segment. An explicit segment override must be used for Turbo
Assembler to recognize the data item as a member of the segment.
MASM mode handles this differently: Sometimes a symbol is
. considered to be part of the segment instead of the group. In
particular, MASM mode treats a symbol as part of a segment
when the symbol is used with the OFFSET operator but as part of
a group when the symbol is used as a pointer in a data allocation.
This can be confusing because, when you directly access the data
without OFFSET, MASM incorrectly generates the reference
relative to the segment instead of the group.
An example will help explain how you can easily get into trouble
with MASM's addressing quirks. Consider the following
incomplete MASM program, which declares three data segments:
470
dsegl
vl
dsegl
SEGMENT PARA PUBLIC 'data'
DB
0
ENDS
dseg2
SEGMENT PARA PUBLIC 'data'
Turbo Assembler User's Guide
v2
dseg2
DB
ENDS
dseg3
v3
dseg3
SEGMENT PARA PUBLIC 'data'
DB
0
ENDS
0
DGROUP GROUP dsegl,dseg2,dseg3
cseg
SEGMENT PARA PUBLIC 'code'
ASSUME cs:cseg,ds:DGROUP
start:
cseg
rnov
rnov
rnov
ENDS
END
ax, OFFSET vl
bx,OFFSET v2
cx,OFFSET v3
start
The three segments, dseg1, dseg2, and dseg3, are grouped under
one name, DGROUP. As a result, all the variables in the individual
segments are stored together in memory. In the program source
text, each of the individual segments declares a byte variable,
labeled v1, v2, and v3.
In the code portion of this MASM program, the offset addresses of
the three variables are loaded into registers AX, BX, and ex.
Because of the earlier ASSUME directive and because the data
segments were grouped together, you might think that MASM
would calculate the offets to the variables relative to the entire
group in which the variables are eventually stored in memory.
But this is not what happens! Despite your intentions, MASM
calculates the offsets of the variables relative to the individual
segments, dseg1, dseg2, and dseg3. It does this even though the
three segments are combined into one data segment in memory,
addressed here by register DS. It makes no sense to take the
offsets of variables relative to individual segments in the program
text when those segments are combined into a single segment in
memory. The only way to address such variables is to refer to
their offsets relative to the entire group.
To fix the problem in MASM requires you to specify the group
name along with the OFFSET keyword:
rnov
rnov
rnov
ax,OFFSET DGROUP:vl
bx,OFFSET DGROUP:v2
cx,OFFSET DGROUP:v3
Chapter 77, Turbo Assembler Ideal Mode
471
Although this now assembles correctly and loads the offsets of vl,
v2, and v3 relative to DGROUP (which collects the individual
segments), you might easily forget to specify the DGROUP
qualifier. If you make this mistake, the offset values will not
correctly locate the variables in memory and you'll receive no
indication from MASM that anything is amiss. In Ideal mode,
there's no need to go to all this trouble:
.
IDEAL
SEGMENT dsegl
vl
DB
ENDS
PARA PUBLIC 'data'
a
SEGMENT dseg2
v2
DB
ENDS
a
SEGMENT dseg3
v3
DB
ENDS
a
PARA PUBLIC 'data'
PARA PUBLIC 'data'
GROUP DGROUP dsegl,dseg2,dseg3
SEGMENT cseg
PARA PUBLIC 'code'
ASSUME cs:cseg, ds:DGROUP
start:
mov
mov
mov
ax,OFFSET vl
ax,OFFSET v2
ax,OFFSET v3
END
start
ENDS
The offsets to vl, v2, and v3 are correctly calculated relative to the
group that collects the individual segments to which the variables
belong. Ideal mode does not require the DGROUP qualifier to
refer to variables in grouped segments. MASM mode does require
the qualifier and, even worse, gives no warning of a serious
problem should you forget to specify the group name in every
single reference.
Defining near or
far code labels
472
When you define near and far LABEL or PROC symbols,
references to a symbol are relative to the group containing the
segment. If a symbol's segment is not part of a group, the symbol
is relative to the segment. This means you do not have to
ASSUME CS to a segment in order to define near or far symbols.
In MASM mode,
Turbo Assembler User's Guide
CODE SEGMENT
ASSUME
cs:CODE
XYZ PROC FAR
iMASM procedure code
XYZ ENDP
CODE ENDS
becomes the following in Ideal mode:
SEGMENT CODE
PROC XYZ FAR
ildeal mode procedure code
ENDP
ENDS
This change doesn't add any new capabilities to MASM mode.
But it does relieve you of telling the assembler something Ideal
mode can usually figure out by itself.
External, public,
and global
symbols
Wherever you must supply a type (BYTE, WORD, and so on), for
example, with the EXTRN or GLOBAL directives, you can use a
structure name:
STRUC MoreStuff
HisStuff DB 0
HerStuff DW 0
ItsStuff DB 0
ENDS
EXTRN SNAME:MoreStuff
This capability, combined with the enhancements to the period (.)
operator described earlier, lets you refer to structure members
that are external to your source module. This is exactly as if you
had declared the members inside both modules. The SIZE
operator also correctly reports the size of external data structures.
Every PUBLIC symbol emitted in Ideal mode occurs where
PUBLIC is specified. This is also useful for redefining variables.
MASM mode emits all the public symbols at the end of the
program, limiting the ways in which you can redefine public
symbols. For example,
Chapter 77, Turbo Assembler Ideal Mode
473
Perfect = 8
PUBLIC Perfect
Perfect = 10
;declare Perfect public
;redefine Perfect's value
In Ideal mode, the PUBLIC Perfect equals 8, even though the
module redefines Perfect after the PUBLIC declaration. In MASM
mode, because the PUBUC symbols are emitted at the end of the
module, another module that imports this symbol via an EXTRN
declaration receives a Perfect 10.
Miscellaneous
differences
Suppressed fixups
This section describes a few additional differences between
MASM and Ideal modes.
Turbo Assembler in Ideal mode does not generate segmentrelative fixups for private segments that are page- or paragraphaligned. Because the linker does not require such fixups,
assembling programs in Ideal mode can result in smaller object
files that also link more quickly than object files generated by
MASM mode. The following demonstrates how superfluous
fixups occur in MASM but not in Ideal mod~:
SEGMENT DATA PRIVATE PARA
VAR1 DB 0
VAR2 DW 0
ENDS
SEGMENT CODE
ASSUME ds:DATA
mov ax,VAR2
ENDS
;no fixup needed
This difference has no effect on code that you write. The
documentation here is simply for your information.
Operand for BOUND
instruction
The BOUND instruction expects a WORD operand, not a DWORD.
This lets you define the lower and upper bounds as two constant
words, eliminating the need to convert the operand to a DWORD
with an explicit DWORD PTR. In MASM mode, you must write
BOUNDS
BOUND
DW 1,4
DWORD PTR BOUNDS
;lower and upper bounds
;required for MASM mode
but, in Ideal mode, you need only write
BOUNDS
DW 1,4
BOUND
[BOUNDS]
474
;lower and upper bounds
;legal in Ideal mode
Turbo Assembler User's Guide
Comments inside
macros
In Ideal mode, comments within macros are treated as strings. To
substitute a dummy parameter within a macro comment, you
must precede the parameter with an ampersand (&):
MACRO DOUBLE ARG
shl ARG,l
ENDM
;rnultiply &ARG by two
When you use this macro in Ideal mode with DOUBLE BX, the
listing file shows
shl bx,l
;rnultiply BX by two
On the other hand, if the macro is defined as
MACRO DOUBLE ARG
shl ARG,l
ENDM
;rnultiply ARG by two
the listing file does not replace ARG:
shl bx,l
Local symbols
;rnultiply ARG by two
Turbo Assembler's local symbol capability is automatically
enabled when you switch to Ideal mode, exactly as if you had
entered the LOCALS directive.
A comparison of MASM and Ideal mode
programming
To wrap up this chapter and give you a feeling for the differences
between Ideal and MASM modes, here is the same program in
both Ideal and MASM mode. By reading through these examples
and by examining the numbered comments after the listings,
you'll be able to appreciate the advantages offered by Ideal mode
syntax.
Please understand that these programs are not intended as
examples of good programming style: The instructions merely
demonstrate the Ideal mode concepts discussed in this chapter,
and show only a sampling of the most common Ideal mode
capabilities and differences from MASM.
Chapter 77 Turbo Assembler Ideal Mode
I
475
The example programs read a single line from the console,
convert the text to uppercase, and then display the result before
returning to 005. To mark where the program code differs in the
MASM and Ideal mode programs, we've added a comment
(beginning with a semicolon) and a number. For example, ; #4
directs you to read the corresponding description number 4
following the listings in the section An Analysis of MASM And
Ideal Modes" on page 479. Also, to make the Ideal mode
differences stand out, we've stripped most of the comments from
its example. Read the first program to understand how the code
operates. Read the second program to compare the Ideal-mode
enhancements.
/I
MASM mode sample program
; File
; MASM mode example program to uppercase a line
TITLE Example MASM Program
;this comment is in the title!
.286
bufsize
=
128
;size of input and output buffers
dosint MACRO intnum
mov ah,intnum
int 21h
ENDM
;assign FN number to AH
;call DOS function &INTNUM
STK SEGMENT STACK
DB 100h DUP (?)
STK ENDS
;reserve stack space
DATA SEGMENT WORD
inbuf
DB bufsize DUP
outbuf
DB bufsize DUP
DATA ENDS
(?)
(?)
DGROUP GROUP STK,DATA
CODE SEGMENT WORD
ASSUME cs:CODE
start: •
mov ax,DGROUP
mov ds,ax
ASSUME ds:DGROUP
mov dx,OFFSET DGROUP:inbuf
xor bx,bx
call readline
mov bx,ax
mov inbuf[bx],O
push ax
476
;input buffer
;output buffer
;group stack and data segs
;assume CS is code seg
;assign address of DGROUP
;segment to DS
;default data segment is DS
;load into DX inbuf offset
;standard input
iread one line
;assign length to BX
;add null terminator
;save AX on stack
Turbo Assembler User's Guide
call mungline
pop cx
mov dx,OFFSET DGROUP:outbuf
mov bx,1
40h
dosint
dosint
4ch
;convert line to uppercase
;restore count
;load into DX outbuf offset
;standard output
;write file function
;exit to OOS
;Read a line, called with dx => buffer, returns count in AX
readline PROC NEAR
mov cx,bufsize
;specify buffer size
dosint
3fh
;read file function
;set zero flag on count
and ax,ax
ret
;return to caller
readline ENDP
;Convert line to uppercase
mungline PROC NEAR
mov si,OFFSET DGROUP:inbuf
mov di,O
@@uloop:
cmp BYTE PTR[si],O
je @@done
mov aI, [si]
and aI, not 'a' - 'A'
mov outbuf[di],al
inc si
inc di
jmp @@uloop
@@done: ret
mung line ENDP
CODE ENDS
END start
;address inbuf with SI
; initialize DI
; end of text?
;if yes, jump to @@done
;else get next character
;convert to uppercase
;store in output buffer
;better to use lodsb,stosb
; .•. this is just an example!
;continue converting text
;end of procedure
;end of code segment
;end of text and OOS entry point
Ideal mode sample program
File
Ideal mode example program to uppercase a line
IDEAL
%TITLE
"Example Ideal-Mode Program"
P286N
BufSize
;n
;12
;#3
128
MACRO dosint intnum
mov ah,intnum
int 21h
ENDM
;14
SEGMENT STK STACK
DB 100h DUP (?)
; 15
Chapter 11, Turbo Assembler Ideal Mode
477
ENDS
i.6
SEGMENT DATA WORD
inbuf
DB Bufsize DUP (1)
outbuf
DB bufSize DUP (1)
ENDS DATA
i
GROUP DGROUP STK,DATA
i.9
SEGMENT CODE WORD
ASSUME cs:CODE
start:
mov ax,DGROUP
mov ds,ax
ASSUME
ds:DGROUP
mov dx,OFFSET inbuf
xor bx,bx
call readline
mov bx,ax
mov [inbuf + bx],O
push ax
call mungline
pop cx
mov dx,OFFSET outbuf
mov bx,1
dosint
40h
dosint
4ch
ino
i'S
in1
in2
in3
iRead a line, called with dx => buffer, returns count in AX
PROC readline NEAR
mov cx,BufSize
3fh
dosint
and ax,ax
ret
ENDP
iConvert line to uppercase
PROC mung line NEAR
mov si,OFFSET inbuf
mov di,O
@@uloop:
cmp [BYTE si],O
je @@done
mov aI, lsi]
and al,not 'a' - 'A'
mov [outbuf + di],al
inc si
inc di
LODSB/STOSB
jmp @@uloop
@@done: ret
478
.7
i.14
ins
in6
in7
iUS
;U9
Turbo Assembler User's Guide
ENDP mung line
ENDS
END start
An analysis of
MASM And Ideal
modes
; 120
; 121
The following paragraphs detail the differences between MASM
and Ideal mode constructions, directives, and operands in the two
previous programs. The numbers refer to the comments in the
Ideal mode example. Compare these lines with the MASM
example.
1. Use the IDEAL directive to switch into Ideal mode. By default,
2.
3.
4.
5.
6.
7.
Turbo Assembler always starts assembling your source file in
MASM mode. You need to use the MASM directive only when
you want to switch back into MASM mode after having earlier
switched to Ideal mode.
The percent sign in front of %TITLE reminds you that this
directive affects the listing file (if you decide to create one by
specifying a listing file name or by using the IL command-line
option when you assemble the program). Ideal mode uses
%TITLE instead of TITLE (without the percent sign) all:d also
requires you to surround the title string with quotes (II II). This
lets you put a comment on the line that, in MASM mode,
becomes part of the title-probably not what you intended.
The .286 directive in MASM mode is P286N in Ideal mode.
Because symbols cannot start with a period (.) in Ideal mode,
all MASM processor and other directives that start with
periods are changed. The statement in the listing does not
serve any useful purpose in this program other than to show
the difference between the two modes. The program does not
use any 80286 instructions.
In Ideal mode, the name of the macro comes after the MACRO
directive, not before as in MASM mode.
The name of the segment in a SEGMENT directive comes after
the directive in Ideal mode.
When you use ENDS to close a segment in Ideal mode, you
don't need to supply the matching segment name as you do in
MASM mode. (You may add the name after the ENDS
directive, however, if you prefer.)
Same as 5. Again, the SEGMENT keyword comes before the
name.
Chapter 77 Turbo Assembler Ideal Mode
I
479
8. If you supply a matching segment name for the ENDS
directive, the name comes after the directive and not before as
in MASM mode. You can delete the name (DATA) if you wish.
9. In Ideal mode, the GROUP directive precedes the name of the
data segment group (which is DGROUP). After this comes the
list of data segments you are grouping under this name. In
MASM, GROUP and the name are reversed.
10. Same as 5. The SEGMENT keyword precedes the name.
11. You don't have to use a group qualifier here with the OFFSET
operator. Ideal mode presumes that INBUF is relative to the
start of DGROUP because INBUF is inside one of the individual
segments collected under this group name. In MASM, you
have to remember to specify DGROUP: inbuf to correctly
locate offsets to variables in grouped segments.
12. The [lNBUF+BX] operand is valid in both Ideal and MASM
modes, but the same line in the MASM mode version,
INBUF[BX], is not valid in Ideal mode. In Ideal mode, all
memory-referencing operands must be surrounded by square
brackets.
13. Same as 11. Here again, you do not need to specify the group
name to reference a variable in a grouped segment. In MASM,
to obtain the correct offset to OUTBUF, you have to write
DGROUP:outbuf. Forget the DGROUP qualifier and, in this
example, you'd store your output in the stack, with no
warning from MASM that something is seriously wrong!
14. The name of a procedure in a PROC directive comes after the
directive, not before as required by MASM mode.
15. When you use ENDP to close a procedure in Ideal mode, you
don't have to supply the matching procedure name as you do
in MASM mode.
16. Same as 14. The PROC directive proceeds the procedure name.
17. Same as 11. Again, you don't need to write DGROUP:inbuf, as
you do in MASM.
18. In Ideal mode, you can optionally omit the PTR operator when
you set the size of an expression. The MASM mode expression
BYTE PTR ABC is identical to BYTE ABC in Ideal mode.
19. Same as 12. In Ideal mode, to refer to the contents of memory,
always put the memory-referencing expression inside
brackets.
20. Optionally place a matching procedure name after the ENDP
directive, not before as in MASM mode.
480
Turbo Assembler User's Guide
21. Same as 6. ENDS does not require a matching segment name,
although you can add the name if you prefer.
481
482
Turbo Assembler User's Guide
References
Crawford, John H., and Patrick P.Gelsinger. Programming the
80386. Alameda: Sybex, Inc., 1987.
Duncan, Ray. Advanced MS-DOS. Redmond: Microsoft Press,
1986.
Lafore, Robert. Assembly Language Primer for the IBM PC & XI.
New York: The Waite Group, 1984.
Murray, William H., and Chris Pappas. 80386/80286 Assembly
Language Programming. Berkeley: Osborne/McGraw-Hill, 1986.
Norton, Peter, and John Socha. Peter Norton's Assembly Language
Book for the IBM PC. New York: Brady Communications, 1986.
Rector, Russell, and George Alexy. The 8086 Book. Berkeley:
Osborne/McGraw-Hill, 1980.
Sargent III, Murray, and Richard L. Shoemaker. The IBM PC from
the Inside Out. Reading: Addison-Wesley, 1986.
Skinner, Thomas P. An Introduction to Assembly Language
Programming for the 8086 Family. New York: John Wiley & Sons,
Inc., 1985.
Turley, James L. Advanced 80386 Programming Techniques.
Berkeley: Osborne/McGraw-Hill, 1988.
Wilton, Richard. Programmer's Guide to PC and PS/2 Video Systems.
Redmond: Microsoft Press, 1987.
References
483
484
Turbo Assembler User's Guide
N
80287 coprocessor 453
protected mode 453
80387 coprocessor 453
new instructions 454
.8086 directive 410
80186 processor 411-417
BOUND instruction 413
ENTER instruction 412
extended 8086 instructions 415
INS instruction 414
LEAVE instruction 412
multiplying immediate values 416
OUTS instruction 414
popA instruction 411
protected mode 33
PUSHA instruction 411
pushing immediate values 415
shifts/ rotates 416
Turbo C and 267
80188 processor 411
80286 processor 417
memory management 418
modes 418
protected mode 33, 418
registers 418
Turbo C and 267
80386 processor 419-453
32-bit instructions 442
48-bit type 424
6-byte variables 425
addressing modes 431
bit scanning 436
bit testing 435
conditional tests 439
converting values 437
DF directive 425
loading pointers 440
LOOP 442
mixing 16-bit/32-bit segments 446
Index
D
E
x
MOV 441
multiple-word shifts 438
multiplication 445
new instructions 434
privileged instructions 419
record size 381
registers 425
segment directives 423
string instructions 444
8087 coprocessor 27, 75
Turbo C and 302
.186 directive 411
.286 directive 418
.386 directive 419
.387 directive 454
8086 processor
architecture 46
1/050
instruction set 67
math capabilities 141
memory 47
registers 51
8088 processor 43, 46, 409, 410
<> (angle brackets)
conditional directives and 221
EQU directive and 178
<> (angle brackets) operator
within macros 373
.386C directive 419
.286P directive 418
.386P directive 420
[] (square brackets)
direct addressing and 100
Ideal mode 461
MASM mode 461
16-bit integers 118
32-bit integers 118
64-bit integers 118
32-bit operations 142
485
10-byte integers 118
\ (line continuation character) 199
;; operator
within macros 373
1 option (Turbo C) 267
. (period) character
Ideal mode 459, 464
in directives 468
. (period) operator 376
& character
parameters and 368
= directive 26, 179
! operator
within macros 373
& operator
within macros 373
% sign
Ideal mode directives and 468
@-sign :39
loca11abels and 351
% sign operator
within macros 373
$ symbol 178
in labels 82
? symbol
in labels 82
uninitialized data and 129
A
/ a option 25, 35
ABStype 195
absolute value 195
accumulator 54, 181
AOC instruction 143
ADD instruction 142
effect on carry flag 245'
addition 142
addressing modes 93
80386431
AH register 54
division 146
AL register 54
division 146
I/O 140
multiplication 145
returning values 167
.ALPHA directive 394
486
ALPHA directive 25, 35
ampersand, parameters and 368
AND instruction 148
records and 382
TESfvs.384
angle brackets 178
conditional directives and 221
angle brackets operator
within macros 373
ANSI.SYS 74
architecture (computer) 41
ARG directive
Turbo C and 300
arithmetic operations 42, 141
signed numbers 136
arrays
initialization 125
asm (keyword) 258
.ASM files 1, 23
assembling
conditional 216
first program 12
inline
Turbo C 258-280
number of passes 31
assembly language
defined 44
Turbo C vs. 45
ASSUME directive 112, 190, 397
CS register and 397
group names and 396
AT combine type 390
at-sign 39
AX register 54
division 146, 147
I/O 140
multiplication 145
returning values 167
B
-B option (Turbo C) 264
/b option 24
/soption 25
backslash
include files and 199
base notation
.RADIX directive and 123
Turbo Assembler User's Guide
BASIC See Turbo Basic
Basic Input/Output System See BIOS
BCD values 118
BH register 55
%BIN directive 214
'BIN files
stack and 231
binary notation 119
BIOS 73
colors 73
OS segment 107
screen mode 73
bit fields 380
bits 116
manipulating 148
scanning (80386) 436
shifting 150
testing (80386) 435
BL register 55
Boolean algebra 148
BOUND instruction
80186413
Ideal mode 474
bounds checking 413
BP register 58
as memory pointer 97, 100
memory operands and 346
branching instructions 153
instruction pointer and 61
BSF instruction 436
BSR instruction 436
BT instruction 435
BTC instruction 435
BTR instruction 435
BTS instruction 435
buffers
copying 182
size of 24
BX register 55, 66
as memory pointer 96
division 147
BYrE PTR operator 134
BYrE type 130
bytes 117
BYTE PTR operator 134
CMPSB 188
converting to words 137, 241
Index
LOOSB instruction 181
MOVSB instruction 183
SCASB instruction 185
setting conditionally 439
STOSB instruction 182
string instructions 189
c
C SeeTurboC
/c option 26
CALL instruction
labels and 93
subroutines and 162
calls, forward references and 361
carry flag 53
32-bit arithmetic 143
unexpected effects 245
case sensitivity
assembler routines and 32
public labels and 193
TurboC291
CBW instruction 137
CDQ instruction 438
CGA68
CH register 55
character constants 123
characters
as operands 90
displaying 71
DOS functions 71
initializing strings 126
CL register 55
rotates 153
shifts 151
CLD instruction 181,239
CMP instruction
SCASvs.185
CMPS instruction 188
ES register and 346
repeating 239
CMPSB instruction 188
CMPSW instruction 188
code-checking 33
.CODE directive 104
code segment 65
multiple 404
Turbo C 282
487
Turbo Pascal 317
@codeSize symbol 110
colon, in labels 84
Color Graphics Adapter See CGA
colors, BIOS and 73
.COM files
stack and 231
combine types 207
AT 390
COMMON 391
MEMORY 391
PRIVATE 392
PUBLIC 391
STACK 391
command files
indirect 39
command-line syntax 22
help screen 27
comments 101
Ideal mode 475
inline assembly and 268
COMMON combine type 391
communications software 74
compact memory model 108
comparisons 185
repeating 239
compatibility with other assemblers 197,210,
221, 253, See also MASM compatibility
compiler options See individual listings
conditional assembly
directives 216
macros and 369
conditional error directives 223
conditional jumps See jumps, conditional
conditional tests (80386) 439
conditionals
assembler vs. other languages 234
in listing files 37
suppressing 211
vs Turbo C's 218
%CONDS directive 211
configuration files 39
.CONST directive 110, 402
constants
$178
as operands 90
character 123
488
expressions 91
restrictions on 90
conversion
bytes to words 137
problems 241
data sizes 136
words to doublewords 137
copying data See data, copying
counting
ex register and 55
%CREF directive 209
.CREF directive 210
CREF table 207
cross-reference
generating 23
in listing files 26, 207
cross-reference utility See TCREF utility, See
OBJXREF utility
CS override 33
CS register 65
ASSUME directive and 112,397
subroutines and 164
CSEG 317
%CTLS directive 212
@curseg symbol 110
CWO instruction 137
CWOE instruction 437
CX register 55
LOOP instructions and 442
loops and 159
repeated string instructions 237
D
/d option 26
--d switch 219
data
allocating 116-141
converting size 136
copying 132
forward references 360
initialization 125
moving 132-141
80386 437
string instructions 181
scanning 185, 436
size 133
BYTE PfR operator 134
Turbo Assembler User's Guide
string instructions 189
WORD PfR operator 134
swapping 139
types 116, 130
labels 195
multiple 387
TurboC292
UNKNOWN 131
uninitialized 129
.DATA? directive 110,402
.DATA directive 105
ES register and 107
data segment 66, 116-141
.CONST directive 110
.DATA? directive 110
.DATA directive 105
.FARDATA directive 110
multiple 404
Turbo Pascal 317
data structures See structures
@data symbol 105
date 74
??Date variable 179
DB directive 125
DD directive 123, 125
debugging 38
DEC instruction 144
effect on carry flag 245
decimal notation 119
decrementing, defined 144
%DEPfH directive 214
DF directive 125
80386 425
DGROUP396
DH register 56
DI register 57, 66
as memory pointer 97
string instructions 185
direct addressing 95
direction flag 53
incorrect setting 239
string instructions and 181
directives See also individual listings
conditional assembly 217
conditional error 223
data definition 125
defined 86
Index
Ideal mode 467
period in 468
processor control 468
processor-type 410
segments 79, 103, 112, 389-408
startup 29
string space 30
symbols 29
Disk Operating System See DOS
DISPLAY directive 216
displaying characters See characters, displaying
DIV instruction 146
divide-by-zero interrupt 147
division 145
AX register and 54
DX register and 56
REPf directive and 363
signed 146
SARand 151
unsigned 147
SHRand 151
DL register 56
dollar sign symbol 178
in labels 82
DOS 69
calling 70
character display 71
keyboard input 70
PSP 316
returning to 226
terminating programs 72
wildcards 22
DOSSEG directive 108, 393
Turbo C and 283
doublewords 118
converting to quad words 437
converting to words 137
DP directive 125
DQ directive 123, 125
DS register 66
ASSUME directive and 112
BIOSand 107
.DATA directive and 105
memory operands and 345
string instructions and 240
DSEG317
DT directive 123
489
DUP operator 126
REPTvs.363
structures and 376
DW directive 125
DWORD type 130
DX register 56
division 147
I/O 140
multiplication 146
E
/eoption 27
EAX register 426
EBP register 426
EBX register 426
ECX register 426
EDI register 426
EDX register 426
ELSE directive 217
ELSEIF directives 222
EMUL directive 27
END directive 79, 86
start address and 87
ENDIF directive 217
ENDM directive 362
ENDP directive
Ideal mode 469
subroutines and 164
ENDS directive 112, 389, 402
Ideal mode 469
ENTER instruction (80186) 412
EQU directive 174
angle brackets and 178
Ideal vs. MASM mode 456, 461
equal (=) directive 26
equate substitutions 174
equates
text and numeric 461
.ERR1 directive 223
.ERR2 directive 223
.ERR directive 223
.ERRB directive 225
.ERRDEF directive 224
.ERRE directive 224
.ERRNB directive 225
.ERRNDEF directive 224
.ERRNZ directive 224
490
error messages 16
conditional 223
source file line display 37
errors, programming 225-256, See also pitfalls
ES register 66
ASSUME directive and 112
.DATA directive and 107
LES instruction and 440
string instructions and 240, 346
ESI register 426
ESP register 426
exclamation mark operator
within macros 373
.EXE files 1, 13
execution, END directive and 87
EXITM directive 371
expressions 91
Ideal mode 461
initializing variables 128
operators in 92
external symbols See symbols, external
extra segment 66
segment overrides and 348
EXTRN directive 194
Ideal vs. MASM mode 473
Turbo C and 293
Turbo Pascal and 321
F
far data 108, 110
Ideal mode 472
FAR PTR operator 135
forward jumps and 360
far subroutines 165
FAR type 130
FAR PTR operator and 135
.FARDATA? directive 110,404
@fardata? symbol 110
.FARDATAdirective 110,402
@fardata symbol 110
FCOS instruction 454
@FileName symbol 110
??Filename variable 179
files
.ASM23
configuration 39
include 198
Turbo Assembler User's Guide
indirect 39
listing 23, 200, See also listing files
flags register 52
80286 418
80386 428
problems 246
string instructions 187
floating-point
emulation 27
Ideal vs. MASM mode 460
instructions 2
numbers 118, 122
operations 75
format, code 81
forward references 358
structures 373
to macros 372
FPREM1 instruction 454
FS register 429
LES instruction and 440
FSEfPM instruction 453
FSIN instruction 454
FSINCOS instruction 454
FUCOM instruction 454
FUCOMP instruction 454
FUCOMPP instruction 454
FWORD type 130
80386 424
G
general-purpose registers 54, See also
individ uallistings
80386 426
AX 54
BP58
BX55
changing sign 147
CX55
0157
DX56
5156
SP59
GLOBAL directive 197
Ideal vs. MASM mode 473
graphics 74
GREP.COM8
GROUP directive 395
Index
Ideal vs. MASM mode 470
grouping segments 252,395
Ideal mode 470
GS register 429
LES instruction and 440
H
/hoption27
hard ware, direct access to 74
heap, Turbo Pascal 318
help screen, displaying 27
hexadecimal notation 62, 121
high-level languages
80186 and 412
linking to 109-110, 115, 253
returning values 166
segment groups 396
segment ordering 393
HIGH operator
Ideal mode 465
huge memory model 109
/i option 28
-i switch 199
iAPx86 processor family 46, 409
IBM PC
features 67
1/068
systems software 68
IBM XT See IBM PC
IDEAL directive 457
Ideal mode 1, 455-481
BOUND instruction 474
directives 467
equates and 461
expressions 461
external symbols 473
floating-point numbers 460
group overrides 253
listing controls 467
local labels 352
local symbols 475
macro comments 475
near/far data 472
operands 461
491
operators 464
segment fixups 474
segment groups 470
speed 457
structures/unions 460, 473
tokens 459
IOIV instruction 147
IFl directive 221
IF2 directive 221
IF directive 217
IFB directive 220
IFDIF directive 220
lFE directive 218
IPION directive 220
IFNB directive 220
IMUL instruction 146
80186 416
80386445
problems 244
IN instruction 56, 140
INC instruction 144
effect on carry flag 245
%INCL directive 212
INCLUDE directive 28, 198
Ideal mode 469
include files 198
GLOBAL directive and 197
setting path 28
suppressing in listing files 212
incrementing, defined 144
indirect addressing 95
indirect command files 39
initialization
arrays 125
character strings 126
data 125
expressions and 128
labels and 128
record variables 381
structures 379
inline assembly
Turbo C 258-280
format 268
limitations 274
input/output See I/O
INS instruction (80186) 414
INSTALL.EXE 8
492
installation 8
instruction mnemonics See mnemonics
instruction pointer 61, 65
80386 428
instruction prefixes 184
instruction set See also individual listings
8086 67
INT instruction
character display and 71
keyboard input and 71
integers
32-bit 118
64-bit 118
lO-byte 118
long 118
short 118
interrupt flag 53
interrupt handler
preserving registers 251
interrupts
0147
divide-by-zero 147
I/O 140
8086 50
80186 414
AL register 140
AX register 140
OX regist~r 56, 140
formatted 71
IBM PC 68
keyboard 70
operations 42
IP register 61, 65
80386 428
subroutines and 164
IRETD instruction
80386 444
IRP instruction
repeat blocks and 364
IRPC instruction
repeat blocks and 364
J
/j option 29
JA instruction 157
JAE instruction 157
JB instruction 157
Turbo Assembler User's Guide
JBE instruction 157
JC instruction 157
JCXZ instruction 160, 238
80386 443
JE instruction 156, 157
JECXZ instruction
80386 443
jEMUL option 27
JG instruction 157
JGE instruction 157.
JL instruction 157
JLE instruction 157
JMP instruction 154
labels and 93
JNA instruction 157
JNAE instruction 157
JNB instruction 157
JNBE instruction 157
JNC instruction 157
JNE instruction 157
JNG instruction 157
JNGE instruction 157
JNL instruction 157
JNLE instruction 157
JNO instruction 158
JNP instruction 157
JNS instruction 157
JNZ instruction 157
80386 443
JO instruction 157
JP instruction 157
JPE instruction 157
JS instruction 157
jumps
80386 428, 443
conditional 156
local labels and 352
problems with 234
size of 353
FARPTRand 155
forward referenced 358
limitations in Turbo C 274
short 154
unconditional 154
JUMPS directive 354
JZ instruction 157, 160
Index
K
keyboard input, DOS functions 70
keywords 31
/kh option 29
/ks option 30
L
$1 directive (Turbo Pascal) 319
/1 option 26, 30, 33
/la option 30
LABEL directive 130,351
labels
= directive and 179
as operators 92
conditional jumps and 352
conflicting definitions 219
data types 195
EQU directive and 174
equating to values/strings 174
external 194
Turbo C and 290
EXTRN directive and 194
for memory locations 130
Ideal mode 352
in macros 371
initializing variables 128
listing 205
loca1349
MASM mode 352
modules and 193-197, 349
PUBLIC directive and 193
redefining 179
requirements 82
undefined 224
language elements 81
large memory model 108
LARGE operator
80386 and 421
LEA instruction
vs. MOV OFFSET 255
LEAVE instruction (80186) 412
LES instruction 440
LFCOND directive 37
line continuation character 199
linking See also TLINK utility
first program 13
493
high-Ievellanguages 115
returning values 166
Turbo C 253, 294, 308
%LINUM directive 214
%LIST directive 211
listing files 23, 200-215
control directives 210
Ideal mode 467
MASM mode 468
%CREF directive 209
cross-reference information 26, 207
false conditionals in 37
format 213
generating 30
high-level code in 30
%NOCREF directive 209
numbers in 199
page size 214
suppressing lines 210
symbol tables
suppressing 33, 205
titles 213
Ideal mode 469
LOCAL directive
in macros 371
Turbo C and 298
local symbols
Ideal vs. MASM mode 475
LOCALS directive 352
LODS instruction 181
multiple segments and 408
LODSB instruction 181
segment override and 347
LODSW instruction 181
logical operations 42, 148
long integers 118
LOOP instruction 56, 159
80386 442
effect on carry flag 245
problems 238
LooPD instruction (80386) 442
LooPDE instruction (80386) 443
LooPDNE instruction (80386) 443
LooPE·instruction 160
80386 442
looping 158
ex register and 55
494
LooPNE instruction 160
80386 442
LooPZ instruction 160
LOW operator
Ideal mode 465
.LST files 23, 200
M
1m option 31
machine language, defined 44
MACRO directive 372
macros 365
comments
Ideal mode 475
conditional assembly directives 369
conditional error directives 225
expansion
EXITM directive 371
suppressing listing 212
forward referenced 372
labels in 371
nested segments 393
operators within 372
subroutines vs. 366
%MACS directive 212
MAKE.EXE 7
MASK operator 384
MASM compatibility 1,82,210,215,253
equates 461
expressions 461
Ideal mode vs. 455-481
listing control directives 468
local labels 352
OFFSET operator bug 397
segment groups 397, 470
segment ordering 108
structures 373
Turbo C and 265
MASM directive 457
MASM mode See MASM compatibility
. math, 8086 141
math coprocessor See numeric coprocessor
medium memory model 108
memory
addressing
80386 431
large blocks 64
Turbo Assembler User's Guide
modes 93
square brackets and 100
blocks of 126, 184
comparing blocks 185
defining variables 125
direct addressing 95
filling blocks of 184
indirect addressing 95
management (80286) 418
mapping 51
models
FAR type and 130
.MODEL directive and 108
NEAR type and 130
PROC type and 131
segments and 400
TurboC282
Turbo Pascal 315
naming locations 130
operands 97
BP ~egister and 346
DS register and 345
ES register and 346
pointers 96
BP register and 58
BX register and 55
CS register and 65
DI register and 57
DS register and 66
ES register and 66
51 register and 56
SP register and 59
55 register and 66
reserving 129
scanning 185
segment names 64
segmentation 61
string instructions and 57, 58
variables
problems 246
MEMORY combine type 391
messages
displaying during assembly 215
suppressing 35
Microsoft Assembler See MASM compatibility
mixed-model programming
segment directives and 115
Index
Iml option 31
-ml switch 193
MMACROS.ARC 8
'mnemonics, defined 85
.MODEL directive 108, 402
modular programming 162, 190
END directive and 87
EXTRN directive 194
.FARDATA segment and 403
GLOBAL directive 197
loca1labels 349
PUBLIC directive 193
MOV instruction 132-139
80386 441
addressing modes and 98
forward references 360
string instructions vs. 181
vs. LEA 255
moving data See data, moving
MOVS instruction 183
ES register and 346
MOVSB instruction 183
80386 444
MOVSD instruction
80386 444
MOVSW instruction 183
MOVSX instruction 437
MOVZX instruction 437
MS-DOS See DOS
.MSFLOAT directive 122
Imu option 32
MUL instruction 145
problems 244
multiple prefixes 242
multiplication 145
80386 445
AX register and 54
DX register and 56
pitfalls 244
REPf directive and 363
SHLand 150
signed 146
unsigned 145
I mV# option 32
I mx option 32
-ml switch 193
495
N
/noption 33
-n switch 205
names 82
near data 108
Ideal mode 472
NEAR PfR operator 135
near subroutines 165
NEAR type 130
NEARPTRoperatorand 135
NEG instruction 147
%NEWPAGE directive 213
%NOCONDS directive 211
.NOCREF directive 210
%NOCREF directive 209
%NOCTLS directive 212
NOEMUL directive 27
%NOINCL directive 212
NO]UMPS directive 357
%NOLISf directive 211
NOLOCALS directive 352
%NOMACS directive 212
%NOSYMS directive 213
NOT instruction 149
NOTHING directive 398
NOTHING keyword 113
%NOTRUNC directive 214
%NOUREF directive 213
NUL device 24
numbers
in labels 82, 174
include files and 199
signed 135
unsigned 135
numeric coprocessor 27, 453
o
.OB] files 1, 13
.OB] files
suppressing 34
object See object files
object files
debugging information in 38
line number information in 38
segment ordering 25, 35
496
object modules
defined 12
OB]XREF.COM 8
octal notation 120
OFFSET operator
MASM vs. Ideal mode 470
problems 247, 253
offsets
$ symbol and 178
operands
character 90
constant 90
defined 88
Ideal mode 461
label 92
limitations in Turbo C 274
memory 97
BP register and 346
ES register and 346
order of 230
register 89
source/ destination 88
string instructions with 189
operators See also individual listings
expressions and 92
Ideal vs. MASM mode 464
macros 372
optimization, Turbo Pascal 325
options, command line See command-line
options
OR instruction 149
OS/2418
OUT instruction 56, 140
%OUT directive 216
OUTS instruction (80186) 414
overflow flag 53
conditional jumps and 157
p
/p option 33
%PAGESIZE directive 214
parameter passing
registers 166
stack 166
testing 220
TurboC 295
Turbo Pascal 325
Turbo Assembler User's Guide
parameters
formal 367
macros and 367
parity flag
conditional jumps and 157
parsing order
Ideal mode 469
Pascal See Turbo Pascal
PC-DOS See DOS
%PCNT directive 214
percent sign
Ideal mode 468
percent sign operator
within macros 373
period
in directives 468
in labels 82
Ideal mode 459
in structures
Ideal mode 464
operator 376
pitfalls 225-256
carry flag 245
conditional jumps 234
converting bytes to words 241
direction flag 239
flags 246
interrupt handler 251
linking to Turbo C 253
LOOP instruction 238
memory variables 246
multiple prefixes 242
multiplication 244
OFFSET operator 247
REP prefix 239
REP string overrun 235
returns 228
reversing operands 230
segment groups 252
segment wraparound 249
stack allocation 230
string comparisons 239
string instruction operands 242
string instructions 235, 245
string segment defaults 240
subroutines 227, 228, 231
termination 226
Index
wiping out registers 231
zeroCX237
plus operator 376
plus sign 23
pointers
defined 44
I/O
OX register 140
OX register and 56
memory 55, 96
BP register and 58
CS register and 65
01 register and 57
OS register and 66
ES register and 66
51 register and 56
SP register and 59
55 register and 66
segment:offset 118
segmentation and 61
to structures
Ideal mode 464
POP instruction 59, 139
POPA instruction (80186) 411
POPAO instruction
80386 445
POPFD instruction
80386 445
%POPLCTL directive 215
#pragma directive 264
predefined symbols See symbols
prefixes 184
local label 353
multiple 242
segment override 345
printing, first program 17
PRIVATE combine type 392
PROC directive 131,351
high-level languages and 109
subroutines and 164, 228
processor, defined 43
Program Segment Prefix (PSP) 316
program structure 78-172
program termination 72, 79, 86
problems with 226
Prolog See Turbo Prolog
protected mode 33
497
80286 418
80287453
protected mode instructions
80286 418
PSP 316
PTR operator
Ideal mode 466
PUBLIC combine type 391
PUBLIC directive 193
Ideal mode 473
Turbo Pascal and 320
public functions
Turbo C and 290
public labels 193-197
Ideal mode 473
PUSH instruction 59, 139
80186 415
PUSHA instruction (80186) 411
PUSHAD instruction
80386445
PUSHFD instruction
80386445
%PUSHLCTL directive 215
PWORDtype 130
Q
Iq option 34
quad words 118
converting to doublewords 437
question mark
in labels 82
uninitialized data and 129
QWORD type 130
R
Ir option 27, 34
.RADIX directive 123
RCL instruction 153
RCR instruction 152
README.COM 7
real mode (80286) 418
real numbers 122
RECORD directive 380
records 381
recursion 163
498
registers See also individual listings
8086 51
80386 425
32-bit 426
as operands 89
incrementing I decrementing 144
parameter passing 166
preserving 167, 231
preserving (Turbo C) 277, 302
setting to zero 149
Turbo Pascal 318
REP prefix
problems 239
segment override and 348
string overruns 235
REPE instruction 187
repeating instructions 184, 362
REPNE instruction 187
REPNZ instruction 187
REPT directive 362
DUPvs.363
REPZ instruction 187
reserved words 80
RET instruction
PROC directive and 228
subroutines and 162,227
ROL instruction 152
ROR instruction 151
rotates 151
5
-5 option (Turbo C) 262
IS option 394
I s option 25, 35
SAL instruction 150
SAR instruction 151
SBB instruction 143
SCAS instruction 185
ES register and 346
repeating 239
SCASB instruction 185
screen mode
ANSI.SYS 74
BIOS and 73
SEG operator
Turbo Pascal and 324
SEGMENT directive 112, 389
Turbo Assembler User's Guide
segment:offset pointers 118, 404
segments
80386 420, 446
accessing multiple 349
alignment
80386 423
types 206, 390
alphabetical order 394
code 65, 104, 404
Turbo Pascal 317
combine types 207, 390
current 112
data 66, 105, 110, 404
directives 79, 103, 389-408
80386 423
high-level languages and 115
simplified 104-110, 399, 400
symbols and 110
standard 111-116
Turbo C and 283
end of 112
extra 66
fixups
Ideal vs. MASM mode 474
groups 395
Ideal mode and 456, 470
MASM mode 470
problems 252
Turbo C and 253
listing 206
multiple 404
names 64, 392
nesting 392
ordering 25, 108, 392, 393
override
prefixes 345
registers 61, See also individ uallistings
80386 429
CS65
DS 66, 105
ES 66, 107
FS429
GS429
memory pointers to 56, 57
moving data between 138
SS66
sequential order 25, 35, 394
Index
stack 66
start of 112
Turbo C and 282
USE16420
USE32420
wraparound problem 249
semicolon 22
inline assembly and 268
semicolon operator
within macros 373
SEQ directive 25
.SEQ directive 394
serial communications 74
SET instruction 440
SETNC instruction 440
SETS instruction 440
SFCONri directive 37
shifts 150
multiple word (80386) 438
SHL instrl:lction 150
80186 416
SHLD instruction 438
short integers 118
SHORT operator 154
forward jumps and 360
SHR instruction 151
80186 416
records and 382
SHRD instruction 439
SI register 56, 66
as memory pointer 97
string instructions 185
sign, changing 147
sign flag
conditional jumps and 157
signed conversion instructions 437
signed division 146
SARand 151
signed multiplication 146
signed numbers 135
simplified segment directives 104-110
80386 423
ASSUME and 399, 400
size of data See data, size
SIZE operator
Ideal mode 466, 473
499
slash
include files and 199
small memory model 108
SMALL operator
80386 and 421
source files
include files 28
symbols 26
SP register 59, 66
speaker 75
square brackets 100
Ideal mode 461
MASM mode 461
SS register
LES instruction and 440
memory operands and 346
memory pointers to 58
stack
80186 instructions 412
allocating 230
combine types and 391
MOV instruction and 139
parameter passing 166
pointer 59, 66
segment 66
segment override and 348
size of 104
Turbo Pascal 317
STACK combine type 391
.STACK directive 104, 230
stack segment register (SS)
memory pointers to 58
standard segment directives 111-116
start address
END directive and 87
statistics, displaying 24
STD instruction 239
STOS instruction 182
ES register and 346
multiple segments and 408
STOSB instruction 58, 182
STOSW instruction 182
pitfalls 241
string instructions 180
80386 444
BP register 58
bytes vs. words 189
500
CMPS 188
data movement 181
decrementing 239
DI register 57
direction flag 239
effect on registers 245
ES register 107
extra segment 66, 348
flags 187
incrementing 239
LODS 181
mixing with non-string 349
MOVS 183
multiple prefixes 242
operands to 242
pitfalls 235
REP prefix 184, 187
repeating 184, 187, 235
SCAS 185
SI register 57
stos 182
strings
.
assigning to hll,Jels 174
comparing 239
displaying 216
initialization 126
quoted (Ideal mode) 469
STRUC directive 374
structures 373-389
DUP operator and 376
forward references 373
Ideal mode 380, 460, 473
initialization 379
MASM mode 373
period operator 376
Turbo C 269, 303
SUB instruction 142
subroutines 161
far 165
local labels 349
macros vs. 366
near 165
preserving registers 231
RET instruction 227
subtraction 142
%SUBTTL directive 213
Turbo Assembler User's Guide
symbol tables
listing files 205
cross-referencing 26
suppressing 33,205,213
symbols
case-sensitive 31, 32
@CodeSize 110
@curseg 110
@data 105
defining 26
external 32
Ideal mode 473
Turbo C and 293
@fardata 110
@fardata? 110
@FileName 110
Ideal mode 459
length of 32
local
Ideal mode 475
public 32
restrictions 26
unreferenced 213
uppercase 32
%SYMS directive 213
.SYMTYPE operator
Ideal mode 465
syntax, command-line See command-line
syntax
system timers 74
systems software
IBM PC 68
T
/t option 35
TASM.CFG files 39
TASM.EXE 7
TBYTE type 130
TCREF 1
TCREF.EXE 7
termination 72
DOS functions 72
END directive and 79, 86
problems with 226
TEST
ANDvs.384
%TEXT directive 214
Index
text strings See strings
TFCOND directive 37
time 74
??Time variable 179
timers 74
tiny memory model 108
%TITLE directive 213
Ideal mode 469
TLIB.EXE 7
TLINK 1, 13, 294, 308
segment ordering 394
Turbo C version 259
TLINK.EXE 7
tokens
Ideal mode 459
TOUCH.COM8
trap flag 53
%TRUNC directive 214
Turbo C 257-313
80186/80286 processor 267
80186 and 412
-1 option 267
ARG directive and 300
assembler modules in 27
assembly language vs 45
-B option 264
case sensitivity 32, 291
code segment 282
data types 292
external symbols 293
floating-point emulation 27
inline assembly 258-280
comments 268
format 268
limitations 274
semicolon 268
jumps 274
linking to 253, 308
LOCAL directive and 298
MASMand 265
memory models 282
operand limitations 274
parameter passing 295
Pascal calling conventions 307
path-naming conventions 199
period operator 376
#pragma directive 264
501
public functions and 290
register preservation 277, 302
returning values 302
-S option 262
segment directives and 283
structures 269, 303
Turbo Debugger 1, 38
Turbo Librarian See TLIB utility
Turbo Link See TLINK utility
Turbo Pascal 315-344
allocating local data 332
data segment 317
EXTRN directive 321
functions results 331
heap 318
making assembler information available to
320
making information available to assembler
321
memory map 315
memory models 315
near / far calls 319
optimization 325
parameter passing 325
PUBLIC directive 320
register usage 318
returning values 331
segment fixups 324
stack 317
tutorial 9-75
two-pass assemblers
compatibility with 223
two's complement arithmetic 136
type-checking
Ideal mode 456
type specifiers 130
.TYPE operator
Ideal mode 465
typefaces in this manual 3
types See data, types
u
unconditional jumps See jumps, unconditional
underscore
Turbo C and 290
uninitialized data 129
UNION directive 387
502
unions
Ideal vs. MASM mode 460
UNKNOWN type 131
unsigned division 147
SHRand 151
unsigned multiplication 145
unsigned numbers 135
%UREF directive 213
USE16 segment 420, 446
USE32 segment 420, 446
V
/v option 24, 35
variables
changing sign 147
converting to strings 72
incrementing/decrementing 144
inline assembly and 276
memory 125
record 381
uninitialized 129
W
/woption 36
warning messages 16
"mild" 36
generating 36
WIDTH operator 383
wildcards See DOS wildcards
WORD PTR operator 134
WORD type 130
words 117
CMPSWand 188
converting to bytes 137
converting to doublewords 137
LODSWand 181
MOVSWand 183
SCASW and 187
srOSWand 182
string instructions 189
WORD PTR operator 134
X
/xoption 37
.xRF files 23
XCHG instruction 139
Turbo Assembler User's Guide
XLAT instruction
operands to 243
XOR instruction 149
Z
/z option 37
Index
/ zd option 38
zero ex value
string instructions and 237
zero flag 53
conditional jumps and 156
loops and 160
/ zi option 38
503
T RB
A E BLERO
BORLAND
scons
1800 GREEN HILLS ROAO, P 0, BOX 660001 ,
VALLEY, CA 95067-0001 , (408)438-5300
UNIT 8 PAVlUONS, RUSCOMBE BUSINES6 PARK, TWYfORD, BERKSHIRE RG 10 9NN, ENGLAND
43 AVENUE DE L'EUROPE - BP 6, 78141 VEUlY VlLLACOUBLAY CEDEX fRANCE
•
PART. 15MN-ASOO2 -20
•
BOR 1491
Source Exif Data:
File Type : PDF
File Type Extension : pdf
MIME Type : application/pdf
PDF Version : 1.3
Linearized : No
XMP Toolkit : Adobe XMP Core 4.2.1-c043 52.372728, 2009/01/18-15:56:37
Producer : Adobe Acrobat 9.12 Paper Capture Plug-in
Modify Date : 2009:07:18 10:09:15-07:00
Create Date : 2009:07:18 10:09:15-07:00
Metadata Date : 2009:07:18 10:09:15-07:00
Format : application/pdf
Document ID : uuid:a3feaab1-3239-4146-8912-00dde15c1a39
Instance ID : uuid:adb48330-da18-4de1-a838-34826e2131fb
Page Layout : SinglePage
Page Mode : UseNone
Page Count : 518
EXIF Metadata provided by EXIF.tools
.