Using Assembler

in Delphi
Fourth revision - May 2010

Guido Gybels
www.guidogybels.eu

The first edition of this paper was created in the late nineties and based on the
behaviour of Delphi 2. While I had a need to use assembler in some of my
projects, I quickly found out that there wasn't much documentation available on
the topic of how to properly integrate assembler code into Delphi programmes.
This paper was written with the intention of bridging that gap. It has since been
through a number of revisions. Processors, and Delphi, have evolved
significantly since I first wrote this paper, but it should retain a lot of its original
value.

Table of Contents
TABLE OF CONTENTS ........................................................................................................... 3
INTRODUCTION...................................................................................................................... 4
BASIC PRINCIPLES................................................................................................................ 6
CHAPTER 1: GENERAL CONTEXT OF ASSEMBLER CODE ................................................ 7
1.1. W HERE TO LOCATE THE ASSEMBLER CODE ......................................................................... 7
1.2. LABELS ........................................................................................................................... 7
1.3. LOOPS............................................................................................................................ 9
1.4. ENTRY AND EXIT CODE ....................................................................................................10
1.5. REGISTER PRESERVATION ...............................................................................................11
CHAPTER 2: PASSING PARAMETERS.................................................................................12
2.1. CALLING CONVENTIONS ..................................................................................................12
2.2. PASSING PARAMETERS IN REGISTERS ...............................................................................12
2.3. USING THE STACK FOR PARAMETER PASSING .....................................................................15
2.4. PASSING BY VALUE VERSUS PASSING BY REFERENCE ..........................................................17
CHAPTER 3: LOCAL VARIABLES ........................................................................................20
3.1. LOCAL VARIABLES AND THE STACK FRAME .........................................................................20
3.2. SIMPLE TYPES AS LOCAL VARIABLES .................................................................................22
3.3. RECORDS AS LOCAL VARIABLES........................................................................................23
3.4. HEAP ALLOCATED TYPES AS LOCAL VARIABLES ...................................................................24
CHAPTER 4: RETURNING RESULTS....................................................................................26
4.1. RETURNING INTEGERS AS IMMEDIATE VALUES ....................................................................26
4.2. RETURNING BOOLEANS AS IMMEDIATE VALUES ...................................................................27
4.3. RETURNING REAL NUMBERS .............................................................................................28
4.4. RETURNING CHARACTERS................................................................................................30
4.5. RETURNING A LONG STRING .............................................................................................31
FURTHER READING ..............................................................................................................35
TABLE 1: USE OF CPU REGISTERS ....................................................................................36
TABLE 2: CALLING CONVENTIONS .....................................................................................38
TABLE 3: PARAMETER PASSING ........................................................................................39
TABLE 4: RETURNING RESULTS .........................................................................................42
ABOUT THE AUTHOR ...........................................................................................................44

Introduction
Many programmers still associate assembler with a difficult, low-level kind of
programming. Most of them also think it is incomprehensible and impossible to
master. In reality, things are not that bad and these perceptions are mostly
founded in unfamiliarity. It is quite possible to learn how to write good assembly
code without being a genius. On the other hand, don't think a few lessons in
assembly will leave you producing faster code than the average Delphi/Pascal
equivalent. The reason is that when you write Delphi code, you are competing
with an efficient and experienced assembler programmer: the compiler. Overall,
the code produced by it is efficient and fast enough for most applications.
Most people grab for assembler in order to achieve better performance for their
software. However, the key to performance is mostly design, not choice of
language. There is little point in trying to fix a bad algorithm through assembler.
So, the first thing to do when you are experiencing bottlenecks, is to review your
code architecture and Pascal implementation, rather than immediately taking
recourse to assembler. Writing it in assembler is not going to turn a bad
algorithm or a bad approach into a good one.
Similarly, most programmers erroneously believe that writing code in assembler
by definition means it's fast and assume that code written in assembler is by
definition faster than compiled Delphi/Pascal code. That is certainly not the
case. Badly written assembler routines will perform on an inferior level and can
even cause strange bugs and problems in your application.
There is however a good case for hand crafting assembler code in certain
cases. Delphi's (Pascal) language is a general programming language and
certain specialised tasks could be done better in assembler. Also, taking
advantage of processor specific features might require manual code design.
It is not within the scope of this article to teach you the principles of assembler
programming. There are other information resources out there discussing
assembler programming. See Further Reading to find some pointers to relevant
material.
Before resorting to rewriting code in assembler, investigate first where your
bottlenecks are and why. A profiler might be a useful tool during this analysis.
Once you have identified the cause, take a step back and look at the structure
and algorithm. Often, you can get much better performance by revising the
algorithm or the application design, rather than just throwing in some
assembler. On the other hand, in some particular cases, assembler might
indeed be a better and even simpler choice.
Once you conclude that assembler is actually needed, you should take time
proper to draw up a plan for your code and design the algorithm. Only when you
Guido Gybels: Using Assembler in Delphi

have a clear idea of what you want to do and how you will be implementing it,
you can start coding. If you don't think about these issues, you'll end up with
spaghetti code, convoluted statements and an unmaintainable program. Not to
mention the possibility of introducing nasty bugs.

Guido Gybels: Using Assembler in Delphi

Basic Principles
During the implementation phase, there are several general principles you
should follow. The first important rule is: keep your routines as short as
possible. Assembler should be used for some real core activity in which
performance and simplicity are essential. So, in most cases, you can manage
with fairly short and specialised routines. If you see that you have plenty of long
pieces of assembler in your project, you are probably over-enthusiastic about it.
Secondly, keep the routines readable by commenting them in a meaningful
way. Because of the basic linear nature of assembly code, the individual
statements are quite easy to read. Comments are needed to clarify how you
want to implement the algorithm, so that a third party will immediately
understand what is going on. Your comments need to add valuable information
for the readers of your code. So, the following is clearly wrong:
inc edx {increase edx}

Such comments are totally pointless, since the instruction itself is making it
blatantly clear that you are incrementing the edx register. Comments should
indicate the inner workings of the algorithm and/or provide information not
otherwise immediately clear from the code and its context, rather than just
rephrase what the mnemonics already show. So, the above could, for instance,
be replaced by something like this:
inc edx {go to the next element in our product table}

Thirdly, you should avoid the use of slow instructions wherever possible. In
general, the simple instructions are preferable over the complex opcodes, since
on modern cpus the latter are implemented in microcode. Google for Agner
Fog's publications on the topic of code optimisation, which discuss this and
many other very useful aspects of how processors behave.
The final general principle is predictable: test your assembler routines
thoroughly and continuously. For routines written in assembler, the compiler will
produce substantially less warnings and error messages, and will offer no
feedback with regard to the basic logic of the assembler code. Unused variables
and wrong use of pointers will not be detected as easily as in Pascal code. So,
prepare for intensive testing and debugging of your code. It will take more time
and effort when compared to debugging regular Pascal code. That is yet
another reason to keep these routines as short and readable as possible.
If after all of this, you still want to go ahead with writing assembler code for your
Delphi/Pascal programs, reading this article is probably a good first step. I have
designed it to be as generic as possible.

Guido Gybels: Using Assembler in Delphi

Chapter 1: General Context of Assembler

Code
To successfully use assembler code inside your Delphi projects, you need to
understand how to call routines written in assembler, have access to the
parameters passed and be able to return a result. This chapter discusses where
assembler code should be located and what its basic structure should look like.
It also explains the compiler's behaviour in generating entry and exit code.
Note: In this document, I will always specifically indicate the calling convention
used in the examples. Although in the case of register this is superfluous
(since it is the default convention that is used when no calling convention is
specifically indicated), it contributes to the readability (look at it as an additional
comment if you want) as it reminds the reader that parameters might be located
in registers. Credit for this tip goes to Christen Fihl.

1.1. Where to locate the assembler code

Delphi uses the asm statement to indicate an assembler block. The end of the
block is indicated by an end statement:
procedure SomeProcedure; register;
asm
{Code goes here}
end;

It is possible to nest asm blocks inside a Pascal function or procedure, but that
approach is not recommended. You should isolate your assembler code inside
separate function or procedure blocks. First of all, inserting assembler inside a
regular Pascal function will interfere with the compiler's optimisation and
variable management activities. As a result, the generated code will be far from
optimal. Variables are likely to be pushed out of their register, requiring saving
on the stack and reloading afterwards. Also, nesting inside a Pascal block
forces the compiler to adapt its generated code to your assembler code. Again,
this interferes with the optimisation logic and the result will be quite inefficient.
So, the rule is to put assembler code in its own separate function/procedure
block. There is also a design aspect: the readability and maintainability of your
code will benefit greatly when all assembler is clearly isolated in dedicated, wellcommented blocks.

1.2. Labels
Labels are tags that mark locations in your code. The most common reason for
having labels is to have a point of reference for branching. There are two kinds

Guido Gybels: Using Assembler in Delphi

of labels you can use in your assembler code: Pascal-style labels and local
assembly labels. The former type requires you to declare them in a label section
first. Once declared, you can use the label in your code. The label must be
followed by a colon:
label
MyLabel;
asm
...
mov ecx, {Counter}
MyLabel:
... {Loop statements}
dec ecx
jnz MyLabel
...
end;

The above example illustrates how to declare a label MyLabel, marking a

position in your program (MyLabel:) and moving to the label's location in a
jump instruction (jnz MyLabel).
The same can be achieved in a slightly simpler way, by using local labels in
your assembler code. Local labels do not require a declaration, rather you
simply insert the label as a separate statement. Local labels must start with the
@ sign, and are again followed by a colon. Because @ can't be part of an
Pascal identifier, you can use local labels only within an asm...end block.
Sometimes, you will see labels prefixed by a double @ sign in code in this
document. This is a convention I use a lot and it draws attention to the labels
immediately, but it is not required (some assemblers use the @@ to identify
special purpose labels, like @@: for an anonymous label). Below is an example
of the same logic as above, but using a local label:
asm
...
mov ecx, {Counter}
@MyLabel:
... {Loop statements}
dec ecx
jnz MyLabel
...
end;

Neither kind of label is intrinsically better than the other. There is no advantage
in code size or speed of course, since labels are only reference points for the
compiler to calculate offsets and jumps. The difference between Pascal-style
and local labels in assembler blocks is a relic from the past and is fading away.
As a consequence, even Pascal-style labels are "local" in the sense that it is not
possible to jump to a label outside the current function or procedure block. That
is just as well, since that would be a perfect scenario for disaster.

Guido Gybels: Using Assembler in Delphi

1.3. Loops
Often, the assembler code will be designed to achieve the highest speed
possible. And quite often also, processing large amounts of data inside loops
will be the task. When loops are involved, you should implement the loop itself
in assembler too. That is not difficult and otherwise you will be wasting a lot of
execution time because of calling overheads. So, instead of doing:
function DoThisFast(...): ...; register;
asm
...
{Here comes your assembler code}
...
end;
procedure SomeRoutine;
var
I: Integer;
begin
I:=0;
...
while I<{NumberOfTimes} do begin
DoThisFast(...);
inc(I);
end;
...
end;

You should implement the loop inside the assembler routine:

function DoThisFast(...): ...; register;
asm
...
mov ecx,{NumberOfTimes}
@@loop:
...
{Here comes your main assembler code}
...
dec ecx
jnz @@loop
...
end;
procedure SomeRoutine;
begin
...
DoThisFast(...);
...
end;

Note that in the example above, the loop counter counts downwards. That is
because in this way, you can simply check the zero flag after decrementing to
see if the end of the loop has been reached. By contrast, if you simply start off

Guido Gybels: Using Assembler in Delphi

with ecx=0 and then count upwards, you will need an additional compare
instruction to check whether or not to continue the loop:
mov ecx,0
@@loop:
...
inc ecx
cmp ecx,{NumberOfTimes}
jne @@loop

Alternatively, you can subtract the NumberOfTimes from 0 and then increase
the loop index until zero is reached. This approach is especially useful if you
use the loop index register simultaneously as an index to some table or array in
memory, since cache performance is better when accessing data in forward
direction:
xor ecx,ecx
sub ecx,{NumberOfTimes}
@@loop:
...
inc ecx
jnz @@loop

Remember however that in this case, your base register or address should
point to the end of the array or table, rather than to the beginning, and you will
be iterating through the elements in reverse order.

1.4. Entry and exit code

Another important aspect to be aware of, is that the compiler will automatically
generate entry and exit code for your assembler block. This will only happen if
there is a need for a stack frame, i.e. if parameters are passed to the routine via
the stack, or if local data is stored on the stack. You'll find that very often this is
indeed the case, and consequently entry and exit code will be generated. The
compiler produces the following entry code:
push ebp
mov ebp,esp
sub esp, {Size of stack space for local variables}

This code preserves ebp, and then copies the stack pointer into the ebp
register. Subsequently ebp can be used as the base register to access
information on the stack frame. The sub esp line reserves space on the stack
for local variables as required. The exit code pattern is as follows:
mov esp,ebp
pop ebp
ret {Size of stack space reserved for parameters}

Guido Gybels: Using Assembler in Delphi

10

This exit sequence will clean up the space allocated for local parameters by
copying ebp (pointing at the beginning of the stack frame) back to the stack
pointer. This deallocates the space used for local variables. Next, ebp is
restored to the value it had upon entry of the routine. Finally, control is returned
to the caller, adjusting the stack again for any space allocated for parameters
passed to the routine. This parameter cleanup in the ret instruction is required
for all calling conventions except cdecl. In all cases except cdecl, the called
function is responsible for cleaning up the stack space allocated for parameters,
and thus the ret instruction will include the necessary adjustment. In case of
cdecl, however, the caller performs the cleanup.
If your function or procedure has neither local variables nor parameters passed
to it via the stack, then no entry and exit code will be produced, except for the
ret instruction that is always generated.

1.5. Register preservation

When using registers inside your function or procedure, please note that only
the registers eax, ecx and edx can be freely modified. All other registers must
be preserved. That means that if you use any of the other registers inside your
routine, you must save them and restore them before returning from your
routine.
You must not change the contents of any of the segment selectors: ds, es and
ss all point to the same segment; cs has its own value; fs refers to the Thread
Information Block (TIB) and gs is reserved. The esp register points to the top of
the stack, of course, and ebp is made upon entry to point to the current stack
frame as a result of the default entry code generated by the compiler. Since
each pop and push operation will change the content of the esp register, it is
usually not a good idea to access the stack frame directly through esp. Rather,
you should reserve ebp for that purpose. Table 1 summarises register usage.
Apart from the register context, you can assume that the direction flag is cleared
upon entry and if you change it (which I don't recommend), you should restore
its cleared state prior to returning (by using the cld instruction). Finally, you
should be careful about changing the FPU control word. Although this allows
you to change the precision and rounding mode for floating point arithmetic, and
permits you to mask certain exceptions, you will be drastically influencing the
way calculations in your entire application are performed. Whenever you decide
it is necessary to change the FPU control word, make sure you restore it as
soon as possible. When you are using Comp or Currency types, make sure
you don't reduce floating point precision.

Guido Gybels: Using Assembler in Delphi

11

Chapter 2: Passing Parameters

Most routines programmers write are passed one or more parameters as input.
This chapter gives an overview of how such parameters can be passed to your
assembler routine. Many routines will be functions that also have to return a
result to the caller, but the process of returning a result is discussed in Chapter
4.

2.1. Calling Conventions

One of the most common operations in the processing of code is calling a
subroutine that carries out a given task. In order to do that, the main code
needs to hand over control to the subroutine, allow the subroutine to execute,
and then return to where it left the main execution path. This process requires
agreement between the caller and the callee about how to pass information to
the subroutine, how to return results where applicable and who is responsible
for the memory allocation and cleanup. Various conventions exist to deal with
invoking subroutines, commonly known as calling conventions. Delphi supports
a wide set of them: register, pascal, cdecl, stdcall, and safecall.
Obviously, caller and callee need to use the same calling convention in order for
this to work properly.
As already mentioned, calling conventions define a number of different aspects
of subroutine invokation:
Where parameters are located: in registers or on the stack
In which order parameters are passed: right-to-left or left-to-right
Who is responsible for cleaning up the parameters afterwards, the caller or
the callee
Table 2 gives an overview of each of the Delphi supported calling conventions.

2.2. Passing parameters in registers

There is only one calling convention in Delphi, register, that uses the CPU
registers to pass parameters to a subroutine. There are three registers available
for this purpose: <eax, edx and ecx and they are used up in this order
(because ecx is used last, that register remains available the longest, which is
quite convenient since you will often want to use it as a loop counter variable). If
you are passing more parameters than there are registers available, the
remaining ones will be passed on the stack as described later. When you use
the register calling convention for methods, however, eax contains a pointer
to Self and thus only two registers are available for parameter passing. Not all
data types can be passed in a register. Table 3 gives a clear overview of which
Guido Gybels: Using Assembler in Delphi

12

types can be passed in a register. Please note that when passing parameters
by reference, you are in fact passing a pointer to the variable in question. Since
pointer types qualify for passing in a register, variables passed by reference
always qualify for passing in a register, with the exception of method pointers.
If the number of parameters passed is lower than or equal to the number of
registers available (three for standalone routines, two for methods), than there
is no need to set up a stack frame for parameter passing. This can save
overhead when calling the routine. Be careful however, because parameter
passing is not the only reason for setting up stack frames: if you declare local
variables, a stack frame is also required and thus extra overhead to manage the
stack frame is still incurred.
In addition, for many structured types, the data itself actually resides on the
stack or on the heap and the variable is a pointer to the actual data. Such a
pointer occupies 32-bits and therefore will fit into a register. This means that
most parameter types will qualify for passing through registers, although
method pointers (consisting of two 32-bit pointers, one to the object instance
and one to the method entry point) will always be passed on the stack.
This article is based on 32-bit modes, so registers are 32 bits wide. When
passing information that doesn't occupy the whole register (byte- and wordsized values for example), the normal rules apply: bytes go in the lowest 8 bits
(for example al) and words in the lower word of the register (for example ax).
Pointers are always 32-bit values and thus occupy the whole register (for
example eax). In case of byte- or word-sized variables, the content of the rest
of the register is unknown and you should not make any assumptions about its
state. For instance, when passing a byte to a function in al, the remaining 24
bits of eax are unknown, so you cannot assume them to be zeroed out. You
can use an and operation to make sure the remaining bits of the register are
reset:
and eax,$FF {Unsigned byte value in AL, clears 24 highest bits}

or
and eax,$FFFF {unsigned word value in AX, clears 16 highest bits}

When passing signed values (shortint and smallint), you might want to
expand them to a 32-bit value for easier computation, but in doing so you need
to retain the sign. To expand a signed byte to a signed double word, you need
two instructions:
cbw {extends al to ax}
cwde {extends ax to eax}

The importance of not relying on the remainder bits having a specific value can
be easily demonstrated. Write the following test routine:
Guido Gybels: Using Assembler in Delphi

13

function Test(Value: ShortInt): LongInt; register;

asm
end;

Next, drop a button and a label on a form and put the following code in the
button's OnClick event:
var
I: ShortInt;
begin
I:=-7;
Label1.Caption:=IntToStr(Test(I));
end;

Run the project and click the button. The Test routine receives a ShortInt
through al. It returns an integer in the eax register (returning results is
discussed in Chapter 4), which should be unchanged since the subroutine
returns immediately. You can easily observe that eax has undefined content
upon return. Now change the test function as follows and run the project again:
function Test(Value: ShortInt): LongInt; register;
asm
cbw
cwde
end;

The Test routine now returns the correct value.

In summary, when using the register calling convention, up to three
parameters can be passed in the eax, edx and ecx registers. So, the following
declaration:
procedure DoSomething(First: Integer; Second: ShortInt;
Third: Pointer); register;
asm
...
end;

will put First in eax, Second in dl and Third in ecx. Next, here is an
example of a method declaration:
procedure
register;
asm
...
end;

TSomeClass.DoSomething(First,

Second:

Integer);

In this case, eax will contain Self, edx contains First, while Second is
stored in ecx.

Guido Gybels: Using Assembler in Delphi

14

Please note that since register will result in parameters being passed in
registers, you will lose that parameter information as soon as you override the
register's contents. Take the following code:
procedure DoSomething(AValue: Integer); register;
asm
{eax will contain AValue}
...
mov eax, [edx+ecx*4] {eax gets overwritten here}
...
end;

After eax gets overwritten, you no longer have access to the AValue
parameter. If you need to preserve that parameter, make sure to save the
contents of eax on the stack or in local storage for use afterwards. And don't fall
into the common trap to do the following later on in your code:
mov eax, AValue

because the compiler will, for the above line, simply generate the following
code:
mov eax, eax

as the compiler only knows, from the chosen calling convention, that AValue
was passed in eax to the subroutine.

2.3. Using the stack for parameter passing

All calling conventions might use the stack for passing some parameters. While
the register convention tries to use CPU registers first, not all variable types
qualify for passing in a register and sometimes you will need to pass more
parameters than there are registers available. All other calling conventions will
pass all their parameters on the stack to the callee.
As explained in the previous chapter, the compiler will generate entry and exit
code to manage the stack frame. As a result, ebp is initialised as base pointer
to the stack frame, allowing easy access to parameters and other information
on the stack (including local variables as explained in Chapter 3). When you
refer to parameters that reside on the stack, the compiler will generate the
appropriate offset from ebp. Have a look at the following declaration:
function Test(First, Second, Third: Integer): Integer; pascal;

The calling convention is pascal, which means that prior to the call to the
subroutine, the caller pushes three parameters on the stack in the order that
they are declared (remember that the stack grows downwards, which means
the first parameter is located at the highest address):
Guido Gybels: Using Assembler in Delphi

15

First

esp

Second
Third

Next, the call instruction will push the return address onto the stack and then
hands over execution to the subroutine, so immediately after entry, the stack
looks as follows:

First
Second

esp

Third
Return Address

The compiler generated entry code (see Chapter 1) saves the current value of
ebp and subsequently copies the value of esp to ebp so that the latter can from
now on be used to access the parameter data on the stack frame:

First
Second

esp

Third
Return Address
Saved ebp

ebp

From this point on, we can access the parameters on the stack frame as offsets
from ebp. Because the return address sits on the stack between the current
top-of-stack and the actual parameters, we access the parameters as follows:
First = ebp + $10 (ebp + 16)
Second = ebp + $0C (ebp + 12)
Third = ebp + $08 (ebp + 8)

However, you can simply refer to these parameters by name. the compiler will
replace each parameter with the correct offset from ebp. So, in the example
above, writing the following:
mov eax, First
will be translated by the compiler into:

Guido Gybels: Using Assembler in Delphi

16

mov eax,[ebp+0x10]

This will save you the headache from calculating the offsets yourself and it is
also much more readable, so you should use the names of the parameters that
are passed on the stack in your code wherever possible (practically always)
instead of hard coding the offsets. Be careful however: if you use the register
calling convention, the first set of parameters will be passed in registers. For
those parameters that are passed in registers, you should use correct register
to refer to the variable, to prevent ambiguities in your code. Take the following
example:
procedure DoSomething(AValue: Integer); register;

Since this declaration uses the register calling convention the AValue
parameter will be passed into the eax register. It is probably wise to explicitely
write eax in your code to refer to this parameter. It will help you to spot the
following potential bug:
mov eax, AValue

which on the basis of the declaration above would result in the following code to
be generated:
mov eax, eax

In summary: for parameters passed in a register, you should use the register to
refer to it. For parameters passed via the stack, use the variable name to refer
to it (and don't use the ebp register, so it remains available for access that
information).
Stack space is always allocated in 32-bit chunks, and therefore the data passed
on the stack will always occupy a dword multiple. Even if you pass a byte to the
procedure, 4 bytes will be allocated on the stack with the three most significant
bytes having undefined content. You should never assume that this undefined
portion is zeroed out or has any other specific value.

2.4. Passing by value versus passing by reference

Earlier on, I mentioned the difference between passing by value and passing by
reference. When passing by reference (using the var directive) or, in some
instances, when using const, no data is copied and handed over to the routine
in question, but rather a pointer to the original data is passed on. This difference
is of course quite important, not in the least because of the non-localised effects
changes to that data might cause. Take for instance the following function
declaration:
function MyFunction(I: Integer): Integer; register;

Guido Gybels: Using Assembler in Delphi

17

As the register calling convention is used, the value of the I parameter will
be passed in the eax register (see table 3). So, given I=254, eax will upon
entry contain the value $000000FE, passing I by value. However, if we change
the declaration as follows:
function MyFunction(var I: Integer): Integer; register;

the eax register no longer contains the value of I, but rather a pointer to the
memory location where I is stored, for example $0066F8BC. Passing
parameters by reference using var or const is done by means of a 32-bit
pointer.
When you use const, indicating that you a variable is used for read-only
access only, the compiler uses either method. The wording in Delphi's online
help can be misleading and some people assume that const always results in
passing a 32-bit pointer to the actual value, but that is not correct. You can use
table 3 for guidance.
By using const, the programmer informs the compiler that the data is only
going to be read. Please note however that within an asm..end block, the
compiler will not prevent you from writing code that violates this read-only
characterisation of const parameters in cases where these are pointers to
structured data like AnsiStrings or records. Be careful to honour the readonly character of the information passed using const in your assembler code,
otherwise you could introduce nasty bugs. And of course, it would be extremely
poor design to label information read-only, yet then proceed to change it. This is
especially important when you are using reference counted types like
AnsiString that use copy-on-write semantics.
All of the above means that you will have to carefully take into account the
differences between passing by value and passing by reference. For example,
imagine a function that calculates the sum of an integer with 12. In case of
passing the integer parameter by value, the code should look as follows:
function MyFunction(I: Integer): Integer; register;
asm
add eax, 12
end;

I will discuss returning results in Chapter 4. For now it is sufficient to know that
the result in this case will be returned to the caller via the eax register. As you
can see, the value of I is taken directly from the eax register. But if we change
the function to pass the information by reference, we would get something like
this:
function MyFunction(var I: Integer): Integer; register;
asm

Guido Gybels: Using Assembler in Delphi

18

mov eax, [eax] {Load the value of I through the pointer}

add eax, 12
end;

Because eax does not contain the value of I, but rather a pointer to the
memory location where I is stored, we retrieve the value through the received
pointer.

Guido Gybels: Using Assembler in Delphi

19

Chapter 3: Local Variables

Just as in any Pascal routine, you can use local variables inside your assembler
code. Local variables are declared in the same way as for Pascal routines, by a
var section. In this chapter, I will give a detailed overview of how local variables
are implemented and used in asm blocks.

3.1. Local variables and the stack frame

Storage space for local variables is allocated on the stack, by means of the
compiler generated entry code (and freed again in exit code). Please note that
for some complex types like AnsiStrings, the space allocated is for a pointer
to the actual data (strings reside on the heap and the string variable is a mere
pointer to the data) and further action will be needed to allocate and assign the
actual data. continuing the narrative from the previous chapter, have a look at
the following code:
procedure DoSomething(First, Second, Third: Integer); pascal;
var
SomeTemp: Integer
asm
...
end;

From that previous chapter, we already know that using the pascal calling
convention will result in parameters being pushed onto the stack prior to
invoking the procedure. The call instruction will push the return address onto the
stack. Next, entry code will cause the value of ebp to also be pushed onto the
stack. Then, ebp is set up as base pointer for accessing the data on the stack
frame. At this point, the stack frame looks therefore as follows:

First
Second

esp

Third
Return Address
Saved ebp

ebp

Because we have also declared a local variable, SomeTemp, the compiler will
add code (for instance push ecx) to reserve space on the stack for said
variable:

Guido Gybels: Using Assembler in Delphi

20

First
Second
Third
Return Address
Saved ebp
esp

SomeTemp

+16
+12
+8
+4

ebp
-4

As stated before, ebp contains a base pointer for accessing data on the stack
frame. Since the stack grows downwards, higher addresses contain
parameters, while lower addresses contain local variables. In our particular
example, the stack frame has the following slots allocated:
Parameters:
First = ebp + $10 (ebp + 16)
Second = ebp + $0C (ebp + 12)
Third = ebp + $08 (ebp + 8)
Local Variables:
SomeTemp = ebp - $04 (ebp - 4)

A next local variable will be allocated at ebp -8 and so on. Just as with
parameters on the stack, you can (and should) use the variable name to refer to
the actual location on the stack:
mov eax, SomeTemp

which will be translated by the compiler into:

mov eax, [ebp-4]

Please note that the content of these variables is generally not initialised, and
you should treat it as being undefined. It is your task to initialise them when and
if required.
Because using local variables cause overhead for creating and managing the
stack frame, it is worth analysing your algorithm carefully to determine whether
or not you need local storage. Clever use of available registers and smart code
design can often avoid the need for local variables altogether. Apart from
avoiding overhead for allocating and managing local variables, moving data
between registers is significantly faster than accessing data in main memory
(but beware of stalls and other performance hits, for instance by reading a
register immediately after writing it). When you are writing Object Pascal code,
the Delphi compiler will perform optimisations by trying to use registers
wherever feasible. Loop counter variables are a particular case in point and
you, too, should favour registers for such usage. Of course, inside an

Guido Gybels: Using Assembler in Delphi

21

asm..end block, you are on your own and the compiler will not perform such
optimisations for you. Well structured code will therefore aim to use registers as
much as possible, especially for data that is used most often.

3.2. Simple types as local variables

Quite a number of data types will require simply allocation of space on the stack
frame when you declare a local variable. ShortInt, SmallInt, LongInt,
Byte, Word, DWord, Boolean, ByteBool, WordBool, LongBool, Char,
Ansichar and Widechar all belong to this category.
While not all of these types are 32-bits wide, reservation of stack space will
always happen in chunks of 32-bits at a time. That means that if you use
smaller types, like for instance byte or word, the unused part of the allocated
space is undefined. For instance, if you declare a local variable as follows:
var
AValue: ShortInt;

While AValue only requires one byte, a full dword is allocated on the stack
frame. This behaviour ensures that data on the stack is always aligned on a
dword boundary, which improves performance and makes the logic to calculate
variable locations easier (and allows for easy use of scaling in indirect
addressing). You should however not use the remainder part of the allocated
space (the padding) since this ultimately an implementation issue. Future
compiler versions might behave differently. If you need additional storage
space, simply use the appropriate, larger type.
Please note that this rule for dword allocation does not apply to local variables
of type record, even though they are also stored on the stack frame. Their
member fields' alignment depends on the state of the alignment switch ({$A}
directive) and the use of the packed modifier. This is discussed in more detail
in the next paragraph.
This alignment behaviour is another good reason to refer to local variables
using their variable name, rather than manually calculating the offset yourself.
The compiler will calculate the correct offset for you. In the example above,
AValue occupies only one byte. Hence, only the lowest byte of the allocated
dword is used. So, this instruction:
mov al, AValue

will result in the compiler generating the following code:

mov al, [epb-$01]

Guido Gybels: Using Assembler in Delphi

22

In Pascal routines, outside asm...end blocks, the compiler might optimise a

local variable into a register, in which case no space for it will be allocated on
the stack frame. While such optimisation does not happen inside your own
asm...end blocks, you should be aware of this behaviour when observing
compiler generated code through the CPU window. Similarly, sometimes the
compiler will generate code that uses esp directly, rather than an offset from
ebp, thus saving the need for initialisation of ebp. As argued before, it's not
generally advisable to use esp directly in your own assembler code, as it makes
it extremely hard to read and maintain the code and it is prone to introducing
subtle coding errors that will be hard to find and debug. While study compiler
generated code can be very instructive, remember that you are not a machine,
but a human programmer. Machines are good in making sure they calculate the
right offsets and the like - humans mostly are not. It's likely that in most cases
stack frame overhead will not constitute a bottleneck, especially if you design
your code carefully. If you identify that stack frame overheads cause
performance issues in your application, you should reconsider your algorithm. It
might sound very obvious, but too many programmers end up at some point
optimising code that has no effect on the overall application's performance.

3.3. Records as local variables

Just as in the case of simple types, local record variables are stored on the
stack frame. In that respect, they are not fundamentally different in their usage
from simple types (see previous paragraph). However, the compiler's record
alignment mechanism is more complex. This can seriously complicate things for
the programmer if he/she is coding offsets directly.
There are two key factors that define the compiler's record alignment behaviour:
the alignment directive ({$A} or {$ALIGN}) and the packed modifier.
Furthermore, the actual alignment of the record member fields is dependent on
the field type. For example, let's consider the following record declaration:
TMyRecord = record
FirstValue: DWord;
SecondValue: Byte;
ThirdValue: DWord;
FourthValue: Byte;
end;

The alignment boundary for each member field of the record depends on its
type and its size. In the example above, Firstvalue and ThirdValue are of
type DWord, which is a 32-bit type. With alignment on, they will be aligned to
dword boundaries. Since in between those two members, there is a byte-sized
field, SecondValue, the compiler will add three padding bytes, thus ensuring

Guido Gybels: Using Assembler in Delphi

23

that ThirdValue is properly aligned. The following picture shows the memory
allocation for this record in the aligned state:

By adding the packed modifier to the record declaration, the record's member
fields are no longer aligned. You can see the result in the following illustration,
as the padding bytes are no longer present:

Similarly, when alignment is turned off by using the {$A-} directive, even without
the packed modifier there will be no padding between record member fields.
Fortunately, just as for simple types, you can refer to record member fields by
their names, and the compiler will calculate the correct offsets for you. However,
always make sure you use operands of the proper size, i.e. specify the operand
size explicitly. In that way, your code will continue to work correctly even when
alignment is changed or the packed modifier is introduced at a later stage:
mov eax, DWORD PTR [ARecord.FirstValue]
mov al, BYTE PTR [ARecord.Byte]

3.4. Heap allocated types as local variables

Dynamic variables, long strings, wide strings, dynamic arrays, variants, and
interfaces in Delphi are all variable types that are stored in heap memory. In
order to use them, you use a reference variable, i.e. a pointer to the actual
variable data. In assembler, you will be responsible for the allocation and
management of the memory and its contents.

Guido Gybels: Using Assembler in Delphi

24

In other words, if you use heap allocated types as local variables, memory will
be allocated for the reference (the pointer) to that variable on the stack frame,
but you are responsible for the actual allocation and deallocation of the memory
and for initialising the contents. In Pascal, most of these types are largely
automatically managed, so allocation and deallocation happens behind the
scenes. In assembler blocks, that is obviously not the case.
You can call GetMem to allocate memory and return a pointer to the newly
allocated memory. You need to pass the amount of memory needed in eax and
upon return from GetMem the eax register will contain the pointer, which you
can then store in the appropriate slot on the stack frame.

Guido Gybels: Using Assembler in Delphi

25

Chapter 4: Returning Results

In many cases, you will return a result from your assembler routines to the
caller. In previous chapters, we have discussed how to pass parameters to a
function and how to use local variables in your assembler code. This chapter
will cover returning results to the caller.

4.1. Returning integers as immediate values

Delphi provides several different integer types. We use the term integer in this
chapter mostly in its general meaning of "whole number", using a standard font
in lowercase. Delphi also has a generic type called Integer. When I refer to
the Delphi-specific generic data type, I will spell it Integer in a monospaced
font and with an uppercase "I".
There are 8-bit, 16-bit, 32-bit and 64-bit integer types in Delphi. Some of these
types are signed, whereas others are unsigned. Unsigned integers always
represent positive whole numbers. Signed types have a sign bit, which if set
indicates a negative value and when cleared indicates a positive value.
Negative values are represented as the two's complement of the absolute
value. The Delphi integer types are Shortint (8-bit, signed), Smallint (16bit, signed), Longint (32-bit, signed), Int64 (64-bit, signed), Byte (8-bit,
unsigned), Word (16-bit, unsigned) and Longword (32-bit, unsigned). In
addition, Delphi also has the generic types Integer and Cardinal, which
correspond on a 32-bit platform to respectively a signed 32-bit value and an
unsigned 32-bit value. So, Integer is on a 32-bit platform the same as
Longint, whereas Cardinal is the same on a 32-bit platform as Longword.
There are several other data types in Delphi that map to one of the above
integer types. Many of these additional types are provided to offer a type of the
same name as in C-declarations, often for conformity with the Windows API.
For example, DWORD and UINT are the same as Longword, whereas SHORT is
the same as Smallint, etc.
In general, returning integers is straightforward: you store the value in the eax
register before returning to the caller. If the return type is smaller than eax, only
the al (8 bits) or ax (16 bits) portion of the register is valid, the contents of the
remainder of the register are ignored. See Table 4 for a detailed overview. The
only exception to this rule are 64-bit integers. On a 32-bit platform, these are
returned in edx:eax, with edx containing the most significant part.
Please note that the Comp type, which also represents a 64-bit integer, does not
behave like other integers. It is a type that uses the floating point unit of the
processor and as such follows the conventions for real types. You should have
Guido Gybels: Using Assembler in Delphi

26

little use for the Comp type, which is maintained for backward compatibility. It is
recommended to use Int64 instead. On several processors, Int64 will also
yield better performance for integer arithmetic, as it uses the CPU registers,
rather than the FPU.
The following code demonstrates how to return an integer from assembly code.
It returns the number of set bits in the AValue parameter as an unsigned 8-bit
value (we don't need the larger range of 16 or 32 bit integers, since the returned
value will fall in the range 0-32).
function CountBits(const AValue: Longword): Byte;
asm
mov ecx, eax
xor al, al
test ecx, ecx
jz
@@ending
@@counting:
shr ecx, 1
adc al, 0
test ecx, ecx
jnz @@counting
@@ending:
end;

4.2. Returning booleans as immediate values

Returning a boolean value is quite simple too. Again, the result goes in the eax
register, or a subset of that register - similar to returning integers. Delphi has
several boolean types. Boolean is the "proper" boolean type: it only has two
possible values, true and false. Within Delphi, you should always use this type.
The other boolean types, ByteBool, WordBool and LongBool are provided
for compatibility with other languages and for calling the Windows API.
The Boolean type is really an enumerated type. It occupies a byte of memory.
As this is not an ordinal type, you should only use the predefined constants
True and False in assignments. Do not assign ordinal values to a Boolean
as this relies on an implementation feature, namely the values that the compiler
uses to represent the True and False values of a Boolean. While it is unlikely
that this will change in future versions of Delphi, relying on it being future proof
is still bad practice, architecturally ugly and also less clear than using the
predefined boolean constants True and False. As a Boolean requires 8 bits,
you return it in al:

Guido Gybels: Using Assembler in Delphi

27

function DoSomething(...): Boolean;

asm
...
mov al, True
...
end;

In contrast, ByteBool, WordBool and LongBool, provided for compatibility

with other languages and calling the Windows API, are essentially ordinal types.
They occupy respectively 8 bits, 16 bits and 32 bits and are thus returned
correspondingly in al, ax or eax. They are considered true if their ordinality is
non-zero and false otherwise. The compiler will perform any necessary
conversions between these types and Boolean where required.
Table 4 provides the full overview.

4.3. Returning real numbers

Real numbers in Delphi are implemented as floating point types. Unfortunately,
all too many programmers nowadays don't properly understand floating point
representation. You can read my brief introduction on the topic.
The basic mechanism for returning real values from your assembler code is to
put the result in the ST(0) register of the FPU, which corresponds to the top of
the FPU stack.
Even though Delphi supports several floating point formats, such as single (78 significant digits, occupies 4 bytes) and double (15-16 significant digits,
occupies 8 bytes of memory), internally the FPU always stores and handles
floating points as 80-bit values. Delphi's Extended type (19-20 significant
digits, uses 10 bytes of memory) maps onto this format. Note that all these real
types are returned to the caller as a value in ST(0). It is only when the result is
subsequently stored in memory or passed along to another part of the program
that it effectively is transformed in its 4, 8 or 10 byte encoding. My article on
floating point values on my website discusses these formats in some more
detail.
There are however other considerations to take into account when working with
real numbers. The Intel FPU has a control register that controls precision and
rounding and also has exception masks to steer FP exception handling. The
different precision and rounding methods allow a programmer fine control over
the behaviour of the FPU and can be important for code compatibility with other
systems or legal and other standards. It is therefore important to understand
that it might not be sufficient to declare a result of a certain data type (say,
single or double) only. Instead, you might need to explicitly configure the
control register.

Guido Gybels: Using Assembler in Delphi

28

The precision control bits in the FPU control register are highly relevant in this
respect. Under normal circumstances, this two bit binary value ought to be set
to 11 at all times, indicating 64-bit mantissa precision. If the precision control
bits are set to a lower precision, the FPU will reduce precision during
computation and hence your result will be less precise than you would have
expected. The theory of floating point arithmetic and the details of the Intel FPU
are out of scope for this article, but you should familiarise yourself intimately
with the topics before attempting to write elaborate floating point code or
whenever you need to produce results in line with specific guidelines or
standards. Delphi has several supporting functions and variables, such as
Get8087CW, Set8087CW, SetPrecisionMode, etc. See online help for more
information. Note that many libraries and even calls to OS functions can change
the value of the FPU control word. If you change the control word inside your
own code, it is good practice to make sure you restore it to its previous state
when you are done. This ought to be done outside your time critical FP code,
since setting the control word causes, on many processors, considerable stall if
the control word is read immediately afterwards, which is the case for most FP
instructions.
In addition to single, double and extended, the Real48, Comp and
Currency types are also returned in ST(0). Even though Comp represents a
64-bit integer, it is a type that uses the FPU, rather than the CPU registers and
as such is manipulated using FPU instructions. Currency is a fixed-point type
mainly designed for monetary calculations, but as with Comp it is in fact a FPU
based type. Note that Currency is scaled by 104. Hence, a Currency value of
5.4321 is stored in ST(0) as 54321.
Anyone wishing to use FP math for monetary applications ought to make sure
they fully understand the nature of floating point arithmetic. My article on floating
point values provides a brief introduction to the topic and contains links to
further reading material. Using scaled integers might be a better approach for
such applications. Also, Intel CPUs support BCD encoding and arithmetic,
which is useful for these purposes. Unfortunately, the Delphi language has no
support for BCD types, so you will need to encode and decode BCD data
yourself.
Unless needed for compatibility with other applications or environments, you
should avoid the non-native Real48 type altogether. This type is not supported
in hardware, so all manipulation has to be done in code, which makes it very
slow. Convert a Real48 into a native float immediately after receiving it, using
the System unit's _Real2Ext function. When invoking _Real2Ext, eax
contains a pointer to the Real48 value. Upon return, ST(0) is loaded with the
value. You can then use the FPU to perform the required calculations. If you
need to hand the result back as a Real48 type, call _Ext2Real, also in the

Guido Gybels: Using Assembler in Delphi

29

System unit, which will convert the value in ST(0) back into a Real48 value.
eax should contain a pointer to a 6-byte wide memory location where the
converted value will be stored. Note that in Delphi versions before Delphi 4, this
non-native 6-byte type was called Real instead of Real48. From Delphi 4
onwards, Real acts as a generic type for real numbers, at present implemented
as a Double.
Table 4 summarises the rules for returning results, including real types.
To conclude this section on returning real numbers, below is a full working
example. The function CalcRelativeMass below demonstrates floating point
arithmetic in assembler within a Delphi environment. The function takes two
parameters, mass and velocity of a body, and calculates the relative mass as
per the theory of relativity.
function CalcRelativeMass(m,v: Double): Double; register;
const
LightVelocity: Integer = 299792500;
asm
{Calculate the relative mass according to the following
formula: Result = m / Sqrt(1-v/c), where c = the
velocity of Light, m the mass and v the velocity of
an object}
fild LightVelocity
fild LightVelocity
fmulp {Calculate c}
fld v
fld v
fmulp {Calculate v}
fxch
fdivp {v/c}
fld1
fxch
fsubp {ST(0)=1-(v/c)}
fsqrt {Root of ST(0)}
fld m
fxch
fdivp {divide mass by root result}
end;

4.4. Returning characters

Delphi currently offers two fundamental character types: AnsiChar is an 8-bit
character, whereas WideChar is a 16-bit Unicode character. There is also a
generic type, Char, which is mapped to AnsiChar. As with integer types, it is
good practice to use the fundamental types in your own assembly code.
As AnsiChar is an 8-bit type, you return it in al. WideChar is a 16-bit Unicode
character and is returned in ax.

Guido Gybels: Using Assembler in Delphi

30

The Delphi online help makes a bit of a mess of things by stating on the one
hand that WideChar is a fundamental type, and on the other hand talking about
its "current implementations [sic]", thereby alluding that future versions might
change its 16-bit nature, as if it were some generic type. In practice, there is
little choice but to consider WideChar as a fundamental, 16-bit Unicode type.
See also Table 4, which provides an overview of the rules for returning results.

4.5. Returning a long string

A very common type used in Delphi applications is the long string:
AnsiString. This type is essentially an array of AnsiChar characters, but
with some additional complexity. The AnsiString type has two important
properties from the point of view of the assembler programmer. Firstly, it is a
reference counted type. The reference counting mechanism has several
advantages. If the same string is used in different places, the same instance
can be shared, rather than having to allocate memory for each identical string.
By using reference counting with a copy-on-write algorithm, a copy of a string in
memory is only made when it is absolutely necessary, which reduces overhead
and enhances performance. Reference counting also allows for automated
memory management. When the reference count reaches zero, indicating no
one is using the string anymore, it is automatically cleaned up. However, within
our assembler code, we don't have the compiler's support for this, so we often
must handle memory allocation and reference counting manually.
The second important feature of long strings is that they are stored in heap
memory. The string variable is a pointer to the string on the heap. As with
reference counting, and in contrast to Pascal code, we will need to explicitly
deal with memory management issues for our long strings.
As we create and manipulate long strings in assembler code, we must ensure
that those strings will continue to function properly for the rest of our Delphi
code, outside the asm block. This requires us to allocate memory for the strings
properly, through Delphi functions, and that we must consider the reference
counting carefully when we create, process and return long strings.
In contrast to ordinal types, long strings are not returned in a register, rather the
function behaves as if an additional var parameter was declared after all the
other parameters. In other words, an additional parameter is passed to your
function by reference. chapter 2 describes parameter passing in detail, including
the the issues related to passing variables by reference and the differences
associated with the various calling conventions.
The approach of using an additional parameter might seem like a needless
complication, but in fact it is quite clever: by passing this additional var

Guido Gybels: Using Assembler in Delphi

31

parameter, responsibility for decreasing the reference count is handed over to

the caller. This makes it quite easy to return long strings from assembler code,
while at the same time making sure that the reference count is suitably adjusted
after the caller is done with it.
With regard to the process of allocating memory for long strings in assembler
code, you should study the various routines provided in System.pas for this
purpose. For example, you can call LStrSetLength to set the length of the
Result string, before filling it with content. A major drawback of this approach
is that it can easily be broken as Delphi itself evolves. System.pas gets
special treatment at compile time and these internal routines might be changed
at some point as Delphi evolves. One way around this is to create the string
elsewhere in Pascal, then just hand an already allocated long string to your own
assembler code. Writing code that ports well is an important consideration.
Clearly the very choice for assembler means that code will be much more
closely tied to a specific platform and compiler, but within those limitations,
programmers should still endeavour to write code that is as readable and future
proof as possible.
The following example illustrates how it can be done. The procedure
FillWithPlusMinus fills the long string passed to it with a pattern. In this
case, the string itself is already allocated before calling the procedure, thus
avoiding the need to call System.pas routines.
procedure FillWithPlusMinus(var AString: Ansistring); register;
asm
push esi
mov esi, [eax] {esi now points to our string}
test esi,esi {if nil, then exit}
jz @@ending
mov edx, [esi-4] {edx = length of the string}
mov eax,'+-+-' {pattern to use}
mov ecx, edx {length in counter register}
shr ecx,2 {divide, we process 4 bytes at once}
test ecx,ecx
jz @@remain
@@loop:
mov [esi],eax
add esi,4
dec ecx
jnz @@loop
@@remain: {fill the remaining bytes}
mov ecx, edx
and ecx, 3
jz @@ending
@@loop2:
mov BYTE PTR [esi],al
shr eax,8
inc esi
dec ecx

Guido Gybels: Using Assembler in Delphi

32

jnz @@loop2
@@ending:
pop esi
end;

The above example does not need to call any of the System.pas routines, but
it still relies on some implementation specific behaviour, namely where it
retrieves the long string's length. Long strings are preceded by two extra
dwords, a 32-bit length indicator at offset -4 and a 32-bit reference count at
offset -8. There is no guarantee that this scheme will remain unchanged forever.
To use the above function, allocate a string and then call the routine:
procedure DoSomething;
var
ALine: AnsiString;
begin
...
SetLength(ALine, {Required Length});
FillWithPlusMinus(ALine);
...
end;

Alternatively, we could opt to call the appropriate System.pas routine for

setting a long string's length (LStrSetLength), as illustrated in the
PlusMinusLine function for Delphi 7 below. From chapter 2 you will
remember that with the register calling convention, LineLength, as the first
parameter, will go in eax. The Result string is, as explained above, passed as
an extra var parameter. In this case our Result parameter is the second
parameter, so in case of the register calling convention it will go into edx.
Note that we also call UniqueStringA to ensure our result has a reference
count of 1. This is necessary, because our manipulation of the string's content
are unknown to the compiler.
function
PlusMinusLine(LineLength:
Integer):
register;
asm
push esi
push ebx
mov ebx, eax {ebx=LineLength}
mov esi, edx {esi=pointer to Result}
xchg edx, eax {eax=pointer to Result, edx=length}
call System.@LStrSetLength
mov ecx, [esi]
jecxz @@ending {if nil, then exit}
mov eax, esi
call System.@UniqueStringA
mov esi, [esi] {esi = first character of string}
mov eax, '+-+-' {pattern to use}
mov ecx, ebx {length in counter register}
shr ecx,2 {divide, we process 4 bytes at once}
test ecx,ecx

Guido Gybels: Using Assembler in Delphi

Ansistring;

33

jz @@remain
@@loop:
mov [esi],eax
add esi,4
dec ecx
jnz @@loop
@@remain: {fill the remaining bytes}
mov ecx, ebx
and ecx, 3
jz @@ending
@@loop2:
mov BYTE PTR [esi],al
shr eax,8
inc esi
dec ecx
jnz @@loop2
@@ending:
pop ebx
pop esi
end;

The example now uses the standard Result mechanism for returning data. It is
called simply with the required length to obtain a string:
procedure DoSomething;
var
ALine: AnsiString;
begin
...
ALine:=PlusMinusLine({Required Length});
...
end;

The above example above was written in Delphi 7. It should be very similar for
most other versions of Delphi, although the names for the internal functions do
differ somewhat between Delphi versions. Inspecting System.pas should help
you in identifying the appropriate function. You can also use the ctrl-left
button shortcut on the name of Pascal functions like UniqueString to jump
to their System.pas implementations.

Guido Gybels: Using Assembler in Delphi

34

Further Reading
The Art of Assembly
http://homepage.mac.com/randyhyde/webster.cs.ucr.edu/www.artofasm.com/index.html

Agner Fog: Optimizing subroutines in assembly language: An optimization

guide for x86 platforms
http://www.agner.org/optimize/optimizing_assembly.pdf

Intel 64 and IA-32 Architectures Software Developer's Manuals

http://www.intel.com/products/processor/manuals/

Guido Gybels: Using Assembler in Delphi

35

Table 1: Use of CPU registers

This table summarises register usage in Delphi 32-bit applications. The first
column lists the different CPU registers. The second column shows what the
register contains upon entry and the third indicates what the register holds upon
exit. The fourth column tells you whether or not you are allowed to use the
register inside your own code and the last column indicates if you need to
preserve the content of the register (save it upon entry and restore it before
leaving).
Use
allowed?

Entry

Exit

eax

Self (1), First parameter (2)

or undefined (3)

Function result (4)

No

ebx

Unknown

n/a

Yes

ecx

Second parameter (1),

Third parameter (2) or
undefined (3)

n/a

No

For Int64 result type:

highest dword,
otherwise n/a

No

edx

First parameter (1), Second

parameter (2) or undefined
(3)

Preserve?

esi

Undefined

n/a

Yes

edi

Undefined

n/a

Yes

ebp

Stack frame pointer

Stack frame pointer

(5)

Yes

esp

Stack pointer

Stack pointer

(5)

Yes

cs

Code selector

n/a

Yes

ds

Data selector

n/a

Yes

es

Data selector

n/a

Yes

fs

Thread information block

(TIB) selector

n/a

Yes

gs

Reserved

n/a

Yes

ss

Stack selector

Stack selector

Yes

(1)

Upon entry of methods when using the Register calling convention.

Upon entry of stand-alone functions and procedures while using the Register calling
convention.
(3)
Upon entry for methods and stand-alone functions and procedures in all calling conventions
(2)

Guido Gybels: Using Assembler in Delphi

36

except Register.
Only for Result types that qualify to go into a register. See table 4 for a complete overview of
how results are returned to the caller.
(5)
While you are allowed to use these registers inside your code, such practise is strongly
discouraged. The stack pointer itself changes with every stack operation and its content is
therefore highly volatile, making it difficult to manage in code. The ebp register is used to
access the stack frame and other uses are therefore to be avoided.
(4)

Guido Gybels: Using Assembler in Delphi

37

Table 2: Calling Conventions

This table gives an overview of the calling conventions supported by the Delphi
compiler. For parameter order, "left-to-right" means that the parameters are
pushed into registers or on the stack in the order that they are declared.
Correspondingly, "right-to-left" means that parameters are pushed in reverse
order.
Parameter order Clean-up by

Common Usage

register

left-to-right

Callee

Delphi applications

pascal

left-to-right

Callee

Backwards compatibility

cdecl

right-to-left

Caller

C and C++ libraries

stdcall

right-to-left

Callee

Windows API calls

safecall

right-to-left

Callee

COM and dual interface routines

Guido Gybels: Using Assembler in Delphi

38

Table 3: Parameter Passing

This table gives an overview of how parameters are passed. A distinction is
made between passing by value, declared as const or by reference. The "Value
in register?" column indicates whether or not a specific type qualifies for passing
in a register if it is passed as a value (applicable only to the "By Value" and
"Const" columns!). Passing by reference in the form of a 32-bit pointer always
qualifies for passing in a register. This is possible for all types, except method
pointers, that are always passed via the stack. For example, an int64 type does
not qualify according to the table below for passing in a register when it is
passed as a value. However, since passing it by reference means passing a 32bit pointer to the int64 value, passing an int64 by reference qualifies for use in a
register.
Value in
Register?

By Value

Const

ShortInt

8-bit value(1)

8-bit value(1)

32-bit pointer to 8bit Value(1)

SmallInt

16-bit value(1)

16-bit value(1)

32-bit pointer to
16-bit Value(1)

LongInt

32-bit value

32-bit value

32-bit pointer to
32-bit Value

Byte

8-bit value(1)

8-bit value(1)

32-bit pointer to 8bit Value(1)

Word

16-bit value(1)

16-bit value(1)

32-bit pointer to
16-bit Value(1)

Dword

32-bit value

32-bit value

32-bit pointer to
32-bit Value

Int64

32-bit pointer to
64-bit value2

32-bit pointer to
64-bit value2

32-bit pointer to
64-bit Value2

Boolean

8-bit value(1)

8-bit value(1)

32-bit pointer to 8bit Value(1)

ByteBool

8-bit value(1)

8-bit value(1)

32-bit pointer to 8bit Value(1)

WordBool

16-bit value(1)

16-bit value(1)

32-bit pointer to
16-bit Value(1)

LongBool

32-bit value

32-bit value

32-bit pointer to
32-bit Value

Guido Gybels: Using Assembler in Delphi

By Reference

39

Value in
Register?

By Value

Const

By Reference

AnsiChar

8-bit value(1)

8-bit value(1)

32-bit pointer to 8bit Value(1)

WideChar

16-bit value(1)

16-bit value(1)

32-bit pointer to
16-bit Value(1)

ShortString

32-bit pointer to
the string

32-bit pointer to
the string

32-bit pointer to
the string

AnsiString

32-bit pointer to
the string

32-bit pointer to
the string

32-bit pointer to a
32-bit pointer to
the string

Variant

32-bit pointer to
the variant

32-bit pointer to
the variant

32-bit pointer to
the variant

Pointer

32-bit pointer

32-bit pointer

32-bit pointer to a
32-bit pointer

Object
reference

32-bit pointer to
the object
instance

32-bit pointer to
the object
instance

32-bit pointer to a
32-bit pointer to
the object instance

Class
reference

32-bit pointer to
the class

32-bit pointer to
the class

32-bit pointer to a
32-bit pointer to
the class

32-bit pointer to
32-bit pointer to
the
the
procedure/function procedure/function

32-bit pointer to a
32-bit pointer to
the
procedure/function

Method
pointer

2x 32-bit pointer(3)

2x 32-bit pointer(3)

2x 32-bit pointer(3)

Set

8/16/32-bit value
or 32-bit pointer(4)

8/16/32-bit value
or 32-bit pointer(4)

32-bit pointer to
the set

Record

8/16/32-bit value
or 32-bit pointer(5)

8/16/32-bit value
or 32-bit pointer(5)

32-bit pointer to
the record

Static
Array

8/16/32-bit value
or 32-bit pointer(6)

8/16/32-bit value
or 32-bit pointer(6)

32-bit pointer to
the array

Dynamic
Array

32-bit pointer to
the array

32-bit pointer to
the array

32-bit pointer to a
32-bit pointer to
the array

Open Array

32-bit pointer to

32-bit pointer to

32-bit pointer to

Procedure
pointer

Guido Gybels: Using Assembler in Delphi

40

Value in
Register?

By Value

Const

By Reference

the array + 32-bit

size value7

the array + 32-bit

size value7

the array + 32-bit

size value7

Single

32-bit value

32-bit value

32-bit pointer to
32-bit Value

Double
(Real48)

64-bit value

64-bit value

32-bit pointer to
64-bit Value

Extended

80-bit value8

80-bit value8

32-bit pointer to
80-bit Value8

Currency

64-bit value

64-bit value

32-bit pointer to
64-bit Value

(1)

Data types that occupy less than 32-bits will still take up 32-bits. The actual value is stored in
the lowest parts of the stack location or register and the content of the remaining part is
undefined and should be treated as such at all times.
(2)
The pointer points to the lowest dword of the value. The highest dword is stored in the next
location.
(3)
Method pointers are always passed on the stack. They consist of an instance pointer, which
is pushed before the actual method pointer, which means the method pointer sits on the lowest
address on the stack.
(4)
If the Set contents fit into a byte/word/dword, its value is passed immediately, respectively as
a 8/16/32 bit value. Otherwise, a 32-bit pointer to the set is passed.
(5)
If the record contents fit into a byte/word/dword, the data is passed immediately, respectively
as a 8/16/32 bit value. Otherwise, a 32-bit pointer to the record is passed.
(6)
If the array contents fit into a byte/word/dword, the data is passed immediately, respectively
as a 8/16/32 bit value. Otherwise, a 32-bit pointer to the array is passed.
(7)
Open arrays are passed as 2 parameters: the first one is the pointer to the actual array, the
second one is the number of elements in the array. As such, passing an open array parameter
actually occupies 2 parameter slots. For instance: if you use the register calling convention
and you pass one open array parameter eax will contain the pointer to the array and edx will
contain the number of elements. See Chapter 2 for details about calling conventions. Also note
that open array parameters reside on the stack, so refrain from using very large arrays.
(8)
While the value itself occupies only 10 bytes, 12 bytes are actually allocated (3 dwords). The
content of the last 2 bytes should be considered undefined.

Guido Gybels: Using Assembler in Delphi

41

Table 4: Returning Results

This table gives an overview of how to return results from your assembler
routines in Delphi 32-bit applications. The first column lists the different data
types. The second column shows where such a type is returned. The third
column indicates the nature of the returned result.
Returned where?

What?

shortInt

al

8-bit signed value

SmallInt(1)

ax

16-bit signed value

LongInt(2)

eax

32-bit signed value

Byte

al

8-bit unsigned value

Word

ax

16-bit unsigned value

LongWord(3)

eax

32-bit unsigned value

Int64

edx:eax

64-bit signed value

Boolean

al

True or False

ByteBool

al

8-bit value, non-zero is true

WordBool

ax

16-bit value, non-zero is true

LongBool

eax

32-bit value, non-zero is true

Single

ST(0)(4)

80-bit floating point value

Double

ST(0)(4)

80-bit floating point value

Extended

ST(0)(4)

80-bit floating point value

Real48

ST(0)(4)

80-bit floating point value

Comp

ST(0)(4) (5)

80-bit floating point value

Currency

ST(0)(4)

80-bit floating point value(6)

AnsiChar

al

8-bit character

WideChar

ax

16-bit Unicode character

Byte

al

8-bit unsigned value

AnsiString

extra var parameter(7)

Pointer to pointer to Ansistring

(1)

The SHORT type is the same as SmallInt

The Integer generic type is currently mapped to LongInt
(3)
The DWORD and UINT types are the same as Longword. Also, the generic type Cardinal is
(2)

Guido Gybels: Using Assembler in Delphi

42

on the 32-bit platform mapped to a Longword

The FPU registers are 80-bits. All results are returned in ST(0). It's only the storage or
manipulation that will define them as single, double, etc.
(5)
Even though Comp represents a 64-bit integer, it is a type that is manipulated in the FPU, as
opposed to the other integer types, which are manipulated using the CPU. As such, Comp
follows the rules for real numbers in terms of manipulation and passing around.
(6)
Whole number scaled by 10,000
(7)
See Chapter 4 for a detailed explanation.
(4)

Guido Gybels: Using Assembler in Delphi

43

About the Author

Guido Gybels is a veteran Information and Communication
Technology (ICT) expert and senior manager, with a proven
track record of delivering award-winning innovation, research
and development activities, software and hardware
engineering projects, standardisation and policy and
regulatory strategy in ICT. A former Director of New
Technologies and Director of Technology, he is an
accomplished senior manager with in-depth experience of
planning, reporting, line management and controlling large
budgets.
Over the last two decades of his professional career, Guido
Gybels has been at the forefront of digital technologies for
desktop and mobile alike, with a special interest in Internet
applications, usability and user-focused design. His ongoing role
as a board member of the Centre for Usable Home
Technologies (CUHTec) at the University of York (UK) is
testament to his strong commitment to harness modern
technology and new media to deliver innovative new solutions in
the real world.
Guido combines strong technical knowledge and experience
with well-developed leadership skills, strategic thinking, an
evidence-based
approach to
management
and
an
uncompromising commitment to excellence.
In addition to being a very accomplished expert and leader in
ICT, Guido Gybels has also demonstrated great political and
public relations skills. He has extensive experience as a media
spokesperson, successfully delivering high profile media
coverage in the broadsheets, on radio and television. He acts as
an expert advisor to both the UK government and the European
Commission, in which role he interacts directly with senior civil
servants, MPs, Ministers and European Commissioners.
While being atypical for the profession, his educational
background indicates an agile and curious mind, having studied
and published in such diverse fields as history, linguistics,
geography, didactics, docimology and psychology.