5. Hints in connection with the Inline-Assembler (Part 1)
Shortindex:
5.1. Principles of the function of the Inline-Assembler
5.2. Assembler Syntax Error
5.3. Faulty passing of Variables in the Inline-Assembler
5.4. Problems with LDS/LES
5.5. Crash after calling own INT-Functions
5.6. Fixup Overflow
5.7. Dividing Variables from WORD to BYTE
5.8. Dividing Variables from DWORD to WORD
5.9. Access to Arrays / Structures with the Inline-Assembler
5.10. Parameter return with the Inline-Assembler
5.11. Parameter return in Interrupt-Procedures
5.12. Creating 32bit-Pointers
5.13. Converting from REG to Inline-ASM
5.14. Converting from A86 to Inline-ASM
The tips shown here are not a tutorial in Inline-Assembler, but may
help solve many beginner mistakes. A more detailes description of the
interactions will not be written here, if it does not belong to the
problem itself.
5.1. Principles of the function of the Inline-Assembler
The PowerBASIC Inline-Assembler contains the functions of the Intel
8086 CPU. This means that you may have to adapt the Inline-Assembler-
Code of other High-Language-Compilers or true Assembler-Code to the
PowerBASIC Inline-Assembler, because they quite often contain 80286
commands. Usually the following commands have to be converted:
Source - > PowerBASIC
shr ax, 2 ! shr ax, 1
! shr ax, 1
or like this:
-----------------------------------------------------------
shl ax, 3 ! push cx
! mov cl, 3
! shl ax, cl
! pop cx
-----------------------------------------------------------
pusha ! push ax
! push bx
and so on until all registers are
saved
-----------------------------------------------------------
popa analog, only restore registers
(End)
5.2. Assembler Syntax Error
When we forget the 'true' Syntax Errors, which usually occurs when you
are not quite used to Assemblercommands, there still is an 'obvious'
Syntax Error which can be marked by the Inline-Assembler of
PowerBASIC. This is the case when the Compiler can't use a Variable in
the Inline-Assembler because it isn't defined in any way.
PowerBASIC usually creates used Variables within true BASIC-Source by
itself and allocates memory for it. You have to do that yourself
within the Inline-Assembler.
This causes a Syntax Error because the Compiler can't do anything with
the Variable 'Demo%'. You should first set the Variable to a value:
Demo% = 1
! mov ax, Demo%
Now the Compiler accepts the Assemblerline. You don't have to assign a
value to a Variable every time, a simple DIM or SHARED, PUBLIC, LOCAL
etc. is enough and initialises 'Demo%'.
5.3. Faulty passing of Variables in the Inline-Assembler
You probably often swore around because a working routine with REG(x)
didn't work crrectly after conversion to the Inline-Assembler or when
your Testroutine wouldn't do its job in a SUB/FUNCTION.
The solution is relatively simple: You have to pass the variables to
the Inline-Assembler BYVAL.
Example:
Demo 1
FUNCTION Demo(BYVAL i%) public
! mov ax, i%
! inc ax
! mov i%, ax
PRINT i%
END FUNCTION
This little Demo simply adds the value '1' using the Inline-Assembler
and then prints it on the screen. Simply leave out the BYVAL and then
test the Demo again!
5.4. Problems with LDS/LES
Compareable to the passing of the parameters is the function of the
commands LDS/LES. It is also essential whether a variable is passed
'BY COPY', 'BY REFERENCE' or 'BY VALUE'. The followingcan be taken as
rule of thumb:
BY REFERENCE: - default in the main program
- or when a variable is declared SHARED/PUBILC etc.
BY COPY: - default in a SUB/FUNCTION, if the variables aren't
passed BY VALUE.
BY VALUE: - always interpreted by the Inline-Assembler as BY VALUE
You should only pass variables of type BY COPY to LDS/LES, because
only then the DS/ES Segmentaddresses will be loaded and the
Offsetaddresses into the other Registers.
When passing BY REFERENCE DS/ES will be loaded with the high value
contents of the variable, if of type Long/DWord, else the DS/ES
Register will contain an irrelevant value. The other Registers will
contain the low value of the variable.
Example:
SHARED DemoSeg%, DemoOff%
i& = &h12345678
Demo1 i&
Demo2 i&
Demo3 i&
FUNCTION Demo1(i&) public
PRINT "PB-Adresse : ";:
PRINT HEX$(VARSEG(i&));":"; HEX$(VARPTR(i&))
END FUNCTION
FUNCTION Demo2(i&) public
! les bx, i&
! mov DemoSeg%, es
! mov DemoOff%, bx
PRINT "LES /BY COPY : ";:
PRINT HEX$(DemoSeg%);":"; HEX$(DemoOff%)
END FUNCTION
FUNCTION Demo3(BYVAL i&) public
! les bx, i&
! mov DemoSeg%, es
! mov DemoOff%, bx
PRINT "LES /BY VALUE: ";:
PRINT HEX$(DemoSeg%);":"; HEX$(DemoOff%)
END FUNCTION
5.5. Crash after calling own INT-Functions
You will ask yourself: Why this section in the FAQ? Does the
PowerBASIC Inline-Assembler have any Bugs? The answer is definetly:
NO.
But many calls using INT-Functions of the BIOS/DOS are connected to
some trouble, because they change important Segments or specially
address them. Many buffers that have to be passed to a function look
for their pointer in the Datasegment-Register (DS). PowerBASIC also
addresses its variables using DS, so that conflicts are 'programmed'
here. It should not be like this, for example:
! mov ax, &h3D90 ; Function File open
! mov ds, FileSeg?? ; Load segment of Filename,
! mov dx, FileOff?? ; First error, because FileOff??
; can't be addressed using DS. DS
; is already pointing somewhere else.
! int &h21 ; INT-Call
! mov Handle%, ax ; Because DS still points to nowhere
; for PowerBASIC, this is wrong, too,
; and PowerBASIC will crash sooner or
; later.
A clean listing should look like this:
! push ds ; Save DS
! mov ax, &h3D90
! mov dx, FileOff?? ; Load Offset of Filename
! mov ds, fileSeg?? ; Load Segment of Filename, not needed
; by PowerBASIC anymore
! int &h21 ; INT-Call
! pop ds ; Restore PowerBASIC Segment
! mov Handle%, ax ; Save Handle% (Or Errorcode)
! jnc ... ; Check Carry-Flag
5.6. Fixup Overflow
The problem is relatively easy and simple: The 8086 CPU only allows
jumps of type SHORT, meaning that you can only jump to labels within
-127/+128 OpCodes directly.
The following example also creates such an error:
DemoLabel:
<more than 128 Bytes Opcode>
! jc DemoLabel
To get around this, you'll just have to address the whole thing a
little different. This is principally no problem once you know it:
DemoLabel:
<more than 128 Bytes Opcode>
! jnc DemoWeiter
! jmp near DemoLabel
DemoWeiter:
But this can be found in any good Assembler book...
5.7. Dividing variables from WORD to BYTE
Should you still turn your 16bit variables into its 8bit parts with
mathematical work, it is time to finally stop it! The CPU can do it by
itsself:
Example:
DIM Demo AS WORD
DIM DemoHigh AS BYTE
DIM DemoLow AS BYTE
Demo?? = &h1234
! mov ax, Demo??
! mov DemoLow? , al
! mov DemoHigh?, ah
5.8. Dividing variables frob DWORD to WORD
More often you have the problem that you have pointers of type DWORD
in PowerBASIC, but you don't know how to pass them to the Inline-
Assembler or turn them into WORD. Principally this is quite simple
(When you know how):
Example:
DIM Demo AS DWORD
DIM DemoHigh AS WORD
DIM DemoLow AS WORD
Demo??? = &h12345678
! mov ax, Demo???[00]
! mov bx, Demo???[02]
! mov DemoLow??, ax
! mov DemoHigh??, bx
5.9. Access to Arrays / Structures with the Inline-Assembler
Access to static Datastructures is relatively easy with the Inline-
Assembler if you already know the correct Offsetaddresses.
PowerBASIC allows the following Syntax:
Example:
! mov ah, byte ptr es:[di][22]
This copies the value to Offset 22 of the address ES:DI into the AH-
Register.
Hints in connection with the Inline-Assembler (Part 2)