>Wongck at 15:31, 12 October 2011

2011-10-12T15:31:30Z

← Older revision		Revision as of 11:31, 12 October 2011
Line 210:		Line 210:

	Back to [[ASM_Tutorial]]		Back to [[ASM_Tutorial]]
	[[Category:~~Programming~~]]		[[Category:Making optimized assembly code by Earx]]

>Silver Surfer: Added category

2009-05-02T17:13:20Z

Added category

← Older revision		Revision as of 13:13, 2 May 2009
Line 210:		Line 210:

	Back to [[ASM_Tutorial]]		Back to [[ASM_Tutorial]]
			[[Category:Programming]]

>Zorro 2 at 08:14, 9 October 2006

2006-10-09T08:14:12Z

← Older revision		Revision as of 04:14, 9 October 2006
Line 208:		Line 208:
	code better readable.		code better readable.
	</pre>		</pre>

			Back to [[ASM_Tutorial]]

>Simonsunnyboy at 17:49, 6 October 2006

2006-10-06T17:49:37Z

New page

<pre>
___ ___
/ \..__ __../ \
<< CHAPTER 8: STRUCTURISING >>
\___/ \___/

When making bigger and bigger demo's and games it, structurising is an
important issue. Contrary to what people say, even the most standard
assembler offers quite alot of functions for doing this:

EQUates:

A good first step in making your code more flexible is using equates. One is
defined by writing down this for instance:

label: EQU 1
label2: EQU 10

Now if you use label in your code:

move.w #label,d0

This moves whatever value you coupled to "label" into d0. The assembler sees
the label and replaces it with the value.

move.w #label*label2,d0

This moves label multiplied by label into d0. The assembler has no problems
with statements like these. You can also add, subtract, and, or, divide
labels with eachother.

Equates are very handy for bigger sourcecodes, where a certain value is used
very often. Equates make code more adaptable and often more readable.

A good example is for instance the size of an effectwindow or the
screendimensions. If you change either, you have a lot of code rewrite. But
simply using EQU's solves the whole thing quite easily. Simply giving the
labels' values will do the trick.

RS:

RS is basicly only used for defining datastructures. I've found them quite
handy. They're defined like this:

RSRESET * Indicate a new structure.
headerid: RS.L 1 * 0 Headerid is one longword.
headerxsize: RS.W 1 * 4 Xsize is one word.
headerysize: RS.W 1 * 6 Ysize is one word.
headersize: RS.B 0 * 8 total size of header

They basicly define offsets for datafield in structures. Now you can use
these offsets like this for instance:

lea header_buffer,a0 * Get address of header.
move.l #"HEAD",headerid(a0) * Move headerid in header.
move.w #16,headerxsize(a0) * Move xsize in header.
move.w #20,headerysize(a0) * Move ysize in header.
lea headersize(a0),a0 * Get startaddress of data.

Ofcourse this isn't very optimal code. The assembler just converts every
label to an offset. So every move is an offset move. You might want to use
"move ..,(a0)+" instead. That's ok, but you will lose out on easy
adaptability of the header.

So using RS is only really handy in situations where speed is not the
leading issue and the code must be very readable.

Directives:

These are often used in combination with equates. Sometimes used to comment
a block of code when an equate doesn't have the right value.

use_testmode: EQU 1

IFNE use_testmode
* These is additional code for the test mode.
;
;
ELSE
* These are instructions for when the testmode is disabled.
;
;
ENDC

"IFNE" is such a directive which includes a block depending on what value
it tests. If the label isn't equal to zero (which it isn't), then the whole
part will be included in the code. This is again done by the assembler. This
is very nifty if your sourcecode can be assembled in different modes (for
instance for black/white or color).

Another little trick which might be very essential for optimisin is the
"REPT" directive.

REPT 4
move.l d0,(a0)+
ENDR

This RePeaTs the instruction inbetween "REPT" and "ENDR" four times. This is
done the hardcoded way. So it will actually look like this in the final
executable:

move.l d0,(a0)+
move.l d0,(a0)+
move.l d0,(a0)+
move.l d0,(a0)+

REPT can also be used in conjunction with a label:

times_to_repeat: EQU 10

REPT times_to_repeat
move.l d0,(a0)+
ENDR

Especially for making innerloops a bit more optimised this can be very, very
handy!

Macro's:

People either love 'em or hate 'em. (Hi evl! =)) Macro's pieces of code much
like the subroutines, but they aren't called, but directly put in the
executable by the assembler. Macro's can have parameters, so in fact you can
use them to make assembler a bit more like a highlevel language.

stupid_macro: MACRO
move.l \1,d0
move.l \2,d1
ENDM

* Actual program goes in here
;
;
;
stupid_marco #1,#2

The problem is that the assembler replaces this with the code in the macro,
but it's mostly hard to know what exactly stands in the macro's. Knowing
your code is essential when debugging, so it can make your source harder to
comprehend.

Oh the other hand, it is the faster solution, because it doesn't suffer from
a bsr/jsr, rts combination. So it is wise to use macro's only for very small
pieces of code where this overhead seems like alot.

Using macro's in combination with IFxx directives, you can construct highly
optimised code for specific immediate-data parameters. Much like high-level
compilers do.

But please don't implement complete >100-line routines in a macro. It simply
makes your code into a mess.

Local labels:

Perhaps one of the most used assembler options ever. And it's explained in
here!! =) (Duuhhh....)

When you've got hundreds of labels already in your sourcecode you might run
out of new labels to use. To avoid your labels getting really big or
strange, you could use local labels.

The nifty extra of local labels is that you can only reference them in a
local area in the code. That is: only inbetween two global labels. If you
use Turbo Assembler, you can skip this. It hasn't got this feature.

******** :ExamplE: ********

bra .local1 * wrong!!

global1:
bra .local2 * correct
.local1:
bra .local2 * correct
.local2:
bra .local1 * correct

global2:
bra .local1 * wrong!!

The local labels always have points "." in front of them. As you can see you
can't reference local labels from behind another global label. Only between
two global labels it's possible. Ofcourse you can declare local labels with
the same names behind every new global label.

Local labels are ideal for use in subroutines. If you force yourself to only
use local labels within a subroutine you can check more easily if there are
any references outside the subroutine perhaps causing bugs. Also it clears
up the code because using 50 characters long names isn't anybody's idea of
having fun.

But beware that local labels don't show up in your debugger, so it might
actually only be good to use if you've already proven your subroutines work
bugfree. With debugging it's always better to use global labels, or at least
until your sure that routine is bugfree.

Register equates:

I haven't seen this used much or used this myself very often. using register
equates has the advantage of both keeping data in the internal registers
(=faster) as well as having easy names for the register (=more readable).

I only recommend to use this when the algorithm uses loads of registers,
but not when the register gets a different function everytime. This because
you can map multiple equated to one register, so sometimes it becomes
unclear when a register is overwritten or not.

A second disadvantage is the increased contrast between sourcecode and the
final code. This can make debugging more painful. Still, a nice way to make
code better readable.
</pre>

Earxtutchap8 - Revision history

>Wongck at 15:31, 12 October 2011

>Silver Surfer: Added category

>Zorro 2 at 08:14, 9 October 2006

>Simonsunnyboy at 17:49, 6 October 2006