Addressing mode syntax question

Can anyone explain this indirect addressing mode syntax I have come across a couple of times? Example:

ChkMouse: move.w $a-2(a6),d3

(This example from one of Photon's old snippets)

It looks like the hardware register base address $dff000 is expected to be in a6 prior to calling this subroutine, and I would expect JOY0DAT ($dff00a) to be read given the purpose of the code. So what is the -2 for? To me it looks like it would read $dff008 (DISKDATR) which cannot be correct!

I think Photon may have tried to explain this briefly in one of his (excellent) YouTube videos, but I didn't get it.

[Thorham]

Yep, it's 8(a6). Really. Who knows what's in a6... Probably Photon

[Kalms]

a6 is intended to point to $dff002 (DMACONR).

The code in question wants to read from $dff00a (JOY0DAT) which contains the current mouse cursor position.


... and setting a6 to $dff002 instead of $dff000 is a micro-optimization: there are probably a couple of places in your program where you want to access $dff002 but none where you want to access $dff000. Therefore your program will become a couple of bytes smaller by using $dff002 as your base pointer... although, as you already noticed, the sourcecode becomes a tad more confusing. Whether or not it's worth it is up to you.

[Codetapper]

In that case, it would be cleaner to make up your own custom.i with values such as:

OFFSET equ 2
custom equ $dff000+OFFSET
dmaconr equ $2-OFFSET
joy0dat equ $a-OFFSET
bltsize equ $58-OFFSET

etc. Then you could still use bltsize(a6) in your code rather than bltsize-2(a6)

[Leffmann]

I agree that hard-coded magical numbers is something you should avoid. I use this:

               ifnd HARDWARE_I

HARDWARE_I = 1

               ifnd hw_base
hw_base = $dff000
               endif
hw_rel = hw_base-$dff000

;------------- Hardware registers and bits -------------------------------------

bltddat  = 0-hw_rel
dmaconr  = 2-hw_rel
vposr    = 4-hw_rel
vhposr   = 6-hw_rel
...

Originally Posted by Kalms

... setting a6 to $dff002 instead of $dff000 is a micro-optimization

That explains all. I guess smaller offsets = smaller instruction lengths = shorter processing time. I should maybe read the optimisation sticky thread before posting again! Thanks.

[Lonewolf10]

Originally Posted by Leffmann

I agree that hard-coded magical numbers is something you should avoid. I use this:

               ifnd HARDWARE_I

HARDWARE_I = 1

               ifnd hw_base
hw_base = $dff000
               endif
hw_rel = hw_base-$dff000

;------------- Hardware registers and bits -------------------------------------

bltddat  = 0-hw_rel
dmaconr  = 2-hw_rel
vposr    = 4-hw_rel
vhposr   = 6-hw_rel
...

Why? I thought all Amiga hardware registers were in the same place for all classic models. The only current exception I am aware of is the interrupt pointers which can be offset using the VBR register on computers using 68020+.

I do tend to code specifically for the A600 (as it's the only real one I own), but try to make my software compatible with other models where possible.

[Kalms]

Many programmers find it easier to understand the intent of this piece of code:

.wait_for_colorful_effect_start
	move.w	vhposr,d0
	cmp.w	#COLOURFUL_EFFECT_START_Y<<8,d0
	blo.s	.wait_for_colorful_effect_start

... than the intent of this piece of code:

.loop	
	move.w	$dff006,d0
	cmp.w	#$4300,d0
	blo.s	.loop

[Leffmann]

Originally Posted by Lonewolf10

Why? I thought all Amiga hardware registers were in the same place for all classic models. The only current exception I am aware of is the interrupt pointers which can be offset using the VBR register on computers using 68020+.

They're in the same place on all models. If you look you'll see it always translates to the correct address, no matter how hw_base is defined. I usually ignore these micro-optimizations and just go with "dmacon" instead of "(dmacon, a6)" etc.

And like Kalms points out, it's about getting cleaner code. It doesn't make sense writing M68K code and Copper programs as lumps and blobs of hexadecimal digits. It's what people had to do in the 70's on their MOS 6502 computer kits, and I don't think it's something that should be taught to people who look to these tutorials for getting into Amiga programming.

[Megol]

Using more verbose code with defined names also reduces the risk of errors - it's very easy to write the wrong hexadecimal digit unnoticed while spelling a defined name wrong will likely result in an error message.

[Lonewolf10]

Originally Posted by Leffmann

They're in the same place on all models. If you look you'll see it always translates to the correct address, no matter how hw_base is defined.

Ahh, yes I see that now.

Originally Posted by Leffmann

And like Kalms points out, it's about getting cleaner code. It doesn't make sense writing M68K code and Copper programs as lumps and blobs of hexadecimal digits. It's what people had to do in the 70's on their MOS 6502 computer kits, and I don't think it's something that should be taught to people who look to these tutorials for getting into Amiga programming.

I have started to clean up my code a bit in recent months (e.g using macro's more), but I still have a lot to learn

Thanks guys.