ASM: Hardware Track Loader - Page 5

Asman · 23 March 2010, 09:55

Quote:

Originally Posted by StingRay

This is kinda useless (sorry) because load address is not constant.

Life is hard to save some bytes

So another possible size optimization: put $55555555 into some free register ( I guess we can use d7 ) and then do andi.l's in .mfm_decode.

pmc · 23 March 2010, 10:00

@ Asman - Correct but if I remember my own source correctly (bad memory - need to check!

) there are no free regs for the MFM decode mask...

pmc · 23 March 2010, 10:02

Asman, I bow - looks like d7 would be OK.

pmc · 23 March 2010, 10:06

Sting & Asman - legendary work guys.

I already implemented Asman's d7 save, I'll implement Sting's pea save too.

416 byte version of the trackloader to come! \o/

EDIT: Where'd Sting's pea save post just go...? Maybe only a 418 byte version to come...

StingRay · 23 March 2010, 10:06

Quote:

Originally Posted by Asman

Life is hard to save some bytes

I know

And I edited my post and added some more explanations.

StingRay · 23 March 2010, 10:08

Quote:

Originally Posted by pmc

Sting & Asman - legendary work guys.

I already implemented Asman's d7 save, I'll implement Sting's pea save too.

416 byte version of the trackloader to come! \o/

Sorry, I was wrong (and deleted my post).

(Didn't have enough coffee yet it seems). But see if you can find a free address register so you can do
lea $8000,a0
...
jmp (a0)

f.e.

Asman · 23 March 2010, 10:08

@StingRay

Awesome

@pmc
Agree, you should put move.l #$5555555,d7 before first bsr.s .mfm_decode.
And please post newest version of source.

pmc · 23 March 2010, 10:09

Sting - after yesterday's shrinking there's two free address regs - a0 and a1 if I remember correctly.

Maybe we might get below 418 yet!

StingRay · 23 March 2010, 10:13

Quote:

Originally Posted by pmc

Sting - after yesterday's shrinking there's two free address regs - a0 and a1 if I remember correctly.

Maybe we might get below 418 yet!

Use the pea optimization, easier. ;D Turns out I wasn't wrong and it is indeed 2 bytes smaller. ;D (too stressful to do "fun code" when boss is around ;D)

And since I deleted my original post, I'll post the pea optimization again.

pea $8000
move.l (a7),a4
...
rts

instead of
lea $8000,a4
....
jmp $8000

pmc · 23 March 2010, 11:08

OK boys, after implementing Sting's and Asman's additional size optimisations here's the 416 byte version!

Anyone who's helped in this is now credited in the source as having done so - thanks guys - much appreciated!

416 byte version removed.

Asman · 23 March 2010, 11:28

So we have one free register a1. Right after comment ; Setup disk controls you should put $4489 into a1 and change a bit source, mean replace all $4489 into a1.

I need some extra coffee. We have also a0 free.

StingRay · 23 March 2010, 11:31

Not tested but:
andi.l #$0000ff00,d1 ;isolate sector number
add.l d1,d1 ;*2 to get byte offset to place this sector in the decode buffer
adda.w d1,a4 ;add the offset to the decode data buffer address

could be done like this:
sf d1
add.w d1,d1
add.w d1,a4

StingRay · 23 March 2010, 11:54

Next one (maybe 400 bytes version is possible

)

moveq.l #71,d7 ;head settle time
bra.s .wait_timer
.upper_head:
bclr.b #2,(a6) ;select upper disk head
moveq.l #71,d7 ;head sett

can be replaced with:
bra.b .w_timer71
.upper_head:
bclr.b #2,(a6) ;select upper disk head
.w_timer71
moveq.l #71,d7 ;head sett

Asman · 23 March 2010, 12:02

Quote:

Originally Posted by StingRay

Next one (maybe 400 bytes version is possible

)

moveq.l #71,d7 ;head settle time
bra.s .wait_timer
.upper_head:
bclr.b #2,(a6) ;select upper disk head
moveq.l #71,d7 ;head sett

can be replaced with:
bra.b .w_timer71
.upper_head:
bclr.b #2,(a6) ;select upper disk head
.w_timer71
moveq.l #71,d7 ;head sett

Damn, I don't catch this ( I'm going to drink coffee ). Could you explain a bit. Thank you.

pmc · 23 March 2010, 12:03

Quote:

Originally Posted by StingRay

Next one (maybe 400 bytes version is possible

)

LOL. Smaller than 416 is definitely possivle - not sure 400 is!

Wait 71 timer branch optimisation implemented - trackloader now 414 bytes.

Quote:

Originally Posted by StingRay

Not tested but:
andi.l #$0000ff00,d1 ;isolate sector number
add.l d1,d1 ;*2 to get byte offset to place this sector in the decode buffer
adda.w d1,a4 ;add the offset to the decode data buffer address

could be done like this:
sf d1
add.w d1,d1
add.w d1,a4

Probably being stupid but - sf d1...?

pmc · 23 March 2010, 12:04

Quote:

Originally Posted by Asman

Damn, I don't catch this ( I'm going to drink coffee ). Could you explain a bit. Thank you.

Sting cleverly spotted that you could use the same moveq.l #71 for both head settle times by using a branch to the second one in place of the first one.

StingRay · 23 March 2010, 12:04

Quote:

Originally Posted by pmc

Probably being stupid but - sf d1...?

s.et f.alse = d1.b = 0.

StingRay · 23 March 2010, 12:06

Quote:

Originally Posted by Asman

Damn, I don't catch this ( I'm going to drink coffee ). Could you explain a bit. Thank you.

I explained this in one of my earlier posts.

Quote:

Originally Posted by StingRay

- "chain" your routines, e.g.
move.w #val,d7
bra.w .wait_timer
; some other code
move.w #val,d7
bra.w .wait_timer
; some other code
move.w #val,d7
bra.w .wait_timer

can be changed to:
move.w #val,d7
bra.b .w_timer
; some other code
move.w #val,d7
bra.b .w_timer
; some other code
move.w #val,d7
.w_timer
bra.w .wait_timer

pmc · 23 March 2010, 12:10

Quote:

Originally Posted by StingRay

s.et f.alse = d1.b = 0.

Ah, OK - set false. I'm being dumb though cos I still don't get how that enables replacing the andi that isolates the sector number from the decoded sector header longword.

StingRay · 23 March 2010, 12:13

Quote:

Originally Posted by pmc

Ah, OK - set false. I'm being dumb though cos I still don't get how that enables replacing the andi that isolates the sector number from the decoded sector header longword.

;d1: $ABCD1234
and.l #$0000ff00,d1 ; d1 = $00001200

;d1: $ABCD1234
sf.b d1 ; d1: $ABCD1200

since we only do .w operations, it doesn't matter what's in the high word of d1

thus: and.l #$0000ff00,d1 = sf.b d1

23 March 2010, 11:08	#90
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	OK boys, after implementing Sting's and Asman's additional size optimisations here's the 416 byte version! Anyone who's helped in this is now credited in the source as having done so - thanks guys - much appreciated! 416 byte version removed. Last edited by pmc; 23 March 2010 at 12:40.

23 March 2010, 11:28	#91
Asman 68k Join Date: Sep 2005 Location: Somewhere Posts: 828	So we have one free register a1. Right after comment ; Setup disk controls you should put $4489 into a1 and change a bit source, mean replace all $4489 into a1. I need some extra coffee. We have also a0 free. Last edited by Asman; 23 March 2010 at 11:49. Reason: added forget register a0

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
HD loader example	st!	Coders. General	4	16 October 2012 21:56
Game loader stuck on certain track after save state restore	andreas	support.WinUAE	2	26 March 2011 19:59
Can't transfer Supaplex cause of CSL track loader !	Vollldo	support.Games	4	12 March 2011 21:51
Hardware File Loader	h0ffman	Coders. General	9	02 December 2010 16:40
mfm/custom regs/track loader	snyp	Coders. General	9	06 June 2006 19:42

23 March 2010, 10:00	#82
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	@ Asman - Correct but if I remember my own source correctly (bad memory - need to check! ) there are no free regs for the MFM decode mask...

23 March 2010, 10:02	#83
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	Asman, I bow - looks like d7 would be OK.

23 March 2010, 10:06	#84
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	Sting & Asman - legendary work guys. I already implemented Asman's d7 save, I'll implement Sting's pea save too. 416 byte version of the trackloader to come! \o/ EDIT: Where'd Sting's pea save post just go...? Maybe only a 418 byte version to come...

23 March 2010, 10:08	#87
Asman 68k Join Date: Sep 2005 Location: Somewhere Posts: 828	@StingRay Awesome @pmc Agree, you should put move.l #$5555555,d7 before first bsr.s .mfm_decode. And please post newest version of source.

23 March 2010, 10:09	#88
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	Sting - after yesterday's shrinking there's two free address regs - a0 and a1 if I remember correctly. Maybe we might get below 418 yet!

23 March 2010, 11:31	#92
StingRay move.l #$c0ff33,throat Join Date: Dec 2005 Location: Berlin/Joymoney Posts: 6,863	Not tested but: andi.l #$0000ff00,d1 ;isolate sector number add.l d1,d1 ;*2 to get byte offset to place this sector in the decode buffer adda.w d1,a4 ;add the offset to the decode data buffer address could be done like this: sf d1 add.w d1,d1 add.w d1,a4

23 March 2010, 11:54	#93
StingRay move.l #$c0ff33,throat Join Date: Dec 2005 Location: Berlin/Joymoney Posts: 6,863	Next one (maybe 400 bytes version is possible ) moveq.l #71,d7 ;head settle time bra.s .wait_timer .upper_head: bclr.b #2,(a6) ;select upper disk head moveq.l #71,d7 ;head sett can be replaced with: bra.b .w_timer71 .upper_head: bclr.b #2,(a6) ;select upper disk head .w_timer71 moveq.l #71,d7 ;head sett

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)