ASM: Hardware Track Loader - Page 6

pmc · 23 March 2010, 12:20

That's why I was being dumb - I didn't even realise sf.b was a valid mnemonic!!

I'd never come across it before!

You're still the man Sting.

pmc · 23 March 2010, 12:41

Tested as working 410 byte version attached.

StingRay is officially a legend.

musashi5150 · 23 March 2010, 13:22

Code:

move.w #%0111111111111111,d0
move.w d0,intena(a5)
move.w d0,dmacon(a5)
move.w d0,intreq(a5)

Code:

move.l #%01111111111111110111111111111111,d0
move.l d0,intena(a5)
move.w d0,dmacon(a5)

Doesn't save any bytes...? but could save you some memory cycles

Asman could be onto something with his $4489 in a register idea, but it might just trade off ? Damn, my kingdom for an assembler

korruptor · 23 March 2010, 13:35

Best thread ever

pmc · 23 March 2010, 13:39

Quote:

Originally Posted by musashi5150

Doesn't save any bytes...?

Does actually - two.

Are you aiming to take the "legend" crown?!

Quote:

Originally Posted by musashi5150

Asman could be onto something with his $4489 in a register idea, but it might just trade off ?

But does this work putting $4489 into an *address* register as suggested rather than a data register?

In an address registers $4489 means something very different, no?

Asman · 23 March 2010, 13:40

Quote:

Originally Posted by musashi5150

Asman could be onto something with his $4489 in a register idea, but it might just trade off ? Damn, my kingdom for an assembler

I thought about something like that

Code:

; Setup disk controls

	move.w	#$4489,a1
	move.w	a1,dsksync(a5)		;MFM sync mark

...

.seek_hdr_strt:	cmp.w	a1,(a3)+		;find start of sector
	bne.s	.seek_hdr_strt
	cmp.w	a1,(a3)

This should save some bytes.

pmc · 23 March 2010, 13:44

OK guys - musashi5150's extra two byte saving has been implemented - trackloader now standing at 408 bytes.

I'll investigate Asman's suggestion after I've eaten some lunch.

New version to come...

Toni Wilen · 23 March 2010, 13:52

Disk DMA enable/disable and DSKLEN $4000 writes can be removed. (Perhaps it is recommend somewhere but there really isn't any difference unless you need to abort active DMA)

StingRay · 23 March 2010, 14:18

I have a "clean" (i.e. Disk DMA still enabled etc) 402 bytes version.

Once I can kill the last 2 bytes I won't touch this code again.

StingRay · 23 March 2010, 14:34

398 bytes version attached. More can be done but I think I did my fair share of killing bytes.

Edit: just noticed there was a small bug in the init part (move.l/move.w) which I introduced when I applied Musashi's optimization.

Fixed version attached. =)

pmc · 23 March 2010, 14:34

Quote:

Originally Posted by StingRay

I have a "clean" (i.e. Disk DMA still enabled etc) 402 bytes version.

Massive power!

The king wants to keep his crown!

Edit:

Quote:

Originally Posted by StingRay

398 bytes version attached. More can be done but I think I did my fair share of killing bytes.

Even more power!

Asman · 23 March 2010, 14:40

We can do another one dirty trick to save some bytes (probably). Just replace

Code:

	lea	cust_regs,a5

into

Code:

	lea	cust_regs+dsklen,a5

and change all references to a5. ( for example all dsklen(a5) ---> (a5)

)

StingRay · 23 March 2010, 14:43

Quote:

Originally Posted by Asman

We can do another one dirty trick to save some bytes (probably). Just replace

Code:

    lea    cust_regs,a5

into

Code:

    lea    cust_regs+dsklen,a5

and change all references to a5. ( for example all dsklen(a5) ---> (a5)

)

Did you check my attached version?

Asman · 23 March 2010, 14:45

Ahhhhh..... Yeeaah. You did that. Awesome

StingRay · 23 March 2010, 14:48

Quote:

Originally Posted by Asman

Ahhhhh..... Yeeaah. You did that. Awesome

Almost.

I did lea dsklen(a5),a0 as I wanted to keep the code at least somewhat readable. Your approach would be smaller (as no additional lea instruction is required) but I didn't do it like that for reasons already mentioned.

phx · 23 March 2010, 14:55

Quote:

Originally Posted by StingRay

398 bytes version attached. More can be done but I think I did my fair share of killing bytes.

Oh no... no PRINTV! Now I cannot assemble it any longer!

Great stuff! I didn't expect that to go lower than 400 bytes!

Some more. You can deselect and disable the diskmotor bit at once:

Code:

trkld_done:    bset.b  #3,(a6)         ;deselect df0:
        bset.b  #7,(a6)         ;dskmotor high
-->
trkld_done:   or.b    #%10001000,(a6)

Asman · 23 March 2010, 14:56

Quote:

Originally Posted by StingRay

Almost.

I did lea dsklen(a5),a0 as I wanted to keep the code at least somewhat readable. Your approach would be smaller (as no additional lea instruction is required) but I didn't do it like that for reasons already mentioned.

I just realized that. The code should be readable, we can change all equates for custom registers

Or better some tool should do this ( first analyze code and count what should be replaced

).

pmc · 23 March 2010, 15:00

This has been a really interesting experience for me.

I'm starting to see now that once you get right down to trying to wring every last space saving out of a piece of code that you start getting into a trade off for those size decreases in terms of both readability of the code and also speed of execution of that code.

As a middle way I think that the version of this code I actually use going forward will be the one I think has the right balance of these trade offs for me personally.

StingRay · 23 March 2010, 15:01

Quote:

Originally Posted by phx

Oh no... no PRINTV! Now I cannot assemble it any longer!

I have attached a new version for you Asm1 haters. :P Hooray for MACRO power. \o/ :P

Asman · 23 March 2010, 15:04

Quote:

Originally Posted by pmc

This has been a really interesting experience for me.

For me too. Now I understand the power of equates in source. ( I always use #include "hardware/custom.i" ).

23 March 2010, 12:41	#102
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	Tested as working 410 byte version attached. StingRay is officially a legend. Last edited by pmc; 23 March 2010 at 20:11.

23 March 2010, 13:22	#103
musashi5150 move.w #$4489,$dff07e Join Date: Sep 2005 Location: Norfolk, UK Age: 43 Posts: 2,351	Code: move.w #%0111111111111111,d0 move.w d0,intena(a5) move.w d0,dmacon(a5) move.w d0,intreq(a5) Code: move.l #%01111111111111110111111111111111,d0 move.l d0,intena(a5) move.w d0,dmacon(a5) Doesn't save any bytes...? but could save you some memory cycles Asman could be onto something with his $4489 in a register idea, but it might just trade off ? Damn, my kingdom for an assembler

23 March 2010, 14:40	#112
Asman 68k Join Date: Sep 2005 Location: Somewhere Posts: 829	We can do another one dirty trick to save some bytes (probably). Just replace Code: lea cust_regs,a5 into Code: lea cust_regs+dsklen,a5 and change all references to a5. ( for example all dsklen(a5) ---> (a5) )

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, 12:20	#101
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	That's why I was being dumb - I didn't even realise sf.b was a valid mnemonic!! I'd never come across it before! You're still the man Sting.

23 March 2010, 13:35	#104
korruptor TDI Join Date: Feb 2007 Location: Blitter Town Posts: 124	Best thread ever

23 March 2010, 13:44	#107
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	OK guys - musashi5150's extra two byte saving has been implemented - trackloader now standing at 408 bytes. I'll investigate Asman's suggestion after I've eaten some lunch. New version to come...

23 March 2010, 13:52	#108
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,574	Disk DMA enable/disable and DSKLEN $4000 writes can be removed. (Perhaps it is recommend somewhere but there really isn't any difference unless you need to abort active DMA)

23 March 2010, 14:18	#109
StingRay move.l #$c0ff33,throat Join Date: Dec 2005 Location: Berlin/Joymoney Posts: 6,865	I have a "clean" (i.e. Disk DMA still enabled etc) 402 bytes version. Once I can kill the last 2 bytes I won't touch this code again.

23 March 2010, 14:45	#114
Asman 68k Join Date: Sep 2005 Location: Somewhere Posts: 829	Ahhhhh..... Yeeaah. You did that. Awesome

23 March 2010, 15:00	#118
pmc gone Join Date: Apr 2007 Location: completely gone Posts: 1,596	This has been a really interesting experience for me. I'm starting to see now that once you get right down to trying to wring every last space saving out of a piece of code that you start getting into a trade off for those size decreases in terms of both readability of the code and also speed of execution of that code. As a middle way I think that the version of this code I actually use going forward will be the one I think has the right balance of these trade offs for me personally.

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