Then perhaps I should build a Jabber "mailing list" which forwards
messages and archives them... Or maybe a gateway that links the Yahoo
Group into Jabber so users can send/receive the ML messages in Jabber...

James Margaris wrote:

> For many of us lurkers the Jabber stuff wouldn't be archived and we would
> end up missing out. Obviously that is no problem for off-topic discussion,
> but for on-topic clarifications if one person asks the question there are
> probably 10 wondering the same thing.
>
> James M
>
>
>
> *Yahoo! Groups Sponsor*
> <http://us.a1.yimg.com/us.yimg.com/a/pl/playstation/468x60_tripleplay.jpg>
>
<http://rd.yahoo.com/M=214508.1858224.3361270.1501205/D=egroupweb/S=1705055196:H\
M/A=950747/R=0/*http://content.search.shopping.yahoo.com/search/tmpl?tmpl=psshow\
case2001.html&query=tag:PSshowcase2001+%23cversion%3A%7Bimage_PSshowcase2001+url\
_PSshowcase2001+desc_PSshowcase2001+title_PSshowcase2001+morehtml_PSshowcase2001\
%7D&q=PSshowcase2001>
>
>
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service
> <http://docs.yahoo.com/info/terms/> .

Hi

Does anyone know how to force GCC to order functions in memory in the order
in which they are declared in the source file?

Thanks

Mat

For many of us lurkers the Jabber stuff wouldn't be archived and we would
end up missing out. Obviously that is no problem for off-topic discussion,
but for on-topic clarifications if one person asks the question there are
probably 10 wondering the same thing.

James M

> Is anyone on this list on jabber? It would be a nice way to have the more
off
> topic discussions or quick chats for clarifications of emails etc.

Heh.  Not that anyone will particularly want to talk to me, but:

YM: johnisaheadcase
ICQ: 1110764
AIM: das monad

However, as I use Trillian, my AIM account is not currently being connected
to - AOL is breaking their setup, trying to get rid of alternate clients.
YM and ICQ are the most reliable ways to get in touch with me, other than
email.

That said, I actually really prefer stuff gonig through the mailing list.
There have been a lot of minor discussions between other people that have
benefitted me as well, as one person will clear up another's
misunderstanding, which happened to be a problem for me too (and I didn't
even realize it yet).

Anybody else have a view on this?

- Fatty


_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com

Erm, then isn't ICQ, with it's huge user base, the best option?

FGL

ICQ: 125566364

:)

[mod note: I'm not sure what the point of this discussion is, but it's starting
to get out of hand]


-----Original Message-----
From: Tom Badran [mailto:tb100@...]
Sent: 28 February 2002 16:09
To: gbadev@yahoogroups.com
Subject: Re: [gbadev] Jabber


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> Isn't IRC suitable for this kind of stuff?

IRC is good for group chats, but is a bit overkill for quick messages to
people, and doesnt support offline messaging.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> Isn't IRC suitable for this kind of stuff?

IRC is good for group chats, but is a bit overkill for quick messages to
people, and doesnt support offline messaging.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE8flYUXCpWOla2mCcRAqy7AJ4oo8FK8qW/ot/X9kHjuhd0PCTO8QCeOakC
/grUap3NlZ+hSvvoCOHGh6Q=
=kBjz
-----END PGP SIGNATURE-----

On Thu, 28 Feb 2002, Dennis Munsie wrote:

> Also, there is the possibility of using Yahoo! Messenger... everyone here
> had to sign up with yahoo to get on the list...

I don't mean to be picky, but that's not strictly true - you can just
email to "gbadev-subscribe@yahoo.egroups.com" or whatever - the moderators
still have to accept you but you don't /need/ a yahoo account.

Cheers, Pete
--
http://www.bits.bris.ac.uk/dooby/

Hmmmm,

Well I don't know anyone that uses Yahoo! Messenger at all..

Isn't IRC suitable for this kind of stuff?

Greetings,

Collin van Ginkel

----- Original Message -----
From: "Dennis Munsie" <munsied@...>
To: <gbadev@yahoogroups.com>
Sent: Thursday, February 28, 2002 4:03 PM
Subject: RE: [gbadev] Jabber


> Also, there is the possibility of using Yahoo! Messenger... everyone here
> had to sign up with yahoo to get on the list...
>
> dennis
>
> -----Original Message-----
> From: Tom Badran [mailto:tb100@...]
> Sent: Thursday, February 28, 2002 8:02 AM
> To: gbadev@yahoogroups.com
> Subject: [gbadev] Jabber
>
>
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Is anyone on this list on jabber? It would be a nice way to have the more
> off
> topic discussions or quick chats for clarifications of emails etc.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


> If i remember correctly a lot of people did something like this with ICQ on
> the old game boy mailing list and it worked quite well. So, if anyone is
> using i, i can be reached on tombadran@..., and ill be happy to
> maintain a list of people on my site so that we can all communicate that

tombadran@... is NOT an email address, it is a jabber user name, just
thought i would clarify this. I dont work for the company (im a student) and
have no other affiliation other than i use there server for my jabber setup.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE8fjMOXCpWOla2mCcRAquLAJ9mv7mk2iXi8Z/P+W5A+SqSBJX1vACfXb42
oONlbSk//xORgsAxlE3HxG8=
=M4p4
-----END PGP SIGNATURE-----

Also, there is the possibility of using Yahoo! Messenger... everyone here
had to sign up with yahoo to get on the list...

dennis

-----Original Message-----
From: Tom Badran [mailto:tb100@...]
Sent: Thursday, February 28, 2002 8:02 AM
To: gbadev@yahoogroups.com
Subject: [gbadev] Jabber


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Is anyone on this list on jabber? It would be a nice way to have the more
off
topic discussions or quick chats for clarifications of emails etc.

I just looked at day 5 of the Pern Project.
Nobody seems to have mentioned here the special
DMA3 mode for updating the screen in mode 3.
Apparently it updates each line just after it's been displayed.
So provided you don't mind dma outside
vertical blanking, (which someone said upsets
comms) this seems a nice way to replace the picture
in one frame.

Anyone tried it?

So, your calculations tally exactly with what I come up with in a practical
appplication.  Don't know whether to be happy because I was right, or sad
because I can't go any faster :) :(

FGL

-----Original Message-----
From: Dennis Ranke [mailto:exoticorn@...]
Sent: 28 February 2002 10:34
To: gbadev@yahoogroups.com
Subject: RE: Re[2]: [gbadev] Mode 3 dma


In message <MGEAIFINLPJAPLPELHMCIEOICEAA.georgea@...>
           "George Andre" <georgea@...> wrote:

> I thought I should contribute to this discussion, since
> I'm doing the DMASpeed test program.. (to be released soon):
>
> mode 3 : 240 halfwords * 160 lines = 38400 halfwords / 19200 words.
>
> vblank time : 68 scanlines * 1232 cycles = 83776 cycles.
>
> DMA speed from ROM to VRAM is 2 cycles per 16 bit transfer respective
> 3 cycles per 32 bit transfer.

actually, it's 3 cycles per 16bit transfer and 6 cycles per 32bit transfer.

> Thus,
> doing 16 bit DMA takes 38400 * 2 = 76800 cycles, while
> doing 32 bit DMA takes 19200 * 3 = 57600 cycles,

which gives 38400 * 3 = 115200 cycles in both cases...

> which is 91.7% respective 68.7% of the total VBlank time.

which is 137.5% of the total vblank time...

> So there you see, it's possible to copy a full mode3 screen
> in Vblank time.

so there you see , it's not possible to copy a full mode3 screen
in vblank time. ;)

--
exoticorn/icebird^farbrausch





Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Is anyone on this list on jabber? It would be a nice way to have the more off
topic discussions or quick chats for clarifications of emails etc.

If i remember correctly a lot of people did something like this with ICQ on
the old game boy mailing list and it worked quite well. So, if anyone is
using i, i can be reached on tombadran@..., and ill be happy to
maintain a list of people on my site so that we can all communicate that way
as well as on this excellent mailing list.

Tom
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE8fio/XCpWOla2mCcRAkYuAKDbDkGILNxVkhBiUpJ/BwrheRxKsQCeKdgb
R2VYi0sleYUVacIfg2XjSsY=
=3uWi
-----END PGP SIGNATURE-----

> -----Original Message-----
>  3 cycles per 32b ROM -> VRAM seems unlikely to me, because:
>
>  VRAM write is 2 cycles
>  ROM read + DMA itselft should be at least 1 cycle
>  = 3 ... till here it looks ok
>  but when screen is displayed, the VRAM is occasionaly slowed
> down, so VRAM

These tests are done with screen forced blank (ir no display).

Aelius-

> -----Original Message-----
> > I thought I should contribute to this discussion, since
> > I'm doing the DMASpeed test program.. (to be released soon):
> >
> > mode 3 : 240 halfwords * 160 lines = 38400 halfwords / 19200 words.
> >
> > vblank time : 68 scanlines * 1232 cycles = 83776 cycles.
> >
> > DMA speed from ROM to VRAM is 2 cycles per 16 bit transfer respective
> > 3 cycles per 32 bit transfer.
>
> actually, it's 3 cycles per 16bit transfer and 6 cycles per 32bit
> transfer.

Humm, exoticorn, since you're a well respected member of this community
I'll revise my program and tests.. AND complete and release it
(as promised to Joat :) ).

This was the reason I wanted to do such a program for testing speed
down to 1-cycle resolution to once and for all have a viable
DMA speed table. I must find time to finish completely this week.

Aelius.

If you have some really cool effects ready next summer, why not
participate the Assembly'02 mobile demo compo. Prizes will be
several hundred euros and you can even participate remotely by
email.

http://www.assembly.org/content/compos/rules_mobile_demo.html
http://www.assembly.org/content/compos/rules_general.html

It'll be very interesting to see how PocketPCs and
Comminicators compete each other :)

If you have something to comment on the compo rules, just email
me before the end of March.


Regards,

- wiba

Ville Vatén
Mobile Compo Organizer
Assembly Organizing

Its probably worth pointing out at this point that doing large DMA
copies/sets etc if you are planning on doing a multiplayer game is not
recommended.  It seems to be ok as long as the dma is all done in the
vertical blank, but large usages during screen time causes a series drop in
the performance of comms stability.

Also, I found that the most data I could transfer during a single frame on a
4 player connection was about 20 bytes.  If I went much higher than this I
could connect 2 players but the third and fourth would not connect properly.
What have other people found with communications transfer limits?

Adrian


[cropped by mod]

In message <MGEAIFINLPJAPLPELHMCIEOICEAA.georgea@...>
           "George Andre" <georgea@...> wrote:

> I thought I should contribute to this discussion, since
> I'm doing the DMASpeed test program.. (to be released soon):
>
> mode 3 : 240 halfwords * 160 lines = 38400 halfwords / 19200 words.
>
> vblank time : 68 scanlines * 1232 cycles = 83776 cycles.
>
> DMA speed from ROM to VRAM is 2 cycles per 16 bit transfer respective
> 3 cycles per 32 bit transfer.

actually, it's 3 cycles per 16bit transfer and 6 cycles per 32bit transfer.

> Thus,
> doing 16 bit DMA takes 38400 * 2 = 76800 cycles, while
> doing 32 bit DMA takes 19200 * 3 = 57600 cycles,

which gives 38400 * 3 = 115200 cycles in both cases...

> which is 91.7% respective 68.7% of the total VBlank time.

which is 137.5% of the total vblank time...

> So there you see, it's possible to copy a full mode3 screen
> in Vblank time.

so there you see , it's not possible to copy a full mode3 screen
in vblank time. ;)

--
exoticorn/icebird^farbrausch

On Visoly cartridges it's ALMOST without penalty... They crash like hell,
just different that Nintendo Cartridges.

- Tom
   -----Oprindelig meddelelse-----
   Fra: ninge [mailto:ninge@...]
   Sendt: 26. februar 2002 10:43
   Til: gbadev@yahoogroups.com
   Emne: RE: [gbadev] sprintf()


   its about time nintendo fixed this problem with their crt0 and
linkscripts! - this must be the 10th time this exact same problem has been
brought up on this list... the solution is to either use a decent crto.s and
linkscript that dosnt stick initialised variables into ROM (like jeff's) or
just dont use sprintf - easy :)

   the reason it works on the devkit and visoly carts is that you can write
to ROM without penalty on them..

   ninge

Hi all,

  3 cycles per 32b ROM -> VRAM seems unlikely to me, because:

  VRAM write is 2 cycles
  ROM read + DMA itselft should be at least 1 cycle
  = 3 ... till here it looks ok

  but when screen is displayed, the VRAM is occasionaly slowed down, so VRAM
write
may be about 2.5 - 3 cycles ... If your DMAspeed could really reach 3cycles,
I'll be glad.
(Couldn't test by myself, I don't have real GBA :))

  And this makes me wonder, is VRAM writes penalised also in H/VBLANK period,
or
only while screen is being drawn ? Anybody knows the timing good enough to
answer me?

                   PED - Peter Helcmanovsky - wishing to have 48h days
               Braniskova 7 Kosice 04001 Slovakia phone: +421 55 6259058
    (_               -- actually in Brno, so don't look after me --
   "~/\~"      -=- deRATized  RAT -=-  Being beyond Yer imagination  -=-
   ,_oo_,      7 GODS demo group WWW pages: http://7gods.sk
     `'

>I thought I should contribute to this discussion, since
>I'm doing the DMASpeed test program.. (to be released soon):

I'm looking forward to it.

I have only one question about DMAs on the GBA, that I'd love to see
someone less lazy than myself investigate:

What is the minimal copy length such that it's worthwhile doing a
hardware DMA transfer?

For example, if you needed to only copy 8 bytes, it would be stupid to
use a DMA transfer, since writing to the DMA registers will take longer
than just copying the memory yourself.

DMAs are undisputably fast for memsetting or memcopying large runs
(much faster even than unrolled stm/ldm, as pointed out), but keep in
mind there's still some overhead associated with setting them up.

In particular, I'm curious about the optimal way to copy data for a
single 16 color tile (for example, as part of a dynamic tile caching
scheme).  That's a mere 32 bytes.


__________________________________________________
Do You Yahoo!?
Yahoo! Greetings - Send FREE e-cards for every occasion!
http://greetings.yahoo.com

> tried with screen on and obj-processing in hblank disabled aswell?
> (now you are actually measuring, i wanna know aswell :O))

No, not yet.  My timigs so far are normal operation screen on with no HBlank
and all screens off, with no objects or HBlank.

As another aside, my colleague wrote a fast fixed 1K copy (for copying OAM
IWRAM data to OAM) in ARM assembly and copied it to IWRAM, and it still
wasn't as quick as a DmaCopy()

ASM_Cpu1K:

		 stmfd   sp!,{r4-r11} 		 @ Push R4, R5, R6, R7, R8, R9, R10, R11

		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}
		 ldmia r0!, {r4-r11}
		 stmia r1!, {r4-r11}

		 ldmfd   sp!,{r4-r11} 		 @ Pop R4, R5, R6, R7, R8, R9, R10, R11
		 bx      lr

So just use DMA.  I suppose that's what it's there for.

----- Original Message -----
From: VBRunner
It's legal, you probably just have to say somewhere that "Gameboy Advance" is
trademarked by Nintendo etc.


Probably so, but if you really want to know the legalities, you need to consult
a lawyer and not a group of developers.  Don't hold what we say as gospel.

-DP

> I think that my name is George Andre... :)

Oops, that's what comes of doing 100 hour weeks.  Many, many apologies.

> Just out of curiosity.. and to make sure... you're sending number
> of halfwords to the DMaCopy routine?

Nope, number of bytes sent to DmaCopy as that's what it takes.  Well, if I
try passing number of halfwords I get half an image.

I usually use DMA to transfer sprite animation data to VRAM and create the
values using:

(w * h) * (NumCols == 16 ? 32 : 64)

Where w and h are sprite character widths and heights.  Unless it's time for
me to quit the industry that's a byte size.

Maybe as this postings going on a bit we should take it on via email and let
everyone know of our findings i.e. when I come back with my tail hung
sullenly between my legs :)

FGL

> It's a lot quicker if you unroll the loop and use a pointer to access vram:
>
>
> screenbuff_temp = screenbufferpointer
> picbuff_temp = picbuffpointer
>
> For(y=0;y<160;y++){
>     For(x=0;x<30;x++){
>
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>         screenbuff_temp++ = picbuff_temp++
>
>     }
> }
>
>

it will be even faster if you use indices instead of pointers
and if you really want some speed you write it in ASM using
ldm/stm instructions :)

Actually I think someone else posted the figures mentioned bellow. I simply
pointed out that the copy doesn't need to complete during vblank for a
non-flickering image replacement/animation. I must agree that it sounds weird
that the Nintendo macro should be broken. On the other hand the macro can be
used wrong I suppose, since you state the width of the transfer. If you use 16
bit width between two types of memmory where you could use 32 bit width I
suppose the transfer will be slower than needed.

Emil
   ----- Original Message -----
   From: Francis Lillie
   To: gbadev@yahoogroups.com
   Sent: Wednesday, February 27, 2002 10:47 PM
   Subject: RE: Re[2]: [gbadev] Mode 3 dma


   Oh, basically as an addition, with the screen display turned off it takes 94
   scan lines to copy a 240*160*2 screen for me and 47 to do a 240*160*1
   screen.  It might just be coincidence, but 47 scan lines is approximately
   69% of a VBlank, the same figure (68.7%) as Emil Arkman quoted for a
   240*160*2 screen.

   Basically, Emil's calculations make DMA twice as fast as I have it.  I hope
   he's right and I'm wrong.  That would make me dance a jig, and those who
   know me would know how funny that would be :)

   FGL

> Oh, basically as an addition, with the screen display turned off
> it takes 94
> scan lines to copy a 240*160*2 screen for me and 47 to do a 240*160*1
> screen.  It might just be coincidence, but 47 scan lines is approximately
> 69% of a VBlank, the same figure (68.7%) as Emil Arkman quoted for a
> 240*160*2 screen.

I think that my name is George Andre... :)


> Basically, Emil's calculations make DMA twice as fast as I have
> it.  I hope
> he's right and I'm wrong.  That would make me dance a jig, and those who
> know me would know how funny that would be :)
>

Just out of curiosity.. and to make sure... you're sending number
of halfwords to the DMaCopy routine?

Aelius

In which case, Nintendo's DmaCopy() is piss poor slow if it takes 102 scan
lines to copy the same data you can do in approximately 47 scan lines.
Where have Nintendo gone wrong?  Or maybe my timings up shit creek :)

I did say I didn't play around with DMA register stuff myself.  Never really
looked at the timings as this project's been a rush from the start (as those
who know who I'm working for will vouch :) ).  This comes as a shock, that
the Nintendo stuff is SOOO slow.

Good job this forum is here :)