Demos using a GFX mode please !

[darkcoder]

yes exactly. I am curious about diferences. I coded by myself a routine similar to that of Azure, after reading his tutorials on amycoders.
(but only after I did my routine I looked at his source code ;-). My routine seems to be a very little faster than the azure one (at least of the PUBLIC Azure code) on 060 but it's a little bit slower on 040.
But now I am curious about the Aki routine. Is the technique similar or different? Is the source code of Aki public?

[darkcoder]

@psyko
many thanks for your interesting comments about c2p!! I hope not to disturb you too much by asking some clarification

>For 1260 the best thing to do is to preload as much data as possible into the cache and then c2p and write to chipram (just write 4 longwords at a time,


I always tried to put as much c2p instructions as
possible between any two longwords write, i.e.

write 1 long
some instruction
write 1 lone
etc.

You say it's better to do:

write 4 long
some instrucions
write 4 long

??

>For 4060 it's totally different. Here a simple loop like: c2p'ing,read16, write16,c2p'ing, read16,write16 would work pretty well, and should

erm..what do you mean with read16 aand write16?
You mean using the MOVE16 instruction?!
I always thought that such instruction was not supported by Amiga architecture, so never used it!
If this is true shame on me and I think I change my nick to crappycoder!!
Does it works in both chip and fast ram?

>special one blueberry
this short one also does cache preloading?
Can you tell me how much bytes? :-)

I use a c2p who preloads 128 bytes, like Azure tutorial. I have almost the same speed as Azure on 060 but I am a little slower on 040. Do you have any hint?
It you will finish the article on this topic, where can I read it?? :-)))

thanks for you help and congratulations for your
wonderful work!

[darkcoder]

@crumb
I think that the CPU tries to write to the ram when the bus is available. Meanwile it keeps values to be written in the write pending buffer.
If the write pending buffer becomes full, and the bus is not yet free to copy values in ram, then the CPU has to stall.

[darkcoder]

@psyko

>>rm..what do you mean with read16 aand write16?

>just the reading of 16 bytes from fastram.

So you suggest that on 4060 cache preloading is not useful? I have a CSMk3 and my routine works better with cache preloading.

> guess that it will at least be on loonies.dk and in some mag.. Actually I feel like writing it

we all wait for it!!
I'll check on the website!

[darkcoder]

@MagicSN

>than for 68k CPUs (as PPC is a "true CPU family").

I don't agree with this. There are many diferences in PPC.
1)Some of them have Altivec and some not. (is it a small difference??)
2) clock speeds range from less than 300 Mhz (I don't remember exacly, I guess the first 603 where even les than 200Mhz) up to 1.4Ghz. And the 970 is coming..
3) the number of functional units=> the digreee of parallelism is different.

Basically you have one line "faster on Floating Point" and one line "faster on Integer" - sure,

"basically"? this is a simplification, i.e. 90%
(or 95%) optimization...

>of Floating Point units inside and such... but >that won't change much the way you code. After  

if you search for 100% optimization they change the way you code. have you read the rest of this thread??
We were discussing having 2 060 c2p: one optimized for Blizzard or Apollo 1260 and onother for 4060 like CSMK2 which behaves differently. That is what i mean, I even have interest in optimizing for a specific board, not just for a specific CPU!


>they are all damned fast CPUs, and limiting
apart from the fact that you sound really unpolite saying that other people are stupid, as I explained in previous messages,being limited by the hardware is EXACTLY what I am searching!

>difference, if it is then this or that chip...

you don't have to trust too much on benchmarks, but if they show 6Mb/sec difference between CV3D and Picasso IV.....THERE IS a  DIFFERENE.

your points were clear, I wonder whether you read what other people answer to you. I don't think so.
Anyway I repeat what I already said (last time):

a) There exists scandoublers. If someone has a real interest in demos, he will have a hardware to watch AGA demos, since there are so many. If not, I don't care. If nobody but me watch my demos, it's sad, but I will still have fun coding!

b) We all agree that RTG is faster than AGA. For me AGA is fast enough the do interesting stuff AND I can use COPPER + SPRITES + DMA control !
For me such things are much important than speed.

c) AmigaOne...ahaha!

Please, MagicSN, answer the following: what's so special with RTG? You can do the same things on a PC. So why use RTG instead of a PC?

[darkcoder]

@crumb & blubble

Indeed this a very interesting thread.

>First, I think Steffen is a bit arrogant/ignorant..

What I find irritating is that it seems he doesn't read what other people write. He simple ties to impose it's ideas (who come from a different contex, that of games/application development) like they're God's words.
What makes threads intersting like this is instead reading and thinking about other ideas.
And in fact I thought about what both of you wrote about PPC cpus and I agree that you are right and I was wrong. I also did superficial and inexact calculation about CPU speeds.
Ok, I apologize, now I have no preclusions in doing PPC coding. The only thing I don't like is this fact of instruciotn reordering. Of course you can schedule inst, but at least from what I read about the G4 the CPU reorders them internally. I red these informations on the ArsTechnica web site. There in non-PPC specific articles they say that this technique, together with the increased number of registers of "Post-RISC" CPU (they mean P4, K6, G4, SPaRC, K7 etc) helps a lot compilers so that it is nearly impossiblre to improve on them. But this was not a PPC specific issue, so if you have direct expirience I trust you.
So if it will be true that PS3 will use a PPC, I would really like to code demos on it.
Of cource I still am in favor of "fixed architecture" and also what blubble wote about it is important: a fixed hardware is more fascinating and inspiring to me (but there are also those technical reason I explained in previous post).

[darkcoder]

@psycho

so thank you very much for this veery interesting numbers. So with 4060 cache preloading is not faster. I hope I will obtain similar results by myself.
How do you use the mmu to improve speed? You detect which part of the buffer have been modified and you only c2p the modified parts?

(oh if you don't have time to answer to all these questions no problem- I still thank you for all the interesting hint!)

[darkcoder]

@lempkee

I love th Amiga (classic) for both its hardware and OS.
I really hope and think that OS4 will be a great development of a still great OS. But the A1, from a hardware perspective, to be honest it doesn't seem so wonderful. It's not bad, if it was out 3 years ago it would have been great but today has no great performances and costs a lot. (BTW just today Apple announced a big cut of their prices..)
From the hardware perspective I also prefer Pegasos to A1.
So I think in the future I will use OS4 for "serious" stuff and a different system (XBox? PS3?GBA?) for coding.
(yeah Amiga is by far my favorite computer, but it's not the only one I use)

[darkcoder]

I am really not interested in A1/Peggy stuff, but to me Peggy is more elegant mtherboard design (is also smaller) it should have a Firewire interface and it seems to cost less. Up to now there is no OS4 for A1 too. But as you know I am interested in AGA, so...

I still think you don't read messages. At least not carefully.

>well, I tried several times to make you aware
                                   that not all people CAN use AGA. And that Systems like the AmigaOne CANNOT run AGA Demos also (while RTG Demos could still be compatible)

You may not belive me, but I am really a coder. So I know by myself that AGA stuff won't run on A1.
(well there is UAE, maybe in future it will support AGA). And in a PREVIOUS MESSAGE that maybe you didn't read I explained why I don't see this as a problem (to me). So it's YOU that says again and again the same things (which I already know by myself).
If you read carefully the post you will notice that as long as the discussion went on, I changed a bit my ideas. I discussed with Crumb, Carls, lempkee and many others I forgot (please excuse me) and we all found the discussion interesting and I think many of us learned something.
I never said anyone shoud optimize for various chipset. (please read carefully, as every other in this post did)
I am still waiting for a response about a question that I directed to you in one of my previous post (do you read my posts?).
There is not much to answare about slow speed of AGA with 18 bit HAM8 mode or hires/superhires displays. IT IS slow, I know. But as I explained (oh, yes...you don't read) a limited speed is what I need for having fun.
You say (again and again) that for you small differences in speed are not important and that you think it is more important that every Amiga user can watch a demo. Ok I UNDERSTAND YOUR OPINION.
And I say (again and again because you don't read) that for me small differences ARE important and is less important that A1 user see demos. I gave arguments for this. Hope now you understand. The others in this post (for example Crumb) perfectly understood my opinion.

Not offended, hopy you also did not offended by me. It seems we have difficulties in understanding each others. Things that happens. Friendship Rulez! :-)

The Dark Coder

PS. I prefer to continue the discussion in public (really no time to follow ALSO a private thread.. :-)

[darkcoder]

@crumb

the truecolor-18bit HAM8 mode is a SHRES screen used to simulate a LORES one. SHRES+8bitplane uses all AGA bandwidth so during displays the CPU cannot write to the mem. And max res. with this technique is 320x512 in PAL.
Unless you mean a more clever scheme (in which I would be very interested) to use "real" LORES
HAM.

best regards

[darkcoder]

@Agony

 >differs from not working . you can use MOVE16 in >chipram with a sliiiiiiiiiiight better speed, but

that's veeery interesting and cool thing!
I have to try. It's a pity for MOVE16 only workin with mem to mem.
Thanks you for the explanation, Agony, also about 18bit HAM8.

@crumb
the usual HAM8 is a planar mode. The 18bit HAM8 is a technique to do a sort of truecolor chunky mode.
(yep a fake one, as Agony says)

[darkcoder]

with ham8 you can change only one component of the previous pixel. So if you have two pixels close to each other with 2 different components you are in trouble. Es
$ff00ff  $ff8888. You can use a "base color" for the second pixel, i.e. not define the second as a difference from the first one. But you only have 64 base colors, so you can do this only 63 times (the first color of the display have to be a base, I think (well, never tried a HAM display starting with a "changed" color...don'tknow what will happen)). Well, of course you can use copper to change the value of base colors.
But then you need a dynamic copperlist which is very complicated to build. And since copper is "8-pixel slow" I can provide easily an example of rasterline which is imposible to exactly reproduce in HAM8: (all values in hex 24bit)
00000 010101 020202 .... ffffff
It is only 255 pixel long, but you cannot obtain any pixel from the previous one. So the first 64 pixel you use all the 64 base colors. During that time you can only reload 8 color registers, so you have enough till color 64+8. During the extra 8 pixel you gained, you can even reload another color register and gain 1 extra pixel. But then it's over!
You can do an approximate conversion to ham8, i.e. changhing color a bit. Ex say you have
008800 00cc33 here 2 components change, you need to use a new base color. But if you transform the 2 pixels in 008800 00cc00 you change a bit the picture but you don't need a base color. I am not expert of color conversion algorithms, but I suspect as Agony says that they are not very fast. You have to analyse all the color used in the picture, make calculation, divisions..
anyway I'll give a look at Aki's work! maybe there is some possibility I don't see.

[darkcoder]

@magicSN

Do you mean that I can add a Firewire interface to an A1 *for free*?? :-D
Os 4 is included in the price but...they don't give it to you because it is no ready! I don't like to pay today for something I will receive in ?? months. And note that the OS4 is included *only* in the earlybird *special* offer.

I think that my best time is the one which i spend enjoying myself. Since I enjoy coding AGA, for me it's not waste of time. Your posting would sound much better if you add "in my opinion" somewhere in between. Your words sounds like orders.
I already explained that for me and my work AGA has many advantages so nothing more to say..