Rotation.
For the Natami this can be a hardware feature with a abstraction.
Why the abstraction layer because this is going to be part of the 3Dcore eventually this doesn't mean another unit can't handle this untill then.

Updating the classic AGA API to fit the AROS needs this means mostly a intergration of the SDL API into the OS API and using a clean room approch.

sorry for the typo's

(Hardware scaling, might be, for example interesting for video playback. With some engineering ingenuity, it might be possible to fit that into the blitter similar to the line mode since it is a linear transformation that can be solved by a variant of the Bresenham algorithm).

So long,
Thomas

Some use cases for LQ/HQ scaling and rotating:
----------------------------------------------
o) Video playback (like ThoR pointed out).
o) Fast HQ font and image rendering for word processing.
o) Quickly change sound data pitch for audio processing.
o) BOB manipulations for games (and demos).

That's what first comes to mind.

Requirements:
-------------
o) Nearest-neighbor approach for LQ (indexed) mode, for best
    quality.
    o) Bilinear remapping in indexed modes to CLUT entries
      would otherwise slow down the whole thing and deliver ugly
      end results on average (probably).
    o) Interleaved planar bitmap format for fast internal P2C
      conversion On-the-fly to get the CLUT indexes.
o) Bilinear approach for HQ (chunky) mode.

I have a 1792x64 pixel animation (30 frames of 32-bit color) of Dragon Breath that I need to 2D rotate so that it goes across the screen at the correct angle. 

I have been thinking about doing that.... but there are many of these breath weapon type fx.  Some of them are 240 frames @60fps and I would have to prerotate all these thousands of frames * all 75000 possibilities.  So that is around 400 Million frames.  At an average filesize of 50K per frame that works out to 20 GB of storage.

> 20 GB of storage is cheap these days.

Yes it is.  I agree 100%.  But Aminet will crash and burn when I put a 40GB game on it. :)

About 8 way til scrolling interresting i must read up on it but as long as it doesn't cause visual artifacts it's great. ;)

Ok, to be serious: There is still a lot of things to do here. And no, I don't consider such a unit essential. If your work requires one, I believe a PC will do just fine as OpenGL texture rendering will do it basically for free. The art of programming: Doing the impossible with as little as you have.

To give you something I consider essential: The line mode. It is really "bare basics" the Amiga had, and these should be working as otherwise existing software simply doesn't work. The ability to connect a modern LCD to it is also essential, as otherwise you don't see anything. The ability to play sound, and the ability to connect a HD that you don't have to steal from a museum, also the ability to connect a mouse and a keyboard.

Everything else is, at best, "extra".

Absolutely! We need the foundation in place before building the roof.  I see the Natami not as a machine requiring ultimate speed but as a new Amiga that can run Amiga software. Additionally, with it's improved spec able to go beyond the original machines.  Expectations are very high, and it's good that people look ahead, but I think it's even better that the Natami team are grounded in reality and undertaking the project in a very professional, methodical way.  I like the idea of a new Amiga machine being professionally developed and that will be released complete, not suffering from commercial dead-lines and short-cuts.  Let the extra features and improvements come after we have a solid working Natami.

Now people back to topic!

What is needed in the API?

Guess what - the latency between your eyes and your consciousness is a lot larger. Or to put it differently, reality as you observe it is a nice fake by your brain, you're necessarely behind by many milliseconds. So don't be so harsh to computers, please.

As far as the API is concerned - there are probably many things wrong with the AmigaOs API, but it isn't terribly bad either. And you can certainly extend it by third party libraries like SDL, for example.

I propose a horizontal API extension to what the OS currently has.
This means the API itself could commandeer the hardware away from the OS like hitting the hardware directly.(not passing through OS calls when it's not needed)
We could add a message port to the API so that AROS can intergrate it as a parallel extension and thus keeping what is good.(relative low latency)
And allowing modernizing of the AOS concept without adding to abstraction.(which would add latency)

Oh Vertical extension would be building ontop of the OS like any normal program would do.(OS friendly)

Similar to a Framework, easy access to functions without being buried in opening thousands of libraries.

1,2,3 - three lines to get access to function for gfx, menu, network etc.

Ha ha, but when I move my head, and I know I moved it, because I see different part of room, it feels pretty snappy, from my desire to seeing other wall :D. It is really fun to imagine that latency, so I've just measured my reaction time, from visual signal to actuating (mouse click, and i have cumbersome mouse buttons); its 201 ms best after 5 tries, and tested population average is reported as about 215 ms. Well, so how much latency there is between my eyes and my consciousness?

- there was some latency in trigger event (screen goes green); don't know how much, but maybe even 16 ms (screen refresh rate). + multitasking, potential page faults, etc.
- there was some latency in timer stopping, after my reaction; don't know how much, but maybe even 8 ms (mouse polling rate, 125 Hz?) + multitasking, potential page faults, etc.

If you consider also that actuating (pressing mouse button) is magnitudes slower than information processing... maybe we are not that much outdated in our minds? Be it 2 or 3 ms, it is really good real-time system :-).

BTW, "real-time system" in my language is defined as "system with rigorous timing requirements", not necessarily 0ms latency requirement :) So yes, we have real-time consciousness even if it is few (but I would not say "many") miliseconds behind :-).

Under normal circumstances my brain runs between 200 fps and 1000 fps.  When it is running at 1000 fps I only have a 1 ms delay at "reality perception".

And I can easily press and release the mouse button 5 times in 1 second.

WTH?!  I just pressed my shift key 5 times in 1 second and my firewall popped up and said "Replacing Application Windows NT High Contrast Application"  What on Earth is that about?  I am on Windoze XP.

We're not gonna have a system that will do 200 frames a second for the simple reason i don't see a team member making the monitor file for it.