RiVA v0.52 Released

[AMIGASYSTEM]

RiVA-0.52: This is the latest update to the fastest MPEG player available on the classic 68k Amiga platform.
Two binaries are provided in this archive, a classic m68k compatible build (68040+) and an AMMX build, exclusive to Apollo Core Gold 2 and newer.

New in this version::

- AMMX usage in motion compensation, residual transform, color conversion
- multibuffered SAGA YUV video output
- Stereo sound output
- audio/video sync
- automatic enabling of 44/48 kHz sound output when P96 with 31 kHz AmigaVideo
environment variable and ECS denise are present
- relaxation of internal data paths, avoidance of unnecessary data copies
- 16 Bit windowed mode quality improvement, along with ordered dithering
- borderless windowed mode (Flype)


RiVA-0.52: Download

[groundplayer]

Everything is working fine here, except with old version (0.50) PIP mode works, but not in new, shows only one colored screen or grey, if windowed (not zoomed PIP). My config is: Mediator TX, Phase5 060 (256mt), Voodoo 3000PCI, OS3.9 BB2.Screenmode 800x600 16bitPC (picasso96). Thankyou for your release !.

edit: Solved problem by changing screenmode to 800x600 16bit not PC mode. Now everything works ok, thankyou again !

[matthey]

This RiVA code is more than a little bit suspect.

CreatePalette	lea	Palette,a0
		move.w	#256,(a0)+			;Number of colours to load
		move.w	#0,(a0)+			;Register number to start from
		move.l	#256-1,d7
.colsloop	move.b	d7,d6
		lsl.l	#8,d6
		move.b	d7,d6
		lsl.l	#8,d6
		move.b	d7,d6
		lsl.l	#8,d6
		move.b	d7,d6
		lsl.l	#8,d6
		not.l	d6
		move.l	d6,(a0)+
		move.l	d6,(a0)+
		move.l	d6,(a0)+
		dbf	d7,.colsloop
		move.l	#0,(a0)				;Close Structure
		rts

Did somebody get carried away with the shifting?

CreatePalette	lea	Palette,a0
		move.l  #256,d7
		move.w	d7,(a0)+			;Number of colours to load
		subq.l  #1,d7				;256-1
		clr.w	(a0)+				;Register number to start from
.colsloop	move.b	d7,d6
		lsl.l	#8,d6
		move.b	d7,d6
		lsl.l	#8,d6
		move.b	d7,d6
		lsl.l	#8,d6
		not.l	d6
		move.l	d6,(a0)+
		move.l	d6,(a0)+
		move.l	d6,(a0)+
		dbf	d7,.colsloop
		clr.l	(a0)				;Close Structure
		rts

Shaved a few cycles and 12 bytes. Even vasm would have saved 6 bytes here if optimizations were on. At least compilers are smart enough to turn on optimizations .

[Don_Adan]

CreatePalette	lea	Palette,a0
		move.w	#256,(a0)+			;Number of colours to load
		clr.w	(a0)+				;Register number to start from
                moveq  #-1,D6
                move.l   #$01010101,D7
.colsloop
		not.l	d6
		move.l	d6,(a0)+
		move.l	d6,(a0)+
		move.l	d6,(a0)+
                not.l d6
                sub.l D7,D6
		bcc.b .colsloop
		clr.l	(a0)				;Close Structure
		rts

Maybe it can works, but I dont checked this.

[buggs]

Originally Posted by matthey

This RiVA code is more than a little bit suspect.

<snip>

Shaved a few cycles and 12 bytes. Even vasm would have saved 6 bytes here if optimizations were on. At least compilers are smart enough to turn on optimizations .

If you want to save cycles that matter, look into the color conversion routines, for a start. Those which I didn't touch with AMMX code bring heavy penalties in terms stalls due to the register indexed addressing (on 060).

[matthey]

Originally Posted by Don_Adan

CreatePalette    lea    Palette,a0
        move.w    #256,(a0)+            ;Number of colours to load
        clr.w    (a0)+                ;Register number to start from
                moveq  #-1,D6
                move.l   #$01010101,D7
.colsloop
        not.l    d6
        move.l    d6,(a0)+
        move.l    d6,(a0)+
        move.l    d6,(a0)+
                not.l d6
                sub.l D7,D6
        bcc.b .colsloop
        clr.l    (a0)                ;Close Structure
        rts

Maybe it can works, but I dont checked this.

Something similar to your code may work (poor mans SIMD). It would be nice to get rid of the original dependencies for superscalar operation but my intention was to fix obvious (to me at least) flaws.

[ShK]

Are these code samples from "Create Palette Structure before opening the screen"?

Where is actual work loop?

[grond]

It's in fact "create a greyscale palette for AGA and poor people that haven't got a Vampire (and thus can't just watch full-color hires movies) code". Legacy stuff that buggs didn't even touch.

[cmsj]

I would like to say thank you to bax/flype/etc. for doing some awesome work on RiVA for AMMX users. It's super impressive what a vampire can do with video now.

[ptyerman]

@cmsj
Right on!

[grelbfarlk]

Great work, can we get a WarpOS version?

[meynaf]

Originally Posted by grelbfarlk

Great work, can we get a WarpOS version?

This is (mostly) 68k code so i doubt it.

[flype]

RiVA 0.52 is based on RiVA 0.50 where Stephen Fellner, years ago, wrote what is still the fastest MPEG1 player for MC680x0. It was written 100% in assembly, has been re rewritten many times at that time, improving it by adding features years after years, for example Picasso IV support. So, of course, some of the code ends on some not super clean code. Well, i know the code in RiVA, it is not that bad. I don't understand such criticismes. At end RiVA 0.50 works as intended and is fast. And this has no relation with the Apollo-Team, with the Vampire. The code that is spotted by Matthey is a detail in the original 0.50 code. We are speaking of a detail. Now regarding the 0.52 re work, also i know the code inside, and i can't let people speculates on "unreadable, unmaintainable, targeted at archaic data types, requiring system patches, exceedingly long, filthy, full of ugly hacks - doesn't matter as long as it executes fast. This isn't the 68k spirit at all - but who cares.". Seriously, this is pure speculation and (not the first time) pure bad faith ; the rework *is* clean, the code *is* maintainable and readable. That is really tiring to be always critized when something comes from the Apollo-Team guys. I am Apollo-Team guy and i do not recognize myself in your description. I like Amiga, i like AmigaOS, i like 68k, i like clean code, we all.

@ flype :
totaly agree. I was beta-testing RIVA on the vampire in the developpement process, and I clearly saw the improvements at each iteration. Same hardware used each time. So, the code was optimised, and delivered what some people claimed to be impossible. Now that RIVA is there and show that's Apollo is much faster than a real 060, Mr. Meynarf complain about the code itself (but don't show his promised optimised and as fast as Apollo one) and Mr. Thoram said "who care" that apollo is faster ? Then in this case, don't stay on Amiga, it is clear that anything else that you can buy now in a shop is way faster than that, but this is not the point. The point is to have the best accelerator for our Amiga. It is so hard to understand ?

[wawa]

the rework *is* clean, the code *is* maintainable and readable

isnt that open for everybody to see?

[flype]

It is available since it is released.

http://aminet.net/package/gfx/show/RiVA-0.52

[wXR]

Hi there, where can I find the RiVA source repo?

[flype]

Originally Posted by wXR

Hi there, where can I find the RiVA source repo?

In first post. On the aminet link.


To stay in topic, Point is to offer to current existing Vampire owners a solution to watch movie at 24fps. They were not able to, at decent resolution at least. Fact is they *can* now, if they likes to. How it is done do not hurts other machines owners.

For non vampire users, there is something that scores/benchs don't shows :
RiVA-0.52-apollo offer a much better visual quality and audio quality than 0.50 and 0.52-m68k.


People that are really interested can start reading this AMMX introduction, many already did :

http://forum.apollo-accelerators.com....php?f=10&t=25

After your reading, if you don't like, don't use. That's all, that's simple, that's clear enough.

[meynaf]

Originally Posted by flype

@meynaf

People that are really interested can start reading this AMMX introduction :

http://forum.apollo-accelerators.com....php?f=10&t=25

After your reading, if you don't like, don't use. That's all, that's simple, that's clear enough.

That's better than all i've seen so far, though looking still not exactly uptodate.

Now if all these 64-bit encodings (and new stuff in general) have to be enabled otherwise they trap, then it becomes perfectly fine for me - won't like, won't use, but also won't care at all.

[flype]

Originally Posted by meynaf

That's better than all i've seen so far, though looking still not exactly uptodate.

Ah, ok, December 2016 is what i call up to date in my world. Current RiVA use this specs.