Mplayer Altivec test version for G4 and PA6T-1682M

@samo79

yes Samo you're right, if we could have a general version LiveForIt will get more testers, i think we will not get a lot of speed improvement from composition version but better than nothing... and however we can resize the video and lose only few details, sdl mode will be slower than composition specially when you try to enlarge the window... i noted on my Sam that use of DMA improve a lot graphics performance and responsiveness

@Liveforit

do not misunderstand, we radeonHD users want composition mode, i'm looking forward to try it, thanks for your work

I have tested a little bit.
Compositing is the fastest here .
Videos until 1280x640 are working well and smooth
Videos from 1280x720 or higher begins to stutter and video/soundouput are not 100 percent synchron.

@tommysammy
Use skiploop filter in the last MPlayer-GUI or "-lavdopts fast:skiploopfilter=all" and synchronization will be good.

Hemmm.... think I fixed the last bug..

It's just aspect ratio and stuff like that is missing in full screen and its all done

@LiveForIt

Quote:

I was just reading some MP4 movies. I could easily reproduce the problem while navigating in the movie (with arrow keys).

@sofistisoftware

Quote:

I do not think it's nice from you to tell this sort of thing. LiveForIt has been the only person trying to improve things and using compositing for those without overlay (with the help of some others).

Let's wait for the final version before making assumptions.

Quote:

What details are you loosing ?

Quote:

It is, of course. And video will be transparent if you are using composition on your WB. SDL doesn't make use of any hardware acceleration (except 2D speedups of the Radeon HD driver).

Quote:

What DMA ?

DMA for hardrives is activated on :

AmigaOnes G3 and G4 with IDE and SATA drives
Pegasos II with IDE and SATA drives
Sam440 with internal SATA drives (sii3114)
Sam460 with SATA (sii3512/3112/3114) cards
AmigaONE X1000 with internal SATA drives

DMA for graphic cards is activated on :

AmigaOne G3 and G4 with either PCI or AGP Cards
Sam440ep with Radeon M9,
Sam440Flex with Radeon 92xx on PCI bus
Sam460 with Radeon 92xx on PCI Bus or RadeonHD on PCI-Express
AmigaONE X1000 with RadeonHD on PCI-Express or Radeon 92xx with PCI<->PCI E adapter.

En résumé, the only way to have no DMA on a graphic card is to use one of the internal PCI slots of an AmigaONE X1000 with a Radeon 92xx attached on it.

@K-L

This is mixed result, for AmigaONE-X1000 users this is major speed up, the harry potter test video, was maxing out the CPU before I started working on it on the X1000.

The speed comes from combination of things.

Video decoding is done by AltiVec (FFMPEG).

Using the correct video format removed the need for conversion before sending it to graphic card.
(This only effects the AmigaONE-X1000, not AmigaONE-XE/Pegasus II)

And then you have hardware overlay that accelerate video scaling.

For now I think this is as good as it gets, whit out a video decoder chip.

https://www.broadcom.com/products/Cabl ... raphics-System-Processors

I don't know how well it will work for Sam460 users, but given that they do not have AltiVec and I'm guessing its only going to minor improvement.

AmigaONE-X1000, handles 720p videos up to 3mbps.

@LiveForIt

Thanks for your explanations.

During Amigateries event last week-end (soon the pictures), I've made some tests with Elwood's Sam460 and diffrerent MPlayer versions (without AltiVec) and here are the results :

Sam460 RadeonHD 3450

MUI-Mplayer CGX_WPA Beta 5:

BENCHMARKs: VC: 27.541s VO: 19.036s A: 0.000s Sys: 0.380s = 46.957s
BENCHMARK%: VC: 58.6511% VO: 40.5387% A: 0.0000% Sys: 0.8102% = 100.0000%

MPlayer MickJT SDL:

BENCHMARKs: VC: 46.107s VO: 12.708s A: 0.000s Sys: 2.627s = 61.442s
BENCHMARK%: VC: 75.0418% VO: 20.6828% A: 0.0000% Sys: 4.2754% = 100.0000%

MPlayer Andrea SDL :

BENCHMARKs: VC: 51.982s VO: 13.677s A: 0.000s Sys: 2.889s = 68.548s
BENCHMARK%: VC: 75.8328% VO: 19.9531% A: 0.0000% Sys: 4.2141% = 100.0000%

It seems that on Sam460, CGX-WPA is the fastest video output.

We'll have to test with Compositing when your version of MPLayer is compiled without AltiVec or when MUI-Mplayer is compiled with compositing video output .

Anyway, thanks for this excellent work !

@LiveForIt

Would be possible to have a special optimized version for the sam 440 and 460. Both processors have 24 DSP instructions that can speedup some operations (up to 50% in certain case). Documentation is available at IBM web site.

Since i read that this is an Altivec test version, maybe a version for the sam's dsp could be compiled too.

@AmigaBlitter

As far as I know, Corto tried to use these instructions to speed up his softwares son Sam440/460 (DosBox amongst others) but did not acheived any speed improvmenents

Does LiveForIt own a Sam4x0 system by the way ?

@K-L

don't know... But someone could try

https://www-01.ibm.com/chips/techlib/t ... PowerPC_440_Embedded_Core

https://www-01.ibm.com/chips/techlib/t ... Optimized_dsp_440_app.pdf

https://www-01.ibm.com/chips/techlib/t ... le/440MacProcessing00.pdf

@K-L

Some tests on each beta with Big Buck Bunny 720p MP4 from here http://www.bigbuckbunny.org/index.php/download/
Video starts windowed by default on my 6850 unless stated

-vo cgx_wpa
-ao sdl

beta1 75-94%
beta2 70-95%
beta3 20-43%
----------------------------------
-vo sdl
-ao sdl

beta1 50-59% starts fullscreen on my 9250
beta2 30-41% starts fullscreen on my 9250
beta3 34-58%
----------------------------------
-vo comp
-ao sdl

beta1 No Video
beta2 35-42%
beta3 35-41%

A couple of beta3 issues with -vo comp.

The diagonal lines occur only on some videos. Dont know why.resizing window, switching to fullscreen and back to window fixes it.


The video image stays on top of the active window.

@ddni

Big Buck Bunny is too big (and too long) to realize some benchmarks. I'm using one video (an FLV trailer of man Of Steel) to make my tests

@K-L

I should have said, I only assessed the first 30 secs of the video.

@K-L

Quote:

Sorry, but that is incorrect. Use of DMA to transfer to/from VRAM is currently only available on Sam4x0 machines. This should change at some point in the future.

Hans

@Hans

Thanks for correcting me. Maybe I did not quite understand how DMA works for graphic cards.

I'll study a little bit more ths specific topic.

Nothing too careful or scientific about my methods, I don’t guess… but here are my findings. :)

Tonight I downloaded the beta3 version of liveforit’s build and played some movie files in resolutions and formats that heretofore brought my X1000 to its knees under AmigaOS. Things are getting better!

These were all with composition selected as output.

A 1280x544 mkv file in h.264 (the movie “Peyton Place”) played back fairly smoothly with a few stutters but sound stayed in sync.

A 1280x694 mkv file in h.264 (the movie “The Paper”) played back smoothly and sound stayed even — did not gain or lose. But sync seemed to lag … lips did not QUITE match but close. It got no worse out of sync and might not have bothered someone unless he was as picky as me.

A 960x720 mkv file labeled as x264 played smoothly but the sound was terribly out of sync. This was the 1950 movie “Union Station” and so was not a widescreen aspect ratio, but source was Blu-ray rip and so it used the full 720 resolution in height. This was not at all watchable because of the audio sync.

I look forward to testing some more! I think this is showing a lot of promise. I never saw my X1000 play back high-def video as well as the first two examples here without having booted into Linux. I hope you guys don’t hit the wall too soon. :)


ADDENDUM: I played a part of another old movie, "Johnny Guitar," at 960x720 and playback was smooth and sound perfectly in sync. That made me doubt my observation about the "Union Station" playback so I retested that one -- still out of sync. Must be something about that file.

@mbrantley

The loopfilter is rally CPU heavy, what it does is some error correction you don't need, or some thing, by turning off loopfilter the movies players better.

mplayer -lavdopts fast:skiploopfilter=all filnavn.avi

if you don't like to type it very time you can add defaults to progdir:conf/config file.

As thing basically done, I strip the exe, compile a none AltiVec version stick it in .lha file and upload it to OS4Dpoet.net tomorrow or latter today.

While people wait for that to happen, try not re-size the window, and press F, that can trigger some nasty bugs and DSI errors.

This will be version 1 from me.

Next will be:
* investigate the mystery CPU too slow message from AHI DEV.
* The aspect correction in full screen.
* Fix full screen and pause for pip.

and to get my changes into the repo, this will take time.

@LiveForIt

And did you find what causes the GrimReaper I posted on G4 ?

Another question : is it possible to avoid tearing on every video outputs ?

And a final question : will v.1 be compiled with the cache option ?

Thanks !

@ live4it. Does the newly uploaded GUI have any performance impact or compatibility issues with your mplayer beta?

http://os4depot.net/?function=showfil ... ideo/misc/mplayer-gui.lha