Mplayer Altivec test version for G4 and PA6T-1682M

@K-L and jabirulo

One thing I have noticed in general is VC score is better in MUI Player always.
I think its time for me to check my compiler optimization flags, it looks like I'm losing out.

It also useful to see that makes a different on Sam460 as it do on X1000,

The results from AmigaONE-XE, I'm sorry to say is expected, it some of the changes I made to cgx_wpa to improve speed on X1000/PCIe, has have the opposite effect on computer whit PCI/AGP.

I'm thinking about checking the CPU, to swap out some code, so it does not suck on Sam440/G3/G4.

@tommysammy

Thank you.

@LiveForIt
How you going with goolge-code commits ? If do not want to worry about for now, maybe you can upload your vo_comp.c so i can add it to muimpayer as well ?

(yesterday was in hope to commit p96-white-strip fix, but run out of time, hope to do so today).

@tommysammy

How can i get german speech?

when you type -h after mplayer you get information about some of the stuff you can do, whit mplayer.

There is a option called -aid i think this the one your looking for.

@kas1e

Yes I should start committing some stuff.

@LiveForIt
Probably we need call diffrent 2 cgx_wpa drivers, my imho is:
vo_cgx_wpa_rgb.c
vo_cgx_wpa_argb.c

and their names then "cgx_wpa_rgb" , "cgx_wpa_argb". As it clear now that on some configs cgx_wpa with your changes is faster, on another original one.

@kas1e

No need to it can be handled in the vo_vgx_wpa, I think it should be.

@LiveForIt
If you can make it works like this then good. But probably it will be mess if you will make split in code based on cpu detection. Cpu all different, there is many of them, and new always come.

Probably better to do checking somehow else to detect when and how and which routines use (imho).

@LiveForIt


Quote:

I'll try to ask for benchmarks with my MPlayer benchmark tests.

Quote:

Never mind since most of the AmigaOne XE G4 do use overlay.

Quote:.

But on Sam440ep o/c at 733 Mga, results are slightly different than with those on AmigaOne XE G4 (thanks to AltiVec ?).

Here are the results:

Sam440ep o/c 733 Mhz and Radeon M9, from the fastest to the slowest:

MUI-Mplayer NON AltiVec OS4Depot 2011 Release p96_pip:

BENCHMARKs: VC: 53.096s VO: 17.799s A: 0.000s Sys: 0.396s = 71.291s
BENCHMARK%: VC: 74.4777% VO: 24.9673% A: 0.0000% Sys: 0.5550% = 100.0000%

MUI-MPlayer **Beta6** NON AltiVec p96_pip:

BENCHMARKs: VC: 55.585s VO: 17.787s A: 0.000s Sys: 0.522s = 73.895s
BENCHMARK%: VC: 75.2218% VO: 24.0712% A: 0.0000% Sys: 0.7070% = 100.0000%

LiveForIt-Mplayer NON AltiVec p96_pip:

BENCHMARKs: VC: 62.005s VO: 17.776s A: 0.000s Sys: 1.932s = 81.713s
BENCHMARK%: VC: 75.8810% VO: 21.7541% A: 0.0000% Sys: 2.3649% = 100.0000%

MUI-Mplayer **Beta6** NON AltIiVec SDL:

BENCHMARKs: VC: 54.330s VO: 36.589s A: 0.000s Sys: 0.880s = 91.798s
BENCHMARK%: VC: 59.1840% VO: 39.8577% A: 0.0000% Sys: 0.9583% = 100.0000%

LiveForIt-Mplayer **Beta6** NON Altivec SDL:

BENCHMARKs: VC: 61.759s VO: 35.906s A: 0.000s Sys: 0.599s = 98.263s
BENCHMARK%: VC: 62.8502% VO: 36.5404% A: 0.0000% Sys: 0.6094% = 100.0000%

LiveForIt-Mplayer NON Altivec compositing:

BENCHMARKs: VC: 62.447s VO: 36.105s A: 0.000s Sys: 0.630s = 99.182s
BENCHMARK%: VC: 62.9619% VO: 36.4029% A: 0.0000% Sys: 0.6352% = 100.0000%

MUI-Mplayer NON AltiVec version cgx_wpa:

BENCHMARKs: VC: 53.385s VO: 46.204s A: 0.000s Sys: 0.452s = 100.041s
BENCHMARK%: VC: 53.3633% VO: 46.1849% A: 0.0000% Sys: 0.4518% = 100.0000%

LiveForIt-Mplayer NON Altivec cgx_wpa:

BENCHMARKs: VC: 62.246s VO: 37.829s A: 0.000s Sys: 0.560s = 100.635s
BENCHMARK%: VC: 61.8533% VO: 37.5899% A: 0.0000% Sys: 0.5568% = 100.0000%

MUI-Mplayer **Beta6** NON AltiVec cgx_wpa:

BENCHMARKs: VC: 55.651s VO: 45.982s A: 0.000s Sys: 0.418s = 102.051s
BENCHMARK%: VC: 54.5321% VO: 45.0582% A: 0.0000% Sys: 0.4097% = 100.0000%

********************************************************************************

Here are the charts :

Sam440ep 733mhz Radeon M9

1° MUI-Mplayer NON AltiVec OS4Depot 2011 Release p96_pip
2° MUI-MPlayer **Beta6** NON AltiVec p96_pip
3° LiveForIt-Mplayer NON AltiVec p96_pip
4° MUI-Mplayer **Beta6** NON AltIiVec SDL
5° LiveForIt-Mplayer **Beta6** NON Altivec SDL
6° LiveForIt-Mplayer NON Altivec compositing
7° MUI-Mplayer NON AltiVec version cgx_wpa
9° MUI-Mplayer **Beta6** NON AltiVec cgx_wpa

AmigaOne XE G4 1,26Ghz Radeon 9000 Pro

1° MUI-MPlayer AltiVec OS4Depot 2011 release p96_pip
2° LiveForIt-Mplayer AltiVec p96_pip
3° MUI-Mplayer NON AltiVec OS4Depot 2011 release p96_pip
4° MUI-Mplayer **Beta6** NON AltiVec p96_pip
5° LiveForIt-Mplayer NON AltiVec p96_pip
6° MUI-Mplayer OS4Depot 2011 release cgx_wpa
7° LiveForIt-Mplayer AltiVec cgx_wpa
8° MUI-MPlayer NON AltiVec OS4Depot 2011 release
9° MUI-Mplayer **Beta6** NON AltiVec cgx_wpa
10° LiveForIt-Mplayer NON AltiVec cgx_wpa
11° LiveForIt-Mplayer AltiVec SDL
12° LiveForIt-Mplayer AltiVec Composititing
13° MUI-Mplayer **Beta6** NON AltiVec SDL
14° LiveForIt-Mplayer NON AltiVec SDL
15° LiveForIt-Mplayer NON AltiVec Compositing

So we can see that on the 440ep, "MUI-Mplayer p96_pip" is way faster than "LiveForIt-Mplayer p96_pip" whereas on AmigaOne XE, there is slightly no difference between both (whether AltiVec is used or not).

On Sam440ep, SDL and compositing are the fastest on a non overlay video output whereas on AmigaOne XE cgx_wpa is the fastest, SDL and Compositing being the slowest (AltiVec helps regarding compositing)on this architecture.

@K-L

Its because Sam4x0 hardware has DMA to aid it, while Pegasus and AmigaONE-XE/SE/Micro does not.

@kas1e

The SDK is not update, can't even see latest CPU models in exectags.h (AMCC460/PA6T)

Maybe your right, maybe its more hazel then its worth.

LiveFotIt : Ok, for the DMA on Sam440 (BTW which video ouput is accelerated by the DMA?

But why is "LiveForIt-Mplayer p96_pip" way slower than 'MUI-MPlayer p69_pip" on Sam440? Since it's the same video output, it should be the same speed, shouldn't it ?

@K-L

Don't know its a mystery, maybe compiler optimize flags, alignment flags, or some thing, different GCC version, different C libs, who knows. You see it in “VC:” column in the benchmark. VO: column is nice.

@LiveForIt

Thanks for all your work and your explanations

Here are the results for an AmigaONE X1000 and RadeonHD 4650:

AmigaONE X1000 and Radeon HD 4650

Results are sorted from the fastest to the slowest


LiveForIt-Mplayer AltiVec cgx_wpa

BENCHMARKs: VC: 6.136s VO: 3.812s A: 0.000s Sys: 0.311s = 10.260s
BENCHMARK%: VC: 59.8084% VO: 37.1585% A: 0.0000% Sys: 3.0331% = 100.0000%

LiveForIt-Mplayer AltiVec compositing:

BENCHMARKs: VC: 6.215s VO: 4.022s A: 0.000s Sys: 0.376s = 10.612s
BENCHMARK%: VC: 58.5613% VO: 37.8961% A: 0.0000% Sys: 3.5426% = 100.0000%

LiveforIt-Mplayer AltiVec SDL:

BENCHMARKs: VC: 6.256s VO: 6.321s A: 0.000s Sys: 0.267s = 12.844s
BENCHMARK%: VC: 48.7080% VO: 49.2132% A: 0.0000% Sys: 2.0788% = 100.0000%

LiveForIt-Mplayer NON Altivec compositing:

BENCHMARKs: VC: 9.010s VO: 5.914s A: 0.000s Sys: 0.314s = 15.238s
BENCHMARK%: VC: 59.1268% VO: 38.8128% A: 0.0000% Sys: 2.0604% = 100.0000%

LiveForIt-Mplayer NON AltiVec cgx_wpa:

BENCHMARKs: VC: 8.897s VO: 5.710s A: 0.000s Sys: 0.249s = 14.856s
BENCHMARK%: VC: 59.8857% VO: 38.4359% A: 0.0000% Sys: 1.6784% = 100.0000%

MUI-Mplayer **Beta6** NON AltiVec SDL:

BENCHMARKs: VC: 9.080s VO: 6.336s A: 0.000s Sys: 0.176s = 15.592s
BENCHMARK%: VC: 58.2346% VO: 40.6344% A: 0.0000% Sys: 1.1310% = 100.0000%

LiveForIt-Mplayer NON AltiVec SDL:

BENCHMARKs: VC: 8.959s VO: 6.336s A: 0.000s Sys: 0.329s = 15.624s
BENCHMARK%: VC: 57.3435% VO: 40.5528% A: 0.0000% Sys: 2.1036% = 100.0000%

MUI-Mplayer AltiVec OS4Depot 2011 release cgx_wpa:

BENCHMARKs: VC: 10.339s VO: 12.053s A: 0.000s Sys: 0.255s = 22.646s
BENCHMARK%: VC: 45.6529% VO: 53.2224% A: 0.0000% Sys: 1.1247% = 100.0000%

MUI-Mplayer **Beta6** NON AltiVec cgx_wpa:

BENCHMARKs: VC: 10.139s VO: 12.166s A: 0.000s Sys: 0.147s = 22.452s
BENCHMARK%: VC: 45.1595% VO: 54.1846% A: 0.0000% Sys: 0.6559% = 100.0000%

MUI-Mplayer NON AltiVec OS4Depot 2011 release cgx_wpa

BENCHMARKs: VC: 13.343s VO: 15.798s A: 0.000s Sys: 0.169s = 29.310s
BENCHMARK%: VC: 45.5235% VO: 53.8995% A: 0.0000% Sys: 0.5770% = 100.0000%

Clearly, LiveForIt's Mplayer port has been heavily optimized for the AmigaONE X1000. And for those using a G4 Amiga-NG, LiveForIt's Mplayer port is also the one to be choosen (the faster with Altivec enabled and p96_pip corrected).

But if you use a Sam440 with Radeon 9xxxx, Kas1e's MUI-Mplayer will be the fastest one with p96_pip (overlay).

Anyway, ATM, LiveForIt's Mplayer is the only choice for those using RadeonHD thanks to the compositing renedering created by LiveForIt, waiting for it to be implemented in MUI-Mplayer.

I will now try to get some benchmarks on Sam460

@LiveForIt
I see you start commit things, good. Can you commit p96_pip.c (so i can commit then white-strip fix over it), and vo_comp (so i can include it to muimplayer) ? If you worry about non-clean code or something, then you may send just vo_comp.c at kas1e@yandex.ru as well.

any news??

@LiveForIt

exectags.h has Sam460 and X1000 in it, if you have the latest SDK.

@LiveForIt

Quote:

The A1-XE RAM is much slower than that on newer machines, so video playback is fastest if each slice that's copied to VRAM is small enough to keep in the CPU cache. That way you avoid the overhead of writing to RAM and then reading it back. So, if you perform a conversion before copying to VRAM then you may be exceeding the cache size and incurring the penalty.

On systems with faster RAM the speed boost of efficient copies outweighs the overhead of exceeding the cache size, especially when DMA is available.

Is it possible to make libavcodec output ARGB directly?

Hans

@K-L

Quote:Which video are you using for your benchmarks?
I've updated my ancient pip vo, but to compare it with other MPlayer versions and vos I have to use the same video everyone else here seems to be using for the MPlayer benchmarks.

@joerg

Actually, I'm using Man Of Steel trailer to do my benchmarks.

This is one 4.6MB FLV file :

http://www.amiga-ng.org/resources/Mplayer/Superman.flv

Hope this can help