• 1
  • 11
  • 12
  • 13(current)
  • 14
  • 15
  • 52
libstagefright - Experimental hardware video decoding builds
(2013-01-28, 03:08)speed32219 Wrote: The latest 1_27_13 build is a Major improvement. Great job!!! I tested all the movies that CrunCher said were failing both in 720P and 1080P from Utube and they are flawless! locked in at 23.98/24 fps stf-h264 with total CPU usage around 36-42%. Note: I don't much like testing across a wireless network using addon/plugin's for testing anyway (way to many variables), so downloaded them to the usb attached mass storage device. (720p streaming worked great though)

My previous samples were using a BD rip to MKV with USB mass storage attached for best performance. This is a HDDVD X264 AC3 that drops frames still using the USB attachment, but not as bad as 1_21_13 release. It pretty much locks at 24 fps with CPU usage around 71-74% for the first 3-5 minutes. Then when fps drops occur pc:1 becomes pc:none (codec o) and fps is erratic (13, 18, 21, 14.5, etc) and cpu usage 70-87%.

You are almost there with the rockchip 3066 dual core, again kudos for doing a fantastic job with achieving hw acceleration on these low end devices.

pastebin:

http://pastebin.com/h7teGysv

What is your Device and Firmware ?

MK808 Rockchip 3066 dual core. Fineless 1.6 Rom i see
it's pity that no one starts Tablet Firmwares and Kernels and all the development concentrates on those Sticks *grml* which is sad if you realize that the first Kernel Sources where Tablet based Sad

I will try some different older Firmware (Kernel releases, there where quiet a few because of all kind of issues) and see if that might bring a improvement on this Tablet for XBMC in the 1080p stability cases please note all these tests where done on the internal SDCard (Hynix Nand) not streaming of a Network.

here is another rendering issue i have to fight with even SD (it renders throughout the file squished like that,and again external hardware decoder player show no issues), it's crazy seeing this fail but 720p work Sad

Image

here the file causing this https://dl.dropbox.com/s/l61vll2vdfi01nk...k.flv?dl=1

All these failing streams have 1 thing in common they all show this strange line on the top

Could it maybe have todo with the Rendering resolution and the Android Nav Bar ?
That's a flash video, JB does not support adobe flash but ICS does. Also what is the rating of that sd card? My read and write speeds are by far much faster than many sd cards connected through a usb 2.0 port from a HD.

Fast tablet and nand SD flash there CrunCher. surprised your having those issues. Good luck and good testing.
What for a nonsense is this it's a H.264 Bitstream one way or another only a different container and as i said externally it works perfectly fine.
The Flash support is meant to be for Adobes Flash Player that's something entirely different it has nothing todo with the file itself being not supported.

Current App and Nand use Stage (1st run):

Image
Image
ODROID-U2

Crash after approx 1h playback. Not the newest build as that one doesn't work for me.

http://xbmclogs.com/show.php?id=30809
(2013-01-28, 03:54)CruNcher Wrote:
(2013-01-28, 03:08)speed32219 Wrote: The latest 1_27_13 build is a Major improvement. Great job!!! I tested all the movies that CrunCher said were failing both in 720P and 1080P from Utube and they are flawless! locked in at 23.98/24 fps stf-h264 with total CPU usage around 36-42%. Note: I don't much like testing across a wireless network using addon/plugin's for testing anyway (way to many variables), so downloaded them to the usb attached mass storage device. (720p streaming worked great though)

My previous samples were using a BD rip to MKV with USB mass storage attached for best performance. This is a HDDVD X264 AC3 that drops frames still using the USB attachment, but not as bad as 1_21_13 release. It pretty much locks at 24 fps with CPU usage around 71-74% for the first 3-5 minutes. Then when fps drops occur pc:1 becomes pc:none (codec o) and fps is erratic (13, 18, 21, 14.5, etc) and cpu usage 70-87%.

You are almost there with the rockchip 3066 dual core, again kudos for doing a fantastic job with achieving hw acceleration on these low end devices.

pastebin:

http://pastebin.com/h7teGysv

What is your Device and Firmware ?

MK808 Rockchip 3066 dual core. Fineless 1.6 Rom i see
it's pity that no one starts Tablet Firmwares and Kernels and all the development concentrates on those Sticks *grml* which is sad if you realize that the first Kernel Sources where Tablet based Sad

I will try some different older Firmware (Kernel releases, there where quiet a few because of all kind of issues) and see if that might bring a improvement on this Tablet for XBMC in the 1080p stability cases please note all these tests where done on the internal SDCard (Hynix Nand) not streaming of a Network.

here is another rendering issue i have to fight with even SD (it renders throughout the file squished like that,and again external hardware decoder player show no issues), it's crazy seeing this fail but 720p work Sad

Image

here the file causing this https://dl.dropbox.com/s/l61vll2vdfi01nk...k.flv?dl=1

All these failing streams have 1 thing in common they all show this strange line on the top

Could it maybe have todo with the Rendering resolution and the Android Nav Bar ?

Classic problem when someone has the width stride wrong. I've not looked at the video yet but sometimes, depending on hw decoder quirks, the width needs to be quantized. ie. 403 -> 404.
Width: 854 pixels

so MOD2 isn't supported by the RockChip Decoder it seems then
eh, I mean it looks like it's decoding fine but the render stride is wrong. We saw this sometimes with VDA hw decode. The width the decoder would output was slightly larger than the width reported by our hints.
Yes i guess MOD8 in this case 856 should work without issues i guess so just padding those 2 pixels

tested a huge amount of 720p streams now and all so far where fine except this one which dropped constantly a huge amount of frames but was decoding with stf-h264.

http://pastebin.com/Tgj8jD1h

http://pastebin.com/Bi4mUWw2

It decodes fine with the Stock Player and MX Player

its a Samsung HMX-H300 Bitstream
(2013-01-28, 06:45)CruNcher Wrote: tested a huge amount of 720p streams now and all so far where fine except this one which dropped constantly a huge amount of frames but was decoding with stf-h264.

http://pastebin.com/Tgj8jD1h

http://pastebin.com/Bi4mUWw2

It decodes fine with the Stock Player and MX Player

its a Samsung HMX-H300 Bitstream
Targeting 60fps is whee bit optimistic Wink
I'm getting failures with stagefright

http://xbmclogs.com/show.php?id=30927[/code]
The RockChip VC-1 Decoder is still not utilized even for .wmv with AP@L2 it loads ff-vc1 as decoder and that is damn slow even @ a bitrate of 4 mbits it just gets around 11 fps

Same situation for the RealVideo part of things

Are the VC-1 and RV parts already neon optimized ?

Maybe this is also the overhead of the Full Debug build ?
neon will not help you with HD video Smile
I see btw here is a full log of one of the 1080p streams (all) in this case the Youtube Zelda-Medley i have severe random issues with

http://pastebin.com/Fcj6k2YT

though i guess much isn't visible as it crash rebooted almost instantly after clicking on it building up in the background this time not even gone as far as playing back 1 frame the same way like the birds stream crashes Sad

I think i have collected enough 1080p test samples now (that randomly fail, even though it looks like every 1080p stream fails with random issues) i will try to test them on different Firmware versions (and there kernels) maybe it changes something.

Also in a clean state maybe really some App in the background does severe things here that make everything unstable though this isn't windows isn't it Wink

But before that i will try the Software Renderer build and yes i disabled libstagefright in the settings also for testing and yes it plays fine with the ff-h264 decoder i mean what you can call fine, performance is off course far away from fine but it's stable that's what i mean Wink

The current most important resume so far of the XMBC Hardware Rendering state for me and my RK 3066 Device (Tablet):

H.264 720p:
runs pretty every 8 bit lossy stream up to complexity of 40 fps bitrate can be quiet high for 720p tested up to 12 Mbit with peaks (pretty much no normal content should go over that) (also tested some 720p MBAFF Broadcast streams some complexitys worked others where to complex because of the FPS and dropped frames like crazy, all the MBAFF streams showed Interlacing)

H.264 1080p:
Pure Horror Wink
Nexus 7 / Android 4.2.1 / Tegra 3

The 1/27 hw build shows massive stuttering on the 1080P/40Mbps Birds video. This same video played fine (albeit with some network buffering) on the 1/18 build.

I tried to download the xbmc.log file using ADB, but all three of the paths listed in the wiki ( http://wiki.xbmc.org/index.php?title=XBM...Q#Userdata ) gave a "file not found" error.

I'm aware there are alternative plugins to download the log file, but I want to use ADB to scrub network passwords before posting.

Has anyone gotten ADB to download the xbmc.log file on a Nexus 7?
@Koying

i tried the Software Renderer version xbmc-fulldebug-stagefright-swrender-20130113 but there it seems the basic stuff fails (Hardware Decode)

http://pastebin.com/yM2mnekV
  • 1
  • 11
  • 12
  • 13(current)
  • 14
  • 15
  • 52

Logout Mark Read Team Forum Stats Members Help
libstagefright - Experimental hardware video decoding builds10