2012-11-03, 03:07
Hi there,
I have just created an account to post my feedback regarding xbmc on A10.
First of all, I would like to thank A LOT empatzero for his work at integrating cedarx in XBMC.
As far as I am concerned, I bought a cheap LY-F1 clone a few weeks ago to hack with A10.
Getting a fully working mediacenter on my linux install was one of the last missing piece and it seems to very close.
On my device, I run :
- A linux sunxi kernel (branch wip/linux-sunxi-3.0/next_mali in which I merged the last commits from the main branch)
- A armhf linaro rootfs (the 20120923 one)
I built empatzero git tree from the 30th of october with no major troubles : I didn't have the same issues as described in this thread. (I simply had to install the required *-dev packages to be able to build in the linaro RFS...).
The resulting XBMC boots fine :
- confluence skin runs fine on my tablet but the resolution of the tablet is low (800x480) : I still have to test how it plays when I use it on HDMI output
- I am able to decode and play HD H264 videos (720 and 1080) through the network (by upnp shares).
Note that I use a AX88772 USB to Ethernet converter to connect to my lan and stream the videos as the tablet has no Ethernet
The rendering is quite fluid (fps almost never drops below 20)
The maximum CPU usage (when decoding 1080 video and displaying subtitles) is less than 50% (to evaluate it I had to use a idle process as xbmc seems to use all cpu if nothing else requires cpu)
On the video rendering side, I am confident that it should be the same on HDMI output. Indeed after some reversing, I think that the video rendering is not performed via a GL texture.
But, as for the android version, the hw decoder puts the decoded pictures in FBM buffers (which are allocated in VE memory section), then xbmc notifies the kernel thanks to DISP_CMD_VIDEO_SET_FB ioctl and the picture will be displayed on the lowest layer at one of the coming vertical sync (The video back end also taking care of scaling) : There is no copy handled by the main cpu nor rendering of GL texture in this process, It just seems to be quite an optimal use of the hw platform...
I may be wrong but what I describe is my current understanding following an analyze of the available source of libcedarv and of the sunxi kernel display drivers...
Yet everything is not perfect :
There are still a few issues depending on the H264 videos. With some of them the whole screen turns pink or white for a little duration and then correct colors come back...
I have not been able to spot the difference between videos which play fine and others but I will investigate...
I have already checked that the issue is not in hardware decoder (I have another android A10 based tablet and I played one of these problematic video with MXplayer -Hw decoding enabled- and it does not exhibit the same issue...)
I also experienced some Audio sync issues (that were solved by setting a manual audio offset...)
I will post updates of my coming tests if there is some interest...
Best regards,
I have just created an account to post my feedback regarding xbmc on A10.
First of all, I would like to thank A LOT empatzero for his work at integrating cedarx in XBMC.
As far as I am concerned, I bought a cheap LY-F1 clone a few weeks ago to hack with A10.
Getting a fully working mediacenter on my linux install was one of the last missing piece and it seems to very close.
On my device, I run :
- A linux sunxi kernel (branch wip/linux-sunxi-3.0/next_mali in which I merged the last commits from the main branch)
- A armhf linaro rootfs (the 20120923 one)
I built empatzero git tree from the 30th of october with no major troubles : I didn't have the same issues as described in this thread. (I simply had to install the required *-dev packages to be able to build in the linaro RFS...).
The resulting XBMC boots fine :
- confluence skin runs fine on my tablet but the resolution of the tablet is low (800x480) : I still have to test how it plays when I use it on HDMI output
- I am able to decode and play HD H264 videos (720 and 1080) through the network (by upnp shares).
Note that I use a AX88772 USB to Ethernet converter to connect to my lan and stream the videos as the tablet has no Ethernet
The rendering is quite fluid (fps almost never drops below 20)
The maximum CPU usage (when decoding 1080 video and displaying subtitles) is less than 50% (to evaluate it I had to use a idle process as xbmc seems to use all cpu if nothing else requires cpu)
On the video rendering side, I am confident that it should be the same on HDMI output. Indeed after some reversing, I think that the video rendering is not performed via a GL texture.
But, as for the android version, the hw decoder puts the decoded pictures in FBM buffers (which are allocated in VE memory section), then xbmc notifies the kernel thanks to DISP_CMD_VIDEO_SET_FB ioctl and the picture will be displayed on the lowest layer at one of the coming vertical sync (The video back end also taking care of scaling) : There is no copy handled by the main cpu nor rendering of GL texture in this process, It just seems to be quite an optimal use of the hw platform...
I may be wrong but what I describe is my current understanding following an analyze of the available source of libcedarv and of the sunxi kernel display drivers...
Yet everything is not perfect :
There are still a few issues depending on the H264 videos. With some of them the whole screen turns pink or white for a little duration and then correct colors come back...
I have not been able to spot the difference between videos which play fine and others but I will investigate...
I have already checked that the issue is not in hardware decoder (I have another android A10 based tablet and I played one of these problematic video with MXplayer -Hw decoding enabled- and it does not exhibit the same issue...)
I also experienced some Audio sync issues (that were solved by setting a manual audio offset...)
I will post updates of my coming tests if there is some interest...
Best regards,
