IRC log of #cubox of Fri 05 Apr 2013. All times are in CEST < Back to index

16:27 dv_ rabeeh: in the kernel subdirectory drivers/video/dovefb , there is the nxp_hdmi folder. is this folder hosted somewhere else? it would be better in cases one just wants this folder, and not the rest of the kernel (for example, for libcec)
17:04 jnettlet dv_, my contract with OLPC is over for now, so I am porting all my new work to the cubox....finally
17:05 jnettlet I have been sick for a week or so, so am a bit behind on everything
17:05 dv_ jnettlet: ah, well, I have been using your driver for my OE layer
17:06 dv_ jnettlet: you might be interested in: https://github.com/dv1/meta-cubox/blob/master/recipes-graphics/xorg-driver/xf86-video-dove_git.bb and https://github.com/dv1/meta-cubox/blob/master/recipes-graphics/xorg-driver/xf86-video-dove/0001-autoconf-related-fixes.patch
17:06 dv_ I determined that video above 1024x768 wont work with DVI displays, but they do with proper HDMI ones
17:06 jnettlet dv_, I will take a look. I actually have a newer driver and galcore that I need to test on the cubox.
17:07 dv_ I havent had the time to test rabeeh's I2C trick yet
17:07 jnettlet there is someone on the forums that says they are successfully using a different in kernel HDMI driver.
17:07 dv_ oh, and the xorg.conf is : https://github.com/dv1/meta-cubox/blob/master/recipes-graphics/xorg-xserver/xserver-xf86-config/xorg.conf
17:07 dv_ but overall, the driver works for me. I have also been able to watch videos with it, using the XVideo extension
17:08 dv_ EXA also worked, but was unstable; for example, matchbox-session caused it to crash for some reason
17:08 jnettlet regardless, I have written a basic HDMI driver for the XO-4 so feel pretty confident we can get the cubox stuff working perfectly
17:08 jnettlet dv_, that is probably due to memory.
17:09 dv_ well, keep me posted about it. I can add new recipes to the layer then.
17:09 jnettlet I also have the newest vmeta gstreamer work...including encoding h264
17:09 dv_ nice! I also added vmeta, but so far only got the decoders for hardfp
17:09 dv_ also check out rabeeh's kernel fix for vmeta performance
17:10 jnettlet I have switched all the memory manager over to ION, backed by CMA. This makes the allocations dynamic for the desktop so you don't get a huge chunk of memory not doing anything if you aren't watching video
17:10 dv_ so the vmalloc parameter will go away?
17:10 jnettlet it should. I have to test on the cubox still :-)
17:11 jnettlet CMA is real, nice where the memory is mapped out in early boot, but can be allocated out if other programs needed, and then reclaimed by drivers later.
17:11 dv_ that would be nice, because the amount of memory to pick for vmalloc is never easy
17:11 dv_ *never easy to estimate
17:11 jnettlet But I have decoded 1080p h264 using only about 40MB's of RAM
17:12 dv_ I currently set it to 384MB, which is probably way too much
17:12 jnettlet it depends on the type of streams.
17:14 dv_ do you have links to your newer xorg dovefb driver and the new vmeta stuff?
17:15 jnettlet dv_, well the dovefb driver, is a revamped pxa168fb driver. I need to merge them. The code for this platform is so fractured at this point it is driver me nuts.
17:16 jnettlet dv_, what kernel version are you working from?
17:16 dv_ rabeeh's 3.6.9 kernel
17:16 dv_ from github
17:17 dv_ exact revision is here: https://github.com/dv1/meta-cubox/blob/master/recipes-kernel/linux-cubox/linux-cubox_3.6.9.bb
17:17 dv_ I have added his vmeta performance patch to master, for testing it out. so far, I havent seen any issues with it, and I have tried out several videos
17:18 jnettlet okay if you are ambitious feel free to grab my recent merge's from http://dev.laptop.org/git/olpc-kernel/log/?h=arm-3.5 that should integrate the ION MM with CMA and an MMP backend which we need
17:18 jnettlet dv_, do you have an image I can download to run on my cubox so we are working from the same thing?
17:20 dv_ yes. uploading it to dropbox right now.
17:20 jnettlet great thanks
17:21 jnettlet dv_, are you using gcc 4.8 or 4.7?
17:22 dv_ 4.7
17:22 dv_ 4.8 will be included in a later OpenEmbedded release
17:23 dv_ I will then add the new -march=marvellpj4 flag to the machine config
17:24 jnettlet okay, I was planning on merging the iwmmxt and marvell tuning patches into the Linaro optimized gcc 4.7 tree this weekend to get all the media stuff compiled.
17:24 jnettlet I fell behind because our new chipset supported both iwmmxt and neon, so the iwmmxt stuff fell to the back burner
17:25 dv_ both? :O
17:25 jnettlet yeah the new mmp3 chip from Marvell supports both
17:25 dv_ okay, here you go, here is the rootfs: https://www.dropbox.com/s/vjn0xmczrhm6hif/cubox-demo-image-sato-cubox-hardfp-3.tar.bz2
17:26 jnettlet coming down now thanks
17:26 dv_ just unpack it on an SD card, first partition. ext2 or ext3. thats it.
17:26 jnettlet will do
17:27 dv_ I will eventually move on to a 2-partition solution, since it is more flexible (for supporting sata disks, ext4/jfs/reiserfs/... root partitions, etc.)
17:36 dv_ jnettlet: also, do you know of the etnaviv project?
17:38 jnettlet dv_, yes but I can't partake in any of the reverse engineering as I have knowledge of the galcore userspace source under NDA
17:38 jnettlet I don't want to taint the project
17:40 jnettlet but I should provide dumps to them for our GC860 and GC2000 chips
17:43 dv_ well, wumpus is the man to talk to about this
17:44 jnettlet dv_, yeah it is on my list of things to do in my month off
17:52 dv_ is the image running? if you need something added or modified, please let me know
17:53 dv_ of course you can set up OE and build it yourself, but that takes several hours on my quad core machine (assuming you have >4GB RAM)
17:54 jnettlet dv_, ha...I am not that far. I am upgrading Twonky on my Verbatim Mediashare right now. Another Marvell ARM NAS :-)
17:54 dv_ plenty busy, hm? :)
17:55 dv_ btw. are the gstreamer vmeta plugins being maintained? or has development ceased, and only vmeta itself is being updated?
17:57 jnettlet dv_, nope this is the latest and greatest. Marvell merged some of my previous patches, and added a bunch of their own. It is still only gstreamer 0.10...I may have a project to pay for a gstreamer-1.0 upgrade...will see
17:58 dv_ yes, support for gstreamer 1.0 is also something I was considering
17:58 dv_ especially adopting it for the new gstreamer base classes, which should cut down code size
18:00 dv_ the current plugins are pretty big, and the xbmc patch for vmeta based playback are much smaller
18:01 dv_ yes, I just checked. the existing plugins are based off GstElement directly. gstreamer core developers advise against doing that these days.
18:04 dv_ jnettlet: also keep me posted about the gstreamer 1.0 thing
18:05 jnettlet dv_, I am just in early dealings so we will see, but will let you know anything I can.
18:05 dv_ alright, thanks
18:05 jnettlet The new plugin is definitely better than the current one in circulation
18:06 jnettlet and that on top of my ION mm changes make it a lot more desktop friendly
18:06 dv_ I thought of setting up a project containing several hw acceleration elements for various platforms. like marvell vmeta, freescale VPUs, TI codecengine etc.
18:06 dv_ because some of them are in a pretty broken state, and none has been updated to 1.0 yet
18:09 dv_ ah, any chance of public access to the new vmeta gstreamer plugin?
18:11 jnettlet dv_, probably this weekend.
18:11 dv_ nice, thanks
19:47 dv_ oh jnettlet, do you know the reason why the vivante drivers are X11 only?