Turl changed the topic of #linux-sunxi to: Allwinner/sunxi development discussion - Don't ask to ask. Just ask and wait! - See http://linux-sunxi.org | https://github.com/linux-sunxi/ | Logs at http://irclog.whitequark.org/linux-sunxi | #blameOliver
tm512 has quit [Ping timeout: 246 seconds]
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
tm512 has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
tm512 has quit [Ping timeout: 246 seconds]
tm512 has joined #linux-sunxi
BJfreeman has quit [Quit: had a good time]
tm512 has quit [Ping timeout: 246 seconds]
tm512 has joined #linux-sunxi
torindel has quit [Ping timeout: 272 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
naobsd has quit [Ping timeout: 250 seconds]
torindel has joined #linux-sunxi
Black_Horseman has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
popolon has quit [Quit: Quitte]
naobsd has joined #linux-sunxi
torindel has quit [Ping timeout: 246 seconds]
torindel has joined #linux-sunxi
egbert has quit [Disconnected by services]
egbert has joined #linux-sunxi
FR^2 has quit [Ping timeout: 240 seconds]
pfdm has quit [Ping timeout: 250 seconds]
deasy has quit [Quit: Nom d'un quark, c'est Edmonton !]
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 246 seconds]
torindel has joined #linux-sunxi
pepee has quit [Quit: goto Quit;]
torindel has quit [Ping timeout: 272 seconds]
torindel has joined #linux-sunxi
[7] has quit [Read error: Operation timed out]
TheSeven has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
eebrah has joined #linux-sunxi
torindel has quit [Remote host closed the connection]
torindel has joined #linux-sunxi
eebrah has quit [Read error: Connection reset by peer]
eebrah has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
eebrah is now known as Guest81696
FDCX has quit [Remote host closed the connection]
wingrime has joined #linux-sunxi
Black_Horseman has quit [Remote host closed the connection]
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
Guest81696 has quit [Read error: Connection reset by peer]
eebrah has joined #linux-sunxi
eebrah is now known as Guest23562
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
JohnDoe_71Rus has joined #linux-sunxi
FR^2 has joined #linux-sunxi
Guest23562 has quit [Ping timeout: 245 seconds]
eebrah_ has joined #linux-sunxi
torindel has quit [Ping timeout: 248 seconds]
torindel has joined #linux-sunxi
eebrah_ has quit [Ping timeout: 245 seconds]
eebrah has joined #linux-sunxi
eebrah is now known as Guest35515
wingrime has quit [Ping timeout: 246 seconds]
pacopad has joined #linux-sunxi
<mnemoc> arokux1: can you sync your own sunxi-next-usb with the one pushed to linux-sunxi. I added developer trees support to the nightles but yours doesn't even have a sunxi_defconfig :<
<mnemoc> arokux1: also, what's your u-boot repo?
torindel has quit [Ping timeout: 246 seconds]
torindel has joined #linux-sunxi
hramrach has quit [Ping timeout: 240 seconds]
hramrach has joined #linux-sunxi
wingrime has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
n01 has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
Guest35515 has quit [Ping timeout: 248 seconds]
torindel has joined #linux-sunxi
eebrah has joined #linux-sunxi
n01 has quit [Ping timeout: 260 seconds]
eebrah is now known as Guest23525
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
rellla has joined #linux-sunxi
n01 has joined #linux-sunxi
ZetaNeta has joined #linux-sunxi
Guest23525 has quit [Ping timeout: 241 seconds]
n01 has quit [Ping timeout: 260 seconds]
rz2k has joined #linux-sunxi
rellla has quit [Remote host closed the connection]
n01 has joined #linux-sunxi
<oliv3r> hramrach: ah yeah that's a fork of my repo; yeah it doesn't work yet ):
torindel has quit [Ping timeout: 248 seconds]
torindel has joined #linux-sunxi
eebrah has joined #linux-sunxi
eebrah is now known as Guest7566
\\Mr_C\\ has quit [Quit: .]
Guest7566 has quit [Ping timeout: 245 seconds]
eebrah_ has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
\\Mr_C\\ has joined #linux-sunxi
<hramrach> oliv3r: it's almost there. it spews junk about division by zero but the clocks still work. And it's missing the pin definition for powering up sata on cubieboard
<hramrach> I am trying the 3.4 kernel on ct
<hramrach> gmac works fine but the wifi is a mess
<hramrach> it alway complains [dhd_module_init] get wl_host_wake gpio failed
<hramrach> I guess I am not going to debug that today
<wingrime> mnemoc: I sended small patch for rfkill
<wingrime> mnemoc: expand it from a10 or a13 to sunxi platform
<mnemoc> wingrime: ok. but need to give people time to comment it ;-)
<mnemoc> Winner's...
<mnemoc> funny people
<wingrime> mnemoc: wifi drvier realy mess
<wingrime> mnemoc: I will wait comments within one mounth
<mnemoc> 2 working days is enough in this case I think
<wingrime> mnemoc: usualy no one comments , if patch not interesting for them
<mnemoc> no comments will imply accepted
<wingrime> mnemoc: new rules)
<mnemoc> very simple patch
<wingrime> mnemoc: also, I not fully agree with SoC parts as modules
eebrah_ has quit [Ping timeout: 245 seconds]
<mnemoc> and I have more time than before
<mnemoc> wingrime: sadly that's the mainline way
<wingrime> mnemoc: nice
<wingrime> mnemoc: we should make something with autodetect
<mnemoc> wingrime: platform-drivers for everything, mach-foo/ been empties and palt-foo killed...
<mnemoc> we do have soc-detect merged
<mnemoc> I'm writting code to turn script.bin into platform-devices in plat-
<mnemoc> as .dts would do
<mnemoc> removing most of the script parsing from the drivers
<mnemoc> and then to migrate most #ifdef to sunxi_is_foo()
<mnemoc> having multi(sunxi)platform in 3.4
<mnemoc> and at the same time turning ugly drivers into a form more friendly to 3.10 and then mainline
<wingrime> mnemoc: lots of work
<mnemoc> yes :|
<wingrime> mnemoc: very sad, that a20 PM stll broken
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
<wingrime> mnemoc: wifi in been's branch, very hackishly connected to sources
HeHoPMaJIeH has joined #linux-sunxi
HeHoPMaJIeH has quit [Changing host]
HeHoPMaJIeH has joined #linux-sunxi
JohnDoe_71Rus has quit [Ping timeout: 252 seconds]
<wingrime> mnemoc: I just sended small patch -- for incoming gmac driver, adds gmac IRQ to plat/irqs.h
<mnemoc> ok
<wingrime> gmac writen by some "shuge"
HeHoPMaJIeH has quit [Ping timeout: 252 seconds]
eebrah has joined #linux-sunxi
torindel has quit [Ping timeout: 272 seconds]
<rz2k> wingrime: shuge is hipboi's friend
torindel has joined #linux-sunxi
<rz2k> if you'll find problems there - you can write to hipboi and he will translate
<rz2k> guess shuge works at allwinner right now
<rz2k> i also saw that email couple of times somewhere
HeHoPMaJIeH has joined #linux-sunxi
HeHoPMaJIeH has quit [Changing host]
HeHoPMaJIeH has joined #linux-sunxi
Daps2 has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
<Daps2> Hi, sorry if i'm in the wrong place but .. can i output CGA 15khz specific resolution with an Alimex A13 board ?
torindel has joined #linux-sunxi
<rz2k> Daps2: olimex a13 board has vga done through the lcd interface
<rz2k> you can setup any clocks you want
<rz2k> editing your script.fex/bin
<Daps2> yes i've seen it .. but i came from the x86 world so .. i'm a bit lost .. i presume that a modeline in xorg.conf is not enough :)
<rz2k> no, use fbdev or fbturbo for xorg and setup everything in the .fex
<Daps2> can i use values from modeline and put them in script.fex/bin ?
<rz2k> yes
<rz2k> about the 15khz
<rz2k> is it pixel clock?
<rz2k> because lowest I saw for lcd is 420x272 screen with 8mhz pixel clock
<rz2k> that is the one olimex is selling actually
<mnemoc> fbturbo sounds fancy :p
<wingrime> ssvb: I see some glitches with CT
<Daps2> ok, i need 6,700 for this particular use
<wingrime> ssvb: in Xvideo
<Daps2> no 15khz is for using an old crt tv
<Daps2> vga = 31
<Daps2> cga = 15
torindel has quit [Ping timeout: 240 seconds]
<ssvb> wingrime: yes, I also observed some lockups which do not happen on A10
torindel has joined #linux-sunxi
<ssvb> wingrime: but need to find a simple way to reproduce them
<wingrime> ssvb: thats happends all time, and I not see normal image
<wingrime> ssvb: It maybe becose I have G2d
<Daps2> rs2k: i've found the [lcd0_para] options but .. i don't know how to translate modelines/DTD parameters to lcd0_* parameters :x
<rz2k> you need to know the horizontal sync, vertical sync, h. front and back porch, v. front and back portch and h/v pulse width
<ssvb> wingrime: yes, it happens quite often on random usage, but still it would be nice to reduce it to a minimal set of 100% reproducible steps
<rz2k> Daps2: I bet for cga standard you can find that somewhere already done
<wingrime> ssvb: 1) run any video with mplayer 2) got that
<rz2k> or figure it out yourself.
<ssvb> wingrime: hmm, this just works fine for me
<Daps2> rs2k: i've searched but didn't find anything for CGA :x ..
<ssvb> wingrime: I get problems only after using it for a while and watching multiple different videos
<rz2k> cant believe that :p I bet someone already done that. If you want to find more help - place question on our mailing list
<ssvb> wingrime: for example, repeatedly starting and stopping playback for the same video does not seem to trigger this bug
<wingrime> ssvb: I not sure that I use fresh driver, but It happends on CubieTruck always, but VDPAU works nice
<rz2k> and google how vga works (the sync pulses and etc.)
<rz2k> you need those values
<Daps2> hehe, ok. but i will try to translate modelines to lcd_* parameters first .. sadly i didn't received yet my board so ...
<Daps2> i have the modelines for the specifics resolutions i want to use, modeline contain all informations, i only need to translate them, i lack knowledge of thoose options
<rz2k> hope you'll figure it out, they should be relatively same to the .fex values, nothing really changed in this area for years.
<Daps2> i will do it, thanks for your help
HeHoPMaJIeH has quit [Remote host closed the connection]
eebrah has quit [Ping timeout: 245 seconds]
geecko has joined #linux-sunxi
eebrah has joined #linux-sunxi
eebrah is now known as Guest14395
fnchen has joined #linux-sunxi
fnchen has left #linux-sunxi [#linux-sunxi]
fnchen has joined #linux-sunxi
fnchen has quit [Remote host closed the connection]
<oliv3r> hramrach: i added some more bits and pieces in my branch, but it was giving me some issues aswell; I was a little to busy the last few weeks, but intend to work on it again versy soon!
Guest14395 has quit [Read error: Connection reset by peer]
rz2k has quit [Read error: Connection reset by peer]
eebrah_ has joined #linux-sunxi
ZetaNeta has quit [Quit: Leaving]
Soru has joined #linux-sunxi
eebrah_ has quit [Read error: Connection reset by peer]
rz2k has joined #linux-sunxi
eebrah_ has joined #linux-sunxi
torindel has quit [Remote host closed the connection]
torindel has joined #linux-sunxi
yo5phz has joined #linux-sunxi
<yo5phz> hey guys, my olinuxino A13 wifi crashes without any logs when I load the sun4i-ts driver...anyone had any similar troubles?
rz2k has quit [Ping timeout: 240 seconds]
torindel has quit [Ping timeout: 246 seconds]
torindel has joined #linux-sunxi
iamfrankenstein has joined #linux-sunxi
<kriegerod> anybody knows if amlogic meson3 is usable only as softfp, or it has FPU?
torindel has quit [Ping timeout: 272 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 245 seconds]
torindel has joined #linux-sunxi
torindel has quit [Ping timeout: 248 seconds]
torindel has joined #linux-sunxi
ZetaNeta has joined #linux-sunxi
torindel has quit [Ping timeout: 272 seconds]
torindel has joined #linux-sunxi
eebrah_ has quit [Ping timeout: 240 seconds]
rz2k has joined #linux-sunxi
eebrah has joined #linux-sunxi
eebrah is now known as Guest91517
iamfrankenstein has quit [Ping timeout: 245 seconds]
torindel has quit [Ping timeout: 260 seconds]
torindel has joined #linux-sunxi
hramrach has quit [Ping timeout: 240 seconds]
hramrach has joined #linux-sunxi
pfdm has joined #linux-sunxi
Guest91517 has quit [Ping timeout: 240 seconds]
vinifr has joined #linux-sunxi
n01 has quit [Ping timeout: 240 seconds]
n01 has joined #linux-sunxi
vinifr has quit [Ping timeout: 245 seconds]
rellla has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
torindel has joined #linux-sunxi
eebrah has joined #linux-sunxi
eebrah is now known as Guest22402
Black_Horseman has joined #linux-sunxi
geecko has quit [Read error: Connection reset by peer]
geecko has joined #linux-sunxi
rellla has quit [Remote host closed the connection]
Guest22402 is now known as eebrah_
vinifr has joined #linux-sunxi
<wingrime> oliv3r: ping
<wingrime> oliv3r: our sata performance so low as 10 Mb/s?
<wingrime> ssvb: ^
<wingrime> ssvb: also, sory, error hapends when video resolution relative big
<wingrime> ssvb: 720p or 1920p
<wingrime> ssvb: I think you should add additonal checks for fb allocs
<ssvb> wingrime: ok, I need to check bigger videos
<ssvb> wingrime: and we don't do framebuffer allocation, if the video is too big and does not fit the offscreen framebuffer area, xv simply does not work
<ssvb> wingrime: what is your screen resolution, framebuffer size and video resolution when it fails?
<wingrime> ssvb: 1920p
<wingrime> ssvb: for monitor
<ssvb> you can find the framebuffer size information in /var/log/Xorg.0.log
<wingrime> ssvb: sunxi_fb_mem_reserve=16
<wingrime> ssvb: If video relatively big, I see somthing strange in video-window
<pfdm> wingrime: are you working on RE in vdpau ? If then do you know why the pointer on the picture is not return as vir2phy directly but only when rendering through /dev/disp ?
<pfdm> Trying to avoid uncesseray re-copy ?
<wingrime> pfdm: not realy understand what are you trying to do
<wingrime> pfdm: in hardware words, nothing bloc me do zero-copy decoding
<wingrime> *block
torindel has quit [Ping timeout: 246 seconds]
<ssvb> wingrime: sunxi_fb_mem_reserve=16 does not mean much, the real framebuffer size may be smaller than this (depending on fb0_framebuffer_num from fex)
torindel has joined #linux-sunxi
<pfdm> wingrime: i'm trying to make libvdpau works in xbmc, i feel like the pointer return in the surface does not point to the decoded picture in yv12. In the presentation_queue_display then for the framebuffer the pointer is transformed bia vir2phy .
<wingrime> pfdm: but SUNXIFB(0): hardware: (video memory: 7200kB)
<wingrime> err
<wingrime> ssvb ^
<ssvb> wingrime: yes, this looks like a problem with the framebuffer size, please check fb0_framebuffer_num in your fex
<ssvb> pfdm: it's not a traditional planar yv12, but a special tiled format
<wingrime> pfdm: thats tiled
<wingrime> pfdm: I think thats done for mem usage optimizaton (memory queue)
<wingrime> pfdm: but, who knows, maybe cedar support something other than tiled format
pfdm_ has joined #linux-sunxi
shineworld has joined #linux-sunxi
<wingrime> pfdm: but control bits are unknown
pfdm has quit [Ping timeout: 250 seconds]
<pfdm_> wingrime: does that means, that we can't convert it and render it through a shader ?
<wingrime> pfdm_: disp directly supports than
<wingrime> pfdm_: why not simply send disp pointer and not care?
<ssvb> pfdm_: I suggest you to look into the allwinner port of xbmc sources and find how this is getting handled
<wingrime> pfdm_: you can use classic hw overlay for render
<wingrime> pfdm_: we have color keys, layers, hw mixer for it
<ssvb> pfdm_: xbmc is probably doing something with the disp overlays to composite the decoded video output with the user interface elements
<pfdm_> ssvb: I understand they use disp directly , in all port, i thought it was possible to use a classic render like the GLES one they have in xbmc which will make it much simpler and we could beneficiate feature of gl
<wingrime> pfdm_: ask rella
<ssvb> pfdm_: you mean to add some sort of gl overhead and "benefit" from it?
<wingrime> pfdm_: he is "they"
<pfdm_> yes for the osd and other. I actually succeed in using libvdpau with xbmc through disp but i need a better integration with the rest of the UI
<wingrime> ssvb: g2d can alpha mix?
<ssvb> wingrime: g2d does not support the tiled yuv format directly
<wingrime> ssvb: thats also possible cedar can produce different format
<wingrime> ssvb: but detals still unknowns
<wingrime> ssvb: also, thats possble ISP engine can do alpha mix, at least a20 manual claim this
<pfdm_> wingrime: there is some new code for the format support in the new blob
<wingrime> pfdm_: can you say more precisely...
<wingrime> CEDARV_PIXEL_FORMAT_PLANNER_YVU420
<wingrime> vs
<wingrime> CEDARV_PIXEL_FORMAT_AW_YUV420
<pfdm_> they add that one https://github.com/patricius972/cedarx-libs/blob/master/libcedarv/linux-armhf2/vdecoder/build/libcedarv.h#L162 and rename #define CEDARV_PIXEL_FORMAT_AW_YUV420 CEDARV_PIXEL_FORMAT_MB_UV_COMBINE_YUV420
<wingrime> comment "1633"
<ssvb> does "1633" look like a cedar hardware revision id?
<wingrime> yeax
<ssvb> A31 maybe?
<wingrime> ssvb: not have a31
lunra has quit [Ping timeout: 264 seconds]
lunra has joined #linux-sunxi
torindel has quit [Ping timeout: 240 seconds]
<wingrime> mnemoc: ping
<wingrime> ssvb: Yes, I need someone who can do simple uboot manipulations
torindel has joined #linux-sunxi
<wingrime> ssvb: but, AW realy lazy, thay use same a10 id with a20 for cedar
<ssvb> is cedar hardware different in a10 and a20?
<wingrime> ssvb: not realy
<mnemoc> wingrime: better than pinging, just tell what you want to say ;-)
<wingrime> mnemoc: 1633 is a31?
<ssvb> well, then keeping the same id makes sense :)
<wingrime> ssvb: a13 hw not much different but need some additionals buffers
<mnemoc> wingrime: sun6i
<wingrime> ssvb: as a20 report same id like a10, old unmodifited a10 blobs are working
<wingrime> ssvb: if you handle cpu cache correctly
<wingrime> pfdm_: That possible, a20/a10 CAN output such way, but questions in bitfileds in _CTRL or _TRIG registers
<wingrime> pfdm_: I need trace with enabled such output from blob, than I can try find where is it
<wingrime> pfdm_: also, there much unknown regs in 00-0xcc block that blob not uses, but there something usualy
<wingrime> last time we find counters and histogram ability
<ssvb> wingrime: I would not have high hopes, G2D in sun6i supposedly supports premultiplied alpha (based on the kernel sources), but no luck with this feature for sun5i/sun4i/sun7i
<ssvb> wingrime: but checking what kind of bits are set for sun6i cedar to get planar yuv420 still makes sense
<ssvb> wingrime: if they do not work on sun7i, then we are probably out of luck
<pfdm_> wingrime: Ok, i do not understand the Register at that time, but from my test changing the format from MB to PLANNER is not really affecting the rendering with /dev/disp when i used the blob.
<wingrime> :(
<ssvb> pfdm_: are you doing all these tests with the new cedar userspace blob?
<pfdm_> wingrime: so if libvdpau sends the good decode surface pointer, the reason i have a green ouput is because the shader is not correct ?
<pfdm_> yes that what i did, and the results are the same has the previous blob, same artifact on h264. which leads me to work with libvdpau
<wingrime> pfdm_: are you on a20?
deasy has joined #linux-sunxi
<pfdm_> wingrime: yes cb2
<wingrime> pfdm_: thats looks cache issue
<pfdm_> ssvb: yes then i change to libvdpau and x11.
<ssvb> pfdm_: does the newest and greatest linux libve blob still has the same weighted P-frame prediction bug? what about the android blobs loaded via libhybris for comparison?
<pfdm_> wingrime: that does not change anything, I applied this patch to the libcedarv , flushing after each mem_cpy .
<pfdm_> ssvb: yes its the prediction bug, libhybris is fine, but too slow
eebrah_ has quit [Ping timeout: 245 seconds]
<pfdm_> wingrime: actually the flush cache patch is necessary on the new kernl with cedarv patch
<ssvb> pfdm_: "slow" does not make any sense, it should be all the same native code with a bit different loading procedure
eebrah_ has joined #linux-sunxi
<wingrime> pfdm_: thats necessary becose, I not let parkic disable caching for cedar
<ssvb> pfdm_: with everything else equal, libhybris libve and native libve should have roughly the same performance
deasy has quit [Quit: Nom d'un quark, c'est Edmonton !]
<pfdm_> ssvb: obviously the movie is not playing at the right speed with some lags. that's my definition of slow
<ssvb> pfdm_: does it play fast with native libve in the same conditions?
<wingrime> pfdm_: yes, vdpau more clear than aw blob
<ssvb> pfdm_: and yes, xbmc is slow, vlc is known to have much less performance problems
<pfdm_> ssvb: yes it plays normally. no lag but white picture
<pfdm_> ssvb: i don't have anyproblem with xbmc and libvdpau actually ! it plays very nicely
<pfdm_> but then i don't have integration of the osd and ui.
<ssvb> pfdm_: I just mean that there is no technical reason for libhybris loaded blob to be slow
<ssvb> pfdm_: no integration for osd and ui sounds like vlc ;) this surely removes a lot of overhead
<pfdm_> ssvb: but there is an additional overhead due to calling the android function ? whatever they are doing in between that remove the bug .
<pfdm_> ssvb: that's because i use the /dev/disp on top layer. i need to figure how to display on the bottom layer with the pixmap, or without.
<wingrime> ssvb: Do you think its possible using many small textures for let cedar with GLES?
popolon has joined #linux-sunxi
<ssvb> wingrime: maybe pfdm_ can explain how to export arbitrary buffers to GLES
<pfdm_> ssvb: I thought we could render through glsl (default in xbmc) or gles, or the /dev/disp (but the wiki is stating it 's gonna disapear ?)
<wingrime> pfdm_: /dev/disp realy will not be in mainline
<ssvb> my understanding is that the physical memory buffers need to be wrapped into UMP, and then the UMP secure id numbers somehow need to be used from GLES code
<wingrime> pfdm_: all things must be done standart way
<wingrime> pfdm_: using /dev/drm for example
<pfdm_> ssvb: lool, I don't know for random buffers. I understand you fully the standard way is good, but then how do I use the result of the decoder to render ? is that result directly usable or not (that's the main question) ?
<ssvb> pfdm_: what is the standard way?
<ssvb> we surely want to avoid "still need to be copied from the application memory to GPU memory" part
<wingrime> pfdm_: you can't directly use render result, My opinion that, aw specialists made this only for mem bandwidth, just for not block memory with hi-priorty DMA transfer for whole frame
<wingrime> pfdm_: and hardware overlay, is usualy fastest way, as it realy zero copy, OSD, if you want so, can be done with color-key
bsdfox\ has joined #linux-sunxi
<pfdm_> ssvb: i guess my understanding is decode the bitstream , get the picture in a known format and then use it in a render GL,GLES, or other.
<ssvb> pfdm_: get the picture into what buffer? and how to make GLES read data from this buffer?
<ssvb> pfdm_: if you don't care about the performance, converting format can be done, for example, on the CPU
<pfdm_> wingrime: ok i'll investigate overlay code of huceke.
<pfdm_> ssvb: i understand that there is a buffer of decoded frame to be display, in xbmc , they call this buffer yuvbuffer in a linuxrender object which (i'm still looking) i guess is filled by the ffmpeg decoder
<pfdm_> then ask to render the picture (and does glsl or other) this part is already well working if the data in the yuvbuffer is in expected state.
boycottg00gle has joined #linux-sunxi
<pfdm_> and then using libvdpau is returning a surface_output which (in what i understand) contains the decoded frame.
<wingrime> pfdm_: you should understand, that actualy using hardware for such thing adds some difficulties, for example, cedarX need physicaly-lineral memory for decoding
<wingrime> pfdm_: as, it simple device, on AMBA bus, CPU not knows about memory changes that done with cedar, so cpu cache is not helpful there
torindel has quit [Ping timeout: 272 seconds]
torindel has joined #linux-sunxi
<pfdm_> wingrime: ok. so if /dev/disp is not going to mainline, the way will be what ssvb said about UMP ?
<ssvb> UMP is also not going to mainline :)
eebrah_ has quit [Read error: Connection reset by peer]
<wingrime> pfdm_: to early saying about disp and cedar, for mainline
<wingrime> pfdm_: we still have no mmc ))
torindel has quit [Ping timeout: 248 seconds]
torindel has joined #linux-sunxi
<ssvb> pfdm_: just looked into XBMC code, it seems to be relying on glTexSubImage2D in CLinuxRendererGLES::LoadPlane
<ssvb> pfdm_: is this how you want this stuff to work?
eebrah_ has joined #linux-sunxi
torindel has quit [Ping timeout: 260 seconds]
torindel has joined #linux-sunxi
shineworld has quit [Remote host closed the connection]
boycottg00gle has quit [Remote host closed the connection]
<pfdm_> the function they endup calling is CLinuxRendererGLES::RenderSinglePass . I have to debug and see if and when they are calling uploadYV12texture.
<pfdm_> ssvb: another one to look at may be this one CLinuxRendererGL::RenderVDPAU
Daps2 has quit [Ping timeout: 250 seconds]
eebrah_ has quit [Read error: Connection reset by peer]
torindel has quit [Ping timeout: 240 seconds]
<ssvb> pfdm_: and then you suddenly have to deal with GLX
shineworld has joined #linux-sunxi
torindel has joined #linux-sunxi
eebrah_ has joined #linux-sunxi
<pfdm_> ssvb: my plan is to first deal with it as a proof that it works. then try to translate the GLX calls in EGL. I know GLX is not supported.
pfdm has joined #linux-sunxi
eebrah_ has quit [Read error: Connection reset by peer]
pfdm_ has quit [Ping timeout: 250 seconds]
eebrah_ has joined #linux-sunxi
<pfdm> ssvb: i'm looking in xbmc master and they have added a function for egl CLinuxRendererGLES::UploadEGLIMGTexture .
deasy has joined #linux-sunxi
<pfdm> ssvb: interesting ! So if i understand , we need to allocate memory by the ump then use it to render ?
<ssvb> pfdm: it should be possible to take UMP secure id (or handle) and use it as a texture, then you need to update the shader to correctly deal with the allwinner specific tiled format
<ssvb> pfdm: the whole memory area reserved by cedar can be wrapped as one big ump buffer (same as framebuffer), there is no need to allocate anything
<ssvb> pfdm: the documentation about the tiled format can be found in a10/a20 manual
<pfdm> ssvb: Can you point me where i can find some example of using this big ump buffer ?
<pfdm> ssvb: thanks i'll take a look at the manual for the tiled format.
<ssvb> pfdm: haha, finding an example of using ump buffer as a texture is the most interesting part, something might be in http://malideveloper.arm.com/develop-for-mali/sample-code/accessing-a-camera-device-under-linux/
<ssvb> pfdm: but I have not checked this example, because downloading requires to agree to some EULA
<libv> "the shader"?
<ssvb> pfdm: an example of mapping a big ump buffer is here - https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_ump.c (the same can be done for cedar)
<ssvb> libv: something to convert yuv to rgb for xbmc purposes
<wingrime> ssvb: we can easy remake allocator around blob to any interface we like
<ssvb> wingrime: allocating/wrapping/mapping ump is not a problem
<libv> the mali should handle such formats just fine, i just haven't bothered, as there are more important things to do first
<wingrime> ssvb: than I wait patchset to UMP for linux-sunxi
<ssvb> wingrime: huh?
<wingrime> ssvb: you talked, thats already backported?
<ssvb> wingrime: an example of *using* ump buffer as a texture from gles code is what we need
<pfdm> libv: such format ? the tiled one ?
<pfdm> ssvb: so mapping the cedar buffer will be on kernel side ?
<ssvb> pfdm: yes
<wingrime> interfaces made simple stuff so complicated
<ssvb> pfdm: just feed a range of physical memory pages to "ump_dd_handle_create_from_phys_blocks" function, and then somehow return the resulting ump secure id to the userspace (via a new ioctl?)
eebrah_ has quit [Read error: Connection reset by peer]
<ssvb> wingrime: which interfaces?
<wingrime> ssvb: we talking about simple allocation, I simply wonder why such things like secureid exsist
<ssvb> wingrime: why do file descriptors exist?
<ssvb> wingrime: we need some sort of a handle to identify the buffer, and ump secure id serves this purpose
<wingrime> ssvb: heh, address already repesent itself
<ssvb> a physical address in the kernel space? you want the userspace to deal with it?
<wingrime> ssvb: cedar blob deal
<ssvb> that would be a big freaking security hole (and allwinner/sunxi kernel code is full of these holes)
torindel has quit [Ping timeout: 248 seconds]
<wingrime> ssvb: but thats bad example
<pfdm> ssvb: ok i think i can do this part. the cedar module should contain start and end of physical block ?
torindel has joined #linux-sunxi
eebrah_ has joined #linux-sunxi
n01 has quit [Ping timeout: 245 seconds]
<wingrime> ssvb: but as I remeber, every have own , presonal adresses, but kernel part mapped for all, and repesent same pages on every process
<wingrime> *every process
<wingrime> but I not sure how it works currently on ARM and linux
<ssvb> ok, imagine that you have a kernel module, and the userspace code asks it to write something to some particular physical address in memory
<wingrime> ssvb: you will get #GPF on x86
<ssvb> heh
<wingrime> and, kernel code will decide whats going on, and what to do
<wingrime> some times it can be not error, just some interface between user-kernel space
<ssvb> ok, imagine that you have a web server, and the remote client submits some sort of memory address in your web server address space where it wants some data to be written to
<wingrime> ssvb: bad example
<ssvb> GPF to the rescue again?
eebrah_ has quit [Read error: Connection reset by peer]
<ssvb> it's exactly the same example
<wingrime> ssvb: but you still can use pointer as handle and request some RPC
<pfdm> wingrime: if i write the mapping for cedar and ump , you will add it the 3.4 branch even though it's gonna disapear in mainline ?
<wingrime> pfdm: if this not break something else , indeed
<wingrime> pfdm: but othere also have opinion
<wingrime> *others
eebrah_ has joined #linux-sunxi
<wingrime> pfdm: so, I will just ACK it,
<pfdm> wingrime: agree, i'm trying to see if the effort is worth, or should i look for an acceptable solution that will last ?
<ssvb> pfdm: we just need a userspace sample gles code which uses ump textures to verify that it is useful for something
<ssvb> pfdm: for testing purposes you can just allocate ump buffers without touching the kernel
<wingrime> pfdm: just for OSD and UI, that not costs time
<ssvb> pfdm: which part of it?
<pfdm> ssvb: i miss read, it does not contains any gles code
eebrah_ has quit [Read error: Connection reset by peer]
<wingrime> pfdm: gralloc is just android allocator for GPU
<ssvb> pfdm: "Updating a texture without copying memory under Android", "In order to create an EGLImage, the native buffer handle is required first"
<ssvb> pfdm: I guess for Linux and Mali400, instead of "native buffer handle" it might be ump handle or ump secure id
<pfdm> so i need the handle from the ump secure id
<pfdm> ssvb : ok
<ssvb> well, you need to do a bit of research, I'm only giving some pointers (maybe even wrong ones)
<pfdm> ssvb: i can look into gralloc function and see how they manage ump.
wigyori has joined #linux-sunxi
<wigyori> hi
<ssvb> pfdm: I only know that this stuff surely works, because UMP buffers for window backing buffers are used by the GLES compositing window managers in X11 desktop
<pfdm> ssvb: can i get a ump handle in userspace if i managed to get the physcal adress ?
<ssvb> pfdm: but for X11 we have a "black box" mali binary blob in between, so it makes things more obscure than necessary
<pfdm> ssvb: this way i don't need to modify the kernel right know, just the ump and get a test file see if i can render it
<ssvb> pfdm: you can't get a ump handle for arbitrary physical address without the assistance from the kernel
eebrah_ has joined #linux-sunxi
<ssvb> pfdm: you don't need the ump buffer to be shared with cedar, then you can just allocate it from userspace (with no guarantee for its physical placement)
<ssvb> pfdm: via something like "ump_buffer_handle = ump_ref_drv_allocate(buffer_size, UMP_REF_DRV_CONSTRAINT_NONE)"
<ssvb> pfdm: you can read libump headers to get the idea about its userspace API
<pfdm> size should be coherent
<wingrime> pfdm: yeax,cedar ioctl give you all info needed for it from cedar side
eebrah_ has quit [Read error: Connection reset by peer]
FR^2 has quit [Quit: und weg...]
<pfdm> wingrime: ok , i remember someone has produced a small player example mpeg2 or h264 using the blob ?
<pfdm> may be in the RE section of the wiki ?
<ssvb> pfdm: yes, the kernel part of cedar ump wrapping is trivial
<wingrime> pfdm: mpeg example behaves using ffmpeg
<wingrime> pfdm: also, there RE about VP8, also simple player
eebrah_ has joined #linux-sunxi
<pfdm> wingrime: the mpeg one is using cedar ?
<wingrime> yes
<pfdm> wingrime: ssvb: great i'll start there . thanks for enriching discussion .
arokux2 has joined #linux-sunxi
<arokux2> oliv3r: ping
rz2k has quit []
<wigyori> hi - on the a13, are there any known issues with the mainline ehci driver? based off sunxi-next-usb, i'm getting "failed to get usb_phy clock"
torindel has quit [Ping timeout: 245 seconds]
<wigyori> i've cherrypicked the clock and the usb commits, and put them into a 3.12-rc7 tree
torindel has joined #linux-sunxi
ZetaNeta has quit [Quit: Leaving]
<arokux2> wigyori: hey
<arokux2> wigyori: where is your sunxi-next-usb from?
<wigyori> github/linux-sunxi
<arokux2> wigyori: what is your board exactly?
arokux2 has quit [Read error: Connection timed out]
<wigyori> that's an a13 olinuxino, one of the older revs
arokux2 has joined #linux-sunxi
<wigyori> that's an a13 olinuxino, one of the older revs
pitillo has quit [Ping timeout: 272 seconds]
<wigyori> wifi-dev rev.b
<arokux2> wigyori: I have the same board and tested my patches on it.
pitillo has joined #linux-sunxi
<arokux2> wigyori: does sunxi-next-usb work if not cherrypicking?
<arokux2> wigyori: i.e. have you tried to build this branch and boot it?
<arokux2> wigyori: (please ping me in your messages)
<wigyori> arokux2: sure, that's what i'll do now
<arokux2> wigyori: ok, tell me how it worked then
pacopad has quit [Quit: pacopad]
Black_Horseman has quit [Remote host closed the connection]
Black_Horseman has joined #linux-sunxi
vinifr has quit [Remote host closed the connection]
Seppoz has quit [Remote host closed the connection]
Seppoz has joined #linux-sunxi
<wigyori> arokux2: [ 0.634983] hub 1-0:1.0: USB hub found
<wigyori> arokux2: so the -next tree gets this far - looks like i screwed up something
<arokux2> wigyori: try lsusb
<wigyori> arokux2: got no rootfs (no emac on this one, only an rtl8188)
<wigyori> but seeing the -usb-next tree with a working usb on this board is fine enough for me now - i'll start looking what i've missed
<arokux2> wigyori: I have added USB support to U-Boot, not published the patches yet. with it you can use usb2ethernet dongle with U-Boot too and have ethernet from the very beginning! :)
<wigyori> heh, nice :)
<arokux2> wigyori: you'd need a special usb2ethernet - supported by U-Boot.
<arokux2> wigyori: everything works, I've tested it.
<arokux2> wigyori: what is you general plan?
<wigyori> arokux2: dusting off the openwrt port i've been working on the summer (since august i had no time to work on it), and push it into owrt trunk
<arokux2> "failed to get usb_phy clock" <-- make sure you have pulled in dts patches - dtsi + board dts
<arokux2> wigyori: I suggest you try this usb2ethernet thing - no need to fiddle with sd card then
<arokux2> wigyori: just tell me to push u-boot patches and i'll do it sonner.
<arokux2> sooner*
<pfdm> ssvb: i guess i can't call that function from userspace ump_dd_handle_create_from_phys_blocks ?
<ssvb> pfdm: right, you can't
<ssvb> pfdm: that would be obviously a major security hole if it worked
<pfdm> ssvb: ok so to bypass that for now, i need to copy my cedar memory to the ump memory reserved by ump_ref_drv_allocate
<pfdm> ssvb: is there a userspace function to do that /
<pfdm> ?
<ssvb> pfdm: memcpy?
<pfdm> ssvb: i thought more complicated...but simple is fine
<ssvb> pfdm: actually you also need to do tiled->planar yuv conversion, otherwise the rest of xmbc will not like it
<ssvb> pfdm: actually I would suggest trying some experiments with a really small gles test application, even initially without getting cedar involved
<arokux2> mnemoc: http://linux-sunxi.org/Windowmanager <-- is this spam?
<pfdm> ssvb: but then i need to find this tiled format texture . the small mpeg decoder is providing it. or you mean try to convert even without allocating in ump ?
<ssvb> pfdm: step one - try to use ump for any texture (even rgb) to see whether it works correctly
<ssvb> pfdm: step two - read the a20 manual about the tiled yuv format and use the decoder to verify that you got it right
<pfdm> ssvb: ;) got you.
wingrime has quit [Ping timeout: 248 seconds]
eebrah_ has quit [Ping timeout: 240 seconds]
eebrah_ has joined #linux-sunxi
wingrime has joined #linux-sunxi
eebrah_ has quit [Read error: Connection reset by peer]
shineworld has quit [Ping timeout: 245 seconds]
eebrah_ has joined #linux-sunxi
<arokux2> hno: please take a look at new USB patches for U-Boot (on the ML)
<arokux2> oliv3r: ^
eebrah_ has quit [Read error: Connection reset by peer]
<arokux2> good night!
arokux2 has quit [Remote host closed the connection]
Sonicadvance1 has joined #linux-sunxi
<wigyori> arokux1: okay, got it working, got missing the sun5i usbclocks support, so "no big deal" :)
<wigyori> arokux1: thanks for the help, i'm fine now
<wigyori> ah, goodnight
Sonicadvance1 has quit [Remote host closed the connection]
eebrah_ has joined #linux-sunxi
techn__ has joined #linux-sunxi
techn_ has quit [Ping timeout: 272 seconds]
eebrah_ has quit [Read error: Connection reset by peer]
<kenny> looks like 3.12 went final. Is there going to be a sunxi branch of it soon?