ChanServ changed the topic of #linux-rockchip to: Rockchip development discussion | Wiki at | Logs at | ML at
premoboss has joined #linux-rockchip
libv_ is now known as libv
premoboss has quit [Quit: Sto andando via]
nighty has joined #linux-rockchip
vagrantc has quit [Quit: leaving]
cnxsoft has joined #linux-rockchip
lkcl has quit [Ping timeout: 255 seconds]
deepbook5broo has joined #linux-rockchip
deepbook5broo has left #linux-rockchip [#linux-rockchip]
indy has quit [Ping timeout: 240 seconds]
indy has joined #linux-rockchip
lkcl has joined #linux-rockchip
descend is now known as descend-irc
beeno has joined #linux-rockchip
zzarr has joined #linux-rockchip
lkcl has quit [Ping timeout: 260 seconds]
muvlon has quit [Ping timeout: 245 seconds]
muvlon has joined #linux-rockchip
jarekin has quit [Ping timeout: 260 seconds]
maz has joined #linux-rockchip
matthias_bgg has quit [Quit: Leaving]
matthias_bgg has joined #linux-rockchip
<LongWork> stdint: Do you know if the mali driver for tinker has dmabuf import opengl support ?
<stdint> LongWork, yes, you could look the gl element in rkvideo
<stdint> rkvideo plugin
<LongWork> and this can't do 4K right ?
<stdint> LongWork, No, but the performance is not enough
<stdint> I don't think the resolution or buffer size would be a limitation
<LongWork> so GPU would use more memory bandwidth than vop does ? or is it a gpu power issue ?
<stdint> GPU can't avoid an memory copy
<LongWork> even with dmabuf ?
<LongWork> i thought that was the whole purpose of dmabuf import
<LongWork> do zerocopy
<stdint> without dmabuf there would be the other copy
<phh> LongWork: you would do vpu => ram => gpu => ram => vop => hdmi
<phh> while scaling with vop you do
<phh> vpu => ram => vop => hdmi
<phh> so first case you do 2 writes 2 reads, first case you do 1 write 1 read
<LongWork> gotcha
<phh> stdint: I guess VPU doesn't support ATSC? (I don't even know if it would make sense)
<LongWork> stdint: i was also thinking abourt AFBC (Arm framebuffer compression)
<phh> ok, not ATSC, AFBC
<phh> I was almost there.
<LongWork> hehe
<LongWork> i'm pretty sure mali supports that
<LongWork> just not sure about vpu
<stdint> the document doesn't cover that
<LongWork> afbc would like squeeze data amont by 2
<LongWork> funny because on amlogic socs, vpu supports afbc, but Mali4(à doesn't
<LongWork> mali450
<phh> LongWork: haha, nice
<LongWork> makes a huge difference though
<LongWork> with afbc would maybe be possible to play 4K thru gpu
<phh> LongWork: but really, noone does that, even the extremely powerful nvidia shield tv
<LongWork> not sure what shield uses
<phh> tegra x1
<LongWork> yeah i mean i know the gpu, just on't know what mechanism they use
<LongWork> @stdint : can still gl + dmabuf output properly 4K video in 1080p ?
<phh> LongWork: no, you'll need to have the vpu scale to 1080p
<phh> (or rga, but it would be better if it is vpu)
shahid_ has joined #linux-rockchip
<LongWork> phh: by the way do you have a 4K display ?
<phh> yes, but no I don't have 4k output from a rk3288 box
<LongWork> you mean it doesn't work ?
<LongWork> stdint: is there any plan to get this fixed ?
<phh> I mean that I still only have mainline kernel booting, which doesn't have 4k
<LongWork> ok
<LongWork> it behaves very unapropriately for me even with rk kernel, no matter what the framerate is
<LongWork> I usually end up in a broken state when i switch to it, that is with a 4K video mode, but X11 will show in like 1/4th of the screen (upper left), upper right part contains garbage, low part is black .. and it's totally unusable
<LongWork> it's pretty wierd to me for a "4K" advertised box , that 4K modes are broken :)
<stdint> LongWork, it can't be
<LongWork> it can't be what ?
<LongWork> it can't be fixed ?
<LongWork> or it can't be broken ?
shahid_ has quit [Ping timeout: 260 seconds]
nighty has quit [Quit: Disappears in a puff of smoke]
<stdint> LongWork, the mali library keep the same to handle the buffer as ARM gave it to rockchip
<stdint> anyway, none of the mali core could support 4K video rendering
<stdint> smoothly
<LongWork> ok
<LongWork> stdint: any work about 4K broken modes ? is this going to be adressed ?
afaerber has quit [Quit: Leaving]
wzyy2 has joined #linux-rockchip
<wzyy2> "X11 will show in like 1/4th of the screen".
<wzyy2> You should use xerver form rockchip-linux which don't enable fake 4k ui.
<LongChair> isn't that what the TinkerOS image uses ?
<wzyy2> TinkerOS use a "fake 4k" xserver and it should use corresponding kernel.
<LongChair> wzyy2: is the xserver replaceable easilly on the tinekr image ?
<LongChair> i mean it might not be as straight forward as build the rockchip one & make install right ?
<ayaka> anyway, I would suggest you to use the kmssink to know benchmark
Myy has joined #linux-rockchip
<LongChair> remove evdev ?
<wzyy2> xserver-xorg-input-evdev
diego71 has joined #linux-rockchip
diego71_ has quit [Ping timeout: 268 seconds]
<wzyy2> GLES have room for improvement.
<wzyy2> gles + dmabuf output should support 1080p 60fps but i don't why it only reach 1080p 20-30fps.
<LongChair> yeah that sounds very odd
<wzyy2> I have test it in qt + eglfs and it can easily reach 1080 60ps.
<wzyy2> I don't know why i can't get this permance in x11.
<wzyy2> performance
<wzyy2> - -There must be something wrong.
<LongChair> wzyy2: do you know if qt can run using gles and using the drm dmabuf at the same time ?
<Myy> Did you try analysing it with streamline ?
<Myy> Or at least try recording a 60 fps session with apitrace, and comparing it with a 20-30 fps session recording
<LongChair> wzyy2: it also looks like the gl drivers i have on teh tinker image don't support dmabuf import :/
<LongChair> i don't have the gl extension for this
<wzyy2> gles
<wzyy2> libmali form tinker image should support it.
<LongChair> egl extension
<LongChair> EGL_KHR_image_pixmap EGL_KHR_partial_update EGL_KHR_config_attribs EGL_KHR_image EGL_KHR_image_base EGL_KHR_fence_sync EGL_KHR_wait_sync EGL_KHR_gl_colorspace EGL_KHR_get_all_proc_addresses EGL_IMG_context_priority EGL_ARM_pixmap_multisample_discard EGL_KHR_gl_texture_2D_image EGL_KHR_gl_renderbuffer_image EGL_KHR_create_context EGL_KHR_surfaceless_context EGL_KHR_gl_texture_cubemap_image EGL_EXT_create_context_robustness EGL_KHR_cl_eve
<LongChair> this is all i have
<LongChair> wzyy2: any ways to overcome this ?
<Myy> LongChair: What's the vendor string of your EGL libraries ?
scelestic has quit [Read error: Connection reset by peer]
<wzyy2> You might use mesa driver.
<LongChair> if i have mesa installed i can't get dmabuf ?
<wzyy2> check symlinks for
<LongChair> lrwxrwxrwx 1 root root 10 Oct 15 07:34 /usr/lib/arm-linux-gnueabihf/ ->
<phh> LongWork: well what does eglinfo says?
<LongChair> i don't seem to have eglinfo
<LongChair> any idea which package this is coming with ?
<phh> I think you need to build it by hand
<wzyy2> "arm-linux-gnueabihf-objdump -s | grep EXT_image_ -C 6"
<wzyy2> 10fb510 70204547 4c5f4558 545f696d 6167655f p EGL_EXT_image_ 10fb520 646d615f 6275665f 696d706f 72740000 dma_buf_import.
<LongChair> hmmm
<LongChair> so there looks to be something in the lib
<LongChair> but eglQueryString will not return anything else than what i pasted above
wzyy2 has quit [Ping timeout: 260 seconds]
<phh> LongWork: and if you try to get EGL's vendor name, you do get arm?
<LongChair> yes
mrjay has joined #linux-rockchip
<LongChair> mpv will list that info :
<LongChair> [vo/opengl] EGL_VERSION=1.4 Midgard-"r9p0-05rel0"
<LongChair> [vo/opengl] Trying to create GLES 3.x context.
<LongChair> [vo/opengl] chose visual 0x21
<LongChair> [vo/opengl] Could not bind API!
<LongChair> [vo/opengl] Trying to create Desktop OpenGL context.
<LongChair> [vo/opengl] EGL_CLIENT_APIS=OpenGL_ES
<LongChair> [vo/opengl] EGL_VENDOR=ARM
<LongChair> [vo/opengl] GL_VERSION='OpenGL ES 3.1 v1.r9p0-05rel0-git(f980191).e4ba9e4c6ff8005348d0332aae160089'
<LongChair> [vo/opengl] Detected GLES 3.1.
<LongChair> [vo/opengl] GL_VENDOR='ARM'
<LongChair> [vo/opengl] GL_RENDERER='Mali-T760'
<LongChair> [vo/opengl] GL_SHADING_LANGUAGE_VERSION='OpenGL ES GLSL ES 3.10'
<phh> funny, I thought current release was r12/r13, but wzyy2 mentioned r9, so I guess it's indeed the current one...
<Myy> If the symbol is in the library, why not try to get it using dlsym, and try to use it ?
<LongChair> well i oculd ignore thes extensions fetching .. but that looks wrong :)
mrjay has quit [Quit: Page closed]
<Myy> Try to md5sum the library and compare it with someone who has a working version. There might be some discrepencies there ( ̄〜 ̄)
Omegamoon has joined #linux-rockchip
<phh> it has the proper version number, so there is not much doubt he has the correct one
<LongChair> the odd thing is that it seems to have the symbo
<LongChair> but it doesn't list it ...
afaerber has joined #linux-rockchip
scelestic has joined #linux-rockchip
<phh> possibly there is a runtime check for something else
nighty has joined #linux-rockchip
<LongChair> yeah, i'm not too sure how to troubleshoot this
afaerber has quit [Quit: Leaving]
afaerber has joined #linux-rockchip
mac-l1 has joined #linux-rockchip
<mac-l1> hi guys. just made a first kodi .deb package for vpu/mpp including zerocopy/eglimage to test. can somebody check if it works?
<mac-l1> just download and dpkg -i the deb package and then apt-get install -f
<rperier> Mark Yao is there ?
<mac-l1> LongChair: i currently have exact same GL_VERSION on my miqi and it supports dma_buf import
<LongWork> mac-l1: well does eglQueryString(...,EGL_EXTENSIONS) returns it properly ?
<mac-l1> LongWork: from kodi i get 13:24:44.877 T:2974375936 DEBUG: EGL_EXTENSIONS: EGL_KHR_image_pixmap EGL_KHR_partial_update EGL_KHR_config_attribs EGL_KHR_image EGL_KHR_image_base EGL_KHR_fence_sync EGL_KHR_wait_sync EGL_KHR_gl_colorspace EGL_KHR_get_all_proc_addresses EGL_IMG_context_priority EGL_ARM_pixmap_multisample_discard EGL_KHR_gl_texture_2D_image EGL_KHR_gl_renderbuffer_image EGL_KHR_create_context EGL_KHR_surfaceless_
<mac-l1> but it is supported as i use it
<LongWork> so you're having the same issue, the symbol is there, but it's not reported as being supported ...
<LongWork> @ayaka ^^ :)
mrjay has joined #linux-rockchip
mrjay has quit [Client Quit]
cnxsoft has quit [Quit: cnxsoft]
lkcl has joined #linux-rockchip
<LongWork> mac-l1: what kind of performance can you achieve with vpu + gl & dmabuf ?
afaerber has quit [Quit: Leaving]
vagrantc has joined #linux-rockchip
afaerber has joined #linux-rockchip
wzyy2 has joined #linux-rockchip
<wzyy2> wow, kodi with mpp.
<wzyy2> mac-l1, could we get the source code?
<wzyy2> hi, rperier, mark don't use IRC, you can send him email.
<rperier> I have a fix for dwhdmi , I will Cc: him
LongWork has quit [Quit: LongWork]
<mac-l1> wzyy2: yes, you can get it, will share later after bit more testing and stable and code cleanup (quite some hacks now...)
<mac-l1> also hope to add rockchip support in mainstream kodi repo
<mac-l1> can you confirm if it works?
<mac-l1> LongChair: not really tested yet. however 4K is not good yet, kodi decides to drop frames quite soon, especially when compared with my simple macffplay
<mac-l1> will try to solve 4K by using RGA to scale down to 1080p
lkcl has quit [Ping timeout: 240 seconds]
<mac-l1> also default display fps reported by kodi seems to default to 30fps (i use drm vsync for master clock for video player)
lkcl has joined #linux-rockchip
<mac-l1> drm vsync clock detects 60fps
Myy has quit [Quit: Leaving]
wzyy2 has quit [Ping timeout: 260 seconds]
matthias_bgg has quit [Quit: Leaving]
paulk-collins has quit [Quit: Leaving]
<LongChair> mac-l1: i also saw soem wierd stuff with fps much lower than expected. just wondering if some kernel stuff isn't waiting for 2 vsync instead of one
lkcl has quit [Ping timeout: 240 seconds]
beeno has quit [Ping timeout: 240 seconds]
jelly has quit [Ping timeout: 240 seconds]
jelly-home has joined #linux-rockchip
jelly-home is now known as jelly
vagrantc has quit [Ping timeout: 255 seconds]
lkcl has joined #linux-rockchip
mac-l1 has quit [Ping timeout: 260 seconds]
Omegamoon has left #linux-rockchip [#linux-rockchip]