As you did see in the post of mglae , i dont have a chance with Kodi 17. So i give up AMD with VAAPI and 4k HW decoding.
piotrasd has build a special AMD version see this thread Experimental build AMD GPU (Vega, Ryzen APU)
Posts by sky42
-
-
Thanks for that info. I will switch off vaapi for AMD in my next releases.
-
some updates ... what was done:
8.2.5-3.42.2-4.14.48
- just the kernel 14.14.48
8.2.5-5.42.2-4.14.48
- just the kernel 4.14.48
8.2.5-6.42.2-4.16.14
- just the kernel 4.16.14
-
xorg 1.20 released, to build for AMD use kernel from here ~agd5f/linux - Unnamed repository; edit this file 'description' to name the repository.
Do you have a PKG_URL for this AMD kernel for me please. I can not figure out what is the right URL for the kernel packae.mk
-
So it was not just me with the refresh rates. I am very happy to hear that it helped.
-
If i find time i will setup a tvheadend on my 2 Intel LE systems and check if i can see a problem with tvheadend. But please be patient. My tvheadend server is running on my backend server for all my LE boxes. The 3.42.2 is the right choise with all you need for Gemini Lake (i hope), but as little as possible change from 8.2.5, except adding encryption.
Pleaee show me a "ldd <tvheadend binary>", so that i can see about linked libraries.
-
My lastest 7.42.0 is with Mesa 18.1.1. and for AMD i do not know what else to do except make a build with a special AMD kernel. I will look into it on the weekend. My 6.42.2 and 7.42.0 (since kernel 4.16.13) are for AMD and Gemini Lake.
-
some updates ... what was done:
a testing version 6.42 with kernel 4.17
8.2.5-6.42.2-201806041418-4.17 HiDrive
- kernel 4.17
- look for -CC suffix if you need CrasyCat DVB drivers, now there are 2 builds for each release
-
correction of the last update ... what was done:
the date needs to be before the kernel version in the filename
8.2.5-7.42.0-201806041407-4.17 HiDrive
- kernel 4.17
- look for -CC suffix if you need CrasyCat DVB drivers, now there are 2 builds for each release
-
some updates ... what was done:
8.2.5-7.42.0-4.17-201806040840deleted look next post- kernel 4.17
- look for -CC suffix if you need CrasyCat DVB drivers, now there are 2 builds for each release
with slightly different timestamps -
I fear that looks like a decoding and/or color space conversion problem and i have no idea what to do. LE 9.0 uses ffmpeg 4.0 and building kodi 17.6 with ffmpeg4 seems not possible, because kodi is kind of hardwired to a ffmpeg version. I just did try and it downloaded 3.1 when only 4.0 was compiled and avariable.
So just for clarification of your test:
You did try 3.42.2, 5.42.2, 6.42.2?
The GLK HMDI bug workaround did also work for 3.42.2 and 5.42.2? That one i needed to rewrite and i am not really a developer and was not sure if it is working at all.
I just build a new version with 4.17 kernel release, but i think that one will not help you.
-
I did not read about you problem in the bug report for 105887 or anywhere else, so i have no idea how to help. You can try both workarounds look in post #80
-
For your Gemini Lake problem you can try
- 8.2.5-5.42.2 (xorg, mesa, xf86 driver close to LE 9.0)
- 8.2.5-3.42.2 (less new software, but i am not sure that it works at all)
- 8.2.5-7.42.0-4.16.13 / 8.2.5-7.42.0-4.17-rc7
- 8.2.5-6.42.2 should be the same as the older build 8.2.5-6.42.1-kernel-4.16.13
and please inform me if there is change or if it is the same
-
some updates ... what was done:
Edit: corrected filename syntax -> the date needs to be before the kernel version in the filename
new file naming scheme
What are these LibreELEC-<Project>.<Architecture>-<LE version>-<F>.42.<V>[-YYYYMMDDHHMM][-<Kernel Version>][-CC] long filenames for
<Project> is like Generic, RPi, RPi2 ...
<Architecture> the CPU architecture of the build
<LE version> what LibreELEC version it is based on
<F> Feature set, higher means more changes/features, but probably less stable
<V> just for counting versions, 0 means testing only
YYYYMMDDHHMM if there is a timestamp -> that is a testing version
<Kernel Version> the kernel version if it is different from the original LE version
-CC a -CC suffix means it is with extra CrasyCat DVB drivers
README.txt update for all download folders
8.2.5-3.42.2-4.14.44
- full build with make distclean before
- new version name .2, because of new naming scheme
- added improve-ir-timeout patches from LE master
- added Gemini Lake HDMI bug 105887 workarounds
to activate put i915.glkhdmi=1 or i915.glkhdmi=2 in /flash/syslinux.cfg
- xf86-video-nvidia 390.59
8.2.5-5.42.2-4.14.47
- full build with make distclean before
- disabled i915 Alpha Features
for Coffee Lake you need now i915.alpha_support=1 in /flash/syslinux.cfg
- new version name .2, because of new naming scheme
- added improve-ir-timeout patches from LE master
- Mesa 3D 18.0.5
- amdgpu r600 / radeonsi now with VAAPI support PR allow using vaapi with r600 and radeonsi by lrusak · Pull Request #2403 · LibreELEC/LibreELEC.tv · GitHub
- PR 2403 did libva --disable-glx and i removed glu dependency, for mesa --enable-va
- added Gemini Lake HDMI bug 105887 workarounds
to activate put i915.glkhdmi=1 or i915.glkhdmi=2 in /flash/syslinux.cfg
8.2.5-6.42.2-4.16.13
- full build with make distclean before
- new version name .2, because of new naming scheme
- Mesa 3D 18.0.5
- amdgpu r600 / radeonsi now with VAAPI support PR allow using vaapi with r600 and radeonsi by lrusak · Pull Request #2403 · LibreELEC/LibreELEC.tv · GitHub
- PR 2403 did libva --disable-glx and i removed glu dependency, for mesa --enable-va
- amdgpu new package amdgpu-firmware with latest firmware for vega10, vega12, vegam, raven from here Index of /~agd5f/radeon_ucode
- package linux now depends on amdgpu-firmware
-
Gemini Lake HDMI bug 105887 workarounds are in the following versions
8.2.5-5.42.2 and later
8.2.5-6.42.2 and later
8.2.5-7.42.0-4.16.13 and later
8.2.5-3.42.2 and later (possibile not working at all, if somebody finds out please inform me)
You need to configure your /flash/syslinux.cfg with i915.glkhdmi=1 or i915.glkhdmi=2 (see below for the parameters)
The patch is taken from here ubuntu-4.15.0-23.25...4.15.0-23.25-glktesting and changed to work with 4.16. For 4.14 i needed to rewrite it more and i am not sure if it is working.
So please tel me if it is working or not and what is the version you are tested.
QuoteDisplay More* Geminilake testing: selectable workaround for the i915 HDMI issue:
- i915.glkhdmi=0 (default/upstream behaviour, will cause broken
output on res change when the pixel clock goes up)
- i915.glkhdmi=1 (effectively disable 12bps support, ie. use
"return false;" at the top of intel_hdmi.c:hdmi_12bpc_possible())
- i915.glkhdmi=2 (insert short delay after disabling the transcoder
to give the level shifter some time to settle)
-
Can somebody please check if i changed the GLK HDMI workarounds correct for kernel 4.14.
The original is from here ubuntu-4.15.0-23.25...4.15.0-23.25-glktesting and for kernel 4.16 the changes are easy.
For 4.14 i needed to change more then i really understand. It looks like the modul parameters have another struct name in 4.14 than in 4.16 and the parameters are totally different to define.
Can please a developer look over my 4.14 patch and check if it should work. The 4.16 does work and i have feedback from the community about that.
Sorry i have no git and you need to download the patch(es) here HiDrive the file kernel-4.16.13-GLK-bug-105887-2-workarounds-2018-05-28.patch
Edit:
It does compile.
-
Settungs -> Player -> Video -> vdpau/vaapi
Try disable vdpau and enable vaapi.
When the video is playing type o on a keyboard for the overlay info. That one shows you what decoding is used and if it is hw accelerated
-
Did you try vdpau and vaapi acceleration? My last build should in theory do vaapi for r600 and radeonsi.
