vp9 high bit depth, vp9 lossless profile
so I played around with vp9 enciding a bit more. apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so. I also tried to remove (currently unused?) libavdevice build from ffmpeg, and fix termux breakage (libandroid-shmem lib was missing) try attached patches and check if vdpau/vaapi still works?
Andrew, Initial test shows Vaapi working with all 3 patches in. Have to boot another computer later this morning to check vdpau and will let you know. On Mon, Mar 7, 2022 at 8:51 PM Andrew Randrianasulu via Cin < [email protected]> wrote:
so I played around with vp9 enciding a bit more. apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so.
I also tried to remove (currently unused?) libavdevice build from ffmpeg, and fix termux breakage (libandroid-shmem lib was missing)
try attached patches and check if vdpau/vaapi still works? -- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
Checked vp9 related patches into GIT. so I played around with vp9 enciding a bit more.
apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so.
Tested on my laptop and the lossless file is, as expected, 15 times larger but looks better. Thank you Andrew. Will compiling with vp9 multibit affect any loading or rendering of vp8/vp9? I will try to check this and test it too.
I also tried to remove (currently unused?) libavdevice build from ffmpeg, and fix termux breakage (libandroid-shmem lib was missing)
This I did not check in yet because my early morning tests were flawed -- I only had the configure.ac patch portion in and not the thirdparty/Makefile patch because I was testing the vp9 multibit. So I have to start over with that and right now am at least compiling on Debian 9.1 32-bit just to make sure it compiles.
try attached patches and check if vdpau/vaapi still works?
Got to redo my tests. I am concerned about potential impact, not to vdpau/vaapi, but rather to v4linux2 and alsa and captures from devices.
On Wednesday, March 9, 2022, Phyllis Smith <[email protected]> wrote:
Checked vp9 related patches into GIT.
so I played around with vp9 enciding a bit more.
apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so.
Tested on my laptop and the lossless file is, as expected, 15 times larger but looks better. Thank you Andrew. Will compiling with vp9 multibit affect any loading or rendering of vp8/vp9? I will try to check this and test it too.
not sure why this still not default - may be due to libvpx size or speed?
I also tried to remove (currently unused?) libavdevice build from ffmpeg, and fix termux breakage (libandroid-shmem lib was missing)
This I did not check in yet because my early morning tests were flawed -- I only had the configure.ac patch portion in and not the thirdparty/Makefile patch because I was testing the vp9 multibit. So I have to start over with that and right now am at least compiling on Debian 9.1 32-bit just to make sure it compiles.
try attached patches and check if vdpau/vaapi still works?
Got to redo my tests. I am concerned about potential impact, not to vdpau/vaapi, but rather to v4linux2 and alsa and captures from devices.
yeah, while I think this specific library not used yet in CinGG at least - I am not sure... so, thanks for testing..
so I played around with vp9 enciding a bit more.
apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so.
Tested on my laptop and the lossless file is, as expected, 15 times larger but looks better. Thank you Andrew. Will compiling with vp9 multibit affect any loading or rendering of vp8/vp9? I will try to check this and test it too.
not sure why this still not default - may be due to libvpx size or speed?
Good news. The directory containing the libvpx code is only 4 M larger and the bin/cin file is just a little bigger (about 150KB if I calculated it correctly) with the patch in. But I did not check the compile speed yet. File size differences are in: 96K ./vpx_scale/generic 144K ./vpx_scale 424K ./vp9/encoder/x86 4.0M ./vp9/encoder 420K ./vp9/decoder 68K ./vp9/common/x86 1.0M ./vp9/common 5.8M ./vp9 2.6M ./vpx_dsp/x86 7.0M ./vpx_dsp There are these files that maybe can be gotten rid of in the compile if there are options for that if necessary later: 3.3M ./docs 240K ./examples 1004K ./third_party/googletest 448K ./tools More good news is that 1 test rendering the same file with the enable in and with the enable out came out exactly the same. Now on to testing 0002-Remove-libavdevice ... by Andrew.
On Wednesday, March 9, 2022, Phyllis Smith <[email protected]> wrote:
so I played around with vp9 enciding a bit more.
apoarently there is vp9 lossless mode, and additionally libvpx can support high bit depth if configured so.
Tested on my laptop and the lossless file is, as expected, 15 times larger but looks better. Thank you Andrew. Will compiling with vp9 multibit affect any loading or rendering of vp8/vp9? I will try to check this and test it too.
not sure why this still not default - may be due to libvpx size or speed?
Good news. The directory containing the libvpx code is only 4 M larger and the bin/cin file is just a little bigger (about 150KB if I calculated it correctly) with the patch in. But I did not check the compile speed yet. File size differences are in: 96K ./vpx_scale/generic 144K ./vpx_scale
424K ./vp9/encoder/x86 4.0M ./vp9/encoder 420K ./vp9/decoder 68K ./vp9/common/x86 1.0M ./vp9/common 5.8M ./vp9
2.6M ./vpx_dsp/x86 7.0M ./vpx_dsp
There are these files that maybe can be gotten rid of in the compile if there are options for that if necessary later: 3.3M ./docs 240K ./examples 1004K ./third_party/googletest 448K ./tools
I think I created small local hack fir deleting all *. o files, just run this line from thirdparty dir after all components built successfully: objrem: find . -type f -name '*.o' -delete {from thirdparty/Makefile} this way you will have a bit more space for building/linking cingg itself
More good news is that 1 test rendering the same file with the enable in and with the enable out came out exactly the same. Now on to testing 0002-Remove-libavdevice ... by Andrew.
if this change does not break normal cingg input/output it may even balance out prev. change in terms of compilation soeed and size I wonder how much slower vp9 profiles become if you choose 10/12 bit vp9? And how quality of encode/decode compare with normal 8bit vp9.. but this is not urgent. ps: you also hoped to push jp2.dfl file?
I think I created small local hack fir deleting all *. o files, just run this line from thirdparty dir after all components built successfully:
objrem: find . -type f -name '*.o' -delete
{from thirdparty/Makefile}
this way you will have a bit more space for building/linking cingg itself
More good news is that 1 test rendering the same file with the enable in and with the enable out came out exactly the same. Now on to testing 0002-Remove-libavdevice ... by Andrew.
if this change does not break normal cingg input/output it may even balance out prev. change in terms of compilation soeed and size
Not sure if I will even bother with this myself was just pondering it.
I wonder how much slower vp9 profiles become if you choose 10/12 bit vp9? And how quality of encode/decode compare with normal 8bit vp9.. but this is not urgent.
I will test speed yet.
ps: you also hoped to push jp2.dfl file?
Oops, thanks for the reminder as I had forgot to push this the other day.
Andrew, still testing this below:
I also tried to remove (currently unused?) libavdevice build from ffmpeg, and fix termux breakage (libandroid-shmem lib was missing)
vaapi and vdpau seem to be working correctly bluray and dvd working good TV record working but I am having trouble with audio on playback for another reason that I have to analyze yet screencapture from the web works well have not yet tested the following because have to get the equipment ready and figure out how to do this based on instructions at: https://cinelerra-gg.org/download/CinelerraGG_Manual/Transfer_VHS_DVD_Media_...
try attached patches and check if vdpau/vaapi still works?
participants (2)
-
Andrew Randrianasulu -
Phyllis Smith