Closed zeule closed 9 years ago
Can you test with this tar ball? https://drive.google.com/file/d/0B7k0IbH_XHW4RlZNSjBUY3VqVEE/view?usp=sharing This is a test-purpose version and lots of functions doesn't work, but I want to know this can help you. To compile and run this cmplayer, please follow next:
$ ./configure
$ make
$ build/cmplayer
This version is free of glitches, but seems to be also free of HW-acceleration. At least the combobox in the settings window is empty and in the playback information acceleration is stated to be disabled. CPU usage is higher also.
Can you see VAAPI[8bit] in playback information like here?: https://cloud.githubusercontent.com/assets/3054729/5619788/43bbfa40-9536-11e4-9267-46cfbb2a9b3d.png All options related with H/W acc are disabled in that version because it requires more changes, so I disabled temporarily. Instead, CMPlayer tries automatic hwacc like mpv's hwdec=auto option. CPU usage might be higher but it should not be over twice of original version. Can you post the actual value of CPU usage?
Indeed, it is not significantly higher. Here is screenshot, please: https://cloud.githubusercontent.com/assets/394621/5620085/82367b96-9528-11e4-9aff-b66486d01191.png
Okay, it seems fine though I cannot check the CPU usage because I think your shcreenshot is taken when paused. Thank you. I'll close this when that version merged into upstream.
Fixed. If the problem still exists, please let me know. Thank you!
Wait-wait! :) Indeed, there are not image distortions anymore, but the CPU usage by CMplayer is incomparable to that of mpv. For example, 1.5-3% by mpv and 8-10% by cmplayer. I see that mpv reports "VO: [vaapi] 720x480 vaapi" while cmplayer "[mpv/cplayer] VO: [opengl-cb] 720x480 vaapi". Could it be the reason?
As total or per core average?
Also, please check the CPU usage of mpv with --vo=opengl --hwdec=vaapi
option.
By the way, basically, CMPlayer use more CPU than mpv because it does more things on top of mpv.
With --vo=opengl mpv uses up to 4%. All of these are numbers given by htop via its meter PERCENT_CPU.
Can you test without hardware hacceleration? run mpv with --hwdec=no --vo=opengl and turn off hwacc in CMPlayer.
mpv: 11-15% cmplayer: 19-22%
In that case it's certain that the hwacc works well because the difference is kept in close range. However, I'm not sure why CMPlayer consumes much more CPU. Of course CPU usage can vary depending the environment. Are you sure there was no such big difference in previous version?
Checked with cmplayer 0.8.16, CPU usage is almost the same as with current git master. I guess this issue may be closed then.
Okay, thank you for checking.
Thank you for working on this issue!
One possibility for higher CPU usage: check the volume normalizer and playback speed. Please turn off volume normalizer and set playback speed to 100% if not and measure the CPU usage.
Volume normalization is disabled, playback speed is default. BTW, when i press 'n', cmplayer crashes with SIGSEGV:
Not reproducible here. Please make a new issue about this with next informations:
Oh, and another idea about CPU usage: channel mixing If the channel layouts of input and out mismatch, CMPlayer mixes them and in order to avoid akward clipping issue, CMPlayer makes a use of logarithm which requires some CPU resource.
I've done check of all settings you've mentioned. Bomi uses more CPU then mpv always, with any of the supported by my VAAPI device codecs.
Well, then I'm out. If you consider it as serious problem, one last option: send me the file so that I can test it, although it doesn't gaurantee anything.
Yes, it seems important to me. Why do I use video decoding acceleration at all? Because it decrease CPU utilization to a level where CPU cooler is almost inactive. However, on my laptop with typical usage, additional load from bomi triggers the cooler. As for the file, I will try to find a sample online.
FYI, activation of cooler means nothing. It'll be turned on anyway for CPU without H/W acc and for GPU with H/W acc.
Let me, please, measure heat, generated by device in front of me, by myself :) I just share my experience with you. If it does not fit into your ideas about the reality, the reality is not guilty :).
Oh, I didn't mean to blame you. If you feel unpleasant, I'm sorry. I just wanted to point it out that hardware acceleration is not a magic.
Thank you for your words and do not worry, please!
I understand why it is (and always was) more efficient than universal CPU. I just pointed out that with my hardware and habits, playing the FHD x264 10bit movie with or without acceleration means active or silent CPU cooler. And since if is VAAPI, discrete GPU and it's cooler are not involved at all.
That's odd. There's no API which supports 10bit format in Linux.
I may be wrong. I was running test for this half of an hour ago. I belive the file was 10 bit but it might be my mistake. Then the same is true for 8 bit. Will check the file tomorrow.
Okay, let me know any update.
I was wrong. It has 8 bit depth.
Any news about measuring temp.?
Not yet. Waiting for enough dust to be collected by the CPU heat sink.
When playing H264 video with High@L4.0 profile, cmplayer shows sporadic glitches on screen (green squares). Mpv plays these video fine (with va-api acceleration).
Here is partial mediainfo:
mpv -v output:
cmplayer --debug output: