MPC-HC v2.2.1

[Zetti]

Media Player Classic - Home Cinema v1.8.0 is released.

Changes from 1.7.18 to 1.8.0:

Updates:
Updated LAV Filters to v0.72-12-g14744
Updated MediaInfo DLL to v18.08.1

Changes/additions/improvements:
Added support for decoding AOMedia AV1 video
Added WMA to internal filters list
Allow using externally installed LAV Filters as internal filters when our own LAV Filters folder doesn't exist.
Added advanced option "AllowInaccurateFastseek", which is enabled by default. When enabled fast seek (to keyframe) is allowed to be very inaccurate (max difference 20 seconds) in files with huge keyframe intervals. When disabled the allowed inaccuracy is much lower, when deciding between a fast and normal seek. For example 30% of jump size.
If a playlist entry points to an URL, then display the title/label in player title bar instead of the URL.
Limit max video height returned by Youtube-DL to 1440 by default. Higher resolutions may not play smoothly due to bandwidth throttling by Youtube.
This value can be customized in: Options > Advanced > YDLMaxHeight

Fixes:
Fixed crash when opening generic URLs. Was a regression since addition of Youtube-DL support in previous version.

thanks, I wish that my lack of change

[LigH]

Just too early for binaries yet...

Oh, did I just see a version 1.8.1?

[clsid]

Files are online now.

[Antonio]

Very nice job, thank you very much for your involvement.

I just wonder if the choice of using own mpc-hc settings for external Lav Filters used as internal ones (where the directory does not exist under mpc-hc) instead of the common Lav Filters settings is very relevant.
For my part, I would have preferred not to have to configure twice the external Lav Filters, but maybe other people are more satisfied with this approach.
Anyway, it is a small detail.

[clsid]

That will not change.

[Antonio]

Originally Posted by Zetti

Media Player Classic - Home Cinema v1.8.0 is released.

Added support for decoding AOMedia AV1 video
.

I just tried to play a AV1 video from https://www.elecard.com/videos, and ouch ! The first UHD AV1 sample gave a very bad time to my 8700k. The CPU was totally unable to decode the video stream in a smooth way, there were a lot of jerks, and the decoder queue was as low as 1 frame. CPU usage was 60%.

It is not an issue with Lav Filters or mpc-hc, as mpv does not get better.

I hope that it is a lack of optimization from the AOM decoder (not using AVX2 ?), or otherwise, the day when this format becomes widespread, current hardware will be good for retirement.

[clsid]

This format is more complex than HEVC. Without hardware acceleration most CPUs will also struggle to decode HEVC at UHD resolution.

Performance will likely get (much) better once FFmpeg gets a native decoder. It will probably be at least 1 or 2 years before AV1 becomes widespread.

[poller]

Originally Posted by clsid

The duration of the video should not matter. I assume you mean you just want more screenshots with shorter durations between the screenshots. How many rows do you suggest?

potplayer goes up to 60 rows.
not sure if we "need" that many, on the other hand... why not? so the user can decide.
i guess technically it shouldn't make much difference if 20 or 60.

[oddball]

Corrupt playback when playing a Youtube AV1 stream.
Video: VP90 3840x2160 23.976fps [V: English [eng] (vp9 profile 0, yuv420p, 3840x2160) [default]]
https://www.youtube.com/watch?v=mIAfxj7nd9k
EDIT: Turned off the VP9 transform filter and all is well. But is it really AV1?
EDIT2: Actually it was not that filter it was the HTTPS filter that caused the corruption.

This is supposedly an AV1 video from their beta playlist.
https://www.youtube.com/playlist?lis...eZUlNUQAVLwrZS

[nevcairiel]

If it says VP9, then you downloaded the wrong encode. There are no 4K AV1 encodes yet, afaik. The videos on that playlist are not exclusively AV1, they just have AV1 encodes in addition to the usual encodes every video has, so you need to make sure you actually grab a AV1 variant.

[Nintendo Maniac 64]

I realize it's still early days, but it's kind of disappointing to see only ~25% CPU utilization on my 4c/8t Nehalem Xeon x3470 while playing back a 1080p 30fps AV1 video in slower than real-time (ended up somewhere between 16 and 20fps).

Mathematically that would equal out to only using 2 threads, but I find a massive performance difference between running 2 threads and 3 threads - it would seem that the 2 less-demanding threads only use around 50% of a CPU core's available utilization even the video you're trying to play is running slower than real-time.

[LigH]

Thanks to a newer libav core, MPC-HC v1.8.1 can now also play a rare ADPCM ADX in MPEG PS (v1.7.8 could not yet play, only detect).

[huhn]

looks like PS2 PCM in mpg didn't make it in.
is that much of a different to ADPCM in mpg?

[LigH]

MediaInfo 18.08.1 did not even detect it yet. And I don't know anything about it, just checked for Selur who linked to a sample in the German Gleitz board.

[littleD]

Originally Posted by Nintendo Maniac 64

I realize it's still early days, but it's kind of disappointing to see only ~25% CPU utilization on my 4c/8t Nehalem Xeon x3470 while playing back a 1080p 30fps AV1 video in slower than real-time (ended up somewhere between 16 and 20fps).

Mathematically that would equal out to only using 2 threads, but I find a massive performance difference between running 2 threads and 3 threads - it would seem that the 2 less-demanding threads only use around 50% of a CPU core's available utilization even the video you're trying to play is running slower than real-time.

Strange, on the low grade 2c/4t processor, i get around 90% utilization, which means decoder must use at least 4 threads. Maybe You have any other bootlenecks

[benwaggoner]

Originally Posted by nevcairiel

If it says VP9, then you downloaded the wrong encode. There are no 4K AV1 encodes yet, afaik. The videos on that playlist are not exclusively AV1, they just have AV1 encodes in addition to the usual encodes every video has, so you need to make sure you actually grab a AV1 variant.

Yes, current AV1 encoders are so slow that there are only a few short test clips created in UHD, and they do stuff like 1 second closed GOP without VBV so they can be run massively parallel. Nothing perceptually tuned yet. How good AV1 would be for UHD and HDR is quite speculative at this point.

Honestly, I'm not expecting to know how "good" AV1 is as a HEVC alternative until late 2019. It'll take at least another year before we have encoders that are fast enough to make a reasonable set of test content and have decent psychovisual tuning.

Decoders are a lot more straightforward, since there's only one right answer. But a good software decoder takes a lot of assembly SIMD optimization, parallelization, fuzz testing, and profiling profiling profiling. The job is harder today as we don't really know what the tools real-world AV1 encodes are likely to use, so knowing where to focus optimization isn't that clear yet. Having LOTS of tools available in a bitstream means LOTS more to optimize in a decoder.

[Nintendo Maniac 64]

Originally Posted by littleD

Strange, on the low grade 2c/4t processor, i get around 90% utilization, which means decoder must use at least 4 threads.

Your CPU wouldn't happen to be support AVX, would it?

In the "Alliance for Open Media codecs" thread, there's someone with a more modern 4c/8t i7 that's seeing 60+% CPU utilization via MPC-HC.

I'm starting to wonder if AV1 has heavy AXV optimizations for its multi-threading which would be very unfortunate for me as I just happened to pick the last generation of 4c/8t processors that didn't have AVX (though even the newest 2c/4t Coffeelake-based Pentiums lack AVX...). A similar thing happened in the past with VP9 where it had some heavy optimizations for SSSE3 (I don't believe this is the case anymore), therefore resulting in something like a 2.5x performance gain simply by having SSSE3 (which wasn't present on AMD CPUs until 2011).

Originally Posted by littleD

Maybe You have any other bottlenecks

The thing is though, if I configure my BIOS to set my CPU to run as 3c/3t, I see the exact same performance when decoding AV1 (though with ~67% due to 5 fewer threads) as I do when my CPU is running in full 4c/8t. By comparison, doing something like playback of 5mbps 1080p 30fps HEVC combined with aggressively-configured 2x framerate interpolation is a complete stuttery mess at 3c/3t (or even at 4c/4t, though to a lesser extent), but at 4c/8t it's perfectly smooth and stutter-free.

Similarly, when running various multithreaded benchmark programs like Cinebench and the like, going from 3c/3t to 4c/8t gives me the kind of performance gains that you would normally expect.

[Klementh]

First of all thanks for your continued work on MPC-HC clsid.

I'm experiencing a bug in version 1.8.1.

When using hotkeys to skip forward or to go back 2 seconds, or anything frame by frame, or even using the seek bar, sometimes the player suddenly closes.

Always happens with this file after a few tries going back and forth: https://expirebox.com/download/e2239...a0b59f452.html

Does not happen with version 1.7.16.

[LigH]

Might be related to inaccurate seeking allowing further distances?