L-SMASH Source

[amayra]

there any reason for not using GitHub Actions or appveyor ?

[Boardlord]

Long time, no post

I'd like to ask for some help. I've just upgraded to an AMD Ryzen 4000-based laptop with integrated Vega graphics, on dGPU. On my previous laptop I had a crappy Nvidia dGPU, but it still was a tremendous help with transcoding stuff through DGIndex. With this laptop this is out of the picture obviously.

What would be the syntax with L-SMASH in Avisynth to utilize the AMD VCN core for HW decoding, to free up the CPU for x264 encoding? If it's not possible with the current compiled binaries, can I build one that includes the capability to use the VCN core? Thanks!

[Atak_Snajpera]

Originally Posted by Boardlord

Long time, no post

I'd like to ask for some help. I've just upgraded to an AMD Ryzen 4000-based laptop with integrated Vega graphics, on dGPU. On my previous laptop I had a crappy Nvidia dGPU, but it still was a tremendous help with transcoding stuff through DGIndex. With this laptop this is out of the picture obviously.

What would be the syntax with L-SMASH in Avisynth to utilize the AMD VCN core for HW decoding, to free up the CPU for x264 encoding? If it's not possible with the current compiled binaries, can I build one that includes the capability to use the VCN core? Thanks!

HW Decoding is not available on AMD GPU. Blame AMD for being stubborn and not providing api to their decoder like Intel or NVidia.

[Boardlord]

Originally Posted by Atak_Snajpera

HW Decoding is not available on AMD GPU. Blame AMD for being stubborn and not providing api to their decoder like Intel or NVidia.

Hmm, but I've seen that LAVFilters can utilize the AMD GPU for HW-accelerated playback...?

Edit: through DXVA and/or D3D11 I believe, unless I've totally misunderstood.

Sent from my SM-G973F using Tapatalk

[Atak_Snajpera]

We need proper api like cuvid or quicksync not some hacks via dxva. AMD should stop acting like some special princes and allow third party programmers access to their decoder.

[jones1913]

Originally Posted by Atak_Snajpera

We need proper api like cuvid or quicksync not some hacks via dxva. AMD should stop acting like some special princes and allow third party programmers access to their decoder.

The API exists: AdvancedMediaFramework (AMF)
However, AMF decoder is not yet integrated in ffmpeg (but encoder is): https://github.com/GPUOpen-Libraries...AMF/issues/199

Btw. on Linux encoding and decoding works with ffmpeg through vaapi.

[real.finder]

there are frame properties PR by Asd-g since a week https://github.com/HolyWu/L-SMASH-Works/pull/12

I think HolyWu is not active these days, so anyone else can build it?

[StvG]

Here - dav1d 0.7.1, ffmpeg@9b72cea446, zlib 1.2.11, xxHash v0.8.0.

[FranceBB]

Thank you!!

[kedautinh12]

Thanks

[real.finder]

Originally Posted by StvG

Here - dav1d 0.7.1, ffmpeg@9b72cea446, zlib 1.2.11, xxHash v0.8.0.

that was fast! thanks!

[kedautinh12]

Originally Posted by StvG

Here - dav1d 0.7.1, ffmpeg@9b72cea446, zlib 1.2.11, xxHash v0.8.0.

I meet error when index with LSMASH, normally with FFMPEG

error: https://drive.google.com/file/d/1ZNB...ew?usp=sharing
file test: https://drive.google.com/file/d/1yHA...ew?usp=sharing

[real.finder]

Originally Posted by kedautinh12

I meet error when index with LSMASH, normally with FFMPEG

error: https://drive.google.com/file/d/1ZNB...ew?usp=sharing
file test: https://drive.google.com/file/d/1yHA...ew?usp=sharing

I think for mkv nothing better than ffms2, LSMASH is not replacement for ffms2 so it's focuses on what ffms2 fails to do

anyway, let's see what the Developers said about it

[manolito]

I don't think that it has anything to do with the MKV container...
The latest HolyWu build from July 2020 does not have this issue, so this is a regression which should be fixed.

[Atak_Snajpera]

Originally Posted by real.finder

I think for mkv nothing better than ffms2, LSMASH is not replacement for ffms2 so it's focuses on what ffms2 fails to do

anyway, let's see what the Developers said about it

Yes It is. L-SMASH is superior in every way. No idea why people are still glued to non frame accurate filter. You are just asking for troubles when editing video in avisynth.

FFMS


L-SMASH

[StvG]

Originally Posted by kedautinh12

I meet error when index with LSMASH, normally with FFMPEG

error: https://drive.google.com/file/d/1ZNB...ew?usp=sharing
file test: https://drive.google.com/file/d/1yHA...ew?usp=sharing

Try with lwlibavvideosource(..., threads=2).

Originally Posted by manolito

I don't think that it has anything to do with the MKV container...
The latest HolyWu build from July 2020 does not have this issue, so this is a regression which should be fixed.

My build with threads=2 doesn't have this issue. Obviously I'm using different building environment than HolyWu.

#The latest official build (20200728)
lwlibavvideosource("Letter - SHE'S.mkv")


Number of frames:                     6518
Length (hh:mm:ss.ms):         00:04:31.855
Frame width:                          1920
Frame height:                         1080
Framerate:                          23.976 (24000/1001)
Colorspace:                           i420

Frames processed:                   6518 (0 - 6517)
FPS (min | max | average):          122.0 | 2021 | 448.1
Process memory usage (max):         105 MiB
Thread count:                       37
CPU usage (average):                13.3%

Time (elapsed):                     00:00:14.547

#my build
lwlibavvideosource("Letter - SHE'S.mkv", threads=2)


Number of frames:                     6518
Length (hh:mm:ss.ms):         00:04:31.855
Frame width:                          1920
Frame height:                         1080
Framerate:                          23.976 (24000/1001)
Colorspace:                           i420

Frames processed:                   6518 (0 - 6517)
FPS (min | max | average):          104.0 | 2487 | 424.9
Process memory usage (max):         104 MiB
Thread count:                       24
CPU usage (average):                10.8%

Time (elapsed):                     00:00:15.340

#ffms2 - no seek issue
ffvideosource("Letter - SHE'S.mkv", threads=8)


Number of frames:                     6518
Length (hh:mm:ss.ms):         00:04:31.855
Frame width:                          1920
Frame height:                         1080
Framerate:                          23.976 (24000/1001)
Colorspace:                           i420

Frames processed:                   6518 (0 - 6517)
FPS (min | max | average):          167.8 | 2205 | 1077
Process memory usage (max):         133 MiB
Thread count:                       30
CPU usage (average):                29.3%

Time (elapsed):                     00:00:06.049

[manolito]

Thanks for looking into it...

I can confirm that the problem disappears when "threads=2" is specified. Still I think that this should be fixed because the HolyWu builds and also FFVideoSource do not have a problem with the default value for Threads (which means to use the number of threads reported by Windows with a ceiling of 16).

[StvG]

What ffms2 version do you use? With this one and default threads I have 6 seek issues.

[kedautinh12]

Originally Posted by StvG

What ffms2 version do you use? With this one and default threads I have 6 seek issues.

I thinks thread=8 very strong for weak computer, can you fixed with default thread for both ffms2 and lsmash??

[manolito]

Originally Posted by StvG

What ffms2 version do you use? With this one and default threads I have 6 seek issues.

Yes, I used the version you linked to from October 2020.

I got curious and made a couple of tests using different source filters with Atak's SeekTester 1.2. I used the default frame limit of 1000 frames. As the source I took the "Letter - SHE'S.mkv" (VP9 Video 1080p). My CPU is a Core i5 third generation (2 physical cores plus hyperthreading). I used the default thread values for all source filters which would be 4 for this CPU.

FFMS2 StvG from Oct. 2020:
There were 12 corrupted frames out of 1000 frames.

FFMS2 StvG from Aug. 2020:
Same

FFMS2 StvG from April 2020:
Same

FFMS2 StvG from Nov. 2019:
Same

FFMS2 C-Plugin latest 2.31 by qyot27
Perfect

FFMS2 C-Plugin by Reino:
Both available versions fail miserably. Mainly corrupted or inserted frames

FFMS2 official version 2.40 by Myrsloik:
Fails miserably. Mostly corrupted or inserted frames

DirectShowSource with LAV Filters:
Many inserted frames

DSS2Mod with LAV Filters (preroll=15):
Perfect

LWLibavVideoSource by HolyWu:
Perfect


Ok, these results are only for this specific VP9 source file. Other source formats may behave differently, but I find this very interesting. Any comments?