LAV Filters - DirectShow Media Splitter and Decoders

[nevcairiel]

If there were progress, it would be mentioned in the issue. Asking is not required, and only adds noise.
I can already tell you that it will probably be weeks before I can even look at the issue, because I simply don't have a external DTS decoder available right now. The code looks fine tho, I suggest to simply decode the audio in your PC.

[theoneofgod]

If there were progress, it would be mentioned in the issue. Asking is not required, and only adds noise.
I can already tell you that it will probably be weeks before I can even look at the issue, because I simply don't have a external DTS decoder available right now. The code looks fine tho, I suggest to simply decode the audio in your PC.

I wouldn't call knowing more than I did already "noise". Thanks for the update.

[Dav3Th3Rav3]

EDIT: Issue is fixed, caused by a feature in NVIDIA Control Panel. To fix, find Content Type Reported To The Display under Adjust Desktop Color Settings and change from Auto-select (Recommended) to Desktop Programs
The green color shift is only caused when NVIDIA detects Fullscreen Video and applies that preset. Since adjusting it to Desktop Programs, no color shift is observed.

Recently, LAV Video Decoder has been shifting ALL my colors (even outside of the video) to a much greener palette on my PC, and for awhile I thought this was a problem on my machine. I have tried a full clean, downloading the latest NVIDIA Drivers, installing the latest K-Lite, etc but nothing worked.

To my surprise, I just noticed this same exact problem on ANOTHER entirely different PC with vastly different specs. The only thing in common on the setups were LAV Filters builds and both using NVIDIA hardware. Remember, this is all on a FRESH install of the latest K-Lite (10.6.3 - July 17th, 2014) on the latest NVIDIA drivers (340.52 - July 29, 2014).

Issue is persistent across x86 & x64, DXVA2 & CUVID

Since printscreen doesn't capture the problem, here's some images from my phone: (http://imgur.com/a/SoPrG/)
http://i.imgur.com/z7oPrAa.jpg

I will try any suggestion or advice, thanks!

[theoneofgod]

Recently, LAV Video Decoder has been shifting ALL my colors (even outside of the video) to a much greener palette on my PC, and for awhile I thought this was a problem on my machine. I have tried a full clean, downloading the latest NVIDIA Drivers, installing the latest K-Lite, etc but nothing worked.

To my surprise, I just noticed this same exact problem on ANOTHER entirely different PC with vastly different specs. The only thing in common on the setups were LAV Filters builds and both using NVIDIA hardware. Remember, this is all on a FRESH install of the latest K-Lite (10.6.3 - July 17th, 2014) on the latest NVIDIA drivers (340.52 - July 29, 2014).

Issue is persistent across x86 & x64, DXVA2 & CUVID

Since printscreen doesn't capture the problem, here's some images from my phone: (http://imgur.com/a/SoPrG/)
http://i.imgur.com/z7oPrAa.jpg

I will try any suggestion or advice, thanks!

K-Lite is an overkill for most people. You'd be much better off reproducing the issue, if you just had MPC-HC with the built in LAV Filters. Try different NVIDIA drivers.

[nevcairiel]

Its impossible for the video decoder to change your colors. Sounds like its something the driver does when it realizes a video is playing.

[Dav3Th3Rav3]

Its impossible for the video decoder to change your colors. Sounds like its something the driver does when it realizes a video is playing. BINGO!

This is absolutely retarded. Must be a new feature, but NVIDIA has a "Content Type Reported To Display" feature that's by default set to "Auto-select (Recommended)". There are two other settings, Desktop Programs and Fullscreen Video.

The green color shift is only caused when NVIDIA detects Fullscreen Video and applies that preset. Since adjusting it to Desktop Programs, no color shift is observed.

THANK YOU FOR THE QUICK RESPONSE! Excellent support, thank you :-)
http://i.imgur.com/qihbtrS.jpg

[Sulik]

AnyETA on DXVA HEVC support ? :)

[truexfan81]

AnyETA on DXVA HEVC support ? :)

whenever the amd/nvidia/intel start releasing cards with support for hevc ;)

[mastan]

whenever the amd/nvidia/intel start releasing cards with support for hevc ;)

Intel already has: http://downloadmirror.intel.com/23885/eng/ReleaseNotes_GFX_64.pdf
HEVC decode now supported by the driver and video players can now take advantage of the GPU accelerated decode support offered by Intel

[Deihmos]

Can anyone assist me with the advance subtitle selection? What I want is if english audio then subs are off - English audio then english forced subs - any other language then english subs.

This is what I have so far.
eng: off;eng:eng|f;*:eng

When audio

[foxyshadis]

Intel already has: http://downloadmirror.intel.com/23885/eng/ReleaseNotes_GFX_64.pdf

Intel removed it from the beta driver update in June, as well as more recent betas. I absolutely wouldn't rely on that update until they come out with something stable.

[NikosD]

Interesting HEVC OpenCL solution from Strongene here:

http://xhevc.com/en/downloads/downloadCenter.jsp

It's a DS filter working with discrete AMD cards (HD 5000 or better) or APU.

I haven't tried it with Nvidia/Intel, it could work.

Using my old Core2Duo and HD 5750 card, it works for most of my H.265 samples, but with no clear performance benefits over the CPU only decoding.

My platform is very old supporting PCI-E v1.1x4, so maybe that's the problem.(it needs PCI-E v2.0)

It's free, you can try it with better AMD cards/platform or with Nvidia/Intel cards.

[nevcairiel]

Any other HEVC decoders are off-topic in the LAV thread, if you want to discuss them, feel free to open a new thread in the HEVC forum.

[Deihmos]

The versions after 0.59.1 introduced a bug when I resume from sleep. last version of lav splitter that worked for me is version 59.1. If I play something in WMC then put my PC to sleep then nothing will play if I resume from sleep until I restart the player. Version 0.59.1 works perfectly fine so something must have changed to cause this behavior. Any ideas what is causing this?

It happens on all my computers and have to upgrade from 0.59.1 because of a few bugs with it that were fixed in the newer releases.

[NikosD]

I would like to discuss yours DXVA HEVC but...:D

[clsid]

Build fails here. Pkg-config is unable to find the speex stuff.


And an on-topic comment regarding HEVC:
There has been an almost 20% increase in decoding performance after today's FFmpeg update ;)
I am getting 98 fps on a 4K sample with the x64 build, and 45 fps with x86 build.

[mastan]

Intel removed it from the beta driver update in June, as well as more recent betas. I absolutely wouldn't rely on that update until they come out with something stable.

DXVA Checkers shows HEVC_VLD_Main with the latest stable drivers https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=23885&lang=eng&ProdId=3719

Seems pretty stable to me.

[qyot27]

Build fails here. Pkg-config is unable to find the speex stuff.
If the Speex build in question is static, did you pass --pkg-config-flags="--static" to FFmpeg?

[detmek]

Those stable drivers are from May. Intel removed it in newer beta drivers from Jun. Some people report Intel also removed 4k H264 in latest beta driver.
BTW, I have those drivers on Pentium G3220 and DXVA Checker does not show HEVC_VLD. But, its Pentum, so...

[SUPERBIF]

Question:
Have just replaced my very old:
Asrock E350 Zacate
- AMD Dual-Core Zacate E350/E350D APU
- Supports AMD's Cool 'n' Quiet Technology
- UMI 2.5 GT/s
- Integrated AMD Radeon HD 6310 graphics
- DX11 class iGPU, Pixel Shader 5.0
- Max. shared memory 512MB

with this new Intel Q1900-ITX:
- Intel® Quad-Core Processor J1900 (2 GHz)
- Intel® 7th generation (Gen 7) graphics
- DirectX 11.0, Pixel Shader 5.0

I am using LAV filter with XBMC but on my new system I get very low frames on 720p hz live TV so video and audio is not in Sync. I have configured scaling to bicubic -0,60

On my old AMD system this was not a problem.
Is it possible to get MPEG4 hardware acceleration on LAVfilter or should how can I get it to run smoothly?

[nevcairiel]

J1900 is a Atom CPU, its pretty slow, so no surprises it doesn't work reliably. In theory its GPU is compatible with DXVA2 for H.264/MPEG2/VC-1, but I've never seen Intel support MPEG-4 hardware acceleration.

[sneaker_ger]

The question would be whether he's talking about H.264 or Mpeg 4 Part 2.

[nevcairiel]

Most people refer to MPEG-4 Pt2 when they say MPEG-4, because if they don't, they are being uncommonly inaccurate. :p

[SUPERBIF]

Thanks for the answer. According to Intel it's a Celeron quad core, so I don't know why it not able to play the mpeg4 codec from live TV?
Is my old E350 really better GPU wise and to use with LAV?

[axlt2002]

Hi guys,

My PC graphic setup comprises an A6-3500 APU and a Radeon HD 6770 1GB GDDR5 GPU. The PC is connected to a Samsung TV through the 6670 HDMI port; the GPU is configured to be the primary graphic card in the BIOS (the motherboard is an ASROCK A75M BIOS v2.00) and this is confirmed also by the Catalyst Control Center. Dual Graphics is also enabled both in the BIOS and in CCC. I then decided to use AMD System Monitor to made some tests.

As you can see from the graphic below, when playing a video game, the 6670 starts to work and sometimes, when needed, the 6530D of the APU works as a supplmentary computational capacity (this is not anyway captured below but I have checked it):

http://i57.tinypic.com/2j0m39y.jpg

So, the Dual Graphics works as expected, considering the 6670 as the primary card.

I have then tried to play a 4K video (125 Mbps) and I was expecting to see the 6670 start to work as well...As you can see from the image below, only the CPUs were used and both GPUs (6670 and 6530D) utilization was 0%. The result was a heavy stuttering playback...

http://i62.tinypic.com/jsevzd.jpg

Just let you know that I was using MPC-HC player with LAV filters and DVXA2 (native) configured to be used also for 4K videos.

Finally the question: is this a correct behaviour? For video playback only the CPU are used? If yes, I'm quite confused to understand the DXVA2 concept...video should be decoded by the GPU...isn't it?

Looking forward for some help!

Thanks in advance!

[LigH]

Video should be decoded by the GPU if the GPU supports the video format.

Hardly any GPU (except the most recent intel chips) supports HEVC decoding yet. So how should the GPU help if it doesn't know how?

[nevcairiel]

Just let you know that I was using MPC-HC player with LAV filters and DVXA2 (native) configured to be used also for 4K videos.

Finally the question: is this a correct behaviour? For video playback only the CPU are used? If yes, I'm quite confused to understand the DXVA2 concept...video should be decoded by the GPU...isn't it?

AMD doesn't support 4K video decoding. If you want 4K don't buy AMD, instead get an NVIDIA card, or even an integrated Intel GPU.

[axlt2002]

Video should be decoded by the GPU if the GPU supports the video format.

Hardly any GPU (except the most recent intel chips) supports HEVC decoding yet. So how should the GPU help if it doesn't know how?

Hi, thanks for the answer!

Here the infos related to the video:

Generale
Complete name : E:\Personale\Multimedia\Test Folder\Audio-Video Check\3840x2160 125 Mbps.mp4
Format : MPEG-4
Format profile : Base Media / Version 2
Codec ID : mp42
File size : 2,69 GiB
Duration : 3min 4s
Overall bit rate mode : Variabile
Overall bit rate : 125 Mbps
Encoded date : UTC 2012-12-31 19:20:50
Tagged date : UTC 2012-12-31 19:20:50

Video
ID : 2
Format : AVC
Format/Info : Advanced Video Codec
Format profile : High@L5.1
Format settings, CABAC : Si
Format settings, ReFrames : 5 frame
Format settings, GOP : M=3, N=33
Codec ID : avc1
Codec ID/Info : Advanced Video Coding
Duration : 3min 4s
Bit rate mode : Costante
Bit rate : 125 Mbps
Width : 3.840 pixel
Height : 2.160 pixel
Display aspect ratio : 16:9
Frame rate mode : Costante
Frame rate : 23,976 fps
Standard : NTSC
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bit
Scan type : Progressivo
Bits/(Pixel*Frame) : 0.629
Stream size : 2,69 GiB (100%)
Language : Inglese
Encoded date : UTC 2012-12-31 19:20:52
Tagged date : UTC 2012-12-31 19:20:52
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

It doesn't seem to be HEVC encoded...do I miss something?

[sneaker_ger]

No, you are correct. You didn't state what codec was used in your initial post so LigH had to guess (and guessed HEVC because of the 4k resolution).

[axlt2002]

No, you are correct. You didn't state what codec was used in your initial post so LigH had to guess (and guessed HEVC because of the 4k resolution).

lol! I read my initial post several times before posting it in order to check if I was missing some important detail...and I missed it!

So, do you have then some other thoghts on this?

[axlt2002]

AMD doesn't support 4K video decoding. If you want 4K don't buy AMD, instead get an NVIDIA card, or even an integrated Intel GPU.

4K in this case is just the resolution, but as written the post above, is encoded in a "traditional" manner...

So I really don't understand why the video doesn't pass through the GPU...any suggestion?

[sneaker_ger]

nevcairiel is referring to the resolution. Everything is correct, it's simply not supported.

[huhn]

it is UHD. the DXVA decoder is limited depending on the GPU. AMD UHD decoding was or is still broken. i just tried it with my r9 270 no chance it uses the CPU.
you can try it with your gpu to you need to enable (UHD) 4k in lav video decoder. it's an AMD so i don't think it will work.
intel and nvidia can decode UHD

[axlt2002]

it is UHD. the DXVA decoder is limited depending on the GPU. AMD UHD decoding was or is still broken. i just tried it with my r9 270 no chance it uses the CPU.
you can try it with your gpu to you need to enable (UHD) 4k in lav video decoder. it's an AMD so i don't think it will work.
intel and nvidia can decode UHD

Thanks huhn for your answer!

Indeed I have already enabled UHD in LAV filters settings and DXVA2 (native) as well. The result is the one reported at the begginning of this thread...stuttering playback! :( Unfortunately I tend to agree with you that 4K videos are decoded using only CPU(s).

According to the discussion so far, if I'm going to play an HD video (lets say 1920x1080), is this going to use DXVA (i.e. the 6670 will do its tasks leaving the CPU resting in peace)?

[GTPVHD]

http://www.guru3d.com/articles_pages/amd_a10_7800_kaveri_apu_review,16.html

Unfortunately Ultra HD videoplayback resulted into stuttering.

The reason why we noticed stuttering seems to be that the trailer is not DXVA encoded, or DXVA at 4K does not yet work.

Kaveri has their latest UVD4 fixed function hardware decoder and still fails miserably at UHD.

[huhn]

According to the discussion so far, if I'm going to play an HD video (lets say 1920x1080), is this going to use DXVA (i.e. the 6670 will do its tasks leaving the CPU resting in peace)?
the CPU is only uploading the stream to the GPU.
and it's not that easy it has to be compatible with the DXVA decoder.

it's not like 1080p will always work. feed it 120 hz it can die or not used with 16 ref frames ,10 bit, other h264 settings to much kbits. it's not reliable on self made videos. but works fine with BD.

so in short the problem is your GPU/CPU not lavfilter.

AMD said they will fixed this issue years ago never happens maybe with the next series i feel sorry for people how believed them :-(

[NikosD]

@clsid and nevcairiel

By disabling Hyper-threading I found out that both LAV x86/x64 didn't drop their performance level, almost at all, during 4K HEVC decoding.

The performance drop of using 4 threads instead of 8, was <10% (!) and the CPU usage of 4 threads was 95%.

The OpenCL decoder had more than 80% increase in CPU usage using 4 threads instead of 8 during playback of 4K HEVC. (from 32% to 58%)

It's clear that the FFMpeg's HEVC decoder (LAV) is weak in multi-threading decoding on more than 4 threads.

[andyvt]

It's clear that the FFMpeg's HEVC decoder (LAV) is weak in multi-threading decoding on more than 4 threads.

That's not necessarily true.

HT is a mechanism to get more throughput from MT workloads (either at the process or system level) where the physical core would otherwise be waiting. In a MT workload where the workload doesn't fit that profile HT has little benefit. TBC, I'm not saying that the hypothesis isn't correct, just that you haven't really proved it because the underlying assumptions are flawed. You would need to re-run the test on a system with more than 4 physical cores and compare runs on all cores vs. runs when limiting processor affinity.

[NikosD]

I don't have such system.

If someone has such system, he could give it a try and tell us.

The fact is that OpenCL decoder handles 8 threads on 4 cores, a lot better/faster.

[James Freeman]

Is the LFE channel boosted by 10db or is it decoded "as is" so I need to boost in my receiver/amplifier?

To simulate this boost in the digital domain, can I just lower -10db all other channels?

[e-t172]

Is the LFE channel boosted by 10db or is it decoded "as is" so I need to boost in my receiver/amplifier?

It is decoded as is, which is how it should be. The 10dB boost should be done at the bass management stage (which is in the AV receiver in most setups), not at the decoding stage.

To simulate this boost in the digital domain, can I just lower -10db all other channels?

Yes.

[andyvt]

The fact is that OpenCL decoder handles 8 threads on 4 cores, a lot better/faster.

In your original post on this topic you made some comments about CPU utilization, and drew some conclusions based on them. While I don't doubt the veracity of your observations I'm not sure you can draw the conclusions that you have based solely on those observations.

If the OpenCL decoder displays greater utilization of cores 5-8 when HT is enabled it doesn't mean that it's necessarily faster (you would need to measure time to prove this, because as I pointed out before you can't assume that utilization equals speed) or better (you'd have to use SSIM or something similar to compare the decoded frames). Again, I'm not saying that your hypothesis is incorrect, just that the information you have provided so far doesn't demonstrate that it is correct.

[NikosD]

Between my initial post here and the last post regarding HyperThreading there were a lot of tests posted in my thread here in doom9 that you can find in my signature.

As I wrote to my original post, the PCI-E v1.0 x4 was a real bottleneck and you definitely can't draw final conclusions based on that only.

I didn't draw any conclusion based on C2D CPU with so slow PCI-E bus.

In fact I wrote just the opposite!

About the 5-8 threads when I wrote better handling I meant better use/optimization of 5-8 threads, not picture quality of course!

About faster, I meant the final result of using LAV vs OpenCL decoder.

If LAV manages to utilise better the 5-8 threads, it could be faster than OpenCL CPU decoding part, or even faster in general.

For now, it's just slower and the x86 version, a lot slower.

[James Freeman]

It is decoded as is, which is how it should be. The 10dB boost should be done at the bass management stage (which is in the AV receiver in most setups), not at the decoding stage.

Thank you e-t172.

I do not have a receiver so I need to manage the bass digitally via plug-in.

What do you think about AC3Filter?
I think it does the job very well after tweaking, clipping prevention and maximizing headroom.
I lower all channels (except LFE) -10db at the input, I set the crossover frequency to 80Hz, then I lower the master volume till my clip audio interface meter shows below -0.2 at peaks.
I created a Surround Test audio file with pink noise (all frequencies equal) every channels reaching -0.1 in his turn, then they are played all together to adjust the master volume maximize headroom.
Note that I down mix the 5.1 to 2.1 via my audio interface control panel which is a huge mixer with everything I need.

[axlt2002]

the CPU is only uploading the stream to the GPU.
and it's not that easy it has to be compatible with the DXVA decoder.

it's not like 1080p will always work. feed it 120 hz it can die or not used with 16 ref frames ,10 bit, other h264 settings to much kbits. it's not reliable on self made videos. but works fine with BD.

so in short the problem is your GPU/CPU not lavfilter.

AMD said they will fixed this issue years ago never happens maybe with the next series i feel sorry for people how believed them :-(

Hi huhn and All!

Just to doulbe check, I have made other trials with some clips with HD resolution and the 6670 is involved in the process of decoding the video stream. I agree on the fact that also for this resolution a lot depends on other parameters...but with my actual playlist, no issues so far (fortunately).

Afterwards i tried also to play an UHD video with around 17 Mbps rate and the CPUs were able to manage it without any stuttering (total load was around 61%). The rate of 125 Mbps of my intial testing clip was too much for them (load 100%)!

Of course I never pointed LAV filters as the guilty for this issue! My trials started as a simple curiosity and landed in the AMD epic fails on the 4K resolution management! :(

Is this an hardware issue or it will be possible to have it fixed through driver updates? Being a complete noob, is not only a question of redirect the video stram to the GPU?

[huhn]

Is this an hardware issue or it will be possible to have it fixed through driver updates? Being a complete noob, is not only a question of redirect the video stram to the GPU?

the problem is the GPU can't decode it. they said they will fix it for years nothing happened. so give up on this this was with 5xxx series or 6xxx. i think UVD 2.2 should handle UHD. AMD is currently limited to level 4.1. you need at least level 5.1 for UHD and for high frame rate UHD you need level 5.2.

this is going off topic so i leave it at this feel free to make a new thread if you like to know more about DXVA and there limits.

[andyvt]

About the 5-8 threads when I wrote better handling I meant better use/optimization of 5-8 threads, not picture quality of course!



On a HT CPU use of threads > physical core count does not necessarily mean that a process is better optimized, it can indicate that or it can indicate the opposite because HT is a mechanism to use HW that would otherwise be idle waiting. If the physical core is fully utilized, and isn't waiting for anything it would be sub-optimal to allow another thread to run.

If you'd like to continue this discussion (since it has veered more towards how CPUs work), perhaps it would be better if we do that somewhere else as to not clutter up this thread. Also, if you can provide your Ffmpeg testing methodology, I have access to a 12/24 box. It would be interesting to see how it actually scales over 4 physical cores.

[axlt2002]

the problem is the GPU can't decode it. they said they will fix it for years nothing happened. so give up on this this was with 5xxx series or 6xxx. i think UVD 2.2 should handle UHD. AMD is currently limited to level 4.1. you need at least level 5.1 for UHD and for high frame rate UHD you need level 5.2.

this is going off topic so i leave it at this feel free to make a new thread if you like to know more about DXVA and there limits.

Thanks a lot for the support! It was a really interesting discussion...a part the related conclusions! ;)

No need to further go in details right now...

Thanks a lot for the work on LAV filters!

[clsid]

To stay a bit on-topic. LAV uses 1.5x logical cores by default for the number of threads. So 12 threads on a 4+4 CPU. This performs a bit better than using 8 threads.

It is normal that HT cores have relatively crappy performance. They are limited to whatever resources are not yet utilized. A low-end CPU usually has less resources per physical core than high-end ones. So HT performance on a Pentium is probably less than with a Core i7.

A simple test on a 4+4 Core i7 with 4k HEVC:
threads - fps
1: 27.3
2: 47.8 (~20 per thread extra)
3: 62.0 (~14 per thread extra)
4: 71.1 (~9 per thread extra)
6: 79.8 (~4.4 per thread extra)
8: 85.4 (~2.8 per thread extra)
12: 98.3 (~3.2 per thread extra)
16: 102.6 (~1 per thread extra)
(I did just 1 pass per test, so the numbers can have a little variance)

[ahaha2013]

The source codes not to support gcc4.9.1? Why when I compiled, there were a lot of warning prompt?

Win8.1 x86
MSYS_MinGW-w64_GCC_491_x86-x64_Full.7z