[Page Previous 1, 2, 3]

Topic: FFMPEG Windows Apache 64bit Limiting CPU Consumption

[3]

Topic: FFMPEG Windows Apache 64bit Limiting CPU Consumption

[C0nw0nk]

Try convert it in ffmpeg it will most likely crash depending on what command line you use compared to me. I used it as a good way to fix my application to start marking information outputs as a single attempt since with ffmpeg the information output will never crash. And with a max attempt number of 5 when it hits 5 attempts if on the final attempt it does not work it marks it as failed.

[jimski]

[C0nw0nk]

Jimski use my soloution if a video is crashing your ffmpeg process like the one i shared it is most likely corrupt or not worth the hassle of trying to get to work.

https://www.apachelounge.com/viewtopic.php?p=29578#29578

To fix it since when ffmpeg crashes not much we can do to fix that but using "-i" we mark the information output of ffmpeg as a single attempt and we have a maximum number of attempts for me it is 5 and when you hit that 5 attempts mark it as failed. It is my soloution to a unique problem and from the file i shared it prooves it is for good reason that you should do things this way since information outputs will never crash. (Thats a fact)

[Jan-E]

[C0nw0nk]

Thanks if you ever compile and start providing Windows ffmpeg builds like zeranoe let me know id be happy to test. Right now i just download Zeranoe's 64bit static builds and use those.

[Jan-E]

FFmpeg 2.5.3 with my patches:
https://phpdev.toolsforresearch.com/ffmpeg64.zip

[C0nw0nk]

[Jan-E]

The only things I disable are w32threads and doc. Doc because it installs the documentation and this takes too much space. And w32threads because pthreads is faster than w32threads. The last time I looked, Zeranoe disabled w32threads as well.

Sandbox: I initially copied the script from Roger Pack. See the readme at https://github.com/Jan-E/MyMpeg
Roger uses sandbox as well: the script installs everything (compilers like gcc and g++, many libraries and the resulting ffmpeg/ffprobe/ffplay.exes) in ~/sandbox. If you want to clean things afterwards: rm -rf ~/sandbox

Faac: I am using this because in my tests it was the fastest audio encoder. However, it makes ffmpeg.exe 'non free and non distributable'. You will not see updates by me.

See for instance http://oss.netfarm.it/mplayer-win32.php

[C0nw0nk]

Yeah i was going to mention it violates the licenses and terms to distribute ffmpeg with those audio encoders in them. (What i find to be silly and hope they change their licenses)

But it is also good to know it never crossed my mind to download zeranoe's 64bit shared build and just manualy add in the .dll of the audio encoder to make it work like that.

Is there any performance between a static and a shared build ? Just seems once saves more disk space than the other to me.

[Jan-E]

[C0nw0nk]

Well when i come to testing it if it does not work with Zeranoes can i do it with your builds ?

[Jan-E]

You can test with the ffmpeg64.exe you've just downloaded. There should be minimal differences between Zeranoe's builds and mine. I only changed some fatal errors into non-fatal with a possible recovery:

https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/asfenc.patch
https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/movenc.patch
https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/mpegvideo_enc.patch
https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/subtitles_non_fatal.patch
https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/swscale.patch

I still have the video where the swscale patch worked, but the newest non-patched FFmpeg transcodes that video also without apparent errors.

Furthermore, I added a volnorm patch to make some audio normalization on the fly:

https://github.com/Jan-E/MyMpeg/blob/master/ffmpeg_patches/volnorm_new.patch

My users upload videos where speech is important and sometimes the volume is just too low. The patch enables them/me to automatically adjust the volume a bit. The FFmpeg developers did not agree with my way of patching, so this patch did not make it into FFmpeg core:

http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/166113/focus=166113
http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/166113/focus=178647

[C0nw0nk]

Thanks Jan i put your ffmpeg build into production on mp4's i use libfaac as my audio encoder and webm it is libopus.

As of your patches above i dont notice any real difference in audio outputs unless i am missing something from my command line ? But all is working well i was actualy supprised at how fast libfaac encoded compared to ffmpeg's experimental audio encoder. (Performance gain.)

And the non fatal recorvery patches is that to prevent ffmpeg from crashing and try to skip the error preventing it from converting ? Wouldnt that cause unexpected behaviour in video playback maybe even a corrupt output file ?

[Jan-E]

Easy solution if you run into trouble with my patches: do not build FFmpeg with them, but use Roger Pack's repo.

https://github.com/rdp/ffmpeg-windows-build-helpers

He also has a non_free option which includes libfaac if you change --disable-libfaac into --enable-libfaac in line 1167:

https://github.com/rdp/ffmpeg-windows-build-helpers/blob/master/cross_compile_ffmpeg.sh#L1167

and uncomment line 1317.

[C0nw0nk]

Since you use libopus i am assuming you generate WebM files but i could be wrong you could be using it in OGG containers but i was just curious what video bitrates you use.

On MP4's i use a minimum of 1000k and currently on webm's it is 700k what is why my webm videos look a little choppy but i have been thinking about increasing it but on the downside WebM's are slow to compile so increasing the video bitrates will make it slower.

[Jan-E]

You will be really surprised: -b:v 336k -ab 96k

The relevant part of our FFmpeg command line:

-movflags +faststart -s 448x256 -aspect 16:9 -sws_flags bicubic -vcodec libx264 -b:v 336k -r 25 -strict -2 -acodec libfaac -ab 96k -ar 48000 -ac 2 -x264opts
global_header=1:partitions=p8x8+b8x8+i8x8:level_idc=30:cabac=0:subq=3:qp_min=0:qp_max=51:qp_step=4:me=dia:\
subme=0:mixed_refs=0:me_range=16:chroma_me=1:trellis=0:8x8dct=0:cqm=flat:chroma_qp_offset=0:nr=0:\
keyint=25:min_keyint=12:scenecut=0:ratetol=1.0:qcomp=0.60:ip_factor=1.4:weightp=0:fast_pskip=1:frameref=1:\
bframes=0:mbtree=1:rc_lookahead=15:sliced_threads=0:threads=8

Our only requirements are that the faces and gestures are recognizable. Speech is far more important. And speed in compression and uploading. As I said: the compression is being done before uploading. Our users upload less than 200MB per hour video.

Nowadays I only compile to mp4. That my FFmpeg builds contain all those codecs is just a strange desire to have a 'complete' build.

[C0nw0nk]

Yeah i used to convert to only MP4 since it was used by everyone until linux and Windows XP users started complaining saying they could not watch so i had to start generating WebM's since i refuse to use the putrid filth known as Shock wave flash or FLV/OGG.

http://en.wikipedia.org/wiki/HTML5_video#Browser_support

MP4 is supported by pretty much all browsers and Operating systems but linux and Windows XP users dont get the luxury so i had to have WebM videos. But WebM is fairly new so hopefully google keep making it better. Since compared to MP4 WebM's do have allot smaller file size.

This is also the media player i use.
http://mediaelementjs.com/

And with a comments feature i have i do a nifty little trick using javascript allowing users to click timestamps and jumping to that time frame in the video like on YouTube.