| View previous topic :: View next topic |
| Author |
Message |
CaptainBlood
Advocate
Joined: 24 Jan 2010
Posts: 3625
|
 Posted: Sat Jun 02, 2018 10:23 am Post subject: ffmpeg@gcc-graphite odd behavior... |
 |
|
Ffmpeg audio file start & end trimming works fine ... except that it often replaces a new end near audio chunk by silence.
The very end of the new file is generally fine though.
I switched back from latest ~am64 to stable, cancelled all /etc/portage.package.env specifics but trial for gcc -O1.
It didn't seem to fix properly.
Anyone with the same issue?
Any idea how to fix?
Or maybe an convertfree alternate tool?
Thks 4 ur attention.
Last edited by CaptainBlood on Tue Jun 05, 2018 8:40 am; edited 5 times in total |
|
| Back to top |
|
 |
CaptainBlood
Advocate
Joined: 24 Jan 2010
Posts: 3625
|
| Posted: Sat Jun 02, 2018 4:48 pm Post subject: |
|
|
Old: Had to fallback to media-video/ffmpeg-3.2.6 which didn't punch the file I was working on.
I will return if fallback reveals to be only partial fix.
New: Maybe a frame rate thing...
AFAIK, codecs have default frame rates, so might be worth trying higher frame rates.
But can't find such a parameter. Any idea?
Thks 4 ur attention, interest & support. |
|
| Back to top |
|
|
CaptainBlood
Advocate
Joined: 24 Jan 2010
Posts: 3625
|
| Posted: Tue Jun 05, 2018 8:17 am Post subject: |
|
|
Real culprit was some among these gcc manually defaulted flags: | Code: | | -floop-block -floop-interchange -floop-strip-mine -ftree-loop-distribution -ftree-loop-linear |
I initially though these could at worst slow down execution.
Now it reveals to modify (corrupt? ) features, possibly.
I now have too little time to investigate which, but will do in a near future.
I will update from then on.
Any advice?
Thks 4 ur attention, interest & support.
EDIT 1: Hearable differences such as silence holes have been confirmed via media-lib/chromaprint features.
After graphite flags cancellation, ffmpeg doesn't generate auditable silences anymore.
Unrebuilded x64 stable chromaprint keeps showing some differences at the concerned audio chunk, flating the wave though.
EDIT 2: Rebuilded chromaprint too...
EDIT 3: Audacity wave views confirms unflated chunk.
EDIT 4: chromaprint display of latest win ffmeg @git under wine processed file shows same erroneous flatness. |
|
| Back to top |
|
|
|
Multimedia
