Support forum for DG Tools

Recently, I have run across this intermittent issue on my Win 10 x64 system when running this command line: Even though file.264 is a well-formed AVC elementary stream, ~5% of the time I get a popup error with the message in the subject line. The issue is completely random. Immediately re-running the exact same command line on the exact same untouched file will work ~95% of the time. And the rerunning it again may result in the same popup again about ~5% of the time, even though previous runs reported no error.
Usually I imagine a sample file would be helpful to diagnose such an issue, but as it seems to happen randomly with any input file and no valid input file produces this error consistently, that would probably be useless here.
This would usually not be a big deal, but is somewhat annoying to me as the popup, when it happens, pauses overnight batch runs.

I really do need you to provide a sample that does this for you. I was not able to duplicate the problem using one of my AVC ES files. Note that I do not run Win10 and we have only a week of experience with it, so that may be a factor.

The code is completely deterministic and I checked carefully for uninited variables, so the randomness is very surprising and may suggest system issues at your end.

Thanks.

That is what I was afraid of. My guess is that the bug is somewhere in NVidia's initialization code and is only triggered by specific system configurations (or more people would see the issue) and system state (or I would see it every time rather that ~5%), such as a race condition.

Sometimes userland code can work around such system bugs by retrying the initialization a couple of times before barfing, but that may be much to ask if it is not a common problem.

For what it is worth, and in case somebody else googles this thread, my system runs Win 10 Pro x64, the current public DGDecNV, and the current public release NVidia driver. The hardware is Core i7 970 (which with 6 cores and HT is still a very fine transcode engine, despite its age) and a newish GTX 970.

The video type detection code has nothing to do with any nVidia code; it runs even before CUDA is inited. It's a simple parsing of the stream and that is why I need to see a sample. You ignored my request for a sample. Please provide a sample and ensure that the sample you provide actually fails for you as you describe. Thank you.

As I said, I thought it was most like a system issue and hence not reproducible , but if you think that a sample would help, I'll DM you a link to a (relatively) small file which has both failed and succeeded with DGIndexNV more than once as well as the DGI and LOG files from a successful run.

Thank you, Carl. If I can't figure out anything from the stream, I'll probably want to make a test build for you.

Hi Carl.

There's nothing suspicious in the file and I can't duplicate it on Win8.1. I would ask you to be patient as I bring up a Win10 machine. Thanks.

Also, make sure everything is on a local hard drive (not a network drive) that is error free.

Patience is no problem. The file is on a local fast SSD without any problems. It may be system, rather than Windows version, related as I *think* I saw these errors before I upgraded from Win 8.1 to Win 10.

Another user has seen this error and troubleshot it to a bad version of Avisynth. What version are you using?

I have the same problem loong time. My workaround is before each cmdline dgi creation execute DGDecNV GUI, then it works in 100% cases.
Note that I am on Windows 7 x64 so this issue is not related neither to Win10 nor any specific stream.

I have notebook graphic card nVidia GeForce GTX 850M. And updating each time new drivers are released (aprox. once a month). Oh and I use avisynth+ because need x64 processing.

Thanks for the report, AJR. That's very interesting that starting the GUI prevents the issue. Maybe I have something uninitialized for the CLI path. I'll look into that and make some more attempts to duplicate this. Which version of Avisynth do you use?

I have started with Avisynth+ r1576 and use it around a year, now I am using r1825.

OK, thanks. I have now reproduced this but it is more like 1% for me and there are several different errors I see. Sometimes it is the MPEG1 error, sometimes CUDA init errors, which might get Guest 2's attention.

Investigating...

Yes agree, it is not very often

I think I'm on to something. The -e was invoking a direct WM_DESTROY instead of the correct WM_CLOSE, so proper cleanup was not done. Initial tests are promising but I want to let it run in a loop for several thousand executions before declaring victory.

Two thousand executions and counting...

Please re-download 2049 DGIndexNV and report your results. I did 5000 repeats without errors whereas previously I was getting errors within a hundred repeats.

First encoding was ok. But only the time will show more
I will let you know in next days/weeks.

Thanks for your testing, AJR, and please let me know if you experience further difficulties. For now, I mark it resolved based on my own testing (5 thousand executions without fail).