Page 1 of 1

Re: Avs2YUV error with DGDecodeNV

Posted: Fri Apr 27, 2018 8:47 am
by Guest 2
admin wrote:
Fri Apr 27, 2018 8:14 am
What is your OS? And what graphics card do you have?
Windows 10 x64 rs5 17655 / GTX 760

Re: Avs2YUV error with DGDecodeNV

Posted: Fri Apr 27, 2018 9:00 am
by admin
That's a Kepler so theoretically OK with 397.31.

Re: Avs2YUV error with DGDecodeNV

Posted: Fri Apr 27, 2018 9:52 am
by Guest 2
admin wrote:
Fri Apr 27, 2018 9:00 am
That's a Kepler so theoretically OK with 397.31.
Not in real life :lol:

Re: Avs2YUV error with DGDecodeNV

Posted: Fri Apr 27, 2018 10:18 am
by admin
So it seems. :cry:

Re: Avs2YUV error with DGDecodeNV

Posted: Fri Apr 27, 2018 11:18 am
by admin
I'm having one issue so far. VirtualDub(2) crashes on exit in a cuda module.

Gonna revert the driver.

EDIT: 391.35 does not crash.

Re: Avs2YUV error with DGDecodeNV

Posted: Fri May 04, 2018 5:00 am
by Guest 2
http://nvidia.custhelp.com/app/answers/detail/a_id/4661

Give them a spin.

P.S: Just discovered that Windows + CTRL + Shift + B restart video driver.

Re: Avs2YUV error with DGDecodeNV

Posted: Fri May 04, 2018 10:24 am
by admin
Does that hotfix driver fix the original issue you had, i.e., crashing on exit in some apps?

Re: Avs2YUV error with DGDecodeNV

Posted: Sat May 05, 2018 4:16 am
by Guest 2
admin wrote:
Fri May 04, 2018 10:24 am
Does that hotfix driver fix the original issue you had, i.e., crashing on exit in some apps?
Nope, just tried and same error when using DGSource.

Re: Avs2YUV error with DGDecodeNV

Posted: Sat May 05, 2018 7:30 am
by admin
Thanks for the testing. Fixed in slipstream 147.

Re: Avs2YUV error with DGDecodeNV

Posted: Sat May 05, 2018 12:59 pm
by Guest 2
admin wrote:
Sat May 05, 2018 7:30 am
Thanks for the testing. Fixed in slipstream 147.
You are always welcome.

What did they change, at the end?

Re: [SOLVED] Avs2YUV error with DGDecodeNV

Posted: Sat May 05, 2018 2:03 pm
by admin
It was my bug and a change in the driver exposed it. I was freeing some CUDA host memory in a thread without setting the correct CUDA context. Previously it apparently didn't matter. One might think that the address is sufficient, but it is not now true.

Still, nVidia should return a proper error and not crash.