Weighted Resize CRASH

[Selur]

Send you links to my current version.
Do these work for you?
Do you still get the non-printable character error?

Cu Selur

Ps.: removing all the whitespace does not fix the crash here when I enable trt.

[Selur]

Found a fix, adding:

os.environ["CUDA_MODULE_LOADING"] = "LAZY"

to the script seems to fix my issue.
-> will compile and upload a new dev and send you a link.

Cu Selur

[Dan64]

Hello Selur,

  the dev version Hybrid_dev_2023.01.30-201642.exe that you sent, is working for me.

Thanks,
Dan

[Selur]

Happy that, that version works for you.
Okay, strange that I need the lazy loading thing.

Cu Selur

[Dan64]

Hello Selur,

   I simple way to check if there are strange issues in the script is to call:

vspipe --info "script.vpy"

   If the output starts with "Width:" it's all Ok, otherwise it is probable that there are issues not allowing vspipe.exe to work.
   I guess that vsViewer is not using vspipe and its implementation is working better.

Dan

---

Hello Selur,

   I'm already using 'CUDA_MODULE_LOADING=LAZY'
   see my post: https://forum.selur.net/thread-2989-post...l#pid17521

Dan

[andry62]

Dear Selur.
I have the same problem. Can you send a link to
new dev version. Thanks.

[Selur]

send you a link.

Cu Selur

[Dan64]

Hello Selur,

  I noted that in the last dev version was added the support to FeMaSR. I'd like to try it, but I noted that the Streams box is locked to "1", so that I can't increase the number of Streams. It is possible to enable this box ?

Thanks,
Dan

[Selur]

Ah, atm. you need to enable CUDA Graph to increase the stream count.

[Dan64]

It is strange, usually the streams are supported with RT.
In any case even if I enable CUDA graph and increase the number of streams, this number is not passed in input to the function.
The generated code is the following

clip = FeMaSR(clip=clip, model=0, device_index=0, trt=False, nvfuser=False, cuda_graphs=True)