CUDA cudaMallocPitch operation not supported

Hello,

I’m currently facing an issue with the call cudaMallocPitch. I get this message:

Exception : OpenCV(4.9.0) /[...]/opencv/modules/core/src/cuda/gpu_mat.cu:116: error: (-217:Gpu API call) operation not supported in function 'allocate'

I straced the execution, and I found out this:

openat(AT_FDCWD, "/dev/nvhost-ctrl", O_RDWR|O_CLOEXEC) = -1 ENOENT (No such file or directory)

I noticed this file no longer exists in Jetson Linux 36.2 (I looked up where it is created in the previous host1x.c file, function nvhost_user_init). I don’t see this in 36.2 (there are a bunch of other nvhost-ctrl-xxx files though)

Am I doing something wrong or this is just an upstream issue with the CUDA version released? (12.2)

Thank you!

Hi,

Just want to confirm first.
Have you built OpenCV with CUDA support on r36 first?

Thanks.

Hello,

I’ve built OpenCV 4.9.0 from source, using the existing CUDA installation on the provided rootfs (CUDA 12.2) I also tried building using 12.4, but it is not compatible (c.f. this issue).

Maybe I wasn’t clear, I am building in r36.2. The missing file was present in the previous release and is missing in r36.2

Hi,

We want to give it a try.
It looks like the building stage is working but fails when declaring a gpumat.
Is that correct? If yes, would you mind sharing a simple OpenCV test to reproduce the error?

Thanks.

Hello,

Sorry for the misinformation, but this was an error on our side: the application had been run on a different platform.

It seems to be running fine in the Orin NX.

Thank you.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.