Pycuda import error

[harskish]

Pycuda import error

[MineThingIssues]

You have to install PyCuda yourself, and make sure that when you run configure.py you enable OpenGL, as that is required for this to work. Instructions can be found here: https://wiki.tiker.net/PyCuda/Installation

[natoucs-datagen]

Thanks!
I managed to install PyCuda through the pip module inside the conda environment.

anaconda/envs/ganspace/bin/pip install pycuda

However this leads to:

ImportError: PyCUDA was compiled without GL extension support

I am not sure how to add GL extension support with the PyCuda installed through pip.
So I will try what you suggested.

[natoucs-datagen]

So I managed to install from your link @MineThingIssues.
And I used:
python configure.py --cuda-root=/usr/local/cuda --cuda-enable-gl
to make sure OpenGL is used.
I did all the steps while keeping my conda env active and then I ran and passed all the unittests in the pycuda /test folder.
However I still get the same error.

Somebody knows why ?

[natoucs-datagen]

Ok I managed to install PyCuda inside the conda environment, deleted PyCuda files from the previous pip install there and succesfully ran interactive.py.

[natoucs-datagen]

Actually now that I have solved some other bugs which where blocking execution and not related to PyCuda, I get:

RuntimeError: make_default_context() wasn't able to create a context on any of the 2 detected devices

This is when running interactive.py.
System details:
2 GPUs GeForce RTX
NVIDIA-SMI 440.64.00 | Driver Version: 440.64.00 | CUDA Version: 10.2
CentOS Linux 8

Anybody went through this and was able to solve it ?

[martindisley]

I'm having also having some trouble installing PYCUDA correctly.

I followed instructions from @MineThingIssues + @natoucs-datagen above.
cd pycuda-2019.1.2/
python configure.py --cuda-root=/usr/local/cuda --cuda-enable-gl
sudo su -c 'make install

It seemed to install correctly (no errors), but I'm still getting this:

Traceback (most recent call last):
  File "interactive.py", line 26, in <module>
    from TkTorchWindow import TorchImageView
  File "/home/martin/workspace/ganspace/TkTorchWindow.py", line 16, in <module>
    from pycuda.gl import graphics_map_flags
  File "/home/martin/anaconda3/envs/ganspace/lib/python3.7/site-packages/pycuda/gl/__init__.py", line 5, in <module>
    raise ImportError("PyCUDA was compiled without GL extension support")
ImportError: PyCUDA was compiled without GL extension support

I previously installed PYCUDA via conda (because su -c "make install" failed for me unless I ran it sudo). Is it possible that it's getting confused between the two versions installed? i.e do I have to remove the env and repo and start again?

[DarkGeekMS]

In case you still have the same issue :

• Remove pycuda :

pip uninstall pycuda

• set CUDA_ENABLE_GL = True in siteconf.py.

• Rebuild pycuda with sudo make install.

This works for me!