-
Notifications
You must be signed in to change notification settings - Fork 65
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ImportError with PyInit_libKeOpsnumpy73a835aa5f module #2
Comments
Hi Hicham, Thanks for your detailed report! According to the information provided by KeOps' verbose mode, it seems that under the hood, CMake decided to use -- The CUDA Host CXX Compiler: /usr/bin/c++ Could you please report the result of the following command? /usr/bin/c++ --version If you get a reference to an old version of |
Hi Jean, Many thanks for your response! Here is a follow-up, as I am using the same machine. I installed gcc, 7.3.0 version, in my conda environment (called
I also checked that the right version of nvcc is installed:
Nevertheless, the same error seems to appear:
Where do you think I should investigate further ? Many thanks! Gwendoline |
Cmake is still using the gcc of your system:
which is the bad one. Can you send the output of
And then try to run your script with the CC and CXX env variable set to link to the right compiler
|
Hi, The command
Folder
|
are you using last version of torch (v1.1)? ps: if the gcc of your system is 7.4 this is not gcc pb and by the way the right command is : |
Hi,
However,
It is unclear to me why CXX compiler stilll uses the old version: |
ok, the gcc of your systeme is still 4.8. So, please :
b. |
It yields the following error:
Is it possible that there are 'incompatibilities' of some sort between my environment gcc (version 7.3.0) and the |
|
I did:
|
Dear Gwendoline, As far as I can see, your gcc are correctly set up but as @fradav pointed out to us, it can be a problem coming from the very last c++ standard. Then, can you perform the following steps:
import pykeops
print(pykeops.__file__) It will return the installation dir that you need.
hope it will work.... b. |
No need to edit the cmake. |
Hi,
|
My best guess is that the cuda compiler doesn't like the gcc 7.4.0 includes. We need to reproduce your error at this point. |
Hi everyone, I did a fresh install this morning, so it might be that there have been some update in pykeops install or so since the beginning of this thread ; but if needed this the current setup on my user session:
Note that I installed cmake3 using
with this, everything seems fine so far. |
Hi all, I'm facing the same problem as well. I installed gcc 7.4 and nvcc 10.0 and still getting the same problem. Any ideas ?
` [ 20%] Building NVCC (Device) object CMakeFiles/keopslibKeOpsnumpy73a835aa5f.dir/keops/core/keopslibKeOpsnumpy73a835aa5f_generated_link_autodiff.cu.o Done. During handling of the above exception, another exception occurred: Traceback (most recent call last): During handling of the above exception, another exception occurred: Traceback (most recent call last): |
Hello, great work !
I couldn't run the sample code below with cuda 10 and and cmake 3.14.4
I made sure I installed pykeops with all dependencies pykeops[full].
Here is the output:
The text was updated successfully, but these errors were encountered: