-
Notifications
You must be signed in to change notification settings - Fork 64
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
Compile Issue - No module named 'libKeOpstorchd5f55273e3' #56
Comments
Hi @kheyer , there is a mismatch between the detected CXX compiler
and the one you are expecting to use
Moreover, from my point of view, using the c++ compiler shipped by anaconda is not a good idea. You better have install a newer g++ from ubuntu apt repo. Here are your error:
|
Hi @bcharlier thanks for responding. I tried setting Anaconda to use the Anaconda install of g++, which uses version 7.3.0
This successfully changes the compiler used:
However the same errors are thrown. I tried using the non-Anaconda version of g++
Although this fails with
I tried totally uninstalling the Anaconda compilers and doing a fresh install of g++ and cmake using apt-install, but the error persisted. Searching around I found one reference to the following error:
On this issue, which points to CUDA compilation issues. Do you have any thoughts on how to go about debugging this? |
No change with your first attempt. This was the anaconda g++ already in use.
You definitely need a working decent c++ compiler on your system.
If you're admin on the machine
where b. |
Installing g++8 fixes the "not able to compile" error but the compilation errors remain.
At the start of compilation I see
Then the following errors are thrown
|
@fradav , an idea ? |
Hello, Please provide the whole cmake output, as it appears that a crucial flag have been borked in the process (-std=c++14). Did you clean your build dir when you upgraded to g++-8 (deleted the CmakeCache.txt file) ? Regards, |
I ran
I noticed that some of the outputs related to nvcc have the |
I figured it out. And wow my GPU memory usage just dropped from 11 GB to 300 MB |
Hi @kheyer , thanks for posting that solution - it works for me as well. Would someone mind explaining how to set the CXXFLAGS environmental variable so that it persists? For instance I can set it once and run the test script, but then every time I re-open a shell or the conda env it returns to the |
I'm trying to get KeOps working and I'm having compiler issues that appear to be related to compiling Pytorch bindings.
I read through similar issues (https://github.com/getkeops/keops/issues/28, https://github.com/getkeops/keops/issues/49, https://github.com/getkeops/keops/issues/8), but these mention fixes in v1.4 release. I'm running into the following issue using the v1.4 pip install.
Test script:
Terminal output:
Compiler settings:
Compilers are installed locally using Anaconda.
cmake version 3.13.3
nvcc Install
Any assistance with this would be helpful.
The text was updated successfully, but these errors were encountered: