-
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
Compilation error on test script #37
Comments
Hi @chaithyagr , thank for the report. I do not have yet cuda 10.2 installed on any system... and will not be able to install it before next week. But commit 185ed8c should have fixed the problem. Can you try if you have time? b. |
Hello @bcharlier , Thank you for the quick response. As I mentioned earlier, I had issues while building pykeops locally also. I fetched the lastest masterm and tried a build. I still have issues with both pykeops and keops build: make VERBOSE=1
Now If I just run the failing command with
|
Thank you for the feedback. Well.. my guess was that this is precisely the expected behavior of CMAKE_CUDA_STANDARD... I will check this later. b |
Some things I tested:
In order to unblock myself, I went ahead and added -std=c++14 as an explicit CUDA_NVCC_FLAG, and now, I see things building just fine. However, in the import I get the classic error : Any idea on how to fix this? Also, my gcc is 5.4, cmake is 3.15 and CUDA is version 10.0 (I went from 10.2 to 10.0 assuming that could help fix things) |
Hi @chaithyagr , I checked on my various system and the compile flag std=c++11 is not explicitly needed with nvcc... Can you send me the id of your linux distro and the full output of
b. |
Here is the attached requested logs:
|
Adding here, However @bcharlier , as I mentioned in my PR, this seems to be a bit hardcoded solution. |
what if you try with a newer gcc ? |
Yes, I have been wanting to do this, but I seem to be facing a frozen ubuntu xenial base, due to which I cannot update gcc that easily. I plan to use a docker for this, however, as everything is working on google colab (that works with docker), I expect everything to work fine and thus I havent really done this. Would you want me to try this? |
@bcharlier do you mean to try the same master codes on later gcc? I think that will work as that is the exact same environment as that in google colab. Also, I wish to use my changes in some codes I am working on. While my changes or a better fix reaches to mainline and in a version of pykeops, can I still point my codes to install pykeops from my github location? |
Hello all, |
Thanks @ReubenDo , so the only question is whether we need my PR at this point, but otherwise I think this issue is resolved. |
Hi @chaithyagr , I close the issue since the problem was configuration specific as far as I understood. Best, b. |
I seem to be facing a rather weird issue on my local test with CUDA 10.2, gcc5.4, cmake3.10 and cmake3.12 (Fails with both cmakes)
I had faced similar issue while trying to build pykeops locally, and I found that adding
-std=c++11
arg for nvcc helped. Now, I am not sure if this is an issue in keops or cmake. Note that this works fine on Google Colab. So I think this is surely more about my environment, and I could not find sufficient resources online other than -std=c++11 to fix this.The text was updated successfully, but these errors were encountered: