Skip to content
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

Make fails for RAJA #9

Closed
tomdeakin opened this issue Aug 26, 2016 · 4 comments
Closed

Make fails for RAJA #9

tomdeakin opened this issue Aug 26, 2016 · 4 comments

Comments

@tomdeakin
Copy link
Contributor

CUDA and RAJA interfeer in CMakeLists.txt

@DavidPoliakoff
Copy link

Hey, I'm running into this as well. I get RAJA to build if I add

set_source_files_properties(main.cpp PROPERTIES CUDA_SOURCE_PROPERTY_FORMAT OBJ)

But then the OMP3 build crashes. Don't know whether it's possible to fix along those lines.

@tomdeakin
Copy link
Contributor Author

We haven't found a way of getting RAJA to play nicely with CMake; we also find that CUDA breaks if you add that line. There are lots of other issues with the CMake too as lots of flags are missing for many of the models. We are just working out what it going to work best and what is needed across the range of devices across the platforms we use.

In the mean time, we're building manually with this command:
nvcc --expt-extended-lambda -arch=sm_52 -O3 -std=c++11 -DUSE_RAJA main.cpp RAJAStream.cpp -x cu -L$RAJA_PATH/lib -lRAJA -Xcompiler -fopenmp

You may also have to make sure common.h exists. It just contains
#define VERSION_STRING 2.1

@DavidPoliakoff
Copy link

Hey Tom, thanks for the update.

In our current implementation we have an (everchanging) custom FindCUDA.cmake which we use to handle a lot of ugliness (including how to do CUDA compilation with the Clang CUDA compiler). I might file an issue on our repo about this, it seems like the kind of thing that ought to work.

Honestly, some of the problem is that we're all reconciling different CUDA oddities, and resolving them in different ways. Let me get back to this when I have time, but thanks again, you've brought a lot of good feedback to the project, it's highly appreciated

@tomdeakin
Copy link
Contributor Author

Fixed in v3.1. We've moved away from CMake so can better control how we build against RAJA. New version works for all combinations we tried in our usual results matrix.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants