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

Build on Linux #2

Closed
jkfindeisen opened this issue Aug 9, 2017 · 11 comments
Closed

Build on Linux #2

jkfindeisen opened this issue Aug 9, 2017 · 11 comments

Comments

@jkfindeisen
Copy link
Collaborator

For example with Ubuntu 16.10 (https://packages.ubuntu.com/yakkety/nvidia-cuda-toolkit)

@jkfindeisen
Copy link
Collaborator Author

And add build instructions to the documentation.

@mscipio
Copy link

mscipio commented Nov 18, 2017

In the guide you just talk about Windows ... do you think it is feasible/simple to compile the code in Linux (I have Debian or CentOS with already cuda and python up and running)?

@superchromix
Copy link
Collaborator

Have a look at the fork of Ingmar Rieger, here: https://github.com/irieger/Gpufit

@mscipio
Copy link

mscipio commented Nov 20, 2017

Hi!
I did and I managed to compile it and to run a couple of examples (there is something to change in the source, at least in my case, I will pull a merge request when I get everything to work)

I have 2 question now:

  1. when I run the comparison tests GPUvsCPU I get a "Floating point exception" as soon as it starts to run code on the GPU. This error seems not to appear in other tests (like SimpleTest, LinearRegression, etc ...). do you think it has to do with my not-so-powerful GPU? Or maybe it is a 32/64 bit issue?

  2. When I try to run the examples built with the python wrapper I got an error related to a Gpufit.dll file not found. As you commented in the code, this file is supposed to be there only on windows ... is there any way I can do something in python on Linux as well?


EDIT: I found an answer for the second issue
You just need to copy the compiled linux library libGpufit.so in the <src_folder>/Gpufit/python/pygpufit and change the following 2 lines of code, before running the setup.py installation:

  • [<src_folder>/Gpufit/python/setup.py # 39] package_data={'pygpufit': ['*.so']},
  • [<src_folder>/Gpufit/python/pygpufit/gpufit.py # 16] lib_path = os.path.join(package_dir, 'libGpufit.so') # this will only work on Linux

--> I made a pull request to the fork of Ingmar Rieger. If you want, give it a look and check if you can merge it with the main one. I didn't check the differences of the fork of Ingmar Rieger with the master, but I tried to keep my changes able to work also for the windows version.

@jkfindeisen
Copy link
Collaborator Author

jkfindeisen commented Nov 28, 2017

I'm in the process of making Gpufit run on Linux. I copy from the forks of Ingmar Rieger and Michele Scipioni and acknowledge their contributions, but do not pull commits directly because partly they are doing other stuff that is not needed and I can test on Windows and Linux at the same time.

See 2a85a80 and before and after.

@mscipio
Copy link

mscipio commented Nov 28, 2017

If you want you can check my current pull request (without accepting it).
I just tested out your current linux-ready version on Debian and it is almost working: I needed to do just a couple of minor changes to make it compile (mostly one import in gpufit.h required by the test files).

@jkfindeisen
Copy link
Collaborator Author

Thanks, I will do this.

@jkfindeisen
Copy link
Collaborator Author

Depends on #36

@jkfindeisen
Copy link
Collaborator Author

jkfindeisen commented Nov 29, 2017

Depends on #38

@jkfindeisen
Copy link
Collaborator Author

Since 5bf8ecf the project builds on Linux and the tests as well as the performance comparison run successfully. There is documentation about the Linux build.

Still to do:

  • test Python binding (should work)
  • fix Matlab binding (should not work, Matlab built with different gcc version)
  • packaging?

@jkfindeisen
Copy link
Collaborator Author

The project compiles and runs on Linux well but we never produced binary releases for Linux and probably never will. Therefore Linux is fully supported.

lsaca05 referenced this issue in lsaca05/Gpufit Aug 3, 2022
Cleaned up some code and progressed a bit on the additional models.
jimkring added a commit to JKISoftware/Gpufit-fork that referenced this issue Oct 20, 2022
More test enviornments, updated find python
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

3 participants