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

added make options for gpu build to mxnet #609

Merged
merged 6 commits into from
Sep 27, 2018
Merged

added make options for gpu build to mxnet #609

merged 6 commits into from
Sep 27, 2018

Conversation

bitstormFA
Copy link
Contributor

I have added options for gpu / cuda build.
To use cuda set CUDA_HOME to the folder containing cuda (e.g. /opt/cuda for arch linux) in the environment of the build
To use cudnn set USE_CUDNN=1

The has only been tested on arch linux_x86_64 where it compiles

@saudet
Copy link
Member

saudet commented Aug 25, 2018

Thanks! That's a good start. Let's include it as an "extension" just like with Caffe and TensorFlow (check their cppbuild.sh, presets/caffe.java, and presets/tensorflow.java files), and add a separate entry for it to .travis.yml as well, so that CUDA gets installed when it gets to build it.

@bitstormFA
Copy link
Contributor Author

I have added the extension handling but I am not sure how to test it. Please check if CUDA home needs to be changed

@saudet
Copy link
Member

saudet commented Aug 26, 2018

Thanks! Let's wait and see what Travis CI has to say about it...

@saudet
Copy link
Member

saudet commented Aug 27, 2018

The build passes, but it doesn't appear to be compiling any CUDA kernels...?

@bitstormFA
Copy link
Contributor Author

I think the problem is the different CUDA_HOME path. I have seen that the cuda project expects the standard /usr/local/cuda, I had /opt/cuda so it worked only on arch.
I have changed this now and it a test build with a softlink works fine here.

Sorry for missing that.

@saudet
Copy link
Member

saudet commented Aug 29, 2018

Looks like the build is failing:

In file included from src/operator/contrib/./../tensor/sort_op.h:104:0,
                 from src/operator/contrib/./bounding_box-inl.h:37,
                 from src/operator/contrib/bounding_box.cu:27:
src/operator/contrib/./../tensor/./sort_op-inl.cuh:34:44: fatal error: cub/device/device_radix_sort.cuh: No such file or directory
 #include <cub/device/device_radix_sort.cuh>

https://travis-ci.org/bytedeco/javacpp-presets/jobs/421178460

BTW, it's basically building inside a Docker container using "nvidia/cuda:9.2-cudnn7-devel-centos7" as image, so you could run with that locally to make sure it works alright.

@saudet
Copy link
Member

saudet commented Sep 27, 2018

Using the release archives seems to have fixed that issue!

@saudet saudet merged commit 4fe8952 into bytedeco:master Sep 27, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants