[incubation] installing Nvidia's CUDA libraries on OpenJ9 CI machines

["Mark Stoodley" <mstoodle@xxxxxxxxxx>]

Hi,

The OpenJ9 project has a special kind
of licensing question that we're not quite sure how to deal with.

The OpenJ9 JVM can build into itself
support for using GPU processors via Nvidia's CUDA libraries, but in order
to do that, the CUDA toolkit needs to be installed on the machine used
to build OpenJ9 and then we configure OpenJ9's build to include certain
header files. OpenJ9 does not absorb anything meaningful (i.e. no source
code is built into OpenJ9) from the CUDA header files but there is an End
User License Agreement for the toolkit and runtime that requires agreement
to some terms in order to be installed anywhere.

We want to build and test this functionality,
so we want to have the library installed on the CI machines where we test
our pull requests before merging them. Of course, we also want to be able
to actually then run Java workloads using the GPU functionality on those
machines so that we can quickly discover any problems being introduced
into our GPU support code.

For reference, here's a link to this
EULA: https://docs.nvidia.com/cuda/eula/index.html

Our project does not distribute binaries
(because OpenJ9 can only be built as a part of OpenJDK, so it's actually
the AdoptOpenJDK community that builds OpenJDK with OpenJ9 binaries) and
we're not "distributing" any part of the CUDA library, so I didn't
think a CQ would be appropriate, though I'm happy to file one if that's
the right way forward.

I'm also not really sure who would be
the one entering into this agreement on behalf of the OpenJ9 project. Would
it be one of the project leads? The foundation?

Thanks for any help!

Mark Stoodley		8200 Warden Avenue
Senior Software Developer		Markham, L6G 1C7
IBM Runtime Technologies		Canada
Phone:	+1-905-413-5831
e-mail:	mstoodle@xxxxxxxxxx
We cannot solve our problems with the same thinking we used when we created them - Albert Einstein