Enable OpenJ9 CUDA support if the SDK is installed on aarch64

[sxa]

We don't enable this as a matter of course, since our cloud machines do not have the CUDA development kit installed, but for anyone that has the CUDA 10.2 toolkit installed on a device such as the Jetson Nano (thread), this will enable the CUDA support to be built into OpenJ9.

Signed-off-by: Stewart X Addison sxa@redhat.com

[sxa]

This is a year old and seems to work (presumably it hasn't been reviewed because it was in draft while it underwent testing) so it would be good to push it in if no-one objects - tagging @andrew-m-leonard
@AdamBrousseau @keithc-ca Any objections to enabling this to be picked up in our build scripts if it exists on aarch64? I seem to remember someone else in the J9 team possibly had one of the devices too so it would be good to allow it to be picked up if someone is using our scripts on a Jetson.

[keithc-ca]

I didn't look, but I would hope there's a way to arrange to use a different version, i.e. the hard-coded path doesn't override an explicitly specified --with-code=/some/other/location. Other than that, I have no objections.

[andrew-m-leonard]

this has been stale for over a year, converting to "draft", if no longer needed please close.

[sxa]

I do want this in and working, although it may be superceded by the recent changes in #3735 ...

[keithc-ca]

Now that #3735 is in, this should just add aarch64 to the set of supported architectures for CUDA.