Paul Richard Thomas wrote:
That surely is a lot of janitorial work :-) LGTM.
Thanks for looking at the patch! :-)
It was committed as r16-5633-g26d41e245dbba3.
While it is more than janitorial, for now ('sorry, unimplemented')
those changes do no really help users that much.
Thanks for pushing forward on this. Do I take it that you will
implement device_type(host) in gcc or is somebody else onto it?
It is marked as currently unimplemented in our
OpenMP implementation status (5.1 feature) at
https://gcc.gnu.org/onlinedocs/libgomp/OpenMP-5_002e1.html#Other-new-OpenMP-5_002e1-features
and will eventually be implemented, whether by me or
someone else remains to be seen.
Additionally, for 'host' we should disable all
device code generation (we currently don't) and
for 'nohost' we also should do better and remove
the symbol node on the host.
There is also some C/C++ work still to be done.
Tobias
