Kris Rowe

After discussing this at the [OCCA TAF meeting](https://github.com/libocca/occa-taf/blob/main/meeting-notes/2022-06-29.md#device-synchronization) we will go with the second option, adding a new function `finishAll()` to the `occa::device` class.

Apologies for delay. The only thing I can see is that this may not be applicable in the Windows case, however adding some precompiler guards for now should likely be...

Currently you could use ``` modeStream_t* backend_stream = occa_device->getStream()->getModeStream(); ``` and then `dynamic_cast` this to the appropriate type before passing it to another library. Are you looking for this to...

You are correct. We will need to add a function to the frontend device API to make this happen.

Other than in the kernel launcher that is generated for the GPU backends, I don't think anything from the OCCA namespace is needed. It is probably reasonable to not include...

> Current state of the PR: I've split the occa dependency options into "include_occa" and "link_occa" since it seems harder to get rid of "include_occa" by default (and the linking...

> Hm, it looks like it's not MacOS, but OpenCL that is the culprit? I have no experience with that; does that one need occa linked in for the launcher...

This is also relevant for OpenCL and SYCL/DPC++ since the innermost `@inner` loop will be mapped to a sub-group. The new versions of the standards support sub-group barriers.

Agreed. Good call.

## Review checklist for @kris-rowe ### Conflict of interest - [x] I confirm that I have read the [JOSS conflict of interest (COI) policy](https://github.com/openjournals/joss/blob/master/COI.md) and that: I have no COIs...