I can't test it and I'm not familiar with Metal, so my feedback is limited, but it looks good to me for the most part.
Is there any case where GPU_use_parallel_compilation() is false for the Metal backend?
Not sure about this semantic.
This is missing specialization_batch_is_ready.
This whole new entry is not needed (and actually causes the infinite processing time of regexes for some reason).
Note that, for me, removing the second entry without accepting an ending ,…
That did the trick! Thanks!
I've also removed the starting \. so it also works with pointers (->) and direct calls.
I thought it was cleaner to simply return, instead of having to do the check in DRW_gpu_context_create/destroy, but we can change it if you prefer it that way.
Done, but please take a look since I'm not familiar with condition_variable :
0c9286599ce49dbfdadf97ba96ff932556bd5a10
AFAIK, that's just the way the BLI_threads API works.
IIRC, there are other places where it's used for a single thread.
I thought about using a std::thread instead, but it doesn't seem to…
