--- Quote Start ---
The cyclical design worked with the emulator because I believe it was simply storing all the channel writes in memory before proceeding with the next kernel. The emulator must have been smart enough to switch kernels upon blocking and return to each afterwards.
--- Quote End ---
OK, that makes sense.
--- Quote Start ---
When you're launching your kernels I'm assuming you're launching them all to run concurrently from the host correct? (As in all of them are using different command queues).
--- Quote End ---
Yes, but they don't necessarily have to run in parallel. Running them sequentially in the order I call them should work just fine, as long as the channel data can be buffered.
--- Quote Start ---
I'm not sure if there was a way we can direct message each other from here, but perhaps we could collaborate via email or google hangouts?
--- Quote End ---
Sure, you can find my contact information here (
https://mscharrer.net/contact/). Shoot me an email and we'll go on from there.
Honored Contributor
