I read the paper GPU Daemon: Road to zero cost submission. It achieves a brilliant result on reducing queue and submission time cost. But I can't find an implementation or a demo to this project. So, could you please share an open sourced version to me? Thanks a lot. @Michal_M_Intel.
Hope you're doing well.
I want to let you know that we are currently checking your case internally.
Please expect a response soon.
Intel® Customer Support Technician
Thank you for your patience and your interest in OpenCL. The technique described in this paper used two bleeding-edge (at the time) features – fine-grain SVM and device-side enqueue – that unfortunately didn’t not see widespread adoption. We’re currently working to integrate a derivative capability based on learnings from the paper *transparently* into the OpenCL and Level Zero driver which can be leveraged directly or via oneAPI. Stay tuned to future driver releases for this new version of “near zero cost submission” capability.
Compute Software Architect