TBB with OpenCL is a great feature. When it will be part of the nominal TBB version?
Are you speaking about tbb::flow::opencl_node? We are looking for the use cases to see whether it is useful to people or not. Based on this information, we are going to make a decision about full support of this feature.
We would love to know your use case and hear why opencl_node is useful to you. Could you please share these details with us?
I hope for it to be useful to me - for the ability to harness a large number of processors with TBB's cache-friendly, work-stealing task queue, for combinatorial optimization. My plan is to compile with SYCL (at least to start with) and load the binary into TBB.
I would recommend to use DPC++ (SYCL) if you are using OpenCL to access GPU (or FPGA). We currently recommend to use DPC++ for work with accelerators instead of OpenCL TBb FG node. I'm not sure what do you mean by "load the binary into TBB"?