Copy between 2 gfx devices


#1

Hello, Is it possible to make use of mgpu power in opencl?
I want to create a context with many devices, allocate different memory objects, init and be able to copy memory objects within the context.

Issues Im facing-

  1. that each queue is associated with 1 device
  2. I cannot dispatch a kernel to a ‘context’ and have the workload be automatically distributed to all devices in the context

How can I copy memory from one buffer to another which are both associated with a single context but may have different devices?
How can I dispatch a kernel and have its compute be distributed to all devices in the context?
thank you for your help