Example code for heterogeneous workflow application

[vrpascuzzi]

N.B. This is more of a discussion than issue.

First, thanks for materializing these ideas.

I'm wondering how one could utilize XACC across heterogeneous resources. For example, something interesting to me is running a VQA with a simulator estimating the cost function and the [also potentially computationally expensive] classical optimization part executing asynchronously using CPUs+GPUs. Going further, I could imagine a larger workflow comprising multiple ansatzes running concurrently on >1 QPU simulators or even across multiple nodes and optimizing concurrently for each ansatz.

What if I had a variety of classical platforms---e.g. CUDA, AMD, Intel--- at my disposal? Is the solution a directive-based (e.g. OpenMP, MPI) approach?

Thanks in advance.