Additional support for qubit array allocation

[ColbyDeLisle]

Wanting to gauge interest in adding Q# support for the __quantum__rt__qubit_allocate_array QIR function.

I think it could be useful in particular for the Backend trait to be extended so that it's able to distinguish between a user allocating single qubits (via use q = Qubit(); in Q#) and them allocating an array of qubits all at once (via use qs = Qubit[10];).

Currently a qubit array allocation is treated as a sequence of calls to the single qubit allocation instruction (this is done here), and Backend in turn has just one function for handling allocation of single qubits at a time. But it seems reasonable to treat __quantum__rt__qubit_allocate_array as its own intrinsic, and add another function to Backend which handles it separately from __quantum__rt__qubit_allocate.

Happy to chat more or pitch in as needed. Thanks!

[swernli]

Hi @ColbyDeLisle, thanks for the feedback. I think what you describe would work fine in simulation but run into some challenges with QIR specification compatibility. Specifically, to allocate an array of qubits the function would need to return an array, but dynamically sized arrays are not part of the Base or Adaptive profile specification. In fact, the calls to __quantum__rt__qubit_allocate get compiled out to comply with this part of the QIR spec:

To be compliant with the Base Profile specification, the program must not make use of dynamic qubit or result management, meaning it must use only the second kind of pointer; qubits and results must be identified by a constant integer value that is bitcast to a pointer to match the expected type. How such an integer value is interpreted and specifically how it relates to hardware resources is ultimately up to the executing backend.

This means that qubits passed to gates must be constant integer identifiers, and can't be pointers to other data structures like an array of qubits that comes from a call to __quantum__rt__qubit_allocate_array. However, this is an ongoing effort to organize a QIR Alliance workstream to define specification for dynamic qubits and array usage that would allow for a re-introduction of array allocation (see qir-alliance/.github#46). We are hoping to participate in that workstream and would include discussion of __quantum__rt__qubit_allocate_array as part of it.

[ColbyDeLisle]

Thanks @swernli, that link clears things up for me. Happy to hear it's being discussed, I'll keep an eye on that workstream!