Classical Control: Feedforward

[balopat]

Problem:

Cirq currently only supports a single unrolled circuit. Measurements results cannot be fed forward to influence the future execution of the circuit. This leads to, for example, Cirq not being able to easily describe a teleportation experiment. There are also use cases for simple feed forward in variational ansatz and simple quantum error correcting protocols.

Depends on feature: Classical data (#3231)

Rough requirements:

• Define how classical data can influence future execution
• how is control specified
• what is changed by the result of the control
• support of multiple cases
• Fit in with the specification of data in the Classical Data feature.
• Support serialization of the specification of control and of the action resulting from the control.

[smitsanghavi]

A previous feature request #762 was closed as intended because we didn't know how the hardware was going to support this. Do we have more idea now? Any specific requirements due to hardware constraints?

[daxfohl]

@95-martin-orion @smitsanghavi @MichaelBroughton Initial RFC here: https://tinyurl.com/cirq-feedforward. I'm reasonably happy with it as a first draft.