`bloq_is_clifford` treats `MultiTargetCNOT` as a single clifford gate

[anurudhp]

https://github.com/quantumlib/Qualtran/blob/683472a25a1bafb64fd5e5c58f79405737a4afd4/qualtran/resource_counting/classify_bloqs.py#L140

Is this the expected behaviour due to some possibly optimized implementation?

@tanujkhattar @mpharrigan

[mpharrigan]

instead of $n$ cliffords?

The notion of clifford counting is pretty poorly defined. In the surface code, you can do a multi-target cnot in 1 step (d cycles) -- the same speed as any other single clifford gate

[anurudhp]

Ah! I think we should document this in the code clearly.

[mpharrigan]

The old t complexity code counts all the constituent CNOTs. Since it has a decomposition (isn't an atomic/leaf bloq) maybe we should keep counting them. However: there's no support for symbolic bitsize, which causes tests to fail. This bloq could probably do with a re-write to make it more readable and including a build_call_graph

[wjhuggins]

This is an example where I think we would benefit from counting things in terms of qubits * surface code cycles, as advocated in #1454.