Per #954, I notice that QBdt::Attach() is very bugged for a ket simulator attachment. In general, we probably need to revert and complete d742e03 for this to work, via PushStateVector() and PopStateVector(), by multiplying out amplitudes to full tree depth. This is surmountable, but the need to multiply out amplitudes would seem to limit our ability to attach stabilizer representation. However, if we limit our gate set to entirely Clifford/Pauli, QStabilizer attachments already work, (about 98% of the time on test_mirror_quantum_volume if "hacked" for only stabilizer gates, to 100%, indicating that there is only a small bug, at most). There seems to be a symmetry to stabilizer sub-states that precludes the need for PushStateVector() and PopStateVector().
Seeing as QBdt::Attach() is really only useful for QStabilizer, we can force this as a design requirement, and exclude ket simulator attachments. (If we first restrict the domain of our implementation just to barely what we need, we might re-orient ourselves faster in debugging.)
Per #954, I notice that
QBdt::Attach()is very bugged for a ket simulator attachment. In general, we probably need to revert and complete d742e03 for this to work, viaPushStateVector()andPopStateVector(), by multiplying out amplitudes to full tree depth. This is surmountable, but the need to multiply out amplitudes would seem to limit our ability to attach stabilizer representation. However, if we limit our gate set to entirely Clifford/Pauli,QStabilizerattachments already work, (about 98% of the time ontest_mirror_quantum_volumeif "hacked" for only stabilizer gates, to 100%, indicating that there is only a small bug, at most). There seems to be a symmetry to stabilizer sub-states that precludes the need forPushStateVector()andPopStateVector().Seeing as
QBdt::Attach()is really only useful forQStabilizer, we can force this as a design requirement, and exclude ket simulator attachments. (If we first restrict the domain of our implementation just to barely what we need, we might re-orient ourselves faster in debugging.)