ub-quantic / dqcsim-iqs Goto Github PK

Blocked as Intel QS does not support control gates with more than one control qubit (e.g. Toffoli).

An alternative solution is displayed in https://github.com/iqusoft/intel-qs/blob/104de88196e6325e7a831591a078c4f43013defb/src/qureg_applytoffoli.cpp

Intel QS has 3 modes of execution depending on the quantum state distribution, as stated in Intel Quantum Simulator: A cloud-ready high-performance simulator of quantum circuits (Guerreschi et al.):

• Single distributed quantum state: A single instance of the quantum state is simulated and it is distributed along all the computers.
  • Used when you are looking for the highest possible amout of qubits.
• Pool of distributed states: Multiple instances of the quantum state distributed into nodes.
• Pool of local states: Multiple instances of the quantum state on local memory.
  • Used when you are looking to run circuits in parallel (not our interest) or to simulate noise effects (may be of interest).

Currently, only Single distributed quantum state is supported in dqcsim-iqs and by the way the dqcsim host plugin is designed, it is difficult to know if we could support the other execution modes. The main drawback is measurements. If we are simulating a hybrid algorithm, the frontend is only instanced once and so, on a measurement, we would expect only one result per measurement but we would have as many results as instances of the quantum state.

A temporal solution is to run different SLURM jobs as the different instances of the quantum state and run the simulator in the Single distributed mode.

An alternative would be to design our own host plugin that communicates with dqcsim-iqs by means of host commands. This host plugins would also be responsible of having as many instances of the frontend as instaces of the quantum state. But are we interested in pool simulations?

QC researchers usually use some popular Python packages such as Qiskit, ProjectQ, Cirq, PennyLane, ... The way this packages deal with the heavy simulation (or actual execution in a quantum computer) is that they abstract the quantum computer as an Engine or Backend. We could write some Python packages that implement the abstraction so we could offer supercomputer power to researchers without refactoring code.

• Which ones can we support?
• Which ones should we support?

How should we deploy our plugin? In which platforms? For which package managers?

• PyPI (Python Package Index):
  • Lets you install it using pip.
  • Looks like the de-facto package manager used by the authors of DQCsim.
  • A lot of QC researchers already use Python for simulations.

We also should automate the deployment linked to GitHub releases.

• Support MPI parallelization
• Support controlled-controlled gates (e.g. Toffoli) #2
• Add option flag for logging macros
• Choose a license