fix precomputation in get_objective_maxcut.py

[danlkv]

fixes #9

[rsln-s]

The tests are failing because the GPU test is not skipped properly if no GPU is available (see logs in the check above):

=================================== FAILURES ===================================
____________________________ test_maxcut_precompute ____________________________

    def test_maxcut_precompute():
        N = 4
        G = nx.random_regular_graph(3, N)
        print(G.edges())
        for u, v, w in G.edges(data=True):
            w["weight"] = np.random.rand()
        precomputed_cuts = precompute_energies(maxcut_obj, N, w=get_adjacency_matrix(G))
        simclass = qokit.fur.choose_simulator("gpu")
        terms = get_maxcut_terms(G)
>       sim = simclass(N, terms=terms)

tests/test_maxcut.py:93: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
qokit/fur/nbcuda/qaoa_simulator.py:24: in __init__
    super().__init__(n_qubits, costs, terms)
qokit/fur/qaoa_simulator_base.py:63: in __init__
    self._hc_diag = self._diag_from_terms(terms)
qokit/fur/nbcuda/qaoa_simulator.py:31: in _diag_from_terms
    out = numba.cuda.device_array(self.n_states, dtype="float32")  # type: ignore
/opt/hostedtoolcache/Python/3.9.18/x64/lib/python3.9/site-packages/numba/cuda/cudadrv/devices.py:231: in _require_cuda_context
    with _runtime.ensure_context():
/opt/hostedtoolcache/Python/3.9.18/x64/lib/python3.9/contextlib.py:119: in __enter__
    return next(self.gen)
/opt/hostedtoolcache/Python/3.9.18/x64/lib/python3.9/site-packages/numba/cuda/cudadrv/devices.py:121: in ensure_context
    with driver.get_active_context():
/opt/hostedtoolcache/Python/3.9.18/x64/lib/python3.9/site-packages/numba/cuda/cudadrv/driver.py:499: in __enter__
    driver.cuCtxGetCurrent(byref(hctx))
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

[rsln-s]

Shouldn't this function call fail if no GPU is present? I checked on a CPU-only instance, and it does not throw an error

[danlkv]

precompute_energies is the CPU-only function, so it shouldn't fail. If you use the simulator from choose_simulator("gpu") it should fail on initialization.

[rsln-s]

Github comments are a bit ambiguous; I was referring to line 91 in particular. Running the following two lines in a CPU-only instance does not throw an error:

import qokit
qokit.fur.choose_simulator("gpu")

[danlkv]

Yes, this is expected, since choose_simulator returns the class, which is then used in later steps of simulation. Using this class will raise an error

[rsln-s]

Tests should pass both on CPU and GPU before the review is requested

[rsln-s]

LGTM, thanks for fixing promptly!