dmikushin/cppsim

ProjectQ-compatible optimized C++/GPU simulator backend replacement

_cppsim

This is a standalone simulator backend for the ProjectQ framework, extended to generate Haener-Steiger quantum kernels for arbitrary number of qubits.

Description

The original code provides handwritten kernels of up to 5 qubits in the following form:

template <class V, class M>
inline void kernel_core(V &psi, std::size_t I, std::size_t d0, M const& m)
{
    std::complex<double> v[2];
    v[0] = psi[I];
    v[1] = psi[I + d0];

    psi[I] = (add(mul(v[0], m[0][0]), mul(v[1], m[0][1])));
    psi[I + d0] = (add(mul(v[0], m[1][0]), mul(v[1], m[1][1])));

}

// bit indices id[.] are given from high to low (e.g. control first for CNOT)
template <class V, class M>
void kernel(V &psi, unsigned id0, M const& m, std::size_t ctrlmask)
{
    std::size_t n = psi.size();
    std::size_t d0 = 1UL << id0;
    std::size_t dsorted[] = {d0 };
    std::sort(dsorted, dsorted + 1, std::greater<std::size_t>());

    if (ctrlmask == 0){
        #pragma omp for collapse(LOOP_COLLAPSE1) schedule(static)
        for (std::size_t i0 = 0; i0 < n; i0 += 2 * dsorted[0]){
            for (std::size_t i1 = 0; i1 < dsorted[0]; ++i1){
                kernel_core(psi, i0 + i1, d0, m);
            }
        }
    }
    else{
        #pragma omp for collapse(LOOP_COLLAPSE1) schedule(static)
        for (std::size_t i0 = 0; i0 < n; i0 += 2 * dsorted[0]){
            for (std::size_t i1 = 0; i1 < dsorted[0]; ++i1){
                if (((i0 + i1)&ctrlmask) == ctrlmask)
                    kernel_core(psi, i0 + i1, d0, m);
            }
        }
    }
}

The proposed generator reproduces the hand-written kernels, and extends support to unlimited number of qubits.

Testing

mkdir build
cd build
cmake .. -G Ninja
ninja
./test_nointrin
./benchmark