Function lu_inplace

src/backend/cuda/lu.cpp:114–136

template<typename T>
Array<int> lu_inplace(Array<T> &in, const bool convert_pivot) {
    dim4 iDims = in.dims();
    int M      = iDims[0];
    int N      = iDims[1];

    Array<int> pivot = createEmptyArray<int>(af::dim4(std::min(M, N), 1, 1, 1));

    int lwork = 0;

    CUSOLVER_CHECK(getrf_buf_func<T>()(solverDnHandle(), M, N, in.get(),
                                       in.strides()[1], &lwork));

    auto workspace = memAlloc<T>(lwork);
    auto info      = memAlloc<int>(1);

    CUSOLVER_CHECK(getrf_func<T>()(solverDnHandle(), M, N, in.get(),
                                   in.strides()[1], workspace.get(),
                                   pivot.get(), info.get()));

    if (convert_pivot) { convertPivot(pivot, M); }

    return pivot;
}

bool isLAPACKAvailable() { return true; }