Method exec

dnn/src/cuda/batched_matrix_mul/brute_force.cpp:66–83

            config.first[0], config.first[1], config.first[2]);
}
void BatchedMatrixMulForwardImpl::AlgoBruteForce::exec(const ExecArgs& args) const {
    auto N = args.layout_a.shape[0];
    auto config = prepare_sub_opr(args);

    rep(n, N) {
        TensorND A_, B_, C_;
        auto tensor_n_from_batch = [n](const TensorND& in, TensorND& out) {
            out.reset_ptr(static_cast<void*>(
                    static_cast<dt_byte*>(in.raw_ptr()) +
                    n * in.layout.stride[0] * in.layout.dtype.size()));
            out.layout = in.layout.remove_axis(0);
        };
        tensor_n_from_batch(args.tensor_a, A_);
        tensor_n_from_batch(args.tensor_b, B_);
        tensor_n_from_batch(args.tensor_c, C_);
        config.second->exec(A_, B_, C_, args.workspace);
    }
}