Flash calculations are a performance bottleneck of compositional flow simulations. Some work has demonstrated the feasibility of using sparse grid techniques to remove the bottleneck, but a complete realisation of the idea is still not available. Thus, this work fills the niche. By introducing a new concept of layer to sparse grid points, the sparse grid construction can become much efficient. As a result, a much easier data structure the array can be used to store the sparse grids. Compared with the popular data structures to store the sparse grids such as the hash table and the tree, the array can minimize the space size and the traversing time, and at the same time reduce the number of points in the sparse grids by removing the architecture ancestors in the tree, which in turn makes parallelization of flash calculations come true. All of them are not only contributions to flash calculations, but also contributions to existing sparse grid techniques. Moreover, both of the sparse grid construction and interpolation algorithms can be done in parallel. Different from the former parallel algorithms in sparse grid techniques, which have troubles in decomposing the domain equally and keeping load balance among the processors, our parallel algorithm can achieve load balance easily among the threads for any sparse grid configurations. Lastly, multicomponent experiments are also carried out to demonstrate the accuracy, correctness and efficiency of the algorithms. (C) 2018 Elsevier B.V. All rights reserved.