Abstract: Numerical simulation is an important research method for laser fusion physics.Running on petaflop computers,the up-to-date numerical simulations usually output large-scale data with complex structures.In detail,the grid number reaches billions,and the data amount to tens of GB at a single time step,which cover a variety of complex meshes including multi-block deformed structured meshes and adaptive hierarchical meshes.Aiming at analysis demands of laser fusion physics,we investigate such large-scale and complex data systematically and propose several techniques of complexity decomposition and scale controlling,such as decomposition of complex data,scalable parallel visual analysis framework,efficient coupling of numerical simulation and visual analysis,and visualization method of high expressiveness.The proposed techniques are demonstrated by the visual analysis experiments with four simulation datasets.