Recently, metal additive manufacturing (AM) has received considerable attention as a manufacturing process with the capacity to form complex shapes. In addition, metal AM can control the material microstructures, which dominate the mechanical properties of AM products, by tuning the scanning strategy. However, due to the numerous combinations of scanning paths, identifying an optimal scanning strategy via experimental trial and error is challenging, and thus, numerical simulations are essential in microstructure prediction. The phase-field method is the most accurate method used in mesoscale microstructure prediction. Chadwick and Voorhees combined the multi-phase-field (MPF) method with the Rosenthal equation to predict microstructure evolution via epitaxial growth during powder bed fusion. However, due to the high computational cost, their simulations were limited to a single path. We scaled up the simulation to multiple layers and tracks using multiple GPUs parallel computing. Nevertheless, due to the high computational cost, it was limited to simulations with low laser powers. In this study, we developed an efficient MPF simulation method that solves only around the melt pool to enable microstructure prediction using an arbitrary scanning strategy and practical scan conditions. Applying the developed method, we investigated the material microstructures obtained using various scanning strategies.