The use of CFD-DEM method to accurately simulate gas-solid flows in complex geometries is challenging, mainly due to the complexity related to the use of unstructured computational grids. In order to solve this problem, researchers have simulated gas-solid fluidized beds with complex geometries using CFD-DEM-IBM method with Cartesian grids. In present study the gas-solid flows in complex geometries were simulated using Cartesian grids following the basic idea of CFD-DEM-IBM method (De Jong et al., 2012), where the interactions between gas phase and complex geometries were firstly modelled using the immersed boundary method (IBM) implemented by Tukovic and Jasak (2012) in OpenFOAM (R). It was found that the computational efficiency is quite low. To improve the efficiency of the CFD-DEM-IBM solver, a new IBM method was then proposed by removing the neighboring immersed boundary cells from the interpolated extended stencil in the reconstruction of the velocity and pressure fields near the wall, and further proposing a new zero-gradient boundary condition to replace the original Neumann boundary condition for reconstructing the pressure field. Single-phase flow past a stationary cylinder, single-phase pipe flow, gas-solid flow in a cylindrical fluidized bed and fluidized bed with immersed tubes were simulated with four different IBM imposition methods to assess the accuracy and efficiency of sharp-interface CFD-DEM-IBM solver. It was shown that (i) the results of CFD-DEM-IBM simulations agree well with the reported experimental, analytical and/or numerical results available in literature; and (ii) the computational efficiency of newly proposed CFD-DEM-IBM solver is one or two order of magnitudes faster than that of the original IBM of Tukovic and Jasak (2012) in OpenFOAM (R), due to the fact that internal iterations are not needed anymore during the reconstruction of velocity and pressure fields.