This contribution studies adaptive power allocation of low-density parity check-code (LDPC) coded modulation on fading channels, where in this case, the symbols within a codeword suffer from different signal-to-noise ratios (SNRs). We consider adjusting the transmission power for each symbol within the codeword so as to improve the transmission performance, through an optimization based on Gaussian approximation. Specially, we first give the expression of the optimization problem, and further propose the solution based on Lagrangian method. Numerical results demonstrate the performance gain brought by the proposed power allocation method.