Irregular Repeat Accumulate (IRA) codes' decoding usually adopts Belief Propagation (BP) decoding algorithm, but BP decoding algorithm needs hyperbolic tangent calculation, so its hardware implementation is very difficult because of the high complexity. A decoding algorithm combining the early detection mechanism and offset polyline was put forward. Its performance would approach to BP algorithm via non-uniform error compensation for polyline approximation decoding algorithm. And the early detection method was introduced which observed the transmitted information of check nodes in advance, judged the lines' log-likelihood value which had negligible influence on the next iteration and moved it out of the iteration. So the computational complexity of next iterations was reduced. The simulation results show that the proposed algorithm greatly reduces the computational complexity through the offset polyline approximating the hyperbolic tangent, and the decoding performance is close to BP algorithm.