Constraint solving has been applied to many domains of program analysis, and deeply applied in concurrent program analysis. Concurrent programs are specific domain software that has been widely used with the rapid development of multi-core processors. However, concurrent defects threaten the security and robustness of concurrent programs, thus it is of great importance to test concurrent programs. Due to the non-deterministic thread scheduling, one of the key challenges for concurrent program testing is how to reduce the thread interleaving space with exponential growth. The state-of-the-art approaches (i.e., J-MCR) tackle the challenge through constraint solving. However, it is found that there exist conflicts and redundancies inside constraints (i.e., the conflict of constraint clauses makes constraints unsatisfiable), solving those unsatisfiable constraints results in lower efficiency. Thus, a directed graph constraint-guided method called GC-MCR (directed graph constraint-guided maximal causality reduction) is proposed. By removing conflictive constraints and simplify redundant constraints, the times of constraint solving are reduced thereby further improving the efficiency. Comparing with state-of-the-art approach J-MCR, GC-MCR reduces the times of constraint solving by 19.36% on average and reduces the testing time on average by 34.01% on 38 concurrent programs.