Swarm intelligence systems realize group-level application tasks by information interaction of individual neighbors, and have sound robustness and flexibility. Meanwhile, most developers struggle to describe distributed and parallel individual interaction mechanisms. Some high-level languages allow users to program parallel swarm intelligence computing tasks in a serial mindset and from a global system perspective, without considering low-level interaction details such as communication protocols and data distribution. However, the huge semantic gap between user-oriented, globally declarative swarm intelligence system applications and individual parallel execution logic makes the compilation process complex and application development inefficient. Thus, this study proposes a compilation system and its supporting tools to support the conversion of high-level swarm intelligence system applications into secure and efficient distributed implementations. By parallel information identification, computing division, and interactive information generation, the compilation system compiles the swarm intelligence application program for global and serial programming into parallel object code for individual execution, and thus users do not have to understand the complex interaction mechanism among individuals. Additionally, a standardized intermediate representation of the compilation system is designed to convert complex swarm intelligence computing tasks into a standardized semantic module sequence composed of swarm intelligence operators and input and output variables, which represents source program information in a platform-independent form and shields the heterogeneity of target hardware platforms. The system is deployed and tested in a case platform of swarm intelligence systems. The results show that the compilation system can compile swarm intelligence applications into platform-executable object code and improve the application development efficiency, and its generated code has better performance than existing compilers in a series of benchmarks.