编译群体智能系统应用程序:以全分布式智能建筑系统为例

群体智能系统通过邻居个体的信息交互实现群体级别的应用任务,具有良好的鲁棒性和灵活性.与此同时,大多数开发人员难以对分布式、并行的个体交互机制进行描述.一些高级语言允许用户以串行思维方式、从系统全局角度来编程并行的群体智能计算任务,而无需考虑通信协议、数据分布等底层交互细节.但面向用户、全局声明式的群体智能系统应用程序与个体并行执行逻辑存在的巨大语义差距,使得编译过程复杂进而导致应用程序开发效率不高.本文提出了一个编译系统及其支撑工具,支持将高级的群体智能系统应用程序转换为安全、高效的分布式实现.该编译系统通过并行信息识别,计算划分,交互信息生成技术,将面向系统全局、串行编程的群体智能应用程序编译为面向个体独立执行的并行目标代码,从而使用户不必了解个体间的复杂交互机制.设计了一种标准化中间表示,将复杂群体智能计算任务转换为群体智能算子和输入输出变量组合而成的标准化语义模块序列,其以独立于平台的形式表示源程序信息,屏蔽了目标硬件平台的异构性.在一个群体智能系统案例平台中部署和测试了该编译系统,结果表明该系统能够有效将群体智能应用程序编译为平台可执行的目标代码并提升应用程序开发效率,其生成的代码在一系列基准测试中具有比现有编译器更好的性能.

Compiling swarm intelligence system application: taking fully distributed intelligent building system as an example

Swarm intelligence system realizes group-level application tasks through the information interaction of individual neighbors, and has good robustness and flexibility. At the same time, most developers struggle to describe distributed, 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. This paper 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 means of 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, so that users do not have to understand the complex interaction mechanism among individuals. 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, shielding the heterogeneity of target hardware platforms. The system is deployed and tested in a swarm intelligence system case platform. The results show that the compilation system can effectively compile swarm intelligence applications into platform-executable object code and improve the efficiency of application development, and its generated code has better performance than existing compilers in a series of benchmarks.