Domain-Specific Language MISPC And Design And Implementation Of The Compiling Framework

With the development of the society, more and more application fields are differentiated, and advanced science and technology of computers is playing an increasingly important role in these applications. Representatively, domain-specific languages are of great significance to the field of application research and development. Based on domain-specific languages, generation technology of high-performance code enhances the value of related studies. It is a hot and difficult study on problems of how to abstract features based on a specific application domain, define a related domain-specific language, and design a corresponding framework of program compilation and high-performance code generation.Firstly, according to multimeida and wireless network domains, the paper defines a new domain-specific language MISPC based on standard C language which introduces new data types such as complex number, vector and matrix, and introduces a series of new operators in accordance with domain requirements and the new data types such as computing inner and part product of vector. MISPC describes characteristics of the mainstream scene of multimedia and wireless network domains. Also, the new domain-specific language shields complexity of data types such as matrix and data operations, and architectural characteristics of base hardwares. With BNF paradigm, the mainly syntactic rules and structure of MISPC are formally described. And we then introduce functions and interfaces of new domain-specific operations associated with the new data types. Meanwhile, in order to better understand the basic semantic meaning from a functional point of MISPC language, the paper describes its semantics by using the denotational semantics methods.Then, the paper proposes a code generation framework for heterogeneous platforms so that high-performance target code coule be generated after compilation and optimization of MISPC programs. Depending on the structure composition of the compilation framework, we introduce the compiler front end, pattern analysis, program lowering, code optimization, and code generation function modules and relations between them. Especially, we illustrate the strategies of local optimization, global optimization, and target optimizaiton in code optimizaiotn module.Finally, experiment is done against two classic programs in the multimedia application domain. The experimental results show that the methods and technolygies of the paper are effective. And the experiment results indicate MIPSC and its compiling framework improve development efficiency and reliability of the domains significantly, and achieve from 1.54 to 21.52x speedup compared with single-thread optimization versions on CPU under different platforms and input sizes.