Compilers for newest architectures using the SSA form intermediate language

在目前的计算机体系结构中，大量使用延迟分支、具有多个功能单元的超标量处理、流水线处理和内部并行处理。在这些体系结构中，只有使用以最佳方式排列目标代码的编译器才能生成有效的目标代码。另一方面，提出了一种使用单赋值变量的静态单赋值形式，作为编译器后端的一种新的中间表示形式。由于它简化了编译器中的数据流分析和优化转换，因此引起了人们的关注。本研究针对上述问题，围绕(1)利用最新体系结构特征生成高效代码的方法和(2)基于SSA形式的优化和代码生成方法。在项目期间，我们取得了以下结果：(1)在异质分布式环境下为分布式编程语言生成高效的本机代码(文献1)，(2)基于处理器描述的超标量机器代码调度器(文6)，(3)具有高度受限处理器的PDA(个人数字助理)语言处理器(文5)，(4)向SSA形式的转换方法，(5)属性文法研究(文2，3)和系统调试(文4)，这是上述优化和代码生成的基础。

In recent computer architectures, delayed branches, superscalar processing with multiple functional units, pipeline processing and internal parallel processing are heavily used. In these architectures, an efficient object code can only be generated by using compilers which optimally arranges the object code. Therefore the role of compilers is becoming more important.On the other hand, the SSA (Static Single Assignment) form, where single assignment variables are used, is proposed as a new intermediate representation in compiler back-ends. It attracts attention because it simplifies dataflow analysis and optimizing transformation in compilers. However, optimizing transformation in SSA form and connection to later phases of compilers such as register allocation and code generation is not fully investigated.In this research, we aim at developing new methods of compilers for the above problems, centering around (1) method of generating efficient code which exploits the characteristic features of newest architectures, and (2) method of optimization and code generation based on the SSA form.During the term of the project, we got the following results :(1) Generation of efficient native code for a distributed programming language on a heterogeneous distributed environment (paper 1), (2) a code scheduler based on processor description for superscalar machines (paper 6), (3) a languase processor for PDA (Personal Digital Assistant) with a highly constrainted processor (paper 5), (4) methods of transformation into the SSA form, (5) research on attribute grammars (paper 2, 3) and systematic debugging (paper 4) which is the basis of the above optimization and code generation.