An APL/370 compiler and some performance comparisons with APL interpreter and FORTRAN
An APL/370 compiler and some performance comparisons with APL interpreter and FORTRAN
复制标题
APL/370 编译器以及与 APL 解释器和 FORTRAN 的一些性能比较
DOI：
--
复制
发表时间：
1986
期刊：
影响因子：
--
通讯作者：
Wai
中科院分区：
文献类型：
--
作者：
Wai
实验APL/370电子编译器将APL的子集编译为足够大的科学和工程用途，直接成370个装配代码。编译器不需要可变声明。前端基于全球数据流量分析采用广泛的类型分析。后端取下前端产生的解析树，图和表，并生成370代码，该编码独立于解释器。编译代码的执行时间是解释器在多个单行功能上的速度的2-10倍，并且该比率随迭代程序大大增加。代码质量与通过与我们的测试案例相对应的FORTRAN程序进行了优化的FORTRAN编译器所产生的质量。这消除了计算密集型应用程序中APL的绩效惩罚。
The experimental APL/370 E-compiler compiles a subset of APL which is large enough for most scientific and engineering uses, directly into 370 assembly code. The compiler does not require variable declarations. The front-end employs extensive type-shape analysis based on global dataflow analysis. The back-end takes the parse trees, graphs and tables produced by the front-end and generates 370-code which runs independently of the interpreter. The compiled-code executes at 2-10 times the speed of the interpreter on several one-line functions, and this ratio increases significantly with iterative programs. The code quality is comparable to that produced by an optimizing FORTRAN compiler on FORTRAN programs corresponding to our test cases. This removes the performance penalty of APL in computation intensive applications.