Cross-file interprocedural optimization method based on algebraic system

Abstract

The invention discloses a cross-file interprocedural optimization method based on an algebraic system. The method comprises steps as follows: according to characteristics of a target machine, instructions relating to stack operation and logical operation are selected, the algebraic system is established, and a mapping relation is established for the instructions and the algebraic system; PCGs (procedure call graphs) are traversed from a program entry, and whether nodes with connected sides belong to different source files is judged; if yes, the next operation is continued, and otherwise, the PCGs are traversed continuously; DDGs (data dependence graphs) are traversed from a function call instruction reversely along CFGs (control flow graphs) in a current function, an algebra expression formula of a stack-push operation instruction is generated, and the expression formula is reduced; a stack-pull operation of a successor node function is analyzed, constant values are read, sequentially transferred, optimized and calculated, and finally, a redundant instruction segment is deleted. According to the method, the optimizable part in a functional stack framework is effectively combined and released; and besides, the better effect is obtained during cross-file interprocedural optimization, constant propagation and constant calculation.

Description

technical field [0001] The invention relates to compiler technology, in particular to an algebraic system-based inter-process optimization method across files. Background technique [0002] The function stack framework is a common protocol in the traditional compilation framework, and is defined by the user in the binary interface (Application Binary Interface, ABI). When the function stack frame processes function calls, it saves temporary variables in temporary registers as much as possible according to the hardware resource constraints of the target machine, and the overflow part is pushed into the stack before the function body is executed, and accessed through storage The instruction performs stack operation, and releases the data in the function stack one by one after the function body is executed. [0003] In this mode, since the maximum number of registers supported by the target machine is limited, and local temporary variables are distributed in multiple basic blo...

AI Technical Summary

Problems solved by technology

[0003] In this mode, since the maximum number of registers supported by the target machine is limited, and local temporary variables are distributed in multiple basic blocks due to the life cycle, frequent stacking, popping and local temporary register multiplexing will directly cause the function stack to Operation is redundant

[0004] The existing technology is mainly based on the optimization of the intermediate language under the compilation framework, and it mainly faces the following main problems: 1. The formulation of the function stack framework protocol is directly related to the target machine, and the amount of code involved in the modification is very large, which will lead to The workload has increased rapidly; 2. It is very complicated to formulate a function stack frame suitable for various application scenarios, which involves multiple stages such as register allocation, and the intermediate language is only an abstract language, which cannot directly and effectively describe the stack operation, so the design complexity of optimizing from the intermediate language is high

Images