Abstract

Abstract We describe an automated way to generate data for a practical LLVM instruction selector based on machine-generated description of the target architecture at register transfer level. The generated instruction selector can handle arbitrarily complex machine instructions with no internal control flow, and can automatically find and take advantage of arithmetic properties of an instructions, specialized pseudo-registers and special cases of immediate operands. Introduction Application-specific processors are often used in embedded applications with large production quantities due to the speed and low power consumption they can provide at modest cost compared to using a generic CPU with higher execution speed. The Lissom project In the compiler front-end it is only necessary to provide information about the application binary interface (ABI) of the target platform; front-ends are already prepared for this, so we only need to describe data types in a predetermined format. The LLVM optimization passes work purely on the internal representation without considering the target architecture. Most of the work therefore involves the compiler back-end, which is described in this article, focusing primarily on the instruction selection component. Related Work A full-featured compiler generator is described in