Translation validation for an optimizing compiler

Abstract

We describe a translation validation infrastructure for the GNU C compiler. During the compilation the infrastructure compares the intermediate form of the program before and after each compiler pass and verifies the preservation of semantics. We discuss a general framework that the optimizer can use to communicate to the validator what transformations were performed. Our implementation however does not rely on help from the optimizer and it is quite successful by using instead a few heuristics to detect the transformations that take place. The main message of this paper is that a practical translation validation infrastructure, able to check the correctness of many of the transformations performed by a realistic compiler, can be implemented with about the effort typically required to implement one compiler pass. We demonstrate this in the context of the GNU C compiler for a number of its optimizations while compiling realistic programs such as the compiler itself or the Linux kernel. We believe that the price of such an infrastructure is small considering the qualitative increase in the ability to isolate compilation errors during compiler testing and maintenance. Copyright 2000 ACM.