Book Image

LLVM Essentials

By : Mayur Pandey, Suyog Sarda, David Farago
Book Image

LLVM Essentials

By: Mayur Pandey, Suyog Sarda, David Farago

Overview of this book

LLVM is currently the point of interest for many firms, and has a very active open source community. It provides us with a compiler infrastructure that can be used to write a compiler for a language. It provides us with a set of reusable libraries that can be used to optimize code, and a target-independent code generator to generate code for different backends. It also provides us with a lot of other utility tools that can be easily integrated into compiler projects. This book details how you can use the LLVM compiler infrastructure libraries effectively, and will enable you to design your own custom compiler with LLVM in a snap. We start with the basics, where you’ll get to know all about LLVM. We then cover how you can use LLVM library calls to emit intermediate representation (IR) of simple and complex high-level language paradigms. Moving on, we show you how to implement optimizations at different levels, write an optimization pass, generate code that is independent of a target, and then map the code generated to a backend. The book also walks you through CLANG, IR to IR transformations, advanced IR block transformations, and target machines. By the end of this book, you’ll be able to easily utilize the LLVM libraries in your own projects.
Table of Contents (14 chapters)
LLVM Essentials
About the Authors
About the Reviewer

Chapter 4. Basic IR Transformations

Until now, we have seen how the IR is independent of its target and how it can be used to generate code for a specific backend. To generate efficient code for the backend, we optimize the IR generated by the frontend by running a series of analysis and transformation passes using the LLVM pass manager. We must note that most of the optimizations that happen in a compiler take place on the IR, one of the reasons being that the IR is retargetable and the same set of optimizations would be valid for a number of targets. It reduces the effort of writing the same optimization for every target. There are some target-specific optimizations too; they happen at the selection DAG level, which we will see later. Another reason for IR being the target of optimization is that LLVM IR is in SSA form, which means every variable is assigned only once and every new assignment to a variable is a new variable itself. One very visible benefit of this representation is that...