Copy-and-patch compilation: a fast compilation algorithm for high-level languages and bytecode

Fast compilation is important when occurs at runtime, such as query compilers in modern database systems and WebAssembly virtual machines browsers. We present copy-and-patch, an extremely fast technique that also produces good quality code. It capable of lowering both high-level languages low-level bytecode programs to binary code, by stitching together code from a large library implementation variants. call these implementations stencils because they have holes where missing values must be inserted during generation. show how construct stencil describe the copy-and-patch algorithm generates optimized demonstrate two use cases copy-and-patch: compiler for C-like language intended metaprogramming WebAssembly. Our has negligible cost: it AST less time than takes AST. implemented SQL on top this system TPC-H benchmarks, orders magnitude faster LLVM -O0 three higher optimization levels. The generated runs order interpretation 14% -O0. 4.9X-6.5X Liftoff, baseline Google Chrome. outperforms Liftoff's 39%-63% Coremark PolyBenchC benchmarks.