Compiling with Continuations and LLVM

Abstract

LLVM is an infrastructure for code generation and low-level optimizations, which has been gaining popularity as a backend for both research and industrial compilers, including many compilers for functional languages. While LLVM provides a relatively easy path to high-quality native code, its design is based on a traditional runtime model which is not well suited to alternative compilation strategies used in high-level language compilers, such as the use of heap-allocated continuation closures.This paper describes a new LLVM-based backend that supports heap-allocated continuation closures, which enables constant-time callcc and very-lightweight multithreading. The backend has been implemented in the Parallel ML compiler, which is part of the Manticore system, but the results should be useful for other compilers, such as Standard ML of New Jersey, that use heap-allocated continuation closures.

A New Backend for Standard ML of New Jersey

“…These numbers are the average of ten runs of each benchmark. Based on our previous experience with the LLVM and MLRisc backends for Manticore [13], we expected to see noticeable improvements for floating-point code (e.g., the mandelbrot and mc-ray benchmarks), which we do see. Surprisingly, we also saw some significant speed up for other benchmarks, but since the running times on these programs are so small, measurement error may be more of an issue.…”

“…Thus, by modifying LLVM to add a new calling convention, we can dictate the register usage at those places. In previous work for the Manticore system [13], we described a new calling convention for LLVM, called Jump With Arguments (JWA), that can be used to support the stackless, heap-allocated-closure runtime model used by both Manticore and SML/NJ. The JWA calling convention has the property that it uses almost all of the available hardware registers for general-purpose parameter passing.…”

“…Generating the SSA form from the CFG IR is quite straightforward. 13 During the initialization pass, we preallocate ϕ nodes for each INTERNAL fragment in a cluster. We define one ϕ node per fragment parameter plus additional nodes for those special registers that are mapped to hardware registers (e.g., alloc, limit, etc.).…”

“…Our work reported here has as its roots the development of the JWA calling convention for use in Manticore's Parallel ML (PML) compiler [13]. As with the other examples above, the PML compiler generates LLVM assembly and uses the llc tool to generate native assembly code.…”

“…Recently, we have used the PML compiler to explore performance and implementation tradeoffs between different runtime strategies for representing continuations and the call stack [14]. The implementation of heap-allocated continuations in that study was the version from our previous work [13], which lacks the more sophisticated closure optimizations implemented by the SML/NJ compiler [8,35,36]. It will be interesting to revisit the experiments using our new LLVM backend for SML/NJ.…”