Jank now has its own custom IR

Jank has introduced a new custom intermediate representation (IR) specifically designed for its Clojure-based language, aiming to improve optimization capabilities and performance.

The developer behind Jank announced the creation of a new IR that operates at a higher level than LLVM IR, focusing on Clojure’s semantics such as vars, transients, and lazy sequences. This IR is stored as C++ data structures but can be rendered as Clojure data for debugging. Unlike previous reliance on LLVM IR, which was low-level and limited in optimization for Clojure’s dynamic features, the new IR allows for language-specific optimizations tailored to Jank’s unique semantics. The IR is SSA-based, with control flow graphs composed of basic blocks, facilitating advanced compiler optimizations. This development aims to enable Jank to compete more effectively with JVM-based solutions by improving runtime and compile-time efficiency.

Why It Matters

This development matters because it could significantly enhance Jank’s performance and optimization potential, making it a more viable alternative to JVM-based languages for Clojure enthusiasts and developers seeking a more tailored compilation approach. A high-level IR aligned with Clojure’s semantics can unlock new optimization strategies, potentially leading to faster execution and more efficient code generation.

Background

Historically, Jank has relied on LLVM IR for optimization, which is low-level and not well-suited to Clojure’s dynamic features. Prior efforts improved runtime and compiler efficiency but did little to optimize the generated code itself. The move to a high-level, language-specific IR represents a strategic shift to better leverage Clojure’s unique features, such as polymorphism and indirection.

“We’ve built a new IR that operates at the level of Clojure’s semantics, giving us more power to optimize the code specifically for Jank.”

— Jank Developer

“This IR is SSA-based and represented as control flow graphs, which will facilitate advanced compiler optimizations.”

— Jank Developer

What Remains Unclear

It is not yet clear how much performance improvement the new IR will deliver in practice, or how stable and mature the IR is expected to become. Further testing and development are ongoing, and the full impact remains to be seen.

What’s Next

The next steps include further refining the IR, integrating it into the Jank compiler pipeline, and benchmarking its performance against existing solutions. Additional optimization passes and stability improvements are expected in upcoming releases.

Key Questions

What is an intermediate representation (IR)?

An IR is a simplified, abstract version of a program used by compilers to perform optimizations and generate machine code more efficiently. It operates at a level between source code and machine instructions.

Why did Jank develop its own IR?

Jank’s previous reliance on LLVM IR limited optimization opportunities for Clojure’s dynamic features. A custom IR allows for language-specific optimizations, potentially improving performance and compiler efficiency.

How does the new IR differ from LLVM IR?

The new IR is higher-level, operating at the level of Clojure’s semantics, and is SSA-based with control flow graphs. LLVM IR is lower-level and less suited to dynamic language features.

Will this change improve runtime performance?

It is expected to, but concrete performance gains will depend on further development, testing, and optimization passes. Results are still being evaluated.

When will the new IR be fully integrated and stable?

There is no fixed timeline yet. The development is ongoing, with future updates expected as the IR matures and is tested in real-world scenarios.