Roc Functional Language Achieves Near C-Level Performance in DEFLATE Compression

Roc's Performance Against C

A DEFLATE compression library, `roc-deflate`, written in the functional programming language Roc, was benchmarked against `libdeflate`, a highly optimized C library. The results showed `roc-deflate` running slower than `libdeflate` by a median of 8.9% on an AMD Ryzen 9 3900X and 2.6% on an M1 MacBook Pro.

This benchmark focused on non-streaming APIs using the Silesia corpus for testing, with verification that identical outputs were produced.

Context and Previous Benchmarks

This effort follows a previous benchmark where Lean 4, another functional language, demonstrated a DEFLATE compression library (`lean-zip`) that outperformed Rust's `miniz_oxide` by being 15% quicker and producing 8.8% smaller output at its fastest setting. However, `lean-zip` did not match the performance of `libdeflate` due to its architecture-specific SIMD optimizations.

Roc, while also functional, is not a theorem prover and explicitly prioritizes runtime performance, with its tagline being "Fast, Friendly, Functional."

Roc's Development Status

Roc is a relatively new language, with its first numbered release (0.1.0) still pending. Its package ecosystem is nascent and has not yet undergone significant optimization work. Despite this, its language design, existing optimizations, and new compiler implementation provided a foundation for this performance comparison.

Implications for Functional Programming

The results indicate that functional programming languages, particularly those designed with performance in mind like Roc, can achieve performance levels very close to highly optimized C implementations for specific computational tasks. This challenges the perception that functional languages inherently incur significant runtime overhead compared to lower-level languages.