This repository contains sample code showcasing differences in performance occuring on most platforms tested.

Rust follows the philosophy of "zero-cost abstrations", which was introduced with C++ to describe higher-level data structures and algorithms built in the language and its standard library providing the same performance as optimal low-level code.

The sample code provided in both stable Rust and C++11 implement a typical audio DSP algorithm called digital biquad filter using Direct Form 2 IIR in a textbook way: without SIMD, parallelization other any optimization.

The C++ code is very close to the Rust one. Both generate an identical output. It can be compiled with either GCC or Clang for speed comparison with the Rust code.

In results/ you can find benchmarks of the Rust and C++ implementation compiled for with arm, armv7, aarch64 architectures, GCC and Clang compilers.

• The same algoritm working on an Array-backed buffer is usually 20% to 60% faster than with a Vector-backed buffer
• Two most recently designed CPUs (1 Intel, 1 ARM64) tested perform identically between Array and Vector buffers
• Only one CPU tested (Intel Core2 Duo) is slower by 10% working on an Array buffer versus Vector
• The performance gap occurs on 32-bit and 64-bit architectures also x86_64 and ARM processors

• Rust performance is comparable to C++ overall
• On the systems exhibiting the Vector/Array perf difference:
  • Rust Vector is slower than GCC C++ Vector
  • Rust Array is faster than GCC C++ Vector
  • Rust Vector is near-identical to Clang C++ Vector