Benchmark Suite Expansion

[clover2123]

Benchmark Suite Expansion

Currently Walrus uses its own C micro-benchmarks (test/wasmBenchmarker/ctests) plus CoreMark, Dhrystone, and smallpt (test/programs).
To better measure interpreter/JIT performance and compare Walrus against other lightweight runtimes (wasm3, WAMR, etc.), the following widely-used WebAssembly benchmarks should be integrated.
Tasks are listed in order of increasing difficulty.

1. Remaining CLBG (Computer Language Benchmarks Game) tests

• ctests already includes nbody, fannkuch, kNucleotide, and mandelbrot
• Add the missing ones: binary-trees (allocation-heavy), spectral-norm, fasta, pidigits (bignum), reverse-complement
• Same style as existing ctests; lowest entry barrier
• https://benchmarksgame-team.pages.debian.net/benchmarksgame/

2. PolyBench/C

• ~30 linear algebra / stencil kernels, a de facto standard in Wasm performance research
• Pure computation kernels with almost no WASI dependency; can run on Walrus as-is
• Useful for measuring JIT loop optimization effects
• https://sourceforge.net/projects/polybench

3. CoreMark-PRO

• Natural extension of the existing CoreMark integration
• Adds diverse workloads: JPEG compression, SHA-256, ZIP, neural net, FFT, etc.
• Well suited for embedded-target comparisons
• https://github.com/eembc/coremark-pro

4. Embenchen (Emscripten benchmark suite)

• Real application-scale C/C++ workloads: box2d, bullet, lua_binarytrees, zlib, memops
• Used by the V8 team and multiple Wasm research papers
• Complements the current kernel-style micro-benchmarks
• Note: some workloads require WASI functions (synergy with ongoing WASI work in [WASI] Implement missing WASI preview1 functions #435)
• https://github.com/emscripten-core/emscripten/tree/main/test

5. Sightglass harness integration

• Standard benchmark suite of the Bytecode Alliance (Wasmtime/Cranelift)
• Includes large real-world modules (e.g. spidermonkey.wasm) — also useful for measuring parse/compile time, a key metric for a lightweight runtime
• Requires registering Walrus as an engine in the Sightglass runner; the integration itself is a standalone task
• https://github.com/bytecodealliance/sightglass

(Optional) Additional candidates

• libsodium benchmark — commonly used by lightweight interpreters (wasm3) for comparison
• JetStream 3 Wasm subtests (argon2, zlib-wasm, gcc-loops, richards-wasm) — extract standalone Wasm modules from the browser suite

Common follow-up work

• Integrate the new benchmarks into test/wasmBenchmarker/benchmark.py (or a unified runner) for performance measurement
• Optionally register them as correctness tests in tools/run-tests.py (verify expected outputs on both interpreter and JIT)