From 41d0ff6c2cd731f62492f1841bb8887991df17b4 Mon Sep 17 00:00:00 2001 From: Henry Gressmann Date: Wed, 15 May 2024 13:47:18 +0200 Subject: [PATCH] chore: update benchmarks Signed-off-by: Henry Gressmann --- BENCHMARKS.md | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) diff --git a/BENCHMARKS.md b/BENCHMARKS.md index 1358f70..50df853 100644 --- a/BENCHMARKS.md +++ b/BENCHMARKS.md @@ -22,21 +22,27 @@ All runtimes are compiled with the following settings: - `opt-level` is set to 3, `lto` is set to `thin`, `codegen-units` is set to 1. - No CPU-specific optimizations are used as AVX2 can reduce performance by more than 50% on some CPUs. +- Default runtime settings are used ## Versions -- `tinywasm`: `0.6.2` +- `tinywasm`: `0.7.0` - `wasmi`: `0.31.2` -- `wasmer`: `4.2.8` +- `wasmer`: `4.3.0` ## Results +> Results include the time it takes to parse/compile the WebAssembly file and execute the function. + | Benchmark | Native | TinyWasm | Wasmi | Wasmer (Single Pass) | | ------------ | -------- | ---------- | --------- | -------------------- | -| `fib` | `0ms` | ` 18.70µs` | `18.53µs` | ` 48.09µs` | -| `fib-rec` | `0.27ms` | ` 16.02ms` | ` 4.96ms` | ` 0.47ms` | -| `argon2id` | `0.53ms` | ` 80.54ms` | `46.36ms` | ` 4.82ms` | -| `selfhosted` | `0.05ms` | ` 7.26ms` | ` 6.51ms` | `446.48ms` | +| `fib` | `6.33µs` | ` 19.18µs` | `18.26µs` | ` 51.20µs` | +| `fib-rec` | `0.27ms` | ` 16.09ms` | ` 5.08ms` | ` 0.47ms` | +| `argon2id` | `0.50ms` | ` 89.52ms` | `45.31ms` | ` 4.74ms` | +| `selfhosted` | `0.05ms` | ` 7.93ms` | ` 7.54ms` | `512.45ms` | + +> Note that parsing is still pretty slow, especially for the `selfhosted` benchmark, taking up `~6ms` for TinyWasm. +> This can be improved by using the `archive` feature, which pre-parses the WebAssembly file into tinywasm's custom bytecode format. ### Fib