Io — Performance

Io VM benchmark results tracked over time. Each point is one run on one commit of master: interpreter operation rates from the classic speed.io microbenchmarks, wall-clock times for shootout programs, and Vector SIMD throughput. CI records a point on every VM change; hover any point for the commit.

Machines

Interpreter ops

millions of operations per second — higher is better

Programs

wall-clock seconds, best of 3 (includes ~30 ms VM startup) — lower is better

Vector SIMD

GFLOPS on 1024-element float vector multiply — higher is better

Microbenchmark rates are measured inside the VM (see bench/micro.io); program times are measured around a wasmtime invocation by bench/run.mjs. Different machines are separate series — compare trends within a machine, not values across machines. CI runner hardware varies between runs, so expect more noise in the CI series than in pinned-hardware ones.