Jairus' Docs

Appearance

Self Time (--time) ​

bash
asb profile --time
asb profile --time --iters 20

Per-function wall-clock self time, recursion-safe and overhead-corrected. Unlike --instr, this measures real nanoseconds — so it captures cache behavior and anything else that makes an instruction cost more than its static weight.

How it works ​

A Binaryen pass outlines each function: the body moves to <name>$tprof_inner and a timing wrapper takes the original name, so direct calls, exports, and call_indirect element segments all flow through it.

  • Self time = a frame's own duration minus its direct wrapped callees (the wrappers maintain a bookkeeping stack mirroring the real call stack).
  • Inclusive time is gated to outermost frames, so recursive functions don't multi-count.
  • Overhead (~2 clock reads + a frame per call) is measured per bench by an injected calibration routine and subtracted, split into per-own-call and per-child-call components.

Flags ​

FlagEffect
--iters <n>Iterations per bench in profile mode (default 10). More iterations → steadier self times.
--min-instrs <w>Don't wrap functions under w static instructions; their time folds into the caller's self time (default 4). Reduces wrapper overhead on trivial helpers.
--top <n> / --allAs in every profile mode.

When to trust it ​

Trust self times ≥ ~1 µs/call. Below that, the per-call overhead correction dominates the signal — use --instr, which is exact there. --time is the node host only (it uses hrtime.bigint).

text
profile: assembly/__benches__/example.ts (wall-clock self time, overhead-corrected; 10 iterations per bench)

A memory-thrashing function will take a larger share of wall-clock than of instructions next to a tight compute loop — that divergence is exactly what this mode exists to surface.

Next ​