Profiling

asb run tells you which bench is slow. asb profile tells you where the cost is — per function, inside one bench. It builds a debug + instrumented wasm, runs each bench, and renders a ranked table.

asb profile                  # --instr (default)
asb profile --time           # wall-clock self time
asb profile --alloc          # bytes allocated  (alias: --heap)

Three lenses

Mode	Flag	Ranks by	Exact?
Instructions	--instr (default)	cost-weighted wasm instructions	yes, deterministic
Self time	--time	overhead-corrected wall-clock self time	measured
Allocations	--alloc / --heap	bytes claimed from the allocator	yes, deterministic

Reach for --instr when you want a stable, machine-independent ranking; --time when cache behavior matters and you want real nanoseconds; --alloc when you're chasing allocation pressure.

The legacy --heaviest=instr|time|alloc forms are still accepted.

Shared flags

Flag	Effect
--top <n>	Rows per bench (default 10).
--all	Include engine/runtime-internal rows (hidden by default).
--iters <n>	(--time / --alloc) iterations per bench.
--min-instrs <w>	(--time) don't wrap functions under w static instructions; their time folds into callers (default 4).

Reading the table

By default each bench shows the rows that account for ≥ 1% of total cost, capped at --top, followed by a coverage line:

sort 1k i32/quick      274,696 weighted · 222,593 instructions
   49.5%   135,997 wt   116,095 instrs    3,317 calls   41 wt/call  StaticArray#__get
   31.8%    87,283 wt    63,349 instrs      361 calls  241 wt/call  quickSort
   16.3%    44,720 wt    38,480 instrs    1,040 calls   43 wt/call  StaticArray#__set
  top 3 rows account for 97.6% of total cost
  (+2 hidden rows, use --all to show)

Tiny profiles collapse to the few rows that matter; large ones shed their sub-1% tail. Pass --all to see everything, including runtime-internal functions.

Pages

• Instruction Counts — --instr
• Self Time — --time
• Allocations — --alloc / --heap