RunMat
GitHub

timeit — Measure function-handle execution time by repeated evaluation with MATLAB-compatible timeit behavior.

t = timeit(f) repeatedly evaluates zero-argument function handle f and returns timing statistics in seconds. It supports MATLAB-compatible optional output-count control for invoked function calls.

Syntax

t = timeit(f)
t = timeit(f, numOutputs)

Inputs

NameTypeRequiredDefaultDescription
fAnyYesZero-input function handle to benchmark.
numOutputsIntegerScalarNo1Requested output count for invoking the benchmarked handle.

Returns

NameTypeDescription
tNumericScalarMedian execution time per invocation in seconds.

Errors

IdentifierWhenMessage
RunMat:timeit:TooManyInputsMore than two input arguments are supplied.timeit: too many input arguments
RunMat:timeit:NumOutputsScalarnumOutputs is not a scalar numeric/integer value.timeit: numOutputs must be a scalar numeric value
RunMat:timeit:NumOutputsFinitenumOutputs is NaN or infinite.timeit: numOutputs must be finite

How timeit works

  • Executes f repeatedly, adjusting the inner loop count until a single batch takes at least a few milliseconds or the function is slow enough.
  • Collects multiple samples (at least seven batches) and returns the median per-invocation time, which is robust against outliers.
  • Drops the outputs produced by f; you should perform any validation that depends on those outputs inside the handle.
  • Leaves GPU residency untouched—if f launches GPU kernels, they execute on the active provider. Insert wait(gpuDevice) inside the handle if you need explicit synchronisation.

Examples

Timing a simple anonymous function

f = @() sin(rand(1000, 1));
t = timeit(f)

Comparing two implementations

A = rand(1e5, 1);
slow = @() sum(A .* A);
fast = @() sumsq(A);

slowTime = timeit(slow);
fastTime = timeit(fast)

Timing a function that returns no outputs

logMessage = @() fprintf("Iteration complete\n");
t = timeit(logMessage, 0)

Timing a multiple-output function

svdTime = timeit(@() svd(rand(256)), 3)

Measuring GPU-bound work

gfun = @() gather(sin(gpuArray.rand(4096, 1)));
tgpu = timeit(gfun)

Timing a preallocation helper

makeMatrix = @() zeros(2048, 2048);
t = timeit(makeMatrix)

Using timeit with coding agents

Open a RunMat example with live inputs, then ask the agent to explain how timeit changes the result.

Run a small timeit example, explain the result, then change one input and compare the output.

FAQ

What does timeit return?

— A scalar double containing the median runtime per invocation in seconds.

How many runs does timeit perform?

— It automatically selects a loop count so each batch lasts a few milliseconds, collecting at least seven batches.

Does timeit synchronise GPU kernels?

— No. Insert wait(gpuDevice) inside the handle when you need explicit synchronisation.

Can I time functions that require inputs?

— Yes. Capture them in the handle, for example timeit(@() myfun(A, B)).

How do I time a function with multiple outputs?

— Pass timeit(@() svd(A), 3) to mirror MATLAB’s call signature. RunMat dispatches the handle with the requested output count and discards returned values.

Why do successive runs differ slightly?

— Normal system jitter, cache effects, and GPU scheduling can change runtimes slightly; the median mitigates outliers.

Can timeit time scripts?

— Wrap the script body in a function handle so it becomes zero-argument, then call timeit on that handle.

Does timeit participate in fusion or JIT tiers?

— It simply executes the provided handle; any tiering or fusion happens inside the timed function.

What happens if the function errors?

— The error is propagated immediately and timing stops, matching MATLAB behaviour.

Is there a limit on runs?

— Yes. timeit caps the inner loop at about one million iterations to avoid runaway measurements.

pause · tic · toc

Open-source implementation

Unlike proprietary runtimes, every RunMat function is open-source. Read exactly how timeit is executed, line by line, in Rust.

About RunMat

RunMat is an open-source runtime that executes MATLAB-syntax code blazing on any GPU. It is licensed under the Apache 2.0 license.

  • RunMat automatically optimizes your math for GPU execution on Apple, Nvidia, and AMD hardware. No code changes needed. Simulations that took hours now take minutes.
  • Start running code in seconds. RunMat runs in the browser, on the desktop, or from the CLI. No license server, no IT ticket.

Getting started · Benchmarks · Pricing

Download RunMat

Download RunMat for full performance, or use RunMat in your browser for zero setup.