Zero-overhead Python Benchmarking

zerobench is a Python benchmarking library with zero overhead, designed for multidimensional performance analysis.

Features

• Context manager API: Benchmark any code block with with bench(...): ...
• Multidimensional: Tag benchmarks with arbitrary keyword arguments
• Zero overhead: Code is passed directly to timeit.Timer, no wrapper function
• Auto-scaling: Automatically determines the number of iterations for reliable measurements
• Multiple exports: CSV, Parquet, Markdown
• Plotting: Built-in visualization with matplotlib

Quick Example

from zerobench import Benchmark

bench = Benchmark()

for n in [100, 1000, 10000]:
    data = list(range(n))
    with bench(method='sum', n=n):
        sum(data)
    with bench(method='len', n=n):
        len(data)

Output:

method=sum, n=100: 0.579 us ± 2.38 ns (median ± std. dev. of 7 runs, 500000 loops each)
method=len, n=100: 0.020 us ± 0.45 ns (median ± std. dev. of 7 runs, 20000000 loops each)
method=sum, n=1000: 5.369 us ± 44.70 ns (median ± std. dev. of 7 runs, 50000 loops each)
method=len, n=1000: 0.029 us ± 0.09 ns (median ± std. dev. of 7 runs, 10000000 loops each)
method=sum, n=10000: 53.728 us ± 69.86 ns (median ± std. dev. of 7 runs, 5000 loops each)
method=len, n=10000: 0.029 us ± 0.25 ns (median ± std. dev. of 7 runs, 10000000 loops each)

print(bench)

┌────────┬────────┬─────────────────────────────────┐
│ method ┆ n      ┆ execution_times                 │
╞════════╪════════╪═════════════════════════════════╡
│ sum    ┆ 100    ┆ [0.577805, 0.57815, … 0.581231… │
│ len    ┆ 100    ┆ [0.019207, 0.019278, … 0.01958… │
│ sum    ┆ 1_000  ┆ [5.417795, 5.33863, … 5.35146]  │
│ len    ┆ 1_000  ┆ [0.028898, 0.030144, … 0.03007… │
│ sum    ┆ 10_000 ┆ [53.743199, 53.664567, … 53.72… │
│ len    ┆ 10_000 ┆ [0.028857, 0.028911, … 0.02942… │
└────────┴────────┴─────────────────────────────────┘

JAX Support

ZeroBench automatically detects JAX arrays and optimizes benchmarking accordingly:

import jax.numpy as jnp
from zerobench import Benchmark

bench = Benchmark()
x = jnp.ones(1000)
y = jnp.ones(1000)

with bench(method='add'):
    x + y

When JAX code is detected, zerobench:

1. Wraps the code in a JIT-compiled function to measure optimized execution
2. Separates compilation from execution by reporting compilation_time separately
3. Captures the StableHLO representation of the compiled function in the hlo field
4. Uses jax.block_until_ready to ensure accurate timing of asynchronous operations

The benchmark report includes additional fields for JAX:

• first_execution_time: Time of the initial (possibly uncompiled) execution
• compilation_time: Time to lower and compile the function
• hlo: The StableHLO text representation of the compiled computation

report = bench.to_dicts()[0]
print(report['compilation_time'])  # e.g., 12345.67 ns
print(report['hlo'][:100])         # HLO module "jit___bench_func" ...

Installation

pip install zerobench

Export and Visualization

# Export results
bench.write_csv('results.csv')
bench.write_parquet('results.parquet')
bench.write_markdown('results.md')

# Plot results
bench.plot()
bench.write_plot('results.pdf')

Configuration

Benchmark(
    repeat=7,                    # Number of measurement repetitions
    min_duration_of_repeat=0.2,  # Minimum duration per repeat (seconds)
    time_units='ns',             # Time units: 'ns', 'us', 'ms', 's'
)

License

MIT