mypyc benchmarks

This is a collection of mypyc benchmarks. They are intended to track performance of mypyc against interpreted CPython. They are also useful for validating that a mypyc enhancement results in a measurable performance improvement.

Some benchmarks are microbenchmarks that are only useful for finding big performance differences related to specific operations or language features. They don't reflect real-world performance.

Benchmark results

We have a service that automatically collects benchmark results for all mypyc and mypy commits:

• Benchmark results
• Microbenchmark results

Running benchmarks

Prerequisites:

• Python 3.7 or later on Linux, macOS, or Windows
• mypyc in PATH
• A working Python C development environment

Display the names of available benchmarks using runbench.py --list:

$ python3 runbench.py --list
binary_trees
bytes_call (micro)
bytes_concat (micro)
bytes_format (micro)
bytes_indexing (micro)
...

Microbenchmarks are distinguished by (micro).

Run a benchmark using runbench.py --mypy-repo <dir> <name>:

$ python3 runbench.py --mypy-repo ~/src/mypy richards
...
running richards
......
interpreted: 0.190326s (avg of 6 iterations; stdev 1%)
compiled:    0.019284s (avg of 6 iterations; stdev 1.6%)

compiled is 9.870x faster

This runs the benchmark in both compiled and interpreted modes using the mypyc from the given mypy repository, and reports the relative performance.

Use runbench.py -c ... to only run the compiled benchmark.

Run a benchmark first using the mypy master branch and then your local branch to see how well your branch does relative to master.

Documentation

There is more information in the documentation.