PHP Generics Benchmark

This repository contains a benchmark for testing the performance of generics in PHP. The benchmark compares the performance of different implementations of generics in PHP.

Running the Benchmark

php src/benchmark.php --scenario=main --iterations=30 \
  --holly-binary=/path/to/holly/php \
  --phpdoc-binary=$(which php) \
  --baseline=phpdoc

At least one of the --phpdoc-binary, --holly-binary, --reified-binary, or --elissa-binary options must be provided. The benchmark will run the specified implementations and compare their performance.

Available scenarios are:

• main: The main benchmark scenario that tests the performance of generics in PHP.
• specialization: A scenario that tests the performance of generics with specialization.

Elissa

Elissa is not PHP: it is a separate language with its own runtime, included here because its generics are reified in the same way the reified implementation proposes. It is benchmarked alongside the PHP implementations so the two can be read against each other.

php src/benchmark.php --scenario=main --iterations=30 \
  --phpdoc-binary=$(which php) \
  --elissa-binary=/path/to/elissa/target/release/elissac \
  --baseline=phpdoc

Because it is a different runtime, a few things differ and the harness accounts for them:

• It needs --elissa-binary; the blanket --binary only covers the PHP implementations, and elissa is skipped without a binary of its own.
• It only runs in plain mode. The jit mode passes opcache ini flags, which mean nothing to it.
• It reports memory as unknown, since the runtime exposes no peak-memory reading.
• It has no specialization scenario, and is skipped for it ( classes have a limit of 64 methods ).