compiler/benchmark/array-count

Known-array count benchmark

Each iteration updates one of 100 array keys and adds count($items) to a checksum. Mutation prevents measuring only a loop-invariant count. This is an isolated array-update/count workload, not an application-wide benchmark.

From this directory with PHP 8.5 and a matching PHPX/embed installation:

php run.php 30000000
php ../../bin/tpc.php project.yml --no-progress -o array_count_benchmark
./array_count_benchmark 30000000

Expected checksum: 2999995050. Compile baseline and candidate into separate binary paths and build directories, keeping PHPX and compiler flags fixed. Alternate execution order, discard a warm-up pair, and compare medians.

Sample result

Linux ARM64 in Docker, PHP 8.5.10 ZTS, PHPX 6a68f38, GCC -O2, no LTO; TypePHP baseline 692841a6 versus direct known-array count lowering. 30 million iterations, nine measured pairs after one discarded warm-up:

Build	Median	Range
Baseline	319.63 ms	301.22–345.40 ms
Direct count	192.73 ms	185.74–196.94 ms

Elapsed time decreased 39.70%; all nine pairs were faster and checksums matched. Raw samples in seconds are in results-arm64.json. Measurements include process startup and exclude compilation, on a shared development machine. They do not imply the same application-wide improvement or a speedup against native PHP.