Alternate implementation
pitrou opened this issue · comments
Antoine Pitrou commented
Not a bug, but for the record, this (branchy) implementation is faster on our benchmarks for random masks:
https://github.com/apache/arrow/blob/493d2c6c998779a15371916e9d673b9f5968062e/cpp/src/parquet/level_conversion_inc.h#L39-L259