This document discusses using Hyperscan to match regular expressions in Rspamd. It describes how Hyperscan can match many regular expressions against the same data faster than joining expressions with pipes in PCRE. However, Hyperscan has slower compile times and does not support some PCRE features. The document proposes a "lazy hyperscan compile" approach where Rspamd initially uses PCRE alone and later switches to the pre-compiled Hyperscan database to get the speed benefits without the initial compile overhead. It shows how this approach improves Rspamd performance for both small and large messages.