Current C1 LIR_Assembler::type_profile_helper performs two linear scans (of TypeProfileWidth width) looking for MDO slot to update the type profile. This takes considerable space in generated code, especially with larger TypeProfileWidth-s.

We can optimize the type profile helper for code density. We do not care *all too much* about minute differences in performance of this check, at it runs only for profiling in tier3 compiled code.

Draft PR: https://github.com/openjdk/jdk/pull/24491