Prevector Quick Destruct #9505

This is a minor performance optimization of prevector. If the stored type is trivially destructible (such as unsigned char which is the main use of prevector), there is no need to call clear on the entries before freeing.

The call to clear seems to not get optimized out completely (likely because of the --_size; side-effect) with -O3.

std::is_trivially_copyable seems to enjoy full cross platform support.

On the included benchmark, I see about a 3% performance improvement.

any chance you get the same gain by specializing erase() with std::enable_if and std::is_trivially_destructible, simply reducing it down to a memset in the specialized case? That way we'd still get the zero'd mem, and it'd get sped up in several other places as well.

@theuni I think that works -- may as well wrap both locations.

utACK 45a5aaf

light utACK 45a5aaf

utACK 45a5aaf