libdl uses a linear symbol search on object file symbols

[Chris Johns]

Symbol searching has two parts, searching the object file and searching the global symbol table. Currently the object file search is linear and the global table search uses a hash table.

A large incrementally linked object file can have a large local and global set of symbols and this can slow the loading process. This issue does not show up for small object files with a few symbols which is typically how our libraries are made.

Change the object file symbol search to a binary search (bsearch). A hash table for each object file would increase the in memory object file footprint by a significant amount and would harm the small object file use case that only have a few symbols. A binary search is a suitable compromise.

[Chris Johns <chrisj@…>]

In 327e45da/rtems:

libdl: Sort object file symbols and use a binary search to find

• Replace the linear object file symbol search with a binary search.
• Sort the object file symbols after loading.

Closes #3748