To count the number of page faults during execution, use:
perf stat <executable>
Page faults occur after anonymous pages are mmap'ed for the heap, either mapping a common zero page, then faulted again for a memory write. Prefaulting the page with MAP_POPULATE flag to mmap can help reduce the number of page faults.
Shared libraries are also mmap'ed and faulted in, and doing it this way saves memory for things that aren't used. But if paying the penalty for faulting the pages when used outweighs the memory savings, it might be better to use MAP_POPULATE here too. It might worth trying to add an LD_LIBRARY_XXX option to tell the loader to use MAP_POPULATE. Statically linking the executable will also reduce the number of faults (sections are combined, etc.)
Shared libraries are also mmap'ed and faulted in, and doing it this way saves memory for things that aren't used. But if paying the penalty for faulting the pages when used outweighs the memory savings, it might be better to use MAP_POPULATE here too. It might worth trying to add an LD_LIBRARY_XXX option to tell the loader to use MAP_POPULATE. Statically linking the executable will also reduce the number of faults (sections are combined, etc.)