SUPERSEDED

Please goto https://github.com/dbj-data/dbj-bench

And look into zallocators folder

comparing_allocation_libs

© 2020 APR by dbj@dbj.org

Quick (hopefully not questionable) comparison of few heap allocation libs

Conclusion?

System memory handling is still the most feasible to use.

Pool allocation for objects should be quicker. But. It is a narrow use case. Also. Implementing object pool array allocator is not trivial. By that I mean proper implementation, starting from new and delete overloads implemeted on the class using it. new [] and delete [] are especially tricky.

Also. System memory allocators are the best for combating the fragementation issue. A must for a long running processes.

Same stands for memory allocation in presence of multiple threads. It is just not feasible to compete with the system memory handling.

Handles instead of pointers

Yes that is superior concept. But that is a paradigm shift for most of the users. And standard libraries have no such concept in use. Thus using handles with standard API's requires a lot of to/from transformations to/from native pointers.

We need to talk about Windows

In my nanolib I have the following:

#define DBJ_NANO_CALLOC(T_,S_) \
(T_*)::HeapAlloc(::GetProcessHeap(), 0, S_ * sizeof(T_))

#define DBJ_NANO_MALLOC(T_,S_) \
(T_*)::HeapAlloc(::GetProcessHeap(), 0, S_)

#define DBJ_NANO_FREE(P_) \
::HeapFree(::GetProcessHeap(), 0, (void*)P_)

It is measured many times over. Tried and tested. On Windows there is no faster system allocation call but HeapAlloc.