How to implement a memory allocation pro-filer that tracks and reports potential memory allocation problems? [on hold]

Memory allocation is one of the more expensive things a program can do. It is possible to make a program run much faster by optimizing how that program uses malloc, and by optimizing malloc itself.

Before optimizing a program, I think, that we must measure that program’s performance. It is super useful, therefore, to have a memory allocation "profile" that tracks and reports potential memory allocation problems.

How to design or implement a particular kind of profiling, heavy-hitter reports, for my own memory allocator? And what they should do?