Human readable BSGP code
Compiler generated C code
Compiler generated C header
The main features are:
Large scale realloc is much faster than dlmalloc and standard Windows 7 malloc.
There is still a slight advantage in some non-realloc-heavy cases, but that's only proven in my artificial test.
The entire free list is managed using a static, 32-way trie, which is much easier to understand/implement than the hybrid data structure of typical modern allocators.
The main limitations are:
It's written in BSGP, though the code is fairly C-like.
It's not optimized for multi-threading (yet).
It consumes 132K+ static memory.
It only runs on 32-bit Windows. The code is written with 64-bit in mind, but the compiler can't produce 64-bit code (yet).