On top of that, this is not a large bitmap split into bitmaps, you are almost half right. This is a special data structure tree that I invented.
This data structure that I thought of will be used on the heap management library, so this will be used on the page allocator, heap allocator, allocations for devices, address space allocations, NUMA will benefit the most of the performance gains.
I'm also aiming to have local and atomic function variants in the heap manager, and seeminly the local (per thread or per processor (like NUMA)) ones will be more than 15 times faster.
The atomic ones will be asynchronous (without spinlock), and in the worst case there will be a spinlock but it will be too quick seemingly
It is a general purpose idea by me, just like the microsoft idea of a n object manager.
I never heard anyone using the optimization that I came out with. everyone uses malloc and keeps searching for chunks and using some sort of cache. Linux splits the pages to power of 2 and keeps searching between chunks.
My allocator will get everything on a blink of an instruction, it won't search for anything ! I thought of that way for a week and came up with the idea !