Is there any reason to think that this algorithm will actually be faster than what is currently done in practice?
For arrays in B-tree nodes, which is the main place where I have encountered this problem, I doubt it will be faster than just using `memmove()`, and for truly large arrays, it would be easier to use a B tree.
If that is the case, this joins a number of algorithms that are asymptotically faster than algorithms in use and paradoxically slower than them. Examples of such algorithms include all of the fast matrix multiplication algorithms, which are slower than good implementations of the the textbook O(n^3) algorithm (GEMM).
For arrays in B-tree nodes, which is the main place where I have encountered this problem, I doubt it will be faster than just using `memmove()`, and for truly large arrays, it would be easier to use a B tree.
If that is the case, this joins a number of algorithms that are asymptotically faster than algorithms in use and paradoxically slower than them. Examples of such algorithms include all of the fast matrix multiplication algorithms, which are slower than good implementations of the the textbook O(n^3) algorithm (GEMM).