How does NumPy handle ufunc chaining optimization internally?

NumPy executes chained ufuncs sequentially and may reuse buffers but does not fully fuse operations by default.

When expressions like a + b + c are evaluated, NumPy breaks them into multiple ufunc calls. Each ufunc operates on raw memory buffers. Although temporary buffer reuse can reduce allocations, full expression fusion is not generally performed in core NumPy. This means intermediate arrays may be created, especially when memory overlap or dtype conversions are involved. Advanced optimization requires libraries like Numba or JIT compilation.