Repeated method lookup via the vtab can be redundant. Consider the following code
cdef class Root:
cdef long i
def __cinit__(self):
self.i = 0
cdef void inc(self):
self.i += 1
cdef class A(Root):
def inc_many(self, long N):
cdef long i = 0
for i from 0 <= i < N:
self.inc()Running A().inc_many(10**9) takes 3.51 seconds on my machine. Assigning the self.vtab to a local pointer, and using that for all the calls to inc cuts down the runtime to 3.39 seconds, and assigning the vtab to a local struct (rather than just a pointer, though this has higher overhead) cuts it down to 3.04 seconds. Given that types of objects are immutable, this might be worth doing in some cases.