I am currently writing a Maya 2013 plugin in c++ that optimizes the geometry of given mesh for specific constraints. i need to achieve maximum performance. for now i only implemented an hard-coded version of the algorithm for a specific constraint and it's very fast, but i need achieve some king of generic implementation of the algorithm for any given constraint (thus i need some kind of polymorphism).
a pseudo code for the general structure of the algorithm is:
(1) do k times:
(2) for every face fj in the mesh do:
(3) some manipulation on the vertices qi incident to the face fj
(4) for every vertex vi in the mesh do:
(5) some manipulation on the faces fj incident to the vertex vi
the problem is that in my specific implementation there is no function calls for the calculations in steps 3 and 5 in the pseudo code, when i tried calculating those steps with auxiliary functions there was a big reduction on performance.
for a generic implementation i need to make function calls in steps 3 and 5 for every constraint.
i thought of two kinds of solutions:
- write a generic class for constraints. derive classes for every constraint i need with methods for calculaing steps 3 and 5 in the algorithm.
- write a general functor (class with only the () operator) and derive from it functors for calculating steps 3 and 5. the reason for this idea is that the call for the () operator in a functor is inlined (answer about reusable code in this link).
first question:
is there any way to reduce the overhead for the function calls?
second question:
is there any way to make the compiler always inline the functions in the two solutions above? from this question i know that virtual functions called from pointers to objects can't be inlined.
third question:
is there a more efficient way then the solutions i proposed?
