CMEX S-function vs. Model blocks: delay in execution?

I am running a Simulink model and an S-Function of that model in parallel. The S-function was generated for real-time using RTW embedded coder. The outputs are almost identical except for a one simulation step delay. My hunch is that Simulink can't use zero-crossing detection in the S-function. Hence, it's causing a 1-step delay. Am I on the right track? Any other ideas? Thanks a lot!

[Kaustubha Govind]

Does your model have any algebraic loops? You can detect these by setting the Algebraic Loop Diagnostic to warning/error: http://www.mathworks.com/help/toolbox/simulink/ug/f7-8243.html#bsn439w

I'm not sure, but if the S-function somehow tries to solve the algebraic loops, it is possible that a delay is introduced.