Alright, well it does not look like you compiled it with -g
which should give more precise line references from .S to .cpp . Anyhow, I doubt we are going to spot the error from the assembly. Back when I suggested you to post it I thought it was a matter of performance and checking whether neon was generated or not, but apparently I misunderstood.
Sure start out with finding out if the coefficients are computed correctly.
If the trouble is only in the coefficient computation (which is likely to require more accurate maths), but (perhaps?) less CPU-time critical(e.g.: hopefully you are not calling it every sample), you may want to try to factor it out to a separate file and build that one with different (less math-extreme) compile options and see if that helps producing more accurate results.