As I expressed in IRC, in this series I was attempting to provide as accurate a chain of ownership for the patches as reasonably possible, as two people have worked on this before me. This is clearly less important that a clean patch series. I am working now to provide something to that end.
For what it's worth, I can't claim to have a good understanding of this procedure, but I get the impression that authorship isn't usually specified with quite this level of granularity, or if it is, it's sufficient to describe co-authorship in the patch subject, e.g. "with fixes by X" or "based on a patch by X".
Anyway, with respect to floating point errors, I'm inclined to say we do care about accuracy as best we can. If there are demonstrable errors compared to native, I think we should try to determine why they're occurring, and if they can be fixed.
One thing that may help, as pointed out to me by Henri, is to perform these calculations in fixed-point instead of floating-point.