I've looked at these crashes, and I can't really find anything wrong
with the library code.

The reason that the fstpl instruction may fault, is because there's a
NaN value on the top of the FPU stack. But, the input parameter doesn't
seem to be a value that would cause this.

All I can think of is that there may be some FPU state save/restore
trouble here, but then again, such problems have not been reported
with the amd64 port for a long time.

So, at this point, my guess is as good as yours, really.

- Frank