The article is doing profiling on x86-64. There is no FMOV. Floating point on x86-64 is done in the SSE or AVX (xmm, ymm) registers loaded with MOVD/Q which has basically the same addressing modes as any of the usual GPR loads/moves. In any case in optimized code, something like a scalar clone is inlined and would not be a consistent accountable and localizable operation (and may even be optimized away).

My point is, whatever it is, it can't possibly be a function call, so how would it ever legitimately show in a profile? I like pcwalton's theory.

The practical implication is that trying to optimize "calls to f32::clone" because they're in a profile is a wild goose chase -- unless you have debug anti-optimization turned on, there shouldn't be any calls to f32::clone.