What are you expecting when you say "handle that"?
If you are expecting the standard C behaviour - that u2 will get part of the underlying memory pattern of the float - then avr-gcc will handle it perfectly well, just like any other C compiler.
But if you are equally wrong about this sort of cast as you were with non-pointer casts, and think the compiler should do a floating-point to integer conversion, then you will be disappointed - that's not how C works, and therefore not how avr-gcc works.