This is needed for processing 32bit floats Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>