Hadoop pipes serializing array of float in C++ mapper

[Charles Earl]

Hi,
Are there examples of serializing array of type float to from C++ pipes?
From the examples, I assume 
    float floatArrayOut[arrayBytes/sizeof(float)];
    // assignment of floatArrayOut entries...
    char dest[arrayBytes];
    memcpy ( dest, floatArrayOut, arrayBytes);
    context.emit(tag,dest);

However in the reduce stage, the sizes of the 'context.getInputValue(); ' from 
the reduce size seem to be random.
I have not looked at the uplink code, but I assume the anything from byte 
ordering to memory assumptions could be off in my model.
Any advice?
C