Hi Christoph,

I think you have misunderstood what I meant, originally, you asked me if I knew the internal low level mechanism for floating point. I explained roughly how this takes place. I am fully aware of what you are saying, basically the mantissa's bit field is limited to 23 bits, idealy it should have at least 127 bits but we have to compromise somewhere, hey?
Although not a be end to the above problem, most compilers do offer rounding which can invoked by the programmer at will.
Could you imagine the added processing time with 127 bit mantissa?
You are quite right about compiler only 'knows how to add,etc' but I thought you meant doing it in low level assembler which would then be a repetative complex task for any asm programmer thinking of doing such things, which is why we use compilers.

Bye for now
Jason