I am looking for an extended precision floating point library with the following features:
It seems that ttmath is very close to what I look for. It allows to specify big floats with exponent and mantissa value and it does not require initialization.