in general floating point operations take longer to carry out than integer in particular where processors do not have floating point coprocessors and use software to do the caluclations (also in such cases the code can be very large)
in general if reading integer data, e.g. from an ADC, keep it as integer data as long as possible.
if there is danger of exceeding the numeric limits of integer values there are alternatives to consider before converting to floating point
1. use unsigned integer types if there are no negative values
2. use long integer types
3. normalise the integer values so that numeric limits will not be exceeded in calculations
4. use processors with hardware suited to specific operations, e.g. DSP processors with built in support for FFTs, etc
if one has to use floating point and run out of program memory move to a processor with more memory
if floating point operations are too slow move to processor with a floating point coprocessor
also remember if one wants to multiply an integer by 2 one can shift left
or divide by 2 shift right
in the latter case be careful - some implementations may use logical shift rather than arithmetic shift so the sign bit becomes 0 - a problem when data may be negative