Not every architecture has an FPU, and doing floating point in software is obviously pretty slow.