That very much depends on the format being decompressed. Even assuming that the format requires non integer calculation, integer-based fixed point arithmetic tends to be used more as it's almost always faster on older processors than floating point, even with an FPU present.
