Are you sure about bit-exact floats? Can you show where Booth's multiplication is done?