quantize_data_code.f90

 
  integer, intent(in) :: nbits
  real(4), intent(out) :: compress_err
  real(4) dataMin, dataMax, scale_fact, offset
  ! if nbits not between 1 and 31, don't do anything
  if (nbits <= 0 .or. nbits > 31) then
     dataout = datain
     compress_err = 0.0
     return
  endif
  datamax = maxval(datain); datamin = minval(datain)
  ! convert data to 32 bit integers in range 0 to 2**nbits-1, then cast
  ! cast back to 32 bit floats (data is then quantized in steps
  ! proportional to 2**nbits so last 32-nbits in floating
  ! point representation should be zero for efficient zlib compression).
  scale_fact = (datamax - datamin) / (2**nbits-1); offset = datamin
  dataout = scale_fact*(nint((datain - offset) / scale_fact)) + offset
  compress_err = maxval(abs(datain-dataout))