I am reading a binary file where I know the byte positions of a large vector that contains more than 10^8 data samples written on 4-byte real in IBM format (described here).
As far as I know the real types in Fortran use the more recent IEEE format of samples (1 bit for sign, 8 bits for the exponent instead of 7-bit exponent for IBM format) as described here. Therefore reading these real values assuming a real format will give me wrong values, so I want to split each real number into 4 integers each occupying a single byte and then making the transformation to get the IEEE sample data from the 4 integer bytes.
I wrote the following test program to investigate this where I declare a positive real a1=1876752.76854 and its negative equivalent a2=-a1 into byte positions 1 and 5 respectively (a1 and a2 were declared to be 4 bytes long (kind=1)).
Later I read the same byte positions from the stream file for two reals of the same kind b1 and b2, but I also read in the same byte position two integers n1 and n2 (4 bytes each) to see if I can get the IEEE real type from the integer type easily. I also read the 4 integers (1 byte each) to use them to compute IEEE values as shown below:
program test_real_types implicit none integer,parameter :: rk=kind(1.0) real(rk) :: a1,a2,b1,b2 integer(kind=1) :: bytes1(4),bytes2(4) integer :: i integer(kind=4) :: n1,n2 a1=1876752.76854 a2=-a1 open(1,file='reals.dat',access="stream",form="unformatted") write(1,pos=1) a1 write(1,pos=5) a2 close(1) open(2,file='reals.dat',access="stream",form="unformatted",status="old") !! reading byte positions: 1-4 do i=1,4 read(2,pos=i) bytes1(i) enddo read(2,pos=1) n1 !! read integer n1 of 4 bytes length in the same position read(2,pos=1) b1 !! read real (4 bytes) in the same position !! reading byte positions: 5-8 do i=5,8 read(2,pos=i) bytes2(i) enddo read(2,pos=5) n2 read(2,pos=5) b2 write(*,*) 'bytes1 = ',bytes1(1:4) write(*,*) ' n1 = ',n1 write(*,*) ' b1 = ',b1 write(*,*) 'bytes2 = ',bytes2(1:4) write(*,*) ' n2 = ',n2 write(*,*) ' b2 = ',b2 end program test_real_types After execution of the code above, I got the following:
bytes1 = -122 24 -27 73 n1 = 1239750790 b1 = 1876752.8 bytes2 = 0 0 0 0 n2 = -907732858 b2 = -1876752.8 As the results indicate, n2 is not equal to -n1 (expected), but I am surprised that bytes 5-8 are all equal to zero when I read them byte by byte while they print b2 correctly as -b1. I tried to read byte 9 but it is empty as expected.
Any help on how I am getting zero bytes for locations 5-8 while I can read correctly the real value of a2 and a non-zero integer n2?
