FORTRAN 77 on PDP-11

Jerome> I have been having some fun with Prime Numbers using FORTRAN Jerome> 77 under RT-11 on a PDP-11. Since I rapidly run out of the Jerome> capacity of 32 bit numbers (INTEGER * 4 NUMBER), I have Jerome> started to use 64 bit values (REAL * 8 NUMBER) but the Jerome> conversion routines leave a lot to be desired on output. Jerome> When using F32.0, I can count on only about 15 digits of Jerome> accuracy even though the values are often accurate to an Jerome> exact integer. Of course. Just because it's exactly representable as an integer doesn't mean it's exactly representable as a 64 bit float. Those have a 56 bit mantissa, which means you have 2^56 bits of resolution, which translates to roughly 16 digits. (The VAX architecture handbook mentions that explicitly, by the way.) Jerome> If needed, I can write my own conversion routines, but I Jerome> though I might inquire if anyone knows of any libraries which Jerome> are exact or better still, multi-precision libraries which Jerome> can handle up to 128 bit integers? If you want exact arithmetic, run, don't walk, away from floating point numbers. Use only integers. There the GNU MP library, and others like it. Long integer arithmetic libraries are commonly found in crypto library packages. Come to think of it, you'll find functions that are helpful in stuff relating to prime numbers -- since prime numbers also figure prominently in cryptography. So you could use MP (which is in C) directly, given a suitable compiler. Or you could port it to Fortran... paul