topic Re: Precision in Intel® Fortran Compiler

Precision

jacques_fontignie — Tue, 31 Jan 2006 21:41:41 GMT

I want to write a double precision number to a file and to read it again later.

Precision is very important and I want to get EXACTLY the same value when reading. I use format(...g) but it seems that sometimes the value is not the same. What format option have I to use?

I use formatted files.

Thanks in advance.

Message Edited by jacques.fontignie on 01-31-2006 05:42 AM

Re: Precision

TimP — Tue, 31 Jan 2006 22:29:50 GMT

g25.17e3 format is sufficient for maximum precision conversion of IEEE754 double precision. If you don't specify the widths in the format, you leave it up to the implementor of the run-time library. The IEEE standard gives range limits within which the READ of the value converted by WRITE can return the value exactly. Outside that range, you could avoid changes in the least significant bit only by the use of unformatted file.

Re: Precision

jacques_fontignie — Tue, 31 Jan 2006 22:36:02 GMT

Thanks a lot for the answer.

Re: Precision

joseph-krahn — Tue, 07 Feb 2006 03:15:57 GMT

But, beware that some variations you see are because the binary form is not exactly the same as the original formatted number, because decimal representations do not map exactly to binary (base 2) values, even for a simple number like 0.1.

Re: Precision

Dayan_S_Intel — Tue, 07 Feb 2006 05:37:04 GMT

test, please disregard