u == 1.4D0produce the expected result?
GFortran, G95, Open Watcom, and Silverfrost are free Fortran compilers, while Absoft, IBM, Intel, Lahey, NAG, Pathscale, PGI, and Oracle produce commercial Fortran compilers. Polyhedron Software provides compiler comparisons at http://www.polyhedron.com/compare0html. See Compilers for more information.
See the Books page.
Standard capitalization is now the preferred way to write Fortran for several reasons, most notably because that is how recent versions of the standard write it. Another reason is due to an effort to standardize the capitalization of the names of programming languages. To quote Walt Brainerd (originally from the 1997-01-03 version of Keith Bierman’s Fortran FAQ):
The rule: if you say the letters, it is all caps (APL); if you pronounce it as a word, it is not (Cobol, Fortran, Ada).
Some choose to write FORTRAN when referring to older versions of the language (prior to Fortran 90) to distinguish them from newer versions.
To build a static library
libfoo.a containing all modules and procedures in the
.f90 files in the current directory on Linux:
% gfortran -c *.f90 % ar cr libfoo.a *.o
The first command builds the object files and the second archives the object files into a static archive.
To build a shared library
% gfortran -shared *.f90 -o libfoo.so -fPIC
In both cases, other compiler flags such as
-O2 can be used.
u == 1.4D0produce the expected result?
This has to do with the representation of real values as (binary) floating point values.
A common Fortran 95 idiom for reading lines until the end of file is
integer :: stat character(len=100) :: buf open(15, file='foo.txt') do read(fh, iostat=stat) buf if (stat /= 0) exit ! process buf end do close(15)
if (stat == iostat_end) exit
There isn’t a portable way to do this either in Fortran or C, in short, because the terminal controls when the input is sent to your program and by default it is buffered. You must request that the terminal send each key and the method for doing so is platform-dependent.
Clive Page’s Fortran Resources has a section on “Reading single keystrokes from Fortran” which provides a couple of short C functions (sys_keyin.c) which can be called from Fortran to achieve the desired behavior on most Unix systems.
John Urban also provides a getkey function, written in C and callable from Fortran.
See Real precision.
There is no intrinsic procedure for converting character strings to numerical values, or vice-versa. However, this can be accomplished using internal file IO. To obtain a string representation of a numeric variable, one can perform a formatted write to a string, just as one does to a file. Similarly, a formatted read from a string can extract a numeric value. See the strnum program for an example.
open(7, form='unformatted', status='scratch')
This will create a temporary file that only lives until it is closed. It doesn’t need a filename as it will not be permanently saved to disk (although it could be stored somewhere as a temporary file). In this example, the unit number is 7.
The file will be deleted when the program terminates (but may not be deleted if the program terminates abnormally, i.e. crashes). Note that it is not permitted to prevent deletion of a scratch file by closing the file using a close statement with
Yes, in Fortran 2003. Declare the variable and allocate with a given length as follows:
character(LEN=:), allocatable :: str integer :: n n = 27 allocate(character(LEN=n) :: str)
Although there are no official file extensions for Fortran code, there are two widely established conventions. Some use
.f for fixed-form source and
.f90 for Free form layout. The latter is a reference to the Fortran 90 standard, when Free form layout was introduced. The code contained in the file can be Fortran 95, Fortran 2003, etc. Others prefer to use file extensions that indicate the standard under which the code was written. For example,
.f03 for Fortran 2003 code and
.f08 for Fortran 2008 code. Unfortunately, this results in a proliferation of file extensions and some compilers may not support the newer extensions yet.
See File extensions for more discussion on these issues.
Intuitively, one might assume that if one wants to initialize a small array by rows that something like the following will work:
! DOES NOT WORK integer :: xx(3,5)= [ 1, 2, 3, 4, 5], & [10,20,30,40,50], & [11,22,33,44,55]
! DOES NOT WORK integer :: xx(3,5)= [ [ 1, 2, 3, 4, 5], & [10,20,30,40,50], & [11,22,33,44,55] ]
Someday something simpler might work, but currently the following syntax is required to specify the values in an intuitive row-column sequence using an array constructor:
integer,save :: xx(3,5)= reshape([& 1, 2, 3, 4, 5, & 10,20,30,40,50, & 11,22,33,44,55 & ],shape(xx),order[2,1])
This is because an array constructor can be used to create and assign values only to rank-one arrays. To define arrays of more than one dimension with an array constructor, you must use the RESHAPE(3f) intrinsic function.
Note that the ORDER= option on RESHAPE(3f) is used to allow the values to be specified in row-column order instead of the default behavior, which fills columns first.
Also note that if the expressions are of type character, Fortran 95/90 requires each expression to have the same character length (there is a common compiler extension that extends all strings to the length of the longest value specified, but depending on it reduces portability).
When working with small arrays the issue that there is no default Fortran routine for printing an array in row-column order becomes apparent. So lets create a simple solution for integer arrays (PRINT_MATRIX_INT(3f)):
program demo_array_constructor ! initializing small arrays implicit none integer,save :: xx(3,5)= reshape([& 1, 2, 3, 4, 5, & 10,20,30,40,50, & 11,22,33,44,-1055 & ],shape(xx),order=[2,1]) call print_matrix_int('xx array:',xx) contains subroutine print_matrix_int(title,arr) implicit none character(len=*),parameter::ident= "@(#)print_matrix_int(3f) - print small 2d integer arrays in row-column format" character(len=*),intent(in) :: title integer,intent(in) :: arr(:,:) integer :: i character(len=:),allocatable :: biggest write(*,*)trim(title) ! print title biggest=' ' ! make buffer to write integer into write(biggest,'(i0)')ceiling(log10(real(maxval(abs(arr)))))+2 ! find how many characters to use for integers biggest='(" > [",*(i'//trim(biggest)//':,","))' ! use this format to write a row do i=1,size(arr,dim=1) ! print one row of array at a time write(*,fmt=biggest,advance='no')arr(i,:) write(*,'(" ]")') enddo end subroutine print_matrix_int end program demo_array_constructor
xx array: > [ 1, 2, 3, 4, 5 ] > [ 10, 20, 30, 40, 50 ] > [ 11, 22, 33, 44, 55 ]
We could do a more robust version that handles REAL and COMPLEX values as well as NaN values, but it has already been done. If you need to print a variety of small matrices see:
dispmodule(3f), "A Fortran 95 module for pretty-printing matrices". Kristjan Jonasson, Department of Computer Science, School of Science and Engineering, University of Iceland, Hjardarhaga 4, 107 Reykjavik, Iceland (firstname.lastname@example.org).
Note that DATA statements are very flexible, and allow for perhaps the most intelligible way of specifying small arrays row by row. For example:
! fill rows using DATA statements integer,save,dimension(3,5) :: gg data gg(1,:)/ 1, 2, 3, 4, 5 / data gg(2,:)/ 10, 20, 30, 40, 50 / data gg(3,:)/ 11, 22, 33, 44, 55 /
There are other ways to use a DATA statement to fill in row-column order, including use of the SIZE(3f) function and an implied-DO:
! use implied-DO so data can be declared in row-column order integer, dimension(3,5) :: ff DATA (( ff(J,I), I=1,size(ff,dim=2)), J=1,size(ff,dim=1)) / & 01,02,03,04,05, & 10,20,30,40,50, & 11,22,33,44,55 /
Sometimes instead of using RESHAPE(3f) you will see someone initialize a vector and then equivalence it to a multi-dimensional array; especially if the code has a reason to access the data as both a vector and a matrix:
! multi-dimensional row1, row2, .... by equivalence integer,parameter :: d1=3,d2=5 integer :: ee(d1,d2) ! note that the DATA statements could be used to initialize the array instead integer :: e(d1*d2) =[1,10,11, 2,20,22, 3,30,33, 4,40,44, 5,50,55] equivalence (e(1),ee(1,1))
Remember that for simple initializations vector statements can be used
real :: arr(10,20)=0.0 ! array constructors can be used to define constants, not just vectors integer,parameter :: ii(10,10)=[(i,i=1,size(ii))] ! odd numbers using implied-DO
and that if things are too complicated you can just set the values in the executable body of the code.
program test_random_number real :: r(5,5) call random_number(r) end program
Remember that a DATA statement does not require that all values be initialized, whereas an array constructor does; and that you cannot initialize values multiple times and be standard-conforming.