# OpenMP error with allocatable character variables **URL:** \ **Category:** Uncategorized\ **Created:** [December 26, 2024, 8:16pm UTC](https://fortran-lang.discourse.group/t/openmp-error-with-allocatable-character-variables/9009 "2024-12-26T20:16:39Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![cwetaski](https://avatars.discourse-cdn.com/v4/letter/c/85f322/32.png) [@cwetaski](https://fortran-lang.discourse.group/u/cwetaski)\ **Post date:** [December 26, 2024, 8:16pm UTC](https://fortran-lang.discourse.group/t/openmp-error-with-allocatable-character-variables/9009/1 "2024-12-26T20:16:39Z")
Hi all, I’m wondering if anyone has any insights as to why openMP parallel blocks tend to behave poorly with allocatable character variables. In this MWE, just the existence of an allocatable character variable (that is allocated at runtime) results in a runtime error (I’m using gfortran 14.2 on Windows, compiling only with the -fopenmp flag) when the code enters the openMP block: ```auto program scratch use omp_lib, only: omp_set_num_threads use iso_fortran_env, only: output_unit integer :: n = 1 character(len=:), allocatable :: char1 character(len=:), allocatable :: char2 integer :: x = 5 call omp_set_num_threads(n) char1 = 'teststring1234' char2 = 'teststring1234' print *, char2 !$OMP parallel ! If I uncomment the following two lines, the error is avoided ! char1 = char1 ! char2 = char2 call do_anything(char1, x) call do_anything(char2, x) print *, x !$OMP end parallel contains subroutine do_anything(char,y) character(len=*), intent(in) :: char integer, intent(inout) :: y y = 2 * y end subroutine do_anything end program ``` The resulting error is: `In file 'scratch.f90', around line 14: Error allocating 2879386768 bytes: Not enough space` Obviously this program doesn’t do anything, and it doesn’t make sense to have a function which doesn’t use one of its arguments, but the point is that the purpose of the function is irrelevant. Interestingly, first assigning the variables to themselves within the omp block prevents the error. Also, the reason I write the MWE with two different allocatable variables is that, at least on my machine, using just one of them doesn’t lead to the runtime error. Although if I were to just use e.g., char1, and then within the omp block I try to assign it to another (non allocatable) character variable, it just assigns gibberish. So I assume there’s still some memory problems happening when I just use 1 variable, and its allocating a huge amount, it’s just not enough to cause the runtime error. In my real code, I can pretty easily avoid this error, but I’m still wondering if anyone knows why this happens in the first place? ---
**Author:** ![gardhor](https://yyz2.discourse-cdn.com/free1/user_avatar/fortran-lang.discourse.group/gardhor/32/88_2.png) [@gardhor](https://fortran-lang.discourse.group/u/gardhor)\ **Post date:** [December 27, 2024, 11:04am UTC](https://fortran-lang.discourse.group/t/openmp-error-with-allocatable-character-variables/9009/2 "2024-12-27T11:04:45Z")
Hi, I have some issues with allocatable character string with openmp and gfortran. See this post : > [@Integer or real to character string: bug with gfortran with OpenMP?](https://fortran-lang.discourse.group/t/integer-or-real-to-character-string-bug-with-gfortran-with-openmp/8408): > > This post is related this “[Equivalent of str function (from python) in Fortran](https://fortran-lang.discourse.group/t/equivalent-of-str-function-from-python-in-fortran/2683/1)”, but I prefer to open a new post. For a large code, I want to use a function equivalent to « TO\_string » to convert an integer or a real to a character string to create some log output. There are several implementations, I’ve got one that is similar to those in the Fortran stdlib or other libraries (fstring or M\_msg). In all these procedures, the integer (except for Fortran stdlib) or the real are written in a chara… As far as I know, this problem is not solved. ---
**Author:** ![MattAl](https://avatars.discourse-cdn.com/v4/letter/m/6a8cbe/32.png) [@MattAl](https://fortran-lang.discourse.group/u/MattAl)\ **Post date:** [December 27, 2024, 1:59pm UTC](https://fortran-lang.discourse.group/t/openmp-error-with-allocatable-character-variables/9009/3 "2024-12-27T13:59:46Z")
~Not sure how the resulting error comes about in your MWE; there’s no `write` or `open` file anywhere.~ Edit: The error of course refers to the _source_ code, my bad. In any case, compiling the MWE with `-Wall` gives a hint: ```sh $ mpif90 -o test -Wall -fopenmp scratch.f90 scratch.f90:24:31: 24 | subroutine do_anything(char,y) | 1 Warning: Unused dummy argument 'char' at (1) [-Wunused-dummy-argument] scratch.f90:13:18: 13 | !$OMP parallel | ^ Warning: '.char1' may be used uninitialized [-Wmaybe-uninitialized] scratch.f90:28:30: 28 | end subroutine do_anything | ^ note: '.char1' was declared here scratch.f90:13:18: 13 | !$OMP parallel | ^ Warning: '.char2' may be used uninitialized [-Wmaybe-uninitialized] scratch.f90:28:30: 28 | end subroutine do_anything | ^ note: '.char2' was declared here ``` In OpenMP land, the `parallel` on its own defines the start of the parallel region. That’s the first problem; the parallel region just says: Run the enclosed code section with all threads. The second problem is that since `char1` and `char2` are initialized before the parallel region, the default behaviour is to enter the region uninitialized. Hence, if you re-initialize them (by uncommenting the code) the problem goes away. You ought to use `default(none)` besides `parallel`, so that the compiler forces you to define everything explicitly. One solution to the problem would be `threadprivate` variables: Edit: `x` has the same problem with initialization, and has to be passed in with `firstprivate` instead: ```fortran module global_strings implicit none character(len=:), allocatable, save :: char1, char2 !$OMP THREADPRIVATE(char1, char2) end module global_strings program scratch use global_strings, only: char1, char2 integer :: x x = 5 char1 = 'test_char1' char2 = 'test_char2' print *, char2 !$OMP parallel default(none) copyin(char1, char2) firstprivate(x) call do_anything(char1, x) call do_anything(char2, x) print *, x !$OMP end parallel ... ``` The output looks like this: ```sh $ export OMP_NUM_THREADS=1 $ ./test test_char2 20 $ export OMP_NUM_THREADS=2 $ ./test test_char2 20 20 ```