Hi Anton,

The problematic statement is

  allocate (c[*], source=b)

The syntax for a coarray reference that includes [ ] is required to 
include the ( ) part of the syntax as well if the object is an array. 
There is no syntax defined for C[*] where C is an array.  This has the 
side effect of not allowing the "get the shape from the source= array" 
mode of allocate for allocating a coarray.  It would be possible to 
extend the standard to allow something like that if additional syntax 
were defined and rules were added to make sure that the shape of the 
source= array was the same on all images.

Another compiler gives this error for the code:

   allocate (c[*], source=b)
             ^
ftn-402 crayftn: ERROR $MAIN, File = test.f90, Line = 12, Column = 13
   The number of extents specified for this allocate object does not 
match the declared rank.

It is possible (though this is just speculation) that the Intel compiler 
treated C as a scalar here and the part that involved copying the values 
from B caused the segfault.

Note that if you opted for this statement instead:

   allocate (c(0:11,3)[*], source=b)

then there is no error at execution.  The part of the allocate that 
involves copying the value of the source to the allocated object is 
purely local, so it does not matter that the allocated object is a coarray.

Cheers,
Bill


On 12/3/13 6:01 AM, Anton Shterenlikht wrote:
> I've asked in comp.lang.fortran usenet group,
> but got no answer, so am trying here.
> Is sourced allocation allowed for coarrays
> but the standard?
>
> This program:
>
>    1 integer, allocatable :: a(:,:)[:], b(:,:)[:], c(:,:)[:]
>    2
>    3 allocate (a(10,10)[*], source=0)
>    4
>    5 a = 0
>    6 if (this_image() .eq. 1) a=1
>    7
>    8 sync all
>    9
>   10 allocate (b(0:11,3)[*], source=0)
>   11 !allocate (c(0:11,3)[*], source=b)
>   12 allocate (c[*], source=b)
>   13
>   14 b(1:10,1:3) = a(1:10,1:3) [1]
>   15 c(1:10,1:3) = a(1:10,4:6) [1]
>   16
>   17 sync all
>   18
>   19 if ( this_image() .gt. 1) b(0,:) = c(10,:) [this_image()-1]
>   20
>   21 end
>
> compiled with ifort 14, and run with shared
> memory gives segfault:
>
> $ ./a.out
> forrtl: severe (174): SIGSEGV, segmentation fault occurred
> In coarray image 2
> Image              PC                Routine            Line        Source
> a.out              000000000046B199  Unknown               Unknown  Unknown
> a.out              0000000000469B10  Unknown               Unknown  Unknown
> a.out              00000000004392E2  Unknown               Unknown  Unknown
> a.out              000000000041F813  Unknown               Unknown  Unknown
> a.out              000000000040730B  Unknown               Unknown  Unknown
> libpthread.so.0    00002AAAAB1B7500  Unknown               Unknown  Unknown
> libicaf.so         00002AAAAACD026D  Unknown               Unknown  Unknown
> a.out              0000000000406866  Unknown               Unknown  Unknown
> a.out              00000000004038C6  Unknown               Unknown  Unknown
> libc.so.6          00002AAAAB3E4CDD  Unknown               Unknown  Unknown
> a.out              00000000004037B9  Unknown               Unknown  Unknown
>
> application called MPI_Abort(comm=0x84000000, 3) - process 1
> rank 1 in job 1  newblue4_35267   caused collective abort of all ranks
>   exit status of rank 1: return code 3
> $
>
> However, when line 12 is commente out
> and line 11 is used instead, the program
> runs fine.
>
> AS far as I understand the standard, the
> meaning of the two statements:
>
> allocate (c(0:11,3)[*], source=b)
> allocate (c[*], source=b)
>
> is identical. Is that not so?
>
> Is the segfault expected?
>
> Thanks
>
> Anton
>

-- 
Bill Long                                           [log in to unmask]
Fortran Technical Support    &                 voice: 651-605-9024
Bioinformatics Software Development            fax:   651-605-9142
Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101