Feature request: Cray Compiler Environment mkmf.template

[hkershaw-brown]

Use case

Compiling with cce on Derecho/Gust

There is an mkmf.tempalate here: https://github.com/NCAR/DART/tree/derecho-cce

Out of the box, get this interesting output of 2*-1 rather than -1 -1 for obs_seq.final:

note, ifort & ifx you get the expected:

  0.899859108742901
   1.04325699682186
  0.784974601221952
  0.000000000000000E+000
  0.000000000000000E+000
       39999          -1          -1
obdef
loc1d
   0.9681760430187080
kind
           1
 57600         41
   1.00000000000000

[hkershaw-brown]

hkershaw@gust02:~/issues/cce-format> cat main.f90 
program write_me

implicit none

real :: a,b,c

a = 1.0
b = 1.0
c = 2.0

print*, a, b
print*, a, b, c
print*, a, c, b
print*, a, b, b, b
write(*,*) a, b
write(6,'(2f8.2)') a, b

end program write_me
hkershaw@gust02:~/issues/cce-format> ftn main.f90 
hkershaw@gust02:~/issues/cce-format> ./a.out 
 2*1.
 2*1.,  2.
 1.,  2.,  1.
 4*1.
 2*1.
    1.00    1.00
hkershaw@gust02:~/issues/cce-format> module -t list
ncarenv/23.04
craype/2.7.20
cray-libsci/23.02.1.1
cce/15.0.1
ncarcompilers/0.8.0
hdf5/1.12.2
netcdf/4.9.1
cray-mpich/8.1.25

[hkershaw-brown]

repeat counts (2*-1) are acceptable in the Fortran standard for list-directed IO. The separators can be blanks, commas, or slashes.
39999 -1 -1 39999, 2*-1 same difference to Fortran
The write_obs is list-directed write(file_id, *) so this needs a format specification if the obs_seq files need to look the same (e.g. if you rely on diff)

DART/assimilation_code/modules/observations/obs_sequence_mod.f90

Lines 2508 to 2512 in 306af91

	write(file_id, *) obs%prev_time, obs%next_time, obs%cov_group
	call write_obs_def(file_id, obs%def, obs%key)
	endif
	end subroutine write_obs

side note here, the obs%cov_group is -1 all the time.

[mjs2369]

Formatting bug (output of 2*-1 rather than -1 -1) moved to separate issue #484