Deprecate nprocs, and better estimate for max_blocks

[anton-seaice]

Deprecate nprocs, and better estimate for max_blocks if not set in ice_in

Closes #145

[micaeljtoliveira]

Overall looks good. Just a few suggestions, mostly optional.

[micaeljtoliveira]

Suggested change

	+ if (nprocs .ne. -1) then
	+ if (nprocs /= -1) then

Easier to read and seems to be the style favored in the rest of the module.

[micaeljtoliveira]

You could move this to be inside the above if statement, so that there's no need to check again if my_task == master_task.

[micaeljtoliveira]

No need to broadcast the number of procs anymore, as all procs are now getting the value with the call to get_num_procs.

[micaeljtoliveira]

I would write this in a slightly different way:

  ! Broadcast data from namelist
   call broadcast_scalar(processor_shape,   master_task)
   call broadcast_scalar(distribution_type, master_task)
   call broadcast_scalar(distribution_wght, master_task)
   call broadcast_scalar(distribution_wght_file, master_task)
   call broadcast_scalar(ew_boundary_type,  master_task)
   call broadcast_scalar(ns_boundary_type,  master_task)
   call broadcast_scalar(maskhalo_dyn,      master_task)
   call broadcast_scalar(maskhalo_remap,    master_task)
   call broadcast_scalar(maskhalo_bound,    master_task)
   call broadcast_scalar(add_mpi_barriers,  master_task)
   call broadcast_scalar(debug_blocks,      master_task)
   call broadcast_scalar(max_blocks,      master_task)
   call broadcast_scalar(block_size_x,      master_task)
   call broadcast_scalar(block_size_y,      master_task)
   call broadcast_scalar(nx_global,         master_task)
   call broadcast_scalar(ny_global,         master_task) 

   nprocs = get_num_procs()

   ! update nprocs_x and nprocs_y
   call proc_decomposition(nprocs, nprocs_x, nprocs_y)

   ! Automatically determine max_blocks if requested by user
   if (max_blocks < 1) then
      max_blocks=((nx_global-1)/block_size_x/nprocs_x+1) * &
                 ((ny_global-1)/block_size_y/nprocs_y+1)
      max_blocks=max(1,max_blocks)
      if (my_task == master_task) then
         write(nu_diag,'(/,a52,i6,/)') &
            '(ice_domain): max_block < 1: max_block estimated to ',max_blocks
      endif
   endif

I think this is easier to read, but in the end this a personal preference, so feel free to ignore this suggestion.

[anton-seaice]

Thanks Micael

So broadcast_scalar calls will wait for that variable to be populated before continuing ?

i.e. On PEs which are not master_task, will this will wait for block_size and max_blocks to be read from the namelist before the max_blocks calculation starts?

[micaeljtoliveira]

i.e. On PEs which are not master_task, will this will wait for block_size and max_blocks to be read from the namelist before the max_blocks calculation starts?

Yes, exactly. Each task will only return from the call to broadcast_scalar and continue execution once it has a local copy of the value being broadcasted.

[minghangli-uni]

Looks good to me. Thank you @anton-seaice.

[anton-seaice]

Made the changes proposed and tested with 1def_iaf and max_blocks not set:

Domain Information
  Processors:  total    =     48
  block_size_x,_y       =     16    15
  max_blocks            =     10

Domain Information
  Processors:  total    =     24
  block_size_x,_y       =     16    15
  max_blocks            =     20

Domain Information
  Processors:  total    =     96
  block_size_x,_y       =     16    15
  max_blocks            =     10