Transformation to move (array?) operations outside call arguments

[sergisiso]

For instance in Nemo, when we find:
call iom_put('empmr', emp - rnf)

we want to transform it to:

tmp = emp - rnf
call iom_put('empmr', tmp)

so then the rest of the script will converted it to:

!$omp parallel do
do i= LBOUND(tmp,1), UBOUND(tmp,1)
  tmp(i) = emp(i) - rnf(i)
enddo
!$omp end parallel do
call iom_put('empmr', tmp)

[arporter]

We're soon going to need this in NG-Arch so we should think about what to call the transformation. Perhaps:
ReplaceArgExpressionsTrans?

[sergisiso]

Since we use "Hoist" to move expressions outside loops bounds and array declarations outside routines maybe we could use the same verb here, so HoistArgumentExpressionsTrans ?

[arporter]

Since @LonelyCat124 is making great progress in #3301, I thought I'd look at some NEMO code to find some real use cases. One is:

module zdfphy
  use oce
  use zdf_oce
  use zdfdrg
  use zdfsh2
  use zdfric
  use zdftke
  use zdfgls
  use zdfosm
  use zdfddm
  use zdfevd
  use zdfmfc
  use zdfiwm
  use zdfswm
  use zdfmxl
  use eosbn2
  use tranpc
  use trc_oce
  use sbc_oce
  use sbcrnf
  use domtile
  use in_out_manager
  use iom
  use lbclnk
  use lib_mpp
  use timing
  implicit none
  integer, parameter, private :: np_cst = 1
  integer, parameter, private :: np_ric = 2
  integer, parameter, private :: np_tke = 3
  integer, parameter, private :: np_gls = 4
  integer, parameter, private :: np_osm = 5
  integer, private :: nzdf_phy
  logical, public :: l_zdfsh2
  real(kind=wp), allocatable, dimension(:,:,:), private :: sh2
  private

contains

  subroutine zdf_nxt(kt, kbb, kaa)
    integer, intent(in) :: kt
    integer, intent(in) :: kbb
    integer, intent(in) :: kaa
    integer :: ji
    integer :: jj
    integer :: jk
    integer :: jtile

    if (ln_timing) then
      call timing_start('zdf_nxt')
    end if
    if (l_zdfsh2) then
      ALLOCATE(sh2(nis0:nie0,njs0:nje0,1:jpk))
      call zdf_sh2(kbb, kaa, avm_k, sh2)
    end if
    if (ln_tile) then
      call dom_tile_start()
    end if

    ! 'dom_tile' is not available on the 'default' accelerator device, and therefore it ca
nnot be called from within an OMP Target 
!& region. Use the 'device_string' option to specify a different device.
    do jtile = 1, nijtile, 1
      if (ln_tile) then
        call dom_tile(ntsi, ntsj, ntei, ntej, ktile=jtile)
      end if
      call eos_rab(ts(:,:,:,:,kaa), rab_n, kaa)
      call bn2(ts(:,:,:,:,kaa), rab_n, rn2, kaa)
      if (nzdf_phy == np_tke) then
        call zdf_tke(kt, kbb, kaa, sh2, avm_k, avt_k)
      else
        if (nzdf_phy == np_gls) then
          call zdf_gls(kt, kbb, kaa, sh2, avm_k, avt_k)
        end if
      end if
    enddo
    if (ln_tile) then
      call dom_tile_stop()
    end if
    if (lrst_oce) then
      if (ln_zdftke) then
        call tke_rst(kt, 'WRITE')
      end if
      if (ln_zdfgls) then
        call gls_rst(kt, 'WRITE')
      end if
    end if
    if (l_zdfsh2) then
      if (iom_use('eshear_k')) then
        !$omp target
        !$omp teams loop collapse(3) default(shared) private(ji,jj,jk)
        do jk = 2, jpkm1, 1
          do jj = ntsj, ntej, 1
            do ji = ntsi, ntei, 1
              sh2(ji,jj,jk) = sh2(ji,jj,jk) * wmask(ji,jj,jk)
            enddo
          enddo
        enddo
        !$omp end teams loop
        !$omp end target
      end if
      if (iom_use('avt_k')) then
        call iom_put('avt_k', avt_k(:,:,:) * wmask(nis0:nie0,njs0:nje0,:))
      end if
      if (iom_use('avm_k')) then
        call iom_put('avm_k', avm_k(nis0:nie0,njs0:nje0,:) * wmask(nis0:nie0,njs0:nje0,:))
      end if
      if (iom_use('estrat_k')) then
        call iom_put('estrat_k', -avt_k(:,:,:) * rn2(nis0:nie0,njs0:nje0,:) * wmask(nis0:nie0,njs0:nje0,:))
      end if
      if (iom_use('eshear_k')) then
        call iom_put('eshear_k', sh2)
      end if
      DEALLOCATE(sh2)
    end if
    if (ln_timing) then
      call timing_stop('zdf_nxt')
    end if
  end subroutine zdf_nxt
end module zdfphy

where it would be the arguments to iom_put that we'd want to transform. @LonelyCat124 could you use this as the basis for an additional test in #3301?