ReserveMessages_DomainDecomposition_t Subroutine

  subroutine ReserveMessages_DomainDecomposition_t(this,nMsg)
  !! Ensures the non-blocking request and status scratch can hold `nMsg` messages
  !! in flight, growing it when it cannot.
  !!
  !! GenerateDecomposition sizes the scratch from the mesh alone (one entry per
  !! unique side), but an exchange posts a send and a receive per rank-remote side
  !! *per variable* -- and, for vectors, per direction. The message count is
  !! therefore a multiple of the side count that the mesh cannot know, and a
  !! solution with enough variables overruns the scratch. Each exchange counts the
  !! sides it will message (CountRemoteSides / CountRemoteMortarSides) and reserves
  !! from that count before posting; the call is a comparison once the scratch is
  !! large enough.
  !!
  !! Reserving from the rank-remote side count rather than from the local side
  !! count matters at scale: the remote sides are the partition boundary, while the
  !! local sides are the whole subdomain, and each entry carries an
  !! MPI_STATUS_SIZE row of `stats` with it.
  !!
  !! No message may be in flight when this is called: the scratch is reallocated,
  !! not grown in place. Every exchange posts and then waits (through
  !! FinalizeMPIExchangeAsync) before another can start, so a reservation at the
  !! head of an exchange is safe.
    implicit none
    class(DomainDecomposition_t),intent(inout) :: this
    integer,intent(in) :: nMsg

    if(nMsg <= this%maxMsg) return

    if(allocated(this%requests)) deallocate(this%requests)
    if(allocated(this%stats)) deallocate(this%stats)

    allocate(this%requests(1:nMsg))
    allocate(this%stats(MPI_STATUS_SIZE,1:nMsg))
    this%maxMsg = nMsg

  endsubroutine ReserveMessages_DomainDecomposition_t