pc_hsmg.f90

Go to the documentation of this file.

! Copyright (c) 2008-2020, UCHICAGO ARGONNE, LLC.
!
! The UChicago Argonne, LLC as Operator of Argonne National
! Laboratory holds copyright in the Software. The copyright holder
! reserves all rights except those expressly granted to licensees,
! and U.S. Government license rights.
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions
! are met:
!
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the disclaimer below.
!
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions and the disclaimer (as noted below)
! in the documentation and/or other materials provided with the
! distribution.
!
! 3. Neither the name of ANL nor the names of its contributors
! may be used to endorse or promote products derived from this software
! without specific prior written permission.
!
! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
! "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
! LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
! FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
! UCHICAGO ARGONNE, LLC, THE U.S. DEPARTMENT OF
! ENERGY OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
! SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
! TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
! DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
! THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
! (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
! OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
!
! Additional BSD Notice
! ---------------------
! 1. This notice is required to be provided under our contract with
! the U.S. Department of Energy (DOE). This work was produced at
! Argonne National Laboratory under Contract
! No. DE-AC02-06CH11357 with the DOE.
!
! 2. Neither the United States Government nor UCHICAGO ARGONNE,
! LLC nor any of their employees, makes any warranty,
! express or implied, or assumes any liability or responsibility for the
! accuracy, completeness, or usefulness of any information, apparatus,
! product, or process disclosed, or represents that its use would not
! infringe privately-owned rights.
!
! 3. Also, reference herein to any specific commercial products, process,
! or services by trade name, trademark, manufacturer or otherwise does
! not necessarily constitute or imply its endorsement, recommendation,
! or favoring by the United States Government or UCHICAGO ARGONNE LLC.
! The views and opinions of authors expressed
! herein do not necessarily state or reflect those of the United States
! Government or UCHICAGO ARGONNE, LLC, and shall
! not be used for advertising or product endorsement purposes.
!
module hsmg
  use neko_config, only : neko_bcknd_device
  use num_types, only : rp
  use math, only : copy, col2, add2
  use utils, only : neko_error
  use precon, only : pc_t, precon_allocator
  use ax_product, only : ax_t, ax_helm_allocator
  use gather_scatter, only : gs_t, gs_op_add
  use interpolation, only : interpolator_t
  use bc, only : bc_t
  use bc_list, only : bc_list_t
  use scalar_bc_projector, only : scalar_bc_projector_t
  use dirichlet, only : dirichlet_t
  use schwarz, only : schwarz_t
  use jacobi, only : jacobi_t
  use sx_jacobi, only : sx_jacobi_t
  use device_jacobi, only : device_jacobi_t
  use device, only : device_map, device_event_create, device_unmap, &
       device_event_destroy, device_get_ptr, device_event_sync
  use device_math, only : device_copy, device_col2, device_add2
  use profiler, only : profiler_start_region, profiler_end_region
  use space, only : space_t, gll
  use dofmap, only : dofmap_t
  use field, only : field_t
  use coefs, only : coef_t, coef_operator
  use mesh, only : mesh_t
  use json_module, only : json_file
  use json_utils, only : json_get_or_default
  use krylov, only : ksp_t, ksp_monitor_t, ksp_max_iter, &
       krylov_solver_factory
  use tree_amg_multigrid, only : tamg_solver_t
  use zero_dirichlet, only : zero_dirichlet_t
  use logger, only : neko_log, log_size
  use, intrinsic :: iso_c_binding, only : c_ptr, c_null_ptr, c_associated
  !$ use omp_lib
  implicit none
  private
 
  !Struct to arrange our multigridlevels
  type, private :: multigrid_t
     type(dofmap_t), pointer :: dof => null()
     type(gs_t), pointer :: gs_h => null()
     type(space_t), pointer :: xh => null()
     type(coef_t), pointer :: coef => null()
     type(scalar_bc_projector_t), pointer :: bc_projector => null()
     type(schwarz_t), pointer :: schwarz => null()
     type(field_t), pointer :: e => null()
  end type multigrid_t
 
  type, public, extends(pc_t) :: hsmg_t
     type(mesh_t), pointer :: msh => null()
     integer :: nlvls
     type(multigrid_t), allocatable :: grids(:)
     type(gs_t) :: gs_crs, gs_mg
     type(space_t) :: xh_crs, xh_mg
     type(dofmap_t) :: dm_crs, dm_mg
     type(coef_t) :: c_crs, c_mg
     type(zero_dirichlet_t) :: bc_crs, bc_mg, bc_reg
     type(scalar_bc_projector_t) :: bc_projector_crs, bc_projector_mg, &
          bc_projector_reg
     type(schwarz_t) :: schwarz, schwarz_mg, schwarz_crs
     type(field_t) :: e, e_mg, e_crs
     type(field_t) :: wf
     class(ksp_t), allocatable :: crs_solver
     type(tamg_solver_t), allocatable :: amg_solver
     integer :: niter
     class(pc_t), allocatable :: pc_crs
     class(ax_t), allocatable :: ax
     real(kind=rp), allocatable :: r(:)
     type(interpolator_t) :: interp_fine_mid
     type(interpolator_t) :: interp_mid_crs
     real(kind=rp), allocatable :: w(:) 
     type(c_ptr) :: w_d = c_null_ptr
     type(c_ptr) :: r_d = c_null_ptr
     type(c_ptr) :: hsmg_event = c_null_ptr
     type(c_ptr) :: gs_event = c_null_ptr
   contains
     procedure, pass(this) :: init => hsmg_init
     procedure, pass(this) :: init_from_components => &
          hsmg_init_from_components
     procedure, pass(this) :: free => hsmg_free
     procedure, pass(this) :: solve => hsmg_solve
     procedure, pass(this) :: update => hsmg_set_h
  end type hsmg_t
 
contains
 
  subroutine hsmg_init(this, coef, bclst, json)
    class(hsmg_t), intent(inout), target :: this
    type(coef_t), intent(in), target :: coef
    type(bc_list_t), intent(inout), target :: bclst
    type(json_file), intent(inout) :: json
    character(len=:), allocatable :: crs_solver, crs_pc
    logical :: crs_monitor
    integer :: crs_tamg_lvls, crs_tamg_itrs, crs_tamg_cheby_degree
 
    ! Exract coarse grid parameters
 
    ! Common parameters for the coarse grid
    call json_get_or_default(json, 'coarse_grid.solver', &
         crs_solver, "cg")
 
    !
    ! Parameters for a Krylov based coarse grid solverthis
    !
    call json_get_or_default(json, 'coarse_grid.iterations', &
         this%niter, 10)
 
    call json_get_or_default(json, 'coarse_grid.preconditioner', &
         crs_pc, "jacobi")
 
    call json_get_or_default(json, 'coarse_grid.monitor', &
         crs_monitor, .false.)
 
    !
    ! Parameters for a tree-amg based coarse grid solver
    !
    call json_get_or_default(json, 'coarse_grid.levels', &
         crs_tamg_lvls, 3)
 
    call json_get_or_default(json, 'coarse_grid.iterations', &
         crs_tamg_itrs, 1)
 
    call json_get_or_default(json, 'coarse_grid.cheby_degree', &
         crs_tamg_cheby_degree, 4)
 
    call this%init_from_components(coef, bclst, crs_solver, crs_pc, &
         crs_monitor, crs_tamg_lvls, crs_tamg_itrs, crs_tamg_cheby_degree)
 
  end subroutine hsmg_init
 
  subroutine hsmg_init_from_components(this, coef, bclst, crs_solver, crs_pc, &
       crs_monitor, crs_tamg_lvls, crs_tamg_itrs, crs_tamg_cheby_degree)
    class(hsmg_t), intent(inout), target :: this
    type(coef_t), intent(in), target :: coef
    type(bc_list_t), intent(inout), target :: bclst
    character(len=:), intent(inout), allocatable :: crs_solver, crs_pc
    logical, intent(inout) :: crs_monitor
    integer, intent(in) :: crs_tamg_lvls, crs_tamg_itrs, crs_tamg_cheby_degree
    integer :: n, i
    integer :: lx_crs, lx_mid
    class(bc_t), pointer :: bc_i
 
    character(len=LOG_SIZE) :: log_buf
 
    call this%free()
    this%nlvls = 3
    lx_crs = 2
    if (coef%Xh%lx .lt. 5) then
       lx_mid = max(coef%Xh%lx-1,3)
 
       if (coef%Xh%lx .le. 2) then
          call neko_error('Polynomial order < 2 not supported for hsmg precon')
       end if
 
    else
       lx_mid = 4
    end if
 
 
    call neko_log%section('HSMG')
    if (this%nlvls .lt. 1e1) then
       write(log_buf, '(A,I1,A)') 'HSMG hierarchy      : ', &
            this%nlvls, ' levels'
    else if (this%nlvls .lt. 1e2) then
       write(log_buf, '(A,I2,A)') 'HSMG hierarchy      : ', &
            this%nlvls, ' levels'
    else if (this%nlvls .lt. 1e3) then
       write(log_buf, '(A,I3,A)') 'HSMG hierarchy      : ', this%nlvls, &
            ' levels'
    else
       write(log_buf, '(A,I6,A)') 'HSMG hierarchy      : ', this%nlvls, &
            ' levels'
    end if
    call neko_log%message(log_buf)
    if (trim(crs_solver) .ne. 'tamg' .or. trim(crs_solver) .eq. 'cheby') then
       call neko_log%message('Coarse grid solver  : (' // trim(crs_solver) // &
            ', ' // trim(crs_pc) // ')')
 
       if (this%niter .lt. 1e1) then
          write(log_buf, '(A,I1)') 'Coarse grid iters.  : ', this%niter
       else if (this%niter .lt. 1e2) then
          write(log_buf, '(A,I2)') 'Coarse grid iters.  : ', this%niter
       else if (this%niter .lt. 1e3) then
          write(log_buf, '(A,I3)') 'Coarse grid iters.  : ', this%niter
       else if (this%niter .lt. 1e4) then
          write(log_buf, '(A,I4)') 'Coarse grid iters.  : ', this%niter
       else
          write(log_buf, '(A,I6)') 'Coarse grid iters.  : ', this%niter
       end if
 
       call neko_log%message(log_buf)
    else
       call neko_log%message('Coarse grid solver  : ' // trim(crs_solver) )
    end if
 
    this%msh => coef%msh
    allocate(this%grids(this%nlvls))
    allocate(this%w(coef%dof%size()))
    allocate(this%r(coef%dof%size()))
 
 
    ! Compute all elements as if they are deformed
    call coef%msh%all_deformed()
 
    n = coef%dof%size()
    call this%e%init(coef%dof, 'work array')
    call this%wf%init(coef%dof, 'work 2')
 
    call this%Xh_crs%init(gll, lx_crs, lx_crs, lx_crs)
    call this%dm_crs%init(coef%msh, this%Xh_crs)
    call this%gs_crs%init(this%dm_crs)
    call this%e_crs%init(this%dm_crs, 'work crs')
    ! Both coarse levels read G_ij, h1, h2, B and mult and nothing else, and
    ! hsmg never rebuilds their geometry, so the rest is dead weight.
    call this%c_crs%init(this%gs_crs, coef_operator)
 
    call this%Xh_mg%init(gll, lx_mid, lx_mid, lx_mid)
    call this%dm_mg%init(coef%msh, this%Xh_mg)
    call this%gs_mg%init(this%dm_mg)
    call this%e_mg%init(this%dm_mg, 'work midl')
    call this%c_mg%init(this%gs_mg, coef_operator)
 
    ! Create backend specific Ax operator. The coarse levels compress their
    ! geometric factors when the fine level does
    if (coef%geo_compression) then
       call this%c_crs%enable_geo_compression(coef%geo_compression_tol)
       call this%c_mg%enable_geo_compression(coef%geo_compression_tol)
       call ax_helm_allocator(this%ax, type_name = "standard_compr")
    else
       call ax_helm_allocator(this%ax, type_name = "standard")
    end if
 
    call this%bc_crs%init_base(this%c_crs)
    call this%bc_mg%init_base(this%c_mg)
    call this%bc_reg%init_base(coef)
    if (bclst%size() .gt. 0) then
       do i = 1, bclst%size()
          bc_i => bclst%get(i)
          call this%bc_reg%mark_facets(bc_i%marked_facet)
          bc_i => bclst%get(i)
          call this%bc_crs%mark_facets(bc_i%marked_facet)
          bc_i => bclst%get(i)
          call this%bc_mg%mark_facets(bc_i%marked_facet)
       end do
    end if
    call this%bc_reg%finalize()
    call this%bc_crs%finalize()
    call this%bc_mg%finalize()
 
    call this%bc_projector_reg%mark(this%bc_reg)
    call this%bc_projector_crs%mark(this%bc_crs)
    call this%bc_projector_mg%mark(this%bc_mg)
 
    call this%schwarz%init(coef%Xh, coef%dof, coef%gs_h, &
         this%bc_projector_reg, coef%msh)
    call this%schwarz_mg%init(this%Xh_mg, this%dm_mg, this%gs_mg,&
         this%bc_projector_mg, coef%msh)
 
    call this%interp_fine_mid%init(coef%Xh, this%Xh_mg)
    call this%interp_mid_crs%init(this%Xh_mg, this%Xh_crs)
 
    call hsmg_fill_grid(coef%dof, coef%gs_h, coef%Xh, coef, &
         this%bc_projector_reg, this%schwarz, this%e, this%grids, 3)
    call hsmg_fill_grid(this%dm_mg, this%gs_mg, this%Xh_mg, this%c_mg, &
         this%bc_projector_mg, this%schwarz_mg, this%e_mg, this%grids, 2)
    call hsmg_fill_grid(this%dm_crs, this%gs_crs, this%Xh_crs, &
         this%c_crs, this%bc_projector_crs, this%schwarz_crs, &
         this%e_crs, this%grids, 1)
 
    call hsmg_set_h(this)
    if (neko_bcknd_device .eq. 1) then
       call device_map(this%w, this%w_d, n)
       call device_map(this%r, this%r_d, n)
    end if
 
    call device_event_create(this%hsmg_event, 2)
    call device_event_create(this%gs_event, 2)
 
 
 
    ! Create a backend specific krylov solver
    if (trim(crs_solver) .eq. 'tamg') then
       allocate(this%amg_solver)
       call this%amg_solver%init(this%ax, this%grids(1)%e%Xh, &
            this%grids(1)%coef, this%msh, this%grids(1)%gs_h, crs_tamg_lvls, &
            this%grids(1)%bc_projector, crs_tamg_itrs, crs_tamg_cheby_degree)
    else
       ! Create a backend specific preconditioner
       call precon_allocator(this%pc_crs, crs_pc)
 
       select type (pc => this%pc_crs)
       type is (jacobi_t)
          call pc%init_from_components(this%c_crs, this%dm_crs, this%gs_crs)
       type is (sx_jacobi_t)
          call pc%init_from_components(this%c_crs, this%dm_crs, this%gs_crs)
       type is (device_jacobi_t)
          call pc%init_from_components(this%c_crs, this%dm_crs, this%gs_crs)
       end select
 
       call krylov_solver_factory(this%crs_solver, &
            this%dm_crs%size(), trim(crs_solver), ksp_max_iter, &
            m = this%pc_crs, monitor = crs_monitor)
    end if
 
    call neko_log%end_section()
 
  end subroutine hsmg_init_from_components
 
  subroutine hsmg_set_h(this)
    class(hsmg_t), intent(inout) :: this
    !    integer :: i
    ! Yeah I dont really know what to do here. For incompressible flow not
    ! much happens
    this%grids(1)%coef%ifh2 = .false.
    call copy(this%grids(1)%coef%h1, this%grids(3)%coef%h1, &
         this%grids(1)%dof%size())
    if (neko_bcknd_device .eq. 1) then
       call device_copy(this%grids(1)%coef%h1_d, this%grids(3)%coef%h1_d, &
            this%grids(1)%dof%size())
    end if
  end subroutine hsmg_set_h
 
 
  subroutine hsmg_fill_grid(dof, gs_h, Xh, coef, bc_projector, schwarz, e, grids, l)
    type(dofmap_t), target, intent(in) :: dof
    type(gs_t), target, intent(in) :: gs_h
    type(space_t), target, intent(in) :: Xh
    type(coef_t), target, intent(in) :: coef
    type(scalar_bc_projector_t), target, intent(in) :: bc_projector
    type(schwarz_t), target, intent(in) :: schwarz
    type(field_t), target, intent(in) :: e
    integer, intent(in) :: l
    type(multigrid_t), intent(inout), dimension(l) :: grids
 
 
    grids(l)%dof => dof
    grids(l)%gs_h => gs_h
    grids(l)%Xh => xh
    grids(l)%coef => coef
    grids(l)%bc_projector => bc_projector
    grids(l)%schwarz => schwarz
    grids(l)%e => e
 
  end subroutine hsmg_fill_grid
 
  subroutine hsmg_free(this)
    class(hsmg_t), intent(inout) :: this
 
    if (allocated(this%ax)) then
       call this%ax%free()
       deallocate(this%ax)
    end if
 
    if (allocated(this%grids)) then
       deallocate(this%grids)
    end if
 
    if (allocated(this%w)) then
       if (c_associated(this%w_d)) then
          call device_unmap(this%w, this%w_d)
       end if
       deallocate(this%w)
    end if
 
    if (allocated(this%r)) then
       if (c_associated(this%r_d)) then
          call device_unmap(this%r, this%r_d)
       end if
       deallocate(this%r)
    end if
 
    call this%schwarz%free()
    call this%schwarz_mg%free()
 
    call this%c_crs%free()
    call this%c_mg%free()
    call this%e%free()
    call this%e_mg%free()
    call this%e_crs%free()
    call this%wf%free()
 
    call this%gs_crs%free()
    call this%gs_mg%free()
    call this%interp_mid_crs%free()
    call this%interp_fine_mid%free()
 
    call this%bc_crs%free()
    call this%bc_mg%free()
    call this%bc_reg%free()
 
    call this%bc_projector_reg%free()
    call this%bc_projector_crs%free()
    call this%bc_projector_mg%free()
 
    if (allocated(this%crs_solver)) then
       call this%crs_solver%free()
       deallocate(this%crs_solver)
    end if
 
    if (allocated(this%amg_solver)) then
       call this%amg_solver%free()
       deallocate(this%amg_solver)
    end if
 
    if (allocated(this%pc_crs)) then
       call this%pc_crs%free()
    end if
 
    if (c_associated(this%hsmg_event)) then
       call device_event_destroy(this%hsmg_event)
    end if
    if (c_associated(this%gs_event)) then
       call device_event_destroy(this%gs_event)
    end if
 
    call this%dm_crs%free()
    call this%dm_mg%free()
    call this%Xh_crs%free()
    call this%Xh_mg%free()
 
  end subroutine hsmg_free
 
  subroutine hsmg_solve(this, z, r, n)
    integer, intent(in) :: n
    class(hsmg_t), intent(inout) :: this
    real(kind=rp), dimension(n), intent(inout) :: z
    real(kind=rp), dimension(n), intent(inout) :: r
    type(c_ptr) :: z_d, r_d
    type(ksp_monitor_t) :: crs_info
    integer :: thrdid, nthrds
 
    call profiler_start_region('HSMG_solve', 8)
    if (neko_bcknd_device .eq. 1) then
       z_d = device_get_ptr(z)
       r_d = device_get_ptr(r)
       !We should not work with the input
       call device_copy(this%r_d, r_d, n)
       call this%bc_projector_reg%apply(this%r, n)
 
       !OVERLAPPING Schwarz exchange and solve
       !! DOWNWARD Leg of V-cycle, we are pretty hardcoded here but w/e
       call device_col2(this%r_d, this%grids(3)%coef%mult_d, &
            this%grids(3)%dof%size())
       !Restrict to middle level
       call this%interp_fine_mid%map(this%e%x, this%r, &
            this%msh%nelv, this%grids(2)%Xh)
       call this%grids(2)%gs_h%op(this%e%x, &
            this%grids(2)%dof%size(), gs_op_add, this%gs_event)
       call device_event_sync(this%gs_event)
       !This should probably be double checked again
       call device_copy(this%r_d, r_d, n)
       call this%bc_projector_reg%apply(this%r, n)
       call device_copy(this%w_d, this%e%x_d, this%grids(2)%dof%size())
       call this%bc_projector_mg%apply(this%w, this%grids(2)%dof%size())
       !OVERLAPPING Schwarz exchange and solve
       call device_col2(this%w_d, this%grids(2)%coef%mult_d, &
            this%grids(2)%dof%size())
       !restrict residual to crs
       call this%interp_mid_crs%map(this%wf%x, this%w, this%msh%nelv, &
            this%grids(1)%Xh)
       !Crs solve
       call device_copy(this%w_d, this%e%x_d, this%grids(2)%dof%size())
       call this%bc_projector_mg%apply(this%w, this%grids(2)%dof%size())
 
       !$omp parallel private(thrdid, nthrds)
 
       thrdid = 0
       nthrds = 1
       !$ thrdid = omp_get_thread_num()
       !$ nthrds = omp_get_num_threads()
 
       if (thrdid .eq. 0) then
          call profiler_start_region('HSMG_schwarz', 9)
          call this%grids(3)%schwarz%compute(z, this%r)
          call this%grids(2)%schwarz%compute(this%grids(2)%e%x, this%w)
          call profiler_end_region('HSMG_schwarz', 9)
       end if
       if (nthrds .eq. 1 .or. thrdid .eq. 1) then
          call profiler_start_region('HSMG_coarse_grid', 10)
          call this%grids(1)%gs_h%op(this%wf%x, &
               this%grids(1)%dof%size(), gs_op_add, this%gs_event)
          call device_event_sync(this%gs_event)
          call this%grids(1)%bc_projector%apply(this%wf%x, &
               this%grids(1)%dof%size())
          call profiler_start_region('HSMG_coarse_solve', 11)
          if (allocated(this%amg_solver)) then
             call this%amg_solver%solve(this%grids(1)%e%x, this%wf%x, &
                  this%grids(1)%dof%size())
          else
             crs_info = this%crs_solver%solve(this%Ax, this%grids(1)%e, &
                  this%wf%x, &
                  this%grids(1)%dof%size(), &
                  this%grids(1)%coef, &
                  this%grids(1)%bc_projector, &
                  this%grids(1)%gs_h, this%niter)
          end if
          call profiler_end_region('HSMG_coarse_solve', 11)
          call this%grids(1)%bc_projector%apply(this%grids(1)%e%x,&
               this%grids(1)%dof%size())
          call profiler_end_region('HSMG_coarse_grid', 10)
       end if
       !$omp end parallel
 
       call this%interp_mid_crs%map(this%w, this%grids(1)%e%x, &
            this%msh%nelv, this%grids(2)%Xh)
       call device_add2(this%grids(2)%e%x_d, this%w_d, this%grids(2)%dof%size())
 
       call this%interp_fine_mid%map(this%w, this%grids(2)%e%x, &
            this%msh%nelv, this%grids(3)%Xh)
       call device_add2(z_d, this%w_d, this%grids(3)%dof%size())
       call this%grids(3)%gs_h%op(z, this%grids(3)%dof%size(), &
            gs_op_add, this%gs_event)
       call device_event_sync(this%gs_event)
       call device_col2(z_d, this%grids(3)%coef%mult_d, &
            this%grids(3)%dof%size())
    else
       !We should not work with the input
       call copy(this%r, r, n)
 
       !OVERLAPPING Schwarz exchange and solve
       call this%grids(3)%schwarz%compute(z, this%r)
       ! DOWNWARD Leg of V-cycle, we are pretty hardcoded here but w/e
       call col2(this%r, this%grids(3)%coef%mult, &
            this%grids(3)%dof%size())
       !Restrict to middle level
       call this%interp_fine_mid%map(this%w, this%r, &
            this%msh%nelv, this%grids(2)%Xh)
       call this%grids(2)%gs_h%op(this%w, this%grids(2)%dof%size(), gs_op_add)
       !OVERLAPPING Schwarz exchange and solve
       call this%grids(2)%schwarz%compute(this%grids(2)%e%x, this%w)
       call col2(this%w, this%grids(2)%coef%mult, this%grids(2)%dof%size())
       !restrict residual to crs
       call this%interp_mid_crs%map(this%r, this%w, &
            this%msh%nelv, this%grids(1)%Xh)
       !Crs solve
 
       call this%grids(1)%gs_h%op(this%r, this%grids(1)%dof%size(), gs_op_add)
       call this%grids(1)%bc_projector%apply(this%r, this%grids(1)%dof%size())
 
       call profiler_start_region('HSMG_coarse-solve', 11)
       if (allocated(this%amg_solver)) then
          call this%amg_solver%solve(this%grids(1)%e%x, this%r, &
               this%grids(1)%dof%size())
       else
          crs_info = this%crs_solver%solve(this%Ax, this%grids(1)%e, this%r, &
               this%grids(1)%dof%size(), &
               this%grids(1)%coef, &
               this%grids(1)%bc_projector, &
               this%grids(1)%gs_h, this%niter)
       end if
       call profiler_end_region('HSMG_coarse-solve', 11)
 
       call this%grids(1)%bc_projector%apply(this%grids(1)%e%x, &
            this%grids(1)%dof%size())
 
 
       call this%interp_mid_crs%map(this%w, this%grids(1)%e%x, &
            this%msh%nelv, this%grids(2)%Xh)
       call add2(this%grids(2)%e%x, this%w, this%grids(2)%dof%size())
 
       call this%interp_fine_mid%map(this%w, this%grids(2)%e%x, &
            this%msh%nelv, this%grids(3)%Xh)
       call add2(z, this%w, this%grids(3)%dof%size())
       call this%grids(3)%gs_h%op(z, this%grids(3)%dof%size(), gs_op_add)
       call col2(z, this%grids(3)%coef%mult, this%grids(3)%dof%size())
 
    end if
    call profiler_end_region('HSMG_solve', 8)
  end subroutine hsmg_solve
end module hsmg