quantum-espresso/UtilXlib/tests/test_mp_gather_gpu.tmpl

! Implemented: i1, iv

#if defined(__CUDA)
#define {vname}
PROGRAM test_mp_gather_{vname}_gpu
!
! Simple program to check the functionalities of mp_gather.
!
!
    USE cudafor
    USE parallel_include
    USE util_param, ONLY : DP
    USE mp, ONLY : mp_gather
    USE mp_world, ONLY : mp_world_start, mp_world_end, mpime, &
                          root, nproc, world_comm
    USE tester
    IMPLICIT NONE
    !
    TYPE(tester_t) :: test
    INTEGER :: rnk, valid_sum, world_group = 0
    INTEGER, PARAMETER :: datasize = {datasize}
    !
    ! Stuff for comparing with CPU implementation
    integer :: i
    REAL(DP) :: rnd{size}
    !
    ! test variable
    {type}, DEVICE :: {vname}_d{size}
    {type}, ALLOCATABLE, DEVICE :: all_{vname}_d{allp1}
    {type} :: {vname}_h{size}
    {type}, ALLOCATABLE :: all_{vname}_h{allp1}
    {type}, ALLOCATABLE :: aux_h{allp1}
    
    !
    CALL test%init()
    
#if defined(__MPI)    
    world_group = MPI_COMM_WORLD
#endif
    CALL mp_world_start(world_group)
    !
    ! pretty ugly hack for a small test case
#if defined(i1)
    ALLOCATE(all_{vname}_h(nproc), all_{vname}_d(nproc), aux_h(nproc))
#else
    ALLOCATE(all_{vname}_h(datasize, nproc), all_{vname}_d(datasize, nproc))
    ALLOCATE(aux_h(datasize, nproc))
#endif
    !
    ! The sum of n ints with rank=1 is (zero based)
    !  sum = (n-1)*n*0.5
    !
    ! For a rank N matrix is size^N * sum
    !
    rnk = SIZE(SHAPE({vname}_h))
    valid_sum = ({datasize}**rnk) * (nproc-1)*nproc/2
    !
    {vname}_h = mpime
    {vname}_d = {vname}_h
    DO i=0, nproc-1
      !mp_gather    mydata, alldata, root, gid)
      CALL mp_gather({vname}_d, all_{vname}_d, i , world_comm)
      all_{vname}_h = all_{vname}_d
      !
      IF (mpime == i) CALL test%assert_equal(SUM(all_{vname}_h) , valid_sum )
    END DO
    !
    !
    ! Test against CPU implementation
    CALL save_random_seed("test_mp_gather_{vname}_gpu", mpime)
    !
    DO i = 0, nproc-1
      CALL RANDOM_NUMBER(rnd)
      {vname}_h = {typeconv} ( 10.0 * rnd )
      {vname}_d = {vname}_h
      aux_h = 0
      CALL mp_gather({vname}_d, all_{vname}_d, i , world_comm)
      CALL mp_gather({vname}_h, all_{vname}_h, i , world_comm)
      aux_h = all_{vname}_d
      CALL test%assert_{compare}( SUM(all_{vname}_h) / (nproc * 10.d0) , &
                              SUM(aux_h) / (nproc * 10.d0) )
    END DO
    !
    DEALLOCATE(all_{vname}_h, all_{vname}_d)
    !
    CALL collect_results(test)
    !
    CALL mp_world_end()
    !
    IF (mpime .eq. 0) CALL test%print()
    !
END PROGRAM test_mp_gather_{vname}_gpu
#else
PROGRAM test_mp_gather_{vname}_gpu
    CALL no_test()
END PROGRAM test_mp_gather_{vname}_gpu
#endif