cuda_backend_t Derived Type

type, public, extends(base_backend_t) :: cuda_backend_t


Components

Type Visibility Attributes Name Initial
real(kind=dp), public :: nu
class(mesh_t), public, pointer :: mesh
class(allocator_t), public, pointer :: allocator
class(poisson_fft_t), public, pointer :: poisson_fft
integer, public :: MPI_FP_PREC = dp
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: u_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: u_recv_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: u_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: u_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: v_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: v_recv_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: v_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: v_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: w_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: w_recv_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: w_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: w_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: du_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: du_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: du_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: du_recv_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: dud_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: dud_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: dud_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: dud_recv_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: d2u_send_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: d2u_send_e_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: d2u_recv_s_dev
real(kind=dp), public, device, allocatable, dimension(:, :, :) :: d2u_recv_e_dev
type(dim3), public :: xblocks
type(dim3), public :: xthreads
type(dim3), public :: yblocks
type(dim3), public :: ythreads
type(dim3), public :: zblocks
type(dim3), public :: zthreads

Constructor

public interface cuda_backend_t

  • public function init(mesh, allocator) result(backend)

    Arguments

    Type IntentOptional Attributes Name
    class(mesh_t), intent(inout), target :: mesh
    class(allocator_t), intent(inout), target :: allocator

    Return Value type(cuda_backend_t)


Type-Bound Procedures

procedure, public :: base_init

procedure, public :: get_field_data

  • public subroutine get_field_data(self, data, f, dir)

    Extract data from field f optionally reordering into dir orientation. To output in same orientation as f, use call ...%get_field_data(data, f, f%dir)

    Arguments

    Type IntentOptional Attributes Name
    class(base_backend_t) :: self
    real(kind=dp), intent(out), dimension(:, :, :) :: data

    Output array
    class(field_t), intent(in) :: f

    Field
    integer, intent(in), optional :: dir

    Desired orientation of output array (defaults to Cartesian)

procedure, public :: set_field_data

  • public subroutine set_field_data(self, f, data, dir)

    Arguments

    Type IntentOptional Attributes Name
    class(base_backend_t) :: self
    class(field_t), intent(inout) :: f

    Field
    real(kind=dp), intent(in), dimension(:, :, :) :: data

    Input array
    integer, intent(in), optional :: dir

    Orientation of input array (defaults to Cartesian)

procedure, public :: alloc_tdsops => alloc_cuda_tdsops

  • public subroutine alloc_cuda_tdsops(self, tdsops, dir, operation, scheme, n_halo, from_to, bc_start, bc_end, sym, c_nu, nu0_nu)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    class(tdsops_t), intent(inout), allocatable :: tdsops
    integer, intent(in) :: dir
    character(len=*), intent(in) :: operation
    character(len=*), intent(in) :: scheme
    integer, intent(in), optional :: n_halo
    character(len=*), intent(in), optional :: from_to
    character(len=*), intent(in), optional :: bc_start
    character(len=*), intent(in), optional :: bc_end
    logical, intent(in), optional :: sym
    real(kind=dp), intent(in), optional :: c_nu
    real(kind=dp), intent(in), optional :: nu0_nu

procedure, public :: transeq_x => transeq_x_cuda

procedure, public :: transeq_y => transeq_y_cuda

procedure, public :: transeq_z => transeq_z_cuda

procedure, public :: tds_solve => tds_solve_cuda

procedure, public :: reorder => reorder_cuda

  • public subroutine reorder_cuda(self, u_o, u_i, direction)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    class(field_t), intent(inout) :: u_o
    class(field_t), intent(in) :: u_i
    integer, intent(in) :: direction

procedure, public :: sum_yintox => sum_yintox_cuda

procedure, public :: sum_zintox => sum_zintox_cuda

procedure, public :: vecadd => vecadd_cuda

  • public subroutine vecadd_cuda(self, a, x, b, y)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    real(kind=dp), intent(in) :: a
    class(field_t), intent(in) :: x
    real(kind=dp), intent(in) :: b
    class(field_t), intent(inout) :: y

procedure, public :: scalar_product => scalar_product_cuda

procedure, public :: copy_data_to_f => copy_data_to_f_cuda

  • public subroutine copy_data_to_f_cuda(self, f, data)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t), intent(inout) :: self
    class(field_t), intent(inout) :: f
    real(kind=dp), intent(inout), dimension(:, :, :) :: data

procedure, public :: copy_f_to_data => copy_f_to_data_cuda

  • public subroutine copy_f_to_data_cuda(self, data, f)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t), intent(inout) :: self
    real(kind=dp), intent(out), dimension(:, :, :) :: data
    class(field_t), intent(in) :: f

procedure, public :: init_poisson_fft => init_cuda_poisson_fft

procedure, public :: transeq_cuda_dist

  • public subroutine transeq_cuda_dist(self, du, dv, dw, u, v, w, dirps, blocks, threads)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    class(field_t), intent(inout) :: du
    class(field_t), intent(inout) :: dv
    class(field_t), intent(inout) :: dw
    class(field_t), intent(in) :: u
    class(field_t), intent(in) :: v
    class(field_t), intent(in) :: w
    type(dirps_t), intent(in) :: dirps
    type(dim3), intent(in) :: blocks
    type(dim3), intent(in) :: threads

procedure, public :: transeq_cuda_thom

  • public subroutine transeq_cuda_thom(self, du, dv, dw, u, v, w, dirps)

    Thomas algorithm implementation. So much more easier than the distributed algorithm. It is intended to work only on a single rank so there is no MPI communication.

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    class(field_t), intent(inout) :: du
    class(field_t), intent(inout) :: dv
    class(field_t), intent(inout) :: dw
    class(field_t), intent(in) :: u
    class(field_t), intent(in) :: v
    class(field_t), intent(in) :: w
    type(dirps_t), intent(in) :: dirps

procedure, public :: tds_solve_dist

  • public subroutine tds_solve_dist(self, du, u, tdsops, blocks, threads)

    Arguments

    Type IntentOptional Attributes Name
    class(cuda_backend_t) :: self
    class(field_t), intent(inout) :: du
    class(field_t), intent(in) :: u
    class(tdsops_t), intent(in) :: tdsops
    type(dim3), intent(in) :: blocks
    type(dim3), intent(in) :: threads