test/Lower/CUDA/cuda-data-transfer.cuf

! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s

! Test CUDA Fortran data transfer using assignment statements.

module mod1
  type :: t1
    integer :: i
  end type
end

subroutine sub1()
  integer, device :: m
  integer, device :: adev(10)
  integer :: i, ahost(10), bhost(10)

  m = 1 + i

  m = 1

  adev = ahost

  adev = ahost + 1

  adev(1:5) = ahost(1:5)

  adev = ahost + bhost

end

! CHECK-LABEL: func.func @_QPsub1()

! CHECK: %[[ADEV:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub1Eadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[AHOST:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {uniq_name = "_QFsub1Eahost"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[I:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFsub1Ei"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[M:.*]]:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub1Em"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)

! CHECK: %[[C1:.*]] = arith.constant 1 : i32
! CHECK: %[[LOADED_I:.*]] = fir.load %[[I]]#0 : !fir.ref<i32>
! CHECK: %[[ADD:.*]] = arith.addi %[[C1]], %[[LOADED_I]] : i32
! CHECK: %[[ASSOC:.*]]:3 = hlfir.associate %[[ADD]] {uniq_name = ".cuf_host_tmp"} : (i32) -> (!fir.ref<i32>, !fir.ref<i32>, i1)
! CHECK: fir.cuda_data_transfer %[[ASSOC]]#0 to %[[M]]#0 {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<i32>, !fir.ref<i32>
! CHECK: hlfir.end_associate %[[ASSOC]]#1, %[[ASSOC]]#2 : !fir.ref<i32>, i1

! CHECK: %[[C1:.*]] = arith.constant 1 : i32
! CHECK: %[[ASSOC:.*]]:3 = hlfir.associate %[[C1]] {uniq_name = ".cuf_host_tmp"} : (i32) -> (!fir.ref<i32>, !fir.ref<i32>, i1)
! CHECK: fir.cuda_data_transfer %[[ASSOC]]#0 to %[[M]]#0 {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<i32>, !fir.ref<i32>
! CHECK: hlfir.end_associate %[[ASSOC]]#1, %[[ASSOC]]#2 : !fir.ref<i32>, i1

! CHECK: fir.cuda_data_transfer %[[AHOST]]#0 to %[[ADEV]]#0 {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>

! CHECK: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<10xi32> {
! CHECK: %[[ASSOC:.*]]:3 = hlfir.associate %[[ELEMENTAL]](%{{.*}}) {uniq_name = ".cuf_host_tmp"} : (!hlfir.expr<10xi32>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>, i1)
! CHECK: fir.cuda_data_transfer %[[ASSOC]]#0 to %[[ADEV]]#0 {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>
! CHECK: hlfir.end_associate %[[ASSOC]]#1, %[[ASSOC]]#2 : !fir.ref<!fir.array<10xi32>>, i1

! CHECK: %[[DES_AHOST:.*]] = hlfir.designate %[[AHOST]]#0 (%c1{{.*}}:%c5{{.*}}:%c1{{.*}})  shape %{{.*}} : (!fir.ref<!fir.array<10xi32>>, index, index, index, !fir.shape<1>) -> !fir.ref<!fir.array<5xi32>>
! CHECK: %[[DES_ADEV:.*]] = hlfir.designate %[[ADEV]]#0 (%c1{{.*}}:%c5{{.*}}:%c1{{.*}})  shape %{{.*}} : (!fir.ref<!fir.array<10xi32>>, index, index, index, !fir.shape<1>) -> !fir.ref<!fir.array<5xi32>>
! CHECK: fir.cuda_data_transfer %[[DES_AHOST]] to %[[DES_ADEV]] {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<!fir.array<5xi32>>, !fir.ref<!fir.array<5xi32>>

! CHECK: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<10xi32>
! CHECK: %[[ASSOC:.*]]:3 = hlfir.associate %[[ELEMENTAL]](%{{.*}}) {uniq_name = ".cuf_host_tmp"} : (!hlfir.expr<10xi32>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>, i1)
! CHECK: fir.cuda_data_transfer %[[ASSOC]]#0 to %[[ADEV]]#0 {transfer_kind = #fir.cuda_transfer<host_device>} : !fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>
! CHECK: hlfir.end_associate %[[ASSOC]]#1, %[[ASSOC]]#2 : !fir.ref<!fir.array<10xi32>>, i1

subroutine sub2()
  integer, device :: m
  integer, device :: adev(10), bdev(10)
  integer :: i, ahost(10), bhost(10)

  ahost = adev

  i = m

  ahost(1:5) = adev(1:5)

  bdev = adev

  ! Implicit data transfer of adev before evaluation.
  bhost = ahost + adev

end

! CHECK-LABEL: func.func @_QPsub2()
! CHECK: %[[ADEV:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Eadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[AHOST:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {uniq_name = "_QFsub2Eahost"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[BDEV:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Ebdev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[BHOST:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {uniq_name = "_QFsub2Ebhost"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[I:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFsub2Ei"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[M:.*]]:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Em"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: fir.cuda_data_transfer %[[ADEV]]#0 to %[[AHOST]]#0 {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>
! CHECK: fir.cuda_data_transfer %[[M]]#0 to %[[I]]#0 {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<i32>, !fir.ref<i32>

! CHECK: %[[DES_ADEV:.*]] = hlfir.designate %[[ADEV]]#0 (%{{.*}}:%{{.*}}:%{{.*}})  shape %{{.*}} : (!fir.ref<!fir.array<10xi32>>, index, index, index, !fir.shape<1>) -> !fir.ref<!fir.array<5xi32>>
! CHECK: %[[DES_AHOST:.*]] = hlfir.designate %[[AHOST]]#0 (%{{.*}}:%{{.*}}:%{{.*}})  shape %{{.*}} : (!fir.ref<!fir.array<10xi32>>, index, index, index, !fir.shape<1>) -> !fir.ref<!fir.array<5xi32>>
! CHECK: fir.cuda_data_transfer %[[DES_ADEV]] to %[[DES_AHOST]] {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<!fir.array<5xi32>>, !fir.ref<!fir.array<5xi32>>

! CHECK: fir.cuda_data_transfer %[[ADEV]]#0 to %[[BDEV]]#0 {transfer_kind = #fir.cuda_transfer<device_device>} : !fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>

! CHECK: %[[TEMP:.*]] = fir.allocmem !fir.array<10xi32> {bindc_name = ".tmp", uniq_name = ""}
! CHECK: %[[DECL_TEMP:.*]]:2 = hlfir.declare %[[TEMP]](%{{.*}}) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.heap<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>)
! CHECK: %[[ADEV_TEMP:.*]]:2 = hlfir.declare %[[DECL_TEMP]]#1(%{{.*}}) {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Eadev"} : (!fir.heap<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.heap<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>)
! CHECK: fir.cuda_data_transfer %[[ADEV]]#1 to %[[DECL_TEMP]]#0 {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>
! CHECK: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<10xi32>
! CHECK: hlfir.assign %[[ELEMENTAL]] to %[[BHOST]]#0 : !hlfir.expr<10xi32>, !fir.ref<!fir.array<10xi32>>
! CHECK: fir.freemem %[[DECL_TEMP]]#0 : !fir.heap<!fir.array<10xi32>>

subroutine sub3()
  use mod1
  type(t1), device :: t
  integer :: ahost(10), bhost(10)

  bhost = ahost + t%i
end

! CHECK-LABEL: func.func @_QPsub3()
! CHECK: %[[TMP:.*]] = fir.alloca !fir.type<_QMmod1Tt1{i:i32}> {bindc_name = ".tmp"}
! CHECK: %[[AHOST:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {uniq_name = "_QFsub3Eahost"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[BHOST:.*]]:2 = hlfir.declare %{{.*}}(%{{.*}}) {uniq_name = "_QFsub3Ebhost"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
! CHECK: %[[T:.*]]:2 = hlfir.declare %7 {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub3Et"} : (!fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>) -> (!fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>, !fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>)
! CHECK: %[[TMP_DECL:.*]]:2 = hlfir.declare %0 {uniq_name = ".tmp"} : (!fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>) -> (!fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>, !fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>)
! CHECK: fir.cuda_data_transfer %[[T]]#1 to %[[TMP_DECL]]#0 {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>, !fir.ref<!fir.type<_QMmod1Tt1{i:i32}>>


! Check that fir.cuda_data_transfer are not generated within cuf kernel
subroutine sub4()
  integer, parameter :: n = 10
  real, device :: adev(n)
  real :: ahost(n)
  real :: b
  integer :: i

  adev = ahost
  !$cuf kernel do <<<*,*>>>
  do i = 1, n
    adev(i) = adev(i) + b
  enddo
end subroutine

! CHECK-LABEL: func.func @_QPsub4()
! CHECK: fir.cuda_data_transfer
! CHECK: fir.cuda_kernel<<<*, *>>>
! CHECK-NOT: fir.cuda_data_transfer
! CHECK: hlfir.assign

attributes(global) subroutine sub5(a)
  integer, device :: a
  integer :: i
  a = i
end subroutine

! CHECK-LABEL: func.func @_QPsub5
! CHECK-NOT: fir.cuda_data_transfer

attributes(host,device) subroutine sub6(a)
  integer, device :: a
  integer :: i
  a = i
end subroutine

! CHECK-LABEL: func.func @_QPsub6
! CHECK: fir.cuda_data_transfer