TheRock/compiler/amd-llvm/flang/test/Lower/CUDA/cuda-data-attribute.cuf

! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s
! RUN: bbc -emit-hlfir -fcuda %s -o - | fir-opt -convert-hlfir-to-fir | FileCheck %s --check-prefix=FIR

! Test lowering of CUDA attribute on variables.

module cuda_var

  type :: t1
    integer :: a
  end type

  real, constant :: mod_a_rc
  real, device :: mod_b_ra
  real, allocatable, managed :: mod_c_rm
  integer, device, dimension(10) :: mod_d_i_init = (/ (i, i = 1, 10) /)
  real, device, dimension(10) :: mod_d_rinit = (/ (i, i = 1, 10) /)
  real, allocatable, pinned :: mod_d_rp
  type(t1), device :: mod_d_t(2)

contains

subroutine local_var_attrs
  real, device :: rd
  real, allocatable, managed :: rm
  real, allocatable, pinned :: rp
  real, unified :: ru
end subroutine

! CHECK-LABEL: func.func @_QMcuda_varPlocal_var_attrs()
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {data_attr = #cuf.cuda<device>, uniq_name = "_QMcuda_varFlocal_var_attrsErd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {data_attr = #cuf.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {data_attr = #cuf.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {data_attr = #cuf.cuda<unified>, uniq_name = "_QMcuda_varFlocal_var_attrsEru"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)


! FIR: %{{.*}} = fir.declare %{{.*}} {data_attr = #cuf.cuda<device>, uniq_name = "_QMcuda_varFlocal_var_attrsErd"} : (!fir.ref<f32>) -> !fir.ref<f32>
! FIR: %{{.*}} = fir.declare %{{.*}} {data_attr = #cuf.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>
! FIR: %{{.*}} = fir.declare %{{.*}} {data_attr = #cuf.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>
! FIR: %{{.*}} = fir.declare %{{.*}} {data_attr = #cuf.cuda<unified>, uniq_name = "_QMcuda_varFlocal_var_attrsEru"} : (!fir.ref<f32>) -> !fir.ref<f32>

subroutine dummy_arg_device(dd)
  real, device :: dd
end subroutine
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_device(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<f32> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "dd"}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] dummy_scope %{{[0-9]+}} arg {{[0-9]+}} {data_attr = #cuf.cuda<device>, uniq_name = "_QMcuda_varFdummy_arg_deviceEdd"} : (!fir.ref<f32>, !fir.dscope) -> (!fir.ref<f32>, !fir.ref<f32>)

subroutine dummy_arg_managed(dm)
  real, allocatable, managed :: dm
end subroutine
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_managed(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<!fir.box<!fir.heap<f32>>> {cuf.data_attr = #cuf.cuda<managed>, fir.bindc_name = "dm"}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] dummy_scope %{{[0-9]+}} arg {{[0-9]+}} {data_attr = #cuf.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFdummy_arg_managedEdm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.dscope) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)

subroutine dummy_arg_pinned(dp)
  real, allocatable, pinned :: dp
end subroutine
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_pinned(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<!fir.box<!fir.heap<f32>>> {cuf.data_attr = #cuf.cuda<pinned>, fir.bindc_name = "dp"}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] dummy_scope %{{[0-9]+}} arg {{[0-9]+}} {data_attr = #cuf.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFdummy_arg_pinnedEdp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.dscope) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)

subroutine dummy_arg_unified(du)
  real, unified :: du
end subroutine
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_unified(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<f32> {cuf.data_attr = #cuf.cuda<unified>, fir.bindc_name = "du"})
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] dummy_scope %{{[0-9]+}} arg {{[0-9]+}} {data_attr = #cuf.cuda<unified>, uniq_name = "_QMcuda_varFdummy_arg_unifiedEdu"} : (!fir.ref<f32>, !fir.dscope) -> (!fir.ref<f32>, !fir.ref<f32>)

subroutine cuda_alloc_free(n)
  integer :: n
  real, device :: a(10)
  integer, unified :: u
  real, managed :: b(n)
end

! CHECK-LABEL: func.func @_QMcuda_varPcuda_alloc_free
! CHECK: %[[ALLOC_A:.*]] = cuf.alloc !fir.array<10xf32> {bindc_name = "a", data_attr = #cuf.cuda<device>, uniq_name = "_QMcuda_varFcuda_alloc_freeEa"} -> !fir.ref<!fir.array<10xf32>>
! CHECK: %[[SHAPE:.*]] = fir.shape %c10{{.*}} : (index) -> !fir.shape<1>
! CHECK: %[[DECL_A:.*]]:2 = hlfir.declare %[[ALLOC_A]](%[[SHAPE]]) {data_attr = #cuf.cuda<device>, uniq_name = "_QMcuda_varFcuda_alloc_freeEa"} : (!fir.ref<!fir.array<10xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xf32>>, !fir.ref<!fir.array<10xf32>>)

! CHECK: %[[ALLOC_U:.*]] = cuf.alloc i32 {bindc_name = "u", data_attr = #cuf.cuda<unified>, uniq_name = "_QMcuda_varFcuda_alloc_freeEu"} -> !fir.ref<i32>
! CHECK: %[[DECL_U:.*]]:2 = hlfir.declare %[[ALLOC_U]] {data_attr = #cuf.cuda<unified>, uniq_name = "_QMcuda_varFcuda_alloc_freeEu"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)

! CHECK: %[[ALLOC_B:.*]] = cuf.alloc !fir.array<?xf32>, %{{.*}} : index {bindc_name = "b", data_attr = #cuf.cuda<managed>, uniq_name = "_QMcuda_varFcuda_alloc_freeEb"} -> !fir.ref<!fir.array<?xf32>>
! CHECK: %[[SHAPE:.*]] = fir.shape %{{.*}} : (index) -> !fir.shape<1>
! CHECK: %[[DECL_B:.*]]:2 = hlfir.declare %[[ALLOC_B]](%[[SHAPE]]) {data_attr = #cuf.cuda<managed>, uniq_name = "_QMcuda_varFcuda_alloc_freeEb"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.array<?xf32>>)

! CHECK: cuf.free %[[DECL_B]]#1 : !fir.ref<!fir.array<?xf32>> {data_attr = #cuf.cuda<managed>}
! CHECK: cuf.free %[[DECL_U]]#0 : !fir.ref<i32> {data_attr = #cuf.cuda<unified>}
! CHECK: cuf.free %[[DECL_A]]#0 : !fir.ref<!fir.array<10xf32>> {data_attr = #cuf.cuda<device>}

subroutine dummy(x)
  real, target, device :: x
end subroutine

! CHECK: func.func @_QMcuda_varPdummy
! CHECK-NOT: cuf.free

end module

subroutine craypointer()
  real, device :: x(*); pointer(px, x)
end

! CHECK-LABEL: func.func @_QPcraypointer()
! CHECK: %{{.*}} = cuf.alloc !fir.box<!fir.ptr<!fir.array<?xf32>>> {data_attr = #cuf.cuda<device>} -> !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>

attributes(global) subroutine attr_value(a)
  integer :: a
  attributes(value) :: a
end subroutine

! CHECK-LABEL: func.func @_QPattr_value(%{{.*}}: i32 {fir.bindc_name = "a"})
! CHECK: hlfir.declare %{{.*}} {fortran_attrs = #fir.var_attrs<value>, uniq_name = "_QFattr_valueEa"}

! CHECK: fir.global @_QMcuda_varEmod_a_rc {data_attr = #cuf.cuda<constant>} : f32 
! CHECK: fir.global @_QMcuda_varEmod_b_ra {data_attr = #cuf.cuda<device>} : f32
! CHECK: fir.global @_QMcuda_varEmod_c_rm {data_attr = #cuf.cuda<managed>} : !fir.box<!fir.heap<f32>>

! CHECK: fir.global @_QMcuda_varEmod_d_i_init(dense<[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]> : tensor<10xi32>) {alignment = 64 : i64, data_attr = #cuf.cuda<device>} : !fir.array<10xi32>
! CHECK: fir.global @_QMcuda_varEmod_d_rinit(dense<[{{.*}}]> : tensor<10xf32>) {alignment = 64 : i64, data_attr = #cuf.cuda<device>} : !fir.array<10xf32>
! CHECK: fir.global @_QMcuda_varEmod_d_rp {data_attr = #cuf.cuda<pinned>} : !fir.box<!fir.heap<f32>>
! CHECK: fir.global @_QMcuda_varEmod_d_t {alignment = 64 : i64, data_attr = #cuf.cuda<device>} : !fir.array<2x!fir.type<_QMcuda_varTt1{a:i32}>>