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

! Test CUDA Fortran atmoicadd functions available cudadevice module

attributes(global) subroutine test_atomicaddvector_r2()
  real(2), device :: a(2), tmp1(2), tmp2(2)
  tmp1 = atomicAddVector(a, tmp2)
end subroutine

! CHECK-LABEL: func.func @_QPtest_atomicaddvector_r2() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: llvm.atomicrmw fadd %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, vector<2xf16>

attributes(global) subroutine test_atomicaddvector_r4()
  real(4), device :: a(2), tmp1(2), tmp2(2)
  tmp1 = atomicAddVector(a, tmp2)
end subroutine

! CHECK-LABEL: func.func @_QPtest_atomicaddvector_r4() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: llvm.atomicrmw fadd %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, vector<2xf32>

attributes(global) subroutine test_atomicadd_r2x4()
  real(4), device :: a(2), tmp1(2), tmp2(2)
  tmp1 = atomicaddreal4x2(a, tmp2)
end subroutine

! CHECK-LABEL: func.func @_QPtest_atomicadd_r2x4() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: llvm.atomicrmw fadd %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, vector<2xf32>

attributes(global) subroutine test_atomicadd_r4x4()
  real(4), device :: a(4), tmp1(4), tmp2(4)
  tmp1 = atomicaddreal4x4(a, tmp2)
end subroutine

! CHECK-LABEL: func.func @_QPtest_atomicadd_r4x4() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: atom.add.v4.f32