[Flang][OpenMP] Issues in copyprivate

[harishch4]

Here is a list of test cases that are failing in flang-new but are working in other compilers.
Godbolt link

subroutine test1
    use omp_lib
    implicit none
    character xz
    common/c/xz
    !$omp threadprivate(/c/)
    !$omp parallel num_threads(4)
        !$omp single
        xz = 'c'
        !$omp end single copyprivate(/c/)
    !$omp end parallel
end

error: COMMON block must be declared in the same scoping unit in which the OpenMP directive or clause appears
          !$omp end single copyprivate(/c/)

!fails in gfortran but all other compilers accepting this

subroutine test2
    real z
    !$omp parallel private(z) num_threads(2)
    !$omp single
        z = -2
    !$omp end single copyprivate(z,z)
    !$omp end parallel
end 

subroutine test3
    use omp_lib
    !$omp parallel private(xz) num_threads(2)
    xz = f()
    !$omp end parallel
contains
    function f() result(rz)
        !$omp single
        rz = 42
        !$omp end single copyprivate(rz)
    end function
end subroutine

error: COPYPRIVATE variable 'rz' is not PRIVATE or THREADPRIVATE in outer context
          !$omp end single copyprivate(rz)
                                       ^^

subroutine test4
    use omp_lib
    implicit none
    call test
contains
    subroutine test
        real, pointer :: xz
        !$omp parallel private(xz) num_threads(2)
        xz => fz()
        !$omp end parallel
    end subroutine
    function fz()
        real, pointer :: fz
        nullify (fz)
        !$omp single
        allocate (fz)
        fz = 42
        !$omp end single copyprivate(fz)
    end function
end subroutine

 error: COPYPRIVATE variable 'fz' is not PRIVATE or THREADPRIVATE in outer context
          !$omp end single copyprivate(fz)
                                       ^^

[luporl]

#88430 was merged, fixing test1.

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: None (harishch4)

Here is a list of test cases that are failing in flang-new but are working in other compilers. Godbolt [link](https://godbolt.org/z/b3hKvKjEj) ``` subroutine test1 use omp_lib implicit none character xz common/c/xz !$omp threadprivate(/c/) !$omp parallel num_threads(4) !$omp single xz = 'c' !$omp end single copyprivate(/c/) !$omp end parallel end

error: COMMON block must be declared in the same scoping unit in which the OpenMP directive or clause appears
!$omp end single copyprivate(/c/)

!fails in gfortran but all other compilers accepting this

subroutine test2
real z
!$omp parallel private(z) num_threads(2)
!$omp single
z = -2
!$omp end single copyprivate(z,z)
!$omp end parallel
end

subroutine test3
use omp_lib
!$omp parallel private(xz) num_threads(2)
xz = f()
!$omp end parallel
contains
function f() result(rz)
!$omp single
rz = 42
!$omp end single copyprivate(rz)
end function
end subroutine

error: COPYPRIVATE variable 'rz' is not PRIVATE or THREADPRIVATE in outer context
!$omp end single copyprivate(rz)
^^

subroutine test4
use omp_lib
implicit none
call test
contains
subroutine test
real, pointer :: xz
!$omp parallel private(xz) num_threads(2)
xz => fz()
!$omp end parallel
end subroutine
function fz()
real, pointer :: fz
nullify (fz)
!$omp single
allocate (fz)
fz = 42
!$omp end single copyprivate(fz)
end function
end subroutine

error: COPYPRIVATE variable 'fz' is not PRIVATE or THREADPRIVATE in outer context
!$omp end single copyprivate(fz)
^^

</details>

[luporl]

#95799 will fix all the remaining issues reported here.
However, in the case of test4 there is also an issue in lowering:

error: loc("test4.f90":15:15): 'fir.convert' op invalid type conversion'f32' / '!fir.box<!fir.ptr<f32>>'
error: Lowering to LLVM IR failed
error: loc("test4.f90":15:15): cannot be converted to LLVM IR: missing `LLVMTranslationDialectInterface` registration for dialect for op: func.func
error: failed to create the LLVM module

I'll open a separate issue for it.

[luporl]

Opened #95801 to track the lowering issue in copyprivate.

[luporl]

Fixed by #95799.

[luporl]

#95799 was reverted. This issue should be fixed by #101009 now.