[flang][openmp] Incorrect diagnostic on Openmp Atomic update statement

[DanielCChen]

Consider the following test case

    integer (kind = 4) :: a
    integer :: c
    a = 1
    c = 3
    d = 4
!$omp atomic
    a = a + c + d
end

Flang now issues an error

./t.f:7:5: error: Atomic update statement should be of form `a = a operator expr` OR `a = expr operator a`
      a = a + c + d
      ^

The code is conforming to the standard.

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: Daniel Chen (DanielCChen)

Consider the following test case ``` integer (kind = 4) :: a integer :: c a = 1 c = 3 d = 4 !$omp atomic a = a + c + d end ```

Flang now issues an error

./t.f:7:5: error: Atomic update statement should be of form `a = a operator expr` OR `a = expr operator a`
      a = a + c + d
      ^

The code is conforming to the standard.

[eugeneepshteyn]

Looking at OpenMP 5.2 spec, section "4.3.1.3 OpenMP Atomic Structured Blocks", page 81, lines 10-13. The following forms are allowed:

x = x operator expr
x = expr operator x

expr here is a scalar expression, and in the reproducer a = a + c + d, where c + d is a scalar expression. So yes, I agree that this should not result in an error.

[eugeneepshteyn]

In OmpStructureChecker::IsOperatorValid(), the code seems to only support cases like a = a + b and a = b + a. Will need to update with more extensive scalar expression support.

[eugeneepshteyn]

(Echoing Peter on the other OpenMP issue)

Kiran, please reassign to whomever takes care of OpenMP parsing.