[flang-commits] [flang] [flang][OpenMP] Identify DO loops affected by loop-associated construct (PR #191719)

[Krzysztof Parzyszek via flang-commits]

================
@@ -1203,6 +1203,97 @@ std::optional<int64_t> GetMinimumSequenceCount(
   return GetMinimumSequenceCount(std::nullopt, std::nullopt);
 }
 
+/// Collect the DO loops that are affected directly by the given loop
+/// transformation. Not all DO loops nested in the associated nest are
+/// affected by the top-level loop transformation, e.g.
+///
+/// !$omp do collapse(5)                           | [2]
+/// !$omp tile sizes(2, 2)  | [1]                  | <- nest of 4 loops
+/// do i = 1, 10            | <- affected by TILE  |    generated by TILE
+///   do j = 1, 10          | <-                   |
+///     do k = 1, 10                               | <- affected by DO
+///     end do
+///   end do
+/// end do
+///
+/// The two DO loops (i and j) in [1] are affected by the TILE construct.
+/// The k DO loop is affected by the DO construct [2].
+/// For the top-level DO COLLAPSE(5) construct, the k loop is the only
+/// directly affected loop.
+std::optional<std::vector<const parser::DoConstruct *>> CollectAffectedDoLoops(
+    const parser::OpenMPLoopConstruct &x, unsigned version,
+    SemanticsContext *semaCtx) {
+  std::vector<const parser::DoConstruct *> result;
+  const parser::OmpDirectiveSpecification &spec{x.BeginDir()};
+
+  auto [depth, _]{GetAffectedNestDepthWithReason(spec, version, semaCtx)};
+
+  // If the depth is absent, then there is some issue. Leave it alone here,
+  // and let the semantic checks diagnose the problem.
+  if (!depth) {
+    return std::nullopt;
+  }
+  if (*depth.value <= 0) {
----------------
kparzysz wrote:

No.  I will change this check to `==` and add assertion to make sure it's not negative.

https://github.com/llvm/llvm-project/pull/191719