[flang-commits] [clang] [flang] [llvm] [clang][flang][OpenMP] Fix context selector matching and scoring (PR #224431)

[via flang-commits]

================
@@ -5389,6 +5406,14 @@ static mlir::omp::DistributeOp genCompositeDistributeParallelDoSimd(
   ConstructQueue::const_iterator doItem = std::next(parallelItem);
   ConstructQueue::const_iterator simdItem = std::next(doItem);
 
+  // Evaluate DISTRIBUTE clauses before entering its source context and then
+  // PARALLEL, as in the explicit nesting.
+  mlir::omp::DistributeOperands distributeClauseOps;
+  genDistributeClauses(converter, semaCtx, stmtCtx, distributeItem->clauses,
+                       loc, distributeClauseOps);
+  mlir::SaveStateStack<OpenMPContextFrame> distributeContext{
----------------
MattPD wrote:

Confirmed at 1aa333a: The fix works for `do simd`, `parallel do simd`, and `teams distribute parallel do simd`.

`genCompositeDistributeSimd` still lowers the SIMD clauses of `distribute simd` without a DISTRIBUTE frame. When `pred` has a `kind(cpu)` variant and a `vendor(score(3))` variant, `teams distribute simd if(simd: pred(n))` selects the vendor variant, while the explicit nest of the three constructs selects the CPU variant. At the merge base, both forms select the vendor variant. Should `genCompositeDistributeSimd` also enter DISTRIBUTE around its SIMD clauses?

https://github.com/llvm/llvm-project/pull/224431