[flang-commits] [flang] [flang][OpenMP] incorrect handling for local variable in OpenMP parallel workshare firstprivate(P) (PR #195616)

[Tom Eccles via flang-commits]

================
@@ -370,10 +387,34 @@ static void parallelizeRegion(Region &sourceRegion, Region &targetRegion,
     SmallVector<Value> copyPrivate;
     bool allParallelized = true;
 
+    // "firstprivate" pointer initialization creates: (1) alloca, (2) store
+    // null box, (3) copy original. If step (2) is duplicated into the
+    // parallel block, it runs after initialization of the private copy and
+    // overwrites the pointer descriptor with null, causing a segfault on
+    // dereference.
+    SmallPtrSet<Value, 4> hoistedCopyprivateAllocas;
+
     for (Operation &op : llvm::make_range(sr.begin, sr.end)) {
       if (isSafeToParallelize(&op)) {
         singleBuilder.clone(op, singleMapping);
-        if (llvm::all_of(op.getOperands(), [&](Value opr) {
+        // Check if this operation writes to a hoisted copyprivate alloca.
+        // Such stores must stay only in the single block; the copyprivate
+        // mechanism handles broadcasting the final value to all threads.
+        bool writesToCopyprivateAlloca = false;
+        if (!hoistedCopyprivateAllocas.empty()) {
+          if (auto memEffects = dyn_cast<MemoryEffectOpInterface>(&op)) {
+            SmallVector<MemoryEffects::EffectInstance> effects;
+            memEffects.getEffects(effects);
+            writesToCopyprivateAlloca =
+                llvm::any_of(effects, [&](const auto &eff) {
+                  return isa<MemoryEffects::Write>(eff.getEffect()) &&
+                         eff.getValue() &&
----------------
tblah wrote:

If the effect has no value then we have to assume that it could write to any value.

https://github.com/llvm/llvm-project/pull/195616