[PATCH] D104863: [LSV] Teach LSV to handle atomic ops.

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp:700-714
+  auto IsAtomicBarrier = [](Instruction *I) {
+    if (!I->isAtomic())
+      return false;
+    if (auto *LD = dyn_cast<LoadInst>(I))
+      return isStrongerThanMonotonic(LD->getOrdering());
+    if (auto *ST = dyn_cast<StoreInst>(I))
+      return isStrongerThanMonotonic(ST->getOrdering());
----------------
It seems like this should be a global utility function somewhere


================
Comment at: llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp:800-810
+          auto getPointerOperand = [](const Instruction *I) -> const Value * {
+            if (auto *Load = dyn_cast<LoadInst>(I))
+              return Load->getPointerOperand();
+            if (auto *Store = dyn_cast<StoreInst>(I))
+              return Store->getPointerOperand();
+            if (auto *RMW = dyn_cast<AtomicRMWInst>(I))
+              return RMW->getPointerOperand();
----------------
I think there's a utility for this osmewhere already (that also will handle the intrinsic calls)


================
Comment at: llvm/test/Transforms/LoadStoreVectorizer/atomic.ll:8-9
+; CHECK-LABEL: t0
+; CHECK: load <4 x float>
+; CHECK: store <4 x float>
+define void @t0(float* noalias %p, i32* noalias %q) {
----------------
Should check more, should show the relation to the atomics


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104863/new/

https://reviews.llvm.org/D104863