[llvm] [LAA] Rewrite findForkedPointer (NFCI) (PR #140298)

[Ramkumar Ramachandra via llvm-commits]

================
@@ -1088,29 +1088,49 @@ static void findForkedSCEVs(
   }
 }
 
-static SmallVector<PointerIntPair<const SCEV *, 1, bool>>
-findForkedPointer(PredicatedScalarEvolution &PSE,
-                  const DenseMap<Value *, const SCEV *> &StridesMap, Value *Ptr,
-                  const Loop *L) {
-  ScalarEvolution *SE = PSE.getSE();
-  assert(SE->isSCEVable(Ptr->getType()) && "Value is not SCEVable!");
-  SmallVector<PointerIntPair<const SCEV *, 1, bool>> Scevs;
-  findForkedSCEVs(SE, L, Ptr, Scevs, MaxForkedSCEVDepth);
-
-  // For now, we will only accept a forked pointer with two possible SCEVs
-  // that are either SCEVAddRecExprs or loop invariant.
-  if (Scevs.size() == 2 &&
-      (isa<SCEVAddRecExpr>(get<0>(Scevs[0])) ||
-       SE->isLoopInvariant(get<0>(Scevs[0]), L)) &&
-      (isa<SCEVAddRecExpr>(get<0>(Scevs[1])) ||
-       SE->isLoopInvariant(get<0>(Scevs[1]), L))) {
-    LLVM_DEBUG(dbgs() << "LAA: Found forked pointer: " << *Ptr << "\n");
-    LLVM_DEBUG(dbgs() << "\t(1) " << *get<0>(Scevs[0]) << "\n");
-    LLVM_DEBUG(dbgs() << "\t(2) " << *get<0>(Scevs[1]) << "\n");
-    return Scevs;
-  }
-
-  return {{replaceSymbolicStrideSCEV(PSE, StridesMap, Ptr), false}};
+/// Given \p ForkedSCEVs corresponding to \p Ptr, get AddRecs from \p Assume and
+/// \p StridesMap, and return SCEVs that could potentially be checked at runtime
+/// (AddRecs and loop-invariants). Returns an empty range as an early exit.
+static iterator_range<PointerIntPair<const SCEV *, 1, bool> *> getRTCheckPtrs(
+    PredicatedScalarEvolution &PSE, const Loop *L, Value *Ptr,
+    MutableArrayRef<PointerIntPair<const SCEV *, 1, bool>> ForkedSCEVs,
+    const DenseMap<Value *, const SCEV *> &StridesMap, bool Assume) {
+  for (auto &P : ForkedSCEVs) {
+    auto *AR = dyn_cast<SCEVAddRecExpr>(P.getPointer());
+    if (!AR && Assume)
+      AR = PSE.getAsAddRec(Ptr);
+
+    // Call replaceSymbolicStrideSCEV only after PSE.getAsAddRec, because
+    // assumptions might have been added to PSE, resulting in simplifications.
+    const SCEV *S = replaceSymbolicStrideSCEV(PSE, StridesMap, Ptr);
+    auto *SAR = dyn_cast<SCEVAddRecExpr>(S);
+
+    if (auto *PtrVal = SAR ? SAR : AR; PtrVal && PtrVal->isAffine())
+      P.setPointer(PtrVal);
+    else if (!PSE.getSE()->isLoopInvariant(P.getPointer(), L))
+      return {ForkedSCEVs.end(), ForkedSCEVs.end()};
+  }
+
+  // De-duplicate the ForkedSCEVs. If two SCEVs are equal, prefer the SCEV that
+  // doesn't need freeze.
+  auto PtrEq = [](const auto &P, const auto &Q) {
+    return get<0>(P) == get<0>(Q);
+  };
+  auto FreezeLess = [PtrEq](const auto &P, const auto &Q) {
+    return PtrEq(P, Q) && get<1>(P) < get<1>(Q);
+  };
+  stable_sort(ForkedSCEVs, FreezeLess);
+  auto UniqPtrs = make_range(ForkedSCEVs.begin(), unique(ForkedSCEVs, PtrEq));
+
+  if (size(UniqPtrs) == 1) {
+    // FIXME: Is this correct?
+    UniqPtrs.begin()->setInt(false);
+    return UniqPtrs;
+  }
----------------
artagnon wrote:

Okay, I thought about it, and I think this is the correct logic.

https://github.com/llvm/llvm-project/pull/140298