!fixu update on top of main

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h

@@ -571,15 +571,11 @@ class LoopVectorizationPlanner {
   /// Update loop metadata and profile info for both the scalar remainder loop
   /// and \p VectorLoop, if it exists. Keeps all loop hints from the original
   /// loop on the vector loop and replaces vectorizer-specific metadata
-void updateLoopMetadata(Loop *VectorLoop, Loop *OrigLoop, ScalarEvolution &SE,
-                               OptimizationRemarkEmitter *ORE,
-                               VPBasicBlock *HeaderVPBB, VPlan &Plan,
-                               const TargetTransformInfo &TTI,
-                               bool VectorizingEpilogue, MDNode *LID,
-                               std::optional<unsigned> OrigAverageTripCount,
-                               unsigned OrigLoopInvocationWeight,
-                               ElementCount BestVF,
-                               unsigned EstimatedVFxUF , bool DisableRuntimeUnroll);
+  void updateLoopMetadata(Loop *VectorLoop, VPBasicBlock *HeaderVPBB,
+                          VPlan &Plan, bool VectorizingEpilogue, MDNode *LID,
+                          std::optional<unsigned> OrigAverageTripCount,
+                          unsigned OrigLoopInvocationWeight,
+                          unsigned EstimatedVFxUF, bool DisableRuntimeUnroll);
 
 protected:
   /// Build VPlans for power-of-2 VF's between \p MinVF and \p MaxVF inclusive,

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -7283,9 +7283,11 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
   updateLoopMetadata(
       HeaderVPBB ? LI->getLoopFor(State.CFG.VPBB2IRBB.lookup(HeaderVPBB))
                  : nullptr,
-      OrigLoop, *PSE.getSE(), ORE, HeaderVPBB, BestVPlan, TTI,
-      VectorizingEpilogue, LID, OrigAverageTripCount,
-      OrigLoopInvocationWeight, BestVF, estimateElementCount(BestVF * BestUF, CM.getVScaleForTuning()), DisableRuntimeUnroll);
+      HeaderVPBB, BestVPlan, VectorizingEpilogue, LID, OrigAverageTripCount,
+      OrigLoopInvocationWeight,
+      estimateElementCount(BestVF * BestUF, CM.getVScaleForTuning()),
+      DisableRuntimeUnroll);
+
   // 3. Fix the vectorized code: take care of header phi's, live-outs,
   //    predication, updating analyses.
   ILV.fixVectorizedLoop(State);

llvm/lib/Transforms/Vectorize/VPlan.cpp

@@ -1677,19 +1677,16 @@ static void addRuntimeUnrollDisableMetaData(Loop *L) {
   }
 }
 
-void LoopVectorizationPlanner::updateLoopMetadata(Loop *VectorLoop, Loop *OrigLoop, ScalarEvolution &SE,
-                               OptimizationRemarkEmitter *ORE,
-                               VPBasicBlock *HeaderVPBB, VPlan &Plan,
-                               const TargetTransformInfo &TTI,
-                               bool VectorizingEpilogue, MDNode *LID,
-                               std::optional<unsigned> OrigAverageTripCount,
-                               unsigned OrigLoopInvocationWeight,
-                               ElementCount BestVF,
-                               unsigned EstimatedVFxUF, bool DisableRuntimeUnroll) {
-  MDNode *LID = OrigLoop->getLoopID();
+void LoopVectorizationPlanner::updateLoopMetadata(
+    Loop *VectorLoop, VPBasicBlock *HeaderVPBB, VPlan &Plan,
+    bool VectorizingEpilogue, MDNode *LID,
+    std::optional<unsigned> OrigAverageTripCount,
+    unsigned OrigLoopInvocationWeight, unsigned EstimatedVFxUF,
+    bool DisableRuntimeUnroll) {
   // Update the metadata of the scalar loop. Skip the update when vectorizing
-  // the epilogue loop, to ensure it is only updated once.
-  if (!VectorizingEpilogue) {
+  // the epilogue loop, to ensure it is only updated once, or when the became
+  // unreachable.
+  if (Plan.getScalarPreheader()->hasPredecessors() && !VectorizingEpilogue) {
     std::optional<MDNode *> RemainderLoopID = makeFollowupLoopID(
         LID, {LLVMLoopVectorizeFollowupAll, LLVMLoopVectorizeFollowupEpilogue});
     if (RemainderLoopID) {
@@ -1759,10 +1756,23 @@ void LoopVectorizationPlanner::updateLoopMetadata(Loop *VectorLoop, Loop *OrigLo
   // For scalable vectorization we can't know at compile time how many
   // iterations of the loop are handled in one vector iteration, so instead
   // use the value of vscale used for tuning.
-  setProfileInfoAfterUnrolling(OrigLoop, VectorLoop, OrigLoop, EstimatedVFxUF);
+  if (OrigAverageTripCount) {
+    // Calculate number of iterations in unrolled loop.
+    unsigned AverageVectorTripCount = *OrigAverageTripCount / EstimatedVFxUF;
+    // Calculate number of iterations for remainder loop.
+    unsigned RemainderAverageTripCount = *OrigAverageTripCount % EstimatedVFxUF;
+
+    if (HeaderVPBB) {
+      setLoopEstimatedTripCount(VectorLoop, AverageVectorTripCount,
+                                OrigLoopInvocationWeight);
+    }
+    if (Plan.getScalarPreheader()->hasPredecessors()) {
+      setLoopEstimatedTripCount(OrigLoop, RemainderAverageTripCount,
+                                OrigLoopInvocationWeight);
+    }
+  }
 }
 
->>>>>>> main
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void LoopVectorizationPlanner::printPlans(raw_ostream &O) {
   if (VPlans.empty()) {