Build the Hopfield network incrementally when splitting global live ranges.

It is common for large live ranges to have few basic blocks with register uses
and many live-through blocks without any uses. This approach grows the Hopfield
network incrementally around the use blocks, completely avoiding checking
interference for some through blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129188 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/CodeGen/EdgeBundles.h

@@ -16,6 +16,7 @@
 #ifndef LLVM_CODEGEN_EDGEBUNDLES_H
 #define LLVM_CODEGEN_EDGEBUNDLES_H
 
+#include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/IntEqClasses.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 
@@ -29,6 +30,9 @@ class EdgeBundles : public MachineFunctionPass {
   ///   2*BB->getNumber()+1 -> Outgoing bundle.
   IntEqClasses EC;
 
+  /// Blocks - Map each bundle to a list of basic block numbers.
+  SmallVector<SmallVector<unsigned, 8>, 4> Blocks;
+
 public:
   static char ID;
   EdgeBundles() : MachineFunctionPass(ID) {}
@@ -40,6 +44,9 @@ public:
   /// getNumBundles - Return the total number of bundles in the CFG.
   unsigned getNumBundles() const { return EC.getNumClasses(); }
 
+  /// getBlocks - Return an array of blocks that are connected to Bundle.
+  ArrayRef<unsigned> getBlocks(unsigned Bundle) { return Blocks[Bundle]; }
+
   /// getMachineFunction - Return the last machine function computed.
   const MachineFunction *getMachineFunction() const { return MF; }
 

lib/CodeGen/EdgeBundles.cpp

@@ -53,6 +53,19 @@ bool EdgeBundles::runOnMachineFunction(MachineFunction &mf) {
   EC.compress();
   if (ViewEdgeBundles)
     view();
+
+  // Compute the reverse mapping.
+  Blocks.clear();
+  Blocks.resize(getNumBundles());
+
+  for (unsigned i = 0, e = MF->getNumBlockIDs(); i != e; ++i) {
+    unsigned b0 = getBundle(i, 0);
+    unsigned b1 = getBundle(i, 1);
+    Blocks[b0].push_back(i);
+    if (b1 != b0)
+      Blocks[b1].push_back(i);
+  }
+
   return false;
 }
 
@@ -82,5 +95,3 @@ raw_ostream &llvm::WriteGraph(raw_ostream &O, const EdgeBundles &G,
   O << "}\n";
   return O;
 }
-
-

lib/CodeGen/RegAllocGreedy.cpp

@@ -22,6 +22,7 @@
 #include "SpillPlacement.h"
 #include "SplitKit.h"
 #include "VirtRegMap.h"
+#include "llvm/ADT/SparseBitVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Function.h"
@@ -126,9 +127,8 @@ class RAGreedy : public MachineFunctionPass,
   /// All basic blocks where the current register has uses.
   SmallVector<SpillPlacement::BlockConstraint, 8> SplitConstraints;
 
-  /// All basic blocks where the current register is live-through and
+  /// Live-through blocks that have already been added to SpillPlacer.
-  /// interference free.
+  SparseBitVector<> ActiveThroughBlocks;
-  SmallVector<unsigned, 8> TransparentBlocks;
 
   /// Global live range splitting candidate info.
   struct GlobalSplitCandidate {
@@ -173,7 +173,9 @@ private:
   void LRE_WillShrinkVirtReg(unsigned);
   void LRE_DidCloneVirtReg(unsigned, unsigned);
 
-  bool addSplitConstraints(unsigned, float&);
+  bool addSplitConstraints(InterferenceCache::Cursor, float&);
+  void addThroughConstraints(InterferenceCache::Cursor, ArrayRef<unsigned>);
+  void growRegion(InterferenceCache::Cursor);
   float calcGlobalSplitCost(unsigned, const BitVector&);
   void splitAroundRegion(LiveInterval&, unsigned, const BitVector&,
                          SmallVectorImpl<LiveInterval*>&);
@@ -417,9 +419,9 @@ unsigned RAGreedy::tryEvict(LiveInterval &VirtReg,
 /// interference pattern in Physreg and its aliases. Add the constraints to
 /// SpillPlacement and return the static cost of this split in Cost, assuming
 /// that all preferences in SplitConstraints are met.
-/// If it is evident that no bundles will be live, abort early and return false.
+/// Return false if there are no bundles with positive bias.
-bool RAGreedy::addSplitConstraints(unsigned PhysReg, float &Cost) {
+bool RAGreedy::addSplitConstraints(InterferenceCache::Cursor Intf,
-  InterferenceCache::Cursor Intf(IntfCache, PhysReg);
+                                   float &Cost) {
   ArrayRef<SplitAnalysis::BlockInfo> UseBlocks = SA->getUseBlocks();
 
   // Reset interference dependent info.
@@ -464,35 +466,41 @@ bool RAGreedy::addSplitConstraints(unsigned PhysReg, float &Cost) {
     if (Ins)
       StaticCost += Ins * SpillPlacer->getBlockFrequency(BC.Number);
   }
+  Cost = StaticCost;
 
   // Add constraints for use-blocks. Note that these are the only constraints
   // that may add a positive bias, it is downhill from here.
   SpillPlacer->addConstraints(SplitConstraints);
-  if (SpillPlacer->getPositiveNodes() == 0)
+  return SpillPlacer->scanActiveBundles();
-    return false;
+}
 
-  Cost = StaticCost;
 
-  // Now handle the live-through blocks without uses. These can only add
+/// addThroughConstraints - Add constraints and links to SpillPlacer from the
-  // negative bias, so we can abort whenever there are no more positive nodes.
+/// live-through blocks in Blocks.
-  // Compute constraints for a group of 8 blocks at a time.
+void RAGreedy::addThroughConstraints(InterferenceCache::Cursor Intf,
+                                     ArrayRef<unsigned> Blocks) {
   const unsigned GroupSize = 8;
   SpillPlacement::BlockConstraint BCS[GroupSize];
-  unsigned B = 0;
+  unsigned TBS[GroupSize];
-  TransparentBlocks.clear();
+  unsigned B = 0, T = 0;
 
-  ArrayRef<unsigned> ThroughBlocks = SA->getThroughBlocks();
+  for (unsigned i = 0; i != Blocks.size(); ++i) {
-  for (unsigned i = 0; i != ThroughBlocks.size(); ++i) {
+    unsigned Number = Blocks[i];
-    unsigned Number = ThroughBlocks[i];
-    assert(B < GroupSize && "Array overflow");
-    BCS[B].Number = Number;
     Intf.moveToBlock(Number);
 
     if (!Intf.hasInterference()) {
-      TransparentBlocks.push_back(Number);
+      assert(T < GroupSize && "Array overflow");
+      TBS[T] = Number;
+      if (++T == GroupSize) {
+        SpillPlacer->addLinks(ArrayRef<unsigned>(TBS, T));
+        T = 0;
+      }
       continue;
     }
 
+    assert(B < GroupSize && "Array overflow");
+    BCS[B].Number = Number;
+
     // Interference for the live-in value.
     if (Intf.first() <= Indexes->getMBBStartIdx(Number))
       BCS[B].Entry = SpillPlacement::MustSpill;
@@ -509,22 +517,55 @@ bool RAGreedy::addSplitConstraints(unsigned PhysReg, float &Cost) {
       ArrayRef<SpillPlacement::BlockConstraint> Array(BCS, B);
       SpillPlacer->addConstraints(Array);
       B = 0;
-      // Abort early when all hope is lost.
-      if (SpillPlacer->getPositiveNodes() == 0)
-        return false;
     }
   }
 
   ArrayRef<SpillPlacement::BlockConstraint> Array(BCS, B);
   SpillPlacer->addConstraints(Array);
-  if (SpillPlacer->getPositiveNodes() == 0)
+  SpillPlacer->addLinks(ArrayRef<unsigned>(TBS, T));
-    return false;
-
-  // There is still some positive bias. Add all the links.
-  SpillPlacer->addLinks(TransparentBlocks);
-  return true;
 }
 
+void RAGreedy::growRegion(InterferenceCache::Cursor Intf) {
+  // Keep track of through blocks that have already been added to SpillPlacer.
+  SparseBitVector<> Added;
+  SmallVector<unsigned, 16> ThroughBlocks;
+#ifndef NDEBUG
+  unsigned Visited = 0;
+#endif
+  for (;;) {
+    ArrayRef<unsigned> NewBundles = SpillPlacer->getRecentPositive();
+    if (NewBundles.empty())
+      break;
+    // Find new through blocks in the periphery of PrefRegBundles.
+    for (int i = 0, e = NewBundles.size(); i != e; ++i) {
+      unsigned Bundle = NewBundles[i];
+      // Look at all blocks connected to Bundle in the full graph.
+      ArrayRef<unsigned> Blocks = Bundles->getBlocks(Bundle);
+      for (ArrayRef<unsigned>::iterator I = Blocks.begin(), E = Blocks.end();
+           I != E; ++I) {
+        unsigned Block = *I;
+        if (!SA->isThroughBlock(Block) || !Added.test_and_set(Block))
+          continue;
+        // This is a new through block. Add it to SpillPlacer later.
+        ThroughBlocks.push_back(Block);
+#ifndef NDEBUG
+        ++Visited;
+#endif
+      }
+    }
+    // Any new blocks to add?
+    if (!ThroughBlocks.empty()) {
+      addThroughConstraints(Intf, ThroughBlocks);
+      ThroughBlocks.clear();
+    }
+    // Perhaps iterating can enable more bundles?
+    SpillPlacer->iterate();
+  }
+
+  // Rememeber the relevant set of through blocks for splitAroundRegion().
+  ActiveThroughBlocks |= Added;
+  DEBUG(dbgs() << ", v=" << Visited);
+}
 
 /// calcGlobalSplitCost - Return the global split cost of following the split
 /// pattern in LiveBundles. This cost should be added to the local cost of the
@@ -550,10 +591,9 @@ float RAGreedy::calcGlobalSplitCost(unsigned PhysReg,
   }
 
   InterferenceCache::Cursor Intf(IntfCache, PhysReg);
-  ArrayRef<unsigned> ThroughBlocks = SA->getThroughBlocks();
+  for (SparseBitVector<>::iterator I = ActiveThroughBlocks.begin(),
-  SplitConstraints.resize(UseBlocks.size() + ThroughBlocks.size());
+       E = ActiveThroughBlocks.end(); I != E; ++I) {
-  for (unsigned i = 0; i != ThroughBlocks.size(); ++i) {
+    unsigned Number = *I;
-    unsigned Number = ThroughBlocks[i];
     bool RegIn  = LiveBundles[Bundles->getBundle(Number, 0)];
     bool RegOut = LiveBundles[Bundles->getBundle(Number, 1)];
     if (!RegIn && !RegOut)
@@ -766,9 +806,9 @@ void RAGreedy::splitAroundRegion(LiveInterval &VirtReg, unsigned PhysReg,
   }
 
   // Handle live-through blocks.
-  ArrayRef<unsigned> ThroughBlocks = SA->getThroughBlocks();
+  for (SparseBitVector<>::iterator I = ActiveThroughBlocks.begin(),
-  for (unsigned i = 0; i != ThroughBlocks.size(); ++i) {
+       E = ActiveThroughBlocks.end(); I != E; ++I) {
-    unsigned Number = ThroughBlocks[i];
+    unsigned Number = *I;
     bool RegIn  = LiveBundles[Bundles->getBundle(Number, 0)];
     bool RegOut = LiveBundles[Bundles->getBundle(Number, 1)];
     DEBUG(dbgs() << "Live through BB#" << Number << '\n');
@@ -804,6 +844,7 @@ unsigned RAGreedy::tryRegionSplit(LiveInterval &VirtReg, AllocationOrder &Order,
   BitVector LiveBundles, BestBundles;
   float BestCost = 0;
   unsigned BestReg = 0;
+  ActiveThroughBlocks.clear();
 
   Order.rewind();
   for (unsigned Cand = 0; unsigned PhysReg = Order.next(); ++Cand) {
@@ -813,16 +854,17 @@ unsigned RAGreedy::tryRegionSplit(LiveInterval &VirtReg, AllocationOrder &Order,
 
     SpillPlacer->prepare(LiveBundles);
     float Cost;
-    if (!addSplitConstraints(PhysReg, Cost)) {
+    InterferenceCache::Cursor Intf(IntfCache, PhysReg);
-      DEBUG(dbgs() << PrintReg(PhysReg, TRI) << "\tno positive bias\n");
+    if (!addSplitConstraints(Intf, Cost)) {
+      DEBUG(dbgs() << PrintReg(PhysReg, TRI) << "\tno positive bundles\n");
       continue;
     }
-    DEBUG(dbgs() << PrintReg(PhysReg, TRI) << "\tbiased = "
+    DEBUG(dbgs() << PrintReg(PhysReg, TRI) << "\tstatic = " << Cost);
-                 << SpillPlacer->getPositiveNodes() << ", static = " << Cost);
     if (BestReg && Cost >= BestCost) {
       DEBUG(dbgs() << " worse than " << PrintReg(BestReg, TRI) << '\n');
       continue;
     }
+    growRegion(Intf);
 
     SpillPlacer->finish();
 

lib/CodeGen/SpillPlacement.cpp

@@ -135,13 +135,10 @@ struct SpillPlacement::Node {
 
   /// addBias - Bias this node from an ingoing[0] or outgoing[1] link.
   /// Return the change to the total number of positive biases.
-  int addBias(float w, bool out) {
+  void addBias(float w, bool out) {
     // Normalize w relative to all connected blocks from that direction.
     w *= Scale[out];
-    int Before = Bias > 0;
     Bias += w;
-    int After = Bias > 0;
-    return After - Before;
   }
 
   /// update - Recompute Value from Bias and Links. Return true when node
@@ -230,14 +227,14 @@ void SpillPlacement::addConstraints(ArrayRef<BlockConstraint> LiveBlocks) {
     if (I->Entry != DontCare) {
       unsigned ib = bundles->getBundle(I->Number, 0);
       activate(ib);
-      PositiveNodes += nodes[ib].addBias(Freq * Bias[I->Entry], 1);
+      nodes[ib].addBias(Freq * Bias[I->Entry], 1);
     }
 
     // Live-out from block?
     if (I->Exit != DontCare) {
       unsigned ob = bundles->getBundle(I->Number, 1);
       activate(ob);
-      PositiveNodes += nodes[ob].addBias(Freq * Bias[I->Exit], 0);
+      nodes[ob].addBias(Freq * Bias[I->Exit], 0);
     }
   }
 }
@@ -254,16 +251,42 @@ void SpillPlacement::addLinks(ArrayRef<unsigned> Links) {
       continue;
     activate(ib);
     activate(ob);
+    if (nodes[ib].Links.empty() && !nodes[ib].mustSpill())
+      Linked.push_back(ib);
+    if (nodes[ob].Links.empty() && !nodes[ob].mustSpill())
+      Linked.push_back(ob);
     float Freq = getBlockFrequency(Number);
     nodes[ib].addLink(ob, Freq, 1);
     nodes[ob].addLink(ib, Freq, 0);
   }
 }
 
+bool SpillPlacement::scanActiveBundles() {
+  Linked.clear();
+  RecentPositive.clear();
+  for (int n = ActiveNodes->find_first(); n>=0; n = ActiveNodes->find_next(n)) {
+    nodes[n].update(nodes);
+    // A node that must spill, or a node without any links is not going to
+    // change its value ever again, so exclude it from iterations.
+    if (nodes[n].mustSpill())
+      continue;
+    if (!nodes[n].Links.empty())
+      Linked.push_back(n);
+    if (nodes[n].preferReg())
+      RecentPositive.push_back(n);
+  }
+  return !RecentPositive.empty();
+}
+
 /// iterate - Repeatedly update the Hopfield nodes until stability or the
 /// maximum number of iterations is reached.
 /// @param Linked - Numbers of linked nodes that need updating.
-void SpillPlacement::iterate(const SmallVectorImpl<unsigned> &Linked) {
+void SpillPlacement::iterate() {
+  // First update the recently positive nodes. They have likely received new
+  // negative bias that will turn them off.
+  while (!RecentPositive.empty())
+    nodes[RecentPositive.pop_back_val()].update(nodes);
+
   if (Linked.empty())
     return;
 
@@ -279,10 +302,13 @@ void SpillPlacement::iterate(const SmallVectorImpl<unsigned> &Linked) {
     for (SmallVectorImpl<unsigned>::const_reverse_iterator I =
            llvm::next(Linked.rbegin()), E = Linked.rend(); I != E; ++I) {
       unsigned n = *I;
-      bool C = nodes[n].update(nodes);
+      if (nodes[n].update(nodes)) {
-      Changed |= C;
+        Changed = true;
+        if (nodes[n].preferReg())
+          RecentPositive.push_back(n);
       }
-    if (!Changed)
+    }
+    if (!Changed || !RecentPositive.empty())
       return;
 
     // Scan forwards, skipping the first node which was just updated.
@@ -290,38 +316,29 @@ void SpillPlacement::iterate(const SmallVectorImpl<unsigned> &Linked) {
     for (SmallVectorImpl<unsigned>::const_iterator I =
            llvm::next(Linked.begin()), E = Linked.end(); I != E; ++I) {
       unsigned n = *I;
-      bool C = nodes[n].update(nodes);
+      if (nodes[n].update(nodes)) {
-      Changed |= C;
+        Changed = true;
+        if (nodes[n].preferReg())
+          RecentPositive.push_back(n);
       }
-    if (!Changed)
+    }
+    if (!Changed || !RecentPositive.empty())
       return;
   }
 }
 
 void SpillPlacement::prepare(BitVector &RegBundles) {
+  Linked.clear();
+  RecentPositive.clear();
   // Reuse RegBundles as our ActiveNodes vector.
   ActiveNodes = &RegBundles;
   ActiveNodes->clear();
   ActiveNodes->resize(bundles->getNumBundles());
-  PositiveNodes = 0;
 }
 
 bool
 SpillPlacement::finish() {
   assert(ActiveNodes && "Call prepare() first");
-  // Update all active nodes, and find the ones that are actually linked to
-  // something so their value may change when iterating.
-  SmallVector<unsigned, 8> Linked;
-  for (int n = ActiveNodes->find_first(); n>=0; n = ActiveNodes->find_next(n)) {
-    nodes[n].update(nodes);
-    // A node that must spill, or a node without any links is not going to
-    // change its value ever again, so exclude it from iterations.
-    if (!nodes[n].Links.empty() && !nodes[n].mustSpill())
-      Linked.push_back(n);
-  }
-
-  // Iterate the network to convergence.
-  iterate(Linked);
 
   // Write preferences back to ActiveNodes.
   bool Perfect = true;

lib/CodeGen/SpillPlacement.h

@@ -49,8 +49,12 @@ class SpillPlacement  : public MachineFunctionPass {
   // caller.
   BitVector *ActiveNodes;
 
-  // The number of active nodes with a positive bias.
+  // Nodes with active links. Populated by scanActiveBundles.
-  unsigned PositiveNodes;
+  SmallVector<unsigned, 8> Linked;
+
+  // Nodes that went positive during the last call to scanActiveBundles or
+  // iterate.
+  SmallVector<unsigned, 8> RecentPositive;
 
   // Block frequencies are computed once. Indexed by block number.
   SmallVector<float, 4> BlockFrequency;
@@ -95,9 +99,20 @@ public:
   /// addLinks - Add transparent blocks with the given numbers.
   void addLinks(ArrayRef<unsigned> Links);
 
-  /// getPositiveNodes - Return the total number of graph nodes with a positive
+  /// scanActiveBundles - Perform an initial scan of all bundles activated by
-  /// bias after adding constraints.
+  /// addConstraints and addLinks, updating their state. Add all the bundles
-  unsigned getPositiveNodes() const { return PositiveNodes; }
+  /// that now prefer a register to RecentPositive.
+  /// Prepare internal data structures for iterate.
+  /// Return true is there are any positive nodes.
+  bool scanActiveBundles();
+
+  /// iterate - Update the network iteratively until convergence, or new bundles
+  /// are found.
+  void iterate();
+
+  /// getRecentPositive - Return an array of bundles that became positive during
+  /// the previous call to scanActiveBundles or iterate.
+  ArrayRef<unsigned> getRecentPositive() { return RecentPositive; }
 
   /// finish - Compute the optimal spill code placement given the
   /// constraints. No MustSpill constraints will be violated, and the smallest
@@ -120,7 +135,6 @@ private:
   virtual void releaseMemory();
 
   void activate(unsigned);
-  void iterate(const SmallVectorImpl<unsigned>&);
 };
 
 } // end namespace llvm

lib/CodeGen/SplitKit.cpp

@@ -132,12 +132,14 @@ void SplitAnalysis::analyzeUses() {
   DEBUG(dbgs() << "Analyze counted "
                << UseSlots.size() << " instrs in "
                << UseBlocks.size() << " blocks, through "
-               << ThroughBlocks.size() << " blocks.\n");
+               << NumThroughBlocks << " blocks.\n");
 }
 
 /// calcLiveBlockInfo - Fill the LiveBlocks array with information about blocks
 /// where CurLI is live.
 bool SplitAnalysis::calcLiveBlockInfo() {
+  ThroughBlocks.resize(MF.getNumBlockIDs());
+  NumThroughBlocks = 0;
   if (CurLI->empty())
     return true;
 
@@ -193,9 +195,10 @@ bool SplitAnalysis::calcLiveBlockInfo() {
     BI.LiveThrough = !hasGap && BI.LiveIn && BI.LiveOut;
     if (Uses)
       UseBlocks.push_back(BI);
-    else
+    else {
-      ThroughBlocks.push_back(BI.MBB->getNumber());
+      ++NumThroughBlocks;
-
+      ThroughBlocks.set(BI.MBB->getNumber());
+    }
     // FIXME: This should never happen. The live range stops or starts without a
     // corresponding use. An earlier pass did something wrong.
     if (!BI.LiveThrough && !Uses)

lib/CodeGen/SplitKit.h

@@ -89,7 +89,10 @@ private:
   SmallVector<BlockInfo, 8> UseBlocks;
 
   /// ThroughBlocks - Block numbers where CurLI is live through without uses.
-  SmallVector<unsigned, 8> ThroughBlocks;
+  BitVector ThroughBlocks;
+
+  /// NumThroughBlocks - Number of live-through blocks.
+  unsigned NumThroughBlocks;
 
   SlotIndex computeLastSplitPoint(unsigned Num);
 
@@ -135,9 +138,11 @@ public:
   /// where CurLI has uses.
   ArrayRef<BlockInfo> getUseBlocks() { return UseBlocks; }
 
-  /// getThroughBlocks - Return an array of block numbers where CurLI is live
+  /// getNumThroughBlocks - Return the number of through blocks.
-  /// through without uses.
+  unsigned getNumThroughBlocks() const { return NumThroughBlocks; }
-  ArrayRef<unsigned> getThroughBlocks() { return ThroughBlocks; }
+
+  /// isThroughBlock - Return true if CurLI is live through MBB without uses.
+  bool isThroughBlock(unsigned MBB) const { return ThroughBlocks.test(MBB); }
 
   typedef SmallPtrSet<const MachineBasicBlock*, 16> BlockPtrSet;