RDF: Dead code elimination

Utility class to perform DFG-based dead code elimination.

llvm-svn: 257485

lib/Target/Hexagon/CMakeLists.txt

@@ -49,6 +49,7 @@ add_llvm_target(HexagonCodeGen
   HexagonTargetObjectFile.cpp
   HexagonTargetTransformInfo.cpp
   HexagonVLIWPacketizer.cpp
+  RDFDeadCode.cpp
   RDFGraph.cpp
   RDFLiveness.cpp
 )

lib/Target/Hexagon/RDFDeadCode.cpp

@@ -0,0 +1,204 @@
+//===--- RDFDeadCode.cpp --------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// RDF-based generic dead code elimination.
+
+#include "RDFGraph.h"
+#include "RDFLiveness.h"
+#include "RDFDeadCode.h"
+
+#include "llvm/ADT/SetVector.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+using namespace llvm;
+using namespace rdf;
+
+// Check if the given instruction has observable side-effects, i.e. if
+// it should be considered "live". It is safe for this function to be
+// overly conservative (i.e. return "true" for all instructions), but it
+// is not safe to return "false" for an instruction that should not be
+// considered removable.
+bool DeadCodeElimination::isLiveInstr(const MachineInstr *MI) const {
+  if (MI->mayStore() || MI->isBranch() || MI->isCall() || MI->isReturn())
+    return true;
+  if (MI->hasOrderedMemoryRef() || MI->hasUnmodeledSideEffects())
+    return true;
+  if (MI->isPHI())
+    return false;
+  for (auto &Op : MI->operands())
+    if (Op.isReg() && MRI.isReserved(Op.getReg()))
+      return true;
+  return false;
+}
+
+void DeadCodeElimination::scanInstr(NodeAddr<InstrNode*> IA,
+      SetVector<NodeId> &WorkQ) {
+  if (!DFG.IsCode<NodeAttrs::Stmt>(IA))
+    return;
+  if (!isLiveInstr(NodeAddr<StmtNode*>(IA).Addr->getCode()))
+    return;
+  for (NodeAddr<RefNode*> RA : IA.Addr->members(DFG)) {
+    if (!LiveNodes.count(RA.Id))
+      WorkQ.insert(RA.Id);
+  }
+}
+
+void DeadCodeElimination::processDef(NodeAddr<DefNode*> DA,
+      SetVector<NodeId> &WorkQ) {
+  NodeAddr<InstrNode*> IA = DA.Addr->getOwner(DFG);
+  for (NodeAddr<UseNode*> UA : IA.Addr->members_if(DFG.IsUse, DFG)) {
+    if (!LiveNodes.count(UA.Id))
+      WorkQ.insert(UA.Id);
+  }
+  for (NodeAddr<DefNode*> TA : DFG.getRelatedRefs(IA, DA))
+    LiveNodes.insert(TA.Id);
+}
+
+void DeadCodeElimination::processUse(NodeAddr<UseNode*> UA,
+      SetVector<NodeId> &WorkQ) {
+  for (NodeAddr<DefNode*> DA : LV.getAllReachingDefs(UA)) {
+    if (!LiveNodes.count(DA.Id))
+      WorkQ.insert(DA.Id);
+  }
+}
+
+// Traverse the DFG and collect the set dead RefNodes and the set of
+// dead instructions. Return "true" if any of these sets is non-empty,
+// "false" otherwise.
+bool DeadCodeElimination::collect() {
+  // This function works by first finding all live nodes. The dead nodes
+  // are then the complement of the set of live nodes.
+  //
+  // Assume that all nodes are dead. Identify instructions which must be
+  // considered live, i.e. instructions with observable side-effects, such
+  // as calls and stores. All arguments of such instructions are considered
+  // live. For each live def, all operands used in the corresponding
+  // instruction are considered live. For each live use, all its reaching
+  // defs are considered live.
+  LiveNodes.clear();
+  SetVector<NodeId> WorkQ;
+  for (NodeAddr<BlockNode*> BA : DFG.getFunc().Addr->members(DFG))
+    for (NodeAddr<InstrNode*> IA : BA.Addr->members(DFG))
+      scanInstr(IA, WorkQ);
+
+  while (!WorkQ.empty()) {
+    NodeId N = *WorkQ.begin();
+    WorkQ.remove(N);
+    LiveNodes.insert(N);
+    auto RA = DFG.addr<RefNode*>(N);
+    if (DFG.IsDef(RA))
+      processDef(RA, WorkQ);
+    else
+      processUse(RA, WorkQ);
+  }
+
+  if (trace()) {
+    dbgs() << "Live nodes:\n";
+    for (NodeId N : LiveNodes) {
+      auto RA = DFG.addr<RefNode*>(N);
+      dbgs() << PrintNode<RefNode*>(RA, DFG) << "\n";
+    }
+  }
+
+  auto IsDead = [this] (NodeAddr<InstrNode*> IA) -> bool {
+    for (NodeAddr<DefNode*> DA : IA.Addr->members_if(DFG.IsDef, DFG))
+      if (LiveNodes.count(DA.Id))
+        return false;
+    return true;
+  };
+
+  for (NodeAddr<BlockNode*> BA : DFG.getFunc().Addr->members(DFG)) {
+    for (NodeAddr<InstrNode*> IA : BA.Addr->members(DFG)) {
+      for (NodeAddr<RefNode*> RA : IA.Addr->members(DFG))
+        if (!LiveNodes.count(RA.Id))
+          DeadNodes.insert(RA.Id);
+      if (DFG.IsCode<NodeAttrs::Stmt>(IA))
+        if (isLiveInstr(NodeAddr<StmtNode*>(IA).Addr->getCode()))
+          continue;
+      if (IsDead(IA)) {
+        DeadInstrs.insert(IA.Id);
+        if (trace())
+          dbgs() << "Dead instr: " << PrintNode<InstrNode*>(IA, DFG) << "\n";
+      }
+    }
+  }
+
+  return !DeadNodes.empty();
+}
+
+// Erase the nodes given in the Nodes set from DFG. In addition to removing
+// them from the DFG, if a node corresponds to a statement, the corresponding
+// machine instruction is erased from the function.
+bool DeadCodeElimination::erase(const SetVector<NodeId> &Nodes) {
+  if (Nodes.empty())
+    return false;
+
+  // Prepare the actual set of ref nodes to remove: ref nodes from Nodes
+  // are included directly, for each InstrNode in Nodes, include the set
+  // of all RefNodes from it.
+  NodeList DRNs, DINs;
+  for (auto I : Nodes) {
+    auto BA = DFG.addr<NodeBase*>(I);
+    uint16_t Type = BA.Addr->getType();
+    if (Type == NodeAttrs::Ref) {
+      DRNs.push_back(DFG.addr<RefNode*>(I));
+      continue;
+    }
+
+    // If it's a code node, add all ref nodes from it.
+    uint16_t Kind = BA.Addr->getKind();
+    if (Kind == NodeAttrs::Stmt || Kind == NodeAttrs::Phi) {
+      for (auto N : NodeAddr<CodeNode*>(BA).Addr->members(DFG))
+        DRNs.push_back(N);
+      DINs.push_back(DFG.addr<InstrNode*>(I));
+    } else {
+      llvm_unreachable("Unexpected code node");
+      return false;
+    }
+  }
+
+  // Sort the list so that use nodes are removed first. This makes the
+  // "unlink" functions a bit faster.
+  auto UsesFirst = [] (NodeAddr<RefNode*> A, NodeAddr<RefNode*> B) -> bool {
+    uint16_t KindA = A.Addr->getKind(), KindB = B.Addr->getKind();
+    if (KindA == NodeAttrs::Use && KindB == NodeAttrs::Def)
+      return true;
+    if (KindA == NodeAttrs::Def && KindB == NodeAttrs::Use)
+      return false;
+    return A.Id < B.Id;
+  };
+  std::sort(DRNs.begin(), DRNs.end(), UsesFirst);
+
+  if (trace())
+    dbgs() << "Removing dead ref nodes:\n";
+  for (NodeAddr<RefNode*> RA : DRNs) {
+    if (trace())
+      dbgs() << "  " << PrintNode<RefNode*>(RA, DFG) << '\n';
+    if (DFG.IsUse(RA))
+      DFG.unlinkUse(RA);
+    else if (DFG.IsDef(RA))
+      DFG.unlinkDef(RA);
+  }
+
+  // Now, remove all dead instruction nodes.
+  for (NodeAddr<InstrNode*> IA : DINs) {
+    NodeAddr<BlockNode*> BA = IA.Addr->getOwner(DFG);
+    BA.Addr->removeMember(IA, DFG);
+    if (!DFG.IsCode<NodeAttrs::Stmt>(IA))
+      continue;
+
+    MachineInstr *MI = NodeAddr<StmtNode*>(IA).Addr->getCode();
+    if (trace())
+      dbgs() << "erasing: " << *MI;
+    MI->eraseFromParent();
+  }
+  return true;
+}

lib/Target/Hexagon/RDFDeadCode.h

@@ -0,0 +1,65 @@
+//===--- RDFDeadCode.h ----------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// RDF-based generic dead code elimination.
+//
+// The main interface of this class are functions "collect" and "erase".
+// This allows custom processing of the function being optimized by a
+// particular consumer. The simplest way to use this class would be to
+// instantiate an object, and then simply call "collect" and "erase",
+// passing the result of "getDeadInstrs()" to it.
+// A more complex scenario would be to call "collect" first, then visit
+// all post-increment instructions to see if the address update is dead
+// or not, and if it is, convert the instruction to a non-updating form.
+// After that "erase" can be called with the set of nodes including both,
+// dead defs from the updating instructions and the nodes corresponding
+// to the dead instructions.
+
+#ifndef RDF_DEADCODE_H
+#define RDF_DEADCODE_H
+
+#include "RDFGraph.h"
+#include "RDFLiveness.h"
+#include "llvm/ADT/SetVector.h"
+
+namespace llvm {
+  class MachineRegisterInfo;
+}
+
+namespace rdf {
+  struct DeadCodeElimination {
+    DeadCodeElimination(DataFlowGraph &dfg, MachineRegisterInfo &mri)
+      : Trace(false), DFG(dfg), MRI(mri), LV(mri, dfg) {}
+
+    bool collect();
+    bool erase(const SetVector<NodeId> &Nodes);
+    void trace(bool On) { Trace = On; }
+    bool trace() const { return Trace; }
+
+    SetVector<NodeId> getDeadNodes() { return DeadNodes; }
+    SetVector<NodeId> getDeadInstrs() { return DeadInstrs; }
+    DataFlowGraph &getDFG() { return DFG; }
+
+  private:
+    bool Trace;
+    SetVector<NodeId> LiveNodes;
+    SetVector<NodeId> DeadNodes;
+    SetVector<NodeId> DeadInstrs;
+    DataFlowGraph &DFG;
+    MachineRegisterInfo &MRI;
+    Liveness LV;
+
+    bool isLiveInstr(const MachineInstr *MI) const;
+    void scanInstr(NodeAddr<InstrNode*> IA, SetVector<NodeId> &WorkQ);
+    void processDef(NodeAddr<DefNode*> DA, SetVector<NodeId> &WorkQ);
+    void processUse(NodeAddr<UseNode*> UA, SetVector<NodeId> &WorkQ);
+  };
+}
+
+#endif