[llvm-commits] CVS: llvm/include/llvm/Analysis/ET-Forest.h Dominators.h PostDominators.h

Chris Lattner lattner at cs.uiuc.edu
Sun Jan 8 00:20:10 PST 2006



Changes in directory llvm/include/llvm/Analysis:

ET-Forest.h added (r1.1)
Dominators.h updated: 1.51 -> 1.52
PostDominators.h updated: 1.9 -> 1.10
---
Log message:

Initial implementation of the ET-Forest data structure for dominators and
post-dominators.  This code was written/adapted by Daniel Berlin!


---
Diffs of the changes:  (+447 -1)

 Dominators.h     |  116 ++++++++++++++++++++
 ET-Forest.h      |  309 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 PostDominators.h |   23 ++++
 3 files changed, 447 insertions(+), 1 deletion(-)


Index: llvm/include/llvm/Analysis/ET-Forest.h
diff -c /dev/null llvm/include/llvm/Analysis/ET-Forest.h:1.1
*** /dev/null	Sun Jan  8 02:20:08 2006
--- llvm/include/llvm/Analysis/ET-Forest.h	Sun Jan  8 02:19:58 2006
***************
*** 0 ****
--- 1,309 ----
+ //===- llvm/Analysis/ET-Forest.h - ET-Forest implementation -----*- C++ -*-===//
+ //
+ //                     The LLVM Compiler Infrastructure
+ //
+ // This file was written by Daniel Berlin from code written by Pavel Nejedy, and
+ // is distributed under the University of Illinois Open Source License. See
+ // LICENSE.TXT for details.
+ //
+ //===----------------------------------------------------------------------===//
+ //
+ // This file defines the following classes:
+ //  1. ETNode:  A node in the ET forest.
+ //  2. ETOccurrence: An occurrence of the node in the splay tree
+ //  storing the DFS path information.
+ //
+ //  The ET-forest structure is described in:
+ //    D. D. Sleator and R. E. Tarjan. A data structure for dynamic trees.
+ //    J.  G'omput. System Sci., 26(3):362 381, 1983.
+ //
+ // Basically, the ET-Forest is storing the dominator tree (ETNode),
+ // and a splay tree containing the depth first path information for
+ // those nodes (ETOccurrence).  This enables us to answer queries
+ // about domination (DominatedBySlow), and ancestry (NCA) in
+ // logarithmic time, and perform updates to the information in
+ // logarithmic time.
+ //
+ //===----------------------------------------------------------------------===//
+ 
+ #ifndef LLVM_ANALYSIS_ETFOREST_H
+ #define LLVM_ANALYSIS_ETFOREST_H
+ 
+ #include <cassert>
+ 
+ namespace llvm {
+ class ETNode;
+ 
+ /// ETOccurrence - An occurrence for a node in the et tree
+ ///
+ /// The et occurrence tree is really storing the sequences you get from
+ /// doing a DFS over the ETNode's.  It is stored as a modified splay
+ /// tree.
+ /// ET occurrences can occur at multiple places in the ordering depending
+ /// on how many ET nodes have it as their father.  To handle
+ /// this, they are separate from the nodes.
+ ///
+ class ETOccurrence {
+ public:
+   ETOccurrence(ETNode *n): OccFor(n), Parent(NULL), Left(NULL), Right(NULL),
+     Depth(0), Min(0), MinOccurrence(this) {};
+ 
+   void setParent(ETOccurrence *n) {
+     Parent = n;
+   }
+ 
+   // Add D to our current depth
+   void setDepthAdd(int d) {
+     Min += d;
+     Depth += d;
+   }
+   
+   // Reset our depth to D
+   void setDepth(int d) {
+     Min += d - Depth;
+     Depth = d;
+   }
+ 
+   // Set Left to N
+   void setLeft(ETOccurrence *n) {
+     assert(n != this && "Trying to set our left to ourselves");
+     Left = n;
+     if (n)
+       n->setParent(this);
+   }
+   
+   // Set Right to N
+   void setRight(ETOccurrence *n) {
+     assert(n != this && "Trying to set our right to ourselves");
+     Right = n;
+     if (n)
+       n->setParent(this);
+   }
+   
+   // Splay us to the root of the tree
+   void Splay(void);
+ 
+   // Recompute the minimum occurrence for this occurrence.
+   void recomputeMin(void) {
+     ETOccurrence *themin = Left;
+     
+     // The min may be our Right, too.
+     if (!themin || (Right && themin->Min > Right->Min))
+       themin = Right;
+     
+     if (themin && themin->Min < 0) {
+       Min = themin->Min + Depth;
+       MinOccurrence = themin->MinOccurrence;
+     } else {
+       Min = Depth;
+       MinOccurrence = this;
+     }
+   }
+  private:
+   friend class ETNode;
+ 
+     // Node we represent
+   ETNode *OccFor;
+ 
+   // Parent in the splay tree
+   ETOccurrence *Parent;
+ 
+   // Left Son in the splay tree
+   ETOccurrence *Left;
+ 
+   // Right Son in the splay tree
+   ETOccurrence *Right;
+ 
+   // Depth of the node is the sum of the depth on the path to the
+   // root.
+   int Depth;
+ 
+   // Subtree occurrence's minimum depth
+   int Min;
+ 
+   // Subtree occurrence with minimum depth
+   ETOccurrence *MinOccurrence;
+ };
+ 
+ 
+ class ETNode {
+ public:
+   ETNode(void *d) : data(d), Father(NULL), Left(NULL),
+                     Right(NULL), Son(NULL), ParentOcc(NULL) {   
+     RightmostOcc = new ETOccurrence(this);
+   };
+ 
+   // This does *not* maintain the tree structure.
+   // If you want to remove a node from the forest structure, use
+   // removeFromForest()
+   ~ETNode() {
+     delete RightmostOcc;
+   }
+ 
+   void removeFromForest() {
+     // Split us away from all our sons.
+     while (Son)
+       Son->Split();
+     
+     // And then split us away from our father.
+     if (Father)
+       Father->Split();
+   }
+ 
+   // Split us away from our parents and children, so that we can be
+   // reparented. NB: setFather WILL NOT DO WHAT YOU WANT IF YOU DO NOT
+   // SPLIT US FIRST.
+   void Split();
+ 
+   // Set our parent node to the passed in node
+   void setFather(ETNode *);
+   
+   // Nearest Common Ancestor of two et nodes.
+   ETNode *NCA(ETNode *);
+   
+   // Return true if we are below the passed in node in the forest.
+   bool Below(ETNode *);
+   /*
+    Given a dominator tree, we can determine whether one thing
+    dominates another in constant time by using two DFS numbers:
+   
+    1. The number for when we visit a node on the way down the tree
+    2. The number for when we visit a node on the way back up the tree
+   
+    You can view these as bounds for the range of dfs numbers the
+    nodes in the subtree of the dominator tree rooted at that node
+    will contain.
+   
+    The dominator tree is always a simple acyclic tree, so there are
+    only three possible relations two nodes in the dominator tree have
+    to each other:
+   
+    1. Node A is above Node B (and thus, Node A dominates node B)
+   
+         A
+         |
+         C
+        / \ 
+       B   D
+   
+   
+    In the above case, DFS_Number_In of A will be <= DFS_Number_In of
+    B, and DFS_Number_Out of A will be >= DFS_Number_Out of B.  This is
+    because we must hit A in the dominator tree *before* B on the walk
+    down, and we will hit A *after* B on the walk back up
+   
+    2. Node A is below node B (and thus, node B dominates node B)
+        
+         B
+         |
+         A
+        / \ 
+       C   D
+   
+    In the above case, DFS_Number_In of A will be >= DFS_Number_In of
+    B, and DFS_Number_Out of A will be <= DFS_Number_Out of B.
+   
+    This is because we must hit A in the dominator tree *after* B on
+    the walk down, and we will hit A *before* B on the walk back up
+   
+    3. Node A and B are siblings (and thus, neither dominates the other)
+   
+         C
+         |
+         D
+        / \                        
+       A   B
+   
+    In the above case, DFS_Number_In of A will *always* be <=
+    DFS_Number_In of B, and DFS_Number_Out of A will *always* be <=
+    DFS_Number_Out of B.  This is because we will always finish the dfs
+    walk of one of the subtrees before the other, and thus, the dfs
+    numbers for one subtree can't intersect with the range of dfs
+    numbers for the other subtree.  If you swap A and B's position in
+    the dominator tree, the comparison changes direction, but the point
+    is that both comparisons will always go the same way if there is no
+    dominance relationship.
+   
+    Thus, it is sufficient to write
+   
+    A_Dominates_B(node A, node B) {
+       return DFS_Number_In(A) <= DFS_Number_In(B) &&
+              DFS_Number_Out(A) >= DFS_Number_Out(B);
+    }
+   
+    A_Dominated_by_B(node A, node B) {
+      return DFS_Number_In(A) >= DFS_Number_In(A) &&
+             DFS_Number_Out(A) <= DFS_Number_Out(B);
+    }
+   */
+   bool DominatedBy(ETNode *other) const {
+     return this->DFSNumIn >= other->DFSNumIn &&
+            this->DFSNumOut <= other->DFSNumOut;
+   }
+   
+   // This method is slower, but doesn't require the DFS numbers to
+   // be up to date.
+   bool DominatedBySlow(ETNode *other) {
+     return this->Below(other);
+   }
+ 
+   void assignDFSNumber(int &num) {
+     DFSNumIn = num++;
+     
+     if (Son) {
+       Son->assignDFSNumber(num);
+       for (ETNode *son = Son->Right; son != Son; son = son->Right)
+         son->assignDFSNumber(num);
+     }
+     DFSNumOut = num++;
+   }
+   
+   bool hasFather() const {
+     return Father != NULL;
+   }
+ 
+   // Do not let people play around with fathers.
+   const ETNode *getFather() const {
+     return Father;
+   }
+ 
+   template <typename T>
+   T *getData() const {
+     return static_cast<T*>(data);
+   }
+   
+   unsigned getDFSNumIn() const {
+     return DFSNumIn;
+   }
+   
+   unsigned getDFSNumOut() const {
+     return DFSNumOut;
+   }
+ 
+  private:
+   // Data represented by the node
+   void *data;
+ 
+   // DFS Numbers
+   unsigned DFSNumIn, DFSNumOut;
+ 
+   // Father
+   ETNode *Father;
+ 
+   // Brothers.  Node, this ends up being a circularly linked list.
+   // Thus, if you want to get all the brothers, you need to stop when
+   // you hit node == this again.
+   ETNode *Left, *Right;
+ 
+   // First Son
+   ETNode *Son;
+ 
+   // Rightmost occurrence for this node
+   ETOccurrence *RightmostOcc;
+ 
+   // Parent occurrence for this node
+   ETOccurrence *ParentOcc;
+ };
+ }  // end llvm namespace
+ 
+ #endif


Index: llvm/include/llvm/Analysis/Dominators.h
diff -u llvm/include/llvm/Analysis/Dominators.h:1.51 llvm/include/llvm/Analysis/Dominators.h:1.52
--- llvm/include/llvm/Analysis/Dominators.h:1.51	Mon Nov 28 19:07:12 2005
+++ llvm/include/llvm/Analysis/Dominators.h	Sun Jan  8 02:19:58 2006
@@ -13,7 +13,9 @@
 //  2. DominatorSet: Calculates the [reverse] dominator set for a function
 //  3. DominatorTree: Represent the ImmediateDominator as an explicit tree
 //     structure.
-//  4. DominanceFrontier: Calculate and hold the dominance frontier for a
+//  4. ETForest: Efficient data structure for dominance comparisons and 
+//     nearest-common-ancestor queries.
+//  5. DominanceFrontier: Calculate and hold the dominance frontier for a
 //     function.
 //
 //  These data structures are listed in increasing order of complexity.  It
@@ -25,6 +27,7 @@
 #ifndef LLVM_ANALYSIS_DOMINATORS_H
 #define LLVM_ANALYSIS_DOMINATORS_H
 
+#include "llvm/Analysis/ET-Forest.h"
 #include "llvm/Pass.h"
 #include <set>
 
@@ -389,6 +392,116 @@
 
 
 //===-------------------------------------
+/// ET-Forest Class - Class used to construct forwards and backwards 
+/// ET-Forests
+///
+struct ETForestBase : public DominatorBase {
+  ETForestBase(bool isPostDom) : DominatorBase(isPostDom), Nodes(), 
+                                 DFSInfoValid(false) {}
+  
+  virtual void releaseMemory() { reset(); }
+
+  typedef std::map<BasicBlock*, ETNode*> ETMapType;
+
+
+  /// dominates - Return true if A dominates B.
+  ///
+  inline bool dominates(BasicBlock *A, BasicBlock *B) const {
+    if (A == B)
+      return true;
+    
+    ETNode *NodeA = getNode(A);
+    ETNode *NodeB = getNode(B);
+    
+    if (DFSInfoValid)
+      return NodeB->DominatedBy(NodeA);
+    else
+      return NodeB->DominatedBySlow(NodeA);
+  }
+
+  /// properlyDominates - Return true if A dominates B and A != B.
+  ///
+  bool properlyDominates(BasicBlock *A, BasicBlock *B) const {
+    return dominates(A, B) && A != B;
+  }
+
+  /// Return the nearest common dominator of A and B.
+  BasicBlock *nearestCommonDominator(BasicBlock *A, BasicBlock *B) const  {
+    ETNode *NodeA = getNode(A);
+    ETNode *NodeB = getNode(B);
+    
+    ETNode *Common = NodeA->NCA(NodeB);
+    if (!Common)
+      return NULL;
+    return Common->getData<BasicBlock>();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    AU.setPreservesAll();
+    AU.addRequired<ImmediateDominators>();
+  }
+  //===--------------------------------------------------------------------===//
+  // API to update Forest information based on modifications
+  // to the CFG...
+
+  /// addNewBlock - Add a new block to the CFG, with the specified immediate
+  /// dominator.
+  ///
+  void addNewBlock(BasicBlock *BB, BasicBlock *IDom);
+
+  /// setImmediateDominator - Update the immediate dominator information to
+  /// change the current immediate dominator for the specified block
+  /// to another block.  This method requires that BB for NewIDom
+  /// already have an ETNode, otherwise just use addNewBlock.
+  ///
+  void setImmediateDominator(BasicBlock *BB, BasicBlock *NewIDom);
+  /// print - Convert to human readable form
+  ///
+  virtual void print(std::ostream &OS, const Module* = 0) const;
+protected:
+  /// getNode - return the (Post)DominatorTree node for the specified basic
+  /// block.  This is the same as using operator[] on this class.
+  ///
+  inline ETNode *getNode(BasicBlock *BB) const {
+    ETMapType::const_iterator i = Nodes.find(BB);
+    return (i != Nodes.end()) ? i->second : 0;
+  }
+
+  inline ETNode *operator[](BasicBlock *BB) const {
+    return getNode(BB);
+  }
+
+  void reset();
+  ETMapType Nodes;
+  bool DFSInfoValid;
+
+};
+
+//==-------------------------------------
+/// ETForest Class - Concrete subclass of ETForestBase that is used to
+/// compute a forwards ET-Forest.
+
+struct ETForest : public ETForestBase {
+  ETForest() : ETForestBase(false) {}
+
+  BasicBlock *getRoot() const {
+    assert(Roots.size() == 1 && "Should always have entry node!");
+    return Roots[0];
+  }
+
+  virtual bool runOnFunction(Function &F) {
+    reset();     // Reset from the last time we were run...
+    ImmediateDominators &ID = getAnalysis<ImmediateDominators>();
+    Roots = ID.getRoots();
+    calculate(ID);
+    return false;
+  }
+
+  void calculate(const ImmediateDominators &ID);
+  ETNode *getNodeForBlock(BasicBlock *BB);
+};
+
+//===-------------------------------------
 /// DominatorTree Class - Concrete subclass of DominatorTreeBase that is used to
 /// compute a normal dominator tree.
 ///
@@ -518,6 +631,7 @@
                               const DominatorTree::Node *Node);
 };
 
+
 // Make sure that any clients of this file link in Dominators.cpp
 static IncludeFile
 DOMINATORS_INCLUDE_FILE((void*)&DominatorSet::stub);


Index: llvm/include/llvm/Analysis/PostDominators.h
diff -u llvm/include/llvm/Analysis/PostDominators.h:1.9 llvm/include/llvm/Analysis/PostDominators.h:1.10
--- llvm/include/llvm/Analysis/PostDominators.h:1.9	Thu Apr 21 15:16:32 2005
+++ llvm/include/llvm/Analysis/PostDominators.h	Sun Jan  8 02:19:58 2006
@@ -84,6 +84,29 @@
 };
 
 
+/// PostETForest Class - Concrete subclass of ETForestBase that is used to
+/// compute a forwards post-dominator ET-Forest.
+struct PostETForest : public ETForestBase {
+  PostETForest() : ETForestBase(true) {}
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    AU.setPreservesAll();
+    AU.addRequired<ImmediatePostDominators>();
+  }
+
+  virtual bool runOnFunction(Function &F) {
+    reset();     // Reset from the last time we were run...
+    ImmediatePostDominators &ID = getAnalysis<ImmediatePostDominators>();
+    Roots = ID.getRoots();
+    calculate(ID);
+    return false;
+  }
+
+  void calculate(const ImmediatePostDominators &ID);
+  ETNode *getNodeForBlock(BasicBlock *BB);
+};
+
+
 /// PostDominanceFrontier Class - Concrete subclass of DominanceFrontier that is
 /// used to compute the a post-dominance frontier.
 ///






More information about the llvm-commits mailing list