aboutsummaryrefslogtreecommitdiffstats
path: root/lib/VMCore/DominatorCalculation.h
blob: afdc92c1d99303157f11e62f9536d4f5cd10d27b (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
//==- llvm/VMCore/DominatorCalculation.h - Dominator Calculation -*- C++ -*-==//
//
//                     The LLVM Compiler Infrastructure
//
// This file was developed by Owen Anderson and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//

#ifndef LLVM_VMCORE_DOMINATOR_CALCULATION_H
#define LLVM_VMCORE_DOMINATOR_CALCULATION_H

#include "llvm/Analysis/Dominators.h"

//===----------------------------------------------------------------------===//
//
// DominatorTree construction - This pass constructs immediate dominator
// information for a flow-graph based on the algorithm described in this
// document:
//
//   A Fast Algorithm for Finding Dominators in a Flowgraph
//   T. Lengauer & R. Tarjan, ACM TOPLAS July 1979, pgs 121-141.
//
// This implements both the O(n*ack(n)) and the O(n*log(n)) versions of EVAL and
// LINK, but it turns out that the theoretically slower O(n*log(n))
// implementation is actually faster than the "efficient" algorithm (even for
// large CFGs) because the constant overheads are substantially smaller.  The
// lower-complexity version can be enabled with the following #define:
//
#define BALANCE_IDOM_TREE 0
//
//===----------------------------------------------------------------------===//

namespace llvm {

void DTCompress(DominatorTree& DT, BasicBlock *VIn) {

  std::vector<BasicBlock *> Work;
  SmallPtrSet<BasicBlock *, 32> Visited;
  BasicBlock *VInAncestor = DT.Info[VIn].Ancestor;
  DominatorTree::InfoRec &VInVAInfo = DT.Info[VInAncestor];

  if (VInVAInfo.Ancestor != 0)
    Work.push_back(VIn);
  
  while (!Work.empty()) {
    BasicBlock *V = Work.back();
    DominatorTree::InfoRec &VInfo = DT.Info[V];
    BasicBlock *VAncestor = VInfo.Ancestor;
    DominatorTree::InfoRec &VAInfo = DT.Info[VAncestor];

    // Process Ancestor first
    if (Visited.insert(VAncestor) &&
        VAInfo.Ancestor != 0) {
      Work.push_back(VAncestor);
      continue;
    } 
    Work.pop_back(); 

    // Update VInfo based on Ancestor info
    if (VAInfo.Ancestor == 0)
      continue;
    BasicBlock *VAncestorLabel = VAInfo.Label;
    BasicBlock *VLabel = VInfo.Label;
    if (DT.Info[VAncestorLabel].Semi < DT.Info[VLabel].Semi)
      VInfo.Label = VAncestorLabel;
    VInfo.Ancestor = VAInfo.Ancestor;
  }
}

BasicBlock *DTEval(DominatorTree& DT, BasicBlock *V) {
  DominatorTree::InfoRec &VInfo = DT.Info[V];
#if !BALANCE_IDOM_TREE
  // Higher-complexity but faster implementation
  if (VInfo.Ancestor == 0)
    return V;
  DTCompress(DT, V);
  return VInfo.Label;
#else
  // Lower-complexity but slower implementation
  if (VInfo.Ancestor == 0)
    return VInfo.Label;
  DTCompress(DT, V);
  BasicBlock *VLabel = VInfo.Label;

  BasicBlock *VAncestorLabel = DT.Info[VInfo.Ancestor].Label;
  if (DT.Info[VAncestorLabel].Semi >= DT.Info[VLabel].Semi)
    return VLabel;
  else
    return VAncestorLabel;
#endif
}

void DTLink(DominatorTree& DT, BasicBlock *V, BasicBlock *W,
            DominatorTree::InfoRec &WInfo) {
#if !BALANCE_IDOM_TREE
  // Higher-complexity but faster implementation
  WInfo.Ancestor = V;
#else
  // Lower-complexity but slower implementation
  BasicBlock *WLabel = WInfo.Label;
  unsigned WLabelSemi = Info[WLabel].Semi;
  BasicBlock *S = W;
  InfoRec *SInfo = &Info[S];

  BasicBlock *SChild = SInfo->Child;
  InfoRec *SChildInfo = &Info[SChild];

  while (WLabelSemi < Info[SChildInfo->Label].Semi) {
    BasicBlock *SChildChild = SChildInfo->Child;
    if (SInfo->Size+Info[SChildChild].Size >= 2*SChildInfo->Size) {
      SChildInfo->Ancestor = S;
      SInfo->Child = SChild = SChildChild;
      SChildInfo = &Info[SChild];
    } else {
      SChildInfo->Size = SInfo->Size;
      S = SInfo->Ancestor = SChild;
      SInfo = SChildInfo;
      SChild = SChildChild;
      SChildInfo = &Info[SChild];
    }
  }

  InfoRec &VInfo = Info[V];
  SInfo->Label = WLabel;

  assert(V != W && "The optimization here will not work in this case!");
  unsigned WSize = WInfo.Size;
  unsigned VSize = (VInfo.Size += WSize);

  if (VSize < 2*WSize)
    std::swap(S, VInfo.Child);

  while (S) {
    SInfo = &Info[S];
    SInfo->Ancestor = V;
    S = SInfo->Child;
  }
#endif
}

void DTcalculate(DominatorTree& DT, Function &F) {
  BasicBlock* Root = DT.Roots[0];

  // Add a node for the root...
  DT.DomTreeNodes[Root] = DT.RootNode = new DomTreeNode(Root, 0);

  DT.Vertex.push_back(0);

  // Step #1: Number blocks in depth-first order and initialize variables used
  // in later stages of the algorithm.
  unsigned N = DT.DFSPass(Root, 0);

  for (unsigned i = N; i >= 2; --i) {
    BasicBlock *W = DT.Vertex[i];
    DominatorTree::InfoRec &WInfo = DT.Info[W];

    // Step #2: Calculate the semidominators of all vertices
    for (pred_iterator PI = pred_begin(W), E = pred_end(W); PI != E; ++PI)
      if (DT.Info.count(*PI)) {  // Only if this predecessor is reachable!
        unsigned SemiU = DT.Info[DTEval(DT, *PI)].Semi;
        if (SemiU < WInfo.Semi)
          WInfo.Semi = SemiU;
      }

    DT.Info[DT.Vertex[WInfo.Semi]].Bucket.push_back(W);

    BasicBlock *WParent = WInfo.Parent;
    DTLink(DT, WParent, W, WInfo);

    // Step #3: Implicitly define the immediate dominator of vertices
    std::vector<BasicBlock*> &WParentBucket = DT.Info[WParent].Bucket;
    while (!WParentBucket.empty()) {
      BasicBlock *V = WParentBucket.back();
      WParentBucket.pop_back();
      BasicBlock *U = DTEval(DT, V);
      DT.IDoms[V] = DT.Info[U].Semi < DT.Info[V].Semi ? U : WParent;
    }
  }

  // Step #4: Explicitly define the immediate dominator of each vertex
  for (unsigned i = 2; i <= N; ++i) {
    BasicBlock *W = DT.Vertex[i];
    BasicBlock *&WIDom = DT.IDoms[W];
    if (WIDom != DT.Vertex[DT.Info[W].Semi])
      WIDom = DT.IDoms[WIDom];
  }

  // Loop over all of the reachable blocks in the function...
  for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
    if (BasicBlock *ImmDom = DT.getIDom(I)) {  // Reachable block.
      DomTreeNode *BBNode = DT.DomTreeNodes[I];
      if (BBNode) continue;  // Haven't calculated this node yet?

      // Get or calculate the node for the immediate dominator
      DomTreeNode *IDomNode = DT.getNodeForBlock(ImmDom);

      // Add a new tree node for this BasicBlock, and link it as a child of
      // IDomNode
      DomTreeNode *C = new DomTreeNode(I, IDomNode);
      DT.DomTreeNodes[I] = IDomNode->addChild(C);
    }

  // Free temporary memory used to construct idom's
  DT.Info.clear();
  DT.IDoms.clear();
  std::vector<BasicBlock*>().swap(DT.Vertex);

  DT.updateDFSNumbers();
}

}
#endif