1 //===-- WebAssemblyCFGSort.cpp - CFG Sorting ------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// \brief This file implements a CFG sorting pass.
12 ///
13 /// This pass reorders the blocks in a function to put them into topological
14 /// order, ignoring loop backedges, and without any loop being interrupted
15 /// by a block not dominated by the loop header, with special care to keep the
16 /// order as similar as possible to the original order.
17 ///
18 ////===----------------------------------------------------------------------===//
19 
20 #include "WebAssembly.h"
21 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
22 #include "WebAssemblySubtarget.h"
23 #include "WebAssemblyUtilities.h"
24 #include "llvm/ADT/PriorityQueue.h"
25 #include "llvm/ADT/SetVector.h"
26 #include "llvm/CodeGen/MachineDominators.h"
27 #include "llvm/CodeGen/MachineFunction.h"
28 #include "llvm/CodeGen/MachineLoopInfo.h"
29 #include "llvm/CodeGen/MachineRegisterInfo.h"
30 #include "llvm/CodeGen/Passes.h"
31 #include "llvm/Support/Debug.h"
32 #include "llvm/Support/raw_ostream.h"
33 using namespace llvm;
34 
35 #define DEBUG_TYPE "wasm-cfg-sort"
36 
37 namespace {
38 class WebAssemblyCFGSort final : public MachineFunctionPass {
39   StringRef getPassName() const override { return "WebAssembly CFG Sort"; }
40 
41   void getAnalysisUsage(AnalysisUsage &AU) const override {
42     AU.setPreservesCFG();
43     AU.addRequired<MachineDominatorTree>();
44     AU.addPreserved<MachineDominatorTree>();
45     AU.addRequired<MachineLoopInfo>();
46     AU.addPreserved<MachineLoopInfo>();
47     MachineFunctionPass::getAnalysisUsage(AU);
48   }
49 
50   bool runOnMachineFunction(MachineFunction &MF) override;
51 
52 public:
53   static char ID; // Pass identification, replacement for typeid
54   WebAssemblyCFGSort() : MachineFunctionPass(ID) {}
55 };
56 } // end anonymous namespace
57 
58 char WebAssemblyCFGSort::ID = 0;
59 FunctionPass *llvm::createWebAssemblyCFGSort() {
60   return new WebAssemblyCFGSort();
61 }
62 
63 static void MaybeUpdateTerminator(MachineBasicBlock *MBB) {
64 #ifndef NDEBUG
65   bool AnyBarrier = false;
66 #endif
67   bool AllAnalyzable = true;
68   for (const MachineInstr &Term : MBB->terminators()) {
69 #ifndef NDEBUG
70     AnyBarrier |= Term.isBarrier();
71 #endif
72     AllAnalyzable &= Term.isBranch() && !Term.isIndirectBranch();
73   }
74   assert((AnyBarrier || AllAnalyzable) &&
75          "AnalyzeBranch needs to analyze any block with a fallthrough");
76   if (AllAnalyzable)
77     MBB->updateTerminator();
78 }
79 
80 namespace {
81 /// Sort blocks by their number.
82 struct CompareBlockNumbers {
83   bool operator()(const MachineBasicBlock *A,
84                   const MachineBasicBlock *B) const {
85     return A->getNumber() > B->getNumber();
86   }
87 };
88 /// Sort blocks by their number in the opposite order..
89 struct CompareBlockNumbersBackwards {
90   bool operator()(const MachineBasicBlock *A,
91                   const MachineBasicBlock *B) const {
92     return A->getNumber() < B->getNumber();
93   }
94 };
95 /// Bookkeeping for a loop to help ensure that we don't mix blocks not dominated
96 /// by the loop header among the loop's blocks.
97 struct Entry {
98   const MachineLoop *Loop;
99   unsigned NumBlocksLeft;
100 
101   /// List of blocks not dominated by Loop's header that are deferred until
102   /// after all of Loop's blocks have been seen.
103   std::vector<MachineBasicBlock *> Deferred;
104 
105   explicit Entry(const MachineLoop *L)
106       : Loop(L), NumBlocksLeft(L->getNumBlocks()) {}
107 };
108 } // end anonymous namespace
109 
110 /// Sort the blocks, taking special care to make sure that loops are not
111 /// interrupted by blocks not dominated by their header.
112 /// TODO: There are many opportunities for improving the heuristics here.
113 /// Explore them.
114 static void SortBlocks(MachineFunction &MF, const MachineLoopInfo &MLI,
115                        const MachineDominatorTree &MDT) {
116   // Prepare for a topological sort: Record the number of predecessors each
117   // block has, ignoring loop backedges.
118   MF.RenumberBlocks();
119   SmallVector<unsigned, 16> NumPredsLeft(MF.getNumBlockIDs(), 0);
120   for (MachineBasicBlock &MBB : MF) {
121     unsigned N = MBB.pred_size();
122     if (MachineLoop *L = MLI.getLoopFor(&MBB))
123       if (L->getHeader() == &MBB)
124         for (const MachineBasicBlock *Pred : MBB.predecessors())
125           if (L->contains(Pred))
126             --N;
127     NumPredsLeft[MBB.getNumber()] = N;
128   }
129 
130   // Topological sort the CFG, with additional constraints:
131   //  - Between a loop header and the last block in the loop, there can be
132   //    no blocks not dominated by the loop header.
133   //  - It's desirable to preserve the original block order when possible.
134   // We use two ready lists; Preferred and Ready. Preferred has recently
135   // processed sucessors, to help preserve block sequences from the original
136   // order. Ready has the remaining ready blocks.
137   PriorityQueue<MachineBasicBlock *, std::vector<MachineBasicBlock *>,
138                 CompareBlockNumbers>
139       Preferred;
140   PriorityQueue<MachineBasicBlock *, std::vector<MachineBasicBlock *>,
141                 CompareBlockNumbersBackwards>
142       Ready;
143   SmallVector<Entry, 4> Loops;
144   for (MachineBasicBlock *MBB = &MF.front();;) {
145     const MachineLoop *L = MLI.getLoopFor(MBB);
146     if (L) {
147       // If MBB is a loop header, add it to the active loop list. We can't put
148       // any blocks that it doesn't dominate until we see the end of the loop.
149       if (L->getHeader() == MBB)
150         Loops.push_back(Entry(L));
151       // For each active loop the block is in, decrement the count. If MBB is
152       // the last block in an active loop, take it off the list and pick up any
153       // blocks deferred because the header didn't dominate them.
154       for (Entry &E : Loops)
155         if (E.Loop->contains(MBB) && --E.NumBlocksLeft == 0)
156           for (auto DeferredBlock : E.Deferred)
157             Ready.push(DeferredBlock);
158       while (!Loops.empty() && Loops.back().NumBlocksLeft == 0)
159         Loops.pop_back();
160     }
161     // The main topological sort logic.
162     for (MachineBasicBlock *Succ : MBB->successors()) {
163       // Ignore backedges.
164       if (MachineLoop *SuccL = MLI.getLoopFor(Succ))
165         if (SuccL->getHeader() == Succ && SuccL->contains(MBB))
166           continue;
167       // Decrement the predecessor count. If it's now zero, it's ready.
168       if (--NumPredsLeft[Succ->getNumber()] == 0)
169         Preferred.push(Succ);
170     }
171     // Determine the block to follow MBB. First try to find a preferred block,
172     // to preserve the original block order when possible.
173     MachineBasicBlock *Next = nullptr;
174     while (!Preferred.empty()) {
175       Next = Preferred.top();
176       Preferred.pop();
177       // If X isn't dominated by the top active loop header, defer it until that
178       // loop is done.
179       if (!Loops.empty() &&
180           !MDT.dominates(Loops.back().Loop->getHeader(), Next)) {
181         Loops.back().Deferred.push_back(Next);
182         Next = nullptr;
183         continue;
184       }
185       // If Next was originally ordered before MBB, and it isn't because it was
186       // loop-rotated above the header, it's not preferred.
187       if (Next->getNumber() < MBB->getNumber() &&
188           (!L || !L->contains(Next) ||
189            L->getHeader()->getNumber() < Next->getNumber())) {
190         Ready.push(Next);
191         Next = nullptr;
192         continue;
193       }
194       break;
195     }
196     // If we didn't find a suitable block in the Preferred list, check the
197     // general Ready list.
198     if (!Next) {
199       // If there are no more blocks to process, we're done.
200       if (Ready.empty()) {
201         MaybeUpdateTerminator(MBB);
202         break;
203       }
204       for (;;) {
205         Next = Ready.top();
206         Ready.pop();
207         // If Next isn't dominated by the top active loop header, defer it until
208         // that loop is done.
209         if (!Loops.empty() &&
210             !MDT.dominates(Loops.back().Loop->getHeader(), Next)) {
211           Loops.back().Deferred.push_back(Next);
212           continue;
213         }
214         break;
215       }
216     }
217     // Move the next block into place and iterate.
218     Next->moveAfter(MBB);
219     MaybeUpdateTerminator(MBB);
220     MBB = Next;
221   }
222   assert(Loops.empty() && "Active loop list not finished");
223   MF.RenumberBlocks();
224 
225 #ifndef NDEBUG
226   SmallSetVector<MachineLoop *, 8> OnStack;
227 
228   // Insert a sentinel representing the degenerate loop that starts at the
229   // function entry block and includes the entire function as a "loop" that
230   // executes once.
231   OnStack.insert(nullptr);
232 
233   for (auto &MBB : MF) {
234     assert(MBB.getNumber() >= 0 && "Renumbered blocks should be non-negative.");
235 
236     MachineLoop *Loop = MLI.getLoopFor(&MBB);
237     if (Loop && &MBB == Loop->getHeader()) {
238       // Loop header. The loop predecessor should be sorted above, and the other
239       // predecessors should be backedges below.
240       for (auto Pred : MBB.predecessors())
241         assert(
242             (Pred->getNumber() < MBB.getNumber() || Loop->contains(Pred)) &&
243             "Loop header predecessors must be loop predecessors or backedges");
244       assert(OnStack.insert(Loop) && "Loops should be declared at most once.");
245     } else {
246       // Not a loop header. All predecessors should be sorted above.
247       for (auto Pred : MBB.predecessors())
248         assert(Pred->getNumber() < MBB.getNumber() &&
249                "Non-loop-header predecessors should be topologically sorted");
250       assert(OnStack.count(MLI.getLoopFor(&MBB)) &&
251              "Blocks must be nested in their loops");
252     }
253     while (OnStack.size() > 1 && &MBB == LoopBottom(OnStack.back()))
254       OnStack.pop_back();
255   }
256   assert(OnStack.pop_back_val() == nullptr &&
257          "The function entry block shouldn't actually be a loop header");
258   assert(OnStack.empty() &&
259          "Control flow stack pushes and pops should be balanced.");
260 #endif
261 }
262 
263 bool WebAssemblyCFGSort::runOnMachineFunction(MachineFunction &MF) {
264   DEBUG(dbgs() << "********** CFG Sorting **********\n"
265                   "********** Function: "
266                << MF.getName() << '\n');
267 
268   const auto &MLI = getAnalysis<MachineLoopInfo>();
269   auto &MDT = getAnalysis<MachineDominatorTree>();
270   // Liveness is not tracked for VALUE_STACK physreg.
271   MF.getRegInfo().invalidateLiveness();
272 
273   // Sort the blocks, with contiguous loops.
274   SortBlocks(MF, MLI, MDT);
275 
276   return true;
277 }
278