//===-- WebAssemblyCFGStackify.cpp - CFG Stackification -------------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// /// /// \file /// \brief This file implements a CFG stacking pass. /// /// This pass reorders the blocks in a function to put them into a reverse /// post-order [0], with special care to keep the order as similar as possible /// to the original order, and to keep loops contiguous even in the case of /// split backedges. /// /// Then, it inserts BLOCK and LOOP markers to mark the start of scopes, since /// scope boundaries serve as the labels for WebAssembly's control transfers. /// /// This is sufficient to convert arbitrary CFGs into a form that works on /// WebAssembly, provided that all loops are single-entry. /// /// [0] https://en.wikipedia.org/wiki/Depth-first_search#Vertex_orderings /// //===----------------------------------------------------------------------===// #include "WebAssembly.h" #include "MCTargetDesc/WebAssemblyMCTargetDesc.h" #include "WebAssemblySubtarget.h" #include "llvm/ADT/SCCIterator.h" #include "llvm/ADT/SetVector.h" #include "llvm/CodeGen/MachineDominators.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/Passes.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" using namespace llvm; #define DEBUG_TYPE "wasm-cfg-stackify" namespace { class WebAssemblyCFGStackify final : public MachineFunctionPass { const char *getPassName() const override { return "WebAssembly CFG Stackify"; } void getAnalysisUsage(AnalysisUsage &AU) const override { AU.setPreservesCFG(); AU.addRequired<MachineDominatorTree>(); AU.addPreserved<MachineDominatorTree>(); AU.addRequired<MachineLoopInfo>(); AU.addPreserved<MachineLoopInfo>(); MachineFunctionPass::getAnalysisUsage(AU); } bool runOnMachineFunction(MachineFunction &MF) override; public: static char ID; // Pass identification, replacement for typeid WebAssemblyCFGStackify() : MachineFunctionPass(ID) {} }; } // end anonymous namespace char WebAssemblyCFGStackify::ID = 0; FunctionPass *llvm::createWebAssemblyCFGStackify() { return new WebAssemblyCFGStackify(); } static void EliminateMultipleEntryLoops(MachineFunction &MF, const MachineLoopInfo &MLI) { SmallPtrSet<MachineBasicBlock *, 8> InSet; for (scc_iterator<MachineFunction *> I = scc_begin(&MF), E = scc_end(&MF); I != E; ++I) { const std::vector<MachineBasicBlock *> &CurrentSCC = *I; // Skip trivial SCCs. if (CurrentSCC.size() == 1) continue; InSet.insert(CurrentSCC.begin(), CurrentSCC.end()); MachineBasicBlock *Header = nullptr; for (MachineBasicBlock *MBB : CurrentSCC) { for (MachineBasicBlock *Pred : MBB->predecessors()) { if (InSet.count(Pred)) continue; if (!Header) { Header = MBB; break; } // TODO: Implement multiple-entry loops. report_fatal_error("multiple-entry loops are not supported yet"); } } assert(MLI.isLoopHeader(Header)); InSet.clear(); } } namespace { /// Post-order traversal stack entry. struct POStackEntry { MachineBasicBlock *MBB; SmallVector<MachineBasicBlock *, 0> Succs; POStackEntry(MachineBasicBlock *MBB, MachineFunction &MF, const MachineLoopInfo &MLI); }; } // end anonymous namespace static bool LoopContains(const MachineLoop *Loop, const MachineBasicBlock *MBB) { return Loop ? Loop->contains(MBB) : true; } POStackEntry::POStackEntry(MachineBasicBlock *MBB, MachineFunction &MF, const MachineLoopInfo &MLI) : MBB(MBB), Succs(MBB->successors()) { // RPO is not a unique form, since at every basic block with multiple // successors, the DFS has to pick which order to visit the successors in. // Sort them strategically (see below). MachineLoop *Loop = MLI.getLoopFor(MBB); MachineFunction::iterator Next = next(MachineFunction::iterator(MBB)); MachineBasicBlock *LayoutSucc = Next == MF.end() ? nullptr : &*Next; std::stable_sort( Succs.begin(), Succs.end(), [=, &MLI](const MachineBasicBlock *A, const MachineBasicBlock *B) { if (A == B) return false; // Keep loops contiguous by preferring the block that's in the same // loop. bool LoopContainsA = LoopContains(Loop, A); bool LoopContainsB = LoopContains(Loop, B); if (LoopContainsA && !LoopContainsB) return true; if (!LoopContainsA && LoopContainsB) return false; // Minimize perturbation by preferring the block which is the immediate // layout successor. if (A == LayoutSucc) return true; if (B == LayoutSucc) return false; // TODO: More sophisticated orderings may be profitable here. return false; }); } /// Return the "bottom" block of a loop. This differs from /// MachineLoop::getBottomBlock in that it works even if the loop is /// discontiguous. static MachineBasicBlock *LoopBottom(const MachineLoop *Loop) { MachineBasicBlock *Bottom = Loop->getHeader(); for (MachineBasicBlock *MBB : Loop->blocks()) if (MBB->getNumber() > Bottom->getNumber()) Bottom = MBB; return Bottom; } /// Sort the blocks in RPO, taking special care to make sure that loops are /// contiguous even in the case of split backedges. /// /// TODO: Determine whether RPO is actually worthwhile, or whether we should /// move to just a stable-topological-sort-based approach that would preserve /// more of the original order. static void SortBlocks(MachineFunction &MF, const MachineLoopInfo &MLI) { // Note that we do our own RPO rather than using // "llvm/ADT/PostOrderIterator.h" because we want control over the order that // successors are visited in (see above). Also, we can sort the blocks in the // MachineFunction as we go. SmallPtrSet<MachineBasicBlock *, 16> Visited; SmallVector<POStackEntry, 16> Stack; MachineBasicBlock *EntryBlock = &*MF.begin(); Visited.insert(EntryBlock); Stack.push_back(POStackEntry(EntryBlock, MF, MLI)); for (;;) { POStackEntry &Entry = Stack.back(); SmallVectorImpl<MachineBasicBlock *> &Succs = Entry.Succs; if (!Succs.empty()) { MachineBasicBlock *Succ = Succs.pop_back_val(); if (Visited.insert(Succ).second) Stack.push_back(POStackEntry(Succ, MF, MLI)); continue; } // Put the block in its position in the MachineFunction. MachineBasicBlock &MBB = *Entry.MBB; MBB.moveBefore(&*MF.begin()); // Branch instructions may utilize a fallthrough, so update them if a // fallthrough has been added or removed. if (!MBB.empty() && MBB.back().isTerminator() && !MBB.back().isBranch() && !MBB.back().isBarrier()) report_fatal_error( "Non-branch terminator with fallthrough cannot yet be rewritten"); if (MBB.empty() || !MBB.back().isTerminator() || MBB.back().isBranch()) MBB.updateTerminator(); Stack.pop_back(); if (Stack.empty()) break; } // Now that we've sorted the blocks in RPO, renumber them. MF.RenumberBlocks(); #ifndef NDEBUG SmallSetVector<MachineLoop *, 8> OnStack; // Insert a sentinel representing the degenerate loop that starts at the // function entry block and includes the entire function as a "loop" that // executes once. OnStack.insert(nullptr); for (auto &MBB : MF) { assert(MBB.getNumber() >= 0 && "Renumbered blocks should be non-negative."); MachineLoop *Loop = MLI.getLoopFor(&MBB); if (Loop && &MBB == Loop->getHeader()) { // Loop header. The loop predecessor should be sorted above, and the other // predecessors should be backedges below. for (auto Pred : MBB.predecessors()) assert( (Pred->getNumber() < MBB.getNumber() || Loop->contains(Pred)) && "Loop header predecessors must be loop predecessors or backedges"); assert(OnStack.insert(Loop) && "Loops should be declared at most once."); } else { // Not a loop header. All predecessors should be sorted above. for (auto Pred : MBB.predecessors()) assert(Pred->getNumber() < MBB.getNumber() && "Non-loop-header predecessors should be topologically sorted"); assert(OnStack.count(MLI.getLoopFor(&MBB)) && "Blocks must be nested in their loops"); } while (OnStack.size() > 1 && &MBB == LoopBottom(OnStack.back())) OnStack.pop_back(); } assert(OnStack.pop_back_val() == nullptr && "The function entry block shouldn't actually be a loop header"); assert(OnStack.empty() && "Control flow stack pushes and pops should be balanced."); #endif } /// Test whether Pred has any terminators explicitly branching to MBB, as /// opposed to falling through. Note that it's possible (eg. in unoptimized /// code) for a branch instruction to both branch to a block and fallthrough /// to it, so we check the actual branch operands to see if there are any /// explicit mentions. static bool ExplicitlyBranchesTo(MachineBasicBlock *Pred, MachineBasicBlock *MBB) { for (MachineInstr &MI : Pred->terminators()) for (MachineOperand &MO : MI.explicit_operands()) if (MO.isMBB() && MO.getMBB() == MBB) return true; return false; } /// Insert a BLOCK marker for branches to MBB (if needed). static void PlaceBlockMarker(MachineBasicBlock &MBB, MachineFunction &MF, SmallVectorImpl<MachineBasicBlock *> &ScopeTops, const WebAssemblyInstrInfo &TII, const MachineLoopInfo &MLI, MachineDominatorTree &MDT) { // First compute the nearest common dominator of all forward non-fallthrough // predecessors so that we minimize the time that the BLOCK is on the stack, // which reduces overall stack height. MachineBasicBlock *Header = nullptr; bool IsBranchedTo = false; int MBBNumber = MBB.getNumber(); for (MachineBasicBlock *Pred : MBB.predecessors()) if (Pred->getNumber() < MBBNumber) { Header = Header ? MDT.findNearestCommonDominator(Header, Pred) : Pred; if (ExplicitlyBranchesTo(Pred, &MBB)) IsBranchedTo = true; } if (!Header) return; if (!IsBranchedTo) return; assert(&MBB != &MF.front() && "Header blocks shouldn't have predecessors"); MachineBasicBlock *LayoutPred = &*prev(MachineFunction::iterator(&MBB)); // If the nearest common dominator is inside a more deeply nested context, // walk out to the nearest scope which isn't more deeply nested. for (MachineFunction::iterator I(LayoutPred), E(Header); I != E; --I) { if (MachineBasicBlock *ScopeTop = ScopeTops[I->getNumber()]) { if (ScopeTop->getNumber() > Header->getNumber()) { // Skip over an intervening scope. I = next(MachineFunction::iterator(ScopeTop)); } else { // We found a scope level at an appropriate depth. Header = ScopeTop; break; } } } // If there's a loop which ends just before MBB which contains Header, we can // reuse its label instead of inserting a new BLOCK. for (MachineLoop *Loop = MLI.getLoopFor(LayoutPred); Loop && Loop->contains(LayoutPred); Loop = Loop->getParentLoop()) if (Loop && LoopBottom(Loop) == LayoutPred && Loop->contains(Header)) return; // Decide where in Header to put the BLOCK. MachineBasicBlock::iterator InsertPos; MachineLoop *HeaderLoop = MLI.getLoopFor(Header); if (HeaderLoop && MBB.getNumber() > LoopBottom(HeaderLoop)->getNumber()) { // Header is the header of a loop that does not lexically contain MBB, so // the BLOCK needs to be above the LOOP. InsertPos = Header->begin(); } else { // Otherwise, insert the BLOCK as late in Header as we can, but before any // existing BLOCKs. InsertPos = Header->getFirstTerminator(); while (InsertPos != Header->begin() && prev(InsertPos)->getOpcode() == WebAssembly::BLOCK) --InsertPos; } // Add the BLOCK. BuildMI(*Header, InsertPos, DebugLoc(), TII.get(WebAssembly::BLOCK)) .addMBB(&MBB); // Track the farthest-spanning scope that ends at this point. int Number = MBB.getNumber(); if (!ScopeTops[Number] || ScopeTops[Number]->getNumber() > Header->getNumber()) ScopeTops[Number] = Header; } /// Insert a LOOP marker for a loop starting at MBB (if it's a loop header). static void PlaceLoopMarker(MachineBasicBlock &MBB, MachineFunction &MF, SmallVectorImpl<MachineBasicBlock *> &ScopeTops, const WebAssemblyInstrInfo &TII, const MachineLoopInfo &MLI) { MachineLoop *Loop = MLI.getLoopFor(&MBB); if (!Loop || Loop->getHeader() != &MBB) return; // The operand of a LOOP is the first block after the loop. If the loop is the // bottom of the function, insert a dummy block at the end. MachineBasicBlock *Bottom = LoopBottom(Loop); auto Iter = next(MachineFunction::iterator(Bottom)); if (Iter == MF.end()) { MachineBasicBlock *Label = MF.CreateMachineBasicBlock(); // Give it a fake predecessor so that AsmPrinter prints its label. Label->addSuccessor(Label); MF.push_back(Label); Iter = next(MachineFunction::iterator(Bottom)); } MachineBasicBlock *AfterLoop = &*Iter; BuildMI(MBB, MBB.begin(), DebugLoc(), TII.get(WebAssembly::LOOP)) .addMBB(AfterLoop); // Emit a special no-op telling the asm printer that we need a label to close // the loop scope, even though the destination is only reachable by // fallthrough. if (!Bottom->back().isBarrier()) BuildMI(*Bottom, Bottom->end(), DebugLoc(), TII.get(WebAssembly::LOOP_END)); assert((!ScopeTops[AfterLoop->getNumber()] || ScopeTops[AfterLoop->getNumber()]->getNumber() < MBB.getNumber()) && "With RPO we should visit the outer-most loop for a block first."); if (!ScopeTops[AfterLoop->getNumber()]) ScopeTops[AfterLoop->getNumber()] = &MBB; } /// Insert LOOP and BLOCK markers at appropriate places. static void PlaceMarkers(MachineFunction &MF, const MachineLoopInfo &MLI, const WebAssemblyInstrInfo &TII, MachineDominatorTree &MDT) { // For each block whose label represents the end of a scope, record the block // which holds the beginning of the scope. This will allow us to quickly skip // over scoped regions when walking blocks. We allocate one more than the // number of blocks in the function to accommodate for the possible fake block // we may insert at the end. SmallVector<MachineBasicBlock *, 8> ScopeTops(MF.getNumBlockIDs() + 1); for (auto &MBB : MF) { // Place the LOOP for MBB if MBB is the header of a loop. PlaceLoopMarker(MBB, MF, ScopeTops, TII, MLI); // Place the BLOCK for MBB if MBB is branched to from above. PlaceBlockMarker(MBB, MF, ScopeTops, TII, MLI, MDT); } } #ifndef NDEBUG static bool IsOnStack(const SmallVectorImpl<std::pair<MachineBasicBlock *, bool>> &Stack, const MachineBasicBlock *MBB) { for (const auto &Pair : Stack) if (Pair.first == MBB) return true; return false; } #endif bool WebAssemblyCFGStackify::runOnMachineFunction(MachineFunction &MF) { DEBUG(dbgs() << "********** CFG Stackifying **********\n" "********** Function: " << MF.getName() << '\n'); const auto &MLI = getAnalysis<MachineLoopInfo>(); auto &MDT = getAnalysis<MachineDominatorTree>(); const auto &TII = *MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo(); // RPO sorting needs all loops to be single-entry. EliminateMultipleEntryLoops(MF, MLI); // Sort the blocks in RPO, with contiguous loops. SortBlocks(MF, MLI); // Place the BLOCK and LOOP markers to indicate the beginnings of scopes. PlaceMarkers(MF, MLI, TII, MDT); #ifndef NDEBUG // Verify that block and loop beginnings and endings are in LIFO order, and // that all references to blocks are to blocks on the stack at the point of // the reference. SmallVector<std::pair<MachineBasicBlock *, bool>, 0> Stack; for (auto &MBB : MF) { while (!Stack.empty() && Stack.back().first == &MBB) if (Stack.back().second) { assert(Stack.size() >= 2); Stack.pop_back(); Stack.pop_back(); } else { assert(Stack.size() >= 1); Stack.pop_back(); } for (auto &MI : MBB) switch (MI.getOpcode()) { case WebAssembly::LOOP: Stack.push_back(std::make_pair(&MBB, false)); Stack.push_back(std::make_pair(MI.getOperand(0).getMBB(), true)); break; case WebAssembly::BLOCK: Stack.push_back(std::make_pair(MI.getOperand(0).getMBB(), false)); break; default: // Verify that all referenced blocks are in scope. A reference to a // block with a negative number is invalid, but can happen with inline // asm, so we shouldn't assert on it, but instead let CodeGen properly // fail on it. for (const MachineOperand &MO : MI.explicit_operands()) if (MO.isMBB() && MO.getMBB()->getNumber() >= 0) assert(IsOnStack(Stack, MO.getMBB())); break; } } assert(Stack.empty()); #endif return true; }