llvm-project/llvm/lib/CodeGen/MachineDominators.cpp

128 lines
4.1 KiB
C++
Raw Normal View History

//===- MachineDominators.cpp - Machine Dominator Calculation --------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements simple dominator construction algorithms for finding
// forward dominators on machine functions.
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/ADT/SmallBitVector.h"
using namespace llvm;
namespace llvm {
template class DomTreeNodeBase<MachineBasicBlock>;
template class DominatorTreeBase<MachineBasicBlock>;
}
char MachineDominatorTree::ID = 0;
INITIALIZE_PASS(MachineDominatorTree, "machinedomtree",
"MachineDominator Tree Construction", true, true)
char &llvm::MachineDominatorsID = MachineDominatorTree::ID;
void MachineDominatorTree::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
MachineFunctionPass::getAnalysisUsage(AU);
}
bool MachineDominatorTree::runOnMachineFunction(MachineFunction &F) {
[MachineDominatorTree] Provide a method to inform a MachineDominatorTree that a critical edge has been split. The MachineDominatorTree will when lazy update the underlying dominance properties when require. ** Context ** This is a follow-up of r215410. Each time a critical edge is split this invalidates the dominator tree information. Thus, subsequent queries of that interface will be slow until the underlying information is actually recomputed (costly). ** Problem ** Prior to this patch, splitting a critical edge needed to query the dominator tree to update the dominator information. Therefore, splitting a bunch of critical edges will likely produce poor performance as each query to the dominator tree will use the slow query path. This happens a lot in passes like MachineSink and PHIElimination. ** Proposed Solution ** Splitting a critical edge is a local modification of the CFG. Moreover, as soon as a critical edge is split, it is not critical anymore and thus cannot be a candidate for critical edge splitting anymore. In other words, the predecessor and successor of a basic block inserted on a critical edge cannot be inserted by critical edge splitting. Using these observations, we can pile up the splitting of critical edge and apply then at once before updating the DT information. The core of this patch moves the update of the MachineDominatorTree information from MachineBasicBlock::SplitCriticalEdge to a lazy MachineDominatorTree. ** Performance ** Thanks to this patch, the motivating example compiles in 4- minutes instead of 6+ minutes. No test case added as the motivating example as nothing special but being huge! The binaries are strictly identical for all the llvm test-suite + SPECs with and without this patch for both Os and O3. Regarding compile time, I observed only noise, although on average I saw a small improvement. <rdar://problem/17894619> llvm-svn: 215576
2014-08-14 05:00:07 +08:00
CriticalEdgesToSplit.clear();
NewBBs.clear();
DT->recalculate(F);
return false;
}
MachineDominatorTree::MachineDominatorTree()
: MachineFunctionPass(ID) {
initializeMachineDominatorTreePass(*PassRegistry::getPassRegistry());
DT = new DominatorTreeBase<MachineBasicBlock>(false);
}
MachineDominatorTree::~MachineDominatorTree() {
delete DT;
}
void MachineDominatorTree::releaseMemory() {
DT->releaseMemory();
}
void MachineDominatorTree::print(raw_ostream &OS, const Module*) const {
DT->print(OS);
}
void MachineDominatorTree::applySplitCriticalEdges() const {
// Bail out early if there is nothing to do.
if (CriticalEdgesToSplit.empty())
return;
// For each element in CriticalEdgesToSplit, remember whether or not element
// is the new immediate domminator of its successor. The mapping is done by
// index, i.e., the information for the ith element of CriticalEdgesToSplit is
// the ith element of IsNewIDom.
SmallBitVector IsNewIDom(CriticalEdgesToSplit.size(), true);
size_t Idx = 0;
// Collect all the dominance properties info, before invalidating
// the underlying DT.
for (CriticalEdge &Edge : CriticalEdgesToSplit) {
// Update dominator information.
MachineBasicBlock *Succ = Edge.ToBB;
MachineDomTreeNode *SuccDTNode = DT->getNode(Succ);
for (MachineBasicBlock *PredBB : Succ->predecessors()) {
if (PredBB == Edge.NewBB)
continue;
// If we are in this situation:
// FromBB1 FromBB2
// + +
// + + + +
// + + + +
// ... Split1 Split2 ...
// + +
// + +
// +
// Succ
// Instead of checking the domiance property with Split2, we check it with
// FromBB2 since Split2 is still unknown of the underlying DT structure.
if (NewBBs.count(PredBB)) {
assert(PredBB->pred_size() == 1 && "A basic block resulting from a "
"critical edge split has more "
"than one predecessor!");
PredBB = *PredBB->pred_begin();
}
if (!DT->dominates(SuccDTNode, DT->getNode(PredBB))) {
IsNewIDom[Idx] = false;
break;
}
}
++Idx;
}
// Now, update DT with the collected dominance properties info.
Idx = 0;
for (CriticalEdge &Edge : CriticalEdgesToSplit) {
// We know FromBB dominates NewBB.
MachineDomTreeNode *NewDTNode = DT->addNewBlock(Edge.NewBB, Edge.FromBB);
// If all the other predecessors of "Succ" are dominated by "Succ" itself
// then the new block is the new immediate dominator of "Succ". Otherwise,
// the new block doesn't dominate anything.
if (IsNewIDom[Idx])
DT->changeImmediateDominator(DT->getNode(Edge.ToBB), NewDTNode);
++Idx;
}
NewBBs.clear();
CriticalEdgesToSplit.clear();
}