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

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1857 lines
66 KiB
C++
Raw Normal View History

//===- SplitKit.cpp - Toolkit for splitting live ranges -------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file contains the SplitAnalysis class as well as mutator functions for
// live range splitting.
//
//===----------------------------------------------------------------------===//
#include "SplitKit.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/LiveInterval.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/LiveRangeCalc.h"
#include "llvm/CodeGen/LiveRangeEdit.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SlotIndexes.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/CodeGen/VirtRegMap.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/MC/LaneBitmask.h"
#include "llvm/Support/Allocator.h"
#include "llvm/Support/BlockFrequency.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <iterator>
#include <limits>
#include <tuple>
#include <utility>
using namespace llvm;
#define DEBUG_TYPE "regalloc"
STATISTIC(NumFinished, "Number of splits finished");
STATISTIC(NumSimple, "Number of splits that were simple");
STATISTIC(NumCopies, "Number of copies inserted for splitting");
STATISTIC(NumRemats, "Number of rematerialized defs for splitting");
STATISTIC(NumRepairs, "Number of invalid live ranges repaired");
//===----------------------------------------------------------------------===//
// Last Insert Point Analysis
//===----------------------------------------------------------------------===//
InsertPointAnalysis::InsertPointAnalysis(const LiveIntervals &lis,
unsigned BBNum)
: LIS(lis), LastInsertPoint(BBNum) {}
SlotIndex
InsertPointAnalysis::computeLastInsertPoint(const LiveInterval &CurLI,
const MachineBasicBlock &MBB) {
unsigned Num = MBB.getNumber();
std::pair<SlotIndex, SlotIndex> &LIP = LastInsertPoint[Num];
SlotIndex MBBEnd = LIS.getMBBEndIdx(&MBB);
SmallVector<const MachineBasicBlock *, 1> EHPadSuccessors;
for (const MachineBasicBlock *SMBB : MBB.successors())
if (SMBB->isEHPad())
EHPadSuccessors.push_back(SMBB);
// Compute insert points on the first call. The pair is independent of the
// current live interval.
if (!LIP.first.isValid()) {
MachineBasicBlock::const_iterator FirstTerm = MBB.getFirstTerminator();
if (FirstTerm == MBB.end())
LIP.first = MBBEnd;
else
LIP.first = LIS.getInstructionIndex(*FirstTerm);
// If there is a landing pad successor, also find the call instruction.
if (EHPadSuccessors.empty())
return LIP.first;
// There may not be a call instruction (?) in which case we ignore LPad.
LIP.second = LIP.first;
for (MachineBasicBlock::const_iterator I = MBB.end(), E = MBB.begin();
I != E;) {
--I;
if (I->isCall()) {
LIP.second = LIS.getInstructionIndex(*I);
break;
}
}
}
// If CurLI is live into a landing pad successor, move the last insert point
// back to the call that may throw.
if (!LIP.second)
return LIP.first;
if (none_of(EHPadSuccessors, [&](const MachineBasicBlock *EHPad) {
return LIS.isLiveInToMBB(CurLI, EHPad);
}))
return LIP.first;
// Find the value leaving MBB.
const VNInfo *VNI = CurLI.getVNInfoBefore(MBBEnd);
if (!VNI)
return LIP.first;
// If the value leaving MBB was defined after the call in MBB, it can't
// really be live-in to the landing pad. This can happen if the landing pad
// has a PHI, and this register is undef on the exceptional edge.
// <rdar://problem/10664933>
if (!SlotIndex::isEarlierInstr(VNI->def, LIP.second) && VNI->def < MBBEnd)
return LIP.first;
// Value is properly live-in to the landing pad.
// Only allow inserts before the call.
return LIP.second;
}
MachineBasicBlock::iterator
InsertPointAnalysis::getLastInsertPointIter(const LiveInterval &CurLI,
MachineBasicBlock &MBB) {
SlotIndex LIP = getLastInsertPoint(CurLI, MBB);
if (LIP == LIS.getMBBEndIdx(&MBB))
return MBB.end();
return LIS.getInstructionFromIndex(LIP);
}
//===----------------------------------------------------------------------===//
// Split Analysis
//===----------------------------------------------------------------------===//
SplitAnalysis::SplitAnalysis(const VirtRegMap &vrm, const LiveIntervals &lis,
const MachineLoopInfo &mli)
: MF(vrm.getMachineFunction()), VRM(vrm), LIS(lis), Loops(mli),
TII(*MF.getSubtarget().getInstrInfo()), IPA(lis, MF.getNumBlockIDs()) {}
void SplitAnalysis::clear() {
UseSlots.clear();
UseBlocks.clear();
ThroughBlocks.clear();
CurLI = nullptr;
DidRepairRange = false;
}
/// analyzeUses - Count instructions, basic blocks, and loops using CurLI.
void SplitAnalysis::analyzeUses() {
assert(UseSlots.empty() && "Call clear first");
// First get all the defs from the interval values. This provides the correct
// slots for early clobbers.
for (const VNInfo *VNI : CurLI->valnos)
if (!VNI->isPHIDef() && !VNI->isUnused())
UseSlots.push_back(VNI->def);
// Get use slots form the use-def chain.
const MachineRegisterInfo &MRI = MF.getRegInfo();
for (MachineOperand &MO : MRI.use_nodbg_operands(CurLI->reg))
if (!MO.isUndef())
UseSlots.push_back(LIS.getInstructionIndex(*MO.getParent()).getRegSlot());
array_pod_sort(UseSlots.begin(), UseSlots.end());
// Remove duplicates, keeping the smaller slot for each instruction.
// That is what we want for early clobbers.
UseSlots.erase(std::unique(UseSlots.begin(), UseSlots.end(),
SlotIndex::isSameInstr),
UseSlots.end());
// Compute per-live block info.
if (!calcLiveBlockInfo()) {
// FIXME: calcLiveBlockInfo found inconsistencies in the live range.
// I am looking at you, RegisterCoalescer!
DidRepairRange = true;
++NumRepairs;
LLVM_DEBUG(dbgs() << "*** Fixing inconsistent live interval! ***\n");
const_cast<LiveIntervals&>(LIS)
.shrinkToUses(const_cast<LiveInterval*>(CurLI));
UseBlocks.clear();
ThroughBlocks.clear();
bool fixed = calcLiveBlockInfo();
(void)fixed;
assert(fixed && "Couldn't fix broken live interval");
}
LLVM_DEBUG(dbgs() << "Analyze counted " << UseSlots.size() << " instrs in "
<< UseBlocks.size() << " blocks, through "
<< NumThroughBlocks << " blocks.\n");
}
/// calcLiveBlockInfo - Fill the LiveBlocks array with information about blocks
/// where CurLI is live.
bool SplitAnalysis::calcLiveBlockInfo() {
ThroughBlocks.resize(MF.getNumBlockIDs());
NumThroughBlocks = NumGapBlocks = 0;
if (CurLI->empty())
return true;
LiveInterval::const_iterator LVI = CurLI->begin();
LiveInterval::const_iterator LVE = CurLI->end();
SmallVectorImpl<SlotIndex>::const_iterator UseI, UseE;
UseI = UseSlots.begin();
UseE = UseSlots.end();
// Loop over basic blocks where CurLI is live.
MachineFunction::iterator MFI =
LIS.getMBBFromIndex(LVI->start)->getIterator();
while (true) {
BlockInfo BI;
BI.MBB = &*MFI;
SlotIndex Start, Stop;
std::tie(Start, Stop) = LIS.getSlotIndexes()->getMBBRange(BI.MBB);
// If the block contains no uses, the range must be live through. At one
// point, RegisterCoalescer could create dangling ranges that ended
// mid-block.
if (UseI == UseE || *UseI >= Stop) {
++NumThroughBlocks;
ThroughBlocks.set(BI.MBB->getNumber());
// The range shouldn't end mid-block if there are no uses. This shouldn't
// happen.
if (LVI->end < Stop)
return false;
} else {
// This block has uses. Find the first and last uses in the block.
BI.FirstInstr = *UseI;
assert(BI.FirstInstr >= Start);
do ++UseI;
while (UseI != UseE && *UseI < Stop);
BI.LastInstr = UseI[-1];
assert(BI.LastInstr < Stop);
// LVI is the first live segment overlapping MBB.
BI.LiveIn = LVI->start <= Start;
// When not live in, the first use should be a def.
if (!BI.LiveIn) {
assert(LVI->start == LVI->valno->def && "Dangling Segment start");
assert(LVI->start == BI.FirstInstr && "First instr should be a def");
BI.FirstDef = BI.FirstInstr;
}
// Look for gaps in the live range.
BI.LiveOut = true;
while (LVI->end < Stop) {
SlotIndex LastStop = LVI->end;
if (++LVI == LVE || LVI->start >= Stop) {
BI.LiveOut = false;
BI.LastInstr = LastStop;
break;
}
if (LastStop < LVI->start) {
// There is a gap in the live range. Create duplicate entries for the
// live-in snippet and the live-out snippet.
++NumGapBlocks;
// Push the Live-in part.
BI.LiveOut = false;
UseBlocks.push_back(BI);
UseBlocks.back().LastInstr = LastStop;
// Set up BI for the live-out part.
BI.LiveIn = false;
BI.LiveOut = true;
BI.FirstInstr = BI.FirstDef = LVI->start;
}
// A Segment that starts in the middle of the block must be a def.
assert(LVI->start == LVI->valno->def && "Dangling Segment start");
if (!BI.FirstDef)
BI.FirstDef = LVI->start;
}
UseBlocks.push_back(BI);
// LVI is now at LVE or LVI->end >= Stop.
if (LVI == LVE)
break;
}
// Live segment ends exactly at Stop. Move to the next segment.
if (LVI->end == Stop && ++LVI == LVE)
break;
// Pick the next basic block.
if (LVI->start < Stop)
++MFI;
else
MFI = LIS.getMBBFromIndex(LVI->start)->getIterator();
}
assert(getNumLiveBlocks() == countLiveBlocks(CurLI) && "Bad block count");
return true;
}
unsigned SplitAnalysis::countLiveBlocks(const LiveInterval *cli) const {
if (cli->empty())
return 0;
LiveInterval *li = const_cast<LiveInterval*>(cli);
LiveInterval::iterator LVI = li->begin();
LiveInterval::iterator LVE = li->end();
unsigned Count = 0;
// Loop over basic blocks where li is live.
MachineFunction::const_iterator MFI =
LIS.getMBBFromIndex(LVI->start)->getIterator();
SlotIndex Stop = LIS.getMBBEndIdx(&*MFI);
while (true) {
++Count;
LVI = li->advanceTo(LVI, Stop);
if (LVI == LVE)
return Count;
do {
++MFI;
Stop = LIS.getMBBEndIdx(&*MFI);
} while (Stop <= LVI->start);
}
}
bool SplitAnalysis::isOriginalEndpoint(SlotIndex Idx) const {
unsigned OrigReg = VRM.getOriginal(CurLI->reg);
const LiveInterval &Orig = LIS.getInterval(OrigReg);
assert(!Orig.empty() && "Splitting empty interval?");
LiveInterval::const_iterator I = Orig.find(Idx);
// Range containing Idx should begin at Idx.
if (I != Orig.end() && I->start <= Idx)
return I->start == Idx;
// Range does not contain Idx, previous must end at Idx.
return I != Orig.begin() && (--I)->end == Idx;
}
void SplitAnalysis::analyze(const LiveInterval *li) {
clear();
CurLI = li;
analyzeUses();
}
//===----------------------------------------------------------------------===//
// Split Editor
//===----------------------------------------------------------------------===//
/// Create a new SplitEditor for editing the LiveInterval analyzed by SA.
SplitEditor::SplitEditor(SplitAnalysis &sa, AliasAnalysis &aa,
LiveIntervals &lis, VirtRegMap &vrm,
MachineDominatorTree &mdt,
MachineBlockFrequencyInfo &mbfi)
: SA(sa), AA(aa), LIS(lis), VRM(vrm),
MRI(vrm.getMachineFunction().getRegInfo()), MDT(mdt),
TII(*vrm.getMachineFunction().getSubtarget().getInstrInfo()),
TRI(*vrm.getMachineFunction().getSubtarget().getRegisterInfo()),
MBFI(mbfi), RegAssign(Allocator) {}
void SplitEditor::reset(LiveRangeEdit &LRE, ComplementSpillMode SM) {
Edit = &LRE;
SpillMode = SM;
OpenIdx = 0;
RegAssign.clear();
Values.clear();
// Reset the LiveRangeCalc instances needed for this spill mode.
LRCalc[0].reset(&VRM.getMachineFunction(), LIS.getSlotIndexes(), &MDT,
&LIS.getVNInfoAllocator());
if (SpillMode)
LRCalc[1].reset(&VRM.getMachineFunction(), LIS.getSlotIndexes(), &MDT,
&LIS.getVNInfoAllocator());
// We don't need an AliasAnalysis since we will only be performing
// cheap-as-a-copy remats anyway.
Edit->anyRematerializable(nullptr);
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
LLVM_DUMP_METHOD void SplitEditor::dump() const {
if (RegAssign.empty()) {
dbgs() << " empty\n";
return;
}
for (RegAssignMap::const_iterator I = RegAssign.begin(); I.valid(); ++I)
dbgs() << " [" << I.start() << ';' << I.stop() << "):" << I.value();
dbgs() << '\n';
}
#endif
LiveInterval::SubRange &SplitEditor::getSubRangeForMask(LaneBitmask LM,
LiveInterval &LI) {
for (LiveInterval::SubRange &S : LI.subranges())
if (S.LaneMask == LM)
return S;
llvm_unreachable("SubRange for this mask not found");
}
void SplitEditor::addDeadDef(LiveInterval &LI, VNInfo *VNI, bool Original) {
if (!LI.hasSubRanges()) {
LI.createDeadDef(VNI);
return;
}
SlotIndex Def = VNI->def;
if (Original) {
// If we are transferring a def from the original interval, make sure
// to only update the subranges for which the original subranges had
// a def at this location.
for (LiveInterval::SubRange &S : LI.subranges()) {
auto &PS = getSubRangeForMask(S.LaneMask, Edit->getParent());
VNInfo *PV = PS.getVNInfoAt(Def);
if (PV != nullptr && PV->def == Def)
S.createDeadDef(Def, LIS.getVNInfoAllocator());
}
} else {
// This is a new def: either from rematerialization, or from an inserted
// copy. Since rematerialization can regenerate a definition of a sub-
// register, we need to check which subranges need to be updated.
const MachineInstr *DefMI = LIS.getInstructionFromIndex(Def);
assert(DefMI != nullptr);
LaneBitmask LM;
for (const MachineOperand &DefOp : DefMI->defs()) {
Apply llvm-prefer-register-over-unsigned from clang-tidy to LLVM Summary: This clang-tidy check is looking for unsigned integer variables whose initializer starts with an implicit cast from llvm::Register and changes the type of the variable to llvm::Register (dropping the llvm:: where possible). Partial reverts in: X86FrameLowering.cpp - Some functions return unsigned and arguably should be MCRegister X86FixupLEAs.cpp - Some functions return unsigned and arguably should be MCRegister X86FrameLowering.cpp - Some functions return unsigned and arguably should be MCRegister HexagonBitSimplify.cpp - Function takes BitTracker::RegisterRef which appears to be unsigned& MachineVerifier.cpp - Ambiguous operator==() given MCRegister and const Register PPCFastISel.cpp - No Register::operator-=() PeepholeOptimizer.cpp - TargetInstrInfo::optimizeLoadInstr() takes an unsigned& MachineTraceMetrics.cpp - MachineTraceMetrics lacks a suitable constructor Manual fixups in: ARMFastISel.cpp - ARMEmitLoad() now takes a Register& instead of unsigned& HexagonSplitDouble.cpp - Ternary operator was ambiguous between unsigned/Register HexagonConstExtenders.cpp - Has a local class named Register, used llvm::Register instead of Register. PPCFastISel.cpp - PPCEmitLoad() now takes a Register& instead of unsigned& Depends on D65919 Reviewers: arsenm, bogner, craig.topper, RKSimon Reviewed By: arsenm Subscribers: RKSimon, craig.topper, lenary, aemerson, wuzish, jholewinski, MatzeB, qcolombet, dschuff, jyknight, dylanmckay, sdardis, nemanjai, jvesely, wdng, nhaehnle, sbc100, jgravelle-google, kristof.beyls, hiraditya, aheejin, kbarton, fedor.sergeev, javed.absar, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, tpr, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, Jim, s.egerton, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65962 llvm-svn: 369041
2019-08-16 03:22:08 +08:00
Register R = DefOp.getReg();
if (R != LI.reg)
continue;
if (unsigned SR = DefOp.getSubReg())
LM |= TRI.getSubRegIndexLaneMask(SR);
else {
LM = MRI.getMaxLaneMaskForVReg(R);
break;
}
}
for (LiveInterval::SubRange &S : LI.subranges())
if ((S.LaneMask & LM).any())
S.createDeadDef(Def, LIS.getVNInfoAllocator());
}
}
VNInfo *SplitEditor::defValue(unsigned RegIdx,
const VNInfo *ParentVNI,
SlotIndex Idx,
bool Original) {
assert(ParentVNI && "Mapping NULL value");
assert(Idx.isValid() && "Invalid SlotIndex");
assert(Edit->getParent().getVNInfoAt(Idx) == ParentVNI && "Bad Parent VNI");
LiveInterval *LI = &LIS.getInterval(Edit->get(RegIdx));
// Create a new value.
VNInfo *VNI = LI->getNextValue(Idx, LIS.getVNInfoAllocator());
bool Force = LI->hasSubRanges();
ValueForcePair FP(Force ? nullptr : VNI, Force);
// Use insert for lookup, so we can add missing values with a second lookup.
std::pair<ValueMap::iterator, bool> InsP =
Values.insert(std::make_pair(std::make_pair(RegIdx, ParentVNI->id), FP));
// This was the first time (RegIdx, ParentVNI) was mapped, and it is not
// forced. Keep it as a simple def without any liveness.
if (!Force && InsP.second)
return VNI;
// If the previous value was a simple mapping, add liveness for it now.
if (VNInfo *OldVNI = InsP.first->second.getPointer()) {
addDeadDef(*LI, OldVNI, Original);
// No longer a simple mapping. Switch to a complex mapping. If the
// interval has subranges, make it a forced mapping.
InsP.first->second = ValueForcePair(nullptr, Force);
}
// This is a complex mapping, add liveness for VNI
addDeadDef(*LI, VNI, Original);
return VNI;
}
void SplitEditor::forceRecompute(unsigned RegIdx, const VNInfo &ParentVNI) {
ValueForcePair &VFP = Values[std::make_pair(RegIdx, ParentVNI.id)];
VNInfo *VNI = VFP.getPointer();
// ParentVNI was either unmapped or already complex mapped. Either way, just
// set the force bit.
if (!VNI) {
VFP.setInt(true);
return;
}
// This was previously a single mapping. Make sure the old def is represented
// by a trivial live range.
addDeadDef(LIS.getInterval(Edit->get(RegIdx)), VNI, false);
// Mark as complex mapped, forced.
VFP = ValueForcePair(nullptr, true);
}
SlotIndex SplitEditor::buildSingleSubRegCopy(unsigned FromReg, unsigned ToReg,
MachineBasicBlock &MBB, MachineBasicBlock::iterator InsertBefore,
unsigned SubIdx, LiveInterval &DestLI, bool Late, SlotIndex Def) {
const MCInstrDesc &Desc = TII.get(TargetOpcode::COPY);
bool FirstCopy = !Def.isValid();
MachineInstr *CopyMI = BuildMI(MBB, InsertBefore, DebugLoc(), Desc)
.addReg(ToReg, RegState::Define | getUndefRegState(FirstCopy)
| getInternalReadRegState(!FirstCopy), SubIdx)
.addReg(FromReg, 0, SubIdx);
BumpPtrAllocator &Allocator = LIS.getVNInfoAllocator();
SlotIndexes &Indexes = *LIS.getSlotIndexes();
if (FirstCopy) {
Def = Indexes.insertMachineInstrInMaps(*CopyMI, Late).getRegSlot();
} else {
CopyMI->bundleWithPred();
}
LaneBitmask LaneMask = TRI.getSubRegIndexLaneMask(SubIdx);
DestLI.refineSubRanges(Allocator, LaneMask,
[Def, &Allocator](LiveInterval::SubRange &SR) {
SR.createDeadDef(Def, Allocator);
},
Indexes, TRI);
return Def;
}
SlotIndex SplitEditor::buildCopy(unsigned FromReg, unsigned ToReg,
LaneBitmask LaneMask, MachineBasicBlock &MBB,
MachineBasicBlock::iterator InsertBefore, bool Late, unsigned RegIdx) {
const MCInstrDesc &Desc = TII.get(TargetOpcode::COPY);
if (LaneMask.all() || LaneMask == MRI.getMaxLaneMaskForVReg(FromReg)) {
// The full vreg is copied.
MachineInstr *CopyMI =
BuildMI(MBB, InsertBefore, DebugLoc(), Desc, ToReg).addReg(FromReg);
SlotIndexes &Indexes = *LIS.getSlotIndexes();
return Indexes.insertMachineInstrInMaps(*CopyMI, Late).getRegSlot();
}
// Only a subset of lanes needs to be copied. The following is a simple
// heuristic to construct a sequence of COPYs. We could add a target
// specific callback if this turns out to be suboptimal.
LiveInterval &DestLI = LIS.getInterval(Edit->get(RegIdx));
// First pass: Try to find a perfectly matching subregister index. If none
// exists find the one covering the most lanemask bits.
SmallVector<unsigned, 8> PossibleIndexes;
unsigned BestIdx = 0;
unsigned BestCover = 0;
const TargetRegisterClass *RC = MRI.getRegClass(FromReg);
assert(RC == MRI.getRegClass(ToReg) && "Should have same reg class");
for (unsigned Idx = 1, E = TRI.getNumSubRegIndices(); Idx < E; ++Idx) {
// Is this index even compatible with the given class?
if (TRI.getSubClassWithSubReg(RC, Idx) != RC)
continue;
LaneBitmask SubRegMask = TRI.getSubRegIndexLaneMask(Idx);
// Early exit if we found a perfect match.
if (SubRegMask == LaneMask) {
BestIdx = Idx;
break;
}
// The index must not cover any lanes outside \p LaneMask.
if ((SubRegMask & ~LaneMask).any())
continue;
unsigned PopCount = SubRegMask.getNumLanes();
PossibleIndexes.push_back(Idx);
if (PopCount > BestCover) {
BestCover = PopCount;
BestIdx = Idx;
}
}
// Abort if we cannot possibly implement the COPY with the given indexes.
if (BestIdx == 0)
report_fatal_error("Impossible to implement partial COPY");
SlotIndex Def = buildSingleSubRegCopy(FromReg, ToReg, MBB, InsertBefore,
BestIdx, DestLI, Late, SlotIndex());
// Greedy heuristic: Keep iterating keeping the best covering subreg index
// each time.
LaneBitmask LanesLeft = LaneMask & ~(TRI.getSubRegIndexLaneMask(BestIdx));
while (LanesLeft.any()) {
unsigned BestIdx = 0;
int BestCover = std::numeric_limits<int>::min();
for (unsigned Idx : PossibleIndexes) {
LaneBitmask SubRegMask = TRI.getSubRegIndexLaneMask(Idx);
// Early exit if we found a perfect match.
if (SubRegMask == LanesLeft) {
BestIdx = Idx;
break;
}
// Try to cover as much of the remaining lanes as possible but
// as few of the already covered lanes as possible.
int Cover = (SubRegMask & LanesLeft).getNumLanes()
- (SubRegMask & ~LanesLeft).getNumLanes();
if (Cover > BestCover) {
BestCover = Cover;
BestIdx = Idx;
}
}
if (BestIdx == 0)
report_fatal_error("Impossible to implement partial COPY");
buildSingleSubRegCopy(FromReg, ToReg, MBB, InsertBefore, BestIdx,
DestLI, Late, Def);
LanesLeft &= ~TRI.getSubRegIndexLaneMask(BestIdx);
}
return Def;
}
2011-02-03 14:18:29 +08:00
VNInfo *SplitEditor::defFromParent(unsigned RegIdx,
VNInfo *ParentVNI,
SlotIndex UseIdx,
MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) {
SlotIndex Def;
LiveInterval *LI = &LIS.getInterval(Edit->get(RegIdx));
// We may be trying to avoid interference that ends at a deleted instruction,
// so always begin RegIdx 0 early and all others late.
bool Late = RegIdx != 0;
// Attempt cheap-as-a-copy rematerialization.
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
unsigned Original = VRM.getOriginal(Edit->get(RegIdx));
LiveInterval &OrigLI = LIS.getInterval(Original);
VNInfo *OrigVNI = OrigLI.getVNInfoAt(UseIdx);
unsigned Reg = LI->reg;
bool DidRemat = false;
if (OrigVNI) {
LiveRangeEdit::Remat RM(ParentVNI);
RM.OrigMI = LIS.getInstructionFromIndex(OrigVNI->def);
if (Edit->canRematerializeAt(RM, OrigVNI, UseIdx, true)) {
Def = Edit->rematerializeAt(MBB, I, Reg, RM, TRI, Late);
++NumRemats;
DidRemat = true;
}
}
if (!DidRemat) {
LaneBitmask LaneMask;
if (LI->hasSubRanges()) {
LaneMask = LaneBitmask::getNone();
for (LiveInterval::SubRange &S : LI->subranges())
LaneMask |= S.LaneMask;
} else {
LaneMask = LaneBitmask::getAll();
}
++NumCopies;
Def = buildCopy(Edit->getReg(), Reg, LaneMask, MBB, I, Late, RegIdx);
}
// Define the value in Reg.
return defValue(RegIdx, ParentVNI, Def, false);
}
/// Create a new virtual register and live interval.
unsigned SplitEditor::openIntv() {
2011-02-03 14:18:29 +08:00
// Create the complement as index 0.
if (Edit->empty())
Edit->createEmptyInterval();
2011-02-03 14:18:29 +08:00
// Create the open interval.
OpenIdx = Edit->size();
Edit->createEmptyInterval();
return OpenIdx;
}
void SplitEditor::selectIntv(unsigned Idx) {
assert(Idx != 0 && "Cannot select the complement interval");
assert(Idx < Edit->size() && "Can only select previously opened interval");
LLVM_DEBUG(dbgs() << " selectIntv " << OpenIdx << " -> " << Idx << '\n');
OpenIdx = Idx;
}
SlotIndex SplitEditor::enterIntvBefore(SlotIndex Idx) {
2011-02-03 14:18:29 +08:00
assert(OpenIdx && "openIntv not called before enterIntvBefore");
LLVM_DEBUG(dbgs() << " enterIntvBefore " << Idx);
Idx = Idx.getBaseIndex();
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Idx);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
return Idx;
}
LLVM_DEBUG(dbgs() << ": valno " << ParentVNI->id << '\n');
MachineInstr *MI = LIS.getInstructionFromIndex(Idx);
assert(MI && "enterIntvBefore called with invalid index");
VNInfo *VNI = defFromParent(OpenIdx, ParentVNI, Idx, *MI->getParent(), MI);
return VNI->def;
}
Reapply r134047 now that the world is ready for it. This patch will sometimes choose live range split points next to interference instead of always splitting next to a register point. That means spill code can now appear almost anywhere, and it was necessary to fix code that didn't expect that. The difficult places were: - Between a CALL returning a value on the x87 stack and the corresponding FpPOP_RETVAL (was FpGET_ST0). Probably also near x87 inline assembly, but that didn't actually show up in testing. - Between a CALL popping arguments off the stack and the corresponding ADJCALLSTACKUP. Both are fixed now. The only place spill code can't appear is after terminators, see SplitAnalysis::getLastSplitPoint. Original commit message: Rewrite RAGreedy::splitAroundRegion, now with cool ASCII art. This function has to deal with a lot of special cases, and the old version got it wrong sometimes. In particular, it would sometimes leave multiple uses in the stack interval in a single block. That causes bad code with multiple reloads in the same basic block. The new version handles block entry and exit in a single pass. It first eliminates all the easy cases, and then goes on to create a local interval for the blocks with difficult interference. Previously, we would only create the local interval for completely isolated blocks. It can happen that the stack interval becomes completely empty because we could allocate a register in all edge bundles, and the new local intervals deal with the interference. The empty stack interval is harmless, but we need to remove a SplitKit assertion that checks for empty intervals. llvm-svn: 134125
2011-06-30 09:30:39 +08:00
SlotIndex SplitEditor::enterIntvAfter(SlotIndex Idx) {
assert(OpenIdx && "openIntv not called before enterIntvAfter");
LLVM_DEBUG(dbgs() << " enterIntvAfter " << Idx);
Reapply r134047 now that the world is ready for it. This patch will sometimes choose live range split points next to interference instead of always splitting next to a register point. That means spill code can now appear almost anywhere, and it was necessary to fix code that didn't expect that. The difficult places were: - Between a CALL returning a value on the x87 stack and the corresponding FpPOP_RETVAL (was FpGET_ST0). Probably also near x87 inline assembly, but that didn't actually show up in testing. - Between a CALL popping arguments off the stack and the corresponding ADJCALLSTACKUP. Both are fixed now. The only place spill code can't appear is after terminators, see SplitAnalysis::getLastSplitPoint. Original commit message: Rewrite RAGreedy::splitAroundRegion, now with cool ASCII art. This function has to deal with a lot of special cases, and the old version got it wrong sometimes. In particular, it would sometimes leave multiple uses in the stack interval in a single block. That causes bad code with multiple reloads in the same basic block. The new version handles block entry and exit in a single pass. It first eliminates all the easy cases, and then goes on to create a local interval for the blocks with difficult interference. Previously, we would only create the local interval for completely isolated blocks. It can happen that the stack interval becomes completely empty because we could allocate a register in all edge bundles, and the new local intervals deal with the interference. The empty stack interval is harmless, but we need to remove a SplitKit assertion that checks for empty intervals. llvm-svn: 134125
2011-06-30 09:30:39 +08:00
Idx = Idx.getBoundaryIndex();
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Idx);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
Reapply r134047 now that the world is ready for it. This patch will sometimes choose live range split points next to interference instead of always splitting next to a register point. That means spill code can now appear almost anywhere, and it was necessary to fix code that didn't expect that. The difficult places were: - Between a CALL returning a value on the x87 stack and the corresponding FpPOP_RETVAL (was FpGET_ST0). Probably also near x87 inline assembly, but that didn't actually show up in testing. - Between a CALL popping arguments off the stack and the corresponding ADJCALLSTACKUP. Both are fixed now. The only place spill code can't appear is after terminators, see SplitAnalysis::getLastSplitPoint. Original commit message: Rewrite RAGreedy::splitAroundRegion, now with cool ASCII art. This function has to deal with a lot of special cases, and the old version got it wrong sometimes. In particular, it would sometimes leave multiple uses in the stack interval in a single block. That causes bad code with multiple reloads in the same basic block. The new version handles block entry and exit in a single pass. It first eliminates all the easy cases, and then goes on to create a local interval for the blocks with difficult interference. Previously, we would only create the local interval for completely isolated blocks. It can happen that the stack interval becomes completely empty because we could allocate a register in all edge bundles, and the new local intervals deal with the interference. The empty stack interval is harmless, but we need to remove a SplitKit assertion that checks for empty intervals. llvm-svn: 134125
2011-06-30 09:30:39 +08:00
return Idx;
}
LLVM_DEBUG(dbgs() << ": valno " << ParentVNI->id << '\n');
Reapply r134047 now that the world is ready for it. This patch will sometimes choose live range split points next to interference instead of always splitting next to a register point. That means spill code can now appear almost anywhere, and it was necessary to fix code that didn't expect that. The difficult places were: - Between a CALL returning a value on the x87 stack and the corresponding FpPOP_RETVAL (was FpGET_ST0). Probably also near x87 inline assembly, but that didn't actually show up in testing. - Between a CALL popping arguments off the stack and the corresponding ADJCALLSTACKUP. Both are fixed now. The only place spill code can't appear is after terminators, see SplitAnalysis::getLastSplitPoint. Original commit message: Rewrite RAGreedy::splitAroundRegion, now with cool ASCII art. This function has to deal with a lot of special cases, and the old version got it wrong sometimes. In particular, it would sometimes leave multiple uses in the stack interval in a single block. That causes bad code with multiple reloads in the same basic block. The new version handles block entry and exit in a single pass. It first eliminates all the easy cases, and then goes on to create a local interval for the blocks with difficult interference. Previously, we would only create the local interval for completely isolated blocks. It can happen that the stack interval becomes completely empty because we could allocate a register in all edge bundles, and the new local intervals deal with the interference. The empty stack interval is harmless, but we need to remove a SplitKit assertion that checks for empty intervals. llvm-svn: 134125
2011-06-30 09:30:39 +08:00
MachineInstr *MI = LIS.getInstructionFromIndex(Idx);
assert(MI && "enterIntvAfter called with invalid index");
VNInfo *VNI = defFromParent(OpenIdx, ParentVNI, Idx, *MI->getParent(),
std::next(MachineBasicBlock::iterator(MI)));
Reapply r134047 now that the world is ready for it. This patch will sometimes choose live range split points next to interference instead of always splitting next to a register point. That means spill code can now appear almost anywhere, and it was necessary to fix code that didn't expect that. The difficult places were: - Between a CALL returning a value on the x87 stack and the corresponding FpPOP_RETVAL (was FpGET_ST0). Probably also near x87 inline assembly, but that didn't actually show up in testing. - Between a CALL popping arguments off the stack and the corresponding ADJCALLSTACKUP. Both are fixed now. The only place spill code can't appear is after terminators, see SplitAnalysis::getLastSplitPoint. Original commit message: Rewrite RAGreedy::splitAroundRegion, now with cool ASCII art. This function has to deal with a lot of special cases, and the old version got it wrong sometimes. In particular, it would sometimes leave multiple uses in the stack interval in a single block. That causes bad code with multiple reloads in the same basic block. The new version handles block entry and exit in a single pass. It first eliminates all the easy cases, and then goes on to create a local interval for the blocks with difficult interference. Previously, we would only create the local interval for completely isolated blocks. It can happen that the stack interval becomes completely empty because we could allocate a register in all edge bundles, and the new local intervals deal with the interference. The empty stack interval is harmless, but we need to remove a SplitKit assertion that checks for empty intervals. llvm-svn: 134125
2011-06-30 09:30:39 +08:00
return VNI->def;
}
SlotIndex SplitEditor::enterIntvAtEnd(MachineBasicBlock &MBB) {
2011-02-03 14:18:29 +08:00
assert(OpenIdx && "openIntv not called before enterIntvAtEnd");
SlotIndex End = LIS.getMBBEndIdx(&MBB);
SlotIndex Last = End.getPrevSlot();
LLVM_DEBUG(dbgs() << " enterIntvAtEnd " << printMBBReference(MBB) << ", "
<< Last);
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Last);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
return End;
}
LLVM_DEBUG(dbgs() << ": valno " << ParentVNI->id);
VNInfo *VNI = defFromParent(OpenIdx, ParentVNI, Last, MBB,
SA.getLastSplitPointIter(&MBB));
RegAssign.insert(VNI->def, End, OpenIdx);
LLVM_DEBUG(dump());
return VNI->def;
}
/// useIntv - indicate that all instructions in MBB should use OpenLI.
void SplitEditor::useIntv(const MachineBasicBlock &MBB) {
useIntv(LIS.getMBBStartIdx(&MBB), LIS.getMBBEndIdx(&MBB));
}
void SplitEditor::useIntv(SlotIndex Start, SlotIndex End) {
2011-02-03 14:18:29 +08:00
assert(OpenIdx && "openIntv not called before useIntv");
LLVM_DEBUG(dbgs() << " useIntv [" << Start << ';' << End << "):");
2011-02-03 14:18:29 +08:00
RegAssign.insert(Start, End, OpenIdx);
LLVM_DEBUG(dump());
}
SlotIndex SplitEditor::leaveIntvAfter(SlotIndex Idx) {
2011-02-03 14:18:29 +08:00
assert(OpenIdx && "openIntv not called before leaveIntvAfter");
LLVM_DEBUG(dbgs() << " leaveIntvAfter " << Idx);
// The interval must be live beyond the instruction at Idx.
SlotIndex Boundary = Idx.getBoundaryIndex();
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Boundary);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
return Boundary.getNextSlot();
}
LLVM_DEBUG(dbgs() << ": valno " << ParentVNI->id << '\n');
MachineInstr *MI = LIS.getInstructionFromIndex(Boundary);
2011-02-09 02:50:18 +08:00
assert(MI && "No instruction at index");
// In spill mode, make live ranges as short as possible by inserting the copy
// before MI. This is only possible if that instruction doesn't redefine the
// value. The inserted COPY is not a kill, and we don't need to recompute
// the source live range. The spiller also won't try to hoist this copy.
if (SpillMode && !SlotIndex::isSameInstr(ParentVNI->def, Idx) &&
MI->readsVirtualRegister(Edit->getReg())) {
forceRecompute(0, *ParentVNI);
defFromParent(0, ParentVNI, Idx, *MI->getParent(), MI);
return Idx;
}
VNInfo *VNI = defFromParent(0, ParentVNI, Boundary, *MI->getParent(),
std::next(MachineBasicBlock::iterator(MI)));
return VNI->def;
}
SlotIndex SplitEditor::leaveIntvBefore(SlotIndex Idx) {
assert(OpenIdx && "openIntv not called before leaveIntvBefore");
LLVM_DEBUG(dbgs() << " leaveIntvBefore " << Idx);
// The interval must be live into the instruction at Idx.
Idx = Idx.getBaseIndex();
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Idx);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
return Idx.getNextSlot();
}
LLVM_DEBUG(dbgs() << ": valno " << ParentVNI->id << '\n');
MachineInstr *MI = LIS.getInstructionFromIndex(Idx);
assert(MI && "No instruction at index");
VNInfo *VNI = defFromParent(0, ParentVNI, Idx, *MI->getParent(), MI);
return VNI->def;
}
SlotIndex SplitEditor::leaveIntvAtTop(MachineBasicBlock &MBB) {
2011-02-03 14:18:29 +08:00
assert(OpenIdx && "openIntv not called before leaveIntvAtTop");
SlotIndex Start = LIS.getMBBStartIdx(&MBB);
LLVM_DEBUG(dbgs() << " leaveIntvAtTop " << printMBBReference(MBB) << ", "
<< Start);
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Start);
if (!ParentVNI) {
LLVM_DEBUG(dbgs() << ": not live\n");
return Start;
}
2011-02-03 14:18:29 +08:00
VNInfo *VNI = defFromParent(0, ParentVNI, Start, MBB,
MBB.SkipPHIsLabelsAndDebug(MBB.begin()));
2011-02-03 14:18:29 +08:00
RegAssign.insert(Start, VNI->def, OpenIdx);
LLVM_DEBUG(dump());
return VNI->def;
}
void SplitEditor::overlapIntv(SlotIndex Start, SlotIndex End) {
assert(OpenIdx && "openIntv not called before overlapIntv");
const VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(Start);
assert(ParentVNI == Edit->getParent().getVNInfoBefore(End) &&
"Parent changes value in extended range");
assert(LIS.getMBBFromIndex(Start) == LIS.getMBBFromIndex(End) &&
"Range cannot span basic blocks");
// The complement interval will be extended as needed by LRCalc.extend().
if (ParentVNI)
forceRecompute(0, *ParentVNI);
LLVM_DEBUG(dbgs() << " overlapIntv [" << Start << ';' << End << "):");
RegAssign.insert(Start, End, OpenIdx);
LLVM_DEBUG(dump());
}
//===----------------------------------------------------------------------===//
// Spill modes
//===----------------------------------------------------------------------===//
void SplitEditor::removeBackCopies(SmallVectorImpl<VNInfo*> &Copies) {
LiveInterval *LI = &LIS.getInterval(Edit->get(0));
LLVM_DEBUG(dbgs() << "Removing " << Copies.size() << " back-copies.\n");
RegAssignMap::iterator AssignI;
AssignI.setMap(RegAssign);
for (unsigned i = 0, e = Copies.size(); i != e; ++i) {
SlotIndex Def = Copies[i]->def;
MachineInstr *MI = LIS.getInstructionFromIndex(Def);
assert(MI && "No instruction for back-copy");
MachineBasicBlock *MBB = MI->getParent();
MachineBasicBlock::iterator MBBI(MI);
bool AtBegin;
do AtBegin = MBBI == MBB->begin();
while (!AtBegin && (--MBBI)->isDebugInstr());
LLVM_DEBUG(dbgs() << "Removing " << Def << '\t' << *MI);
LIS.removeVRegDefAt(*LI, Def);
LIS.RemoveMachineInstrFromMaps(*MI);
MI->eraseFromParent();
// Adjust RegAssign if a register assignment is killed at Def. We want to
// avoid calculating the live range of the source register if possible.
AssignI.find(Def.getPrevSlot());
if (!AssignI.valid() || AssignI.start() >= Def)
continue;
// If MI doesn't kill the assigned register, just leave it.
if (AssignI.stop() != Def)
continue;
unsigned RegIdx = AssignI.value();
if (AtBegin || !MBBI->readsVirtualRegister(Edit->getReg())) {
LLVM_DEBUG(dbgs() << " cannot find simple kill of RegIdx " << RegIdx
<< '\n');
forceRecompute(RegIdx, *Edit->getParent().getVNInfoAt(Def));
} else {
SlotIndex Kill = LIS.getInstructionIndex(*MBBI).getRegSlot();
LLVM_DEBUG(dbgs() << " move kill to " << Kill << '\t' << *MBBI);
AssignI.setStop(Kill);
}
}
}
MachineBasicBlock*
SplitEditor::findShallowDominator(MachineBasicBlock *MBB,
MachineBasicBlock *DefMBB) {
if (MBB == DefMBB)
return MBB;
assert(MDT.dominates(DefMBB, MBB) && "MBB must be dominated by the def.");
const MachineLoopInfo &Loops = SA.Loops;
const MachineLoop *DefLoop = Loops.getLoopFor(DefMBB);
MachineDomTreeNode *DefDomNode = MDT[DefMBB];
// Best candidate so far.
MachineBasicBlock *BestMBB = MBB;
unsigned BestDepth = std::numeric_limits<unsigned>::max();
while (true) {
const MachineLoop *Loop = Loops.getLoopFor(MBB);
// MBB isn't in a loop, it doesn't get any better. All dominators have a
// higher frequency by definition.
if (!Loop) {
LLVM_DEBUG(dbgs() << "Def in " << printMBBReference(*DefMBB)
<< " dominates " << printMBBReference(*MBB)
<< " at depth 0\n");
return MBB;
}
// We'll never be able to exit the DefLoop.
if (Loop == DefLoop) {
LLVM_DEBUG(dbgs() << "Def in " << printMBBReference(*DefMBB)
<< " dominates " << printMBBReference(*MBB)
<< " in the same loop\n");
return MBB;
}
// Least busy dominator seen so far.
unsigned Depth = Loop->getLoopDepth();
if (Depth < BestDepth) {
BestMBB = MBB;
BestDepth = Depth;
LLVM_DEBUG(dbgs() << "Def in " << printMBBReference(*DefMBB)
<< " dominates " << printMBBReference(*MBB)
<< " at depth " << Depth << '\n');
}
// Leave loop by going to the immediate dominator of the loop header.
// This is a bigger stride than simply walking up the dominator tree.
MachineDomTreeNode *IDom = MDT[Loop->getHeader()]->getIDom();
// Too far up the dominator tree?
if (!IDom || !MDT.dominates(DefDomNode, IDom))
return BestMBB;
MBB = IDom->getBlock();
}
}
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
void SplitEditor::computeRedundantBackCopies(
DenseSet<unsigned> &NotToHoistSet, SmallVectorImpl<VNInfo *> &BackCopies) {
LiveInterval *LI = &LIS.getInterval(Edit->get(0));
LiveInterval *Parent = &Edit->getParent();
SmallVector<SmallPtrSet<VNInfo *, 8>, 8> EqualVNs(Parent->getNumValNums());
SmallPtrSet<VNInfo *, 8> DominatedVNIs;
// Aggregate VNIs having the same value as ParentVNI.
for (VNInfo *VNI : LI->valnos) {
if (VNI->isUnused())
continue;
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(VNI->def);
EqualVNs[ParentVNI->id].insert(VNI);
}
// For VNI aggregation of each ParentVNI, collect dominated, i.e.,
// redundant VNIs to BackCopies.
for (unsigned i = 0, e = Parent->getNumValNums(); i != e; ++i) {
VNInfo *ParentVNI = Parent->getValNumInfo(i);
if (!NotToHoistSet.count(ParentVNI->id))
continue;
SmallPtrSetIterator<VNInfo *> It1 = EqualVNs[ParentVNI->id].begin();
SmallPtrSetIterator<VNInfo *> It2 = It1;
for (; It1 != EqualVNs[ParentVNI->id].end(); ++It1) {
It2 = It1;
for (++It2; It2 != EqualVNs[ParentVNI->id].end(); ++It2) {
if (DominatedVNIs.count(*It1) || DominatedVNIs.count(*It2))
continue;
MachineBasicBlock *MBB1 = LIS.getMBBFromIndex((*It1)->def);
MachineBasicBlock *MBB2 = LIS.getMBBFromIndex((*It2)->def);
if (MBB1 == MBB2) {
DominatedVNIs.insert((*It1)->def < (*It2)->def ? (*It2) : (*It1));
} else if (MDT.dominates(MBB1, MBB2)) {
DominatedVNIs.insert(*It2);
} else if (MDT.dominates(MBB2, MBB1)) {
DominatedVNIs.insert(*It1);
}
}
}
if (!DominatedVNIs.empty()) {
forceRecompute(0, *ParentVNI);
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
for (auto VNI : DominatedVNIs) {
BackCopies.push_back(VNI);
}
DominatedVNIs.clear();
}
}
}
/// For SM_Size mode, find a common dominator for all the back-copies for
/// the same ParentVNI and hoist the backcopies to the dominator BB.
/// For SM_Speed mode, if the common dominator is hot and it is not beneficial
/// to do the hoisting, simply remove the dominated backcopies for the same
/// ParentVNI.
void SplitEditor::hoistCopies() {
// Get the complement interval, always RegIdx 0.
LiveInterval *LI = &LIS.getInterval(Edit->get(0));
LiveInterval *Parent = &Edit->getParent();
// Track the nearest common dominator for all back-copies for each ParentVNI,
// indexed by ParentVNI->id.
using DomPair = std::pair<MachineBasicBlock *, SlotIndex>;
SmallVector<DomPair, 8> NearestDom(Parent->getNumValNums());
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
// The total cost of all the back-copies for each ParentVNI.
SmallVector<BlockFrequency, 8> Costs(Parent->getNumValNums());
// The ParentVNI->id set for which hoisting back-copies are not beneficial
// for Speed.
DenseSet<unsigned> NotToHoistSet;
// Find the nearest common dominator for parent values with multiple
// back-copies. If a single back-copy dominates, put it in DomPair.second.
for (VNInfo *VNI : LI->valnos) {
if (VNI->isUnused())
continue;
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(VNI->def);
assert(ParentVNI && "Parent not live at complement def");
// Don't hoist remats. The complement is probably going to disappear
// completely anyway.
if (Edit->didRematerialize(ParentVNI))
continue;
MachineBasicBlock *ValMBB = LIS.getMBBFromIndex(VNI->def);
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
DomPair &Dom = NearestDom[ParentVNI->id];
// Keep directly defined parent values. This is either a PHI or an
// instruction in the complement range. All other copies of ParentVNI
// should be eliminated.
if (VNI->def == ParentVNI->def) {
LLVM_DEBUG(dbgs() << "Direct complement def at " << VNI->def << '\n');
Dom = DomPair(ValMBB, VNI->def);
continue;
}
// Skip the singly mapped values. There is nothing to gain from hoisting a
// single back-copy.
if (Values.lookup(std::make_pair(0, ParentVNI->id)).getPointer()) {
LLVM_DEBUG(dbgs() << "Single complement def at " << VNI->def << '\n');
continue;
}
if (!Dom.first) {
// First time we see ParentVNI. VNI dominates itself.
Dom = DomPair(ValMBB, VNI->def);
} else if (Dom.first == ValMBB) {
// Two defs in the same block. Pick the earlier def.
if (!Dom.second.isValid() || VNI->def < Dom.second)
Dom.second = VNI->def;
} else {
// Different basic blocks. Check if one dominates.
MachineBasicBlock *Near =
MDT.findNearestCommonDominator(Dom.first, ValMBB);
if (Near == ValMBB)
// Def ValMBB dominates.
Dom = DomPair(ValMBB, VNI->def);
else if (Near != Dom.first)
// None dominate. Hoist to common dominator, need new def.
Dom = DomPair(Near, SlotIndex());
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
Costs[ParentVNI->id] += MBFI.getBlockFreq(ValMBB);
}
LLVM_DEBUG(dbgs() << "Multi-mapped complement " << VNI->id << '@'
<< VNI->def << " for parent " << ParentVNI->id << '@'
<< ParentVNI->def << " hoist to "
<< printMBBReference(*Dom.first) << ' ' << Dom.second
<< '\n');
}
// Insert the hoisted copies.
for (unsigned i = 0, e = Parent->getNumValNums(); i != e; ++i) {
DomPair &Dom = NearestDom[i];
if (!Dom.first || Dom.second.isValid())
continue;
// This value needs a hoisted copy inserted at the end of Dom.first.
VNInfo *ParentVNI = Parent->getValNumInfo(i);
MachineBasicBlock *DefMBB = LIS.getMBBFromIndex(ParentVNI->def);
// Get a less loopy dominator than Dom.first.
Dom.first = findShallowDominator(Dom.first, DefMBB);
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
if (SpillMode == SM_Speed &&
MBFI.getBlockFreq(Dom.first) > Costs[ParentVNI->id]) {
NotToHoistSet.insert(ParentVNI->id);
continue;
}
SlotIndex Last = LIS.getMBBEndIdx(Dom.first).getPrevSlot();
Dom.second =
defFromParent(0, ParentVNI, Last, *Dom.first,
SA.getLastSplitPointIter(Dom.first))->def;
}
// Remove redundant back-copies that are now known to be dominated by another
// def with the same value.
SmallVector<VNInfo*, 8> BackCopies;
for (VNInfo *VNI : LI->valnos) {
if (VNI->isUnused())
continue;
VNInfo *ParentVNI = Edit->getParent().getVNInfoAt(VNI->def);
const DomPair &Dom = NearestDom[ParentVNI->id];
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
if (!Dom.first || Dom.second == VNI->def ||
NotToHoistSet.count(ParentVNI->id))
continue;
BackCopies.push_back(VNI);
forceRecompute(0, *ParentVNI);
}
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
// If it is not beneficial to hoist all the BackCopies, simply remove
// redundant BackCopies in speed mode.
if (SpillMode == SM_Speed && !NotToHoistSet.empty())
computeRedundantBackCopies(NotToHoistSet, BackCopies);
removeBackCopies(BackCopies);
}
/// transferValues - Transfer all possible values to the new live ranges.
/// Values that were rematerialized are left alone, they need LRCalc.extend().
bool SplitEditor::transferValues() {
bool Skipped = false;
RegAssignMap::const_iterator AssignI = RegAssign.begin();
for (const LiveRange::Segment &S : Edit->getParent()) {
LLVM_DEBUG(dbgs() << " blit " << S << ':');
VNInfo *ParentVNI = S.valno;
// RegAssign has holes where RegIdx 0 should be used.
SlotIndex Start = S.start;
AssignI.advanceTo(Start);
do {
unsigned RegIdx;
SlotIndex End = S.end;
if (!AssignI.valid()) {
RegIdx = 0;
} else if (AssignI.start() <= Start) {
RegIdx = AssignI.value();
if (AssignI.stop() < End) {
End = AssignI.stop();
++AssignI;
}
} else {
RegIdx = 0;
End = std::min(End, AssignI.start());
}
// The interval [Start;End) is continuously mapped to RegIdx, ParentVNI.
LLVM_DEBUG(dbgs() << " [" << Start << ';' << End << ")=" << RegIdx << '('
<< printReg(Edit->get(RegIdx)) << ')');
LiveInterval &LI = LIS.getInterval(Edit->get(RegIdx));
// Check for a simply defined value that can be blitted directly.
ValueForcePair VFP = Values.lookup(std::make_pair(RegIdx, ParentVNI->id));
if (VNInfo *VNI = VFP.getPointer()) {
LLVM_DEBUG(dbgs() << ':' << VNI->id);
LI.addSegment(LiveInterval::Segment(Start, End, VNI));
Start = End;
continue;
}
// Skip values with forced recomputation.
if (VFP.getInt()) {
LLVM_DEBUG(dbgs() << "(recalc)");
Skipped = true;
Start = End;
continue;
}
LiveRangeCalc &LRC = getLRCalc(RegIdx);
// This value has multiple defs in RegIdx, but it wasn't rematerialized,
// so the live range is accurate. Add live-in blocks in [Start;End) to the
// LiveInBlocks.
MachineFunction::iterator MBB = LIS.getMBBFromIndex(Start)->getIterator();
SlotIndex BlockStart, BlockEnd;
std::tie(BlockStart, BlockEnd) = LIS.getSlotIndexes()->getMBBRange(&*MBB);
// The first block may be live-in, or it may have its own def.
if (Start != BlockStart) {
VNInfo *VNI = LI.extendInBlock(BlockStart, std::min(BlockEnd, End));
assert(VNI && "Missing def for complex mapped value");
LLVM_DEBUG(dbgs() << ':' << VNI->id << "*" << printMBBReference(*MBB));
// MBB has its own def. Is it also live-out?
if (BlockEnd <= End)
LRC.setLiveOutValue(&*MBB, VNI);
// Skip to the next block for live-in.
++MBB;
BlockStart = BlockEnd;
}
// Handle the live-in blocks covered by [Start;End).
assert(Start <= BlockStart && "Expected live-in block");
while (BlockStart < End) {
LLVM_DEBUG(dbgs() << ">" << printMBBReference(*MBB));
BlockEnd = LIS.getMBBEndIdx(&*MBB);
if (BlockStart == ParentVNI->def) {
// This block has the def of a parent PHI, so it isn't live-in.
assert(ParentVNI->isPHIDef() && "Non-phi defined at block start?");
VNInfo *VNI = LI.extendInBlock(BlockStart, std::min(BlockEnd, End));
assert(VNI && "Missing def for complex mapped parent PHI");
if (End >= BlockEnd)
LRC.setLiveOutValue(&*MBB, VNI); // Live-out as well.
} else {
// This block needs a live-in value. The last block covered may not
// be live-out.
if (End < BlockEnd)
LRC.addLiveInBlock(LI, MDT[&*MBB], End);
else {
// Live-through, and we don't know the value.
LRC.addLiveInBlock(LI, MDT[&*MBB]);
LRC.setLiveOutValue(&*MBB, nullptr);
}
}
BlockStart = BlockEnd;
++MBB;
}
Start = End;
} while (Start != S.end);
LLVM_DEBUG(dbgs() << '\n');
}
LRCalc[0].calculateValues();
if (SpillMode)
LRCalc[1].calculateValues();
return Skipped;
}
static bool removeDeadSegment(SlotIndex Def, LiveRange &LR) {
const LiveRange::Segment *Seg = LR.getSegmentContaining(Def);
if (Seg == nullptr)
return true;
if (Seg->end != Def.getDeadSlot())
return false;
// This is a dead PHI. Remove it.
LR.removeSegment(*Seg, true);
return true;
}
void SplitEditor::extendPHIRange(MachineBasicBlock &B, LiveRangeCalc &LRC,
LiveRange &LR, LaneBitmask LM,
ArrayRef<SlotIndex> Undefs) {
for (MachineBasicBlock *P : B.predecessors()) {
SlotIndex End = LIS.getMBBEndIdx(P);
SlotIndex LastUse = End.getPrevSlot();
// The predecessor may not have a live-out value. That is OK, like an
// undef PHI operand.
LiveInterval &PLI = Edit->getParent();
// Need the cast because the inputs to ?: would otherwise be deemed
// "incompatible": SubRange vs LiveInterval.
LiveRange &PSR = !LM.all() ? getSubRangeForMask(LM, PLI)
: static_cast<LiveRange&>(PLI);
if (PSR.liveAt(LastUse))
LRC.extend(LR, End, /*PhysReg=*/0, Undefs);
}
}
void SplitEditor::extendPHIKillRanges() {
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
// Extend live ranges to be live-out for successor PHI values.
// Visit each PHI def slot in the parent live interval. If the def is dead,
// remove it. Otherwise, extend the live interval to reach the end indexes
// of all predecessor blocks.
LiveInterval &ParentLI = Edit->getParent();
for (const VNInfo *V : ParentLI.valnos) {
if (V->isUnused() || !V->isPHIDef())
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
continue;
unsigned RegIdx = RegAssign.lookup(V->def);
LiveInterval &LI = LIS.getInterval(Edit->get(RegIdx));
LiveRangeCalc &LRC = getLRCalc(RegIdx);
MachineBasicBlock &B = *LIS.getMBBFromIndex(V->def);
if (!removeDeadSegment(V->def, LI))
extendPHIRange(B, LRC, LI, LaneBitmask::getAll(), /*Undefs=*/{});
}
SmallVector<SlotIndex, 4> Undefs;
LiveRangeCalc SubLRC;
for (LiveInterval::SubRange &PS : ParentLI.subranges()) {
for (const VNInfo *V : PS.valnos) {
if (V->isUnused() || !V->isPHIDef())
continue;
unsigned RegIdx = RegAssign.lookup(V->def);
LiveInterval &LI = LIS.getInterval(Edit->get(RegIdx));
LiveInterval::SubRange &S = getSubRangeForMask(PS.LaneMask, LI);
if (removeDeadSegment(V->def, S))
continue;
MachineBasicBlock &B = *LIS.getMBBFromIndex(V->def);
SubLRC.reset(&VRM.getMachineFunction(), LIS.getSlotIndexes(), &MDT,
&LIS.getVNInfoAllocator());
Undefs.clear();
LI.computeSubRangeUndefs(Undefs, PS.LaneMask, MRI, *LIS.getSlotIndexes());
extendPHIRange(B, SubLRC, S, PS.LaneMask, Undefs);
}
}
}
/// rewriteAssigned - Rewrite all uses of Edit->getReg().
void SplitEditor::rewriteAssigned(bool ExtendRanges) {
struct ExtPoint {
ExtPoint(const MachineOperand &O, unsigned R, SlotIndex N)
: MO(O), RegIdx(R), Next(N) {}
MachineOperand MO;
unsigned RegIdx;
SlotIndex Next;
};
SmallVector<ExtPoint,4> ExtPoints;
for (MachineRegisterInfo::reg_iterator RI = MRI.reg_begin(Edit->getReg()),
RE = MRI.reg_end(); RI != RE;) {
MachineOperand &MO = *RI;
MachineInstr *MI = MO.getParent();
++RI;
2011-02-03 14:18:29 +08:00
// LiveDebugVariables should have handled all DBG_VALUE instructions.
if (MI->isDebugValue()) {
LLVM_DEBUG(dbgs() << "Zapping " << *MI);
MO.setReg(0);
continue;
}
// <undef> operands don't really read the register, so it doesn't matter
// which register we choose. When the use operand is tied to a def, we must
// use the same register as the def, so just do that always.
SlotIndex Idx = LIS.getInstructionIndex(*MI);
if (MO.isDef() || MO.isUndef())
Idx = Idx.getRegSlot(MO.isEarlyClobber());
2011-02-03 14:18:29 +08:00
// Rewrite to the mapped register at Idx.
unsigned RegIdx = RegAssign.lookup(Idx);
LiveInterval &LI = LIS.getInterval(Edit->get(RegIdx));
MO.setReg(LI.reg);
LLVM_DEBUG(dbgs() << " rewr " << printMBBReference(*MI->getParent())
<< '\t' << Idx << ':' << RegIdx << '\t' << *MI);
2011-02-03 14:18:29 +08:00
// Extend liveness to Idx if the instruction reads reg.
if (!ExtendRanges || MO.isUndef())
continue;
// Skip instructions that don't read Reg.
if (MO.isDef()) {
if (!MO.getSubReg() && !MO.isEarlyClobber())
continue;
// We may want to extend a live range for a partial redef, or for a use
// tied to an early clobber.
Idx = Idx.getPrevSlot();
if (!Edit->getParent().liveAt(Idx))
continue;
} else
Idx = Idx.getRegSlot(true);
SlotIndex Next = Idx.getNextSlot();
if (LI.hasSubRanges()) {
// We have to delay extending subranges until we have seen all operands
// defining the register. This is because a <def,read-undef> operand
// will create an "undef" point, and we cannot extend any subranges
// until all of them have been accounted for.
if (MO.isUse())
ExtPoints.push_back(ExtPoint(MO, RegIdx, Next));
} else {
LiveRangeCalc &LRC = getLRCalc(RegIdx);
LRC.extend(LI, Next, 0, ArrayRef<SlotIndex>());
}
}
for (ExtPoint &EP : ExtPoints) {
LiveInterval &LI = LIS.getInterval(Edit->get(EP.RegIdx));
assert(LI.hasSubRanges());
LiveRangeCalc SubLRC;
Apply llvm-prefer-register-over-unsigned from clang-tidy to LLVM Summary: This clang-tidy check is looking for unsigned integer variables whose initializer starts with an implicit cast from llvm::Register and changes the type of the variable to llvm::Register (dropping the llvm:: where possible). Partial reverts in: X86FrameLowering.cpp - Some functions return unsigned and arguably should be MCRegister X86FixupLEAs.cpp - Some functions return unsigned and arguably should be MCRegister X86FrameLowering.cpp - Some functions return unsigned and arguably should be MCRegister HexagonBitSimplify.cpp - Function takes BitTracker::RegisterRef which appears to be unsigned& MachineVerifier.cpp - Ambiguous operator==() given MCRegister and const Register PPCFastISel.cpp - No Register::operator-=() PeepholeOptimizer.cpp - TargetInstrInfo::optimizeLoadInstr() takes an unsigned& MachineTraceMetrics.cpp - MachineTraceMetrics lacks a suitable constructor Manual fixups in: ARMFastISel.cpp - ARMEmitLoad() now takes a Register& instead of unsigned& HexagonSplitDouble.cpp - Ternary operator was ambiguous between unsigned/Register HexagonConstExtenders.cpp - Has a local class named Register, used llvm::Register instead of Register. PPCFastISel.cpp - PPCEmitLoad() now takes a Register& instead of unsigned& Depends on D65919 Reviewers: arsenm, bogner, craig.topper, RKSimon Reviewed By: arsenm Subscribers: RKSimon, craig.topper, lenary, aemerson, wuzish, jholewinski, MatzeB, qcolombet, dschuff, jyknight, dylanmckay, sdardis, nemanjai, jvesely, wdng, nhaehnle, sbc100, jgravelle-google, kristof.beyls, hiraditya, aheejin, kbarton, fedor.sergeev, javed.absar, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, tpr, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, Jim, s.egerton, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65962 llvm-svn: 369041
2019-08-16 03:22:08 +08:00
Register Reg = EP.MO.getReg(), Sub = EP.MO.getSubReg();
LaneBitmask LM = Sub != 0 ? TRI.getSubRegIndexLaneMask(Sub)
: MRI.getMaxLaneMaskForVReg(Reg);
for (LiveInterval::SubRange &S : LI.subranges()) {
if ((S.LaneMask & LM).none())
continue;
// The problem here can be that the new register may have been created
// for a partially defined original register. For example:
// %0:subreg_hireg<def,read-undef> = ...
// ...
[CodeGen] Use MachineOperand::print in the MIRPrinter for MO_Register. Work towards the unification of MIR and debug output by refactoring the interfaces. For MachineOperand::print, keep a simple version that can be easily called from `dump()`, and a more complex one which will be called from both the MIRPrinter and MachineInstr::print. Add extra checks inside MachineOperand for detached operands (operands with getParent() == nullptr). https://reviews.llvm.org/D40836 * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/kill: ([^ ]+) ([^ ]+)<def> ([^ ]+)/kill: \1 def \2 \3/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/kill: ([^ ]+) ([^ ]+) ([^ ]+)<def>/kill: \1 \2 def \3/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/kill: def ([^ ]+) ([^ ]+) ([^ ]+)<def>/kill: def \1 \2 def \3/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/<def>//g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<kill>/killed \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<imp-use,kill>/implicit killed \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<dead>/dead \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<def[ ]*,[ ]*dead>/dead \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<imp-def[ ]*,[ ]*dead>/implicit-def dead \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<imp-def>/implicit-def \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<imp-use>/implicit \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<internal>/internal \1/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" -o -name "*.s" \) -type f -print0 | xargs -0 sed -i '' -E 's/([^ ]+)<undef>/undef \1/g' llvm-svn: 320022
2017-12-07 18:40:31 +08:00
// %1 = COPY %0
if (S.empty())
continue;
SubLRC.reset(&VRM.getMachineFunction(), LIS.getSlotIndexes(), &MDT,
&LIS.getVNInfoAllocator());
SmallVector<SlotIndex, 4> Undefs;
LI.computeSubRangeUndefs(Undefs, S.LaneMask, MRI, *LIS.getSlotIndexes());
SubLRC.extend(S, EP.Next, 0, Undefs);
}
}
for (unsigned R : *Edit) {
LiveInterval &LI = LIS.getInterval(R);
if (!LI.hasSubRanges())
continue;
LI.clear();
LI.removeEmptySubRanges();
LIS.constructMainRangeFromSubranges(LI);
}
}
void SplitEditor::deleteRematVictims() {
SmallVector<MachineInstr*, 8> Dead;
for (LiveRangeEdit::iterator I = Edit->begin(), E = Edit->end(); I != E; ++I){
LiveInterval *LI = &LIS.getInterval(*I);
for (const LiveRange::Segment &S : LI->segments) {
// Dead defs end at the dead slot.
if (S.end != S.valno->def.getDeadSlot())
continue;
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
if (S.valno->isPHIDef())
continue;
MachineInstr *MI = LIS.getInstructionFromIndex(S.valno->def);
assert(MI && "Missing instruction for dead def");
MI->addRegisterDead(LI->reg, &TRI);
if (!MI->allDefsAreDead())
continue;
LLVM_DEBUG(dbgs() << "All defs dead: " << *MI);
Dead.push_back(MI);
}
}
if (Dead.empty())
return;
Edit->eliminateDeadDefs(Dead, None, &AA);
}
void SplitEditor::forceRecomputeVNI(const VNInfo &ParentVNI) {
// Fast-path for common case.
if (!ParentVNI.isPHIDef()) {
for (unsigned I = 0, E = Edit->size(); I != E; ++I)
forceRecompute(I, ParentVNI);
return;
}
// Trace value through phis.
SmallPtrSet<const VNInfo *, 8> Visited; ///< whether VNI was/is in worklist.
SmallVector<const VNInfo *, 4> WorkList;
Visited.insert(&ParentVNI);
WorkList.push_back(&ParentVNI);
const LiveInterval &ParentLI = Edit->getParent();
const SlotIndexes &Indexes = *LIS.getSlotIndexes();
do {
const VNInfo &VNI = *WorkList.back();
WorkList.pop_back();
for (unsigned I = 0, E = Edit->size(); I != E; ++I)
forceRecompute(I, VNI);
if (!VNI.isPHIDef())
continue;
MachineBasicBlock &MBB = *Indexes.getMBBFromIndex(VNI.def);
for (const MachineBasicBlock *Pred : MBB.predecessors()) {
SlotIndex PredEnd = Indexes.getMBBEndIdx(Pred);
VNInfo *PredVNI = ParentLI.getVNInfoBefore(PredEnd);
assert(PredVNI && "Value available in PhiVNI predecessor");
if (Visited.insert(PredVNI).second)
WorkList.push_back(PredVNI);
}
} while(!WorkList.empty());
}
void SplitEditor::finish(SmallVectorImpl<unsigned> *LRMap) {
++NumFinished;
2011-02-03 14:18:29 +08:00
// At this point, the live intervals in Edit contain VNInfos corresponding to
// the inserted copies.
// Add the original defs from the parent interval.
for (const VNInfo *ParentVNI : Edit->getParent().valnos) {
2011-02-04 08:59:23 +08:00
if (ParentVNI->isUnused())
continue;
unsigned RegIdx = RegAssign.lookup(ParentVNI->def);
defValue(RegIdx, ParentVNI, ParentVNI->def, true);
// Force rematted values to be recomputed everywhere.
// The new live ranges may be truncated.
if (Edit->didRematerialize(ParentVNI))
forceRecomputeVNI(*ParentVNI);
}
// Hoist back-copies to the complement interval when in spill mode.
switch (SpillMode) {
case SM_Partition:
// Leave all back-copies as is.
break;
case SM_Size:
case SM_Speed:
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
// hoistCopies will behave differently between size and speed.
hoistCopies();
}
// Transfer the simply mapped values, check if any are skipped.
bool Skipped = transferValues();
Recommit r265547, and r265610,r265639,r265657 on top of it, plus two fixes with one about error verify-regalloc reported, and another about live range update of phi after rematerialization. r265547: Replace analyzeSiblingValues with new algorithm to fix its compile time issue. The patch is to solve PR17409 and its duplicates. analyzeSiblingValues is a N x N complexity algorithm where N is the number of siblings generated by reg splitting. Although it causes siginificant compile time issue when N is large, it is also important for performance since it removes redundent spills and enables rematerialization. To solve the compile time issue, the patch removes analyzeSiblingValues and replaces it with lower cost alternatives containing two parts. The first part creates a new spill hoisting method in postOptimization of register allocation. It does spill hoisting at once after all the spills are generated instead of inside every instance of selectOrSplit. The second part queries the define expr of the original register for rematerializaiton and keep it always available during register allocation even if it is already dead. It deletes those dead instructions only in postOptimization. With the two parts in the patch, it can remove analyzeSiblingValues without sacrificing performance. Patches on top of r265547: r265610 "Fix the compare-clang diff error introduced by r265547." r265639 "Fix the sanitizer bootstrap error in r265547." r265657 "InlineSpiller.cpp: Escap \@ in r265547. [-Wdocumentation]" Differential Revision: http://reviews.llvm.org/D15302 Differential Revision: http://reviews.llvm.org/D18934 Differential Revision: http://reviews.llvm.org/D18935 Differential Revision: http://reviews.llvm.org/D18936 llvm-svn: 266162
2016-04-13 11:08:27 +08:00
// Rewrite virtual registers, possibly extending ranges.
rewriteAssigned(Skipped);
if (Skipped)
extendPHIKillRanges();
else
++NumSimple;
// Delete defs that were rematted everywhere.
if (Skipped)
deleteRematVictims();
// Get rid of unused values and set phi-kill flags.
for (unsigned Reg : *Edit) {
LiveInterval &LI = LIS.getInterval(Reg);
LI.removeEmptySubRanges();
LI.RenumberValues();
}
// Provide a reverse mapping from original indices to Edit ranges.
if (LRMap) {
LRMap->clear();
for (unsigned i = 0, e = Edit->size(); i != e; ++i)
LRMap->push_back(i);
}
// Now check if any registers were separated into multiple components.
ConnectedVNInfoEqClasses ConEQ(LIS);
for (unsigned i = 0, e = Edit->size(); i != e; ++i) {
// Don't use iterators, they are invalidated by create() below.
unsigned VReg = Edit->get(i);
LiveInterval &LI = LIS.getInterval(VReg);
SmallVector<LiveInterval*, 8> SplitLIs;
LIS.splitSeparateComponents(LI, SplitLIs);
unsigned Original = VRM.getOriginal(VReg);
for (LiveInterval *SplitLI : SplitLIs)
VRM.setIsSplitFromReg(SplitLI->reg, Original);
// The new intervals all map back to i.
if (LRMap)
LRMap->resize(Edit->size(), i);
}
// Calculate spill weight and allocation hints for new intervals.
Edit->calculateRegClassAndHint(VRM.getMachineFunction(), SA.Loops, MBFI);
assert(!LRMap || LRMap->size() == Edit->size());
}
//===----------------------------------------------------------------------===//
// Single Block Splitting
//===----------------------------------------------------------------------===//
bool SplitAnalysis::shouldSplitSingleBlock(const BlockInfo &BI,
bool SingleInstrs) const {
// Always split for multiple instructions.
if (!BI.isOneInstr())
return true;
// Don't split for single instructions unless explicitly requested.
if (!SingleInstrs)
return false;
// Splitting a live-through range always makes progress.
if (BI.LiveIn && BI.LiveOut)
return true;
// No point in isolating a copy. It has no register class constraints.
if (LIS.getInstructionFromIndex(BI.FirstInstr)->isCopyLike())
return false;
// Finally, don't isolate an end point that was created by earlier splits.
return isOriginalEndpoint(BI.FirstInstr);
}
void SplitEditor::splitSingleBlock(const SplitAnalysis::BlockInfo &BI) {
openIntv();
SlotIndex LastSplitPoint = SA.getLastSplitPoint(BI.MBB->getNumber());
SlotIndex SegStart = enterIntvBefore(std::min(BI.FirstInstr,
LastSplitPoint));
if (!BI.LiveOut || BI.LastInstr < LastSplitPoint) {
useIntv(SegStart, leaveIntvAfter(BI.LastInstr));
} else {
// The last use is after the last valid split point.
SlotIndex SegStop = leaveIntvBefore(LastSplitPoint);
useIntv(SegStart, SegStop);
overlapIntv(SegStop, BI.LastInstr);
}
}
//===----------------------------------------------------------------------===//
// Global Live Range Splitting Support
//===----------------------------------------------------------------------===//
// These methods support a method of global live range splitting that uses a
// global algorithm to decide intervals for CFG edges. They will insert split
// points and color intervals in basic blocks while avoiding interference.
//
// Note that splitSingleBlock is also useful for blocks where both CFG edges
// are on the stack.
void SplitEditor::splitLiveThroughBlock(unsigned MBBNum,
unsigned IntvIn, SlotIndex LeaveBefore,
unsigned IntvOut, SlotIndex EnterAfter){
SlotIndex Start, Stop;
std::tie(Start, Stop) = LIS.getSlotIndexes()->getMBBRange(MBBNum);
LLVM_DEBUG(dbgs() << "%bb." << MBBNum << " [" << Start << ';' << Stop
<< ") intf " << LeaveBefore << '-' << EnterAfter
<< ", live-through " << IntvIn << " -> " << IntvOut);
assert((IntvIn || IntvOut) && "Use splitSingleBlock for isolated blocks");
assert((!LeaveBefore || LeaveBefore < Stop) && "Interference after block");
assert((!IntvIn || !LeaveBefore || LeaveBefore > Start) && "Impossible intf");
assert((!EnterAfter || EnterAfter >= Start) && "Interference before block");
MachineBasicBlock *MBB = VRM.getMachineFunction().getBlockNumbered(MBBNum);
if (!IntvOut) {
LLVM_DEBUG(dbgs() << ", spill on entry.\n");
//
// <<<<<<<<< Possible LeaveBefore interference.
// |-----------| Live through.
// -____________ Spill on entry.
//
selectIntv(IntvIn);
SlotIndex Idx = leaveIntvAtTop(*MBB);
assert((!LeaveBefore || Idx <= LeaveBefore) && "Interference");
(void)Idx;
return;
}
if (!IntvIn) {
LLVM_DEBUG(dbgs() << ", reload on exit.\n");
//
// >>>>>>> Possible EnterAfter interference.
// |-----------| Live through.
// ___________-- Reload on exit.
//
selectIntv(IntvOut);
SlotIndex Idx = enterIntvAtEnd(*MBB);
assert((!EnterAfter || Idx >= EnterAfter) && "Interference");
(void)Idx;
return;
}
if (IntvIn == IntvOut && !LeaveBefore && !EnterAfter) {
LLVM_DEBUG(dbgs() << ", straight through.\n");
//
// |-----------| Live through.
// ------------- Straight through, same intv, no interference.
//
selectIntv(IntvOut);
useIntv(Start, Stop);
return;
}
// We cannot legally insert splits after LSP.
SlotIndex LSP = SA.getLastSplitPoint(MBBNum);
assert((!IntvOut || !EnterAfter || EnterAfter < LSP) && "Impossible intf");
if (IntvIn != IntvOut && (!LeaveBefore || !EnterAfter ||
LeaveBefore.getBaseIndex() > EnterAfter.getBoundaryIndex())) {
LLVM_DEBUG(dbgs() << ", switch avoiding interference.\n");
//
// >>>> <<<< Non-overlapping EnterAfter/LeaveBefore interference.
// |-----------| Live through.
// ------======= Switch intervals between interference.
//
selectIntv(IntvOut);
SlotIndex Idx;
if (LeaveBefore && LeaveBefore < LSP) {
Idx = enterIntvBefore(LeaveBefore);
useIntv(Idx, Stop);
} else {
Idx = enterIntvAtEnd(*MBB);
}
selectIntv(IntvIn);
useIntv(Start, Idx);
assert((!LeaveBefore || Idx <= LeaveBefore) && "Interference");
assert((!EnterAfter || Idx >= EnterAfter) && "Interference");
return;
}
LLVM_DEBUG(dbgs() << ", create local intv for interference.\n");
//
// >>><><><><<<< Overlapping EnterAfter/LeaveBefore interference.
// |-----------| Live through.
// ==---------== Switch intervals before/after interference.
//
assert(LeaveBefore <= EnterAfter && "Missed case");
selectIntv(IntvOut);
SlotIndex Idx = enterIntvAfter(EnterAfter);
useIntv(Idx, Stop);
assert((!EnterAfter || Idx >= EnterAfter) && "Interference");
selectIntv(IntvIn);
Idx = leaveIntvBefore(LeaveBefore);
useIntv(Start, Idx);
assert((!LeaveBefore || Idx <= LeaveBefore) && "Interference");
}
void SplitEditor::splitRegInBlock(const SplitAnalysis::BlockInfo &BI,
unsigned IntvIn, SlotIndex LeaveBefore) {
SlotIndex Start, Stop;
std::tie(Start, Stop) = LIS.getSlotIndexes()->getMBBRange(BI.MBB);
LLVM_DEBUG(dbgs() << printMBBReference(*BI.MBB) << " [" << Start << ';'
<< Stop << "), uses " << BI.FirstInstr << '-'
<< BI.LastInstr << ", reg-in " << IntvIn
<< ", leave before " << LeaveBefore
<< (BI.LiveOut ? ", stack-out" : ", killed in block"));
assert(IntvIn && "Must have register in");
assert(BI.LiveIn && "Must be live-in");
assert((!LeaveBefore || LeaveBefore > Start) && "Bad interference");
if (!BI.LiveOut && (!LeaveBefore || LeaveBefore >= BI.LastInstr)) {
LLVM_DEBUG(dbgs() << " before interference.\n");
//
// <<< Interference after kill.
// |---o---x | Killed in block.
// ========= Use IntvIn everywhere.
//
selectIntv(IntvIn);
useIntv(Start, BI.LastInstr);
return;
}
SlotIndex LSP = SA.getLastSplitPoint(BI.MBB->getNumber());
if (!LeaveBefore || LeaveBefore > BI.LastInstr.getBoundaryIndex()) {
//
// <<< Possible interference after last use.
// |---o---o---| Live-out on stack.
// =========____ Leave IntvIn after last use.
//
// < Interference after last use.
// |---o---o--o| Live-out on stack, late last use.
// ============ Copy to stack after LSP, overlap IntvIn.
// \_____ Stack interval is live-out.
//
if (BI.LastInstr < LSP) {
LLVM_DEBUG(dbgs() << ", spill after last use before interference.\n");
selectIntv(IntvIn);
SlotIndex Idx = leaveIntvAfter(BI.LastInstr);
useIntv(Start, Idx);
assert((!LeaveBefore || Idx <= LeaveBefore) && "Interference");
} else {
LLVM_DEBUG(dbgs() << ", spill before last split point.\n");
selectIntv(IntvIn);
SlotIndex Idx = leaveIntvBefore(LSP);
overlapIntv(Idx, BI.LastInstr);
useIntv(Start, Idx);
assert((!LeaveBefore || Idx <= LeaveBefore) && "Interference");
}
return;
}
// The interference is overlapping somewhere we wanted to use IntvIn. That
// means we need to create a local interval that can be allocated a
// different register.
unsigned LocalIntv = openIntv();
(void)LocalIntv;
LLVM_DEBUG(dbgs() << ", creating local interval " << LocalIntv << ".\n");
if (!BI.LiveOut || BI.LastInstr < LSP) {
//
// <<<<<<< Interference overlapping uses.
// |---o---o---| Live-out on stack.
// =====----____ Leave IntvIn before interference, then spill.
//
SlotIndex To = leaveIntvAfter(BI.LastInstr);
SlotIndex From = enterIntvBefore(LeaveBefore);
useIntv(From, To);
selectIntv(IntvIn);
useIntv(Start, From);
assert((!LeaveBefore || From <= LeaveBefore) && "Interference");
return;
}
// <<<<<<< Interference overlapping uses.
// |---o---o--o| Live-out on stack, late last use.
// =====------- Copy to stack before LSP, overlap LocalIntv.
// \_____ Stack interval is live-out.
//
SlotIndex To = leaveIntvBefore(LSP);
overlapIntv(To, BI.LastInstr);
SlotIndex From = enterIntvBefore(std::min(To, LeaveBefore));
useIntv(From, To);
selectIntv(IntvIn);
useIntv(Start, From);
assert((!LeaveBefore || From <= LeaveBefore) && "Interference");
}
void SplitEditor::splitRegOutBlock(const SplitAnalysis::BlockInfo &BI,
unsigned IntvOut, SlotIndex EnterAfter) {
SlotIndex Start, Stop;
std::tie(Start, Stop) = LIS.getSlotIndexes()->getMBBRange(BI.MBB);
LLVM_DEBUG(dbgs() << printMBBReference(*BI.MBB) << " [" << Start << ';'
<< Stop << "), uses " << BI.FirstInstr << '-'
<< BI.LastInstr << ", reg-out " << IntvOut
<< ", enter after " << EnterAfter
<< (BI.LiveIn ? ", stack-in" : ", defined in block"));
SlotIndex LSP = SA.getLastSplitPoint(BI.MBB->getNumber());
assert(IntvOut && "Must have register out");
assert(BI.LiveOut && "Must be live-out");
assert((!EnterAfter || EnterAfter < LSP) && "Bad interference");
if (!BI.LiveIn && (!EnterAfter || EnterAfter <= BI.FirstInstr)) {
LLVM_DEBUG(dbgs() << " after interference.\n");
//
// >>>> Interference before def.
// | o---o---| Defined in block.
// ========= Use IntvOut everywhere.
//
selectIntv(IntvOut);
useIntv(BI.FirstInstr, Stop);
return;
}
if (!EnterAfter || EnterAfter < BI.FirstInstr.getBaseIndex()) {
LLVM_DEBUG(dbgs() << ", reload after interference.\n");
//
// >>>> Interference before def.
// |---o---o---| Live-through, stack-in.
// ____========= Enter IntvOut before first use.
//
selectIntv(IntvOut);
SlotIndex Idx = enterIntvBefore(std::min(LSP, BI.FirstInstr));
useIntv(Idx, Stop);
assert((!EnterAfter || Idx >= EnterAfter) && "Interference");
return;
}
// The interference is overlapping somewhere we wanted to use IntvOut. That
// means we need to create a local interval that can be allocated a
// different register.
LLVM_DEBUG(dbgs() << ", interference overlaps uses.\n");
//
// >>>>>>> Interference overlapping uses.
// |---o---o---| Live-through, stack-in.
// ____---====== Create local interval for interference range.
//
selectIntv(IntvOut);
SlotIndex Idx = enterIntvAfter(EnterAfter);
useIntv(Idx, Stop);
assert((!EnterAfter || Idx >= EnterAfter) && "Interference");
openIntv();
SlotIndex From = enterIntvBefore(std::min(Idx, BI.FirstInstr));
useIntv(From, Idx);
}