1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318
|
//=- llvm/CodeGen/DFAPacketizer.cpp - DFA Packetizer for VLIW -*- C++ -*-=====//
//
// 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 class implements a deterministic finite automaton (DFA) based
// packetizing mechanism for VLIW architectures. It provides APIs to
// determine whether there exists a legal mapping of instructions to
// functional unit assignments in a packet. The DFA is auto-generated from
// the target's Schedule.td file.
//
// A DFA consists of 3 major elements: states, inputs, and transitions. For
// the packetizing mechanism, the input is the set of instruction classes for
// a target. The state models all possible combinations of functional unit
// consumption for a given set of instructions in a packet. A transition
// models the addition of an instruction to a packet. In the DFA constructed
// by this class, if an instruction can be added to a packet, then a valid
// transition exists from the corresponding state. Invalid transitions
// indicate that the instruction cannot be added to the current packet.
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/DFAPacketizer.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBundle.h"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/ScheduleDAGInstrs.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/MC/MCInstrItineraries.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <iterator>
#include <memory>
#include <vector>
using namespace llvm;
#define DEBUG_TYPE "packets"
static cl::opt<unsigned> InstrLimit("dfa-instr-limit", cl::Hidden,
cl::init(0), cl::desc("If present, stops packetizing after N instructions"));
static unsigned InstrCount = 0;
// Check if the resources occupied by a MCInstrDesc are available in the
// current state.
bool DFAPacketizer::canReserveResources(const MCInstrDesc *MID) {
unsigned Action = ItinActions[MID->getSchedClass()];
if (MID->getSchedClass() == 0 || Action == 0)
return false;
return A.canAdd(Action);
}
// Reserve the resources occupied by a MCInstrDesc and change the current
// state to reflect that change.
void DFAPacketizer::reserveResources(const MCInstrDesc *MID) {
unsigned Action = ItinActions[MID->getSchedClass()];
if (MID->getSchedClass() == 0 || Action == 0)
return;
A.add(Action);
}
// Check if the resources occupied by a machine instruction are available
// in the current state.
bool DFAPacketizer::canReserveResources(MachineInstr &MI) {
const MCInstrDesc &MID = MI.getDesc();
return canReserveResources(&MID);
}
// Reserve the resources occupied by a machine instruction and change the
// current state to reflect that change.
void DFAPacketizer::reserveResources(MachineInstr &MI) {
const MCInstrDesc &MID = MI.getDesc();
reserveResources(&MID);
}
unsigned DFAPacketizer::getUsedResources(unsigned InstIdx) {
ArrayRef<NfaPath> NfaPaths = A.getNfaPaths();
assert(!NfaPaths.empty() && "Invalid bundle!");
const NfaPath &RS = NfaPaths.front();
// RS stores the cumulative resources used up to and including the I'th
// instruction. The 0th instruction is the base case.
if (InstIdx == 0)
return RS[0];
// Return the difference between the cumulative resources used by InstIdx and
// its predecessor.
return RS[InstIdx] ^ RS[InstIdx - 1];
}
namespace llvm {
// This class extends ScheduleDAGInstrs and overrides the schedule method
// to build the dependence graph.
class DefaultVLIWScheduler : public ScheduleDAGInstrs {
private:
AAResults *AA;
/// Ordered list of DAG postprocessing steps.
std::vector<std::unique_ptr<ScheduleDAGMutation>> Mutations;
public:
DefaultVLIWScheduler(MachineFunction &MF, MachineLoopInfo &MLI,
AAResults *AA);
// Actual scheduling work.
void schedule() override;
/// DefaultVLIWScheduler takes ownership of the Mutation object.
void addMutation(std::unique_ptr<ScheduleDAGMutation> Mutation) {
Mutations.push_back(std::move(Mutation));
}
protected:
void postprocessDAG();
};
} // end namespace llvm
DefaultVLIWScheduler::DefaultVLIWScheduler(MachineFunction &MF,
MachineLoopInfo &MLI,
AAResults *AA)
: ScheduleDAGInstrs(MF, &MLI), AA(AA) {
CanHandleTerminators = true;
}
/// Apply each ScheduleDAGMutation step in order.
void DefaultVLIWScheduler::postprocessDAG() {
for (auto &M : Mutations)
M->apply(this);
}
void DefaultVLIWScheduler::schedule() {
// Build the scheduling graph.
buildSchedGraph(AA);
postprocessDAG();
}
VLIWPacketizerList::VLIWPacketizerList(MachineFunction &mf,
MachineLoopInfo &mli, AAResults *aa)
: MF(mf), TII(mf.getSubtarget().getInstrInfo()), AA(aa) {
ResourceTracker = TII->CreateTargetScheduleState(MF.getSubtarget());
ResourceTracker->setTrackResources(true);
VLIWScheduler = new DefaultVLIWScheduler(MF, mli, AA);
}
VLIWPacketizerList::~VLIWPacketizerList() {
delete VLIWScheduler;
delete ResourceTracker;
}
// End the current packet, bundle packet instructions and reset DFA state.
void VLIWPacketizerList::endPacket(MachineBasicBlock *MBB,
MachineBasicBlock::iterator MI) {
LLVM_DEBUG({
if (!CurrentPacketMIs.empty()) {
dbgs() << "Finalizing packet:\n";
unsigned Idx = 0;
for (MachineInstr *MI : CurrentPacketMIs) {
unsigned R = ResourceTracker->getUsedResources(Idx++);
dbgs() << " * [res:0x" << utohexstr(R) << "] " << *MI;
}
}
});
if (CurrentPacketMIs.size() > 1) {
MachineInstr &MIFirst = *CurrentPacketMIs.front();
finalizeBundle(*MBB, MIFirst.getIterator(), MI.getInstrIterator());
}
CurrentPacketMIs.clear();
ResourceTracker->clearResources();
LLVM_DEBUG(dbgs() << "End packet\n");
}
// Bundle machine instructions into packets.
void VLIWPacketizerList::PacketizeMIs(MachineBasicBlock *MBB,
MachineBasicBlock::iterator BeginItr,
MachineBasicBlock::iterator EndItr) {
assert(VLIWScheduler && "VLIW Scheduler is not initialized!");
VLIWScheduler->startBlock(MBB);
VLIWScheduler->enterRegion(MBB, BeginItr, EndItr,
std::distance(BeginItr, EndItr));
VLIWScheduler->schedule();
LLVM_DEBUG({
dbgs() << "Scheduling DAG of the packetize region\n";
VLIWScheduler->dump();
});
// Generate MI -> SU map.
MIToSUnit.clear();
for (SUnit &SU : VLIWScheduler->SUnits)
MIToSUnit[SU.getInstr()] = &SU;
bool LimitPresent = InstrLimit.getPosition();
// The main packetizer loop.
for (; BeginItr != EndItr; ++BeginItr) {
if (LimitPresent) {
if (InstrCount >= InstrLimit) {
EndItr = BeginItr;
break;
}
InstrCount++;
}
MachineInstr &MI = *BeginItr;
initPacketizerState();
// End the current packet if needed.
if (isSoloInstruction(MI)) {
endPacket(MBB, MI);
continue;
}
// Ignore pseudo instructions.
if (ignorePseudoInstruction(MI, MBB))
continue;
SUnit *SUI = MIToSUnit[&MI];
assert(SUI && "Missing SUnit Info!");
// Ask DFA if machine resource is available for MI.
LLVM_DEBUG(dbgs() << "Checking resources for adding MI to packet " << MI);
bool ResourceAvail = ResourceTracker->canReserveResources(MI);
LLVM_DEBUG({
if (ResourceAvail)
dbgs() << " Resources are available for adding MI to packet\n";
else
dbgs() << " Resources NOT available\n";
});
if (ResourceAvail && shouldAddToPacket(MI)) {
// Dependency check for MI with instructions in CurrentPacketMIs.
for (auto MJ : CurrentPacketMIs) {
SUnit *SUJ = MIToSUnit[MJ];
assert(SUJ && "Missing SUnit Info!");
LLVM_DEBUG(dbgs() << " Checking against MJ " << *MJ);
// Is it legal to packetize SUI and SUJ together.
if (!isLegalToPacketizeTogether(SUI, SUJ)) {
LLVM_DEBUG(dbgs() << " Not legal to add MI, try to prune\n");
// Allow packetization if dependency can be pruned.
if (!isLegalToPruneDependencies(SUI, SUJ)) {
// End the packet if dependency cannot be pruned.
LLVM_DEBUG(dbgs()
<< " Could not prune dependencies for adding MI\n");
endPacket(MBB, MI);
break;
}
LLVM_DEBUG(dbgs() << " Pruned dependence for adding MI\n");
}
}
} else {
LLVM_DEBUG(if (ResourceAvail) dbgs()
<< "Resources are available, but instruction should not be "
"added to packet\n "
<< MI);
// End the packet if resource is not available, or if the instruction
// shoud not be added to the current packet.
endPacket(MBB, MI);
}
// Add MI to the current packet.
LLVM_DEBUG(dbgs() << "* Adding MI to packet " << MI << '\n');
BeginItr = addToPacket(MI);
} // For all instructions in the packetization range.
// End any packet left behind.
endPacket(MBB, EndItr);
VLIWScheduler->exitRegion();
VLIWScheduler->finishBlock();
}
bool VLIWPacketizerList::alias(const MachineMemOperand &Op1,
const MachineMemOperand &Op2,
bool UseTBAA) const {
if (!Op1.getValue() || !Op2.getValue())
return true;
int64_t MinOffset = std::min(Op1.getOffset(), Op2.getOffset());
int64_t Overlapa = Op1.getSize() + Op1.getOffset() - MinOffset;
int64_t Overlapb = Op2.getSize() + Op2.getOffset() - MinOffset;
AliasResult AAResult =
AA->alias(MemoryLocation(Op1.getValue(), Overlapa,
UseTBAA ? Op1.getAAInfo() : AAMDNodes()),
MemoryLocation(Op2.getValue(), Overlapb,
UseTBAA ? Op2.getAAInfo() : AAMDNodes()));
return AAResult != AliasResult::NoAlias;
}
bool VLIWPacketizerList::alias(const MachineInstr &MI1,
const MachineInstr &MI2,
bool UseTBAA) const {
if (MI1.memoperands_empty() || MI2.memoperands_empty())
return true;
for (const MachineMemOperand *Op1 : MI1.memoperands())
for (const MachineMemOperand *Op2 : MI2.memoperands())
if (alias(*Op1, *Op2, UseTBAA))
return true;
return false;
}
// Add a DAG mutation object to the ordered list.
void VLIWPacketizerList::addMutation(
std::unique_ptr<ScheduleDAGMutation> Mutation) {
VLIWScheduler->addMutation(std::move(Mutation));
}
|