| File: | build/source/llvm/lib/CodeGen/MachinePipeliner.cpp |
| Warning: | line 1922, column 13 Value stored to 'Order' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- MachinePipeliner.cpp - Machine Software Pipeliner Pass -------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // An implementation of the Swing Modulo Scheduling (SMS) software pipeliner. |
| 10 | // |
| 11 | // This SMS implementation is a target-independent back-end pass. When enabled, |
| 12 | // the pass runs just prior to the register allocation pass, while the machine |
| 13 | // IR is in SSA form. If software pipelining is successful, then the original |
| 14 | // loop is replaced by the optimized loop. The optimized loop contains one or |
| 15 | // more prolog blocks, the pipelined kernel, and one or more epilog blocks. If |
| 16 | // the instructions cannot be scheduled in a given MII, we increase the MII by |
| 17 | // one and try again. |
| 18 | // |
| 19 | // The SMS implementation is an extension of the ScheduleDAGInstrs class. We |
| 20 | // represent loop carried dependences in the DAG as order edges to the Phi |
| 21 | // nodes. We also perform several passes over the DAG to eliminate unnecessary |
| 22 | // edges that inhibit the ability to pipeline. The implementation uses the |
| 23 | // DFAPacketizer class to compute the minimum initiation interval and the check |
| 24 | // where an instruction may be inserted in the pipelined schedule. |
| 25 | // |
| 26 | // In order for the SMS pass to work, several target specific hooks need to be |
| 27 | // implemented to get information about the loop structure and to rewrite |
| 28 | // instructions. |
| 29 | // |
| 30 | //===----------------------------------------------------------------------===// |
| 31 | |
| 32 | #include "llvm/CodeGen/MachinePipeliner.h" |
| 33 | #include "llvm/ADT/ArrayRef.h" |
| 34 | #include "llvm/ADT/BitVector.h" |
| 35 | #include "llvm/ADT/DenseMap.h" |
| 36 | #include "llvm/ADT/MapVector.h" |
| 37 | #include "llvm/ADT/PriorityQueue.h" |
| 38 | #include "llvm/ADT/SetOperations.h" |
| 39 | #include "llvm/ADT/SetVector.h" |
| 40 | #include "llvm/ADT/SmallPtrSet.h" |
| 41 | #include "llvm/ADT/SmallSet.h" |
| 42 | #include "llvm/ADT/SmallVector.h" |
| 43 | #include "llvm/ADT/Statistic.h" |
| 44 | #include "llvm/ADT/iterator_range.h" |
| 45 | #include "llvm/Analysis/AliasAnalysis.h" |
| 46 | #include "llvm/Analysis/CycleAnalysis.h" |
| 47 | #include "llvm/Analysis/MemoryLocation.h" |
| 48 | #include "llvm/Analysis/OptimizationRemarkEmitter.h" |
| 49 | #include "llvm/Analysis/ValueTracking.h" |
| 50 | #include "llvm/CodeGen/DFAPacketizer.h" |
| 51 | #include "llvm/CodeGen/LiveIntervals.h" |
| 52 | #include "llvm/CodeGen/MachineBasicBlock.h" |
| 53 | #include "llvm/CodeGen/MachineDominators.h" |
| 54 | #include "llvm/CodeGen/MachineFunction.h" |
| 55 | #include "llvm/CodeGen/MachineFunctionPass.h" |
| 56 | #include "llvm/CodeGen/MachineInstr.h" |
| 57 | #include "llvm/CodeGen/MachineInstrBuilder.h" |
| 58 | #include "llvm/CodeGen/MachineLoopInfo.h" |
| 59 | #include "llvm/CodeGen/MachineMemOperand.h" |
| 60 | #include "llvm/CodeGen/MachineOperand.h" |
| 61 | #include "llvm/CodeGen/MachineRegisterInfo.h" |
| 62 | #include "llvm/CodeGen/ModuloSchedule.h" |
| 63 | #include "llvm/CodeGen/RegisterPressure.h" |
| 64 | #include "llvm/CodeGen/ScheduleDAG.h" |
| 65 | #include "llvm/CodeGen/ScheduleDAGMutation.h" |
| 66 | #include "llvm/CodeGen/TargetOpcodes.h" |
| 67 | #include "llvm/CodeGen/TargetRegisterInfo.h" |
| 68 | #include "llvm/CodeGen/TargetSubtargetInfo.h" |
| 69 | #include "llvm/Config/llvm-config.h" |
| 70 | #include "llvm/IR/Attributes.h" |
| 71 | #include "llvm/IR/Function.h" |
| 72 | #include "llvm/MC/LaneBitmask.h" |
| 73 | #include "llvm/MC/MCInstrDesc.h" |
| 74 | #include "llvm/MC/MCInstrItineraries.h" |
| 75 | #include "llvm/MC/MCRegisterInfo.h" |
| 76 | #include "llvm/Pass.h" |
| 77 | #include "llvm/Support/CommandLine.h" |
| 78 | #include "llvm/Support/Compiler.h" |
| 79 | #include "llvm/Support/Debug.h" |
| 80 | #include "llvm/Support/MathExtras.h" |
| 81 | #include "llvm/Support/raw_ostream.h" |
| 82 | #include <algorithm> |
| 83 | #include <cassert> |
| 84 | #include <climits> |
| 85 | #include <cstdint> |
| 86 | #include <deque> |
| 87 | #include <functional> |
| 88 | #include <iomanip> |
| 89 | #include <iterator> |
| 90 | #include <map> |
| 91 | #include <memory> |
| 92 | #include <sstream> |
| 93 | #include <tuple> |
| 94 | #include <utility> |
| 95 | #include <vector> |
| 96 | |
| 97 | using namespace llvm; |
| 98 | |
| 99 | #define DEBUG_TYPE"pipeliner" "pipeliner" |
| 100 | |
| 101 | STATISTIC(NumTrytoPipeline, "Number of loops that we attempt to pipeline")static llvm::Statistic NumTrytoPipeline = {"pipeliner", "NumTrytoPipeline" , "Number of loops that we attempt to pipeline"}; |
| 102 | STATISTIC(NumPipelined, "Number of loops software pipelined")static llvm::Statistic NumPipelined = {"pipeliner", "NumPipelined" , "Number of loops software pipelined"}; |
| 103 | STATISTIC(NumNodeOrderIssues, "Number of node order issues found")static llvm::Statistic NumNodeOrderIssues = {"pipeliner", "NumNodeOrderIssues" , "Number of node order issues found"}; |
| 104 | STATISTIC(NumFailBranch, "Pipeliner abort due to unknown branch")static llvm::Statistic NumFailBranch = {"pipeliner", "NumFailBranch" , "Pipeliner abort due to unknown branch"}; |
| 105 | STATISTIC(NumFailLoop, "Pipeliner abort due to unsupported loop")static llvm::Statistic NumFailLoop = {"pipeliner", "NumFailLoop" , "Pipeliner abort due to unsupported loop"}; |
| 106 | STATISTIC(NumFailPreheader, "Pipeliner abort due to missing preheader")static llvm::Statistic NumFailPreheader = {"pipeliner", "NumFailPreheader" , "Pipeliner abort due to missing preheader"}; |
| 107 | STATISTIC(NumFailLargeMaxMII, "Pipeliner abort due to MaxMII too large")static llvm::Statistic NumFailLargeMaxMII = {"pipeliner", "NumFailLargeMaxMII" , "Pipeliner abort due to MaxMII too large"}; |
| 108 | STATISTIC(NumFailZeroMII, "Pipeliner abort due to zero MII")static llvm::Statistic NumFailZeroMII = {"pipeliner", "NumFailZeroMII" , "Pipeliner abort due to zero MII"}; |
| 109 | STATISTIC(NumFailNoSchedule, "Pipeliner abort due to no schedule found")static llvm::Statistic NumFailNoSchedule = {"pipeliner", "NumFailNoSchedule" , "Pipeliner abort due to no schedule found"}; |
| 110 | STATISTIC(NumFailZeroStage, "Pipeliner abort due to zero stage")static llvm::Statistic NumFailZeroStage = {"pipeliner", "NumFailZeroStage" , "Pipeliner abort due to zero stage"}; |
| 111 | STATISTIC(NumFailLargeMaxStage, "Pipeliner abort due to too many stages")static llvm::Statistic NumFailLargeMaxStage = {"pipeliner", "NumFailLargeMaxStage" , "Pipeliner abort due to too many stages"}; |
| 112 | |
| 113 | /// A command line option to turn software pipelining on or off. |
| 114 | static cl::opt<bool> EnableSWP("enable-pipeliner", cl::Hidden, cl::init(true), |
| 115 | cl::desc("Enable Software Pipelining")); |
| 116 | |
| 117 | /// A command line option to enable SWP at -Os. |
| 118 | static cl::opt<bool> EnableSWPOptSize("enable-pipeliner-opt-size", |
| 119 | cl::desc("Enable SWP at Os."), cl::Hidden, |
| 120 | cl::init(false)); |
| 121 | |
| 122 | /// A command line argument to limit minimum initial interval for pipelining. |
| 123 | static cl::opt<int> SwpMaxMii("pipeliner-max-mii", |
| 124 | cl::desc("Size limit for the MII."), |
| 125 | cl::Hidden, cl::init(27)); |
| 126 | |
| 127 | /// A command line argument to force pipeliner to use specified initial |
| 128 | /// interval. |
| 129 | static cl::opt<int> SwpForceII("pipeliner-force-ii", |
| 130 | cl::desc("Force pipeliner to use specified II."), |
| 131 | cl::Hidden, cl::init(-1)); |
| 132 | |
| 133 | /// A command line argument to limit the number of stages in the pipeline. |
| 134 | static cl::opt<int> |
| 135 | SwpMaxStages("pipeliner-max-stages", |
| 136 | cl::desc("Maximum stages allowed in the generated scheduled."), |
| 137 | cl::Hidden, cl::init(3)); |
| 138 | |
| 139 | /// A command line option to disable the pruning of chain dependences due to |
| 140 | /// an unrelated Phi. |
| 141 | static cl::opt<bool> |
| 142 | SwpPruneDeps("pipeliner-prune-deps", |
| 143 | cl::desc("Prune dependences between unrelated Phi nodes."), |
| 144 | cl::Hidden, cl::init(true)); |
| 145 | |
| 146 | /// A command line option to disable the pruning of loop carried order |
| 147 | /// dependences. |
| 148 | static cl::opt<bool> |
| 149 | SwpPruneLoopCarried("pipeliner-prune-loop-carried", |
| 150 | cl::desc("Prune loop carried order dependences."), |
| 151 | cl::Hidden, cl::init(true)); |
| 152 | |
| 153 | #ifndef NDEBUG |
| 154 | static cl::opt<int> SwpLoopLimit("pipeliner-max", cl::Hidden, cl::init(-1)); |
| 155 | #endif |
| 156 | |
| 157 | static cl::opt<bool> SwpIgnoreRecMII("pipeliner-ignore-recmii", |
| 158 | cl::ReallyHidden, |
| 159 | cl::desc("Ignore RecMII")); |
| 160 | |
| 161 | static cl::opt<bool> SwpShowResMask("pipeliner-show-mask", cl::Hidden, |
| 162 | cl::init(false)); |
| 163 | static cl::opt<bool> SwpDebugResource("pipeliner-dbg-res", cl::Hidden, |
| 164 | cl::init(false)); |
| 165 | |
| 166 | static cl::opt<bool> EmitTestAnnotations( |
| 167 | "pipeliner-annotate-for-testing", cl::Hidden, cl::init(false), |
| 168 | cl::desc("Instead of emitting the pipelined code, annotate instructions " |
| 169 | "with the generated schedule for feeding into the " |
| 170 | "-modulo-schedule-test pass")); |
| 171 | |
| 172 | static cl::opt<bool> ExperimentalCodeGen( |
| 173 | "pipeliner-experimental-cg", cl::Hidden, cl::init(false), |
| 174 | cl::desc( |
| 175 | "Use the experimental peeling code generator for software pipelining")); |
| 176 | |
| 177 | namespace llvm { |
| 178 | |
| 179 | // A command line option to enable the CopyToPhi DAG mutation. |
| 180 | cl::opt<bool> SwpEnableCopyToPhi("pipeliner-enable-copytophi", cl::ReallyHidden, |
| 181 | cl::init(true), |
| 182 | cl::desc("Enable CopyToPhi DAG Mutation")); |
| 183 | |
| 184 | /// A command line argument to force pipeliner to use specified issue |
| 185 | /// width. |
| 186 | cl::opt<int> SwpForceIssueWidth( |
| 187 | "pipeliner-force-issue-width", |
| 188 | cl::desc("Force pipeliner to use specified issue width."), cl::Hidden, |
| 189 | cl::init(-1)); |
| 190 | |
| 191 | } // end namespace llvm |
| 192 | |
| 193 | unsigned SwingSchedulerDAG::Circuits::MaxPaths = 5; |
| 194 | char MachinePipeliner::ID = 0; |
| 195 | #ifndef NDEBUG |
| 196 | int MachinePipeliner::NumTries = 0; |
| 197 | #endif |
| 198 | char &llvm::MachinePipelinerID = MachinePipeliner::ID; |
| 199 | |
| 200 | INITIALIZE_PASS_BEGIN(MachinePipeliner, DEBUG_TYPE,static void *initializeMachinePipelinerPassOnce(PassRegistry & Registry) { |
| 201 | "Modulo Software Pipelining", false, false)static void *initializeMachinePipelinerPassOnce(PassRegistry & Registry) { |
| 202 | INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)initializeAAResultsWrapperPassPass(Registry); |
| 203 | INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)initializeMachineLoopInfoPass(Registry); |
| 204 | INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)initializeMachineDominatorTreePass(Registry); |
| 205 | INITIALIZE_PASS_DEPENDENCY(LiveIntervals)initializeLiveIntervalsPass(Registry); |
| 206 | INITIALIZE_PASS_END(MachinePipeliner, DEBUG_TYPE,PassInfo *PI = new PassInfo( "Modulo Software Pipelining", "pipeliner" , &MachinePipeliner::ID, PassInfo::NormalCtor_t(callDefaultCtor <MachinePipeliner>), false, false); Registry.registerPass (*PI, true); return PI; } static llvm::once_flag InitializeMachinePipelinerPassFlag ; void llvm::initializeMachinePipelinerPass(PassRegistry & Registry) { llvm::call_once(InitializeMachinePipelinerPassFlag , initializeMachinePipelinerPassOnce, std::ref(Registry)); } |
| 207 | "Modulo Software Pipelining", false, false)PassInfo *PI = new PassInfo( "Modulo Software Pipelining", "pipeliner" , &MachinePipeliner::ID, PassInfo::NormalCtor_t(callDefaultCtor <MachinePipeliner>), false, false); Registry.registerPass (*PI, true); return PI; } static llvm::once_flag InitializeMachinePipelinerPassFlag ; void llvm::initializeMachinePipelinerPass(PassRegistry & Registry) { llvm::call_once(InitializeMachinePipelinerPassFlag , initializeMachinePipelinerPassOnce, std::ref(Registry)); } |
| 208 | |
| 209 | /// The "main" function for implementing Swing Modulo Scheduling. |
| 210 | bool MachinePipeliner::runOnMachineFunction(MachineFunction &mf) { |
| 211 | if (skipFunction(mf.getFunction())) |
| 212 | return false; |
| 213 | |
| 214 | if (!EnableSWP) |
| 215 | return false; |
| 216 | |
| 217 | if (mf.getFunction().getAttributes().hasFnAttr(Attribute::OptimizeForSize) && |
| 218 | !EnableSWPOptSize.getPosition()) |
| 219 | return false; |
| 220 | |
| 221 | if (!mf.getSubtarget().enableMachinePipeliner()) |
| 222 | return false; |
| 223 | |
| 224 | // Cannot pipeline loops without instruction itineraries if we are using |
| 225 | // DFA for the pipeliner. |
| 226 | if (mf.getSubtarget().useDFAforSMS() && |
| 227 | (!mf.getSubtarget().getInstrItineraryData() || |
| 228 | mf.getSubtarget().getInstrItineraryData()->isEmpty())) |
| 229 | return false; |
| 230 | |
| 231 | MF = &mf; |
| 232 | MLI = &getAnalysis<MachineLoopInfo>(); |
| 233 | MDT = &getAnalysis<MachineDominatorTree>(); |
| 234 | ORE = &getAnalysis<MachineOptimizationRemarkEmitterPass>().getORE(); |
| 235 | TII = MF->getSubtarget().getInstrInfo(); |
| 236 | RegClassInfo.runOnMachineFunction(*MF); |
| 237 | |
| 238 | for (const auto &L : *MLI) |
| 239 | scheduleLoop(*L); |
| 240 | |
| 241 | return false; |
| 242 | } |
| 243 | |
| 244 | /// Attempt to perform the SMS algorithm on the specified loop. This function is |
| 245 | /// the main entry point for the algorithm. The function identifies candidate |
| 246 | /// loops, calculates the minimum initiation interval, and attempts to schedule |
| 247 | /// the loop. |
| 248 | bool MachinePipeliner::scheduleLoop(MachineLoop &L) { |
| 249 | bool Changed = false; |
| 250 | for (const auto &InnerLoop : L) |
| 251 | Changed |= scheduleLoop(*InnerLoop); |
| 252 | |
| 253 | #ifndef NDEBUG |
| 254 | // Stop trying after reaching the limit (if any). |
| 255 | int Limit = SwpLoopLimit; |
| 256 | if (Limit >= 0) { |
| 257 | if (NumTries >= SwpLoopLimit) |
| 258 | return Changed; |
| 259 | NumTries++; |
| 260 | } |
| 261 | #endif |
| 262 | |
| 263 | setPragmaPipelineOptions(L); |
| 264 | if (!canPipelineLoop(L)) { |
| 265 | LLVM_DEBUG(dbgs() << "\n!!! Can not pipeline loop.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "\n!!! Can not pipeline loop.\n" ; } } while (false); |
| 266 | ORE->emit([&]() { |
| 267 | return MachineOptimizationRemarkMissed(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 268 | L.getStartLoc(), L.getHeader()) |
| 269 | << "Failed to pipeline loop"; |
| 270 | }); |
| 271 | |
| 272 | LI.LoopPipelinerInfo.reset(); |
| 273 | return Changed; |
| 274 | } |
| 275 | |
| 276 | ++NumTrytoPipeline; |
| 277 | |
| 278 | Changed = swingModuloScheduler(L); |
| 279 | |
| 280 | LI.LoopPipelinerInfo.reset(); |
| 281 | return Changed; |
| 282 | } |
| 283 | |
| 284 | void MachinePipeliner::setPragmaPipelineOptions(MachineLoop &L) { |
| 285 | // Reset the pragma for the next loop in iteration. |
| 286 | disabledByPragma = false; |
| 287 | II_setByPragma = 0; |
| 288 | |
| 289 | MachineBasicBlock *LBLK = L.getTopBlock(); |
| 290 | |
| 291 | if (LBLK == nullptr) |
| 292 | return; |
| 293 | |
| 294 | const BasicBlock *BBLK = LBLK->getBasicBlock(); |
| 295 | if (BBLK == nullptr) |
| 296 | return; |
| 297 | |
| 298 | const Instruction *TI = BBLK->getTerminator(); |
| 299 | if (TI == nullptr) |
| 300 | return; |
| 301 | |
| 302 | MDNode *LoopID = TI->getMetadata(LLVMContext::MD_loop); |
| 303 | if (LoopID == nullptr) |
| 304 | return; |
| 305 | |
| 306 | assert(LoopID->getNumOperands() > 0 && "requires atleast one operand")(static_cast <bool> (LoopID->getNumOperands() > 0 && "requires atleast one operand") ? void (0) : __assert_fail ("LoopID->getNumOperands() > 0 && \"requires atleast one operand\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 306, __extension__ __PRETTY_FUNCTION__)); |
| 307 | assert(LoopID->getOperand(0) == LoopID && "invalid loop")(static_cast <bool> (LoopID->getOperand(0) == LoopID && "invalid loop") ? void (0) : __assert_fail ("LoopID->getOperand(0) == LoopID && \"invalid loop\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 307, __extension__ __PRETTY_FUNCTION__)); |
| 308 | |
| 309 | for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) { |
| 310 | MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i)); |
| 311 | |
| 312 | if (MD == nullptr) |
| 313 | continue; |
| 314 | |
| 315 | MDString *S = dyn_cast<MDString>(MD->getOperand(0)); |
| 316 | |
| 317 | if (S == nullptr) |
| 318 | continue; |
| 319 | |
| 320 | if (S->getString() == "llvm.loop.pipeline.initiationinterval") { |
| 321 | assert(MD->getNumOperands() == 2 &&(static_cast <bool> (MD->getNumOperands() == 2 && "Pipeline initiation interval hint metadata should have two operands." ) ? void (0) : __assert_fail ("MD->getNumOperands() == 2 && \"Pipeline initiation interval hint metadata should have two operands.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 322, __extension__ __PRETTY_FUNCTION__)) |
| 322 | "Pipeline initiation interval hint metadata should have two operands.")(static_cast <bool> (MD->getNumOperands() == 2 && "Pipeline initiation interval hint metadata should have two operands." ) ? void (0) : __assert_fail ("MD->getNumOperands() == 2 && \"Pipeline initiation interval hint metadata should have two operands.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 322, __extension__ __PRETTY_FUNCTION__)); |
| 323 | II_setByPragma = |
| 324 | mdconst::extract<ConstantInt>(MD->getOperand(1))->getZExtValue(); |
| 325 | assert(II_setByPragma >= 1 && "Pipeline initiation interval must be positive.")(static_cast <bool> (II_setByPragma >= 1 && "Pipeline initiation interval must be positive." ) ? void (0) : __assert_fail ("II_setByPragma >= 1 && \"Pipeline initiation interval must be positive.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 325, __extension__ __PRETTY_FUNCTION__)); |
| 326 | } else if (S->getString() == "llvm.loop.pipeline.disable") { |
| 327 | disabledByPragma = true; |
| 328 | } |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | /// Return true if the loop can be software pipelined. The algorithm is |
| 333 | /// restricted to loops with a single basic block. Make sure that the |
| 334 | /// branch in the loop can be analyzed. |
| 335 | bool MachinePipeliner::canPipelineLoop(MachineLoop &L) { |
| 336 | if (L.getNumBlocks() != 1) { |
| 337 | ORE->emit([&]() { |
| 338 | return MachineOptimizationRemarkAnalysis(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 339 | L.getStartLoc(), L.getHeader()) |
| 340 | << "Not a single basic block: " |
| 341 | << ore::NV("NumBlocks", L.getNumBlocks()); |
| 342 | }); |
| 343 | return false; |
| 344 | } |
| 345 | |
| 346 | if (disabledByPragma) { |
| 347 | ORE->emit([&]() { |
| 348 | return MachineOptimizationRemarkAnalysis(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 349 | L.getStartLoc(), L.getHeader()) |
| 350 | << "Disabled by Pragma."; |
| 351 | }); |
| 352 | return false; |
| 353 | } |
| 354 | |
| 355 | // Check if the branch can't be understood because we can't do pipelining |
| 356 | // if that's the case. |
| 357 | LI.TBB = nullptr; |
| 358 | LI.FBB = nullptr; |
| 359 | LI.BrCond.clear(); |
| 360 | if (TII->analyzeBranch(*L.getHeader(), LI.TBB, LI.FBB, LI.BrCond)) { |
| 361 | LLVM_DEBUG(dbgs() << "Unable to analyzeBranch, can NOT pipeline Loop\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Unable to analyzeBranch, can NOT pipeline Loop\n" ; } } while (false); |
| 362 | NumFailBranch++; |
| 363 | ORE->emit([&]() { |
| 364 | return MachineOptimizationRemarkAnalysis(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 365 | L.getStartLoc(), L.getHeader()) |
| 366 | << "The branch can't be understood"; |
| 367 | }); |
| 368 | return false; |
| 369 | } |
| 370 | |
| 371 | LI.LoopInductionVar = nullptr; |
| 372 | LI.LoopCompare = nullptr; |
| 373 | LI.LoopPipelinerInfo = TII->analyzeLoopForPipelining(L.getTopBlock()); |
| 374 | if (!LI.LoopPipelinerInfo) { |
| 375 | LLVM_DEBUG(dbgs() << "Unable to analyzeLoop, can NOT pipeline Loop\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Unable to analyzeLoop, can NOT pipeline Loop\n" ; } } while (false); |
| 376 | NumFailLoop++; |
| 377 | ORE->emit([&]() { |
| 378 | return MachineOptimizationRemarkAnalysis(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 379 | L.getStartLoc(), L.getHeader()) |
| 380 | << "The loop structure is not supported"; |
| 381 | }); |
| 382 | return false; |
| 383 | } |
| 384 | |
| 385 | if (!L.getLoopPreheader()) { |
| 386 | LLVM_DEBUG(dbgs() << "Preheader not found, can NOT pipeline Loop\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Preheader not found, can NOT pipeline Loop\n" ; } } while (false); |
| 387 | NumFailPreheader++; |
| 388 | ORE->emit([&]() { |
| 389 | return MachineOptimizationRemarkAnalysis(DEBUG_TYPE"pipeliner", "canPipelineLoop", |
| 390 | L.getStartLoc(), L.getHeader()) |
| 391 | << "No loop preheader found"; |
| 392 | }); |
| 393 | return false; |
| 394 | } |
| 395 | |
| 396 | // Remove any subregisters from inputs to phi nodes. |
| 397 | preprocessPhiNodes(*L.getHeader()); |
| 398 | return true; |
| 399 | } |
| 400 | |
| 401 | void MachinePipeliner::preprocessPhiNodes(MachineBasicBlock &B) { |
| 402 | MachineRegisterInfo &MRI = MF->getRegInfo(); |
| 403 | SlotIndexes &Slots = *getAnalysis<LiveIntervals>().getSlotIndexes(); |
| 404 | |
| 405 | for (MachineInstr &PI : B.phis()) { |
| 406 | MachineOperand &DefOp = PI.getOperand(0); |
| 407 | assert(DefOp.getSubReg() == 0)(static_cast <bool> (DefOp.getSubReg() == 0) ? void (0) : __assert_fail ("DefOp.getSubReg() == 0", "llvm/lib/CodeGen/MachinePipeliner.cpp" , 407, __extension__ __PRETTY_FUNCTION__)); |
| 408 | auto *RC = MRI.getRegClass(DefOp.getReg()); |
| 409 | |
| 410 | for (unsigned i = 1, n = PI.getNumOperands(); i != n; i += 2) { |
| 411 | MachineOperand &RegOp = PI.getOperand(i); |
| 412 | if (RegOp.getSubReg() == 0) |
| 413 | continue; |
| 414 | |
| 415 | // If the operand uses a subregister, replace it with a new register |
| 416 | // without subregisters, and generate a copy to the new register. |
| 417 | Register NewReg = MRI.createVirtualRegister(RC); |
| 418 | MachineBasicBlock &PredB = *PI.getOperand(i+1).getMBB(); |
| 419 | MachineBasicBlock::iterator At = PredB.getFirstTerminator(); |
| 420 | const DebugLoc &DL = PredB.findDebugLoc(At); |
| 421 | auto Copy = BuildMI(PredB, At, DL, TII->get(TargetOpcode::COPY), NewReg) |
| 422 | .addReg(RegOp.getReg(), getRegState(RegOp), |
| 423 | RegOp.getSubReg()); |
| 424 | Slots.insertMachineInstrInMaps(*Copy); |
| 425 | RegOp.setReg(NewReg); |
| 426 | RegOp.setSubReg(0); |
| 427 | } |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | /// The SMS algorithm consists of the following main steps: |
| 432 | /// 1. Computation and analysis of the dependence graph. |
| 433 | /// 2. Ordering of the nodes (instructions). |
| 434 | /// 3. Attempt to Schedule the loop. |
| 435 | bool MachinePipeliner::swingModuloScheduler(MachineLoop &L) { |
| 436 | assert(L.getBlocks().size() == 1 && "SMS works on single blocks only.")(static_cast <bool> (L.getBlocks().size() == 1 && "SMS works on single blocks only.") ? void (0) : __assert_fail ("L.getBlocks().size() == 1 && \"SMS works on single blocks only.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 436, __extension__ __PRETTY_FUNCTION__)); |
| 437 | |
| 438 | SwingSchedulerDAG SMS(*this, L, getAnalysis<LiveIntervals>(), RegClassInfo, |
| 439 | II_setByPragma, LI.LoopPipelinerInfo.get()); |
| 440 | |
| 441 | MachineBasicBlock *MBB = L.getHeader(); |
| 442 | // The kernel should not include any terminator instructions. These |
| 443 | // will be added back later. |
| 444 | SMS.startBlock(MBB); |
| 445 | |
| 446 | // Compute the number of 'real' instructions in the basic block by |
| 447 | // ignoring terminators. |
| 448 | unsigned size = MBB->size(); |
| 449 | for (MachineBasicBlock::iterator I = MBB->getFirstTerminator(), |
| 450 | E = MBB->instr_end(); |
| 451 | I != E; ++I, --size) |
| 452 | ; |
| 453 | |
| 454 | SMS.enterRegion(MBB, MBB->begin(), MBB->getFirstTerminator(), size); |
| 455 | SMS.schedule(); |
| 456 | SMS.exitRegion(); |
| 457 | |
| 458 | SMS.finishBlock(); |
| 459 | return SMS.hasNewSchedule(); |
| 460 | } |
| 461 | |
| 462 | void MachinePipeliner::getAnalysisUsage(AnalysisUsage &AU) const { |
| 463 | AU.addRequired<AAResultsWrapperPass>(); |
| 464 | AU.addPreserved<AAResultsWrapperPass>(); |
| 465 | AU.addRequired<MachineLoopInfo>(); |
| 466 | AU.addRequired<MachineDominatorTree>(); |
| 467 | AU.addRequired<LiveIntervals>(); |
| 468 | AU.addRequired<MachineOptimizationRemarkEmitterPass>(); |
| 469 | MachineFunctionPass::getAnalysisUsage(AU); |
| 470 | } |
| 471 | |
| 472 | void SwingSchedulerDAG::setMII(unsigned ResMII, unsigned RecMII) { |
| 473 | if (SwpForceII > 0) |
| 474 | MII = SwpForceII; |
| 475 | else if (II_setByPragma > 0) |
| 476 | MII = II_setByPragma; |
| 477 | else |
| 478 | MII = std::max(ResMII, RecMII); |
| 479 | } |
| 480 | |
| 481 | void SwingSchedulerDAG::setMAX_II() { |
| 482 | if (SwpForceII > 0) |
| 483 | MAX_II = SwpForceII; |
| 484 | else if (II_setByPragma > 0) |
| 485 | MAX_II = II_setByPragma; |
| 486 | else |
| 487 | MAX_II = MII + 10; |
| 488 | } |
| 489 | |
| 490 | /// We override the schedule function in ScheduleDAGInstrs to implement the |
| 491 | /// scheduling part of the Swing Modulo Scheduling algorithm. |
| 492 | void SwingSchedulerDAG::schedule() { |
| 493 | AliasAnalysis *AA = &Pass.getAnalysis<AAResultsWrapperPass>().getAAResults(); |
| 494 | buildSchedGraph(AA); |
| 495 | addLoopCarriedDependences(AA); |
| 496 | updatePhiDependences(); |
| 497 | Topo.InitDAGTopologicalSorting(); |
| 498 | changeDependences(); |
| 499 | postProcessDAG(); |
| 500 | LLVM_DEBUG(dump())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dump(); } } while (false); |
| 501 | |
| 502 | NodeSetType NodeSets; |
| 503 | findCircuits(NodeSets); |
| 504 | NodeSetType Circuits = NodeSets; |
| 505 | |
| 506 | // Calculate the MII. |
| 507 | unsigned ResMII = calculateResMII(); |
| 508 | unsigned RecMII = calculateRecMII(NodeSets); |
| 509 | |
| 510 | fuseRecs(NodeSets); |
| 511 | |
| 512 | // This flag is used for testing and can cause correctness problems. |
| 513 | if (SwpIgnoreRecMII) |
| 514 | RecMII = 0; |
| 515 | |
| 516 | setMII(ResMII, RecMII); |
| 517 | setMAX_II(); |
| 518 | |
| 519 | LLVM_DEBUG(dbgs() << "MII = " << MII << " MAX_II = " << MAX_IIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "MII = " << MII << " MAX_II = " << MAX_II << " (rec=" << RecMII << ", res=" << ResMII << ")\n"; } } while ( false) |
| 520 | << " (rec=" << RecMII << ", res=" << ResMII << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "MII = " << MII << " MAX_II = " << MAX_II << " (rec=" << RecMII << ", res=" << ResMII << ")\n"; } } while ( false); |
| 521 | |
| 522 | // Can't schedule a loop without a valid MII. |
| 523 | if (MII == 0) { |
| 524 | LLVM_DEBUG(dbgs() << "Invalid Minimal Initiation Interval: 0\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Invalid Minimal Initiation Interval: 0\n" ; } } while (false); |
| 525 | NumFailZeroMII++; |
| 526 | Pass.ORE->emit([&]() { |
| 527 | return MachineOptimizationRemarkAnalysis( |
| 528 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 529 | << "Invalid Minimal Initiation Interval: 0"; |
| 530 | }); |
| 531 | return; |
| 532 | } |
| 533 | |
| 534 | // Don't pipeline large loops. |
| 535 | if (SwpMaxMii != -1 && (int)MII > SwpMaxMii) { |
| 536 | LLVM_DEBUG(dbgs() << "MII > " << SwpMaxMiido { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "MII > " << SwpMaxMii << ", we don't pipeline large loops\n"; } } while (false ) |
| 537 | << ", we don't pipeline large loops\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "MII > " << SwpMaxMii << ", we don't pipeline large loops\n"; } } while (false ); |
| 538 | NumFailLargeMaxMII++; |
| 539 | Pass.ORE->emit([&]() { |
| 540 | return MachineOptimizationRemarkAnalysis( |
| 541 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 542 | << "Minimal Initiation Interval too large: " |
| 543 | << ore::NV("MII", (int)MII) << " > " |
| 544 | << ore::NV("SwpMaxMii", SwpMaxMii) << "." |
| 545 | << "Refer to -pipeliner-max-mii."; |
| 546 | }); |
| 547 | return; |
| 548 | } |
| 549 | |
| 550 | computeNodeFunctions(NodeSets); |
| 551 | |
| 552 | registerPressureFilter(NodeSets); |
| 553 | |
| 554 | colocateNodeSets(NodeSets); |
| 555 | |
| 556 | checkNodeSets(NodeSets); |
| 557 | |
| 558 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false) |
| 559 | for (auto &I : NodeSets) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false) |
| 560 | dbgs() << " Rec NodeSet ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false) |
| 561 | I.dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false) |
| 562 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false) |
| 563 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " Rec NodeSet "; I.dump(); } }; } } while (false); |
| 564 | |
| 565 | llvm::stable_sort(NodeSets, std::greater<NodeSet>()); |
| 566 | |
| 567 | groupRemainingNodes(NodeSets); |
| 568 | |
| 569 | removeDuplicateNodes(NodeSets); |
| 570 | |
| 571 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false) |
| 572 | for (auto &I : NodeSets) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false) |
| 573 | dbgs() << " NodeSet ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false) |
| 574 | I.dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false) |
| 575 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false) |
| 576 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (auto &I : NodeSets) { dbgs() << " NodeSet "; I.dump(); } }; } } while (false); |
| 577 | |
| 578 | computeNodeOrder(NodeSets); |
| 579 | |
| 580 | // check for node order issues |
| 581 | checkValidNodeOrder(Circuits); |
| 582 | |
| 583 | SMSchedule Schedule(Pass.MF, this); |
| 584 | Scheduled = schedulePipeline(Schedule); |
| 585 | |
| 586 | if (!Scheduled){ |
| 587 | LLVM_DEBUG(dbgs() << "No schedule found, return\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "No schedule found, return\n" ; } } while (false); |
| 588 | NumFailNoSchedule++; |
| 589 | Pass.ORE->emit([&]() { |
| 590 | return MachineOptimizationRemarkAnalysis( |
| 591 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 592 | << "Unable to find schedule"; |
| 593 | }); |
| 594 | return; |
| 595 | } |
| 596 | |
| 597 | unsigned numStages = Schedule.getMaxStageCount(); |
| 598 | // No need to generate pipeline if there are no overlapped iterations. |
| 599 | if (numStages == 0) { |
| 600 | LLVM_DEBUG(dbgs() << "No overlapped iterations, skip.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "No overlapped iterations, skip.\n" ; } } while (false); |
| 601 | NumFailZeroStage++; |
| 602 | Pass.ORE->emit([&]() { |
| 603 | return MachineOptimizationRemarkAnalysis( |
| 604 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 605 | << "No need to pipeline - no overlapped iterations in schedule."; |
| 606 | }); |
| 607 | return; |
| 608 | } |
| 609 | // Check that the maximum stage count is less than user-defined limit. |
| 610 | if (SwpMaxStages > -1 && (int)numStages > SwpMaxStages) { |
| 611 | LLVM_DEBUG(dbgs() << "numStages:" << numStages << ">" << SwpMaxStagesdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "numStages:" << numStages << ">" << SwpMaxStages << " : too many stages, abort\n" ; } } while (false) |
| 612 | << " : too many stages, abort\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "numStages:" << numStages << ">" << SwpMaxStages << " : too many stages, abort\n" ; } } while (false); |
| 613 | NumFailLargeMaxStage++; |
| 614 | Pass.ORE->emit([&]() { |
| 615 | return MachineOptimizationRemarkAnalysis( |
| 616 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 617 | << "Too many stages in schedule: " |
| 618 | << ore::NV("numStages", (int)numStages) << " > " |
| 619 | << ore::NV("SwpMaxStages", SwpMaxStages) |
| 620 | << ". Refer to -pipeliner-max-stages."; |
| 621 | }); |
| 622 | return; |
| 623 | } |
| 624 | |
| 625 | Pass.ORE->emit([&]() { |
| 626 | return MachineOptimizationRemark(DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), |
| 627 | Loop.getHeader()) |
| 628 | << "Pipelined succesfully!"; |
| 629 | }); |
| 630 | |
| 631 | // Generate the schedule as a ModuloSchedule. |
| 632 | DenseMap<MachineInstr *, int> Cycles, Stages; |
| 633 | std::vector<MachineInstr *> OrderedInsts; |
| 634 | for (int Cycle = Schedule.getFirstCycle(); Cycle <= Schedule.getFinalCycle(); |
| 635 | ++Cycle) { |
| 636 | for (SUnit *SU : Schedule.getInstructions(Cycle)) { |
| 637 | OrderedInsts.push_back(SU->getInstr()); |
| 638 | Cycles[SU->getInstr()] = Cycle; |
| 639 | Stages[SU->getInstr()] = Schedule.stageScheduled(SU); |
| 640 | } |
| 641 | } |
| 642 | DenseMap<MachineInstr *, std::pair<unsigned, int64_t>> NewInstrChanges; |
| 643 | for (auto &KV : NewMIs) { |
| 644 | Cycles[KV.first] = Cycles[KV.second]; |
| 645 | Stages[KV.first] = Stages[KV.second]; |
| 646 | NewInstrChanges[KV.first] = InstrChanges[getSUnit(KV.first)]; |
| 647 | } |
| 648 | |
| 649 | ModuloSchedule MS(MF, &Loop, std::move(OrderedInsts), std::move(Cycles), |
| 650 | std::move(Stages)); |
| 651 | if (EmitTestAnnotations) { |
| 652 | assert(NewInstrChanges.empty() &&(static_cast <bool> (NewInstrChanges.empty() && "Cannot serialize a schedule with InstrChanges!") ? void (0) : __assert_fail ("NewInstrChanges.empty() && \"Cannot serialize a schedule with InstrChanges!\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 653, __extension__ __PRETTY_FUNCTION__)) |
| 653 | "Cannot serialize a schedule with InstrChanges!")(static_cast <bool> (NewInstrChanges.empty() && "Cannot serialize a schedule with InstrChanges!") ? void (0) : __assert_fail ("NewInstrChanges.empty() && \"Cannot serialize a schedule with InstrChanges!\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 653, __extension__ __PRETTY_FUNCTION__)); |
| 654 | ModuloScheduleTestAnnotater MSTI(MF, MS); |
| 655 | MSTI.annotate(); |
| 656 | return; |
| 657 | } |
| 658 | // The experimental code generator can't work if there are InstChanges. |
| 659 | if (ExperimentalCodeGen && NewInstrChanges.empty()) { |
| 660 | PeelingModuloScheduleExpander MSE(MF, MS, &LIS); |
| 661 | MSE.expand(); |
| 662 | } else { |
| 663 | ModuloScheduleExpander MSE(MF, MS, LIS, std::move(NewInstrChanges)); |
| 664 | MSE.expand(); |
| 665 | MSE.cleanup(); |
| 666 | } |
| 667 | ++NumPipelined; |
| 668 | } |
| 669 | |
| 670 | /// Clean up after the software pipeliner runs. |
| 671 | void SwingSchedulerDAG::finishBlock() { |
| 672 | for (auto &KV : NewMIs) |
| 673 | MF.deleteMachineInstr(KV.second); |
| 674 | NewMIs.clear(); |
| 675 | |
| 676 | // Call the superclass. |
| 677 | ScheduleDAGInstrs::finishBlock(); |
| 678 | } |
| 679 | |
| 680 | /// Return the register values for the operands of a Phi instruction. |
| 681 | /// This function assume the instruction is a Phi. |
| 682 | static void getPhiRegs(MachineInstr &Phi, MachineBasicBlock *Loop, |
| 683 | unsigned &InitVal, unsigned &LoopVal) { |
| 684 | assert(Phi.isPHI() && "Expecting a Phi.")(static_cast <bool> (Phi.isPHI() && "Expecting a Phi." ) ? void (0) : __assert_fail ("Phi.isPHI() && \"Expecting a Phi.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 684, __extension__ __PRETTY_FUNCTION__)); |
| 685 | |
| 686 | InitVal = 0; |
| 687 | LoopVal = 0; |
| 688 | for (unsigned i = 1, e = Phi.getNumOperands(); i != e; i += 2) |
| 689 | if (Phi.getOperand(i + 1).getMBB() != Loop) |
| 690 | InitVal = Phi.getOperand(i).getReg(); |
| 691 | else |
| 692 | LoopVal = Phi.getOperand(i).getReg(); |
| 693 | |
| 694 | assert(InitVal != 0 && LoopVal != 0 && "Unexpected Phi structure.")(static_cast <bool> (InitVal != 0 && LoopVal != 0 && "Unexpected Phi structure.") ? void (0) : __assert_fail ("InitVal != 0 && LoopVal != 0 && \"Unexpected Phi structure.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 694, __extension__ __PRETTY_FUNCTION__)); |
| 695 | } |
| 696 | |
| 697 | /// Return the Phi register value that comes the loop block. |
| 698 | static unsigned getLoopPhiReg(MachineInstr &Phi, MachineBasicBlock *LoopBB) { |
| 699 | for (unsigned i = 1, e = Phi.getNumOperands(); i != e; i += 2) |
| 700 | if (Phi.getOperand(i + 1).getMBB() == LoopBB) |
| 701 | return Phi.getOperand(i).getReg(); |
| 702 | return 0; |
| 703 | } |
| 704 | |
| 705 | /// Return true if SUb can be reached from SUa following the chain edges. |
| 706 | static bool isSuccOrder(SUnit *SUa, SUnit *SUb) { |
| 707 | SmallPtrSet<SUnit *, 8> Visited; |
| 708 | SmallVector<SUnit *, 8> Worklist; |
| 709 | Worklist.push_back(SUa); |
| 710 | while (!Worklist.empty()) { |
| 711 | const SUnit *SU = Worklist.pop_back_val(); |
| 712 | for (const auto &SI : SU->Succs) { |
| 713 | SUnit *SuccSU = SI.getSUnit(); |
| 714 | if (SI.getKind() == SDep::Order) { |
| 715 | if (Visited.count(SuccSU)) |
| 716 | continue; |
| 717 | if (SuccSU == SUb) |
| 718 | return true; |
| 719 | Worklist.push_back(SuccSU); |
| 720 | Visited.insert(SuccSU); |
| 721 | } |
| 722 | } |
| 723 | } |
| 724 | return false; |
| 725 | } |
| 726 | |
| 727 | /// Return true if the instruction causes a chain between memory |
| 728 | /// references before and after it. |
| 729 | static bool isDependenceBarrier(MachineInstr &MI) { |
| 730 | return MI.isCall() || MI.mayRaiseFPException() || |
| 731 | MI.hasUnmodeledSideEffects() || |
| 732 | (MI.hasOrderedMemoryRef() && |
| 733 | (!MI.mayLoad() || !MI.isDereferenceableInvariantLoad())); |
| 734 | } |
| 735 | |
| 736 | /// Return the underlying objects for the memory references of an instruction. |
| 737 | /// This function calls the code in ValueTracking, but first checks that the |
| 738 | /// instruction has a memory operand. |
| 739 | static void getUnderlyingObjects(const MachineInstr *MI, |
| 740 | SmallVectorImpl<const Value *> &Objs) { |
| 741 | if (!MI->hasOneMemOperand()) |
| 742 | return; |
| 743 | MachineMemOperand *MM = *MI->memoperands_begin(); |
| 744 | if (!MM->getValue()) |
| 745 | return; |
| 746 | getUnderlyingObjects(MM->getValue(), Objs); |
| 747 | for (const Value *V : Objs) { |
| 748 | if (!isIdentifiedObject(V)) { |
| 749 | Objs.clear(); |
| 750 | return; |
| 751 | } |
| 752 | Objs.push_back(V); |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | /// Add a chain edge between a load and store if the store can be an |
| 757 | /// alias of the load on a subsequent iteration, i.e., a loop carried |
| 758 | /// dependence. This code is very similar to the code in ScheduleDAGInstrs |
| 759 | /// but that code doesn't create loop carried dependences. |
| 760 | void SwingSchedulerDAG::addLoopCarriedDependences(AliasAnalysis *AA) { |
| 761 | MapVector<const Value *, SmallVector<SUnit *, 4>> PendingLoads; |
| 762 | Value *UnknownValue = |
| 763 | UndefValue::get(Type::getVoidTy(MF.getFunction().getContext())); |
| 764 | for (auto &SU : SUnits) { |
| 765 | MachineInstr &MI = *SU.getInstr(); |
| 766 | if (isDependenceBarrier(MI)) |
| 767 | PendingLoads.clear(); |
| 768 | else if (MI.mayLoad()) { |
| 769 | SmallVector<const Value *, 4> Objs; |
| 770 | ::getUnderlyingObjects(&MI, Objs); |
| 771 | if (Objs.empty()) |
| 772 | Objs.push_back(UnknownValue); |
| 773 | for (const auto *V : Objs) { |
| 774 | SmallVector<SUnit *, 4> &SUs = PendingLoads[V]; |
| 775 | SUs.push_back(&SU); |
| 776 | } |
| 777 | } else if (MI.mayStore()) { |
| 778 | SmallVector<const Value *, 4> Objs; |
| 779 | ::getUnderlyingObjects(&MI, Objs); |
| 780 | if (Objs.empty()) |
| 781 | Objs.push_back(UnknownValue); |
| 782 | for (const auto *V : Objs) { |
| 783 | MapVector<const Value *, SmallVector<SUnit *, 4>>::iterator I = |
| 784 | PendingLoads.find(V); |
| 785 | if (I == PendingLoads.end()) |
| 786 | continue; |
| 787 | for (auto *Load : I->second) { |
| 788 | if (isSuccOrder(Load, &SU)) |
| 789 | continue; |
| 790 | MachineInstr &LdMI = *Load->getInstr(); |
| 791 | // First, perform the cheaper check that compares the base register. |
| 792 | // If they are the same and the load offset is less than the store |
| 793 | // offset, then mark the dependence as loop carried potentially. |
| 794 | const MachineOperand *BaseOp1, *BaseOp2; |
| 795 | int64_t Offset1, Offset2; |
| 796 | bool Offset1IsScalable, Offset2IsScalable; |
| 797 | if (TII->getMemOperandWithOffset(LdMI, BaseOp1, Offset1, |
| 798 | Offset1IsScalable, TRI) && |
| 799 | TII->getMemOperandWithOffset(MI, BaseOp2, Offset2, |
| 800 | Offset2IsScalable, TRI)) { |
| 801 | if (BaseOp1->isIdenticalTo(*BaseOp2) && |
| 802 | Offset1IsScalable == Offset2IsScalable && |
| 803 | (int)Offset1 < (int)Offset2) { |
| 804 | assert(TII->areMemAccessesTriviallyDisjoint(LdMI, MI) &&(static_cast <bool> (TII->areMemAccessesTriviallyDisjoint (LdMI, MI) && "What happened to the chain edge?") ? void (0) : __assert_fail ("TII->areMemAccessesTriviallyDisjoint(LdMI, MI) && \"What happened to the chain edge?\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 805, __extension__ __PRETTY_FUNCTION__)) |
| 805 | "What happened to the chain edge?")(static_cast <bool> (TII->areMemAccessesTriviallyDisjoint (LdMI, MI) && "What happened to the chain edge?") ? void (0) : __assert_fail ("TII->areMemAccessesTriviallyDisjoint(LdMI, MI) && \"What happened to the chain edge?\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 805, __extension__ __PRETTY_FUNCTION__)); |
| 806 | SDep Dep(Load, SDep::Barrier); |
| 807 | Dep.setLatency(1); |
| 808 | SU.addPred(Dep); |
| 809 | continue; |
| 810 | } |
| 811 | } |
| 812 | // Second, the more expensive check that uses alias analysis on the |
| 813 | // base registers. If they alias, and the load offset is less than |
| 814 | // the store offset, the mark the dependence as loop carried. |
| 815 | if (!AA) { |
| 816 | SDep Dep(Load, SDep::Barrier); |
| 817 | Dep.setLatency(1); |
| 818 | SU.addPred(Dep); |
| 819 | continue; |
| 820 | } |
| 821 | MachineMemOperand *MMO1 = *LdMI.memoperands_begin(); |
| 822 | MachineMemOperand *MMO2 = *MI.memoperands_begin(); |
| 823 | if (!MMO1->getValue() || !MMO2->getValue()) { |
| 824 | SDep Dep(Load, SDep::Barrier); |
| 825 | Dep.setLatency(1); |
| 826 | SU.addPred(Dep); |
| 827 | continue; |
| 828 | } |
| 829 | if (MMO1->getValue() == MMO2->getValue() && |
| 830 | MMO1->getOffset() <= MMO2->getOffset()) { |
| 831 | SDep Dep(Load, SDep::Barrier); |
| 832 | Dep.setLatency(1); |
| 833 | SU.addPred(Dep); |
| 834 | continue; |
| 835 | } |
| 836 | if (!AA->isNoAlias( |
| 837 | MemoryLocation::getAfter(MMO1->getValue(), MMO1->getAAInfo()), |
| 838 | MemoryLocation::getAfter(MMO2->getValue(), |
| 839 | MMO2->getAAInfo()))) { |
| 840 | SDep Dep(Load, SDep::Barrier); |
| 841 | Dep.setLatency(1); |
| 842 | SU.addPred(Dep); |
| 843 | } |
| 844 | } |
| 845 | } |
| 846 | } |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | /// Update the phi dependences to the DAG because ScheduleDAGInstrs no longer |
| 851 | /// processes dependences for PHIs. This function adds true dependences |
| 852 | /// from a PHI to a use, and a loop carried dependence from the use to the |
| 853 | /// PHI. The loop carried dependence is represented as an anti dependence |
| 854 | /// edge. This function also removes chain dependences between unrelated |
| 855 | /// PHIs. |
| 856 | void SwingSchedulerDAG::updatePhiDependences() { |
| 857 | SmallVector<SDep, 4> RemoveDeps; |
| 858 | const TargetSubtargetInfo &ST = MF.getSubtarget<TargetSubtargetInfo>(); |
| 859 | |
| 860 | // Iterate over each DAG node. |
| 861 | for (SUnit &I : SUnits) { |
| 862 | RemoveDeps.clear(); |
| 863 | // Set to true if the instruction has an operand defined by a Phi. |
| 864 | unsigned HasPhiUse = 0; |
| 865 | unsigned HasPhiDef = 0; |
| 866 | MachineInstr *MI = I.getInstr(); |
| 867 | // Iterate over each operand, and we process the definitions. |
| 868 | for (const MachineOperand &MO : MI->operands()) { |
| 869 | if (!MO.isReg()) |
| 870 | continue; |
| 871 | Register Reg = MO.getReg(); |
| 872 | if (MO.isDef()) { |
| 873 | // If the register is used by a Phi, then create an anti dependence. |
| 874 | for (MachineRegisterInfo::use_instr_iterator |
| 875 | UI = MRI.use_instr_begin(Reg), |
| 876 | UE = MRI.use_instr_end(); |
| 877 | UI != UE; ++UI) { |
| 878 | MachineInstr *UseMI = &*UI; |
| 879 | SUnit *SU = getSUnit(UseMI); |
| 880 | if (SU != nullptr && UseMI->isPHI()) { |
| 881 | if (!MI->isPHI()) { |
| 882 | SDep Dep(SU, SDep::Anti, Reg); |
| 883 | Dep.setLatency(1); |
| 884 | I.addPred(Dep); |
| 885 | } else { |
| 886 | HasPhiDef = Reg; |
| 887 | // Add a chain edge to a dependent Phi that isn't an existing |
| 888 | // predecessor. |
| 889 | if (SU->NodeNum < I.NodeNum && !I.isPred(SU)) |
| 890 | I.addPred(SDep(SU, SDep::Barrier)); |
| 891 | } |
| 892 | } |
| 893 | } |
| 894 | } else if (MO.isUse()) { |
| 895 | // If the register is defined by a Phi, then create a true dependence. |
| 896 | MachineInstr *DefMI = MRI.getUniqueVRegDef(Reg); |
| 897 | if (DefMI == nullptr) |
| 898 | continue; |
| 899 | SUnit *SU = getSUnit(DefMI); |
| 900 | if (SU != nullptr && DefMI->isPHI()) { |
| 901 | if (!MI->isPHI()) { |
| 902 | SDep Dep(SU, SDep::Data, Reg); |
| 903 | Dep.setLatency(0); |
| 904 | ST.adjustSchedDependency(SU, 0, &I, MO.getOperandNo(), Dep); |
| 905 | I.addPred(Dep); |
| 906 | } else { |
| 907 | HasPhiUse = Reg; |
| 908 | // Add a chain edge to a dependent Phi that isn't an existing |
| 909 | // predecessor. |
| 910 | if (SU->NodeNum < I.NodeNum && !I.isPred(SU)) |
| 911 | I.addPred(SDep(SU, SDep::Barrier)); |
| 912 | } |
| 913 | } |
| 914 | } |
| 915 | } |
| 916 | // Remove order dependences from an unrelated Phi. |
| 917 | if (!SwpPruneDeps) |
| 918 | continue; |
| 919 | for (auto &PI : I.Preds) { |
| 920 | MachineInstr *PMI = PI.getSUnit()->getInstr(); |
| 921 | if (PMI->isPHI() && PI.getKind() == SDep::Order) { |
| 922 | if (I.getInstr()->isPHI()) { |
| 923 | if (PMI->getOperand(0).getReg() == HasPhiUse) |
| 924 | continue; |
| 925 | if (getLoopPhiReg(*PMI, PMI->getParent()) == HasPhiDef) |
| 926 | continue; |
| 927 | } |
| 928 | RemoveDeps.push_back(PI); |
| 929 | } |
| 930 | } |
| 931 | for (int i = 0, e = RemoveDeps.size(); i != e; ++i) |
| 932 | I.removePred(RemoveDeps[i]); |
| 933 | } |
| 934 | } |
| 935 | |
| 936 | /// Iterate over each DAG node and see if we can change any dependences |
| 937 | /// in order to reduce the recurrence MII. |
| 938 | void SwingSchedulerDAG::changeDependences() { |
| 939 | // See if an instruction can use a value from the previous iteration. |
| 940 | // If so, we update the base and offset of the instruction and change |
| 941 | // the dependences. |
| 942 | for (SUnit &I : SUnits) { |
| 943 | unsigned BasePos = 0, OffsetPos = 0, NewBase = 0; |
| 944 | int64_t NewOffset = 0; |
| 945 | if (!canUseLastOffsetValue(I.getInstr(), BasePos, OffsetPos, NewBase, |
| 946 | NewOffset)) |
| 947 | continue; |
| 948 | |
| 949 | // Get the MI and SUnit for the instruction that defines the original base. |
| 950 | Register OrigBase = I.getInstr()->getOperand(BasePos).getReg(); |
| 951 | MachineInstr *DefMI = MRI.getUniqueVRegDef(OrigBase); |
| 952 | if (!DefMI) |
| 953 | continue; |
| 954 | SUnit *DefSU = getSUnit(DefMI); |
| 955 | if (!DefSU) |
| 956 | continue; |
| 957 | // Get the MI and SUnit for the instruction that defins the new base. |
| 958 | MachineInstr *LastMI = MRI.getUniqueVRegDef(NewBase); |
| 959 | if (!LastMI) |
| 960 | continue; |
| 961 | SUnit *LastSU = getSUnit(LastMI); |
| 962 | if (!LastSU) |
| 963 | continue; |
| 964 | |
| 965 | if (Topo.IsReachable(&I, LastSU)) |
| 966 | continue; |
| 967 | |
| 968 | // Remove the dependence. The value now depends on a prior iteration. |
| 969 | SmallVector<SDep, 4> Deps; |
| 970 | for (const SDep &P : I.Preds) |
| 971 | if (P.getSUnit() == DefSU) |
| 972 | Deps.push_back(P); |
| 973 | for (int i = 0, e = Deps.size(); i != e; i++) { |
| 974 | Topo.RemovePred(&I, Deps[i].getSUnit()); |
| 975 | I.removePred(Deps[i]); |
| 976 | } |
| 977 | // Remove the chain dependence between the instructions. |
| 978 | Deps.clear(); |
| 979 | for (auto &P : LastSU->Preds) |
| 980 | if (P.getSUnit() == &I && P.getKind() == SDep::Order) |
| 981 | Deps.push_back(P); |
| 982 | for (int i = 0, e = Deps.size(); i != e; i++) { |
| 983 | Topo.RemovePred(LastSU, Deps[i].getSUnit()); |
| 984 | LastSU->removePred(Deps[i]); |
| 985 | } |
| 986 | |
| 987 | // Add a dependence between the new instruction and the instruction |
| 988 | // that defines the new base. |
| 989 | SDep Dep(&I, SDep::Anti, NewBase); |
| 990 | Topo.AddPred(LastSU, &I); |
| 991 | LastSU->addPred(Dep); |
| 992 | |
| 993 | // Remember the base and offset information so that we can update the |
| 994 | // instruction during code generation. |
| 995 | InstrChanges[&I] = std::make_pair(NewBase, NewOffset); |
| 996 | } |
| 997 | } |
| 998 | |
| 999 | namespace { |
| 1000 | |
| 1001 | // FuncUnitSorter - Comparison operator used to sort instructions by |
| 1002 | // the number of functional unit choices. |
| 1003 | struct FuncUnitSorter { |
| 1004 | const InstrItineraryData *InstrItins; |
| 1005 | const MCSubtargetInfo *STI; |
| 1006 | DenseMap<InstrStage::FuncUnits, unsigned> Resources; |
| 1007 | |
| 1008 | FuncUnitSorter(const TargetSubtargetInfo &TSI) |
| 1009 | : InstrItins(TSI.getInstrItineraryData()), STI(&TSI) {} |
| 1010 | |
| 1011 | // Compute the number of functional unit alternatives needed |
| 1012 | // at each stage, and take the minimum value. We prioritize the |
| 1013 | // instructions by the least number of choices first. |
| 1014 | unsigned minFuncUnits(const MachineInstr *Inst, |
| 1015 | InstrStage::FuncUnits &F) const { |
| 1016 | unsigned SchedClass = Inst->getDesc().getSchedClass(); |
| 1017 | unsigned min = UINT_MAX(2147483647 *2U +1U); |
| 1018 | if (InstrItins && !InstrItins->isEmpty()) { |
| 1019 | for (const InstrStage &IS : |
| 1020 | make_range(InstrItins->beginStage(SchedClass), |
| 1021 | InstrItins->endStage(SchedClass))) { |
| 1022 | InstrStage::FuncUnits funcUnits = IS.getUnits(); |
| 1023 | unsigned numAlternatives = llvm::popcount(funcUnits); |
| 1024 | if (numAlternatives < min) { |
| 1025 | min = numAlternatives; |
| 1026 | F = funcUnits; |
| 1027 | } |
| 1028 | } |
| 1029 | return min; |
| 1030 | } |
| 1031 | if (STI && STI->getSchedModel().hasInstrSchedModel()) { |
| 1032 | const MCSchedClassDesc *SCDesc = |
| 1033 | STI->getSchedModel().getSchedClassDesc(SchedClass); |
| 1034 | if (!SCDesc->isValid()) |
| 1035 | // No valid Schedule Class Desc for schedClass, should be |
| 1036 | // Pseudo/PostRAPseudo |
| 1037 | return min; |
| 1038 | |
| 1039 | for (const MCWriteProcResEntry &PRE : |
| 1040 | make_range(STI->getWriteProcResBegin(SCDesc), |
| 1041 | STI->getWriteProcResEnd(SCDesc))) { |
| 1042 | if (!PRE.Cycles) |
| 1043 | continue; |
| 1044 | const MCProcResourceDesc *ProcResource = |
| 1045 | STI->getSchedModel().getProcResource(PRE.ProcResourceIdx); |
| 1046 | unsigned NumUnits = ProcResource->NumUnits; |
| 1047 | if (NumUnits < min) { |
| 1048 | min = NumUnits; |
| 1049 | F = PRE.ProcResourceIdx; |
| 1050 | } |
| 1051 | } |
| 1052 | return min; |
| 1053 | } |
| 1054 | llvm_unreachable("Should have non-empty InstrItins or hasInstrSchedModel!")::llvm::llvm_unreachable_internal("Should have non-empty InstrItins or hasInstrSchedModel!" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 1054); |
| 1055 | } |
| 1056 | |
| 1057 | // Compute the critical resources needed by the instruction. This |
| 1058 | // function records the functional units needed by instructions that |
| 1059 | // must use only one functional unit. We use this as a tie breaker |
| 1060 | // for computing the resource MII. The instrutions that require |
| 1061 | // the same, highly used, functional unit have high priority. |
| 1062 | void calcCriticalResources(MachineInstr &MI) { |
| 1063 | unsigned SchedClass = MI.getDesc().getSchedClass(); |
| 1064 | if (InstrItins && !InstrItins->isEmpty()) { |
| 1065 | for (const InstrStage &IS : |
| 1066 | make_range(InstrItins->beginStage(SchedClass), |
| 1067 | InstrItins->endStage(SchedClass))) { |
| 1068 | InstrStage::FuncUnits FuncUnits = IS.getUnits(); |
| 1069 | if (llvm::popcount(FuncUnits) == 1) |
| 1070 | Resources[FuncUnits]++; |
| 1071 | } |
| 1072 | return; |
| 1073 | } |
| 1074 | if (STI && STI->getSchedModel().hasInstrSchedModel()) { |
| 1075 | const MCSchedClassDesc *SCDesc = |
| 1076 | STI->getSchedModel().getSchedClassDesc(SchedClass); |
| 1077 | if (!SCDesc->isValid()) |
| 1078 | // No valid Schedule Class Desc for schedClass, should be |
| 1079 | // Pseudo/PostRAPseudo |
| 1080 | return; |
| 1081 | |
| 1082 | for (const MCWriteProcResEntry &PRE : |
| 1083 | make_range(STI->getWriteProcResBegin(SCDesc), |
| 1084 | STI->getWriteProcResEnd(SCDesc))) { |
| 1085 | if (!PRE.Cycles) |
| 1086 | continue; |
| 1087 | Resources[PRE.ProcResourceIdx]++; |
| 1088 | } |
| 1089 | return; |
| 1090 | } |
| 1091 | llvm_unreachable("Should have non-empty InstrItins or hasInstrSchedModel!")::llvm::llvm_unreachable_internal("Should have non-empty InstrItins or hasInstrSchedModel!" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 1091); |
| 1092 | } |
| 1093 | |
| 1094 | /// Return true if IS1 has less priority than IS2. |
| 1095 | bool operator()(const MachineInstr *IS1, const MachineInstr *IS2) const { |
| 1096 | InstrStage::FuncUnits F1 = 0, F2 = 0; |
| 1097 | unsigned MFUs1 = minFuncUnits(IS1, F1); |
| 1098 | unsigned MFUs2 = minFuncUnits(IS2, F2); |
| 1099 | if (MFUs1 == MFUs2) |
| 1100 | return Resources.lookup(F1) < Resources.lookup(F2); |
| 1101 | return MFUs1 > MFUs2; |
| 1102 | } |
| 1103 | }; |
| 1104 | |
| 1105 | } // end anonymous namespace |
| 1106 | |
| 1107 | /// Calculate the resource constrained minimum initiation interval for the |
| 1108 | /// specified loop. We use the DFA to model the resources needed for |
| 1109 | /// each instruction, and we ignore dependences. A different DFA is created |
| 1110 | /// for each cycle that is required. When adding a new instruction, we attempt |
| 1111 | /// to add it to each existing DFA, until a legal space is found. If the |
| 1112 | /// instruction cannot be reserved in an existing DFA, we create a new one. |
| 1113 | unsigned SwingSchedulerDAG::calculateResMII() { |
| 1114 | LLVM_DEBUG(dbgs() << "calculateResMII:\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "calculateResMII:\n"; } } while (false); |
| 1115 | ResourceManager RM(&MF.getSubtarget(), this); |
| 1116 | return RM.calculateResMII(); |
| 1117 | } |
| 1118 | |
| 1119 | /// Calculate the recurrence-constrainted minimum initiation interval. |
| 1120 | /// Iterate over each circuit. Compute the delay(c) and distance(c) |
| 1121 | /// for each circuit. The II needs to satisfy the inequality |
| 1122 | /// delay(c) - II*distance(c) <= 0. For each circuit, choose the smallest |
| 1123 | /// II that satisfies the inequality, and the RecMII is the maximum |
| 1124 | /// of those values. |
| 1125 | unsigned SwingSchedulerDAG::calculateRecMII(NodeSetType &NodeSets) { |
| 1126 | unsigned RecMII = 0; |
| 1127 | |
| 1128 | for (NodeSet &Nodes : NodeSets) { |
| 1129 | if (Nodes.empty()) |
| 1130 | continue; |
| 1131 | |
| 1132 | unsigned Delay = Nodes.getLatency(); |
| 1133 | unsigned Distance = 1; |
| 1134 | |
| 1135 | // ii = ceil(delay / distance) |
| 1136 | unsigned CurMII = (Delay + Distance - 1) / Distance; |
| 1137 | Nodes.setRecMII(CurMII); |
| 1138 | if (CurMII > RecMII) |
| 1139 | RecMII = CurMII; |
| 1140 | } |
| 1141 | |
| 1142 | return RecMII; |
| 1143 | } |
| 1144 | |
| 1145 | /// Swap all the anti dependences in the DAG. That means it is no longer a DAG, |
| 1146 | /// but we do this to find the circuits, and then change them back. |
| 1147 | static void swapAntiDependences(std::vector<SUnit> &SUnits) { |
| 1148 | SmallVector<std::pair<SUnit *, SDep>, 8> DepsAdded; |
| 1149 | for (SUnit &SU : SUnits) { |
| 1150 | for (SDep &Pred : SU.Preds) |
| 1151 | if (Pred.getKind() == SDep::Anti) |
| 1152 | DepsAdded.push_back(std::make_pair(&SU, Pred)); |
| 1153 | } |
| 1154 | for (std::pair<SUnit *, SDep> &P : DepsAdded) { |
| 1155 | // Remove this anti dependency and add one in the reverse direction. |
| 1156 | SUnit *SU = P.first; |
| 1157 | SDep &D = P.second; |
| 1158 | SUnit *TargetSU = D.getSUnit(); |
| 1159 | unsigned Reg = D.getReg(); |
| 1160 | unsigned Lat = D.getLatency(); |
| 1161 | SU->removePred(D); |
| 1162 | SDep Dep(SU, SDep::Anti, Reg); |
| 1163 | Dep.setLatency(Lat); |
| 1164 | TargetSU->addPred(Dep); |
| 1165 | } |
| 1166 | } |
| 1167 | |
| 1168 | /// Create the adjacency structure of the nodes in the graph. |
| 1169 | void SwingSchedulerDAG::Circuits::createAdjacencyStructure( |
| 1170 | SwingSchedulerDAG *DAG) { |
| 1171 | BitVector Added(SUnits.size()); |
| 1172 | DenseMap<int, int> OutputDeps; |
| 1173 | for (int i = 0, e = SUnits.size(); i != e; ++i) { |
| 1174 | Added.reset(); |
| 1175 | // Add any successor to the adjacency matrix and exclude duplicates. |
| 1176 | for (auto &SI : SUnits[i].Succs) { |
| 1177 | // Only create a back-edge on the first and last nodes of a dependence |
| 1178 | // chain. This records any chains and adds them later. |
| 1179 | if (SI.getKind() == SDep::Output) { |
| 1180 | int N = SI.getSUnit()->NodeNum; |
| 1181 | int BackEdge = i; |
| 1182 | auto Dep = OutputDeps.find(BackEdge); |
| 1183 | if (Dep != OutputDeps.end()) { |
| 1184 | BackEdge = Dep->second; |
| 1185 | OutputDeps.erase(Dep); |
| 1186 | } |
| 1187 | OutputDeps[N] = BackEdge; |
| 1188 | } |
| 1189 | // Do not process a boundary node, an artificial node. |
| 1190 | // A back-edge is processed only if it goes to a Phi. |
| 1191 | if (SI.getSUnit()->isBoundaryNode() || SI.isArtificial() || |
| 1192 | (SI.getKind() == SDep::Anti && !SI.getSUnit()->getInstr()->isPHI())) |
| 1193 | continue; |
| 1194 | int N = SI.getSUnit()->NodeNum; |
| 1195 | if (!Added.test(N)) { |
| 1196 | AdjK[i].push_back(N); |
| 1197 | Added.set(N); |
| 1198 | } |
| 1199 | } |
| 1200 | // A chain edge between a store and a load is treated as a back-edge in the |
| 1201 | // adjacency matrix. |
| 1202 | for (auto &PI : SUnits[i].Preds) { |
| 1203 | if (!SUnits[i].getInstr()->mayStore() || |
| 1204 | !DAG->isLoopCarriedDep(&SUnits[i], PI, false)) |
| 1205 | continue; |
| 1206 | if (PI.getKind() == SDep::Order && PI.getSUnit()->getInstr()->mayLoad()) { |
| 1207 | int N = PI.getSUnit()->NodeNum; |
| 1208 | if (!Added.test(N)) { |
| 1209 | AdjK[i].push_back(N); |
| 1210 | Added.set(N); |
| 1211 | } |
| 1212 | } |
| 1213 | } |
| 1214 | } |
| 1215 | // Add back-edges in the adjacency matrix for the output dependences. |
| 1216 | for (auto &OD : OutputDeps) |
| 1217 | if (!Added.test(OD.second)) { |
| 1218 | AdjK[OD.first].push_back(OD.second); |
| 1219 | Added.set(OD.second); |
| 1220 | } |
| 1221 | } |
| 1222 | |
| 1223 | /// Identify an elementary circuit in the dependence graph starting at the |
| 1224 | /// specified node. |
| 1225 | bool SwingSchedulerDAG::Circuits::circuit(int V, int S, NodeSetType &NodeSets, |
| 1226 | bool HasBackedge) { |
| 1227 | SUnit *SV = &SUnits[V]; |
| 1228 | bool F = false; |
| 1229 | Stack.insert(SV); |
| 1230 | Blocked.set(V); |
| 1231 | |
| 1232 | for (auto W : AdjK[V]) { |
| 1233 | if (NumPaths > MaxPaths) |
| 1234 | break; |
| 1235 | if (W < S) |
| 1236 | continue; |
| 1237 | if (W == S) { |
| 1238 | if (!HasBackedge) |
| 1239 | NodeSets.push_back(NodeSet(Stack.begin(), Stack.end())); |
| 1240 | F = true; |
| 1241 | ++NumPaths; |
| 1242 | break; |
| 1243 | } else if (!Blocked.test(W)) { |
| 1244 | if (circuit(W, S, NodeSets, |
| 1245 | Node2Idx->at(W) < Node2Idx->at(V) ? true : HasBackedge)) |
| 1246 | F = true; |
| 1247 | } |
| 1248 | } |
| 1249 | |
| 1250 | if (F) |
| 1251 | unblock(V); |
| 1252 | else { |
| 1253 | for (auto W : AdjK[V]) { |
| 1254 | if (W < S) |
| 1255 | continue; |
| 1256 | B[W].insert(SV); |
| 1257 | } |
| 1258 | } |
| 1259 | Stack.pop_back(); |
| 1260 | return F; |
| 1261 | } |
| 1262 | |
| 1263 | /// Unblock a node in the circuit finding algorithm. |
| 1264 | void SwingSchedulerDAG::Circuits::unblock(int U) { |
| 1265 | Blocked.reset(U); |
| 1266 | SmallPtrSet<SUnit *, 4> &BU = B[U]; |
| 1267 | while (!BU.empty()) { |
| 1268 | SmallPtrSet<SUnit *, 4>::iterator SI = BU.begin(); |
| 1269 | assert(SI != BU.end() && "Invalid B set.")(static_cast <bool> (SI != BU.end() && "Invalid B set." ) ? void (0) : __assert_fail ("SI != BU.end() && \"Invalid B set.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 1269, __extension__ __PRETTY_FUNCTION__)); |
| 1270 | SUnit *W = *SI; |
| 1271 | BU.erase(W); |
| 1272 | if (Blocked.test(W->NodeNum)) |
| 1273 | unblock(W->NodeNum); |
| 1274 | } |
| 1275 | } |
| 1276 | |
| 1277 | /// Identify all the elementary circuits in the dependence graph using |
| 1278 | /// Johnson's circuit algorithm. |
| 1279 | void SwingSchedulerDAG::findCircuits(NodeSetType &NodeSets) { |
| 1280 | // Swap all the anti dependences in the DAG. That means it is no longer a DAG, |
| 1281 | // but we do this to find the circuits, and then change them back. |
| 1282 | swapAntiDependences(SUnits); |
| 1283 | |
| 1284 | Circuits Cir(SUnits, Topo); |
| 1285 | // Create the adjacency structure. |
| 1286 | Cir.createAdjacencyStructure(this); |
| 1287 | for (int i = 0, e = SUnits.size(); i != e; ++i) { |
| 1288 | Cir.reset(); |
| 1289 | Cir.circuit(i, i, NodeSets); |
| 1290 | } |
| 1291 | |
| 1292 | // Change the dependences back so that we've created a DAG again. |
| 1293 | swapAntiDependences(SUnits); |
| 1294 | } |
| 1295 | |
| 1296 | // Create artificial dependencies between the source of COPY/REG_SEQUENCE that |
| 1297 | // is loop-carried to the USE in next iteration. This will help pipeliner avoid |
| 1298 | // additional copies that are needed across iterations. An artificial dependence |
| 1299 | // edge is added from USE to SOURCE of COPY/REG_SEQUENCE. |
| 1300 | |
| 1301 | // PHI-------Anti-Dep-----> COPY/REG_SEQUENCE (loop-carried) |
| 1302 | // SRCOfCopY------True-Dep---> COPY/REG_SEQUENCE |
| 1303 | // PHI-------True-Dep------> USEOfPhi |
| 1304 | |
| 1305 | // The mutation creates |
| 1306 | // USEOfPHI -------Artificial-Dep---> SRCOfCopy |
| 1307 | |
| 1308 | // This overall will ensure, the USEOfPHI is scheduled before SRCOfCopy |
| 1309 | // (since USE is a predecessor), implies, the COPY/ REG_SEQUENCE is scheduled |
| 1310 | // late to avoid additional copies across iterations. The possible scheduling |
| 1311 | // order would be |
| 1312 | // USEOfPHI --- SRCOfCopy--- COPY/REG_SEQUENCE. |
| 1313 | |
| 1314 | void SwingSchedulerDAG::CopyToPhiMutation::apply(ScheduleDAGInstrs *DAG) { |
| 1315 | for (SUnit &SU : DAG->SUnits) { |
| 1316 | // Find the COPY/REG_SEQUENCE instruction. |
| 1317 | if (!SU.getInstr()->isCopy() && !SU.getInstr()->isRegSequence()) |
| 1318 | continue; |
| 1319 | |
| 1320 | // Record the loop carried PHIs. |
| 1321 | SmallVector<SUnit *, 4> PHISUs; |
| 1322 | // Record the SrcSUs that feed the COPY/REG_SEQUENCE instructions. |
| 1323 | SmallVector<SUnit *, 4> SrcSUs; |
| 1324 | |
| 1325 | for (auto &Dep : SU.Preds) { |
| 1326 | SUnit *TmpSU = Dep.getSUnit(); |
| 1327 | MachineInstr *TmpMI = TmpSU->getInstr(); |
| 1328 | SDep::Kind DepKind = Dep.getKind(); |
| 1329 | // Save the loop carried PHI. |
| 1330 | if (DepKind == SDep::Anti && TmpMI->isPHI()) |
| 1331 | PHISUs.push_back(TmpSU); |
| 1332 | // Save the source of COPY/REG_SEQUENCE. |
| 1333 | // If the source has no pre-decessors, we will end up creating cycles. |
| 1334 | else if (DepKind == SDep::Data && !TmpMI->isPHI() && TmpSU->NumPreds > 0) |
| 1335 | SrcSUs.push_back(TmpSU); |
| 1336 | } |
| 1337 | |
| 1338 | if (PHISUs.size() == 0 || SrcSUs.size() == 0) |
| 1339 | continue; |
| 1340 | |
| 1341 | // Find the USEs of PHI. If the use is a PHI or REG_SEQUENCE, push back this |
| 1342 | // SUnit to the container. |
| 1343 | SmallVector<SUnit *, 8> UseSUs; |
| 1344 | // Do not use iterator based loop here as we are updating the container. |
| 1345 | for (size_t Index = 0; Index < PHISUs.size(); ++Index) { |
| 1346 | for (auto &Dep : PHISUs[Index]->Succs) { |
| 1347 | if (Dep.getKind() != SDep::Data) |
| 1348 | continue; |
| 1349 | |
| 1350 | SUnit *TmpSU = Dep.getSUnit(); |
| 1351 | MachineInstr *TmpMI = TmpSU->getInstr(); |
| 1352 | if (TmpMI->isPHI() || TmpMI->isRegSequence()) { |
| 1353 | PHISUs.push_back(TmpSU); |
| 1354 | continue; |
| 1355 | } |
| 1356 | UseSUs.push_back(TmpSU); |
| 1357 | } |
| 1358 | } |
| 1359 | |
| 1360 | if (UseSUs.size() == 0) |
| 1361 | continue; |
| 1362 | |
| 1363 | SwingSchedulerDAG *SDAG = cast<SwingSchedulerDAG>(DAG); |
| 1364 | // Add the artificial dependencies if it does not form a cycle. |
| 1365 | for (auto *I : UseSUs) { |
| 1366 | for (auto *Src : SrcSUs) { |
| 1367 | if (!SDAG->Topo.IsReachable(I, Src) && Src != I) { |
| 1368 | Src->addPred(SDep(I, SDep::Artificial)); |
| 1369 | SDAG->Topo.AddPred(Src, I); |
| 1370 | } |
| 1371 | } |
| 1372 | } |
| 1373 | } |
| 1374 | } |
| 1375 | |
| 1376 | /// Return true for DAG nodes that we ignore when computing the cost functions. |
| 1377 | /// We ignore the back-edge recurrence in order to avoid unbounded recursion |
| 1378 | /// in the calculation of the ASAP, ALAP, etc functions. |
| 1379 | static bool ignoreDependence(const SDep &D, bool isPred) { |
| 1380 | if (D.isArtificial() || D.getSUnit()->isBoundaryNode()) |
| 1381 | return true; |
| 1382 | return D.getKind() == SDep::Anti && isPred; |
| 1383 | } |
| 1384 | |
| 1385 | /// Compute several functions need to order the nodes for scheduling. |
| 1386 | /// ASAP - Earliest time to schedule a node. |
| 1387 | /// ALAP - Latest time to schedule a node. |
| 1388 | /// MOV - Mobility function, difference between ALAP and ASAP. |
| 1389 | /// D - Depth of each node. |
| 1390 | /// H - Height of each node. |
| 1391 | void SwingSchedulerDAG::computeNodeFunctions(NodeSetType &NodeSets) { |
| 1392 | ScheduleInfo.resize(SUnits.size()); |
| 1393 | |
| 1394 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false) |
| 1395 | for (int I : Topo) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false) |
| 1396 | const SUnit &SU = SUnits[I];do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false) |
| 1397 | dumpNode(SU);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false) |
| 1398 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false) |
| 1399 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (int I : Topo) { const SUnit &SU = SUnits[I]; dumpNode(SU); } }; } } while (false); |
| 1400 | |
| 1401 | int maxASAP = 0; |
| 1402 | // Compute ASAP and ZeroLatencyDepth. |
| 1403 | for (int I : Topo) { |
| 1404 | int asap = 0; |
| 1405 | int zeroLatencyDepth = 0; |
| 1406 | SUnit *SU = &SUnits[I]; |
| 1407 | for (const SDep &P : SU->Preds) { |
| 1408 | SUnit *pred = P.getSUnit(); |
| 1409 | if (P.getLatency() == 0) |
| 1410 | zeroLatencyDepth = |
| 1411 | std::max(zeroLatencyDepth, getZeroLatencyDepth(pred) + 1); |
| 1412 | if (ignoreDependence(P, true)) |
| 1413 | continue; |
| 1414 | asap = std::max(asap, (int)(getASAP(pred) + P.getLatency() - |
| 1415 | getDistance(pred, SU, P) * MII)); |
| 1416 | } |
| 1417 | maxASAP = std::max(maxASAP, asap); |
| 1418 | ScheduleInfo[I].ASAP = asap; |
| 1419 | ScheduleInfo[I].ZeroLatencyDepth = zeroLatencyDepth; |
| 1420 | } |
| 1421 | |
| 1422 | // Compute ALAP, ZeroLatencyHeight, and MOV. |
| 1423 | for (int I : llvm::reverse(Topo)) { |
| 1424 | int alap = maxASAP; |
| 1425 | int zeroLatencyHeight = 0; |
| 1426 | SUnit *SU = &SUnits[I]; |
| 1427 | for (const SDep &S : SU->Succs) { |
| 1428 | SUnit *succ = S.getSUnit(); |
| 1429 | if (succ->isBoundaryNode()) |
| 1430 | continue; |
| 1431 | if (S.getLatency() == 0) |
| 1432 | zeroLatencyHeight = |
| 1433 | std::max(zeroLatencyHeight, getZeroLatencyHeight(succ) + 1); |
| 1434 | if (ignoreDependence(S, true)) |
| 1435 | continue; |
| 1436 | alap = std::min(alap, (int)(getALAP(succ) - S.getLatency() + |
| 1437 | getDistance(SU, succ, S) * MII)); |
| 1438 | } |
| 1439 | |
| 1440 | ScheduleInfo[I].ALAP = alap; |
| 1441 | ScheduleInfo[I].ZeroLatencyHeight = zeroLatencyHeight; |
| 1442 | } |
| 1443 | |
| 1444 | // After computing the node functions, compute the summary for each node set. |
| 1445 | for (NodeSet &I : NodeSets) |
| 1446 | I.computeNodeSetInfo(this); |
| 1447 | |
| 1448 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1449 | for (unsigned i = 0; i < SUnits.size(); i++) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1450 | dbgs() << "\tNode " << i << ":\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1451 | dbgs() << "\t ASAP = " << getASAP(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1452 | dbgs() << "\t ALAP = " << getALAP(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1453 | dbgs() << "\t MOV = " << getMOV(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1454 | dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1455 | dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1456 | dbgs() << "\t ZLD = " << getZeroLatencyDepth(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1457 | dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1458 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false) |
| 1459 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { for (unsigned i = 0; i < SUnits.size(); i++) { dbgs() << "\tNode " << i << ":\n"; dbgs () << "\t ASAP = " << getASAP(&SUnits[i]) << "\n"; dbgs() << "\t ALAP = " << getALAP(&SUnits [i]) << "\n"; dbgs() << "\t MOV = " << getMOV (&SUnits[i]) << "\n"; dbgs() << "\t D = " << getDepth(&SUnits[i]) << "\n"; dbgs() << "\t H = " << getHeight(&SUnits[i]) << "\n" ; dbgs() << "\t ZLD = " << getZeroLatencyDepth (&SUnits[i]) << "\n"; dbgs() << "\t ZLH = " << getZeroLatencyHeight(&SUnits[i]) << "\n"; } }; } } while (false); |
| 1460 | } |
| 1461 | |
| 1462 | /// Compute the Pred_L(O) set, as defined in the paper. The set is defined |
| 1463 | /// as the predecessors of the elements of NodeOrder that are not also in |
| 1464 | /// NodeOrder. |
| 1465 | static bool pred_L(SetVector<SUnit *> &NodeOrder, |
| 1466 | SmallSetVector<SUnit *, 8> &Preds, |
| 1467 | const NodeSet *S = nullptr) { |
| 1468 | Preds.clear(); |
| 1469 | for (const SUnit *SU : NodeOrder) { |
| 1470 | for (const SDep &Pred : SU->Preds) { |
| 1471 | if (S && S->count(Pred.getSUnit()) == 0) |
| 1472 | continue; |
| 1473 | if (ignoreDependence(Pred, true)) |
| 1474 | continue; |
| 1475 | if (NodeOrder.count(Pred.getSUnit()) == 0) |
| 1476 | Preds.insert(Pred.getSUnit()); |
| 1477 | } |
| 1478 | // Back-edges are predecessors with an anti-dependence. |
| 1479 | for (const SDep &Succ : SU->Succs) { |
| 1480 | if (Succ.getKind() != SDep::Anti) |
| 1481 | continue; |
| 1482 | if (S && S->count(Succ.getSUnit()) == 0) |
| 1483 | continue; |
| 1484 | if (NodeOrder.count(Succ.getSUnit()) == 0) |
| 1485 | Preds.insert(Succ.getSUnit()); |
| 1486 | } |
| 1487 | } |
| 1488 | return !Preds.empty(); |
| 1489 | } |
| 1490 | |
| 1491 | /// Compute the Succ_L(O) set, as defined in the paper. The set is defined |
| 1492 | /// as the successors of the elements of NodeOrder that are not also in |
| 1493 | /// NodeOrder. |
| 1494 | static bool succ_L(SetVector<SUnit *> &NodeOrder, |
| 1495 | SmallSetVector<SUnit *, 8> &Succs, |
| 1496 | const NodeSet *S = nullptr) { |
| 1497 | Succs.clear(); |
| 1498 | for (const SUnit *SU : NodeOrder) { |
| 1499 | for (const SDep &Succ : SU->Succs) { |
| 1500 | if (S && S->count(Succ.getSUnit()) == 0) |
| 1501 | continue; |
| 1502 | if (ignoreDependence(Succ, false)) |
| 1503 | continue; |
| 1504 | if (NodeOrder.count(Succ.getSUnit()) == 0) |
| 1505 | Succs.insert(Succ.getSUnit()); |
| 1506 | } |
| 1507 | for (const SDep &Pred : SU->Preds) { |
| 1508 | if (Pred.getKind() != SDep::Anti) |
| 1509 | continue; |
| 1510 | if (S && S->count(Pred.getSUnit()) == 0) |
| 1511 | continue; |
| 1512 | if (NodeOrder.count(Pred.getSUnit()) == 0) |
| 1513 | Succs.insert(Pred.getSUnit()); |
| 1514 | } |
| 1515 | } |
| 1516 | return !Succs.empty(); |
| 1517 | } |
| 1518 | |
| 1519 | /// Return true if there is a path from the specified node to any of the nodes |
| 1520 | /// in DestNodes. Keep track and return the nodes in any path. |
| 1521 | static bool computePath(SUnit *Cur, SetVector<SUnit *> &Path, |
| 1522 | SetVector<SUnit *> &DestNodes, |
| 1523 | SetVector<SUnit *> &Exclude, |
| 1524 | SmallPtrSet<SUnit *, 8> &Visited) { |
| 1525 | if (Cur->isBoundaryNode()) |
| 1526 | return false; |
| 1527 | if (Exclude.contains(Cur)) |
| 1528 | return false; |
| 1529 | if (DestNodes.contains(Cur)) |
| 1530 | return true; |
| 1531 | if (!Visited.insert(Cur).second) |
| 1532 | return Path.contains(Cur); |
| 1533 | bool FoundPath = false; |
| 1534 | for (auto &SI : Cur->Succs) |
| 1535 | if (!ignoreDependence(SI, false)) |
| 1536 | FoundPath |= |
| 1537 | computePath(SI.getSUnit(), Path, DestNodes, Exclude, Visited); |
| 1538 | for (auto &PI : Cur->Preds) |
| 1539 | if (PI.getKind() == SDep::Anti) |
| 1540 | FoundPath |= |
| 1541 | computePath(PI.getSUnit(), Path, DestNodes, Exclude, Visited); |
| 1542 | if (FoundPath) |
| 1543 | Path.insert(Cur); |
| 1544 | return FoundPath; |
| 1545 | } |
| 1546 | |
| 1547 | /// Compute the live-out registers for the instructions in a node-set. |
| 1548 | /// The live-out registers are those that are defined in the node-set, |
| 1549 | /// but not used. Except for use operands of Phis. |
| 1550 | static void computeLiveOuts(MachineFunction &MF, RegPressureTracker &RPTracker, |
| 1551 | NodeSet &NS) { |
| 1552 | const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); |
| 1553 | MachineRegisterInfo &MRI = MF.getRegInfo(); |
| 1554 | SmallVector<RegisterMaskPair, 8> LiveOutRegs; |
| 1555 | SmallSet<unsigned, 4> Uses; |
| 1556 | for (SUnit *SU : NS) { |
| 1557 | const MachineInstr *MI = SU->getInstr(); |
| 1558 | if (MI->isPHI()) |
| 1559 | continue; |
| 1560 | for (const MachineOperand &MO : MI->operands()) |
| 1561 | if (MO.isReg() && MO.isUse()) { |
| 1562 | Register Reg = MO.getReg(); |
| 1563 | if (Reg.isVirtual()) |
| 1564 | Uses.insert(Reg); |
| 1565 | else if (MRI.isAllocatable(Reg)) |
| 1566 | for (MCRegUnitIterator Units(Reg.asMCReg(), TRI); Units.isValid(); |
| 1567 | ++Units) |
| 1568 | Uses.insert(*Units); |
| 1569 | } |
| 1570 | } |
| 1571 | for (SUnit *SU : NS) |
| 1572 | for (const MachineOperand &MO : SU->getInstr()->operands()) |
| 1573 | if (MO.isReg() && MO.isDef() && !MO.isDead()) { |
| 1574 | Register Reg = MO.getReg(); |
| 1575 | if (Reg.isVirtual()) { |
| 1576 | if (!Uses.count(Reg)) |
| 1577 | LiveOutRegs.push_back(RegisterMaskPair(Reg, |
| 1578 | LaneBitmask::getNone())); |
| 1579 | } else if (MRI.isAllocatable(Reg)) { |
| 1580 | for (MCRegUnitIterator Units(Reg.asMCReg(), TRI); Units.isValid(); |
| 1581 | ++Units) |
| 1582 | if (!Uses.count(*Units)) |
| 1583 | LiveOutRegs.push_back(RegisterMaskPair(*Units, |
| 1584 | LaneBitmask::getNone())); |
| 1585 | } |
| 1586 | } |
| 1587 | RPTracker.addLiveRegs(LiveOutRegs); |
| 1588 | } |
| 1589 | |
| 1590 | /// A heuristic to filter nodes in recurrent node-sets if the register |
| 1591 | /// pressure of a set is too high. |
| 1592 | void SwingSchedulerDAG::registerPressureFilter(NodeSetType &NodeSets) { |
| 1593 | for (auto &NS : NodeSets) { |
| 1594 | // Skip small node-sets since they won't cause register pressure problems. |
| 1595 | if (NS.size() <= 2) |
| 1596 | continue; |
| 1597 | IntervalPressure RecRegPressure; |
| 1598 | RegPressureTracker RecRPTracker(RecRegPressure); |
| 1599 | RecRPTracker.init(&MF, &RegClassInfo, &LIS, BB, BB->end(), false, true); |
| 1600 | computeLiveOuts(MF, RecRPTracker, NS); |
| 1601 | RecRPTracker.closeBottom(); |
| 1602 | |
| 1603 | std::vector<SUnit *> SUnits(NS.begin(), NS.end()); |
| 1604 | llvm::sort(SUnits, [](const SUnit *A, const SUnit *B) { |
| 1605 | return A->NodeNum > B->NodeNum; |
| 1606 | }); |
| 1607 | |
| 1608 | for (auto &SU : SUnits) { |
| 1609 | // Since we're computing the register pressure for a subset of the |
| 1610 | // instructions in a block, we need to set the tracker for each |
| 1611 | // instruction in the node-set. The tracker is set to the instruction |
| 1612 | // just after the one we're interested in. |
| 1613 | MachineBasicBlock::const_iterator CurInstI = SU->getInstr(); |
| 1614 | RecRPTracker.setPos(std::next(CurInstI)); |
| 1615 | |
| 1616 | RegPressureDelta RPDelta; |
| 1617 | ArrayRef<PressureChange> CriticalPSets; |
| 1618 | RecRPTracker.getMaxUpwardPressureDelta(SU->getInstr(), nullptr, RPDelta, |
| 1619 | CriticalPSets, |
| 1620 | RecRegPressure.MaxSetPressure); |
| 1621 | if (RPDelta.Excess.isValid()) { |
| 1622 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Excess register pressure: SU(" << SU->NodeNum << ") " << TRI->getRegPressureSetName (RPDelta.Excess.getPSet()) << ":" << RPDelta.Excess .getUnitInc() << "\n"; } } while (false) |
| 1623 | dbgs() << "Excess register pressure: SU(" << SU->NodeNum << ") "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Excess register pressure: SU(" << SU->NodeNum << ") " << TRI->getRegPressureSetName (RPDelta.Excess.getPSet()) << ":" << RPDelta.Excess .getUnitInc() << "\n"; } } while (false) |
| 1624 | << TRI->getRegPressureSetName(RPDelta.Excess.getPSet())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Excess register pressure: SU(" << SU->NodeNum << ") " << TRI->getRegPressureSetName (RPDelta.Excess.getPSet()) << ":" << RPDelta.Excess .getUnitInc() << "\n"; } } while (false) |
| 1625 | << ":" << RPDelta.Excess.getUnitInc() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Excess register pressure: SU(" << SU->NodeNum << ") " << TRI->getRegPressureSetName (RPDelta.Excess.getPSet()) << ":" << RPDelta.Excess .getUnitInc() << "\n"; } } while (false); |
| 1626 | NS.setExceedPressure(SU); |
| 1627 | break; |
| 1628 | } |
| 1629 | RecRPTracker.recede(); |
| 1630 | } |
| 1631 | } |
| 1632 | } |
| 1633 | |
| 1634 | /// A heuristic to colocate node sets that have the same set of |
| 1635 | /// successors. |
| 1636 | void SwingSchedulerDAG::colocateNodeSets(NodeSetType &NodeSets) { |
| 1637 | unsigned Colocate = 0; |
| 1638 | for (int i = 0, e = NodeSets.size(); i < e; ++i) { |
| 1639 | NodeSet &N1 = NodeSets[i]; |
| 1640 | SmallSetVector<SUnit *, 8> S1; |
| 1641 | if (N1.empty() || !succ_L(N1, S1)) |
| 1642 | continue; |
| 1643 | for (int j = i + 1; j < e; ++j) { |
| 1644 | NodeSet &N2 = NodeSets[j]; |
| 1645 | if (N1.compareRecMII(N2) != 0) |
| 1646 | continue; |
| 1647 | SmallSetVector<SUnit *, 8> S2; |
| 1648 | if (N2.empty() || !succ_L(N2, S2)) |
| 1649 | continue; |
| 1650 | if (llvm::set_is_subset(S1, S2) && S1.size() == S2.size()) { |
| 1651 | N1.setColocate(++Colocate); |
| 1652 | N2.setColocate(Colocate); |
| 1653 | break; |
| 1654 | } |
| 1655 | } |
| 1656 | } |
| 1657 | } |
| 1658 | |
| 1659 | /// Check if the existing node-sets are profitable. If not, then ignore the |
| 1660 | /// recurrent node-sets, and attempt to schedule all nodes together. This is |
| 1661 | /// a heuristic. If the MII is large and all the recurrent node-sets are small, |
| 1662 | /// then it's best to try to schedule all instructions together instead of |
| 1663 | /// starting with the recurrent node-sets. |
| 1664 | void SwingSchedulerDAG::checkNodeSets(NodeSetType &NodeSets) { |
| 1665 | // Look for loops with a large MII. |
| 1666 | if (MII < 17) |
| 1667 | return; |
| 1668 | // Check if the node-set contains only a simple add recurrence. |
| 1669 | for (auto &NS : NodeSets) { |
| 1670 | if (NS.getRecMII() > 2) |
| 1671 | return; |
| 1672 | if (NS.getMaxDepth() > MII) |
| 1673 | return; |
| 1674 | } |
| 1675 | NodeSets.clear(); |
| 1676 | LLVM_DEBUG(dbgs() << "Clear recurrence node-sets\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Clear recurrence node-sets\n" ; } } while (false); |
| 1677 | } |
| 1678 | |
| 1679 | /// Add the nodes that do not belong to a recurrence set into groups |
| 1680 | /// based upon connected components. |
| 1681 | void SwingSchedulerDAG::groupRemainingNodes(NodeSetType &NodeSets) { |
| 1682 | SetVector<SUnit *> NodesAdded; |
| 1683 | SmallPtrSet<SUnit *, 8> Visited; |
| 1684 | // Add the nodes that are on a path between the previous node sets and |
| 1685 | // the current node set. |
| 1686 | for (NodeSet &I : NodeSets) { |
| 1687 | SmallSetVector<SUnit *, 8> N; |
| 1688 | // Add the nodes from the current node set to the previous node set. |
| 1689 | if (succ_L(I, N)) { |
| 1690 | SetVector<SUnit *> Path; |
| 1691 | for (SUnit *NI : N) { |
| 1692 | Visited.clear(); |
| 1693 | computePath(NI, Path, NodesAdded, I, Visited); |
| 1694 | } |
| 1695 | if (!Path.empty()) |
| 1696 | I.insert(Path.begin(), Path.end()); |
| 1697 | } |
| 1698 | // Add the nodes from the previous node set to the current node set. |
| 1699 | N.clear(); |
| 1700 | if (succ_L(NodesAdded, N)) { |
| 1701 | SetVector<SUnit *> Path; |
| 1702 | for (SUnit *NI : N) { |
| 1703 | Visited.clear(); |
| 1704 | computePath(NI, Path, I, NodesAdded, Visited); |
| 1705 | } |
| 1706 | if (!Path.empty()) |
| 1707 | I.insert(Path.begin(), Path.end()); |
| 1708 | } |
| 1709 | NodesAdded.insert(I.begin(), I.end()); |
| 1710 | } |
| 1711 | |
| 1712 | // Create a new node set with the connected nodes of any successor of a node |
| 1713 | // in a recurrent set. |
| 1714 | NodeSet NewSet; |
| 1715 | SmallSetVector<SUnit *, 8> N; |
| 1716 | if (succ_L(NodesAdded, N)) |
| 1717 | for (SUnit *I : N) |
| 1718 | addConnectedNodes(I, NewSet, NodesAdded); |
| 1719 | if (!NewSet.empty()) |
| 1720 | NodeSets.push_back(NewSet); |
| 1721 | |
| 1722 | // Create a new node set with the connected nodes of any predecessor of a node |
| 1723 | // in a recurrent set. |
| 1724 | NewSet.clear(); |
| 1725 | if (pred_L(NodesAdded, N)) |
| 1726 | for (SUnit *I : N) |
| 1727 | addConnectedNodes(I, NewSet, NodesAdded); |
| 1728 | if (!NewSet.empty()) |
| 1729 | NodeSets.push_back(NewSet); |
| 1730 | |
| 1731 | // Create new nodes sets with the connected nodes any remaining node that |
| 1732 | // has no predecessor. |
| 1733 | for (SUnit &SU : SUnits) { |
| 1734 | if (NodesAdded.count(&SU) == 0) { |
| 1735 | NewSet.clear(); |
| 1736 | addConnectedNodes(&SU, NewSet, NodesAdded); |
| 1737 | if (!NewSet.empty()) |
| 1738 | NodeSets.push_back(NewSet); |
| 1739 | } |
| 1740 | } |
| 1741 | } |
| 1742 | |
| 1743 | /// Add the node to the set, and add all of its connected nodes to the set. |
| 1744 | void SwingSchedulerDAG::addConnectedNodes(SUnit *SU, NodeSet &NewSet, |
| 1745 | SetVector<SUnit *> &NodesAdded) { |
| 1746 | NewSet.insert(SU); |
| 1747 | NodesAdded.insert(SU); |
| 1748 | for (auto &SI : SU->Succs) { |
| 1749 | SUnit *Successor = SI.getSUnit(); |
| 1750 | if (!SI.isArtificial() && !Successor->isBoundaryNode() && |
| 1751 | NodesAdded.count(Successor) == 0) |
| 1752 | addConnectedNodes(Successor, NewSet, NodesAdded); |
| 1753 | } |
| 1754 | for (auto &PI : SU->Preds) { |
| 1755 | SUnit *Predecessor = PI.getSUnit(); |
| 1756 | if (!PI.isArtificial() && NodesAdded.count(Predecessor) == 0) |
| 1757 | addConnectedNodes(Predecessor, NewSet, NodesAdded); |
| 1758 | } |
| 1759 | } |
| 1760 | |
| 1761 | /// Return true if Set1 contains elements in Set2. The elements in common |
| 1762 | /// are returned in a different container. |
| 1763 | static bool isIntersect(SmallSetVector<SUnit *, 8> &Set1, const NodeSet &Set2, |
| 1764 | SmallSetVector<SUnit *, 8> &Result) { |
| 1765 | Result.clear(); |
| 1766 | for (SUnit *SU : Set1) { |
| 1767 | if (Set2.count(SU) != 0) |
| 1768 | Result.insert(SU); |
| 1769 | } |
| 1770 | return !Result.empty(); |
| 1771 | } |
| 1772 | |
| 1773 | /// Merge the recurrence node sets that have the same initial node. |
| 1774 | void SwingSchedulerDAG::fuseRecs(NodeSetType &NodeSets) { |
| 1775 | for (NodeSetType::iterator I = NodeSets.begin(), E = NodeSets.end(); I != E; |
| 1776 | ++I) { |
| 1777 | NodeSet &NI = *I; |
| 1778 | for (NodeSetType::iterator J = I + 1; J != E;) { |
| 1779 | NodeSet &NJ = *J; |
| 1780 | if (NI.getNode(0)->NodeNum == NJ.getNode(0)->NodeNum) { |
| 1781 | if (NJ.compareRecMII(NI) > 0) |
| 1782 | NI.setRecMII(NJ.getRecMII()); |
| 1783 | for (SUnit *SU : *J) |
| 1784 | I->insert(SU); |
| 1785 | NodeSets.erase(J); |
| 1786 | E = NodeSets.end(); |
| 1787 | } else { |
| 1788 | ++J; |
| 1789 | } |
| 1790 | } |
| 1791 | } |
| 1792 | } |
| 1793 | |
| 1794 | /// Remove nodes that have been scheduled in previous NodeSets. |
| 1795 | void SwingSchedulerDAG::removeDuplicateNodes(NodeSetType &NodeSets) { |
| 1796 | for (NodeSetType::iterator I = NodeSets.begin(), E = NodeSets.end(); I != E; |
| 1797 | ++I) |
| 1798 | for (NodeSetType::iterator J = I + 1; J != E;) { |
| 1799 | J->remove_if([&](SUnit *SUJ) { return I->count(SUJ); }); |
| 1800 | |
| 1801 | if (J->empty()) { |
| 1802 | NodeSets.erase(J); |
| 1803 | E = NodeSets.end(); |
| 1804 | } else { |
| 1805 | ++J; |
| 1806 | } |
| 1807 | } |
| 1808 | } |
| 1809 | |
| 1810 | /// Compute an ordered list of the dependence graph nodes, which |
| 1811 | /// indicates the order that the nodes will be scheduled. This is a |
| 1812 | /// two-level algorithm. First, a partial order is created, which |
| 1813 | /// consists of a list of sets ordered from highest to lowest priority. |
| 1814 | void SwingSchedulerDAG::computeNodeOrder(NodeSetType &NodeSets) { |
| 1815 | SmallSetVector<SUnit *, 8> R; |
| 1816 | NodeOrder.clear(); |
| 1817 | |
| 1818 | for (auto &Nodes : NodeSets) { |
| 1819 | LLVM_DEBUG(dbgs() << "NodeSet size " << Nodes.size() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "NodeSet size " << Nodes .size() << "\n"; } } while (false); |
| 1820 | OrderKind Order; |
| 1821 | SmallSetVector<SUnit *, 8> N; |
| 1822 | if (pred_L(NodeOrder, N) && llvm::set_is_subset(N, Nodes)) { |
| 1823 | R.insert(N.begin(), N.end()); |
| 1824 | Order = BottomUp; |
| 1825 | LLVM_DEBUG(dbgs() << " Bottom up (preds) ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << " Bottom up (preds) "; } } while (false); |
| 1826 | } else if (succ_L(NodeOrder, N) && llvm::set_is_subset(N, Nodes)) { |
| 1827 | R.insert(N.begin(), N.end()); |
| 1828 | Order = TopDown; |
| 1829 | LLVM_DEBUG(dbgs() << " Top down (succs) ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << " Top down (succs) "; } } while (false); |
| 1830 | } else if (isIntersect(N, Nodes, R)) { |
| 1831 | // If some of the successors are in the existing node-set, then use the |
| 1832 | // top-down ordering. |
| 1833 | Order = TopDown; |
| 1834 | LLVM_DEBUG(dbgs() << " Top down (intersect) ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << " Top down (intersect) "; } } while (false); |
| 1835 | } else if (NodeSets.size() == 1) { |
| 1836 | for (const auto &N : Nodes) |
| 1837 | if (N->Succs.size() == 0) |
| 1838 | R.insert(N); |
| 1839 | Order = BottomUp; |
| 1840 | LLVM_DEBUG(dbgs() << " Bottom up (all) ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << " Bottom up (all) "; } } while (false); |
| 1841 | } else { |
| 1842 | // Find the node with the highest ASAP. |
| 1843 | SUnit *maxASAP = nullptr; |
| 1844 | for (SUnit *SU : Nodes) { |
| 1845 | if (maxASAP == nullptr || getASAP(SU) > getASAP(maxASAP) || |
| 1846 | (getASAP(SU) == getASAP(maxASAP) && SU->NodeNum > maxASAP->NodeNum)) |
| 1847 | maxASAP = SU; |
| 1848 | } |
| 1849 | R.insert(maxASAP); |
| 1850 | Order = BottomUp; |
| 1851 | LLVM_DEBUG(dbgs() << " Bottom up (default) ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << " Bottom up (default) "; } } while (false); |
| 1852 | } |
| 1853 | |
| 1854 | while (!R.empty()) { |
| 1855 | if (Order == TopDown) { |
| 1856 | // Choose the node with the maximum height. If more than one, choose |
| 1857 | // the node wiTH the maximum ZeroLatencyHeight. If still more than one, |
| 1858 | // choose the node with the lowest MOV. |
| 1859 | while (!R.empty()) { |
| 1860 | SUnit *maxHeight = nullptr; |
| 1861 | for (SUnit *I : R) { |
| 1862 | if (maxHeight == nullptr || getHeight(I) > getHeight(maxHeight)) |
| 1863 | maxHeight = I; |
| 1864 | else if (getHeight(I) == getHeight(maxHeight) && |
| 1865 | getZeroLatencyHeight(I) > getZeroLatencyHeight(maxHeight)) |
| 1866 | maxHeight = I; |
| 1867 | else if (getHeight(I) == getHeight(maxHeight) && |
| 1868 | getZeroLatencyHeight(I) == |
| 1869 | getZeroLatencyHeight(maxHeight) && |
| 1870 | getMOV(I) < getMOV(maxHeight)) |
| 1871 | maxHeight = I; |
| 1872 | } |
| 1873 | NodeOrder.insert(maxHeight); |
| 1874 | LLVM_DEBUG(dbgs() << maxHeight->NodeNum << " ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << maxHeight->NodeNum << " "; } } while (false); |
| 1875 | R.remove(maxHeight); |
| 1876 | for (const auto &I : maxHeight->Succs) { |
| 1877 | if (Nodes.count(I.getSUnit()) == 0) |
| 1878 | continue; |
| 1879 | if (NodeOrder.contains(I.getSUnit())) |
| 1880 | continue; |
| 1881 | if (ignoreDependence(I, false)) |
| 1882 | continue; |
| 1883 | R.insert(I.getSUnit()); |
| 1884 | } |
| 1885 | // Back-edges are predecessors with an anti-dependence. |
| 1886 | for (const auto &I : maxHeight->Preds) { |
| 1887 | if (I.getKind() != SDep::Anti) |
| 1888 | continue; |
| 1889 | if (Nodes.count(I.getSUnit()) == 0) |
| 1890 | continue; |
| 1891 | if (NodeOrder.contains(I.getSUnit())) |
| 1892 | continue; |
| 1893 | R.insert(I.getSUnit()); |
| 1894 | } |
| 1895 | } |
| 1896 | Order = BottomUp; |
| 1897 | LLVM_DEBUG(dbgs() << "\n Switching order to bottom up ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "\n Switching order to bottom up " ; } } while (false); |
| 1898 | SmallSetVector<SUnit *, 8> N; |
| 1899 | if (pred_L(NodeOrder, N, &Nodes)) |
| 1900 | R.insert(N.begin(), N.end()); |
| 1901 | } else { |
| 1902 | // Choose the node with the maximum depth. If more than one, choose |
| 1903 | // the node with the maximum ZeroLatencyDepth. If still more than one, |
| 1904 | // choose the node with the lowest MOV. |
| 1905 | while (!R.empty()) { |
| 1906 | SUnit *maxDepth = nullptr; |
| 1907 | for (SUnit *I : R) { |
| 1908 | if (maxDepth == nullptr || getDepth(I) > getDepth(maxDepth)) |
| 1909 | maxDepth = I; |
| 1910 | else if (getDepth(I) == getDepth(maxDepth) && |
| 1911 | getZeroLatencyDepth(I) > getZeroLatencyDepth(maxDepth)) |
| 1912 | maxDepth = I; |
| 1913 | else if (getDepth(I) == getDepth(maxDepth) && |
| 1914 | getZeroLatencyDepth(I) == getZeroLatencyDepth(maxDepth) && |
| 1915 | getMOV(I) < getMOV(maxDepth)) |
| 1916 | maxDepth = I; |
| 1917 | } |
| 1918 | NodeOrder.insert(maxDepth); |
| 1919 | LLVM_DEBUG(dbgs() << maxDepth->NodeNum << " ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << maxDepth->NodeNum << " "; } } while (false); |
| 1920 | R.remove(maxDepth); |
| 1921 | if (Nodes.isExceedSU(maxDepth)) { |
| 1922 | Order = TopDown; |
Value stored to 'Order' is never read | |
| 1923 | R.clear(); |
| 1924 | R.insert(Nodes.getNode(0)); |
| 1925 | break; |
| 1926 | } |
| 1927 | for (const auto &I : maxDepth->Preds) { |
| 1928 | if (Nodes.count(I.getSUnit()) == 0) |
| 1929 | continue; |
| 1930 | if (NodeOrder.contains(I.getSUnit())) |
| 1931 | continue; |
| 1932 | R.insert(I.getSUnit()); |
| 1933 | } |
| 1934 | // Back-edges are predecessors with an anti-dependence. |
| 1935 | for (const auto &I : maxDepth->Succs) { |
| 1936 | if (I.getKind() != SDep::Anti) |
| 1937 | continue; |
| 1938 | if (Nodes.count(I.getSUnit()) == 0) |
| 1939 | continue; |
| 1940 | if (NodeOrder.contains(I.getSUnit())) |
| 1941 | continue; |
| 1942 | R.insert(I.getSUnit()); |
| 1943 | } |
| 1944 | } |
| 1945 | Order = TopDown; |
| 1946 | LLVM_DEBUG(dbgs() << "\n Switching order to top down ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "\n Switching order to top down " ; } } while (false); |
| 1947 | SmallSetVector<SUnit *, 8> N; |
| 1948 | if (succ_L(NodeOrder, N, &Nodes)) |
| 1949 | R.insert(N.begin(), N.end()); |
| 1950 | } |
| 1951 | } |
| 1952 | LLVM_DEBUG(dbgs() << "\nDone with Nodeset\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "\nDone with Nodeset\n"; } } while (false); |
| 1953 | } |
| 1954 | |
| 1955 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false) |
| 1956 | dbgs() << "Node order: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false) |
| 1957 | for (SUnit *I : NodeOrder)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false) |
| 1958 | dbgs() << " " << I->NodeNum << " ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false) |
| 1959 | dbgs() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false) |
| 1960 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Node order: "; for (SUnit *I : NodeOrder) dbgs() << " " << I->NodeNum << " "; dbgs() << "\n"; }; } } while (false); |
| 1961 | } |
| 1962 | |
| 1963 | /// Process the nodes in the computed order and create the pipelined schedule |
| 1964 | /// of the instructions, if possible. Return true if a schedule is found. |
| 1965 | bool SwingSchedulerDAG::schedulePipeline(SMSchedule &Schedule) { |
| 1966 | |
| 1967 | if (NodeOrder.empty()){ |
| 1968 | LLVM_DEBUG(dbgs() << "NodeOrder is empty! abort scheduling\n" )do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "NodeOrder is empty! abort scheduling\n" ; } } while (false); |
| 1969 | return false; |
| 1970 | } |
| 1971 | |
| 1972 | bool scheduleFound = false; |
| 1973 | // Keep increasing II until a valid schedule is found. |
| 1974 | for (unsigned II = MII; II <= MAX_II && !scheduleFound; ++II) { |
| 1975 | Schedule.reset(); |
| 1976 | Schedule.setInitiationInterval(II); |
| 1977 | LLVM_DEBUG(dbgs() << "Try to schedule with " << II << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Try to schedule with " << II << "\n"; } } while (false); |
| 1978 | |
| 1979 | SetVector<SUnit *>::iterator NI = NodeOrder.begin(); |
| 1980 | SetVector<SUnit *>::iterator NE = NodeOrder.end(); |
| 1981 | do { |
| 1982 | SUnit *SU = *NI; |
| 1983 | |
| 1984 | // Compute the schedule time for the instruction, which is based |
| 1985 | // upon the scheduled time for any predecessors/successors. |
| 1986 | int EarlyStart = INT_MIN(-2147483647 -1); |
| 1987 | int LateStart = INT_MAX2147483647; |
| 1988 | // These values are set when the size of the schedule window is limited |
| 1989 | // due to chain dependences. |
| 1990 | int SchedEnd = INT_MAX2147483647; |
| 1991 | int SchedStart = INT_MIN(-2147483647 -1); |
| 1992 | Schedule.computeStart(SU, &EarlyStart, &LateStart, &SchedEnd, &SchedStart, |
| 1993 | II, this); |
| 1994 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false) |
| 1995 | dbgs() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false) |
| 1996 | dbgs() << "Inst (" << SU->NodeNum << ") ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false) |
| 1997 | SU->getInstr()->dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false) |
| 1998 | dbgs() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false) |
| 1999 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\n"; dbgs() << "Inst (" << SU->NodeNum << ") "; SU->getInstr()-> dump(); dbgs() << "\n"; }; } } while (false); |
| 2000 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << format("\tes: %8x ls: %8x me: %8x ms: %8x\n" , EarlyStart, LateStart, SchedEnd, SchedStart); }; } } while ( false) |
| 2001 | dbgs() << format("\tes: %8x ls: %8x me: %8x ms: %8x\n", EarlyStart,do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << format("\tes: %8x ls: %8x me: %8x ms: %8x\n" , EarlyStart, LateStart, SchedEnd, SchedStart); }; } } while ( false) |
| 2002 | LateStart, SchedEnd, SchedStart);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << format("\tes: %8x ls: %8x me: %8x ms: %8x\n" , EarlyStart, LateStart, SchedEnd, SchedStart); }; } } while ( false) |
| 2003 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << format("\tes: %8x ls: %8x me: %8x ms: %8x\n" , EarlyStart, LateStart, SchedEnd, SchedStart); }; } } while ( false); |
| 2004 | |
| 2005 | if (EarlyStart > LateStart || SchedEnd < EarlyStart || |
| 2006 | SchedStart > LateStart) |
| 2007 | scheduleFound = false; |
| 2008 | else if (EarlyStart != INT_MIN(-2147483647 -1) && LateStart == INT_MAX2147483647) { |
| 2009 | SchedEnd = std::min(SchedEnd, EarlyStart + (int)II - 1); |
| 2010 | scheduleFound = Schedule.insert(SU, EarlyStart, SchedEnd, II); |
| 2011 | } else if (EarlyStart == INT_MIN(-2147483647 -1) && LateStart != INT_MAX2147483647) { |
| 2012 | SchedStart = std::max(SchedStart, LateStart - (int)II + 1); |
| 2013 | scheduleFound = Schedule.insert(SU, LateStart, SchedStart, II); |
| 2014 | } else if (EarlyStart != INT_MIN(-2147483647 -1) && LateStart != INT_MAX2147483647) { |
| 2015 | SchedEnd = |
| 2016 | std::min(SchedEnd, std::min(LateStart, EarlyStart + (int)II - 1)); |
| 2017 | // When scheduling a Phi it is better to start at the late cycle and go |
| 2018 | // backwards. The default order may insert the Phi too far away from |
| 2019 | // its first dependence. |
| 2020 | if (SU->getInstr()->isPHI()) |
| 2021 | scheduleFound = Schedule.insert(SU, SchedEnd, EarlyStart, II); |
| 2022 | else |
| 2023 | scheduleFound = Schedule.insert(SU, EarlyStart, SchedEnd, II); |
| 2024 | } else { |
| 2025 | int FirstCycle = Schedule.getFirstCycle(); |
| 2026 | scheduleFound = Schedule.insert(SU, FirstCycle + getASAP(SU), |
| 2027 | FirstCycle + getASAP(SU) + II - 1, II); |
| 2028 | } |
| 2029 | // Even if we find a schedule, make sure the schedule doesn't exceed the |
| 2030 | // allowable number of stages. We keep trying if this happens. |
| 2031 | if (scheduleFound) |
| 2032 | if (SwpMaxStages > -1 && |
| 2033 | Schedule.getMaxStageCount() > (unsigned)SwpMaxStages) |
| 2034 | scheduleFound = false; |
| 2035 | |
| 2036 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!scheduleFound) dbgs() << "\tCan't schedule\n" ; }; } } while (false) |
| 2037 | if (!scheduleFound)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!scheduleFound) dbgs() << "\tCan't schedule\n" ; }; } } while (false) |
| 2038 | dbgs() << "\tCan't schedule\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!scheduleFound) dbgs() << "\tCan't schedule\n" ; }; } } while (false) |
| 2039 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!scheduleFound) dbgs() << "\tCan't schedule\n" ; }; } } while (false); |
| 2040 | } while (++NI != NE && scheduleFound); |
| 2041 | |
| 2042 | // If a schedule is found, ensure non-pipelined instructions are in stage 0 |
| 2043 | if (scheduleFound) |
| 2044 | scheduleFound = |
| 2045 | Schedule.normalizeNonPipelinedInstructions(this, LoopPipelinerInfo); |
| 2046 | |
| 2047 | // If a schedule is found, check if it is a valid schedule too. |
| 2048 | if (scheduleFound) |
| 2049 | scheduleFound = Schedule.isValidSchedule(this); |
| 2050 | } |
| 2051 | |
| 2052 | LLVM_DEBUG(dbgs() << "Schedule Found? " << scheduleFounddo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Schedule Found? " << scheduleFound << " (II=" << Schedule.getInitiationInterval() << ")\n"; } } while (false) |
| 2053 | << " (II=" << Schedule.getInitiationInterval()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Schedule Found? " << scheduleFound << " (II=" << Schedule.getInitiationInterval() << ")\n"; } } while (false) |
| 2054 | << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Schedule Found? " << scheduleFound << " (II=" << Schedule.getInitiationInterval() << ")\n"; } } while (false); |
| 2055 | |
| 2056 | if (scheduleFound) { |
| 2057 | scheduleFound = LoopPipelinerInfo->shouldUseSchedule(*this, Schedule); |
| 2058 | if (!scheduleFound) |
| 2059 | LLVM_DEBUG(dbgs() << "Target rejected schedule\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Target rejected schedule\n" ; } } while (false); |
| 2060 | } |
| 2061 | |
| 2062 | if (scheduleFound) { |
| 2063 | Schedule.finalizeSchedule(this); |
| 2064 | Pass.ORE->emit([&]() { |
| 2065 | return MachineOptimizationRemarkAnalysis( |
| 2066 | DEBUG_TYPE"pipeliner", "schedule", Loop.getStartLoc(), Loop.getHeader()) |
| 2067 | << "Schedule found with Initiation Interval: " |
| 2068 | << ore::NV("II", Schedule.getInitiationInterval()) |
| 2069 | << ", MaxStageCount: " |
| 2070 | << ore::NV("MaxStageCount", Schedule.getMaxStageCount()); |
| 2071 | }); |
| 2072 | } else |
| 2073 | Schedule.reset(); |
| 2074 | |
| 2075 | return scheduleFound && Schedule.getMaxStageCount() > 0; |
| 2076 | } |
| 2077 | |
| 2078 | /// Return true if we can compute the amount the instruction changes |
| 2079 | /// during each iteration. Set Delta to the amount of the change. |
| 2080 | bool SwingSchedulerDAG::computeDelta(MachineInstr &MI, unsigned &Delta) { |
| 2081 | const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); |
| 2082 | const MachineOperand *BaseOp; |
| 2083 | int64_t Offset; |
| 2084 | bool OffsetIsScalable; |
| 2085 | if (!TII->getMemOperandWithOffset(MI, BaseOp, Offset, OffsetIsScalable, TRI)) |
| 2086 | return false; |
| 2087 | |
| 2088 | // FIXME: This algorithm assumes instructions have fixed-size offsets. |
| 2089 | if (OffsetIsScalable) |
| 2090 | return false; |
| 2091 | |
| 2092 | if (!BaseOp->isReg()) |
| 2093 | return false; |
| 2094 | |
| 2095 | Register BaseReg = BaseOp->getReg(); |
| 2096 | |
| 2097 | MachineRegisterInfo &MRI = MF.getRegInfo(); |
| 2098 | // Check if there is a Phi. If so, get the definition in the loop. |
| 2099 | MachineInstr *BaseDef = MRI.getVRegDef(BaseReg); |
| 2100 | if (BaseDef && BaseDef->isPHI()) { |
| 2101 | BaseReg = getLoopPhiReg(*BaseDef, MI.getParent()); |
| 2102 | BaseDef = MRI.getVRegDef(BaseReg); |
| 2103 | } |
| 2104 | if (!BaseDef) |
| 2105 | return false; |
| 2106 | |
| 2107 | int D = 0; |
| 2108 | if (!TII->getIncrementValue(*BaseDef, D) && D >= 0) |
| 2109 | return false; |
| 2110 | |
| 2111 | Delta = D; |
| 2112 | return true; |
| 2113 | } |
| 2114 | |
| 2115 | /// Check if we can change the instruction to use an offset value from the |
| 2116 | /// previous iteration. If so, return true and set the base and offset values |
| 2117 | /// so that we can rewrite the load, if necessary. |
| 2118 | /// v1 = Phi(v0, v3) |
| 2119 | /// v2 = load v1, 0 |
| 2120 | /// v3 = post_store v1, 4, x |
| 2121 | /// This function enables the load to be rewritten as v2 = load v3, 4. |
| 2122 | bool SwingSchedulerDAG::canUseLastOffsetValue(MachineInstr *MI, |
| 2123 | unsigned &BasePos, |
| 2124 | unsigned &OffsetPos, |
| 2125 | unsigned &NewBase, |
| 2126 | int64_t &Offset) { |
| 2127 | // Get the load instruction. |
| 2128 | if (TII->isPostIncrement(*MI)) |
| 2129 | return false; |
| 2130 | unsigned BasePosLd, OffsetPosLd; |
| 2131 | if (!TII->getBaseAndOffsetPosition(*MI, BasePosLd, OffsetPosLd)) |
| 2132 | return false; |
| 2133 | Register BaseReg = MI->getOperand(BasePosLd).getReg(); |
| 2134 | |
| 2135 | // Look for the Phi instruction. |
| 2136 | MachineRegisterInfo &MRI = MI->getMF()->getRegInfo(); |
| 2137 | MachineInstr *Phi = MRI.getVRegDef(BaseReg); |
| 2138 | if (!Phi || !Phi->isPHI()) |
| 2139 | return false; |
| 2140 | // Get the register defined in the loop block. |
| 2141 | unsigned PrevReg = getLoopPhiReg(*Phi, MI->getParent()); |
| 2142 | if (!PrevReg) |
| 2143 | return false; |
| 2144 | |
| 2145 | // Check for the post-increment load/store instruction. |
| 2146 | MachineInstr *PrevDef = MRI.getVRegDef(PrevReg); |
| 2147 | if (!PrevDef || PrevDef == MI) |
| 2148 | return false; |
| 2149 | |
| 2150 | if (!TII->isPostIncrement(*PrevDef)) |
| 2151 | return false; |
| 2152 | |
| 2153 | unsigned BasePos1 = 0, OffsetPos1 = 0; |
| 2154 | if (!TII->getBaseAndOffsetPosition(*PrevDef, BasePos1, OffsetPos1)) |
| 2155 | return false; |
| 2156 | |
| 2157 | // Make sure that the instructions do not access the same memory location in |
| 2158 | // the next iteration. |
| 2159 | int64_t LoadOffset = MI->getOperand(OffsetPosLd).getImm(); |
| 2160 | int64_t StoreOffset = PrevDef->getOperand(OffsetPos1).getImm(); |
| 2161 | MachineInstr *NewMI = MF.CloneMachineInstr(MI); |
| 2162 | NewMI->getOperand(OffsetPosLd).setImm(LoadOffset + StoreOffset); |
| 2163 | bool Disjoint = TII->areMemAccessesTriviallyDisjoint(*NewMI, *PrevDef); |
| 2164 | MF.deleteMachineInstr(NewMI); |
| 2165 | if (!Disjoint) |
| 2166 | return false; |
| 2167 | |
| 2168 | // Set the return value once we determine that we return true. |
| 2169 | BasePos = BasePosLd; |
| 2170 | OffsetPos = OffsetPosLd; |
| 2171 | NewBase = PrevReg; |
| 2172 | Offset = StoreOffset; |
| 2173 | return true; |
| 2174 | } |
| 2175 | |
| 2176 | /// Apply changes to the instruction if needed. The changes are need |
| 2177 | /// to improve the scheduling and depend up on the final schedule. |
| 2178 | void SwingSchedulerDAG::applyInstrChange(MachineInstr *MI, |
| 2179 | SMSchedule &Schedule) { |
| 2180 | SUnit *SU = getSUnit(MI); |
| 2181 | DenseMap<SUnit *, std::pair<unsigned, int64_t>>::iterator It = |
| 2182 | InstrChanges.find(SU); |
| 2183 | if (It != InstrChanges.end()) { |
| 2184 | std::pair<unsigned, int64_t> RegAndOffset = It->second; |
| 2185 | unsigned BasePos, OffsetPos; |
| 2186 | if (!TII->getBaseAndOffsetPosition(*MI, BasePos, OffsetPos)) |
| 2187 | return; |
| 2188 | Register BaseReg = MI->getOperand(BasePos).getReg(); |
| 2189 | MachineInstr *LoopDef = findDefInLoop(BaseReg); |
| 2190 | int DefStageNum = Schedule.stageScheduled(getSUnit(LoopDef)); |
| 2191 | int DefCycleNum = Schedule.cycleScheduled(getSUnit(LoopDef)); |
| 2192 | int BaseStageNum = Schedule.stageScheduled(SU); |
| 2193 | int BaseCycleNum = Schedule.cycleScheduled(SU); |
| 2194 | if (BaseStageNum < DefStageNum) { |
| 2195 | MachineInstr *NewMI = MF.CloneMachineInstr(MI); |
| 2196 | int OffsetDiff = DefStageNum - BaseStageNum; |
| 2197 | if (DefCycleNum < BaseCycleNum) { |
| 2198 | NewMI->getOperand(BasePos).setReg(RegAndOffset.first); |
| 2199 | if (OffsetDiff > 0) |
| 2200 | --OffsetDiff; |
| 2201 | } |
| 2202 | int64_t NewOffset = |
| 2203 | MI->getOperand(OffsetPos).getImm() + RegAndOffset.second * OffsetDiff; |
| 2204 | NewMI->getOperand(OffsetPos).setImm(NewOffset); |
| 2205 | SU->setInstr(NewMI); |
| 2206 | MISUnitMap[NewMI] = SU; |
| 2207 | NewMIs[MI] = NewMI; |
| 2208 | } |
| 2209 | } |
| 2210 | } |
| 2211 | |
| 2212 | /// Return the instruction in the loop that defines the register. |
| 2213 | /// If the definition is a Phi, then follow the Phi operand to |
| 2214 | /// the instruction in the loop. |
| 2215 | MachineInstr *SwingSchedulerDAG::findDefInLoop(Register Reg) { |
| 2216 | SmallPtrSet<MachineInstr *, 8> Visited; |
| 2217 | MachineInstr *Def = MRI.getVRegDef(Reg); |
| 2218 | while (Def->isPHI()) { |
| 2219 | if (!Visited.insert(Def).second) |
| 2220 | break; |
| 2221 | for (unsigned i = 1, e = Def->getNumOperands(); i < e; i += 2) |
| 2222 | if (Def->getOperand(i + 1).getMBB() == BB) { |
| 2223 | Def = MRI.getVRegDef(Def->getOperand(i).getReg()); |
| 2224 | break; |
| 2225 | } |
| 2226 | } |
| 2227 | return Def; |
| 2228 | } |
| 2229 | |
| 2230 | /// Return true for an order or output dependence that is loop carried |
| 2231 | /// potentially. A dependence is loop carried if the destination defines a valu |
| 2232 | /// that may be used or defined by the source in a subsequent iteration. |
| 2233 | bool SwingSchedulerDAG::isLoopCarriedDep(SUnit *Source, const SDep &Dep, |
| 2234 | bool isSucc) { |
| 2235 | if ((Dep.getKind() != SDep::Order && Dep.getKind() != SDep::Output) || |
| 2236 | Dep.isArtificial() || Dep.getSUnit()->isBoundaryNode()) |
| 2237 | return false; |
| 2238 | |
| 2239 | if (!SwpPruneLoopCarried) |
| 2240 | return true; |
| 2241 | |
| 2242 | if (Dep.getKind() == SDep::Output) |
| 2243 | return true; |
| 2244 | |
| 2245 | MachineInstr *SI = Source->getInstr(); |
| 2246 | MachineInstr *DI = Dep.getSUnit()->getInstr(); |
| 2247 | if (!isSucc) |
| 2248 | std::swap(SI, DI); |
| 2249 | assert(SI != nullptr && DI != nullptr && "Expecting SUnit with an MI.")(static_cast <bool> (SI != nullptr && DI != nullptr && "Expecting SUnit with an MI.") ? void (0) : __assert_fail ("SI != nullptr && DI != nullptr && \"Expecting SUnit with an MI.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 2249, __extension__ __PRETTY_FUNCTION__)); |
| 2250 | |
| 2251 | // Assume ordered loads and stores may have a loop carried dependence. |
| 2252 | if (SI->hasUnmodeledSideEffects() || DI->hasUnmodeledSideEffects() || |
| 2253 | SI->mayRaiseFPException() || DI->mayRaiseFPException() || |
| 2254 | SI->hasOrderedMemoryRef() || DI->hasOrderedMemoryRef()) |
| 2255 | return true; |
| 2256 | |
| 2257 | // Only chain dependences between a load and store can be loop carried. |
| 2258 | if (!DI->mayStore() || !SI->mayLoad()) |
| 2259 | return false; |
| 2260 | |
| 2261 | unsigned DeltaS, DeltaD; |
| 2262 | if (!computeDelta(*SI, DeltaS) || !computeDelta(*DI, DeltaD)) |
| 2263 | return true; |
| 2264 | |
| 2265 | const MachineOperand *BaseOpS, *BaseOpD; |
| 2266 | int64_t OffsetS, OffsetD; |
| 2267 | bool OffsetSIsScalable, OffsetDIsScalable; |
| 2268 | const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); |
| 2269 | if (!TII->getMemOperandWithOffset(*SI, BaseOpS, OffsetS, OffsetSIsScalable, |
| 2270 | TRI) || |
| 2271 | !TII->getMemOperandWithOffset(*DI, BaseOpD, OffsetD, OffsetDIsScalable, |
| 2272 | TRI)) |
| 2273 | return true; |
| 2274 | |
| 2275 | assert(!OffsetSIsScalable && !OffsetDIsScalable &&(static_cast <bool> (!OffsetSIsScalable && !OffsetDIsScalable && "Expected offsets to be byte offsets") ? void (0) : __assert_fail ("!OffsetSIsScalable && !OffsetDIsScalable && \"Expected offsets to be byte offsets\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 2276, __extension__ __PRETTY_FUNCTION__)) |
| 2276 | "Expected offsets to be byte offsets")(static_cast <bool> (!OffsetSIsScalable && !OffsetDIsScalable && "Expected offsets to be byte offsets") ? void (0) : __assert_fail ("!OffsetSIsScalable && !OffsetDIsScalable && \"Expected offsets to be byte offsets\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 2276, __extension__ __PRETTY_FUNCTION__)); |
| 2277 | |
| 2278 | MachineInstr *DefS = MRI.getVRegDef(BaseOpS->getReg()); |
| 2279 | MachineInstr *DefD = MRI.getVRegDef(BaseOpD->getReg()); |
| 2280 | if (!DefS || !DefD || !DefS->isPHI() || !DefD->isPHI()) |
| 2281 | return true; |
| 2282 | |
| 2283 | unsigned InitValS = 0; |
| 2284 | unsigned LoopValS = 0; |
| 2285 | unsigned InitValD = 0; |
| 2286 | unsigned LoopValD = 0; |
| 2287 | getPhiRegs(*DefS, BB, InitValS, LoopValS); |
| 2288 | getPhiRegs(*DefD, BB, InitValD, LoopValD); |
| 2289 | MachineInstr *InitDefS = MRI.getVRegDef(InitValS); |
| 2290 | MachineInstr *InitDefD = MRI.getVRegDef(InitValD); |
| 2291 | |
| 2292 | if (!InitDefS->isIdenticalTo(*InitDefD)) |
| 2293 | return true; |
| 2294 | |
| 2295 | // Check that the base register is incremented by a constant value for each |
| 2296 | // iteration. |
| 2297 | MachineInstr *LoopDefS = MRI.getVRegDef(LoopValS); |
| 2298 | int D = 0; |
| 2299 | if (!LoopDefS || !TII->getIncrementValue(*LoopDefS, D)) |
| 2300 | return true; |
| 2301 | |
| 2302 | uint64_t AccessSizeS = (*SI->memoperands_begin())->getSize(); |
| 2303 | uint64_t AccessSizeD = (*DI->memoperands_begin())->getSize(); |
| 2304 | |
| 2305 | // This is the main test, which checks the offset values and the loop |
| 2306 | // increment value to determine if the accesses may be loop carried. |
| 2307 | if (AccessSizeS == MemoryLocation::UnknownSize || |
| 2308 | AccessSizeD == MemoryLocation::UnknownSize) |
| 2309 | return true; |
| 2310 | |
| 2311 | if (DeltaS != DeltaD || DeltaS < AccessSizeS || DeltaD < AccessSizeD) |
| 2312 | return true; |
| 2313 | |
| 2314 | return (OffsetS + (int64_t)AccessSizeS < OffsetD + (int64_t)AccessSizeD); |
| 2315 | } |
| 2316 | |
| 2317 | void SwingSchedulerDAG::postProcessDAG() { |
| 2318 | for (auto &M : Mutations) |
| 2319 | M->apply(this); |
| 2320 | } |
| 2321 | |
| 2322 | /// Try to schedule the node at the specified StartCycle and continue |
| 2323 | /// until the node is schedule or the EndCycle is reached. This function |
| 2324 | /// returns true if the node is scheduled. This routine may search either |
| 2325 | /// forward or backward for a place to insert the instruction based upon |
| 2326 | /// the relative values of StartCycle and EndCycle. |
| 2327 | bool SMSchedule::insert(SUnit *SU, int StartCycle, int EndCycle, int II) { |
| 2328 | bool forward = true; |
| 2329 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to insert node between " << StartCycle << " and " << EndCycle << " II: " << II << "\n"; }; } } while (false) |
| 2330 | dbgs() << "Trying to insert node between " << StartCycle << " and "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to insert node between " << StartCycle << " and " << EndCycle << " II: " << II << "\n"; }; } } while (false) |
| 2331 | << EndCycle << " II: " << II << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to insert node between " << StartCycle << " and " << EndCycle << " II: " << II << "\n"; }; } } while (false) |
| 2332 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to insert node between " << StartCycle << " and " << EndCycle << " II: " << II << "\n"; }; } } while (false); |
| 2333 | if (StartCycle > EndCycle) |
| 2334 | forward = false; |
| 2335 | |
| 2336 | // The terminating condition depends on the direction. |
| 2337 | int termCycle = forward ? EndCycle + 1 : EndCycle - 1; |
| 2338 | for (int curCycle = StartCycle; curCycle != termCycle; |
| 2339 | forward ? ++curCycle : --curCycle) { |
| 2340 | |
| 2341 | if (ST.getInstrInfo()->isZeroCost(SU->getInstr()->getOpcode()) || |
| 2342 | ProcItinResources.canReserveResources(*SU, curCycle)) { |
| 2343 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tinsert at cycle " << curCycle << " "; SU->getInstr()->dump(); }; } } while (false) |
| 2344 | dbgs() << "\tinsert at cycle " << curCycle << " ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tinsert at cycle " << curCycle << " "; SU->getInstr()->dump(); }; } } while (false) |
| 2345 | SU->getInstr()->dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tinsert at cycle " << curCycle << " "; SU->getInstr()->dump(); }; } } while (false) |
| 2346 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tinsert at cycle " << curCycle << " "; SU->getInstr()->dump(); }; } } while (false); |
| 2347 | |
| 2348 | if (!ST.getInstrInfo()->isZeroCost(SU->getInstr()->getOpcode())) |
| 2349 | ProcItinResources.reserveResources(*SU, curCycle); |
| 2350 | ScheduledInstrs[curCycle].push_back(SU); |
| 2351 | InstrToCycle.insert(std::make_pair(SU, curCycle)); |
| 2352 | if (curCycle > LastCycle) |
| 2353 | LastCycle = curCycle; |
| 2354 | if (curCycle < FirstCycle) |
| 2355 | FirstCycle = curCycle; |
| 2356 | return true; |
| 2357 | } |
| 2358 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tfailed to insert at cycle " << curCycle << " "; SU->getInstr()->dump() ; }; } } while (false) |
| 2359 | dbgs() << "\tfailed to insert at cycle " << curCycle << " ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tfailed to insert at cycle " << curCycle << " "; SU->getInstr()->dump() ; }; } } while (false) |
| 2360 | SU->getInstr()->dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tfailed to insert at cycle " << curCycle << " "; SU->getInstr()->dump() ; }; } } while (false) |
| 2361 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "\tfailed to insert at cycle " << curCycle << " "; SU->getInstr()->dump() ; }; } } while (false); |
| 2362 | } |
| 2363 | return false; |
| 2364 | } |
| 2365 | |
| 2366 | // Return the cycle of the earliest scheduled instruction in the chain. |
| 2367 | int SMSchedule::earliestCycleInChain(const SDep &Dep) { |
| 2368 | SmallPtrSet<SUnit *, 8> Visited; |
| 2369 | SmallVector<SDep, 8> Worklist; |
| 2370 | Worklist.push_back(Dep); |
| 2371 | int EarlyCycle = INT_MAX2147483647; |
| 2372 | while (!Worklist.empty()) { |
| 2373 | const SDep &Cur = Worklist.pop_back_val(); |
| 2374 | SUnit *PrevSU = Cur.getSUnit(); |
| 2375 | if (Visited.count(PrevSU)) |
| 2376 | continue; |
| 2377 | std::map<SUnit *, int>::const_iterator it = InstrToCycle.find(PrevSU); |
| 2378 | if (it == InstrToCycle.end()) |
| 2379 | continue; |
| 2380 | EarlyCycle = std::min(EarlyCycle, it->second); |
| 2381 | for (const auto &PI : PrevSU->Preds) |
| 2382 | if (PI.getKind() == SDep::Order || PI.getKind() == SDep::Output) |
| 2383 | Worklist.push_back(PI); |
| 2384 | Visited.insert(PrevSU); |
| 2385 | } |
| 2386 | return EarlyCycle; |
| 2387 | } |
| 2388 | |
| 2389 | // Return the cycle of the latest scheduled instruction in the chain. |
| 2390 | int SMSchedule::latestCycleInChain(const SDep &Dep) { |
| 2391 | SmallPtrSet<SUnit *, 8> Visited; |
| 2392 | SmallVector<SDep, 8> Worklist; |
| 2393 | Worklist.push_back(Dep); |
| 2394 | int LateCycle = INT_MIN(-2147483647 -1); |
| 2395 | while (!Worklist.empty()) { |
| 2396 | const SDep &Cur = Worklist.pop_back_val(); |
| 2397 | SUnit *SuccSU = Cur.getSUnit(); |
| 2398 | if (Visited.count(SuccSU) || SuccSU->isBoundaryNode()) |
| 2399 | continue; |
| 2400 | std::map<SUnit *, int>::const_iterator it = InstrToCycle.find(SuccSU); |
| 2401 | if (it == InstrToCycle.end()) |
| 2402 | continue; |
| 2403 | LateCycle = std::max(LateCycle, it->second); |
| 2404 | for (const auto &SI : SuccSU->Succs) |
| 2405 | if (SI.getKind() == SDep::Order || SI.getKind() == SDep::Output) |
| 2406 | Worklist.push_back(SI); |
| 2407 | Visited.insert(SuccSU); |
| 2408 | } |
| 2409 | return LateCycle; |
| 2410 | } |
| 2411 | |
| 2412 | /// If an instruction has a use that spans multiple iterations, then |
| 2413 | /// return true. These instructions are characterized by having a back-ege |
| 2414 | /// to a Phi, which contains a reference to another Phi. |
| 2415 | static SUnit *multipleIterations(SUnit *SU, SwingSchedulerDAG *DAG) { |
| 2416 | for (auto &P : SU->Preds) |
| 2417 | if (DAG->isBackedge(SU, P) && P.getSUnit()->getInstr()->isPHI()) |
| 2418 | for (auto &S : P.getSUnit()->Succs) |
| 2419 | if (S.getKind() == SDep::Data && S.getSUnit()->getInstr()->isPHI()) |
| 2420 | return P.getSUnit(); |
| 2421 | return nullptr; |
| 2422 | } |
| 2423 | |
| 2424 | /// Compute the scheduling start slot for the instruction. The start slot |
| 2425 | /// depends on any predecessor or successor nodes scheduled already. |
| 2426 | void SMSchedule::computeStart(SUnit *SU, int *MaxEarlyStart, int *MinLateStart, |
| 2427 | int *MinEnd, int *MaxStart, int II, |
| 2428 | SwingSchedulerDAG *DAG) { |
| 2429 | // Iterate over each instruction that has been scheduled already. The start |
| 2430 | // slot computation depends on whether the previously scheduled instruction |
| 2431 | // is a predecessor or successor of the specified instruction. |
| 2432 | for (int cycle = getFirstCycle(); cycle <= LastCycle; ++cycle) { |
| 2433 | |
| 2434 | // Iterate over each instruction in the current cycle. |
| 2435 | for (SUnit *I : getInstructions(cycle)) { |
| 2436 | // Because we're processing a DAG for the dependences, we recognize |
| 2437 | // the back-edge in recurrences by anti dependences. |
| 2438 | for (unsigned i = 0, e = (unsigned)SU->Preds.size(); i != e; ++i) { |
| 2439 | const SDep &Dep = SU->Preds[i]; |
| 2440 | if (Dep.getSUnit() == I) { |
| 2441 | if (!DAG->isBackedge(SU, Dep)) { |
| 2442 | int EarlyStart = cycle + Dep.getLatency() - |
| 2443 | DAG->getDistance(Dep.getSUnit(), SU, Dep) * II; |
| 2444 | *MaxEarlyStart = std::max(*MaxEarlyStart, EarlyStart); |
| 2445 | if (DAG->isLoopCarriedDep(SU, Dep, false)) { |
| 2446 | int End = earliestCycleInChain(Dep) + (II - 1); |
| 2447 | *MinEnd = std::min(*MinEnd, End); |
| 2448 | } |
| 2449 | } else { |
| 2450 | int LateStart = cycle - Dep.getLatency() + |
| 2451 | DAG->getDistance(SU, Dep.getSUnit(), Dep) * II; |
| 2452 | *MinLateStart = std::min(*MinLateStart, LateStart); |
| 2453 | } |
| 2454 | } |
| 2455 | // For instruction that requires multiple iterations, make sure that |
| 2456 | // the dependent instruction is not scheduled past the definition. |
| 2457 | SUnit *BE = multipleIterations(I, DAG); |
| 2458 | if (BE && Dep.getSUnit() == BE && !SU->getInstr()->isPHI() && |
| 2459 | !SU->isPred(I)) |
| 2460 | *MinLateStart = std::min(*MinLateStart, cycle); |
| 2461 | } |
| 2462 | for (unsigned i = 0, e = (unsigned)SU->Succs.size(); i != e; ++i) { |
| 2463 | if (SU->Succs[i].getSUnit() == I) { |
| 2464 | const SDep &Dep = SU->Succs[i]; |
| 2465 | if (!DAG->isBackedge(SU, Dep)) { |
| 2466 | int LateStart = cycle - Dep.getLatency() + |
| 2467 | DAG->getDistance(SU, Dep.getSUnit(), Dep) * II; |
| 2468 | *MinLateStart = std::min(*MinLateStart, LateStart); |
| 2469 | if (DAG->isLoopCarriedDep(SU, Dep)) { |
| 2470 | int Start = latestCycleInChain(Dep) + 1 - II; |
| 2471 | *MaxStart = std::max(*MaxStart, Start); |
| 2472 | } |
| 2473 | } else { |
| 2474 | int EarlyStart = cycle + Dep.getLatency() - |
| 2475 | DAG->getDistance(Dep.getSUnit(), SU, Dep) * II; |
| 2476 | *MaxEarlyStart = std::max(*MaxEarlyStart, EarlyStart); |
| 2477 | } |
| 2478 | } |
| 2479 | } |
| 2480 | } |
| 2481 | } |
| 2482 | } |
| 2483 | |
| 2484 | /// Order the instructions within a cycle so that the definitions occur |
| 2485 | /// before the uses. Returns true if the instruction is added to the start |
| 2486 | /// of the list, or false if added to the end. |
| 2487 | void SMSchedule::orderDependence(SwingSchedulerDAG *SSD, SUnit *SU, |
| 2488 | std::deque<SUnit *> &Insts) { |
| 2489 | MachineInstr *MI = SU->getInstr(); |
| 2490 | bool OrderBeforeUse = false; |
| 2491 | bool OrderAfterDef = false; |
| 2492 | bool OrderBeforeDef = false; |
| 2493 | unsigned MoveDef = 0; |
| 2494 | unsigned MoveUse = 0; |
| 2495 | int StageInst1 = stageScheduled(SU); |
| 2496 | |
| 2497 | unsigned Pos = 0; |
| 2498 | for (std::deque<SUnit *>::iterator I = Insts.begin(), E = Insts.end(); I != E; |
| 2499 | ++I, ++Pos) { |
| 2500 | for (MachineOperand &MO : MI->operands()) { |
| 2501 | if (!MO.isReg() || !MO.getReg().isVirtual()) |
| 2502 | continue; |
| 2503 | |
| 2504 | Register Reg = MO.getReg(); |
| 2505 | unsigned BasePos, OffsetPos; |
| 2506 | if (ST.getInstrInfo()->getBaseAndOffsetPosition(*MI, BasePos, OffsetPos)) |
| 2507 | if (MI->getOperand(BasePos).getReg() == Reg) |
| 2508 | if (unsigned NewReg = SSD->getInstrBaseReg(SU)) |
| 2509 | Reg = NewReg; |
| 2510 | bool Reads, Writes; |
| 2511 | std::tie(Reads, Writes) = |
| 2512 | (*I)->getInstr()->readsWritesVirtualRegister(Reg); |
| 2513 | if (MO.isDef() && Reads && stageScheduled(*I) <= StageInst1) { |
| 2514 | OrderBeforeUse = true; |
| 2515 | if (MoveUse == 0) |
| 2516 | MoveUse = Pos; |
| 2517 | } else if (MO.isDef() && Reads && stageScheduled(*I) > StageInst1) { |
| 2518 | // Add the instruction after the scheduled instruction. |
| 2519 | OrderAfterDef = true; |
| 2520 | MoveDef = Pos; |
| 2521 | } else if (MO.isUse() && Writes && stageScheduled(*I) == StageInst1) { |
| 2522 | if (cycleScheduled(*I) == cycleScheduled(SU) && !(*I)->isSucc(SU)) { |
| 2523 | OrderBeforeUse = true; |
| 2524 | if (MoveUse == 0) |
| 2525 | MoveUse = Pos; |
| 2526 | } else { |
| 2527 | OrderAfterDef = true; |
| 2528 | MoveDef = Pos; |
| 2529 | } |
| 2530 | } else if (MO.isUse() && Writes && stageScheduled(*I) > StageInst1) { |
| 2531 | OrderBeforeUse = true; |
| 2532 | if (MoveUse == 0) |
| 2533 | MoveUse = Pos; |
| 2534 | if (MoveUse != 0) { |
| 2535 | OrderAfterDef = true; |
| 2536 | MoveDef = Pos - 1; |
| 2537 | } |
| 2538 | } else if (MO.isUse() && Writes && stageScheduled(*I) < StageInst1) { |
| 2539 | // Add the instruction before the scheduled instruction. |
| 2540 | OrderBeforeUse = true; |
| 2541 | if (MoveUse == 0) |
| 2542 | MoveUse = Pos; |
| 2543 | } else if (MO.isUse() && stageScheduled(*I) == StageInst1 && |
| 2544 | isLoopCarriedDefOfUse(SSD, (*I)->getInstr(), MO)) { |
| 2545 | if (MoveUse == 0) { |
| 2546 | OrderBeforeDef = true; |
| 2547 | MoveUse = Pos; |
| 2548 | } |
| 2549 | } |
| 2550 | } |
| 2551 | // Check for order dependences between instructions. Make sure the source |
| 2552 | // is ordered before the destination. |
| 2553 | for (auto &S : SU->Succs) { |
| 2554 | if (S.getSUnit() != *I) |
| 2555 | continue; |
| 2556 | if (S.getKind() == SDep::Order && stageScheduled(*I) == StageInst1) { |
| 2557 | OrderBeforeUse = true; |
| 2558 | if (Pos < MoveUse) |
| 2559 | MoveUse = Pos; |
| 2560 | } |
| 2561 | // We did not handle HW dependences in previous for loop, |
| 2562 | // and we normally set Latency = 0 for Anti deps, |
| 2563 | // so may have nodes in same cycle with Anti denpendent on HW regs. |
| 2564 | else if (S.getKind() == SDep::Anti && stageScheduled(*I) == StageInst1) { |
| 2565 | OrderBeforeUse = true; |
| 2566 | if ((MoveUse == 0) || (Pos < MoveUse)) |
| 2567 | MoveUse = Pos; |
| 2568 | } |
| 2569 | } |
| 2570 | for (auto &P : SU->Preds) { |
| 2571 | if (P.getSUnit() != *I) |
| 2572 | continue; |
| 2573 | if (P.getKind() == SDep::Order && stageScheduled(*I) == StageInst1) { |
| 2574 | OrderAfterDef = true; |
| 2575 | MoveDef = Pos; |
| 2576 | } |
| 2577 | } |
| 2578 | } |
| 2579 | |
| 2580 | // A circular dependence. |
| 2581 | if (OrderAfterDef && OrderBeforeUse && MoveUse == MoveDef) |
| 2582 | OrderBeforeUse = false; |
| 2583 | |
| 2584 | // OrderAfterDef takes precedences over OrderBeforeDef. The latter is due |
| 2585 | // to a loop-carried dependence. |
| 2586 | if (OrderBeforeDef) |
| 2587 | OrderBeforeUse = !OrderAfterDef || (MoveUse > MoveDef); |
| 2588 | |
| 2589 | // The uncommon case when the instruction order needs to be updated because |
| 2590 | // there is both a use and def. |
| 2591 | if (OrderBeforeUse && OrderAfterDef) { |
| 2592 | SUnit *UseSU = Insts.at(MoveUse); |
| 2593 | SUnit *DefSU = Insts.at(MoveDef); |
| 2594 | if (MoveUse > MoveDef) { |
| 2595 | Insts.erase(Insts.begin() + MoveUse); |
| 2596 | Insts.erase(Insts.begin() + MoveDef); |
| 2597 | } else { |
| 2598 | Insts.erase(Insts.begin() + MoveDef); |
| 2599 | Insts.erase(Insts.begin() + MoveUse); |
| 2600 | } |
| 2601 | orderDependence(SSD, UseSU, Insts); |
| 2602 | orderDependence(SSD, SU, Insts); |
| 2603 | orderDependence(SSD, DefSU, Insts); |
| 2604 | return; |
| 2605 | } |
| 2606 | // Put the new instruction first if there is a use in the list. Otherwise, |
| 2607 | // put it at the end of the list. |
| 2608 | if (OrderBeforeUse) |
| 2609 | Insts.push_front(SU); |
| 2610 | else |
| 2611 | Insts.push_back(SU); |
| 2612 | } |
| 2613 | |
| 2614 | /// Return true if the scheduled Phi has a loop carried operand. |
| 2615 | bool SMSchedule::isLoopCarried(SwingSchedulerDAG *SSD, MachineInstr &Phi) { |
| 2616 | if (!Phi.isPHI()) |
| 2617 | return false; |
| 2618 | assert(Phi.isPHI() && "Expecting a Phi.")(static_cast <bool> (Phi.isPHI() && "Expecting a Phi." ) ? void (0) : __assert_fail ("Phi.isPHI() && \"Expecting a Phi.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 2618, __extension__ __PRETTY_FUNCTION__)); |
| 2619 | SUnit *DefSU = SSD->getSUnit(&Phi); |
| 2620 | unsigned DefCycle = cycleScheduled(DefSU); |
| 2621 | int DefStage = stageScheduled(DefSU); |
| 2622 | |
| 2623 | unsigned InitVal = 0; |
| 2624 | unsigned LoopVal = 0; |
| 2625 | getPhiRegs(Phi, Phi.getParent(), InitVal, LoopVal); |
| 2626 | SUnit *UseSU = SSD->getSUnit(MRI.getVRegDef(LoopVal)); |
| 2627 | if (!UseSU) |
| 2628 | return true; |
| 2629 | if (UseSU->getInstr()->isPHI()) |
| 2630 | return true; |
| 2631 | unsigned LoopCycle = cycleScheduled(UseSU); |
| 2632 | int LoopStage = stageScheduled(UseSU); |
| 2633 | return (LoopCycle > DefCycle) || (LoopStage <= DefStage); |
| 2634 | } |
| 2635 | |
| 2636 | /// Return true if the instruction is a definition that is loop carried |
| 2637 | /// and defines the use on the next iteration. |
| 2638 | /// v1 = phi(v2, v3) |
| 2639 | /// (Def) v3 = op v1 |
| 2640 | /// (MO) = v1 |
| 2641 | /// If MO appears before Def, then then v1 and v3 may get assigned to the same |
| 2642 | /// register. |
| 2643 | bool SMSchedule::isLoopCarriedDefOfUse(SwingSchedulerDAG *SSD, |
| 2644 | MachineInstr *Def, MachineOperand &MO) { |
| 2645 | if (!MO.isReg()) |
| 2646 | return false; |
| 2647 | if (Def->isPHI()) |
| 2648 | return false; |
| 2649 | MachineInstr *Phi = MRI.getVRegDef(MO.getReg()); |
| 2650 | if (!Phi || !Phi->isPHI() || Phi->getParent() != Def->getParent()) |
| 2651 | return false; |
| 2652 | if (!isLoopCarried(SSD, *Phi)) |
| 2653 | return false; |
| 2654 | unsigned LoopReg = getLoopPhiReg(*Phi, Phi->getParent()); |
| 2655 | for (MachineOperand &DMO : Def->operands()) { |
| 2656 | if (!DMO.isReg() || !DMO.isDef()) |
| 2657 | continue; |
| 2658 | if (DMO.getReg() == LoopReg) |
| 2659 | return true; |
| 2660 | } |
| 2661 | return false; |
| 2662 | } |
| 2663 | |
| 2664 | /// Determine transitive dependences of unpipelineable instructions |
| 2665 | SmallSet<SUnit *, 8> SMSchedule::computeUnpipelineableNodes( |
| 2666 | SwingSchedulerDAG *SSD, TargetInstrInfo::PipelinerLoopInfo *PLI) { |
| 2667 | SmallSet<SUnit *, 8> DoNotPipeline; |
| 2668 | SmallVector<SUnit *, 8> Worklist; |
| 2669 | |
| 2670 | for (auto &SU : SSD->SUnits) |
| 2671 | if (SU.isInstr() && PLI->shouldIgnoreForPipelining(SU.getInstr())) |
| 2672 | Worklist.push_back(&SU); |
| 2673 | |
| 2674 | while (!Worklist.empty()) { |
| 2675 | auto SU = Worklist.pop_back_val(); |
| 2676 | if (DoNotPipeline.count(SU)) |
| 2677 | continue; |
| 2678 | LLVM_DEBUG(dbgs() << "Do not pipeline SU(" << SU->NodeNum << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Do not pipeline SU(" << SU->NodeNum << ")\n"; } } while (false); |
| 2679 | DoNotPipeline.insert(SU); |
| 2680 | for (auto &Dep : SU->Preds) |
| 2681 | Worklist.push_back(Dep.getSUnit()); |
| 2682 | if (SU->getInstr()->isPHI()) |
| 2683 | for (auto &Dep : SU->Succs) |
| 2684 | if (Dep.getKind() == SDep::Anti) |
| 2685 | Worklist.push_back(Dep.getSUnit()); |
| 2686 | } |
| 2687 | return DoNotPipeline; |
| 2688 | } |
| 2689 | |
| 2690 | // Determine all instructions upon which any unpipelineable instruction depends |
| 2691 | // and ensure that they are in stage 0. If unable to do so, return false. |
| 2692 | bool SMSchedule::normalizeNonPipelinedInstructions( |
| 2693 | SwingSchedulerDAG *SSD, TargetInstrInfo::PipelinerLoopInfo *PLI) { |
| 2694 | SmallSet<SUnit *, 8> DNP = computeUnpipelineableNodes(SSD, PLI); |
| 2695 | |
| 2696 | int NewLastCycle = INT_MIN(-2147483647 -1); |
| 2697 | for (SUnit &SU : SSD->SUnits) { |
| 2698 | if (!SU.isInstr()) |
| 2699 | continue; |
| 2700 | if (!DNP.contains(&SU) || stageScheduled(&SU) == 0) { |
| 2701 | NewLastCycle = std::max(NewLastCycle, InstrToCycle[&SU]); |
| 2702 | continue; |
| 2703 | } |
| 2704 | |
| 2705 | // Put the non-pipelined instruction as early as possible in the schedule |
| 2706 | int NewCycle = getFirstCycle(); |
| 2707 | for (auto &Dep : SU.Preds) |
| 2708 | NewCycle = std::max(InstrToCycle[Dep.getSUnit()], NewCycle); |
| 2709 | |
| 2710 | int OldCycle = InstrToCycle[&SU]; |
| 2711 | if (OldCycle != NewCycle) { |
| 2712 | InstrToCycle[&SU] = NewCycle; |
| 2713 | auto &OldS = getInstructions(OldCycle); |
| 2714 | llvm::erase_value(OldS, &SU); |
| 2715 | getInstructions(NewCycle).emplace_back(&SU); |
| 2716 | LLVM_DEBUG(dbgs() << "SU(" << SU.NodeNumdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "SU(" << SU.NodeNum << ") is not pipelined; moving from cycle " << OldCycle << " to " << NewCycle << " Instr:" << *SU.getInstr (); } } while (false) |
| 2717 | << ") is not pipelined; moving from cycle " << OldCycledo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "SU(" << SU.NodeNum << ") is not pipelined; moving from cycle " << OldCycle << " to " << NewCycle << " Instr:" << *SU.getInstr (); } } while (false) |
| 2718 | << " to " << NewCycle << " Instr:" << *SU.getInstr())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "SU(" << SU.NodeNum << ") is not pipelined; moving from cycle " << OldCycle << " to " << NewCycle << " Instr:" << *SU.getInstr (); } } while (false); |
| 2719 | } |
| 2720 | NewLastCycle = std::max(NewLastCycle, NewCycle); |
| 2721 | } |
| 2722 | LastCycle = NewLastCycle; |
| 2723 | return true; |
| 2724 | } |
| 2725 | |
| 2726 | // Check if the generated schedule is valid. This function checks if |
| 2727 | // an instruction that uses a physical register is scheduled in a |
| 2728 | // different stage than the definition. The pipeliner does not handle |
| 2729 | // physical register values that may cross a basic block boundary. |
| 2730 | // Furthermore, if a physical def/use pair is assigned to the same |
| 2731 | // cycle, orderDependence does not guarantee def/use ordering, so that |
| 2732 | // case should be considered invalid. (The test checks for both |
| 2733 | // earlier and same-cycle use to be more robust.) |
| 2734 | bool SMSchedule::isValidSchedule(SwingSchedulerDAG *SSD) { |
| 2735 | for (SUnit &SU : SSD->SUnits) { |
| 2736 | if (!SU.hasPhysRegDefs) |
| 2737 | continue; |
| 2738 | int StageDef = stageScheduled(&SU); |
| 2739 | int CycleDef = InstrToCycle[&SU]; |
| 2740 | assert(StageDef != -1 && "Instruction should have been scheduled.")(static_cast <bool> (StageDef != -1 && "Instruction should have been scheduled." ) ? void (0) : __assert_fail ("StageDef != -1 && \"Instruction should have been scheduled.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 2740, __extension__ __PRETTY_FUNCTION__)); |
| 2741 | for (auto &SI : SU.Succs) |
| 2742 | if (SI.isAssignedRegDep() && !SI.getSUnit()->isBoundaryNode()) |
| 2743 | if (Register::isPhysicalRegister(SI.getReg())) { |
| 2744 | if (stageScheduled(SI.getSUnit()) != StageDef) |
| 2745 | return false; |
| 2746 | if (InstrToCycle[SI.getSUnit()] <= CycleDef) |
| 2747 | return false; |
| 2748 | } |
| 2749 | } |
| 2750 | return true; |
| 2751 | } |
| 2752 | |
| 2753 | /// A property of the node order in swing-modulo-scheduling is |
| 2754 | /// that for nodes outside circuits the following holds: |
| 2755 | /// none of them is scheduled after both a successor and a |
| 2756 | /// predecessor. |
| 2757 | /// The method below checks whether the property is met. |
| 2758 | /// If not, debug information is printed and statistics information updated. |
| 2759 | /// Note that we do not use an assert statement. |
| 2760 | /// The reason is that although an invalid node oder may prevent |
| 2761 | /// the pipeliner from finding a pipelined schedule for arbitrary II, |
| 2762 | /// it does not lead to the generation of incorrect code. |
| 2763 | void SwingSchedulerDAG::checkValidNodeOrder(const NodeSetType &Circuits) const { |
| 2764 | |
| 2765 | // a sorted vector that maps each SUnit to its index in the NodeOrder |
| 2766 | typedef std::pair<SUnit *, unsigned> UnitIndex; |
| 2767 | std::vector<UnitIndex> Indices(NodeOrder.size(), std::make_pair(nullptr, 0)); |
| 2768 | |
| 2769 | for (unsigned i = 0, s = NodeOrder.size(); i < s; ++i) |
| 2770 | Indices.push_back(std::make_pair(NodeOrder[i], i)); |
| 2771 | |
| 2772 | auto CompareKey = [](UnitIndex i1, UnitIndex i2) { |
| 2773 | return std::get<0>(i1) < std::get<0>(i2); |
| 2774 | }; |
| 2775 | |
| 2776 | // sort, so that we can perform a binary search |
| 2777 | llvm::sort(Indices, CompareKey); |
| 2778 | |
| 2779 | bool Valid = true; |
| 2780 | (void)Valid; |
| 2781 | // for each SUnit in the NodeOrder, check whether |
| 2782 | // it appears after both a successor and a predecessor |
| 2783 | // of the SUnit. If this is the case, and the SUnit |
| 2784 | // is not part of circuit, then the NodeOrder is not |
| 2785 | // valid. |
| 2786 | for (unsigned i = 0, s = NodeOrder.size(); i < s; ++i) { |
| 2787 | SUnit *SU = NodeOrder[i]; |
| 2788 | unsigned Index = i; |
| 2789 | |
| 2790 | bool PredBefore = false; |
| 2791 | bool SuccBefore = false; |
| 2792 | |
| 2793 | SUnit *Succ; |
| 2794 | SUnit *Pred; |
| 2795 | (void)Succ; |
| 2796 | (void)Pred; |
| 2797 | |
| 2798 | for (SDep &PredEdge : SU->Preds) { |
| 2799 | SUnit *PredSU = PredEdge.getSUnit(); |
| 2800 | unsigned PredIndex = std::get<1>( |
| 2801 | *llvm::lower_bound(Indices, std::make_pair(PredSU, 0), CompareKey)); |
| 2802 | if (!PredSU->getInstr()->isPHI() && PredIndex < Index) { |
| 2803 | PredBefore = true; |
| 2804 | Pred = PredSU; |
| 2805 | break; |
| 2806 | } |
| 2807 | } |
| 2808 | |
| 2809 | for (SDep &SuccEdge : SU->Succs) { |
| 2810 | SUnit *SuccSU = SuccEdge.getSUnit(); |
| 2811 | // Do not process a boundary node, it was not included in NodeOrder, |
| 2812 | // hence not in Indices either, call to std::lower_bound() below will |
| 2813 | // return Indices.end(). |
| 2814 | if (SuccSU->isBoundaryNode()) |
| 2815 | continue; |
| 2816 | unsigned SuccIndex = std::get<1>( |
| 2817 | *llvm::lower_bound(Indices, std::make_pair(SuccSU, 0), CompareKey)); |
| 2818 | if (!SuccSU->getInstr()->isPHI() && SuccIndex < Index) { |
| 2819 | SuccBefore = true; |
| 2820 | Succ = SuccSU; |
| 2821 | break; |
| 2822 | } |
| 2823 | } |
| 2824 | |
| 2825 | if (PredBefore && SuccBefore && !SU->getInstr()->isPHI()) { |
| 2826 | // instructions in circuits are allowed to be scheduled |
| 2827 | // after both a successor and predecessor. |
| 2828 | bool InCircuit = llvm::any_of( |
| 2829 | Circuits, [SU](const NodeSet &Circuit) { return Circuit.count(SU); }); |
| 2830 | if (InCircuit) |
| 2831 | LLVM_DEBUG(dbgs() << "In a circuit, predecessor ";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "In a circuit, predecessor " ;; } } while (false); |
| 2832 | else { |
| 2833 | Valid = false; |
| 2834 | NumNodeOrderIssues++; |
| 2835 | LLVM_DEBUG(dbgs() << "Predecessor ";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Predecessor ";; } } while ( false); |
| 2836 | } |
| 2837 | LLVM_DEBUG(dbgs() << Pred->NodeNum << " and successor " << Succ->NodeNumdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << Pred->NodeNum << " and successor " << Succ->NodeNum << " are scheduled before node " << SU->NodeNum << "\n";; } } while (false) |
| 2838 | << " are scheduled before node " << SU->NodeNumdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << Pred->NodeNum << " and successor " << Succ->NodeNum << " are scheduled before node " << SU->NodeNum << "\n";; } } while (false) |
| 2839 | << "\n";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << Pred->NodeNum << " and successor " << Succ->NodeNum << " are scheduled before node " << SU->NodeNum << "\n";; } } while (false); |
| 2840 | } |
| 2841 | } |
| 2842 | |
| 2843 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!Valid) dbgs() << "Invalid node order found!\n" ; }; } } while (false) |
| 2844 | if (!Valid)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!Valid) dbgs() << "Invalid node order found!\n" ; }; } } while (false) |
| 2845 | dbgs() << "Invalid node order found!\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!Valid) dbgs() << "Invalid node order found!\n" ; }; } } while (false) |
| 2846 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (!Valid) dbgs() << "Invalid node order found!\n" ; }; } } while (false); |
| 2847 | } |
| 2848 | |
| 2849 | /// Attempt to fix the degenerate cases when the instruction serialization |
| 2850 | /// causes the register lifetimes to overlap. For example, |
| 2851 | /// p' = store_pi(p, b) |
| 2852 | /// = load p, offset |
| 2853 | /// In this case p and p' overlap, which means that two registers are needed. |
| 2854 | /// Instead, this function changes the load to use p' and updates the offset. |
| 2855 | void SwingSchedulerDAG::fixupRegisterOverlaps(std::deque<SUnit *> &Instrs) { |
| 2856 | unsigned OverlapReg = 0; |
| 2857 | unsigned NewBaseReg = 0; |
| 2858 | for (SUnit *SU : Instrs) { |
| 2859 | MachineInstr *MI = SU->getInstr(); |
| 2860 | for (unsigned i = 0, e = MI->getNumOperands(); i < e; ++i) { |
| 2861 | const MachineOperand &MO = MI->getOperand(i); |
| 2862 | // Look for an instruction that uses p. The instruction occurs in the |
| 2863 | // same cycle but occurs later in the serialized order. |
| 2864 | if (MO.isReg() && MO.isUse() && MO.getReg() == OverlapReg) { |
| 2865 | // Check that the instruction appears in the InstrChanges structure, |
| 2866 | // which contains instructions that can have the offset updated. |
| 2867 | DenseMap<SUnit *, std::pair<unsigned, int64_t>>::iterator It = |
| 2868 | InstrChanges.find(SU); |
| 2869 | if (It != InstrChanges.end()) { |
| 2870 | unsigned BasePos, OffsetPos; |
| 2871 | // Update the base register and adjust the offset. |
| 2872 | if (TII->getBaseAndOffsetPosition(*MI, BasePos, OffsetPos)) { |
| 2873 | MachineInstr *NewMI = MF.CloneMachineInstr(MI); |
| 2874 | NewMI->getOperand(BasePos).setReg(NewBaseReg); |
| 2875 | int64_t NewOffset = |
| 2876 | MI->getOperand(OffsetPos).getImm() - It->second.second; |
| 2877 | NewMI->getOperand(OffsetPos).setImm(NewOffset); |
| 2878 | SU->setInstr(NewMI); |
| 2879 | MISUnitMap[NewMI] = SU; |
| 2880 | NewMIs[MI] = NewMI; |
| 2881 | } |
| 2882 | } |
| 2883 | OverlapReg = 0; |
| 2884 | NewBaseReg = 0; |
| 2885 | break; |
| 2886 | } |
| 2887 | // Look for an instruction of the form p' = op(p), which uses and defines |
| 2888 | // two virtual registers that get allocated to the same physical register. |
| 2889 | unsigned TiedUseIdx = 0; |
| 2890 | if (MI->isRegTiedToUseOperand(i, &TiedUseIdx)) { |
| 2891 | // OverlapReg is p in the example above. |
| 2892 | OverlapReg = MI->getOperand(TiedUseIdx).getReg(); |
| 2893 | // NewBaseReg is p' in the example above. |
| 2894 | NewBaseReg = MI->getOperand(i).getReg(); |
| 2895 | break; |
| 2896 | } |
| 2897 | } |
| 2898 | } |
| 2899 | } |
| 2900 | |
| 2901 | /// After the schedule has been formed, call this function to combine |
| 2902 | /// the instructions from the different stages/cycles. That is, this |
| 2903 | /// function creates a schedule that represents a single iteration. |
| 2904 | void SMSchedule::finalizeSchedule(SwingSchedulerDAG *SSD) { |
| 2905 | // Move all instructions to the first stage from later stages. |
| 2906 | for (int cycle = getFirstCycle(); cycle <= getFinalCycle(); ++cycle) { |
| 2907 | for (int stage = 1, lastStage = getMaxStageCount(); stage <= lastStage; |
| 2908 | ++stage) { |
| 2909 | std::deque<SUnit *> &cycleInstrs = |
| 2910 | ScheduledInstrs[cycle + (stage * InitiationInterval)]; |
| 2911 | for (SUnit *SU : llvm::reverse(cycleInstrs)) |
| 2912 | ScheduledInstrs[cycle].push_front(SU); |
| 2913 | } |
| 2914 | } |
| 2915 | |
| 2916 | // Erase all the elements in the later stages. Only one iteration should |
| 2917 | // remain in the scheduled list, and it contains all the instructions. |
| 2918 | for (int cycle = getFinalCycle() + 1; cycle <= LastCycle; ++cycle) |
| 2919 | ScheduledInstrs.erase(cycle); |
| 2920 | |
| 2921 | // Change the registers in instruction as specified in the InstrChanges |
| 2922 | // map. We need to use the new registers to create the correct order. |
| 2923 | for (const SUnit &SU : SSD->SUnits) |
| 2924 | SSD->applyInstrChange(SU.getInstr(), *this); |
| 2925 | |
| 2926 | // Reorder the instructions in each cycle to fix and improve the |
| 2927 | // generated code. |
| 2928 | for (int Cycle = getFirstCycle(), E = getFinalCycle(); Cycle <= E; ++Cycle) { |
| 2929 | std::deque<SUnit *> &cycleInstrs = ScheduledInstrs[Cycle]; |
| 2930 | std::deque<SUnit *> newOrderPhi; |
| 2931 | for (SUnit *SU : cycleInstrs) { |
| 2932 | if (SU->getInstr()->isPHI()) |
| 2933 | newOrderPhi.push_back(SU); |
| 2934 | } |
| 2935 | std::deque<SUnit *> newOrderI; |
| 2936 | for (SUnit *SU : cycleInstrs) { |
| 2937 | if (!SU->getInstr()->isPHI()) |
| 2938 | orderDependence(SSD, SU, newOrderI); |
| 2939 | } |
| 2940 | // Replace the old order with the new order. |
| 2941 | cycleInstrs.swap(newOrderPhi); |
| 2942 | llvm::append_range(cycleInstrs, newOrderI); |
| 2943 | SSD->fixupRegisterOverlaps(cycleInstrs); |
| 2944 | } |
| 2945 | |
| 2946 | LLVM_DEBUG(dump();)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dump();; } } while (false); |
| 2947 | } |
| 2948 | |
| 2949 | void NodeSet::print(raw_ostream &os) const { |
| 2950 | os << "Num nodes " << size() << " rec " << RecMII << " mov " << MaxMOV |
| 2951 | << " depth " << MaxDepth << " col " << Colocate << "\n"; |
| 2952 | for (const auto &I : Nodes) |
| 2953 | os << " SU(" << I->NodeNum << ") " << *(I->getInstr()); |
| 2954 | os << "\n"; |
| 2955 | } |
| 2956 | |
| 2957 | #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) |
| 2958 | /// Print the schedule information to the given output. |
| 2959 | void SMSchedule::print(raw_ostream &os) const { |
| 2960 | // Iterate over each cycle. |
| 2961 | for (int cycle = getFirstCycle(); cycle <= getFinalCycle(); ++cycle) { |
| 2962 | // Iterate over each instruction in the cycle. |
| 2963 | const_sched_iterator cycleInstrs = ScheduledInstrs.find(cycle); |
| 2964 | for (SUnit *CI : cycleInstrs->second) { |
| 2965 | os << "cycle " << cycle << " (" << stageScheduled(CI) << ") "; |
| 2966 | os << "(" << CI->NodeNum << ") "; |
| 2967 | CI->getInstr()->print(os); |
| 2968 | os << "\n"; |
| 2969 | } |
| 2970 | } |
| 2971 | } |
| 2972 | |
| 2973 | /// Utility function used for debugging to print the schedule. |
| 2974 | LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void SMSchedule::dump() const { print(dbgs()); } |
| 2975 | LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void NodeSet::dump() const { print(dbgs()); } |
| 2976 | |
| 2977 | void ResourceManager::dumpMRT() const { |
| 2978 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2979 | if (UseDFA)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2980 | return;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2981 | std::stringstream SS;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2982 | SS << "MRT:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2983 | SS << std::setw(4) << "Slot";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2984 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2985 | SS << std::setw(3) << I;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2986 | SS << std::setw(7) << "#Mops"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2987 | << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2988 | for (int Slot = 0; Slot < InitiationInterval; ++Slot) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2989 | SS << std::setw(4) << Slot;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2990 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2991 | SS << std::setw(3) << MRT[Slot][I];do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2992 | SS << std::setw(7) << NumScheduledMops[Slot] << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2993 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2994 | dbgs() << SS.str();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false) |
| 2995 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (UseDFA) return; std::stringstream SS; SS << "MRT:\n"; SS << std::setw(4) << "Slot"; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << I; SS << std ::setw(7) << "#Mops" << "\n"; for (int Slot = 0; Slot < InitiationInterval; ++Slot) { SS << std::setw(4) << Slot; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) SS << std::setw(3) << MRT[Slot][I ]; SS << std::setw(7) << NumScheduledMops[Slot] << "\n"; } dbgs() << SS.str(); }; } } while (false); |
| 2996 | } |
| 2997 | #endif |
| 2998 | |
| 2999 | void ResourceManager::initProcResourceVectors( |
| 3000 | const MCSchedModel &SM, SmallVectorImpl<uint64_t> &Masks) { |
| 3001 | unsigned ProcResourceID = 0; |
| 3002 | |
| 3003 | // We currently limit the resource kinds to 64 and below so that we can use |
| 3004 | // uint64_t for Masks |
| 3005 | assert(SM.getNumProcResourceKinds() < 64 &&(static_cast <bool> (SM.getNumProcResourceKinds() < 64 && "Too many kinds of resources, unsupported") ? void (0) : __assert_fail ("SM.getNumProcResourceKinds() < 64 && \"Too many kinds of resources, unsupported\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 3006, __extension__ __PRETTY_FUNCTION__)) |
| 3006 | "Too many kinds of resources, unsupported")(static_cast <bool> (SM.getNumProcResourceKinds() < 64 && "Too many kinds of resources, unsupported") ? void (0) : __assert_fail ("SM.getNumProcResourceKinds() < 64 && \"Too many kinds of resources, unsupported\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 3006, __extension__ __PRETTY_FUNCTION__)); |
| 3007 | // Create a unique bitmask for every processor resource unit. |
| 3008 | // Skip resource at index 0, since it always references 'InvalidUnit'. |
| 3009 | Masks.resize(SM.getNumProcResourceKinds()); |
| 3010 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { |
| 3011 | const MCProcResourceDesc &Desc = *SM.getProcResource(I); |
| 3012 | if (Desc.SubUnitsIdxBegin) |
| 3013 | continue; |
| 3014 | Masks[I] = 1ULL << ProcResourceID; |
| 3015 | ProcResourceID++; |
| 3016 | } |
| 3017 | // Create a unique bitmask for every processor resource group. |
| 3018 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { |
| 3019 | const MCProcResourceDesc &Desc = *SM.getProcResource(I); |
| 3020 | if (!Desc.SubUnitsIdxBegin) |
| 3021 | continue; |
| 3022 | Masks[I] = 1ULL << ProcResourceID; |
| 3023 | for (unsigned U = 0; U < Desc.NumUnits; ++U) |
| 3024 | Masks[I] |= Masks[Desc.SubUnitsIdxBegin[U]]; |
| 3025 | ProcResourceID++; |
| 3026 | } |
| 3027 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3028 | if (SwpShowResMask) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3029 | dbgs() << "ProcResourceDesc:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3030 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3031 | const MCProcResourceDesc *ProcResource = SM.getProcResource(I);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3032 | dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n",do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3033 | ProcResource->Name, I, Masks[I],do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3034 | ProcResource->NumUnits);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3035 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3036 | dbgs() << " -----------------\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3037 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false) |
| 3038 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpShowResMask) { dbgs() << "ProcResourceDesc:\n" ; for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { const MCProcResourceDesc *ProcResource = SM.getProcResource (I); dbgs() << format(" %16s(%2d): Mask: 0x%08x, NumUnits:%2d\n" , ProcResource->Name, I, Masks[I], ProcResource->NumUnits ); } dbgs() << " -----------------\n"; } }; } } while ( false); |
| 3039 | } |
| 3040 | |
| 3041 | bool ResourceManager::canReserveResources(SUnit &SU, int Cycle) { |
| 3042 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "canReserveResources:\n" ; }; } } while (false) |
| 3043 | if (SwpDebugResource)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "canReserveResources:\n" ; }; } } while (false) |
| 3044 | dbgs() << "canReserveResources:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "canReserveResources:\n" ; }; } } while (false) |
| 3045 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "canReserveResources:\n" ; }; } } while (false); |
| 3046 | if (UseDFA) |
| 3047 | return DFAResources[positiveModulo(Cycle, InitiationInterval)] |
| 3048 | ->canReserveResources(&SU.getInstr()->getDesc()); |
| 3049 | |
| 3050 | const MCSchedClassDesc *SCDesc = DAG->getSchedClass(&SU); |
| 3051 | if (!SCDesc->isValid()) { |
| 3052 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3053 | dbgs() << "No valid Schedule Class Desc for schedClass!\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3054 | dbgs() << "isPseudo:" << SU.getInstr()->isPseudo() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3055 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false); |
| 3056 | return true; |
| 3057 | } |
| 3058 | |
| 3059 | reserveResources(SCDesc, Cycle); |
| 3060 | bool Result = !isOverbooked(); |
| 3061 | unreserveResources(SCDesc, Cycle); |
| 3062 | |
| 3063 | LLVM_DEBUG(if (SwpDebugResource) dbgs() << "return " << Result << "\n\n";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { if (SwpDebugResource) dbgs() << "return " << Result << "\n\n";; } } while (false); |
| 3064 | return Result; |
| 3065 | } |
| 3066 | |
| 3067 | void ResourceManager::reserveResources(SUnit &SU, int Cycle) { |
| 3068 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "reserveResources:\n" ; }; } } while (false) |
| 3069 | if (SwpDebugResource)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "reserveResources:\n" ; }; } } while (false) |
| 3070 | dbgs() << "reserveResources:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "reserveResources:\n" ; }; } } while (false) |
| 3071 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "reserveResources:\n" ; }; } } while (false); |
| 3072 | if (UseDFA) |
| 3073 | return DFAResources[positiveModulo(Cycle, InitiationInterval)] |
| 3074 | ->reserveResources(&SU.getInstr()->getDesc()); |
| 3075 | |
| 3076 | const MCSchedClassDesc *SCDesc = DAG->getSchedClass(&SU); |
| 3077 | if (!SCDesc->isValid()) { |
| 3078 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3079 | dbgs() << "No valid Schedule Class Desc for schedClass!\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3080 | dbgs() << "isPseudo:" << SU.getInstr()->isPseudo() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false) |
| 3081 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "No valid Schedule Class Desc for schedClass!\n" ; dbgs() << "isPseudo:" << SU.getInstr()->isPseudo () << "\n"; }; } } while (false); |
| 3082 | return; |
| 3083 | } |
| 3084 | |
| 3085 | reserveResources(SCDesc, Cycle); |
| 3086 | |
| 3087 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false) |
| 3088 | if (SwpDebugResource) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false) |
| 3089 | dumpMRT();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false) |
| 3090 | dbgs() << "reserveResources: done!\n\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false) |
| 3091 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false) |
| 3092 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { dumpMRT(); dbgs() << "reserveResources: done!\n\n"; } }; } } while (false); |
| 3093 | } |
| 3094 | |
| 3095 | void ResourceManager::reserveResources(const MCSchedClassDesc *SCDesc, |
| 3096 | int Cycle) { |
| 3097 | assert(!UseDFA)(static_cast <bool> (!UseDFA) ? void (0) : __assert_fail ("!UseDFA", "llvm/lib/CodeGen/MachinePipeliner.cpp", 3097, __extension__ __PRETTY_FUNCTION__)); |
| 3098 | for (const MCWriteProcResEntry &PRE : make_range( |
| 3099 | STI->getWriteProcResBegin(SCDesc), STI->getWriteProcResEnd(SCDesc))) |
| 3100 | for (int C = Cycle; C < Cycle + PRE.Cycles; ++C) |
| 3101 | ++MRT[positiveModulo(C, InitiationInterval)][PRE.ProcResourceIdx]; |
| 3102 | |
| 3103 | for (int C = Cycle; C < Cycle + SCDesc->NumMicroOps; ++C) |
| 3104 | ++NumScheduledMops[positiveModulo(C, InitiationInterval)]; |
| 3105 | } |
| 3106 | |
| 3107 | void ResourceManager::unreserveResources(const MCSchedClassDesc *SCDesc, |
| 3108 | int Cycle) { |
| 3109 | assert(!UseDFA)(static_cast <bool> (!UseDFA) ? void (0) : __assert_fail ("!UseDFA", "llvm/lib/CodeGen/MachinePipeliner.cpp", 3109, __extension__ __PRETTY_FUNCTION__)); |
| 3110 | for (const MCWriteProcResEntry &PRE : make_range( |
| 3111 | STI->getWriteProcResBegin(SCDesc), STI->getWriteProcResEnd(SCDesc))) |
| 3112 | for (int C = Cycle; C < Cycle + PRE.Cycles; ++C) |
| 3113 | --MRT[positiveModulo(C, InitiationInterval)][PRE.ProcResourceIdx]; |
| 3114 | |
| 3115 | for (int C = Cycle; C < Cycle + SCDesc->NumMicroOps; ++C) |
| 3116 | --NumScheduledMops[positiveModulo(C, InitiationInterval)]; |
| 3117 | } |
| 3118 | |
| 3119 | bool ResourceManager::isOverbooked() const { |
| 3120 | assert(!UseDFA)(static_cast <bool> (!UseDFA) ? void (0) : __assert_fail ("!UseDFA", "llvm/lib/CodeGen/MachinePipeliner.cpp", 3120, __extension__ __PRETTY_FUNCTION__)); |
| 3121 | for (int Slot = 0; Slot < InitiationInterval; ++Slot) { |
| 3122 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { |
| 3123 | const MCProcResourceDesc *Desc = SM.getProcResource(I); |
| 3124 | if (MRT[Slot][I] > Desc->NumUnits) |
| 3125 | return true; |
| 3126 | } |
| 3127 | if (NumScheduledMops[Slot] > IssueWidth) |
| 3128 | return true; |
| 3129 | } |
| 3130 | return false; |
| 3131 | } |
| 3132 | |
| 3133 | int ResourceManager::calculateResMIIDFA() const { |
| 3134 | assert(UseDFA)(static_cast <bool> (UseDFA) ? void (0) : __assert_fail ("UseDFA", "llvm/lib/CodeGen/MachinePipeliner.cpp", 3134, __extension__ __PRETTY_FUNCTION__)); |
| 3135 | |
| 3136 | // Sort the instructions by the number of available choices for scheduling, |
| 3137 | // least to most. Use the number of critical resources as the tie breaker. |
| 3138 | FuncUnitSorter FUS = FuncUnitSorter(*ST); |
| 3139 | for (SUnit &SU : DAG->SUnits) |
| 3140 | FUS.calcCriticalResources(*SU.getInstr()); |
| 3141 | PriorityQueue<MachineInstr *, std::vector<MachineInstr *>, FuncUnitSorter> |
| 3142 | FuncUnitOrder(FUS); |
| 3143 | |
| 3144 | for (SUnit &SU : DAG->SUnits) |
| 3145 | FuncUnitOrder.push(SU.getInstr()); |
| 3146 | |
| 3147 | SmallVector<std::unique_ptr<DFAPacketizer>, 8> Resources; |
| 3148 | Resources.push_back( |
| 3149 | std::unique_ptr<DFAPacketizer>(TII->CreateTargetScheduleState(*ST))); |
| 3150 | |
| 3151 | while (!FuncUnitOrder.empty()) { |
| 3152 | MachineInstr *MI = FuncUnitOrder.top(); |
| 3153 | FuncUnitOrder.pop(); |
| 3154 | if (TII->isZeroCost(MI->getOpcode())) |
| 3155 | continue; |
| 3156 | |
| 3157 | // Attempt to reserve the instruction in an existing DFA. At least one |
| 3158 | // DFA is needed for each cycle. |
| 3159 | unsigned NumCycles = DAG->getSUnit(MI)->Latency; |
| 3160 | unsigned ReservedCycles = 0; |
| 3161 | auto *RI = Resources.begin(); |
| 3162 | auto *RE = Resources.end(); |
| 3163 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to reserve resource for " << NumCycles << " cycles for \n"; MI->dump(); }; } } while (false) |
| 3164 | dbgs() << "Trying to reserve resource for " << NumCyclesdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to reserve resource for " << NumCycles << " cycles for \n"; MI->dump(); }; } } while (false) |
| 3165 | << " cycles for \n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to reserve resource for " << NumCycles << " cycles for \n"; MI->dump(); }; } } while (false) |
| 3166 | MI->dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to reserve resource for " << NumCycles << " cycles for \n"; MI->dump(); }; } } while (false) |
| 3167 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { dbgs() << "Trying to reserve resource for " << NumCycles << " cycles for \n"; MI->dump(); }; } } while (false); |
| 3168 | for (unsigned C = 0; C < NumCycles; ++C) |
| 3169 | while (RI != RE) { |
| 3170 | if ((*RI)->canReserveResources(*MI)) { |
| 3171 | (*RI)->reserveResources(*MI); |
| 3172 | ++ReservedCycles; |
| 3173 | break; |
| 3174 | } |
| 3175 | RI++; |
| 3176 | } |
| 3177 | LLVM_DEBUG(dbgs() << "ReservedCycles:" << ReservedCyclesdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "ReservedCycles:" << ReservedCycles << ", NumCycles:" << NumCycles << "\n"; } } while (false) |
| 3178 | << ", NumCycles:" << NumCycles << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "ReservedCycles:" << ReservedCycles << ", NumCycles:" << NumCycles << "\n"; } } while (false); |
| 3179 | // Add new DFAs, if needed, to reserve resources. |
| 3180 | for (unsigned C = ReservedCycles; C < NumCycles; ++C) { |
| 3181 | LLVM_DEBUG(if (SwpDebugResource) dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { if (SwpDebugResource) dbgs() << "NewResource created to reserve resources" << "\n"; } } while (false) |
| 3182 | << "NewResource created to reserve resources"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { if (SwpDebugResource) dbgs() << "NewResource created to reserve resources" << "\n"; } } while (false) |
| 3183 | << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { if (SwpDebugResource) dbgs() << "NewResource created to reserve resources" << "\n"; } } while (false); |
| 3184 | auto *NewResource = TII->CreateTargetScheduleState(*ST); |
| 3185 | assert(NewResource->canReserveResources(*MI) && "Reserve error.")(static_cast <bool> (NewResource->canReserveResources (*MI) && "Reserve error.") ? void (0) : __assert_fail ("NewResource->canReserveResources(*MI) && \"Reserve error.\"" , "llvm/lib/CodeGen/MachinePipeliner.cpp", 3185, __extension__ __PRETTY_FUNCTION__)); |
| 3186 | NewResource->reserveResources(*MI); |
| 3187 | Resources.push_back(std::unique_ptr<DFAPacketizer>(NewResource)); |
| 3188 | } |
| 3189 | } |
| 3190 | |
| 3191 | int Resmii = Resources.size(); |
| 3192 | LLVM_DEBUG(dbgs() << "Return Res MII:" << Resmii << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { dbgs() << "Return Res MII:" << Resmii << "\n"; } } while (false); |
| 3193 | return Resmii; |
| 3194 | } |
| 3195 | |
| 3196 | int ResourceManager::calculateResMII() const { |
| 3197 | if (UseDFA) |
| 3198 | return calculateResMIIDFA(); |
| 3199 | |
| 3200 | // Count each resource consumption and divide it by the number of units. |
| 3201 | // ResMII is the max value among them. |
| 3202 | |
| 3203 | int NumMops = 0; |
| 3204 | SmallVector<uint64_t> ResourceCount(SM.getNumProcResourceKinds()); |
| 3205 | for (SUnit &SU : DAG->SUnits) { |
| 3206 | if (TII->isZeroCost(SU.getInstr()->getOpcode())) |
| 3207 | continue; |
| 3208 | |
| 3209 | const MCSchedClassDesc *SCDesc = DAG->getSchedClass(&SU); |
| 3210 | if (!SCDesc->isValid()) |
| 3211 | continue; |
| 3212 | |
| 3213 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3214 | if (SwpDebugResource) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3215 | DAG->dumpNode(SU);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3216 | dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3217 | << " WriteProcRes: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3218 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ) |
| 3219 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { DAG->dumpNode(SU ); dbgs() << " #Mops: " << SCDesc->NumMicroOps << "\n" << " WriteProcRes: "; } }; } } while (false ); |
| 3220 | NumMops += SCDesc->NumMicroOps; |
| 3221 | for (const MCWriteProcResEntry &PRE : |
| 3222 | make_range(STI->getWriteProcResBegin(SCDesc), |
| 3223 | STI->getWriteProcResEnd(SCDesc))) { |
| 3224 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3225 | if (SwpDebugResource) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3226 | const MCProcResourceDesc *Desc =do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3227 | SM.getProcResource(PRE.ProcResourceIdx);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3228 | dbgs() << Desc->Name << ": " << PRE.Cycles << ", ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3229 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false) |
| 3230 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { const MCProcResourceDesc *Desc = SM.getProcResource(PRE.ProcResourceIdx); dbgs() << Desc->Name << ": " << PRE.Cycles << ", " ; } }; } } while (false); |
| 3231 | ResourceCount[PRE.ProcResourceIdx] += PRE.Cycles; |
| 3232 | } |
| 3233 | LLVM_DEBUG(if (SwpDebugResource) dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { if (SwpDebugResource) dbgs() << "\n"; } } while (false); |
| 3234 | } |
| 3235 | |
| 3236 | int Result = (NumMops + IssueWidth - 1) / IssueWidth; |
| 3237 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false) |
| 3238 | if (SwpDebugResource)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false) |
| 3239 | dbgs() << "#Mops: " << NumMops << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false) |
| 3240 | << "IssueWidth: " << IssueWidth << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false) |
| 3241 | << "Cycles: " << Result << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false) |
| 3242 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) dbgs() << "#Mops: " << NumMops << ", " << "IssueWidth: " << IssueWidth << ", " << "Cycles: " << Result << "\n"; }; } } while (false); |
| 3243 | |
| 3244 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3245 | if (SwpDebugResource) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3246 | std::stringstream SS;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3247 | SS << std::setw(2) << "ID" << std::setw(16) << "Name" << std::setw(10)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3248 | << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3249 | << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3250 | dbgs() << SS.str();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3251 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false) |
| 3252 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << "ID" << std::setw(16 ) << "Name" << std::setw(10) << "Units" << std::setw(10) << "Consumed" << std::setw(10) << "Cycles" << "\n"; dbgs() << SS.str(); } }; } } while (false); |
| 3253 | for (unsigned I = 1, E = SM.getNumProcResourceKinds(); I < E; ++I) { |
| 3254 | const MCProcResourceDesc *Desc = SM.getProcResource(I); |
| 3255 | int Cycles = (ResourceCount[I] + Desc->NumUnits - 1) / Desc->NumUnits; |
| 3256 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3257 | if (SwpDebugResource) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3258 | std::stringstream SS;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3259 | SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3260 | << Desc->NumUnits << std::setw(10) << ResourceCount[I]do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3261 | << std::setw(10) << Cycles << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3262 | dbgs() << SS.str();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3263 | }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false) |
| 3264 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("pipeliner")) { { if (SwpDebugResource) { std::stringstream SS ; SS << std::setw(2) << I << std::setw(16) << Desc->Name << std::setw(10) << Desc->NumUnits << std::setw(10) << ResourceCount[I] << std ::setw(10) << Cycles << "\n"; dbgs() << SS. str(); } }; } } while (false); |
| 3265 | if (Cycles > Result) |
| 3266 | Result = Cycles; |
| 3267 | } |
| 3268 | return Result; |
| 3269 | } |
| 3270 | |
| 3271 | void ResourceManager::init(int II) { |
| 3272 | InitiationInterval = II; |
| 3273 | DFAResources.clear(); |
| 3274 | DFAResources.resize(II); |
| 3275 | for (auto &I : DFAResources) |
| 3276 | I.reset(ST->getInstrInfo()->CreateTargetScheduleState(*ST)); |
| 3277 | MRT.clear(); |
| 3278 | MRT.resize(II, SmallVector<uint64_t>(SM.getNumProcResourceKinds())); |
| 3279 | NumScheduledMops.clear(); |
| 3280 | NumScheduledMops.resize(II); |
| 3281 | } |