lib/Transforms/Instrumentation/RSProfiling.cpp

   1 //===- RSProfiling.cpp - Various profiling using random sampling ----------===//
   2 //
   3 //                      The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // These passes implement a random sampling based profiling.  Different methods
  11 // of choosing when to sample are supported, as well as different types of
  12 // profiling.  This is done as two passes.  The first is a sequence of profiling
  13 // passes which insert profiling into the program, and remember what they
  14 // inserted.
  15 //
  16 // The second stage duplicates all instructions in a function, ignoring the
  17 // profiling code, then connects the two versions togeather at the entry and at
  18 // backedges.  At each connection point a choice is made as to whether to jump
  19 // to the profiled code (take a sample) or execute the unprofiled code.
  20 //
  21 // It is highly recommended that after this pass one runs mem2reg and adce
  22 // (instcombine load-vn gdce dse also are good to run afterwards)
  23 //
  24 // This design is intended to make the profiling passes independent of the RS
  25 // framework, but any profiling pass that implements the RSProfiling interface
  26 // is compatible with the rs framework (and thus can be sampled)
  27 //
  28 // TODO: obviously the block and function profiling are almost identical to the
  29 // existing ones, so they can be unified (esp since these passes are valid
  30 // without the rs framework).
  31 // TODO: Fix choice code so that frequency is not hard coded
  32 //
  33 //===----------------------------------------------------------------------===//
  34
  35 #include "llvm/Pass.h"
  36 #include "llvm/LLVMContext.h"
  37 #include "llvm/Module.h"
  38 #include "llvm/Instructions.h"
  39 #include "llvm/Constants.h"
  40 #include "llvm/DerivedTypes.h"
  41 #include "llvm/Intrinsics.h"
  42 #include "llvm/Transforms/Scalar.h"
  43 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
  44 #include "llvm/Support/CommandLine.h"
  45 #include "llvm/Support/Compiler.h"
  46 #include "llvm/Support/Debug.h"
  47 #include "llvm/Support/ErrorHandling.h"
  48 #include "llvm/Support/raw_ostream.h"
  49 #include "llvm/Transforms/Instrumentation.h"
  50 #include "RSProfiling.h"
  51 #include <set>
  52 #include <map>
  53 #include <queue>
  54 using namespace llvm;
  55
  56 namespace {
  57   enum RandomMeth {
  58     GBV, GBVO, HOSTCC
  59   };
  60 }
  61
  62 static cl::opt<RandomMeth> RandomMethod("profile-randomness",
  63     cl::desc("How to randomly choose to profile:"),
  64     cl::values(
  65                clEnumValN(GBV, "global", "global counter"),
  66                clEnumValN(GBVO, "ra_global",
  67                           "register allocated global counter"),
  68                clEnumValN(HOSTCC, "rdcc", "cycle counter"),
  69                clEnumValEnd));
  70
  71 namespace {
  72   /// NullProfilerRS - The basic profiler that does nothing.  It is the default
  73   /// profiler and thus terminates RSProfiler chains.  It is useful for
  74   /// measuring framework overhead
  75   class VISIBILITY_HIDDEN NullProfilerRS : public RSProfilers {
  76   public:
  77     static char ID; // Pass identification, replacement for typeid
  78     bool isProfiling(Value* v) {
  79       return false;
  80     }
  81     bool runOnModule(Module &M) {
  82       return false;
  83     }
  84     void getAnalysisUsage(AnalysisUsage &AU) const {
  85       AU.setPreservesAll();
  86     }
  87   };
  88 }
  89
  90 static RegisterAnalysisGroup<RSProfilers> A("Profiling passes");
  91 static RegisterPass<NullProfilerRS> NP("insert-null-profiling-rs",
  92                                        "Measure profiling framework overhead");
  93 static RegisterAnalysisGroup<RSProfilers, true> NPT(NP);
  94
  95 namespace {
  96   /// Chooser - Something that chooses when to make a sample of the profiled code
  97   class VISIBILITY_HIDDEN Chooser {
  98   public:
  99     /// ProcessChoicePoint - is called for each basic block inserted to choose
 100     /// between normal and sample code
 101     virtual void ProcessChoicePoint(BasicBlock*) = 0;
 102     /// PrepFunction - is called once per function before other work is done.
 103     /// This gives the opertunity to insert new allocas and such.
 104     virtual void PrepFunction(Function*) = 0;
 105     virtual ~Chooser() {}
 106   };
 107
 108   //Things that implement sampling policies
 109   //A global value that is read-mod-stored to choose when to sample.
 110   //A sample is taken when the global counter hits 0
 111   class VISIBILITY_HIDDEN GlobalRandomCounter : public Chooser {
 112     GlobalVariable* Counter;
 113     Value* ResetValue;
 114     const IntegerType* T;
 115   public:
 116     GlobalRandomCounter(Module& M, const IntegerType* t, uint64_t resetval);
 117     virtual ~GlobalRandomCounter();
 118     virtual void PrepFunction(Function* F);
 119     virtual void ProcessChoicePoint(BasicBlock* bb);
 120   };
 121
 122   //Same is GRC, but allow register allocation of the global counter
 123   class VISIBILITY_HIDDEN GlobalRandomCounterOpt : public Chooser {
 124     GlobalVariable* Counter;
 125     Value* ResetValue;
 126     AllocaInst* AI;
 127     const IntegerType* T;
 128   public:
 129     GlobalRandomCounterOpt(Module& M, const IntegerType* t, uint64_t resetval);
 130     virtual ~GlobalRandomCounterOpt();
 131     virtual void PrepFunction(Function* F);
 132     virtual void ProcessChoicePoint(BasicBlock* bb);
 133   };
 134
 135   //Use the cycle counter intrinsic as a source of pseudo randomness when
 136   //deciding when to sample.
 137   class VISIBILITY_HIDDEN CycleCounter : public Chooser {
 138     uint64_t rm;
 139     Constant *F;
 140   public:
 141     CycleCounter(Module& m, uint64_t resetmask);
 142     virtual ~CycleCounter();
 143     virtual void PrepFunction(Function* F);
 144     virtual void ProcessChoicePoint(BasicBlock* bb);
 145   };
 146
 147   /// ProfilerRS - Insert the random sampling framework
 148   struct VISIBILITY_HIDDEN ProfilerRS : public FunctionPass {
 149     static char ID; // Pass identification, replacement for typeid
 150     ProfilerRS() : FunctionPass(&ID) {}
 151
 152     std::map<Value*, Value*> TransCache;
 153     std::set<BasicBlock*> ChoicePoints;
 154     Chooser* c;
 155
 156     //Translate and duplicate values for the new profile free version of stuff
 157     Value* Translate(Value* v);
 158     //Duplicate an entire function (with out profiling)
 159     void Duplicate(Function& F, RSProfilers& LI);
 160     //Called once for each backedge, handle the insertion of choice points and
 161     //the interconection of the two versions of the code
 162     void ProcessBackEdge(BasicBlock* src, BasicBlock* dst, Function& F);
 163     bool runOnFunction(Function& F);
 164     bool doInitialization(Module &M);
 165     virtual void getAnalysisUsage(AnalysisUsage &AU) const;
 166   };
 167 }
 168
 169 static RegisterPass<ProfilerRS>
 170 X("insert-rs-profiling-framework",
 171   "Insert random sampling instrumentation framework");
 172
 173 char RSProfilers::ID = 0;
 174 char NullProfilerRS::ID = 0;
 175 char ProfilerRS::ID = 0;
 176
 177 //Local utilities
 178 static void ReplacePhiPred(BasicBlock* btarget,
 179                            BasicBlock* bold, BasicBlock* bnew);
 180
 181 static void CollapsePhi(BasicBlock* btarget, BasicBlock* bsrc);
 182
 183 template<class T>
 184 static void recBackEdge(BasicBlock* bb, T& BackEdges,
 185                         std::map<BasicBlock*, int>& color,
 186                         std::map<BasicBlock*, int>& depth,
 187                         std::map<BasicBlock*, int>& finish,
 188                         int& time);
 189
 190 //find the back edges and where they go to
 191 template<class T>
 192 static void getBackEdges(Function& F, T& BackEdges);
 193
 194
 195 ///////////////////////////////////////
 196 // Methods of choosing when to profile
 197 ///////////////////////////////////////
 198
 199 GlobalRandomCounter::GlobalRandomCounter(Module& M, const IntegerType* t,
 200                                          uint64_t resetval) : T(t) {
 201   ConstantInt* Init = ConstantInt::get(T, resetval);
 202   ResetValue = Init;
 203   Counter = new GlobalVariable(M, T, false, GlobalValue::InternalLinkage,
 204                                Init, "RandomSteeringCounter");
 205 }
 206
 207 GlobalRandomCounter::~GlobalRandomCounter() {}
 208
 209 void GlobalRandomCounter::PrepFunction(Function* F) {}
 210
 211 void GlobalRandomCounter::ProcessChoicePoint(BasicBlock* bb) {
 212   BranchInst* t = cast<BranchInst>(bb->getTerminator());
 213
 214   //decrement counter
 215   LoadInst* l = new LoadInst(Counter, "counter", t);
 216
 217   ICmpInst* s = new ICmpInst(t, ICmpInst::ICMP_EQ, l,
 218                              ConstantInt::get(T, 0),
 219                              "countercc");
 220
 221   Value* nv = BinaryOperator::CreateSub(l, ConstantInt::get(T, 1),
 222                                         "counternew", t);
 223   new StoreInst(nv, Counter, t);
 224   t->setCondition(s);
 225
 226   //reset counter
 227   BasicBlock* oldnext = t->getSuccessor(0);
 228   BasicBlock* resetblock = BasicBlock::Create("reset", oldnext->getParent(),
 229                                               oldnext);
 230   TerminatorInst* t2 = BranchInst::Create(oldnext, resetblock);
 231   t->setSuccessor(0, resetblock);
 232   new StoreInst(ResetValue, Counter, t2);
 233   ReplacePhiPred(oldnext, bb, resetblock);
 234 }
 235
 236 GlobalRandomCounterOpt::GlobalRandomCounterOpt(Module& M, const IntegerType* t,
 237                                                uint64_t resetval)
 238   : AI(0), T(t) {
 239   ConstantInt* Init = ConstantInt::get(T, resetval);
 240   ResetValue  = Init;
 241   Counter = new GlobalVariable(M, T, false, GlobalValue::InternalLinkage,
 242                                Init, "RandomSteeringCounter");
 243 }
 244
 245 GlobalRandomCounterOpt::~GlobalRandomCounterOpt() {}
 246
 247 void GlobalRandomCounterOpt::PrepFunction(Function* F) {
 248   //make a local temporary to cache the global
 249   BasicBlock& bb = F->getEntryBlock();
 250   BasicBlock::iterator InsertPt = bb.begin();
 251   AI = new AllocaInst(T, 0, "localcounter", InsertPt);
 252   LoadInst* l = new LoadInst(Counter, "counterload", InsertPt);
 253   new StoreInst(l, AI, InsertPt);
 254
 255   //modify all functions and return values to restore the local variable to/from
 256   //the global variable
 257   for(Function::iterator fib = F->begin(), fie = F->end();
 258       fib != fie; ++fib)
 259     for(BasicBlock::iterator bib = fib->begin(), bie = fib->end();
 260         bib != bie; ++bib)
 261       if (isa<CallInst>(bib)) {
 262         LoadInst* l = new LoadInst(AI, "counter", bib);
 263         new StoreInst(l, Counter, bib);
 264         l = new LoadInst(Counter, "counter", ++bib);
 265         new StoreInst(l, AI, bib--);
 266       } else if (isa<InvokeInst>(bib)) {
 267         LoadInst* l = new LoadInst(AI, "counter", bib);
 268         new StoreInst(l, Counter, bib);
 269
 270         BasicBlock* bb = cast<InvokeInst>(bib)->getNormalDest();
 271         BasicBlock::iterator i = bb->getFirstNonPHI();
 272         l = new LoadInst(Counter, "counter", i);
 273
 274         bb = cast<InvokeInst>(bib)->getUnwindDest();
 275         i = bb->getFirstNonPHI();
 276         l = new LoadInst(Counter, "counter", i);
 277         new StoreInst(l, AI, i);
 278       } else if (isa<UnwindInst>(&*bib) || isa<ReturnInst>(&*bib)) {
 279         LoadInst* l = new LoadInst(AI, "counter", bib);
 280         new StoreInst(l, Counter, bib);
 281       }
 282 }
 283
 284 void GlobalRandomCounterOpt::ProcessChoicePoint(BasicBlock* bb) {
 285   BranchInst* t = cast<BranchInst>(bb->getTerminator());
 286
 287   //decrement counter
 288   LoadInst* l = new LoadInst(AI, "counter", t);
 289
 290   ICmpInst* s = new ICmpInst(t, ICmpInst::ICMP_EQ, l,
 291                              ConstantInt::get(T, 0),
 292                              "countercc");
 293
 294   Value* nv = BinaryOperator::CreateSub(l, ConstantInt::get(T, 1),
 295                                         "counternew", t);
 296   new StoreInst(nv, AI, t);
 297   t->setCondition(s);
 298
 299   //reset counter
 300   BasicBlock* oldnext = t->getSuccessor(0);
 301   BasicBlock* resetblock = BasicBlock::Create("reset", oldnext->getParent(),
 302                                               oldnext);
 303   TerminatorInst* t2 = BranchInst::Create(oldnext, resetblock);
 304   t->setSuccessor(0, resetblock);
 305   new StoreInst(ResetValue, AI, t2);
 306   ReplacePhiPred(oldnext, bb, resetblock);
 307 }
 308
 309
 310 CycleCounter::CycleCounter(Module& m, uint64_t resetmask) : rm(resetmask) {
 311   F = Intrinsic::getDeclaration(&m, Intrinsic::readcyclecounter);
 312 }
 313
 314 CycleCounter::~CycleCounter() {}
 315
 316 void CycleCounter::PrepFunction(Function* F) {}
 317
 318 void CycleCounter::ProcessChoicePoint(BasicBlock* bb) {
 319   BranchInst* t = cast<BranchInst>(bb->getTerminator());
 320
 321   CallInst* c = CallInst::Create(F, "rdcc", t);
 322   BinaryOperator* b =
 323     BinaryOperator::CreateAnd(c, ConstantInt::get(Type::Int64Ty, rm),
 324                               "mrdcc", t);
 325
 326   ICmpInst *s = new ICmpInst(t, ICmpInst::ICMP_EQ, b,
 327                              ConstantInt::get(Type::Int64Ty, 0),
 328                              "mrdccc");
 329
 330   t->setCondition(s);
 331 }
 332
 333 ///////////////////////////////////////
 334 // Profiling:
 335 ///////////////////////////////////////
 336 bool RSProfilers_std::isProfiling(Value* v) {
 337   if (profcode.find(v) != profcode.end())
 338     return true;
 339   //else
 340   RSProfilers& LI = getAnalysis<RSProfilers>();
 341   return LI.isProfiling(v);
 342 }
 343
 344 void RSProfilers_std::IncrementCounterInBlock(BasicBlock *BB, unsigned CounterNum,
 345                                           GlobalValue *CounterArray) {
 346   // Insert the increment after any alloca or PHI instructions...
 347   BasicBlock::iterator InsertPos = BB->getFirstNonPHI();
 348   while (isa<AllocaInst>(InsertPos))
 349     ++InsertPos;
 350
 351   // Create the getelementptr constant expression
 352   std::vector<Constant*> Indices(2);
 353   Indices[0] = BB->getContext().getNullValue(Type::Int32Ty);
 354   Indices[1] = ConstantInt::get(Type::Int32Ty, CounterNum);
 355   Constant *ElementPtr =ConstantExpr::getGetElementPtr(CounterArray,
 356                                                         &Indices[0], 2);
 357
 358   // Load, increment and store the value back.
 359   Value *OldVal = new LoadInst(ElementPtr, "OldCounter", InsertPos);
 360   profcode.insert(OldVal);
 361   Value *NewVal = BinaryOperator::CreateAdd(OldVal,
 362                               ConstantInt::get(Type::Int32Ty, 1),
 363                                             "NewCounter", InsertPos);
 364   profcode.insert(NewVal);
 365   profcode.insert(new StoreInst(NewVal, ElementPtr, InsertPos));
 366 }
 367
 368 void RSProfilers_std::getAnalysisUsage(AnalysisUsage &AU) const {
 369   //grab any outstanding profiler, or get the null one
 370   AU.addRequired<RSProfilers>();
 371 }
 372
 373 ///////////////////////////////////////
 374 // RS Framework
 375 ///////////////////////////////////////
 376
 377 Value* ProfilerRS::Translate(Value* v) {
 378   if(TransCache[v])
 379     return TransCache[v];
 380
 381   if (BasicBlock* bb = dyn_cast<BasicBlock>(v)) {
 382     if (bb == &bb->getParent()->getEntryBlock())
 383       TransCache[bb] = bb; //don't translate entry block
 384     else
 385       TransCache[bb] = BasicBlock::Create("dup_" + bb->getName(),
 386                                           bb->getParent(), NULL);
 387     return TransCache[bb];
 388   } else if (Instruction* i = dyn_cast<Instruction>(v)) {
 389     //we have already translated this
 390     //do not translate entry block allocas
 391     if(&i->getParent()->getParent()->getEntryBlock() == i->getParent()) {
 392       TransCache[i] = i;
 393       return i;
 394     } else {
 395       //translate this
 396       Instruction* i2 = i->clone(v->getContext());
 397       if (i->hasName())
 398         i2->setName("dup_" + i->getName());
 399       TransCache[i] = i2;
 400       //NumNewInst++;
 401       for (unsigned x = 0; x < i2->getNumOperands(); ++x)
 402         i2->setOperand(x, Translate(i2->getOperand(x)));
 403       return i2;
 404     }
 405   } else if (isa<Function>(v) || isa<Constant>(v) || isa<Argument>(v)) {
 406     TransCache[v] = v;
 407     return v;
 408   }
 409   llvm_unreachable("Value not handled");
 410   return 0;
 411 }
 412
 413 void ProfilerRS::Duplicate(Function& F, RSProfilers& LI)
 414 {
 415   //perform a breadth first search, building up a duplicate of the code
 416   std::queue<BasicBlock*> worklist;
 417   std::set<BasicBlock*> seen;
 418
 419   //This loop ensures proper BB order, to help performance
 420   for (Function::iterator fib = F.begin(), fie = F.end(); fib != fie; ++fib)
 421     worklist.push(fib);
 422   while (!worklist.empty()) {
 423     Translate(worklist.front());
 424     worklist.pop();
 425   }
 426
 427   //remember than reg2mem created a new entry block we don't want to duplicate
 428   worklist.push(F.getEntryBlock().getTerminator()->getSuccessor(0));
 429   seen.insert(&F.getEntryBlock());
 430
 431   while (!worklist.empty()) {
 432     BasicBlock* bb = worklist.front();
 433     worklist.pop();
 434     if(seen.find(bb) == seen.end()) {
 435       BasicBlock* bbtarget = cast<BasicBlock>(Translate(bb));
 436       BasicBlock::InstListType& instlist = bbtarget->getInstList();
 437       for (BasicBlock::iterator iib = bb->begin(), iie = bb->end();
 438            iib != iie; ++iib) {
 439         //NumOldInst++;
 440         if (!LI.isProfiling(&*iib)) {
 441           Instruction* i = cast<Instruction>(Translate(iib));
 442           instlist.insert(bbtarget->end(), i);
 443         }
 444       }
 445       //updated search state;
 446       seen.insert(bb);
 447       TerminatorInst* ti = bb->getTerminator();
 448       for (unsigned x = 0; x < ti->getNumSuccessors(); ++x) {
 449         BasicBlock* bbs = ti->getSuccessor(x);
 450         if (seen.find(bbs) == seen.end()) {
 451           worklist.push(bbs);
 452         }
 453       }
 454     }
 455   }
 456 }
 457
 458 void ProfilerRS::ProcessBackEdge(BasicBlock* src, BasicBlock* dst, Function& F) {
 459   //given a backedge from B -> A, and translations A' and B',
 460   //a: insert C and C'
 461   //b: add branches in C to A and A' and in C' to A and A'
 462   //c: mod terminators@B, replace A with C
 463   //d: mod terminators@B', replace A' with C'
 464   //e: mod phis@A for pred B to be pred C
 465   //       if multiple entries, simplify to one
 466   //f: mod phis@A' for pred B' to be pred C'
 467   //       if multiple entries, simplify to one
 468   //g: for all phis@A with pred C using x
 469   //       add in edge from C' using x'
 470   //       add in edge from C using x in A'
 471
 472   //a:
 473   Function::iterator BBN = src; ++BBN;
 474   BasicBlock* bbC = BasicBlock::Create("choice", &F, BBN);
 475   //ChoicePoints.insert(bbC);
 476   BBN = cast<BasicBlock>(Translate(src));
 477   BasicBlock* bbCp = BasicBlock::Create("choice", &F, ++BBN);
 478   ChoicePoints.insert(bbCp);
 479
 480   //b:
 481   BranchInst::Create(cast<BasicBlock>(Translate(dst)), bbC);
 482   BranchInst::Create(dst, cast<BasicBlock>(Translate(dst)),
 483                      ConstantInt::get(Type::Int1Ty, true), bbCp);
 484   //c:
 485   {
 486     TerminatorInst* iB = src->getTerminator();
 487     for (unsigned x = 0; x < iB->getNumSuccessors(); ++x)
 488       if (iB->getSuccessor(x) == dst)
 489         iB->setSuccessor(x, bbC);
 490   }
 491   //d:
 492   {
 493     TerminatorInst* iBp = cast<TerminatorInst>(Translate(src->getTerminator()));
 494     for (unsigned x = 0; x < iBp->getNumSuccessors(); ++x)
 495       if (iBp->getSuccessor(x) == cast<BasicBlock>(Translate(dst)))
 496         iBp->setSuccessor(x, bbCp);
 497   }
 498   //e:
 499   ReplacePhiPred(dst, src, bbC);
 500   //src could be a switch, in which case we are replacing several edges with one
 501   //thus collapse those edges int the Phi
 502   CollapsePhi(dst, bbC);
 503   //f:
 504   ReplacePhiPred(cast<BasicBlock>(Translate(dst)),
 505                  cast<BasicBlock>(Translate(src)),bbCp);
 506   CollapsePhi(cast<BasicBlock>(Translate(dst)), bbCp);
 507   //g:
 508   for(BasicBlock::iterator ib = dst->begin(), ie = dst->end(); ib != ie;
 509       ++ib)
 510     if (PHINode* phi = dyn_cast<PHINode>(&*ib)) {
 511       for(unsigned x = 0; x < phi->getNumIncomingValues(); ++x)
 512         if(bbC == phi->getIncomingBlock(x)) {
 513           phi->addIncoming(Translate(phi->getIncomingValue(x)), bbCp);
 514           cast<PHINode>(Translate(phi))->addIncoming(phi->getIncomingValue(x),
 515                                                      bbC);
 516         }
 517       phi->removeIncomingValue(bbC);
 518     }
 519 }
 520
 521 bool ProfilerRS::runOnFunction(Function& F) {
 522   if (!F.isDeclaration()) {
 523     std::set<std::pair<BasicBlock*, BasicBlock*> > BackEdges;
 524     RSProfilers& LI = getAnalysis<RSProfilers>();
 525
 526     getBackEdges(F, BackEdges);
 527     Duplicate(F, LI);
 528     //assume that stuff worked.  now connect the duplicated basic blocks
 529     //with the originals in such a way as to preserve ssa.  yuk!
 530     for (std::set<std::pair<BasicBlock*, BasicBlock*> >::iterator
 531            ib = BackEdges.begin(), ie = BackEdges.end(); ib != ie; ++ib)
 532       ProcessBackEdge(ib->first, ib->second, F);
 533
 534     //oh, and add the edge from the reg2mem created entry node to the
 535     //duplicated second node
 536     TerminatorInst* T = F.getEntryBlock().getTerminator();
 537     ReplaceInstWithInst(T, BranchInst::Create(T->getSuccessor(0),
 538                                               cast<BasicBlock>(
 539                                                 Translate(T->getSuccessor(0))),
 540                                     ConstantInt::get(Type::Int1Ty, true)));
 541
 542     //do whatever is needed now that the function is duplicated
 543     c->PrepFunction(&F);
 544
 545     //add entry node to choice points
 546     ChoicePoints.insert(&F.getEntryBlock());
 547
 548     for (std::set<BasicBlock*>::iterator
 549            ii = ChoicePoints.begin(), ie = ChoicePoints.end(); ii != ie; ++ii)
 550       c->ProcessChoicePoint(*ii);
 551
 552     ChoicePoints.clear();
 553     TransCache.clear();
 554
 555     return true;
 556   }
 557   return false;
 558 }
 559
 560 bool ProfilerRS::doInitialization(Module &M) {
 561   switch (RandomMethod) {
 562   case GBV:
 563     c = new GlobalRandomCounter(M, Type::Int32Ty, (1 << 14) - 1);
 564     break;
 565   case GBVO:
 566     c = new GlobalRandomCounterOpt(M, Type::Int32Ty, (1 << 14) - 1);
 567     break;
 568   case HOSTCC:
 569     c = new CycleCounter(M, (1 << 14) - 1);
 570     break;
 571   };
 572   return true;
 573 }
 574
 575 void ProfilerRS::getAnalysisUsage(AnalysisUsage &AU) const {
 576   AU.addRequired<RSProfilers>();
 577   AU.addRequiredID(DemoteRegisterToMemoryID);
 578 }
 579
 580 ///////////////////////////////////////
 581 // Utilities:
 582 ///////////////////////////////////////
 583 static void ReplacePhiPred(BasicBlock* btarget,
 584                            BasicBlock* bold, BasicBlock* bnew) {
 585   for(BasicBlock::iterator ib = btarget->begin(), ie = btarget->end();
 586       ib != ie; ++ib)
 587     if (PHINode* phi = dyn_cast<PHINode>(&*ib)) {
 588       for(unsigned x = 0; x < phi->getNumIncomingValues(); ++x)
 589         if(bold == phi->getIncomingBlock(x))
 590           phi->setIncomingBlock(x, bnew);
 591     }
 592 }
 593
 594 static void CollapsePhi(BasicBlock* btarget, BasicBlock* bsrc) {
 595   for(BasicBlock::iterator ib = btarget->begin(), ie = btarget->end();
 596       ib != ie; ++ib)
 597     if (PHINode* phi = dyn_cast<PHINode>(&*ib)) {
 598       std::map<BasicBlock*, Value*> counter;
 599       for(unsigned i = 0; i < phi->getNumIncomingValues(); ) {
 600         if (counter[phi->getIncomingBlock(i)]) {
 601           assert(phi->getIncomingValue(i) == counter[phi->getIncomingBlock(i)]);
 602           phi->removeIncomingValue(i, false);
 603         } else {
 604           counter[phi->getIncomingBlock(i)] = phi->getIncomingValue(i);
 605           ++i;
 606         }
 607       }
 608     }
 609 }
 610
 611 template<class T>
 612 static void recBackEdge(BasicBlock* bb, T& BackEdges,
 613                         std::map<BasicBlock*, int>& color,
 614                         std::map<BasicBlock*, int>& depth,
 615                         std::map<BasicBlock*, int>& finish,
 616                         int& time)
 617 {
 618   color[bb] = 1;
 619   ++time;
 620   depth[bb] = time;
 621   TerminatorInst* t= bb->getTerminator();
 622   for(unsigned i = 0; i < t->getNumSuccessors(); ++i) {
 623     BasicBlock* bbnew = t->getSuccessor(i);
 624     if (color[bbnew] == 0)
 625       recBackEdge(bbnew, BackEdges, color, depth, finish, time);
 626     else if (color[bbnew] == 1) {
 627       BackEdges.insert(std::make_pair(bb, bbnew));
 628       //NumBackEdges++;
 629     }
 630   }
 631   color[bb] = 2;
 632   ++time;
 633   finish[bb] = time;
 634 }
 635
 636
 637
 638 //find the back edges and where they go to
 639 template<class T>
 640 static void getBackEdges(Function& F, T& BackEdges) {
 641   std::map<BasicBlock*, int> color;
 642   std::map<BasicBlock*, int> depth;
 643   std::map<BasicBlock*, int> finish;
 644   int time = 0;
 645   recBackEdge(&F.getEntryBlock(), BackEdges, color, depth, finish, time);
 646   DEBUG(errs() << F.getName() << " " << BackEdges.size() << "\n");
 647 }
 648
 649
 650 //Creation functions
 651 ModulePass* llvm::createNullProfilerRSPass() {
 652   return new NullProfilerRS();
 653 }
 654
 655 FunctionPass* llvm::createRSProfilingPass() {
 656   return new ProfilerRS();
 657 }