lib/Transforms/Utils/BypassSlowDivision.cpp

   1 //===- BypassSlowDivision.cpp - Bypass slow division ----------------------===//
   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 // This file contains an optimization for div and rem on architectures that
  10 // execute short instructions significantly faster than longer instructions.
  11 // For example, on Intel Atom 32-bit divides are slow enough that during
  12 // runtime it is profitable to check the value of the operands, and if they are
  13 // positive and less than 256 use an unsigned 8-bit divide.
  14 //
  15 //===----------------------------------------------------------------------===//
  16
  17 #include "llvm/Transforms/Utils/BypassSlowDivision.h"
  18 #include "llvm/ADT/DenseMap.h"
  19 #include "llvm/ADT/None.h"
  20 #include "llvm/ADT/Optional.h"
  21 #include "llvm/ADT/STLExtras.h"
  22 #include "llvm/ADT/SmallPtrSet.h"
  23 #include "llvm/Transforms/Utils/Local.h"
  24 #include "llvm/Analysis/ValueTracking.h"
  25 #include "llvm/IR/BasicBlock.h"
  26 #include "llvm/IR/Constants.h"
  27 #include "llvm/IR/DerivedTypes.h"
  28 #include "llvm/IR/Function.h"
  29 #include "llvm/IR/IRBuilder.h"
  30 #include "llvm/IR/Instruction.h"
  31 #include "llvm/IR/Instructions.h"
  32 #include "llvm/IR/Module.h"
  33 #include "llvm/IR/Type.h"
  34 #include "llvm/IR/Value.h"
  35 #include "llvm/Support/Casting.h"
  36 #include "llvm/Support/KnownBits.h"
  37 #include <cassert>
  38 #include <cstdint>
  39
  40 using namespace llvm;
  41
  42 #define DEBUG_TYPE "bypass-slow-division"
  43
  44 namespace {
  45
  46   struct QuotRemPair {
  47     Value *Quotient;
  48     Value *Remainder;
  49
  50     QuotRemPair(Value *InQuotient, Value *InRemainder)
  51         : Quotient(InQuotient), Remainder(InRemainder) {}
  52   };
  53
  54   /// A quotient and remainder, plus a BB from which they logically "originate".
  55   /// If you use Quotient or Remainder in a Phi node, you should use BB as its
  56   /// corresponding predecessor.
  57   struct QuotRemWithBB {
  58     BasicBlock *BB = nullptr;
  59     Value *Quotient = nullptr;
  60     Value *Remainder = nullptr;
  61   };
  62
  63 using DivCacheTy = DenseMap<DivRemMapKey, QuotRemPair>;
  64 using BypassWidthsTy = DenseMap<unsigned, unsigned>;
  65 using VisitedSetTy = SmallPtrSet<Instruction *, 4>;
  66
  67 enum ValueRange {
  68   /// Operand definitely fits into BypassType. No runtime checks are needed.
  69   VALRNG_KNOWN_SHORT,
  70   /// A runtime check is required, as value range is unknown.
  71   VALRNG_UNKNOWN,
  72   /// Operand is unlikely to fit into BypassType. The bypassing should be
  73   /// disabled.
  74   VALRNG_LIKELY_LONG
  75 };
  76
  77 class FastDivInsertionTask {
  78   bool IsValidTask = false;
  79   Instruction *SlowDivOrRem = nullptr;
  80   IntegerType *BypassType = nullptr;
  81   BasicBlock *MainBB = nullptr;
  82
  83   bool isHashLikeValue(Value *V, VisitedSetTy &Visited);
  84   ValueRange getValueRange(Value *Op, VisitedSetTy &Visited);
  85   QuotRemWithBB createSlowBB(BasicBlock *Successor);
  86   QuotRemWithBB createFastBB(BasicBlock *Successor);
  87   QuotRemPair createDivRemPhiNodes(QuotRemWithBB &LHS, QuotRemWithBB &RHS,
  88                                    BasicBlock *PhiBB);
  89   Value *insertOperandRuntimeCheck(Value *Op1, Value *Op2);
  90   Optional<QuotRemPair> insertFastDivAndRem();
  91
  92   bool isSignedOp() {
  93     return SlowDivOrRem->getOpcode() == Instruction::SDiv ||
  94            SlowDivOrRem->getOpcode() == Instruction::SRem;
  95   }
  96
  97   bool isDivisionOp() {
  98     return SlowDivOrRem->getOpcode() == Instruction::SDiv ||
  99            SlowDivOrRem->getOpcode() == Instruction::UDiv;
 100   }
 101
 102   Type *getSlowType() { return SlowDivOrRem->getType(); }
 103
 104 public:
 105   FastDivInsertionTask(Instruction *I, const BypassWidthsTy &BypassWidths);
 106
 107   Value *getReplacement(DivCacheTy &Cache);
 108 };
 109
 110 } // end anonymous namespace
 111
 112 FastDivInsertionTask::FastDivInsertionTask(Instruction *I,
 113                                            const BypassWidthsTy &BypassWidths) {
 114   switch (I->getOpcode()) {
 115   case Instruction::UDiv:
 116   case Instruction::SDiv:
 117   case Instruction::URem:
 118   case Instruction::SRem:
 119     SlowDivOrRem = I;
 120     break;
 121   default:
 122     // I is not a div/rem operation.
 123     return;
 124   }
 125
 126   // Skip division on vector types. Only optimize integer instructions.
 127   IntegerType *SlowType = dyn_cast<IntegerType>(SlowDivOrRem->getType());
 128   if (!SlowType)
 129     return;
 130
 131   // Skip if this bitwidth is not bypassed.
 132   auto BI = BypassWidths.find(SlowType->getBitWidth());
 133   if (BI == BypassWidths.end())
 134     return;
 135
 136   // Get type for div/rem instruction with bypass bitwidth.
 137   IntegerType *BT = IntegerType::get(I->getContext(), BI->second);
 138   BypassType = BT;
 139
 140   // The original basic block.
 141   MainBB = I->getParent();
 142
 143   // The instruction is indeed a slow div or rem operation.
 144   IsValidTask = true;
 145 }
 146
 147 /// Reuses previously-computed dividend or remainder from the current BB if
 148 /// operands and operation are identical. Otherwise calls insertFastDivAndRem to
 149 /// perform the optimization and caches the resulting dividend and remainder.
 150 /// If no replacement can be generated, nullptr is returned.
 151 Value *FastDivInsertionTask::getReplacement(DivCacheTy &Cache) {
 152   // First, make sure that the task is valid.
 153   if (!IsValidTask)
 154     return nullptr;
 155
 156   // Then, look for a value in Cache.
 157   Value *Dividend = SlowDivOrRem->getOperand(0);
 158   Value *Divisor = SlowDivOrRem->getOperand(1);
 159   DivRemMapKey Key(isSignedOp(), Dividend, Divisor);
 160   auto CacheI = Cache.find(Key);
 161
 162   if (CacheI == Cache.end()) {
 163     // If previous instance does not exist, try to insert fast div.
 164     Optional<QuotRemPair> OptResult = insertFastDivAndRem();
 165     // Bail out if insertFastDivAndRem has failed.
 166     if (!OptResult)
 167       return nullptr;
 168     CacheI = Cache.insert({Key, *OptResult}).first;
 169   }
 170
 171   QuotRemPair &Value = CacheI->second;
 172   return isDivisionOp() ? Value.Quotient : Value.Remainder;
 173 }
 174
 175 /// Check if a value looks like a hash.
 176 ///
 177 /// The routine is expected to detect values computed using the most common hash
 178 /// algorithms. Typically, hash computations end with one of the following
 179 /// instructions:
 180 ///
 181 /// 1) MUL with a constant wider than BypassType
 182 /// 2) XOR instruction
 183 ///
 184 /// And even if we are wrong and the value is not a hash, it is still quite
 185 /// unlikely that such values will fit into BypassType.
 186 ///
 187 /// To detect string hash algorithms like FNV we have to look through PHI-nodes.
 188 /// It is implemented as a depth-first search for values that look neither long
 189 /// nor hash-like.
 190 bool FastDivInsertionTask::isHashLikeValue(Value *V, VisitedSetTy &Visited) {
 191   Instruction *I = dyn_cast<Instruction>(V);
 192   if (!I)
 193     return false;
 194
 195   switch (I->getOpcode()) {
 196   case Instruction::Xor:
 197     return true;
 198   case Instruction::Mul: {
 199     // After Constant Hoisting pass, long constants may be represented as
 200     // bitcast instructions. As a result, some constants may look like an
 201     // instruction at first, and an additional check is necessary to find out if
 202     // an operand is actually a constant.
 203     Value *Op1 = I->getOperand(1);
 204     ConstantInt *C = dyn_cast<ConstantInt>(Op1);
 205     if (!C && isa<BitCastInst>(Op1))
 206       C = dyn_cast<ConstantInt>(cast<BitCastInst>(Op1)->getOperand(0));
 207     return C && C->getValue().getMinSignedBits() > BypassType->getBitWidth();
 208   }
 209   case Instruction::PHI:
 210     // Stop IR traversal in case of a crazy input code. This limits recursion
 211     // depth.
 212     if (Visited.size() >= 16)
 213       return false;
 214     // Do not visit nodes that have been visited already. We return true because
 215     // it means that we couldn't find any value that doesn't look hash-like.
 216     if (Visited.find(I) != Visited.end())
 217       return true;
 218     Visited.insert(I);
 219     return llvm::all_of(cast<PHINode>(I)->incoming_values(), [&](Value *V) {
 220       // Ignore undef values as they probably don't affect the division
 221       // operands.
 222       return getValueRange(V, Visited) == VALRNG_LIKELY_LONG ||
 223              isa<UndefValue>(V);
 224     });
 225   default:
 226     return false;
 227   }
 228 }
 229
 230 /// Check if an integer value fits into our bypass type.
 231 ValueRange FastDivInsertionTask::getValueRange(Value *V,
 232                                                VisitedSetTy &Visited) {
 233   unsigned ShortLen = BypassType->getBitWidth();
 234   unsigned LongLen = V->getType()->getIntegerBitWidth();
 235
 236   assert(LongLen > ShortLen && "Value type must be wider than BypassType");
 237   unsigned HiBits = LongLen - ShortLen;
 238
 239   const DataLayout &DL = SlowDivOrRem->getModule()->getDataLayout();
 240   KnownBits Known(LongLen);
 241
 242   computeKnownBits(V, Known, DL);
 243
 244   if (Known.countMinLeadingZeros() >= HiBits)
 245     return VALRNG_KNOWN_SHORT;
 246
 247   if (Known.countMaxLeadingZeros() < HiBits)
 248     return VALRNG_LIKELY_LONG;
 249
 250   // Long integer divisions are often used in hashtable implementations. It's
 251   // not worth bypassing such divisions because hash values are extremely
 252   // unlikely to have enough leading zeros. The call below tries to detect
 253   // values that are unlikely to fit BypassType (including hashes).
 254   if (isHashLikeValue(V, Visited))
 255     return VALRNG_LIKELY_LONG;
 256
 257   return VALRNG_UNKNOWN;
 258 }
 259
 260 /// Add new basic block for slow div and rem operations and put it before
 261 /// SuccessorBB.
 262 QuotRemWithBB FastDivInsertionTask::createSlowBB(BasicBlock *SuccessorBB) {
 263   QuotRemWithBB DivRemPair;
 264   DivRemPair.BB = BasicBlock::Create(MainBB->getParent()->getContext(), "",
 265                                      MainBB->getParent(), SuccessorBB);
 266   IRBuilder<> Builder(DivRemPair.BB, DivRemPair.BB->begin());
 267
 268   Value *Dividend = SlowDivOrRem->getOperand(0);
 269   Value *Divisor = SlowDivOrRem->getOperand(1);
 270
 271   if (isSignedOp()) {
 272     DivRemPair.Quotient = Builder.CreateSDiv(Dividend, Divisor);
 273     DivRemPair.Remainder = Builder.CreateSRem(Dividend, Divisor);
 274   } else {
 275     DivRemPair.Quotient = Builder.CreateUDiv(Dividend, Divisor);
 276     DivRemPair.Remainder = Builder.CreateURem(Dividend, Divisor);
 277   }
 278
 279   Builder.CreateBr(SuccessorBB);
 280   return DivRemPair;
 281 }
 282
 283 /// Add new basic block for fast div and rem operations and put it before
 284 /// SuccessorBB.
 285 QuotRemWithBB FastDivInsertionTask::createFastBB(BasicBlock *SuccessorBB) {
 286   QuotRemWithBB DivRemPair;
 287   DivRemPair.BB = BasicBlock::Create(MainBB->getParent()->getContext(), "",
 288                                      MainBB->getParent(), SuccessorBB);
 289   IRBuilder<> Builder(DivRemPair.BB, DivRemPair.BB->begin());
 290
 291   Value *Dividend = SlowDivOrRem->getOperand(0);
 292   Value *Divisor = SlowDivOrRem->getOperand(1);
 293   Value *ShortDivisorV =
 294       Builder.CreateCast(Instruction::Trunc, Divisor, BypassType);
 295   Value *ShortDividendV =
 296       Builder.CreateCast(Instruction::Trunc, Dividend, BypassType);
 297
 298   // udiv/urem because this optimization only handles positive numbers.
 299   Value *ShortQV = Builder.CreateUDiv(ShortDividendV, ShortDivisorV);
 300   Value *ShortRV = Builder.CreateURem(ShortDividendV, ShortDivisorV);
 301   DivRemPair.Quotient =
 302       Builder.CreateCast(Instruction::ZExt, ShortQV, getSlowType());
 303   DivRemPair.Remainder =
 304       Builder.CreateCast(Instruction::ZExt, ShortRV, getSlowType());
 305   Builder.CreateBr(SuccessorBB);
 306
 307   return DivRemPair;
 308 }
 309
 310 /// Creates Phi nodes for result of Div and Rem.
 311 QuotRemPair FastDivInsertionTask::createDivRemPhiNodes(QuotRemWithBB &LHS,
 312                                                        QuotRemWithBB &RHS,
 313                                                        BasicBlock *PhiBB) {
 314   IRBuilder<> Builder(PhiBB, PhiBB->begin());
 315   PHINode *QuoPhi = Builder.CreatePHI(getSlowType(), 2);
 316   QuoPhi->addIncoming(LHS.Quotient, LHS.BB);
 317   QuoPhi->addIncoming(RHS.Quotient, RHS.BB);
 318   PHINode *RemPhi = Builder.CreatePHI(getSlowType(), 2);
 319   RemPhi->addIncoming(LHS.Remainder, LHS.BB);
 320   RemPhi->addIncoming(RHS.Remainder, RHS.BB);
 321   return QuotRemPair(QuoPhi, RemPhi);
 322 }
 323
 324 /// Creates a runtime check to test whether both the divisor and dividend fit
 325 /// into BypassType. The check is inserted at the end of MainBB. True return
 326 /// value means that the operands fit. Either of the operands may be NULL if it
 327 /// doesn't need a runtime check.
 328 Value *FastDivInsertionTask::insertOperandRuntimeCheck(Value *Op1, Value *Op2) {
 329   assert((Op1 || Op2) && "Nothing to check");
 330   IRBuilder<> Builder(MainBB, MainBB->end());
 331
 332   Value *OrV;
 333   if (Op1 && Op2)
 334     OrV = Builder.CreateOr(Op1, Op2);
 335   else
 336     OrV = Op1 ? Op1 : Op2;
 337
 338   // BitMask is inverted to check if the operands are
 339   // larger than the bypass type
 340   uint64_t BitMask = ~BypassType->getBitMask();
 341   Value *AndV = Builder.CreateAnd(OrV, BitMask);
 342
 343   // Compare operand values
 344   Value *ZeroV = ConstantInt::getSigned(getSlowType(), 0);
 345   return Builder.CreateICmpEQ(AndV, ZeroV);
 346 }
 347
 348 /// Substitutes the div/rem instruction with code that checks the value of the
 349 /// operands and uses a shorter-faster div/rem instruction when possible.
 350 Optional<QuotRemPair> FastDivInsertionTask::insertFastDivAndRem() {
 351   Value *Dividend = SlowDivOrRem->getOperand(0);
 352   Value *Divisor = SlowDivOrRem->getOperand(1);
 353
 354   VisitedSetTy SetL;
 355   ValueRange DividendRange = getValueRange(Dividend, SetL);
 356   if (DividendRange == VALRNG_LIKELY_LONG)
 357     return None;
 358
 359   VisitedSetTy SetR;
 360   ValueRange DivisorRange = getValueRange(Divisor, SetR);
 361   if (DivisorRange == VALRNG_LIKELY_LONG)
 362     return None;
 363
 364   bool DividendShort = (DividendRange == VALRNG_KNOWN_SHORT);
 365   bool DivisorShort = (DivisorRange == VALRNG_KNOWN_SHORT);
 366
 367   if (DividendShort && DivisorShort) {
 368     // If both operands are known to be short then just replace the long
 369     // division with a short one in-place.  Since we're not introducing control
 370     // flow in this case, narrowing the division is always a win, even if the
 371     // divisor is a constant (and will later get replaced by a multiplication).
 372
 373     IRBuilder<> Builder(SlowDivOrRem);
 374     Value *TruncDividend = Builder.CreateTrunc(Dividend, BypassType);
 375     Value *TruncDivisor = Builder.CreateTrunc(Divisor, BypassType);
 376     Value *TruncDiv = Builder.CreateUDiv(TruncDividend, TruncDivisor);
 377     Value *TruncRem = Builder.CreateURem(TruncDividend, TruncDivisor);
 378     Value *ExtDiv = Builder.CreateZExt(TruncDiv, getSlowType());
 379     Value *ExtRem = Builder.CreateZExt(TruncRem, getSlowType());
 380     return QuotRemPair(ExtDiv, ExtRem);
 381   }
 382
 383   if (isa<ConstantInt>(Divisor)) {
 384     // If the divisor is not a constant, DAGCombiner will convert it to a
 385     // multiplication by a magic constant.  It isn't clear if it is worth
 386     // introducing control flow to get a narrower multiply.
 387     return None;
 388   }
 389
 390   // After Constant Hoisting pass, long constants may be represented as
 391   // bitcast instructions. As a result, some constants may look like an
 392   // instruction at first, and an additional check is necessary to find out if
 393   // an operand is actually a constant.
 394   if (auto *BCI = dyn_cast<BitCastInst>(Divisor))
 395     if (BCI->getParent() == SlowDivOrRem->getParent() &&
 396         isa<ConstantInt>(BCI->getOperand(0)))
 397       return None;
 398
 399   if (DividendShort && !isSignedOp()) {
 400     // If the division is unsigned and Dividend is known to be short, then
 401     // either
 402     // 1) Divisor is less or equal to Dividend, and the result can be computed
 403     //    with a short division.
 404     // 2) Divisor is greater than Dividend. In this case, no division is needed
 405     //    at all: The quotient is 0 and the remainder is equal to Dividend.
 406     //
 407     // So instead of checking at runtime whether Divisor fits into BypassType,
 408     // we emit a runtime check to differentiate between these two cases. This
 409     // lets us entirely avoid a long div.
 410
 411     // Split the basic block before the div/rem.
 412     BasicBlock *SuccessorBB = MainBB->splitBasicBlock(SlowDivOrRem);
 413     // Remove the unconditional branch from MainBB to SuccessorBB.
 414     MainBB->getInstList().back().eraseFromParent();
 415     QuotRemWithBB Long;
 416     Long.BB = MainBB;
 417     Long.Quotient = ConstantInt::get(getSlowType(), 0);
 418     Long.Remainder = Dividend;
 419     QuotRemWithBB Fast = createFastBB(SuccessorBB);
 420     QuotRemPair Result = createDivRemPhiNodes(Fast, Long, SuccessorBB);
 421     IRBuilder<> Builder(MainBB, MainBB->end());
 422     Value *CmpV = Builder.CreateICmpUGE(Dividend, Divisor);
 423     Builder.CreateCondBr(CmpV, Fast.BB, SuccessorBB);
 424     return Result;
 425   } else {
 426     // General case. Create both slow and fast div/rem pairs and choose one of
 427     // them at runtime.
 428
 429     // Split the basic block before the div/rem.
 430     BasicBlock *SuccessorBB = MainBB->splitBasicBlock(SlowDivOrRem);
 431     // Remove the unconditional branch from MainBB to SuccessorBB.
 432     MainBB->getInstList().back().eraseFromParent();
 433     QuotRemWithBB Fast = createFastBB(SuccessorBB);
 434     QuotRemWithBB Slow = createSlowBB(SuccessorBB);
 435     QuotRemPair Result = createDivRemPhiNodes(Fast, Slow, SuccessorBB);
 436     Value *CmpV = insertOperandRuntimeCheck(DividendShort ? nullptr : Dividend,
 437                                             DivisorShort ? nullptr : Divisor);
 438     IRBuilder<> Builder(MainBB, MainBB->end());
 439     Builder.CreateCondBr(CmpV, Fast.BB, Slow.BB);
 440     return Result;
 441   }
 442 }
 443
 444 /// This optimization identifies DIV/REM instructions in a BB that can be
 445 /// profitably bypassed and carried out with a shorter, faster divide.
 446 bool llvm::bypassSlowDivision(BasicBlock *BB,
 447                               const BypassWidthsTy &BypassWidths) {
 448   DivCacheTy PerBBDivCache;
 449
 450   bool MadeChange = false;
 451   Instruction* Next = &*BB->begin();
 452   while (Next != nullptr) {
 453     // We may add instructions immediately after I, but we want to skip over
 454     // them.
 455     Instruction* I = Next;
 456     Next = Next->getNextNode();
 457
 458     FastDivInsertionTask Task(I, BypassWidths);
 459     if (Value *Replacement = Task.getReplacement(PerBBDivCache)) {
 460       I->replaceAllUsesWith(Replacement);
 461       I->eraseFromParent();
 462       MadeChange = true;
 463     }
 464   }
 465
 466   // Above we eagerly create divs and rems, as pairs, so that we can efficiently
 467   // create divrem machine instructions.  Now erase any unused divs / rems so we
 468   // don't leave extra instructions sitting around.
 469   for (auto &KV : PerBBDivCache)
 470     for (Value *V : {KV.second.Quotient, KV.second.Remainder})
 471       RecursivelyDeleteTriviallyDeadInstructions(V);
 472
 473   return MadeChange;
 474 }