David Blaikie | 1213dbf | 2015-06-26 16:57:30 +0000 | [diff] [blame] | 1 | //===----------- VectorUtils.cpp - Vectorizer utility functions -----------===// |
| 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 | // This file defines vectorizer utilities. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 14 | #include "llvm/ADT/EquivalenceClasses.h" |
| 15 | #include "llvm/Analysis/DemandedBits.h" |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 16 | #include "llvm/Analysis/LoopInfo.h" |
| 17 | #include "llvm/Analysis/ScalarEvolutionExpressions.h" |
| 18 | #include "llvm/Analysis/ScalarEvolution.h" |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 19 | #include "llvm/Analysis/TargetTransformInfo.h" |
David Blaikie | b447ac6 | 2015-06-26 18:02:52 +0000 | [diff] [blame] | 20 | #include "llvm/Analysis/VectorUtils.h" |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 21 | #include "llvm/IR/GetElementPtrTypeIterator.h" |
| 22 | #include "llvm/IR/PatternMatch.h" |
| 23 | #include "llvm/IR/Value.h" |
Renato Golin | 3b1d3b0 | 2015-08-30 10:49:04 +0000 | [diff] [blame] | 24 | #include "llvm/IR/Constants.h" |
| 25 | |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 26 | using namespace llvm; |
| 27 | using namespace llvm::PatternMatch; |
David Blaikie | 1213dbf | 2015-06-26 16:57:30 +0000 | [diff] [blame] | 28 | |
| 29 | /// \brief Identify if the intrinsic is trivially vectorizable. |
| 30 | /// This method returns true if the intrinsic's argument types are all |
| 31 | /// scalars for the scalar form of the intrinsic and all vectors for |
| 32 | /// the vector form of the intrinsic. |
| 33 | bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) { |
| 34 | switch (ID) { |
| 35 | case Intrinsic::sqrt: |
| 36 | case Intrinsic::sin: |
| 37 | case Intrinsic::cos: |
| 38 | case Intrinsic::exp: |
| 39 | case Intrinsic::exp2: |
| 40 | case Intrinsic::log: |
| 41 | case Intrinsic::log10: |
| 42 | case Intrinsic::log2: |
| 43 | case Intrinsic::fabs: |
| 44 | case Intrinsic::minnum: |
| 45 | case Intrinsic::maxnum: |
| 46 | case Intrinsic::copysign: |
| 47 | case Intrinsic::floor: |
| 48 | case Intrinsic::ceil: |
| 49 | case Intrinsic::trunc: |
| 50 | case Intrinsic::rint: |
| 51 | case Intrinsic::nearbyint: |
| 52 | case Intrinsic::round: |
| 53 | case Intrinsic::bswap: |
| 54 | case Intrinsic::ctpop: |
| 55 | case Intrinsic::pow: |
| 56 | case Intrinsic::fma: |
| 57 | case Intrinsic::fmuladd: |
| 58 | case Intrinsic::ctlz: |
| 59 | case Intrinsic::cttz: |
| 60 | case Intrinsic::powi: |
| 61 | return true; |
| 62 | default: |
| 63 | return false; |
| 64 | } |
| 65 | } |
| 66 | |
| 67 | /// \brief Identifies if the intrinsic has a scalar operand. It check for |
| 68 | /// ctlz,cttz and powi special intrinsics whose argument is scalar. |
| 69 | bool llvm::hasVectorInstrinsicScalarOpd(Intrinsic::ID ID, |
| 70 | unsigned ScalarOpdIdx) { |
| 71 | switch (ID) { |
| 72 | case Intrinsic::ctlz: |
| 73 | case Intrinsic::cttz: |
| 74 | case Intrinsic::powi: |
| 75 | return (ScalarOpdIdx == 1); |
| 76 | default: |
| 77 | return false; |
| 78 | } |
| 79 | } |
| 80 | |
| 81 | /// \brief Check call has a unary float signature |
| 82 | /// It checks following: |
| 83 | /// a) call should have a single argument |
| 84 | /// b) argument type should be floating point type |
| 85 | /// c) call instruction type and argument type should be same |
| 86 | /// d) call should only reads memory. |
| 87 | /// If all these condition is met then return ValidIntrinsicID |
| 88 | /// else return not_intrinsic. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 89 | Intrinsic::ID |
David Blaikie | 1213dbf | 2015-06-26 16:57:30 +0000 | [diff] [blame] | 90 | llvm::checkUnaryFloatSignature(const CallInst &I, |
| 91 | Intrinsic::ID ValidIntrinsicID) { |
| 92 | if (I.getNumArgOperands() != 1 || |
| 93 | !I.getArgOperand(0)->getType()->isFloatingPointTy() || |
| 94 | I.getType() != I.getArgOperand(0)->getType() || !I.onlyReadsMemory()) |
| 95 | return Intrinsic::not_intrinsic; |
| 96 | |
| 97 | return ValidIntrinsicID; |
| 98 | } |
| 99 | |
| 100 | /// \brief Check call has a binary float signature |
| 101 | /// It checks following: |
| 102 | /// a) call should have 2 arguments. |
| 103 | /// b) arguments type should be floating point type |
| 104 | /// c) call instruction type and arguments type should be same |
| 105 | /// d) call should only reads memory. |
| 106 | /// If all these condition is met then return ValidIntrinsicID |
| 107 | /// else return not_intrinsic. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 108 | Intrinsic::ID |
David Blaikie | 1213dbf | 2015-06-26 16:57:30 +0000 | [diff] [blame] | 109 | llvm::checkBinaryFloatSignature(const CallInst &I, |
| 110 | Intrinsic::ID ValidIntrinsicID) { |
| 111 | if (I.getNumArgOperands() != 2 || |
| 112 | !I.getArgOperand(0)->getType()->isFloatingPointTy() || |
| 113 | !I.getArgOperand(1)->getType()->isFloatingPointTy() || |
| 114 | I.getType() != I.getArgOperand(0)->getType() || |
| 115 | I.getType() != I.getArgOperand(1)->getType() || !I.onlyReadsMemory()) |
| 116 | return Intrinsic::not_intrinsic; |
| 117 | |
| 118 | return ValidIntrinsicID; |
| 119 | } |
| 120 | |
| 121 | /// \brief Returns intrinsic ID for call. |
| 122 | /// For the input call instruction it finds mapping intrinsic and returns |
| 123 | /// its ID, in case it does not found it return not_intrinsic. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 124 | Intrinsic::ID llvm::getIntrinsicIDForCall(CallInst *CI, |
| 125 | const TargetLibraryInfo *TLI) { |
David Blaikie | 1213dbf | 2015-06-26 16:57:30 +0000 | [diff] [blame] | 126 | // If we have an intrinsic call, check if it is trivially vectorizable. |
| 127 | if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) { |
| 128 | Intrinsic::ID ID = II->getIntrinsicID(); |
| 129 | if (isTriviallyVectorizable(ID) || ID == Intrinsic::lifetime_start || |
| 130 | ID == Intrinsic::lifetime_end || ID == Intrinsic::assume) |
| 131 | return ID; |
| 132 | return Intrinsic::not_intrinsic; |
| 133 | } |
| 134 | |
| 135 | if (!TLI) |
| 136 | return Intrinsic::not_intrinsic; |
| 137 | |
| 138 | LibFunc::Func Func; |
| 139 | Function *F = CI->getCalledFunction(); |
| 140 | // We're going to make assumptions on the semantics of the functions, check |
| 141 | // that the target knows that it's available in this environment and it does |
| 142 | // not have local linkage. |
| 143 | if (!F || F->hasLocalLinkage() || !TLI->getLibFunc(F->getName(), Func)) |
| 144 | return Intrinsic::not_intrinsic; |
| 145 | |
| 146 | // Otherwise check if we have a call to a function that can be turned into a |
| 147 | // vector intrinsic. |
| 148 | switch (Func) { |
| 149 | default: |
| 150 | break; |
| 151 | case LibFunc::sin: |
| 152 | case LibFunc::sinf: |
| 153 | case LibFunc::sinl: |
| 154 | return checkUnaryFloatSignature(*CI, Intrinsic::sin); |
| 155 | case LibFunc::cos: |
| 156 | case LibFunc::cosf: |
| 157 | case LibFunc::cosl: |
| 158 | return checkUnaryFloatSignature(*CI, Intrinsic::cos); |
| 159 | case LibFunc::exp: |
| 160 | case LibFunc::expf: |
| 161 | case LibFunc::expl: |
| 162 | return checkUnaryFloatSignature(*CI, Intrinsic::exp); |
| 163 | case LibFunc::exp2: |
| 164 | case LibFunc::exp2f: |
| 165 | case LibFunc::exp2l: |
| 166 | return checkUnaryFloatSignature(*CI, Intrinsic::exp2); |
| 167 | case LibFunc::log: |
| 168 | case LibFunc::logf: |
| 169 | case LibFunc::logl: |
| 170 | return checkUnaryFloatSignature(*CI, Intrinsic::log); |
| 171 | case LibFunc::log10: |
| 172 | case LibFunc::log10f: |
| 173 | case LibFunc::log10l: |
| 174 | return checkUnaryFloatSignature(*CI, Intrinsic::log10); |
| 175 | case LibFunc::log2: |
| 176 | case LibFunc::log2f: |
| 177 | case LibFunc::log2l: |
| 178 | return checkUnaryFloatSignature(*CI, Intrinsic::log2); |
| 179 | case LibFunc::fabs: |
| 180 | case LibFunc::fabsf: |
| 181 | case LibFunc::fabsl: |
| 182 | return checkUnaryFloatSignature(*CI, Intrinsic::fabs); |
| 183 | case LibFunc::fmin: |
| 184 | case LibFunc::fminf: |
| 185 | case LibFunc::fminl: |
| 186 | return checkBinaryFloatSignature(*CI, Intrinsic::minnum); |
| 187 | case LibFunc::fmax: |
| 188 | case LibFunc::fmaxf: |
| 189 | case LibFunc::fmaxl: |
| 190 | return checkBinaryFloatSignature(*CI, Intrinsic::maxnum); |
| 191 | case LibFunc::copysign: |
| 192 | case LibFunc::copysignf: |
| 193 | case LibFunc::copysignl: |
| 194 | return checkBinaryFloatSignature(*CI, Intrinsic::copysign); |
| 195 | case LibFunc::floor: |
| 196 | case LibFunc::floorf: |
| 197 | case LibFunc::floorl: |
| 198 | return checkUnaryFloatSignature(*CI, Intrinsic::floor); |
| 199 | case LibFunc::ceil: |
| 200 | case LibFunc::ceilf: |
| 201 | case LibFunc::ceill: |
| 202 | return checkUnaryFloatSignature(*CI, Intrinsic::ceil); |
| 203 | case LibFunc::trunc: |
| 204 | case LibFunc::truncf: |
| 205 | case LibFunc::truncl: |
| 206 | return checkUnaryFloatSignature(*CI, Intrinsic::trunc); |
| 207 | case LibFunc::rint: |
| 208 | case LibFunc::rintf: |
| 209 | case LibFunc::rintl: |
| 210 | return checkUnaryFloatSignature(*CI, Intrinsic::rint); |
| 211 | case LibFunc::nearbyint: |
| 212 | case LibFunc::nearbyintf: |
| 213 | case LibFunc::nearbyintl: |
| 214 | return checkUnaryFloatSignature(*CI, Intrinsic::nearbyint); |
| 215 | case LibFunc::round: |
| 216 | case LibFunc::roundf: |
| 217 | case LibFunc::roundl: |
| 218 | return checkUnaryFloatSignature(*CI, Intrinsic::round); |
| 219 | case LibFunc::pow: |
| 220 | case LibFunc::powf: |
| 221 | case LibFunc::powl: |
| 222 | return checkBinaryFloatSignature(*CI, Intrinsic::pow); |
| 223 | } |
| 224 | |
| 225 | return Intrinsic::not_intrinsic; |
| 226 | } |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 227 | |
| 228 | /// \brief Find the operand of the GEP that should be checked for consecutive |
| 229 | /// stores. This ignores trailing indices that have no effect on the final |
| 230 | /// pointer. |
| 231 | unsigned llvm::getGEPInductionOperand(const GetElementPtrInst *Gep) { |
| 232 | const DataLayout &DL = Gep->getModule()->getDataLayout(); |
| 233 | unsigned LastOperand = Gep->getNumOperands() - 1; |
Eduard Burtescu | 19eb031 | 2016-01-19 17:28:00 +0000 | [diff] [blame] | 234 | unsigned GEPAllocSize = DL.getTypeAllocSize(Gep->getResultElementType()); |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 235 | |
| 236 | // Walk backwards and try to peel off zeros. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 237 | while (LastOperand > 1 && match(Gep->getOperand(LastOperand), m_Zero())) { |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 238 | // Find the type we're currently indexing into. |
| 239 | gep_type_iterator GEPTI = gep_type_begin(Gep); |
| 240 | std::advance(GEPTI, LastOperand - 1); |
| 241 | |
| 242 | // If it's a type with the same allocation size as the result of the GEP we |
| 243 | // can peel off the zero index. |
| 244 | if (DL.getTypeAllocSize(*GEPTI) != GEPAllocSize) |
| 245 | break; |
| 246 | --LastOperand; |
| 247 | } |
| 248 | |
| 249 | return LastOperand; |
| 250 | } |
| 251 | |
| 252 | /// \brief If the argument is a GEP, then returns the operand identified by |
| 253 | /// getGEPInductionOperand. However, if there is some other non-loop-invariant |
| 254 | /// operand, it returns that instead. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 255 | Value *llvm::stripGetElementPtr(Value *Ptr, ScalarEvolution *SE, Loop *Lp) { |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 256 | GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr); |
| 257 | if (!GEP) |
| 258 | return Ptr; |
| 259 | |
| 260 | unsigned InductionOperand = getGEPInductionOperand(GEP); |
| 261 | |
| 262 | // Check that all of the gep indices are uniform except for our induction |
| 263 | // operand. |
| 264 | for (unsigned i = 0, e = GEP->getNumOperands(); i != e; ++i) |
| 265 | if (i != InductionOperand && |
| 266 | !SE->isLoopInvariant(SE->getSCEV(GEP->getOperand(i)), Lp)) |
| 267 | return Ptr; |
| 268 | return GEP->getOperand(InductionOperand); |
| 269 | } |
| 270 | |
| 271 | /// \brief If a value has only one user that is a CastInst, return it. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 272 | Value *llvm::getUniqueCastUse(Value *Ptr, Loop *Lp, Type *Ty) { |
| 273 | Value *UniqueCast = nullptr; |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 274 | for (User *U : Ptr->users()) { |
| 275 | CastInst *CI = dyn_cast<CastInst>(U); |
| 276 | if (CI && CI->getType() == Ty) { |
| 277 | if (!UniqueCast) |
| 278 | UniqueCast = CI; |
| 279 | else |
| 280 | return nullptr; |
| 281 | } |
| 282 | } |
| 283 | return UniqueCast; |
| 284 | } |
| 285 | |
| 286 | /// \brief Get the stride of a pointer access in a loop. Looks for symbolic |
| 287 | /// strides "a[i*stride]". Returns the symbolic stride, or null otherwise. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 288 | Value *llvm::getStrideFromPointer(Value *Ptr, ScalarEvolution *SE, Loop *Lp) { |
Craig Topper | e3dcce9 | 2015-08-01 22:20:21 +0000 | [diff] [blame] | 289 | auto *PtrTy = dyn_cast<PointerType>(Ptr->getType()); |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 290 | if (!PtrTy || PtrTy->isAggregateType()) |
| 291 | return nullptr; |
| 292 | |
| 293 | // Try to remove a gep instruction to make the pointer (actually index at this |
| 294 | // point) easier analyzable. If OrigPtr is equal to Ptr we are analzying the |
| 295 | // pointer, otherwise, we are analyzing the index. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 296 | Value *OrigPtr = Ptr; |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 297 | |
| 298 | // The size of the pointer access. |
| 299 | int64_t PtrAccessSize = 1; |
| 300 | |
| 301 | Ptr = stripGetElementPtr(Ptr, SE, Lp); |
| 302 | const SCEV *V = SE->getSCEV(Ptr); |
| 303 | |
| 304 | if (Ptr != OrigPtr) |
| 305 | // Strip off casts. |
| 306 | while (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(V)) |
| 307 | V = C->getOperand(); |
| 308 | |
| 309 | const SCEVAddRecExpr *S = dyn_cast<SCEVAddRecExpr>(V); |
| 310 | if (!S) |
| 311 | return nullptr; |
| 312 | |
| 313 | V = S->getStepRecurrence(*SE); |
| 314 | if (!V) |
| 315 | return nullptr; |
| 316 | |
| 317 | // Strip off the size of access multiplication if we are still analyzing the |
| 318 | // pointer. |
| 319 | if (OrigPtr == Ptr) { |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 320 | if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(V)) { |
| 321 | if (M->getOperand(0)->getSCEVType() != scConstant) |
| 322 | return nullptr; |
| 323 | |
Sanjoy Das | 0de2fec | 2015-12-17 20:28:46 +0000 | [diff] [blame] | 324 | const APInt &APStepVal = cast<SCEVConstant>(M->getOperand(0))->getAPInt(); |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 325 | |
| 326 | // Huge step value - give up. |
| 327 | if (APStepVal.getBitWidth() > 64) |
| 328 | return nullptr; |
| 329 | |
| 330 | int64_t StepVal = APStepVal.getSExtValue(); |
| 331 | if (PtrAccessSize != StepVal) |
| 332 | return nullptr; |
| 333 | V = M->getOperand(1); |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | // Strip off casts. |
| 338 | Type *StripedOffRecurrenceCast = nullptr; |
| 339 | if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(V)) { |
| 340 | StripedOffRecurrenceCast = C->getType(); |
| 341 | V = C->getOperand(); |
| 342 | } |
| 343 | |
| 344 | // Look for the loop invariant symbolic value. |
| 345 | const SCEVUnknown *U = dyn_cast<SCEVUnknown>(V); |
| 346 | if (!U) |
| 347 | return nullptr; |
| 348 | |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 349 | Value *Stride = U->getValue(); |
Hal Finkel | 9cf58c4 | 2015-07-11 10:52:42 +0000 | [diff] [blame] | 350 | if (!Lp->isLoopInvariant(Stride)) |
| 351 | return nullptr; |
| 352 | |
| 353 | // If we have stripped off the recurrence cast we have to make sure that we |
| 354 | // return the value that is used in this loop so that we can replace it later. |
| 355 | if (StripedOffRecurrenceCast) |
| 356 | Stride = getUniqueCastUse(Stride, Lp, StripedOffRecurrenceCast); |
| 357 | |
| 358 | return Stride; |
| 359 | } |
David Majnemer | 599ca44 | 2015-07-13 01:15:53 +0000 | [diff] [blame] | 360 | |
| 361 | /// \brief Given a vector and an element number, see if the scalar value is |
| 362 | /// already around as a register, for example if it were inserted then extracted |
| 363 | /// from the vector. |
David Majnemer | 5eaf08f | 2015-08-18 22:07:20 +0000 | [diff] [blame] | 364 | Value *llvm::findScalarElement(Value *V, unsigned EltNo) { |
David Majnemer | 599ca44 | 2015-07-13 01:15:53 +0000 | [diff] [blame] | 365 | assert(V->getType()->isVectorTy() && "Not looking at a vector?"); |
| 366 | VectorType *VTy = cast<VectorType>(V->getType()); |
| 367 | unsigned Width = VTy->getNumElements(); |
| 368 | if (EltNo >= Width) // Out of range access. |
| 369 | return UndefValue::get(VTy->getElementType()); |
| 370 | |
| 371 | if (Constant *C = dyn_cast<Constant>(V)) |
| 372 | return C->getAggregateElement(EltNo); |
| 373 | |
| 374 | if (InsertElementInst *III = dyn_cast<InsertElementInst>(V)) { |
| 375 | // If this is an insert to a variable element, we don't know what it is. |
| 376 | if (!isa<ConstantInt>(III->getOperand(2))) |
| 377 | return nullptr; |
| 378 | unsigned IIElt = cast<ConstantInt>(III->getOperand(2))->getZExtValue(); |
| 379 | |
| 380 | // If this is an insert to the element we are looking for, return the |
| 381 | // inserted value. |
| 382 | if (EltNo == IIElt) |
| 383 | return III->getOperand(1); |
| 384 | |
| 385 | // Otherwise, the insertelement doesn't modify the value, recurse on its |
| 386 | // vector input. |
| 387 | return findScalarElement(III->getOperand(0), EltNo); |
| 388 | } |
| 389 | |
| 390 | if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(V)) { |
| 391 | unsigned LHSWidth = SVI->getOperand(0)->getType()->getVectorNumElements(); |
| 392 | int InEl = SVI->getMaskValue(EltNo); |
| 393 | if (InEl < 0) |
| 394 | return UndefValue::get(VTy->getElementType()); |
| 395 | if (InEl < (int)LHSWidth) |
| 396 | return findScalarElement(SVI->getOperand(0), InEl); |
| 397 | return findScalarElement(SVI->getOperand(1), InEl - LHSWidth); |
| 398 | } |
| 399 | |
| 400 | // Extract a value from a vector add operation with a constant zero. |
| 401 | Value *Val = nullptr; Constant *Con = nullptr; |
David Majnemer | c6bb0e2 | 2015-08-18 22:07:25 +0000 | [diff] [blame] | 402 | if (match(V, m_Add(m_Value(Val), m_Constant(Con)))) |
| 403 | if (Constant *Elt = Con->getAggregateElement(EltNo)) |
| 404 | if (Elt->isNullValue()) |
| 405 | return findScalarElement(Val, EltNo); |
David Majnemer | 599ca44 | 2015-07-13 01:15:53 +0000 | [diff] [blame] | 406 | |
| 407 | // Otherwise, we don't know. |
| 408 | return nullptr; |
| 409 | } |
Renato Golin | 3b1d3b0 | 2015-08-30 10:49:04 +0000 | [diff] [blame] | 410 | |
| 411 | /// \brief Get splat value if the input is a splat vector or return nullptr. |
Elena Demikhovsky | 63a7ca9 | 2015-08-30 13:48:02 +0000 | [diff] [blame] | 412 | /// This function is not fully general. It checks only 2 cases: |
| 413 | /// the input value is (1) a splat constants vector or (2) a sequence |
| 414 | /// of instructions that broadcast a single value into a vector. |
| 415 | /// |
Elena Demikhovsky | 0781d7b | 2015-12-01 12:08:36 +0000 | [diff] [blame] | 416 | const llvm::Value *llvm::getSplatValue(const Value *V) { |
| 417 | |
| 418 | if (auto *C = dyn_cast<Constant>(V)) |
Elena Demikhovsky | 47fa271 | 2015-12-01 12:30:40 +0000 | [diff] [blame] | 419 | if (isa<VectorType>(V->getType())) |
| 420 | return C->getSplatValue(); |
Elena Demikhovsky | 63a7ca9 | 2015-08-30 13:48:02 +0000 | [diff] [blame] | 421 | |
| 422 | auto *ShuffleInst = dyn_cast<ShuffleVectorInst>(V); |
Renato Golin | 3b1d3b0 | 2015-08-30 10:49:04 +0000 | [diff] [blame] | 423 | if (!ShuffleInst) |
| 424 | return nullptr; |
Elena Demikhovsky | 63a7ca9 | 2015-08-30 13:48:02 +0000 | [diff] [blame] | 425 | // All-zero (or undef) shuffle mask elements. |
| 426 | for (int MaskElt : ShuffleInst->getShuffleMask()) |
| 427 | if (MaskElt != 0 && MaskElt != -1) |
Renato Golin | 3b1d3b0 | 2015-08-30 10:49:04 +0000 | [diff] [blame] | 428 | return nullptr; |
| 429 | // The first shuffle source is 'insertelement' with index 0. |
Elena Demikhovsky | 63a7ca9 | 2015-08-30 13:48:02 +0000 | [diff] [blame] | 430 | auto *InsertEltInst = |
| 431 | dyn_cast<InsertElementInst>(ShuffleInst->getOperand(0)); |
Renato Golin | 3b1d3b0 | 2015-08-30 10:49:04 +0000 | [diff] [blame] | 432 | if (!InsertEltInst || !isa<ConstantInt>(InsertEltInst->getOperand(2)) || |
| 433 | !cast<ConstantInt>(InsertEltInst->getOperand(2))->isNullValue()) |
| 434 | return nullptr; |
| 435 | |
| 436 | return InsertEltInst->getOperand(1); |
| 437 | } |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 438 | |
Charlie Turner | 54336a5 | 2015-11-26 20:39:51 +0000 | [diff] [blame] | 439 | MapVector<Instruction *, uint64_t> |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 440 | llvm::computeMinimumValueSizes(ArrayRef<BasicBlock *> Blocks, DemandedBits &DB, |
| 441 | const TargetTransformInfo *TTI) { |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 442 | |
| 443 | // DemandedBits will give us every value's live-out bits. But we want |
| 444 | // to ensure no extra casts would need to be inserted, so every DAG |
| 445 | // of connected values must have the same minimum bitwidth. |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 446 | EquivalenceClasses<Value *> ECs; |
| 447 | SmallVector<Value *, 16> Worklist; |
| 448 | SmallPtrSet<Value *, 4> Roots; |
| 449 | SmallPtrSet<Value *, 16> Visited; |
| 450 | DenseMap<Value *, uint64_t> DBits; |
| 451 | SmallPtrSet<Instruction *, 4> InstructionSet; |
Charlie Turner | 54336a5 | 2015-11-26 20:39:51 +0000 | [diff] [blame] | 452 | MapVector<Instruction *, uint64_t> MinBWs; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 453 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 454 | // Determine the roots. We work bottom-up, from truncs or icmps. |
| 455 | bool SeenExtFromIllegalType = false; |
| 456 | for (auto *BB : Blocks) |
| 457 | for (auto &I : *BB) { |
| 458 | InstructionSet.insert(&I); |
| 459 | |
| 460 | if (TTI && (isa<ZExtInst>(&I) || isa<SExtInst>(&I)) && |
| 461 | !TTI->isTypeLegal(I.getOperand(0)->getType())) |
| 462 | SeenExtFromIllegalType = true; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 463 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 464 | // Only deal with non-vector integers up to 64-bits wide. |
| 465 | if ((isa<TruncInst>(&I) || isa<ICmpInst>(&I)) && |
| 466 | !I.getType()->isVectorTy() && |
| 467 | I.getOperand(0)->getType()->getScalarSizeInBits() <= 64) { |
| 468 | // Don't make work for ourselves. If we know the loaded type is legal, |
| 469 | // don't add it to the worklist. |
| 470 | if (TTI && isa<TruncInst>(&I) && TTI->isTypeLegal(I.getType())) |
| 471 | continue; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 472 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 473 | Worklist.push_back(&I); |
| 474 | Roots.insert(&I); |
| 475 | } |
| 476 | } |
| 477 | // Early exit. |
| 478 | if (Worklist.empty() || (TTI && !SeenExtFromIllegalType)) |
| 479 | return MinBWs; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 480 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 481 | // Now proceed breadth-first, unioning values together. |
| 482 | while (!Worklist.empty()) { |
| 483 | Value *Val = Worklist.pop_back_val(); |
| 484 | Value *Leader = ECs.getOrInsertLeaderValue(Val); |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 485 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 486 | if (Visited.count(Val)) |
| 487 | continue; |
| 488 | Visited.insert(Val); |
| 489 | |
| 490 | // Non-instructions terminate a chain successfully. |
| 491 | if (!isa<Instruction>(Val)) |
| 492 | continue; |
| 493 | Instruction *I = cast<Instruction>(Val); |
| 494 | |
| 495 | // If we encounter a type that is larger than 64 bits, we can't represent |
| 496 | // it so bail out. |
| 497 | if (DB.getDemandedBits(I).getBitWidth() > 64) |
Charlie Turner | 54336a5 | 2015-11-26 20:39:51 +0000 | [diff] [blame] | 498 | return MapVector<Instruction *, uint64_t>(); |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 499 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 500 | uint64_t V = DB.getDemandedBits(I).getZExtValue(); |
| 501 | DBits[Leader] |= V; |
James Molloy | 8e46cd0 | 2016-03-30 10:11:43 +0000 | [diff] [blame^] | 502 | DBits[I] = V; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 503 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 504 | // Casts, loads and instructions outside of our range terminate a chain |
| 505 | // successfully. |
| 506 | if (isa<SExtInst>(I) || isa<ZExtInst>(I) || isa<LoadInst>(I) || |
| 507 | !InstructionSet.count(I)) |
| 508 | continue; |
| 509 | |
| 510 | // Unsafe casts terminate a chain unsuccessfully. We can't do anything |
| 511 | // useful with bitcasts, ptrtoints or inttoptrs and it'd be unsafe to |
| 512 | // transform anything that relies on them. |
| 513 | if (isa<BitCastInst>(I) || isa<PtrToIntInst>(I) || isa<IntToPtrInst>(I) || |
| 514 | !I->getType()->isIntegerTy()) { |
| 515 | DBits[Leader] |= ~0ULL; |
| 516 | continue; |
| 517 | } |
| 518 | |
| 519 | // We don't modify the types of PHIs. Reductions will already have been |
| 520 | // truncated if possible, and inductions' sizes will have been chosen by |
| 521 | // indvars. |
| 522 | if (isa<PHINode>(I)) |
| 523 | continue; |
| 524 | |
| 525 | if (DBits[Leader] == ~0ULL) |
| 526 | // All bits demanded, no point continuing. |
| 527 | continue; |
| 528 | |
| 529 | for (Value *O : cast<User>(I)->operands()) { |
| 530 | ECs.unionSets(Leader, O); |
| 531 | Worklist.push_back(O); |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | // Now we've discovered all values, walk them to see if there are |
| 536 | // any users we didn't see. If there are, we can't optimize that |
| 537 | // chain. |
| 538 | for (auto &I : DBits) |
| 539 | for (auto *U : I.first->users()) |
| 540 | if (U->getType()->isIntegerTy() && DBits.count(U) == 0) |
| 541 | DBits[ECs.getOrInsertLeaderValue(I.first)] |= ~0ULL; |
James Molloy | 45f67d5 | 2015-11-09 14:32:05 +0000 | [diff] [blame] | 542 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 543 | for (auto I = ECs.begin(), E = ECs.end(); I != E; ++I) { |
| 544 | uint64_t LeaderDemandedBits = 0; |
| 545 | for (auto MI = ECs.member_begin(I), ME = ECs.member_end(); MI != ME; ++MI) |
| 546 | LeaderDemandedBits |= DBits[*MI]; |
| 547 | |
| 548 | uint64_t MinBW = (sizeof(LeaderDemandedBits) * 8) - |
| 549 | llvm::countLeadingZeros(LeaderDemandedBits); |
| 550 | // Round up to a power of 2 |
| 551 | if (!isPowerOf2_64((uint64_t)MinBW)) |
| 552 | MinBW = NextPowerOf2(MinBW); |
James Molloy | 8e46cd0 | 2016-03-30 10:11:43 +0000 | [diff] [blame^] | 553 | |
| 554 | // We don't modify the types of PHIs. Reductions will already have been |
| 555 | // truncated if possible, and inductions' sizes will have been chosen by |
| 556 | // indvars. |
| 557 | // If we are required to shrink a PHI, abandon this entire equivalence class. |
| 558 | bool Abort = false; |
| 559 | for (auto MI = ECs.member_begin(I), ME = ECs.member_end(); MI != ME; ++MI) |
| 560 | if (isa<PHINode>(*MI) && MinBW < (*MI)->getType()->getScalarSizeInBits()) { |
| 561 | Abort = true; |
| 562 | break; |
| 563 | } |
| 564 | if (Abort) |
| 565 | continue; |
| 566 | |
James Molloy | 55d633b | 2015-10-12 12:34:45 +0000 | [diff] [blame] | 567 | for (auto MI = ECs.member_begin(I), ME = ECs.member_end(); MI != ME; ++MI) { |
| 568 | if (!isa<Instruction>(*MI)) |
| 569 | continue; |
| 570 | Type *Ty = (*MI)->getType(); |
| 571 | if (Roots.count(*MI)) |
| 572 | Ty = cast<Instruction>(*MI)->getOperand(0)->getType(); |
| 573 | if (MinBW < Ty->getScalarSizeInBits()) |
| 574 | MinBWs[cast<Instruction>(*MI)] = MinBW; |
| 575 | } |
| 576 | } |
| 577 | |
| 578 | return MinBWs; |
| 579 | } |