llvm/lib/CodeGen/IntrinsicLowering.cpp

   1 //===-- IntrinsicLowering.cpp - Intrinsic Lowering default implementation -===//
   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 implements the IntrinsicLowering class.
  10 //
  11 //===----------------------------------------------------------------------===//
  12
  13 #include "llvm/CodeGen/IntrinsicLowering.h"
  14 #include "llvm/ADT/SmallVector.h"
  15 #include "llvm/IR/Constants.h"
  16 #include "llvm/IR/DataLayout.h"
  17 #include "llvm/IR/DerivedTypes.h"
  18 #include "llvm/IR/IRBuilder.h"
  19 #include "llvm/IR/Module.h"
  20 #include "llvm/IR/Type.h"
  21 #include "llvm/Support/ErrorHandling.h"
  22 #include "llvm/Support/raw_ostream.h"
  23 using namespace llvm;
  24
  25 /// This function is used when we want to lower an intrinsic call to a call of
  26 /// an external function. This handles hard cases such as when there was already
  27 /// a prototype for the external function, but that prototype doesn't match the
  28 /// arguments we expect to pass in.
  29 template <class ArgIt>
  30 static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI,
  31                                  ArgIt ArgBegin, ArgIt ArgEnd,
  32                                  Type *RetTy) {
  33   // If we haven't already looked up this function, check to see if the
  34   // program already contains a function with this name.
  35   Module *M = CI->getModule();
  36   // Get or insert the definition now.
  37   std::vector<Type *> ParamTys;
  38   for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
  39     ParamTys.push_back((*I)->getType());
  40   FunctionCallee FCache =
  41       M->getOrInsertFunction(NewFn, FunctionType::get(RetTy, ParamTys, false));
  42
  43   IRBuilder<> Builder(CI->getParent(), CI->getIterator());
  44   SmallVector<Value *, 8> Args(ArgBegin, ArgEnd);
  45   CallInst *NewCI = Builder.CreateCall(FCache, Args);
  46   NewCI->setName(CI->getName());
  47   if (!CI->use_empty())
  48     CI->replaceAllUsesWith(NewCI);
  49   return NewCI;
  50 }
  51
  52 /// Emit the code to lower bswap of V before the specified instruction IP.
  53 static Value *LowerBSWAP(LLVMContext &Context, Value *V, Instruction *IP) {
  54   assert(V->getType()->isIntOrIntVectorTy() && "Can't bswap a non-integer type!");
  55
  56   unsigned BitSize = V->getType()->getScalarSizeInBits();
  57
  58   IRBuilder<> Builder(IP);
  59
  60   switch(BitSize) {
  61   default: llvm_unreachable("Unhandled type size of value to byteswap!");
  62   case 16: {
  63     Value *Tmp1 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
  64                                     "bswap.2");
  65     Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
  66                                      "bswap.1");
  67     V = Builder.CreateOr(Tmp1, Tmp2, "bswap.i16");
  68     break;
  69   }
  70   case 32: {
  71     Value *Tmp4 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
  72                                     "bswap.4");
  73     Value *Tmp3 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
  74                                     "bswap.3");
  75     Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
  76                                      "bswap.2");
  77     Value *Tmp1 = Builder.CreateLShr(V,ConstantInt::get(V->getType(), 24),
  78                                      "bswap.1");
  79     Tmp3 = Builder.CreateAnd(Tmp3,
  80                          ConstantInt::get(V->getType(), 0xFF0000),
  81                              "bswap.and3");
  82     Tmp2 = Builder.CreateAnd(Tmp2,
  83                            ConstantInt::get(V->getType(), 0xFF00),
  84                              "bswap.and2");
  85     Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or1");
  86     Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or2");
  87     V = Builder.CreateOr(Tmp4, Tmp2, "bswap.i32");
  88     break;
  89   }
  90   case 64: {
  91     Value *Tmp8 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 56),
  92                                     "bswap.8");
  93     Value *Tmp7 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 40),
  94                                     "bswap.7");
  95     Value *Tmp6 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
  96                                     "bswap.6");
  97     Value *Tmp5 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
  98                                     "bswap.5");
  99     Value* Tmp4 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
 100                                      "bswap.4");
 101     Value* Tmp3 = Builder.CreateLShr(V,
 102                                      ConstantInt::get(V->getType(), 24),
 103                                      "bswap.3");
 104     Value* Tmp2 = Builder.CreateLShr(V,
 105                                      ConstantInt::get(V->getType(), 40),
 106                                      "bswap.2");
 107     Value* Tmp1 = Builder.CreateLShr(V,
 108                                      ConstantInt::get(V->getType(), 56),
 109                                      "bswap.1");
 110     Tmp7 = Builder.CreateAnd(Tmp7,
 111                              ConstantInt::get(V->getType(),
 112                                               0xFF000000000000ULL),
 113                              "bswap.and7");
 114     Tmp6 = Builder.CreateAnd(Tmp6,
 115                              ConstantInt::get(V->getType(),
 116                                               0xFF0000000000ULL),
 117                              "bswap.and6");
 118     Tmp5 = Builder.CreateAnd(Tmp5,
 119                         ConstantInt::get(V->getType(),
 120                              0xFF00000000ULL),
 121                              "bswap.and5");
 122     Tmp4 = Builder.CreateAnd(Tmp4,
 123                         ConstantInt::get(V->getType(),
 124                              0xFF000000ULL),
 125                              "bswap.and4");
 126     Tmp3 = Builder.CreateAnd(Tmp3,
 127                              ConstantInt::get(V->getType(),
 128                              0xFF0000ULL),
 129                              "bswap.and3");
 130     Tmp2 = Builder.CreateAnd(Tmp2,
 131                              ConstantInt::get(V->getType(),
 132                              0xFF00ULL),
 133                              "bswap.and2");
 134     Tmp8 = Builder.CreateOr(Tmp8, Tmp7, "bswap.or1");
 135     Tmp6 = Builder.CreateOr(Tmp6, Tmp5, "bswap.or2");
 136     Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or3");
 137     Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or4");
 138     Tmp8 = Builder.CreateOr(Tmp8, Tmp6, "bswap.or5");
 139     Tmp4 = Builder.CreateOr(Tmp4, Tmp2, "bswap.or6");
 140     V = Builder.CreateOr(Tmp8, Tmp4, "bswap.i64");
 141     break;
 142   }
 143   }
 144   return V;
 145 }
 146
 147 /// Emit the code to lower ctpop of V before the specified instruction IP.
 148 static Value *LowerCTPOP(LLVMContext &Context, Value *V, Instruction *IP) {
 149   assert(V->getType()->isIntegerTy() && "Can't ctpop a non-integer type!");
 150
 151   static const uint64_t MaskValues[6] = {
 152     0x5555555555555555ULL, 0x3333333333333333ULL,
 153     0x0F0F0F0F0F0F0F0FULL, 0x00FF00FF00FF00FFULL,
 154     0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL
 155   };
 156
 157   IRBuilder<> Builder(IP);
 158
 159   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
 160   unsigned WordSize = (BitSize + 63) / 64;
 161   Value *Count = ConstantInt::get(V->getType(), 0);
 162
 163   for (unsigned n = 0; n < WordSize; ++n) {
 164     Value *PartValue = V;
 165     for (unsigned i = 1, ct = 0; i < (BitSize>64 ? 64 : BitSize);
 166          i <<= 1, ++ct) {
 167       Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]);
 168       Value *LHS = Builder.CreateAnd(PartValue, MaskCst, "cppop.and1");
 169       Value *VShift = Builder.CreateLShr(PartValue,
 170                                         ConstantInt::get(V->getType(), i),
 171                                          "ctpop.sh");
 172       Value *RHS = Builder.CreateAnd(VShift, MaskCst, "cppop.and2");
 173       PartValue = Builder.CreateAdd(LHS, RHS, "ctpop.step");
 174     }
 175     Count = Builder.CreateAdd(PartValue, Count, "ctpop.part");
 176     if (BitSize > 64) {
 177       V = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 64),
 178                              "ctpop.part.sh");
 179       BitSize -= 64;
 180     }
 181   }
 182
 183   return Count;
 184 }
 185
 186 /// Emit the code to lower ctlz of V before the specified instruction IP.
 187 static Value *LowerCTLZ(LLVMContext &Context, Value *V, Instruction *IP) {
 188
 189   IRBuilder<> Builder(IP);
 190
 191   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
 192   for (unsigned i = 1; i < BitSize; i <<= 1) {
 193     Value *ShVal = ConstantInt::get(V->getType(), i);
 194     ShVal = Builder.CreateLShr(V, ShVal, "ctlz.sh");
 195     V = Builder.CreateOr(V, ShVal, "ctlz.step");
 196   }
 197
 198   V = Builder.CreateNot(V);
 199   return LowerCTPOP(Context, V, IP);
 200 }
 201
 202 static void ReplaceFPIntrinsicWithCall(CallInst *CI, const char *Fname,
 203                                        const char *Dname,
 204                                        const char *LDname) {
 205   switch (CI->getArgOperand(0)->getType()->getTypeID()) {
 206   default: llvm_unreachable("Invalid type in intrinsic");
 207   case Type::FloatTyID:
 208     ReplaceCallWith(Fname, CI, CI->arg_begin(), CI->arg_end(),
 209                     Type::getFloatTy(CI->getContext()));
 210     break;
 211   case Type::DoubleTyID:
 212     ReplaceCallWith(Dname, CI, CI->arg_begin(), CI->arg_end(),
 213                     Type::getDoubleTy(CI->getContext()));
 214     break;
 215   case Type::X86_FP80TyID:
 216   case Type::FP128TyID:
 217   case Type::PPC_FP128TyID:
 218     ReplaceCallWith(LDname, CI, CI->arg_begin(), CI->arg_end(),
 219                     CI->getArgOperand(0)->getType());
 220     break;
 221   }
 222 }
 223
 224 void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
 225   IRBuilder<> Builder(CI);
 226   LLVMContext &Context = CI->getContext();
 227
 228   const Function *Callee = CI->getCalledFunction();
 229   assert(Callee && "Cannot lower an indirect call!");
 230
 231   switch (Callee->getIntrinsicID()) {
 232   case Intrinsic::not_intrinsic:
 233     report_fatal_error("Cannot lower a call to a non-intrinsic function '"+
 234                       Callee->getName() + "'!");
 235   default:
 236     report_fatal_error("Code generator does not support intrinsic function '"+
 237                       Callee->getName()+"'!");
 238
 239   case Intrinsic::expect:
 240   case Intrinsic::expect_with_probability: {
 241     // Just replace __builtin_expect(exp, c) and
 242     // __builtin_expect_with_probability(exp, c, p) with EXP.
 243     Value *V = CI->getArgOperand(0);
 244     CI->replaceAllUsesWith(V);
 245     break;
 246   }
 247
 248   case Intrinsic::allow_runtime_check:
 249   case Intrinsic::allow_ubsan_check:
 250     CI->replaceAllUsesWith(ConstantInt::getTrue(CI->getType()));
 251     return;
 252
 253   case Intrinsic::ctpop:
 254     CI->replaceAllUsesWith(LowerCTPOP(Context, CI->getArgOperand(0), CI));
 255     break;
 256
 257   case Intrinsic::bswap:
 258     CI->replaceAllUsesWith(LowerBSWAP(Context, CI->getArgOperand(0), CI));
 259     break;
 260
 261   case Intrinsic::ctlz:
 262     CI->replaceAllUsesWith(LowerCTLZ(Context, CI->getArgOperand(0), CI));
 263     break;
 264
 265   case Intrinsic::cttz: {
 266     // cttz(x) -> ctpop(~X & (X-1))
 267     Value *Src = CI->getArgOperand(0);
 268     Value *NotSrc = Builder.CreateNot(Src);
 269     NotSrc->setName(Src->getName() + ".not");
 270     Value *SrcM1 = ConstantInt::get(Src->getType(), 1);
 271     SrcM1 = Builder.CreateSub(Src, SrcM1);
 272     Src = LowerCTPOP(Context, Builder.CreateAnd(NotSrc, SrcM1), CI);
 273     CI->replaceAllUsesWith(Src);
 274     break;
 275   }
 276
 277   case Intrinsic::stacksave:
 278   case Intrinsic::stackrestore: {
 279     if (!Warned)
 280       errs() << "WARNING: this target does not support the llvm.stack"
 281              << (Callee->getIntrinsicID() == Intrinsic::stacksave ?
 282                "save" : "restore") << " intrinsic.\n";
 283     Warned = true;
 284     if (Callee->getIntrinsicID() == Intrinsic::stacksave)
 285       CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
 286     break;
 287   }
 288
 289   case Intrinsic::get_dynamic_area_offset:
 290     errs() << "WARNING: this target does not support the custom llvm.get."
 291               "dynamic.area.offset.  It is being lowered to a constant 0\n";
 292     // Just lower it to a constant 0 because for most targets
 293     // @llvm.get.dynamic.area.offset is lowered to zero.
 294     CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 0));
 295     break;
 296   case Intrinsic::returnaddress:
 297   case Intrinsic::frameaddress:
 298     errs() << "WARNING: this target does not support the llvm."
 299            << (Callee->getIntrinsicID() == Intrinsic::returnaddress ?
 300              "return" : "frame") << "address intrinsic.\n";
 301     CI->replaceAllUsesWith(
 302         ConstantPointerNull::get(cast<PointerType>(CI->getType())));
 303     break;
 304   case Intrinsic::addressofreturnaddress:
 305     errs() << "WARNING: this target does not support the "
 306               "llvm.addressofreturnaddress intrinsic.\n";
 307     CI->replaceAllUsesWith(
 308         ConstantPointerNull::get(cast<PointerType>(CI->getType())));
 309     break;
 310
 311   case Intrinsic::prefetch:
 312     break;    // Simply strip out prefetches on unsupported architectures
 313
 314   case Intrinsic::pcmarker:
 315     break;    // Simply strip out pcmarker on unsupported architectures
 316   case Intrinsic::readcyclecounter: {
 317     errs() << "WARNING: this target does not support the llvm.readcyclecoun"
 318            << "ter intrinsic.  It is being lowered to a constant 0\n";
 319     CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
 320     break;
 321   }
 322   case Intrinsic::readsteadycounter: {
 323     errs() << "WARNING: this target does not support the llvm.readsteadycounter"
 324            << " intrinsic.  It is being lowered to a constant 0\n";
 325     CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
 326     break;
 327   }
 328
 329   case Intrinsic::dbg_declare:
 330   case Intrinsic::dbg_label:
 331     break;    // Simply strip out debugging intrinsics
 332
 333   case Intrinsic::eh_typeid_for:
 334     // Return something different to eh_selector.
 335     CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
 336     break;
 337
 338   case Intrinsic::annotation:
 339   case Intrinsic::ptr_annotation:
 340     // Just drop the annotation, but forward the value
 341     CI->replaceAllUsesWith(CI->getOperand(0));
 342     break;
 343
 344   case Intrinsic::assume:
 345   case Intrinsic::experimental_noalias_scope_decl:
 346   case Intrinsic::var_annotation:
 347     break;   // Strip out these intrinsics
 348
 349   case Intrinsic::memcpy: {
 350     Type *IntPtr = DL.getIntPtrType(Context);
 351     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
 352                                         /* isSigned */ false);
 353     Value *Ops[3];
 354     Ops[0] = CI->getArgOperand(0);
 355     Ops[1] = CI->getArgOperand(1);
 356     Ops[2] = Size;
 357     ReplaceCallWith("memcpy", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
 358     break;
 359   }
 360   case Intrinsic::memmove: {
 361     Type *IntPtr = DL.getIntPtrType(Context);
 362     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
 363                                         /* isSigned */ false);
 364     Value *Ops[3];
 365     Ops[0] = CI->getArgOperand(0);
 366     Ops[1] = CI->getArgOperand(1);
 367     Ops[2] = Size;
 368     ReplaceCallWith("memmove", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
 369     break;
 370   }
 371   case Intrinsic::memset: {
 372     Value *Op0 = CI->getArgOperand(0);
 373     Type *IntPtr = DL.getIntPtrType(Op0->getType());
 374     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
 375                                         /* isSigned */ false);
 376     Value *Ops[3];
 377     Ops[0] = Op0;
 378     // Extend the amount to i32.
 379     Ops[1] = Builder.CreateIntCast(CI->getArgOperand(1),
 380                                    Type::getInt32Ty(Context),
 381                                    /* isSigned */ false);
 382     Ops[2] = Size;
 383     ReplaceCallWith("memset", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
 384     break;
 385   }
 386   case Intrinsic::sqrt: {
 387     ReplaceFPIntrinsicWithCall(CI, "sqrtf", "sqrt", "sqrtl");
 388     break;
 389   }
 390   case Intrinsic::log: {
 391     ReplaceFPIntrinsicWithCall(CI, "logf", "log", "logl");
 392     break;
 393   }
 394   case Intrinsic::log2: {
 395     ReplaceFPIntrinsicWithCall(CI, "log2f", "log2", "log2l");
 396     break;
 397   }
 398   case Intrinsic::log10: {
 399     ReplaceFPIntrinsicWithCall(CI, "log10f", "log10", "log10l");
 400     break;
 401   }
 402   case Intrinsic::exp: {
 403     ReplaceFPIntrinsicWithCall(CI, "expf", "exp", "expl");
 404     break;
 405   }
 406   case Intrinsic::exp2: {
 407     ReplaceFPIntrinsicWithCall(CI, "exp2f", "exp2", "exp2l");
 408     break;
 409   }
 410   case Intrinsic::pow: {
 411     ReplaceFPIntrinsicWithCall(CI, "powf", "pow", "powl");
 412     break;
 413   }
 414   case Intrinsic::sin: {
 415     ReplaceFPIntrinsicWithCall(CI, "sinf", "sin", "sinl");
 416     break;
 417   }
 418   case Intrinsic::cos: {
 419     ReplaceFPIntrinsicWithCall(CI, "cosf", "cos", "cosl");
 420     break;
 421   }
 422   case Intrinsic::floor: {
 423     ReplaceFPIntrinsicWithCall(CI, "floorf", "floor", "floorl");
 424     break;
 425   }
 426   case Intrinsic::ceil: {
 427     ReplaceFPIntrinsicWithCall(CI, "ceilf", "ceil", "ceill");
 428     break;
 429   }
 430   case Intrinsic::trunc: {
 431     ReplaceFPIntrinsicWithCall(CI, "truncf", "trunc", "truncl");
 432     break;
 433   }
 434   case Intrinsic::round: {
 435     ReplaceFPIntrinsicWithCall(CI, "roundf", "round", "roundl");
 436     break;
 437   }
 438   case Intrinsic::roundeven: {
 439     ReplaceFPIntrinsicWithCall(CI, "roundevenf", "roundeven", "roundevenl");
 440     break;
 441   }
 442   case Intrinsic::copysign: {
 443     ReplaceFPIntrinsicWithCall(CI, "copysignf", "copysign", "copysignl");
 444     break;
 445   }
 446   case Intrinsic::get_rounding:
 447      // Lower to "round to the nearest"
 448      if (!CI->getType()->isVoidTy())
 449        CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
 450      break;
 451   case Intrinsic::invariant_start:
 452   case Intrinsic::lifetime_start:
 453     // Discard region information.
 454     CI->replaceAllUsesWith(PoisonValue::get(CI->getType()));
 455     break;
 456   case Intrinsic::invariant_end:
 457   case Intrinsic::lifetime_end:
 458     // Discard region information.
 459     break;
 460   }
 461
 462   assert(CI->use_empty() &&
 463          "Lowering should have eliminated any uses of the intrinsic call!");
 464   CI->eraseFromParent();
 465 }
 466
 467 bool IntrinsicLowering::LowerToByteSwap(CallInst *CI) {
 468   // Verify this is a simple bswap.
 469   if (CI->arg_size() != 1 || CI->getType() != CI->getArgOperand(0)->getType() ||
 470       !CI->getType()->isIntegerTy())
 471     return false;
 472
 473   IntegerType *Ty = dyn_cast<IntegerType>(CI->getType());
 474   if (!Ty)
 475     return false;
 476
 477   // Okay, we can do this xform, do so now.
 478   Module *M = CI->getModule();
 479   Function *Int = Intrinsic::getOrInsertDeclaration(M, Intrinsic::bswap, Ty);
 480
 481   Value *Op = CI->getArgOperand(0);
 482   Op = CallInst::Create(Int, Op, CI->getName(), CI->getIterator());
 483
 484   CI->replaceAllUsesWith(Op);
 485   CI->eraseFromParent();
 486   return true;
 487 }