CGBuiltin.cpp revision 24512501319aa6d7f70b1bd79470df0197b832c8
1//===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===// 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 contains code to emit Builtin calls as LLVM code. 11// 12//===----------------------------------------------------------------------===// 13 14#include "CodeGenFunction.h" 15#include "CodeGenModule.h" 16#include "clang/Basic/TargetInfo.h" 17#include "clang/AST/ASTContext.h" 18#include "clang/AST/Builtins.h" 19#include "clang/AST/Expr.h" 20#include "clang/AST/TargetBuiltins.h" 21#include "llvm/Constants.h" 22#include "llvm/Function.h" 23#include "llvm/Intrinsics.h" 24using namespace clang; 25using namespace CodeGen; 26using namespace llvm; 27 28/// Utility to insert an atomic instruction based Instrinsic::ID and 29// the expression node 30static RValue EmitBinaryAtomic(CodeGenFunction& CFG, 31 Intrinsic::ID Id, const CallExpr *E) { 32 const llvm::Type *ResType = CFG.ConvertType(E->getType()); 33 Value *AtomF = CFG.CGM.getIntrinsic(Id, &ResType, 1); 34 return RValue::get(CFG.Builder.CreateCall2(AtomF, 35 CFG.EmitScalarExpr(E->getArg(0)), 36 CFG.EmitScalarExpr(E->getArg(1)))); 37} 38 39RValue CodeGenFunction::EmitBuiltinExpr(unsigned BuiltinID, const CallExpr *E) { 40 switch (BuiltinID) { 41 default: { 42 if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) 43 return EmitCallExpr(CGM.getBuiltinLibFunction(BuiltinID), 44 E->getCallee()->getType(), E->arg_begin(), 45 E->getNumArgs()); 46 47 // See if we have a target specific intrinsic. 48 Intrinsic::ID IntrinsicID; 49 const char *TargetPrefix = Target.getTargetPrefix(); 50 const char *BuiltinName = getContext().BuiltinInfo.GetName(BuiltinID); 51#define GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN 52#include "llvm/Intrinsics.gen" 53#undef GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN 54 55 if (IntrinsicID != Intrinsic::not_intrinsic) { 56 SmallVector<Value*, 16> Args; 57 58 Function *F = CGM.getIntrinsic(IntrinsicID); 59 const llvm::FunctionType *FTy = F->getFunctionType(); 60 61 for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { 62 Value *ArgValue = EmitScalarExpr(E->getArg(i)); 63 64 // If the intrinsic arg type is different from the builtin arg type 65 // we need to do a bit cast. 66 const llvm::Type *PTy = FTy->getParamType(i); 67 if (PTy != ArgValue->getType()) { 68 assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && 69 "Must be able to losslessly bit cast to param"); 70 ArgValue = Builder.CreateBitCast(ArgValue, PTy); 71 } 72 73 Args.push_back(ArgValue); 74 } 75 76 Value *V = Builder.CreateCall(F, &Args[0], &Args[0] + Args.size()); 77 QualType BuiltinRetType = E->getType(); 78 79 const llvm::Type *RetTy = llvm::Type::VoidTy; 80 if (!BuiltinRetType->isVoidType()) RetTy = ConvertType(BuiltinRetType); 81 82 if (RetTy != V->getType()) { 83 assert(V->getType()->canLosslesslyBitCastTo(RetTy) && 84 "Must be able to losslessly bit cast result type"); 85 V = Builder.CreateBitCast(V, RetTy); 86 } 87 88 return RValue::get(V); 89 } 90 91 // See if we have a target specific builtin that needs to be lowered. 92 Value *V = 0; 93 94 if (strcmp(TargetPrefix, "x86") == 0) 95 V = EmitX86BuiltinExpr(BuiltinID, E); 96 else if (strcmp(TargetPrefix, "ppc") == 0) 97 V = EmitPPCBuiltinExpr(BuiltinID, E); 98 99 if (V) 100 return RValue::get(V); 101 102 WarnUnsupported(E, "builtin function"); 103 104 // Unknown builtin, for now just dump it out and return undef. 105 if (hasAggregateLLVMType(E->getType())) 106 return RValue::getAggregate(CreateTempAlloca(ConvertType(E->getType()))); 107 return RValue::get(UndefValue::get(ConvertType(E->getType()))); 108 } 109 case Builtin::BI__builtin___CFStringMakeConstantString: { 110 const Expr *Arg = E->getArg(0); 111 112 while (1) { 113 if (const ParenExpr *PE = dyn_cast<ParenExpr>(Arg)) 114 Arg = PE->getSubExpr(); 115 else if (const ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(Arg)) 116 Arg = CE->getSubExpr(); 117 else 118 break; 119 } 120 121 const StringLiteral *Literal = cast<StringLiteral>(Arg); 122 std::string S(Literal->getStrData(), Literal->getByteLength()); 123 124 return RValue::get(CGM.GetAddrOfConstantCFString(S)); 125 } 126 case Builtin::BI__builtin_va_start: 127 case Builtin::BI__builtin_va_end: { 128 Value *ArgValue = EmitScalarExpr(E->getArg(0)); 129 const llvm::Type *DestType = 130 llvm::PointerType::getUnqual(llvm::Type::Int8Ty); 131 if (ArgValue->getType() != DestType) 132 ArgValue = Builder.CreateBitCast(ArgValue, DestType, 133 ArgValue->getNameStart()); 134 135 Intrinsic::ID inst = (BuiltinID == Builtin::BI__builtin_va_start) ? 136 Intrinsic::vastart : Intrinsic::vaend; 137 return RValue::get(Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue)); 138 } 139 case Builtin::BI__builtin_va_copy: { 140 // FIXME: This does not yet handle architectures where va_list is a struct. 141 Value *DstPtr = EmitScalarExpr(E->getArg(0)); 142 Value *SrcValue = EmitScalarExpr(E->getArg(1)); 143 144 Value *SrcPtr = CreateTempAlloca(SrcValue->getType(), "dst_ptr"); 145 146 // FIXME: Volatile 147 Builder.CreateStore(SrcValue, SrcPtr, false); 148 149 const llvm::Type *Type = 150 llvm::PointerType::getUnqual(llvm::Type::Int8Ty); 151 152 DstPtr = Builder.CreateBitCast(DstPtr, Type); 153 SrcPtr = Builder.CreateBitCast(SrcPtr, Type); 154 return RValue::get(Builder.CreateCall2(CGM.getIntrinsic(Intrinsic::vacopy), 155 DstPtr, SrcPtr)); 156 } 157 case Builtin::BI__builtin_classify_type: { 158 APSInt Result(32); 159 if (!E->isBuiltinClassifyType(Result)) 160 assert(0 && "Expr not __builtin_classify_type!"); 161 return RValue::get(ConstantInt::get(Result)); 162 } 163 case Builtin::BI__builtin_constant_p: { 164 APSInt Result(32); 165 // FIXME: Analyze the parameter and check if it is a constant. 166 Result = 0; 167 return RValue::get(ConstantInt::get(Result)); 168 } 169 case Builtin::BI__builtin_abs: { 170 Value *ArgValue = EmitScalarExpr(E->getArg(0)); 171 172 llvm::BinaryOperator *NegOp = 173 Builder.CreateNeg(ArgValue, (ArgValue->getName() + "neg").c_str()); 174 Value *CmpResult = 175 Builder.CreateICmpSGE(ArgValue, NegOp->getOperand(0), "abscond"); 176 Value *Result = 177 Builder.CreateSelect(CmpResult, ArgValue, NegOp, "abs"); 178 179 return RValue::get(Result); 180 } 181 case Builtin::BI__builtin_ctz: 182 case Builtin::BI__builtin_ctzl: 183 case Builtin::BI__builtin_ctzll: { 184 Value *ArgValue = EmitScalarExpr(E->getArg(0)); 185 186 const llvm::Type *ArgType = ArgValue->getType(); 187 Value *F = CGM.getIntrinsic(Intrinsic::cttz, &ArgType, 1); 188 189 const llvm::Type *ResultType = ConvertType(E->getType()); 190 Value *Result = Builder.CreateCall(F, ArgValue, "tmp"); 191 if (Result->getType() != ResultType) 192 Result = Builder.CreateIntCast(Result, ResultType, "cast"); 193 return RValue::get(Result); 194 } 195 case Builtin::BI__builtin_expect: 196 return RValue::get(EmitScalarExpr(E->getArg(0))); 197 case Builtin::BI__builtin_bswap32: 198 case Builtin::BI__builtin_bswap64: { 199 Value *ArgValue = EmitScalarExpr(E->getArg(0)); 200 const llvm::Type *ArgType = ArgValue->getType(); 201 Value *F = CGM.getIntrinsic(Intrinsic::bswap, &ArgType, 1); 202 return RValue::get(Builder.CreateCall(F, ArgValue, "tmp")); 203 } 204 case Builtin::BI__builtin_inff: { 205 APFloat f(APFloat::IEEEsingle, APFloat::fcInfinity, false); 206 return RValue::get(ConstantFP::get(f)); 207 } 208 case Builtin::BI__builtin_huge_val: 209 case Builtin::BI__builtin_inf: 210 // FIXME: mapping long double onto double. 211 case Builtin::BI__builtin_infl: { 212 APFloat f(APFloat::IEEEdouble, APFloat::fcInfinity, false); 213 return RValue::get(ConstantFP::get(f)); 214 } 215 case Builtin::BI__builtin_isgreater: 216 case Builtin::BI__builtin_isgreaterequal: 217 case Builtin::BI__builtin_isless: 218 case Builtin::BI__builtin_islessequal: 219 case Builtin::BI__builtin_islessgreater: 220 case Builtin::BI__builtin_isunordered: { 221 // Ordered comparisons: we know the arguments to these are matching scalar 222 // floating point values. 223 Value *LHS = EmitScalarExpr(E->getArg(0)); 224 Value *RHS = EmitScalarExpr(E->getArg(1)); 225 226 switch (BuiltinID) { 227 default: assert(0 && "Unknown ordered comparison"); 228 case Builtin::BI__builtin_isgreater: 229 LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); 230 break; 231 case Builtin::BI__builtin_isgreaterequal: 232 LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); 233 break; 234 case Builtin::BI__builtin_isless: 235 LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); 236 break; 237 case Builtin::BI__builtin_islessequal: 238 LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); 239 break; 240 case Builtin::BI__builtin_islessgreater: 241 LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); 242 break; 243 case Builtin::BI__builtin_isunordered: 244 LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); 245 break; 246 } 247 // ZExt bool to int type. 248 return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()), 249 "tmp")); 250 } 251 case Builtin::BI__builtin_alloca: 252 return RValue::get(Builder.CreateAlloca(llvm::Type::Int8Ty, 253 EmitScalarExpr(E->getArg(0)), 254 "tmp")); 255 case Builtin::BI__sync_fetch_and_add: 256 return EmitBinaryAtomic(*this, Intrinsic::atomic_las, E); 257 case Builtin::BI__sync_fetch_and_sub: 258 return EmitBinaryAtomic(*this, Intrinsic::atomic_lss, E); 259 case Builtin::BI__sync_fetch_and_min: 260 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_min, E); 261 case Builtin::BI__sync_fetch_and_max: 262 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_max, E); 263 case Builtin::BI__sync_fetch_and_umin: 264 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_umin, E); 265 case Builtin::BI__sync_fetch_and_umax: 266 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_umax, E); 267 case Builtin::BI__sync_fetch_and_and: 268 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_and, E); 269 case Builtin::BI__sync_fetch_and_or: 270 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_or, E); 271 case Builtin::BI__sync_fetch_and_xor: 272 return EmitBinaryAtomic(*this, Intrinsic::atomic_load_xor, E); 273 case Builtin::BI__sync_val_compare_and_swap: { 274 Value *Args[3]; 275 Args[0]= EmitScalarExpr(E->getArg(0)); 276 Args[1] = EmitScalarExpr(E->getArg(1)); 277 Args[2] = EmitScalarExpr(E->getArg(2)); 278 const llvm::Type *ResType = ConvertType(E->getType()); 279 Value *AtomF = CGM.getIntrinsic(Intrinsic::atomic_lcs, &ResType, 1); 280 return RValue::get(Builder.CreateCall(AtomF, &Args[0], &Args[1]+2)); 281 } 282 case Builtin::BI__sync_lock_test_and_set: 283 return EmitBinaryAtomic(*this, Intrinsic::atomic_swap, E); 284 } 285 return RValue::get(0); 286} 287 288Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, 289 const CallExpr *E) { 290 291 llvm::SmallVector<Value*, 4> Ops; 292 293 for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) 294 Ops.push_back(EmitScalarExpr(E->getArg(i))); 295 296 switch (BuiltinID) { 297 default: return 0; 298 case X86::BI__builtin_ia32_mulps: 299 return Builder.CreateMul(Ops[0], Ops[1], "mulps"); 300 case X86::BI__builtin_ia32_mulpd: 301 return Builder.CreateMul(Ops[0], Ops[1], "mulpd"); 302 case X86::BI__builtin_ia32_pand: 303 case X86::BI__builtin_ia32_pand128: 304 return Builder.CreateAnd(Ops[0], Ops[1], "pand"); 305 case X86::BI__builtin_ia32_por: 306 case X86::BI__builtin_ia32_por128: 307 return Builder.CreateOr(Ops[0], Ops[1], "por"); 308 case X86::BI__builtin_ia32_pxor: 309 case X86::BI__builtin_ia32_pxor128: 310 return Builder.CreateXor(Ops[0], Ops[1], "pxor"); 311 case X86::BI__builtin_ia32_pandn: 312 case X86::BI__builtin_ia32_pandn128: 313 Ops[0] = Builder.CreateNot(Ops[0], "tmp"); 314 return Builder.CreateAnd(Ops[0], Ops[1], "pandn"); 315 case X86::BI__builtin_ia32_paddb: 316 case X86::BI__builtin_ia32_paddb128: 317 case X86::BI__builtin_ia32_paddd: 318 case X86::BI__builtin_ia32_paddd128: 319 case X86::BI__builtin_ia32_paddq: 320 case X86::BI__builtin_ia32_paddq128: 321 case X86::BI__builtin_ia32_paddw: 322 case X86::BI__builtin_ia32_paddw128: 323 case X86::BI__builtin_ia32_addps: 324 case X86::BI__builtin_ia32_addpd: 325 return Builder.CreateAdd(Ops[0], Ops[1], "add"); 326 case X86::BI__builtin_ia32_psubb: 327 case X86::BI__builtin_ia32_psubb128: 328 case X86::BI__builtin_ia32_psubd: 329 case X86::BI__builtin_ia32_psubd128: 330 case X86::BI__builtin_ia32_psubq: 331 case X86::BI__builtin_ia32_psubq128: 332 case X86::BI__builtin_ia32_psubw: 333 case X86::BI__builtin_ia32_psubw128: 334 case X86::BI__builtin_ia32_subps: 335 case X86::BI__builtin_ia32_subpd: 336 return Builder.CreateSub(Ops[0], Ops[1], "sub"); 337 case X86::BI__builtin_ia32_divps: 338 return Builder.CreateFDiv(Ops[0], Ops[1], "divps"); 339 case X86::BI__builtin_ia32_divpd: 340 return Builder.CreateFDiv(Ops[0], Ops[1], "divpd"); 341 case X86::BI__builtin_ia32_pmullw: 342 case X86::BI__builtin_ia32_pmullw128: 343 return Builder.CreateMul(Ops[0], Ops[1], "pmul"); 344 case X86::BI__builtin_ia32_punpckhbw: 345 return EmitShuffleVector(Ops[0], Ops[1], 4, 12, 5, 13, 6, 14, 7, 15, 346 "punpckhbw"); 347 case X86::BI__builtin_ia32_punpckhbw128: 348 return EmitShuffleVector(Ops[0], Ops[1], 8, 24, 9, 25, 10, 26, 11, 27, 349 12, 28, 13, 29, 14, 30, 15, 31, 350 "punpckhbw"); 351 case X86::BI__builtin_ia32_punpckhwd: 352 return EmitShuffleVector(Ops[0], Ops[1], 2, 6, 3, 7, "punpckhwd"); 353 case X86::BI__builtin_ia32_punpckhwd128: 354 return EmitShuffleVector(Ops[0], Ops[1], 4, 12, 5, 13, 6, 14, 7, 15, 355 "punpckhwd"); 356 case X86::BI__builtin_ia32_punpckhdq: 357 return EmitShuffleVector(Ops[0], Ops[1], 1, 3, "punpckhdq"); 358 case X86::BI__builtin_ia32_punpckhdq128: 359 return EmitShuffleVector(Ops[0], Ops[1], 2, 6, 3, 7, "punpckhdq"); 360 case X86::BI__builtin_ia32_punpcklbw: 361 return EmitShuffleVector(Ops[0], Ops[1], 0, 8, 1, 9, 2, 10, 3, 11, 362 "punpcklbw"); 363 case X86::BI__builtin_ia32_punpcklwd: 364 return EmitShuffleVector(Ops[0], Ops[1], 0, 4, 1, 5, "punpcklwd"); 365 case X86::BI__builtin_ia32_punpckldq: 366 return EmitShuffleVector(Ops[0], Ops[1], 0, 2, "punpckldq"); 367 case X86::BI__builtin_ia32_punpckldq128: 368 return EmitShuffleVector(Ops[0], Ops[1], 0, 4, 1, 5, "punpckldq"); 369 case X86::BI__builtin_ia32_pslldi128: 370 case X86::BI__builtin_ia32_psllqi128: 371 case X86::BI__builtin_ia32_psllwi128: 372 case X86::BI__builtin_ia32_psradi128: 373 case X86::BI__builtin_ia32_psrawi128: 374 case X86::BI__builtin_ia32_psrldi128: 375 case X86::BI__builtin_ia32_psrlqi128: 376 case X86::BI__builtin_ia32_psrlwi128: { 377 Ops[1] = Builder.CreateZExt(Ops[1], llvm::Type::Int64Ty, "zext"); 378 const llvm::Type *Ty = llvm::VectorType::get(llvm::Type::Int64Ty, 2); 379 llvm::Value *Zero = llvm::ConstantInt::get(llvm::Type::Int32Ty, 0); 380 Ops[1] = Builder.CreateInsertElement(llvm::UndefValue::get(Ty), 381 Ops[1], Zero, "insert"); 382 Ops[1] = Builder.CreateBitCast(Ops[1], Ops[0]->getType(), "bitcast"); 383 const char *name = 0; 384 Intrinsic::ID ID = Intrinsic::not_intrinsic; 385 386 switch (BuiltinID) { 387 default: assert(0 && "Unsupported shift intrinsic!"); 388 case X86::BI__builtin_ia32_pslldi128: 389 name = "pslldi"; 390 ID = Intrinsic::x86_sse2_psll_d; 391 break; 392 case X86::BI__builtin_ia32_psllqi128: 393 name = "psllqi"; 394 ID = Intrinsic::x86_sse2_psll_q; 395 break; 396 case X86::BI__builtin_ia32_psllwi128: 397 name = "psllwi"; 398 ID = Intrinsic::x86_sse2_psll_w; 399 break; 400 case X86::BI__builtin_ia32_psradi128: 401 name = "psradi"; 402 ID = Intrinsic::x86_sse2_psra_d; 403 break; 404 case X86::BI__builtin_ia32_psrawi128: 405 name = "psrawi"; 406 ID = Intrinsic::x86_sse2_psra_w; 407 break; 408 case X86::BI__builtin_ia32_psrldi128: 409 name = "psrldi"; 410 ID = Intrinsic::x86_sse2_psrl_d; 411 break; 412 case X86::BI__builtin_ia32_psrlqi128: 413 name = "psrlqi"; 414 ID = Intrinsic::x86_sse2_psrl_q; 415 break; 416 case X86::BI__builtin_ia32_psrlwi128: 417 name = "psrlwi"; 418 ID = Intrinsic::x86_sse2_psrl_w; 419 break; 420 } 421 llvm::Function *F = CGM.getIntrinsic(ID); 422 return Builder.CreateCall(F, &Ops[0], &Ops[0] + Ops.size(), name); 423 } 424 case X86::BI__builtin_ia32_pslldi: 425 case X86::BI__builtin_ia32_psllqi: 426 case X86::BI__builtin_ia32_psllwi: 427 case X86::BI__builtin_ia32_psradi: 428 case X86::BI__builtin_ia32_psrawi: 429 case X86::BI__builtin_ia32_psrldi: 430 case X86::BI__builtin_ia32_psrlqi: 431 case X86::BI__builtin_ia32_psrlwi: { 432 Ops[1] = Builder.CreateZExt(Ops[1], llvm::Type::Int64Ty, "zext"); 433 const llvm::Type *Ty = llvm::VectorType::get(llvm::Type::Int64Ty, 1); 434 Ops[1] = Builder.CreateBitCast(Ops[1], Ty, "bitcast"); 435 const char *name = 0; 436 Intrinsic::ID ID = Intrinsic::not_intrinsic; 437 438 switch (BuiltinID) { 439 default: assert(0 && "Unsupported shift intrinsic!"); 440 case X86::BI__builtin_ia32_pslldi: 441 name = "pslldi"; 442 ID = Intrinsic::x86_mmx_psll_d; 443 break; 444 case X86::BI__builtin_ia32_psllqi: 445 name = "psllqi"; 446 ID = Intrinsic::x86_mmx_psll_q; 447 break; 448 case X86::BI__builtin_ia32_psllwi: 449 name = "psllwi"; 450 ID = Intrinsic::x86_mmx_psll_w; 451 break; 452 case X86::BI__builtin_ia32_psradi: 453 name = "psradi"; 454 ID = Intrinsic::x86_mmx_psra_d; 455 break; 456 case X86::BI__builtin_ia32_psrawi: 457 name = "psrawi"; 458 ID = Intrinsic::x86_mmx_psra_w; 459 break; 460 case X86::BI__builtin_ia32_psrldi: 461 name = "psrldi"; 462 ID = Intrinsic::x86_mmx_psrl_d; 463 break; 464 case X86::BI__builtin_ia32_psrlqi: 465 name = "psrlqi"; 466 ID = Intrinsic::x86_mmx_psrl_q; 467 break; 468 case X86::BI__builtin_ia32_psrlwi: 469 name = "psrlwi"; 470 ID = Intrinsic::x86_mmx_psrl_w; 471 break; 472 case X86::BI__builtin_ia32_pslldi128: 473 name = "pslldi"; 474 ID = Intrinsic::x86_sse2_psll_d; 475 break; 476 case X86::BI__builtin_ia32_psllqi128: 477 name = "psllqi"; 478 ID = Intrinsic::x86_sse2_psll_q; 479 break; 480 case X86::BI__builtin_ia32_psllwi128: 481 name = "psllwi"; 482 ID = Intrinsic::x86_sse2_psll_w; 483 break; 484 case X86::BI__builtin_ia32_psradi128: 485 name = "psradi"; 486 ID = Intrinsic::x86_sse2_psra_d; 487 break; 488 case X86::BI__builtin_ia32_psrawi128: 489 name = "psrawi"; 490 ID = Intrinsic::x86_sse2_psra_w; 491 break; 492 case X86::BI__builtin_ia32_psrldi128: 493 name = "psrldi"; 494 ID = Intrinsic::x86_sse2_psrl_d; 495 break; 496 case X86::BI__builtin_ia32_psrlqi128: 497 name = "psrlqi"; 498 ID = Intrinsic::x86_sse2_psrl_q; 499 break; 500 case X86::BI__builtin_ia32_psrlwi128: 501 name = "psrlwi"; 502 ID = Intrinsic::x86_sse2_psrl_w; 503 break; 504 } 505 llvm::Function *F = CGM.getIntrinsic(ID); 506 return Builder.CreateCall(F, &Ops[0], &Ops[0] + Ops.size(), name); 507 } 508 case X86::BI__builtin_ia32_pshuflw: { 509 unsigned i = cast<ConstantInt>(Ops[1])->getZExtValue(); 510 return EmitShuffleVector(Ops[0], Ops[0], 511 i & 0x3, (i & 0xc) >> 2, 512 (i & 0x30) >> 4, (i & 0xc0) >> 6, 4, 5, 6, 7, 513 "pshuflw"); 514 } 515 case X86::BI__builtin_ia32_pshufhw: { 516 unsigned i = cast<ConstantInt>(Ops[1])->getZExtValue(); 517 return EmitShuffleVector(Ops[0], Ops[0], 0, 1, 2, 3, 518 4 + (i & 0x3), 4 + ((i & 0xc) >> 2), 519 4 + ((i & 0x30) >> 4), 4 + ((i & 0xc0) >> 6), 520 "pshufhw"); 521 } 522 case X86::BI__builtin_ia32_pshufd: { 523 unsigned i = cast<ConstantInt>(Ops[1])->getZExtValue(); 524 return EmitShuffleVector(Ops[0], Ops[0], 525 i & 0x3, (i & 0xc) >> 2, 526 (i & 0x30) >> 4, (i & 0xc0) >> 6, 527 "pshufd"); 528 } 529 case X86::BI__builtin_ia32_vec_init_v4hi: 530 case X86::BI__builtin_ia32_vec_init_v8qi: 531 case X86::BI__builtin_ia32_vec_init_v2si: 532 return EmitVector(&Ops[0], Ops.size()); 533 case X86::BI__builtin_ia32_vec_ext_v2si: 534 case X86::BI__builtin_ia32_vec_ext_v2di: 535 case X86::BI__builtin_ia32_vec_ext_v4sf: 536 case X86::BI__builtin_ia32_vec_ext_v4si: 537 case X86::BI__builtin_ia32_vec_ext_v2df: 538 return Builder.CreateExtractElement(Ops[0], Ops[1], "result"); 539 case X86::BI__builtin_ia32_cmpordss: 540 case X86::BI__builtin_ia32_cmpordsd: 541 case X86::BI__builtin_ia32_cmpunordss: 542 case X86::BI__builtin_ia32_cmpunordsd: 543 case X86::BI__builtin_ia32_cmpeqss: 544 case X86::BI__builtin_ia32_cmpeqsd: 545 case X86::BI__builtin_ia32_cmpltss: 546 case X86::BI__builtin_ia32_cmpltsd: 547 case X86::BI__builtin_ia32_cmpless: 548 case X86::BI__builtin_ia32_cmplesd: 549 case X86::BI__builtin_ia32_cmpneqss: 550 case X86::BI__builtin_ia32_cmpneqsd: 551 case X86::BI__builtin_ia32_cmpnltss: 552 case X86::BI__builtin_ia32_cmpnltsd: 553 case X86::BI__builtin_ia32_cmpnless: 554 case X86::BI__builtin_ia32_cmpnlesd: { 555 unsigned i = 0; 556 const char *name = 0; 557 switch (BuiltinID) { 558 default: assert(0 && "Unknown compare builtin!"); 559 case X86::BI__builtin_ia32_cmpeqss: 560 case X86::BI__builtin_ia32_cmpeqsd: 561 i = 0; 562 name = "cmpeq"; 563 break; 564 case X86::BI__builtin_ia32_cmpltss: 565 case X86::BI__builtin_ia32_cmpltsd: 566 i = 1; 567 name = "cmplt"; 568 break; 569 case X86::BI__builtin_ia32_cmpless: 570 case X86::BI__builtin_ia32_cmplesd: 571 i = 2; 572 name = "cmple"; 573 break; 574 case X86::BI__builtin_ia32_cmpunordss: 575 case X86::BI__builtin_ia32_cmpunordsd: 576 i = 3; 577 name = "cmpunord"; 578 break; 579 case X86::BI__builtin_ia32_cmpneqss: 580 case X86::BI__builtin_ia32_cmpneqsd: 581 i = 4; 582 name = "cmpneq"; 583 break; 584 case X86::BI__builtin_ia32_cmpnltss: 585 case X86::BI__builtin_ia32_cmpnltsd: 586 i = 5; 587 name = "cmpntl"; 588 break; 589 case X86::BI__builtin_ia32_cmpnless: 590 case X86::BI__builtin_ia32_cmpnlesd: 591 i = 6; 592 name = "cmpnle"; 593 break; 594 case X86::BI__builtin_ia32_cmpordss: 595 case X86::BI__builtin_ia32_cmpordsd: 596 i = 7; 597 name = "cmpord"; 598 break; 599 } 600 601 llvm::Function *F; 602 if (cast<llvm::VectorType>(Ops[0]->getType())->getElementType() == 603 llvm::Type::FloatTy) 604 F = CGM.getIntrinsic(Intrinsic::x86_sse_cmp_ss); 605 else 606 F = CGM.getIntrinsic(Intrinsic::x86_sse2_cmp_sd); 607 608 Ops.push_back(llvm::ConstantInt::get(llvm::Type::Int8Ty, i)); 609 return Builder.CreateCall(F, &Ops[0], &Ops[0] + Ops.size(), name); 610 } 611 case X86::BI__builtin_ia32_ldmxcsr: { 612 llvm::Type *PtrTy = llvm::PointerType::getUnqual(llvm::Type::Int8Ty); 613 Value *One = llvm::ConstantInt::get(llvm::Type::Int32Ty, 1); 614 Value *Tmp = Builder.CreateAlloca(llvm::Type::Int32Ty, One, "tmp"); 615 Builder.CreateStore(Ops[0], Tmp); 616 return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr), 617 Builder.CreateBitCast(Tmp, PtrTy)); 618 } 619 case X86::BI__builtin_ia32_stmxcsr: { 620 llvm::Type *PtrTy = llvm::PointerType::getUnqual(llvm::Type::Int8Ty); 621 Value *One = llvm::ConstantInt::get(llvm::Type::Int32Ty, 1); 622 Value *Tmp = Builder.CreateAlloca(llvm::Type::Int32Ty, One, "tmp"); 623 One = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr), 624 Builder.CreateBitCast(Tmp, PtrTy)); 625 return Builder.CreateLoad(Tmp, "stmxcsr"); 626 } 627 case X86::BI__builtin_ia32_cmpordps: 628 case X86::BI__builtin_ia32_cmpordpd: 629 case X86::BI__builtin_ia32_cmpunordps: 630 case X86::BI__builtin_ia32_cmpunordpd: 631 case X86::BI__builtin_ia32_cmpeqps: 632 case X86::BI__builtin_ia32_cmpeqpd: 633 case X86::BI__builtin_ia32_cmpltps: 634 case X86::BI__builtin_ia32_cmpltpd: 635 case X86::BI__builtin_ia32_cmpleps: 636 case X86::BI__builtin_ia32_cmplepd: 637 case X86::BI__builtin_ia32_cmpneqps: 638 case X86::BI__builtin_ia32_cmpneqpd: 639 case X86::BI__builtin_ia32_cmpngtps: 640 case X86::BI__builtin_ia32_cmpngtpd: 641 case X86::BI__builtin_ia32_cmpnltps: 642 case X86::BI__builtin_ia32_cmpnltpd: 643 case X86::BI__builtin_ia32_cmpgtps: 644 case X86::BI__builtin_ia32_cmpgtpd: 645 case X86::BI__builtin_ia32_cmpgeps: 646 case X86::BI__builtin_ia32_cmpgepd: 647 case X86::BI__builtin_ia32_cmpngeps: 648 case X86::BI__builtin_ia32_cmpngepd: 649 case X86::BI__builtin_ia32_cmpnleps: 650 case X86::BI__builtin_ia32_cmpnlepd: { 651 unsigned i = 0; 652 const char *name = 0; 653 bool ShouldSwap = false; 654 switch (BuiltinID) { 655 default: assert(0 && "Unknown compare builtin!"); 656 case X86::BI__builtin_ia32_cmpeqps: 657 case X86::BI__builtin_ia32_cmpeqpd: i = 0; name = "cmpeq"; break; 658 case X86::BI__builtin_ia32_cmpltps: 659 case X86::BI__builtin_ia32_cmpltpd: i = 1; name = "cmplt"; break; 660 case X86::BI__builtin_ia32_cmpleps: 661 case X86::BI__builtin_ia32_cmplepd: i = 2; name = "cmple"; break; 662 case X86::BI__builtin_ia32_cmpunordps: 663 case X86::BI__builtin_ia32_cmpunordpd: i = 3; name = "cmpunord"; break; 664 case X86::BI__builtin_ia32_cmpneqps: 665 case X86::BI__builtin_ia32_cmpneqpd: i = 4; name = "cmpneq"; break; 666 case X86::BI__builtin_ia32_cmpnltps: 667 case X86::BI__builtin_ia32_cmpnltpd: i = 5; name = "cmpntl"; break; 668 case X86::BI__builtin_ia32_cmpnleps: 669 case X86::BI__builtin_ia32_cmpnlepd: i = 6; name = "cmpnle"; break; 670 case X86::BI__builtin_ia32_cmpordps: 671 case X86::BI__builtin_ia32_cmpordpd: i = 7; name = "cmpord"; break; 672 case X86::BI__builtin_ia32_cmpgtps: 673 case X86::BI__builtin_ia32_cmpgtpd: 674 ShouldSwap = true; 675 i = 1; 676 name = "cmpgt"; 677 break; 678 case X86::BI__builtin_ia32_cmpgeps: 679 case X86::BI__builtin_ia32_cmpgepd: 680 i = 2; 681 name = "cmpge"; 682 ShouldSwap = true; 683 break; 684 case X86::BI__builtin_ia32_cmpngtps: 685 case X86::BI__builtin_ia32_cmpngtpd: 686 i = 5; 687 name = "cmpngt"; 688 ShouldSwap = true; 689 break; 690 case X86::BI__builtin_ia32_cmpngeps: 691 case X86::BI__builtin_ia32_cmpngepd: 692 i = 6; 693 name = "cmpnge"; 694 ShouldSwap = true; 695 break; 696 } 697 698 if (ShouldSwap) 699 std::swap(Ops[0], Ops[1]); 700 701 llvm::Function *F; 702 if (cast<llvm::VectorType>(Ops[0]->getType())->getElementType() == 703 llvm::Type::FloatTy) 704 F = CGM.getIntrinsic(Intrinsic::x86_sse_cmp_ps); 705 else 706 F = CGM.getIntrinsic(Intrinsic::x86_sse2_cmp_pd); 707 708 Ops.push_back(llvm::ConstantInt::get(llvm::Type::Int8Ty, i)); 709 return Builder.CreateCall(F, &Ops[0], &Ops[0] + Ops.size(), name); 710 } 711 case X86::BI__builtin_ia32_movss: 712 return EmitShuffleVector(Ops[0], Ops[1], 4, 1, 2, 3, "movss"); 713 case X86::BI__builtin_ia32_shufps: { 714 unsigned i = cast<ConstantInt>(Ops[2])->getZExtValue(); 715 return EmitShuffleVector(Ops[0], Ops[1], 716 i & 0x3, (i & 0xc) >> 2, 717 ((i & 0x30) >> 4) + 4, 718 ((i & 0x60) >> 6) + 4, "shufps"); 719 } 720 case X86::BI__builtin_ia32_punpcklbw128: 721 return EmitShuffleVector(Ops[0], Ops[1], 0, 16, 1, 17, 2, 18, 3, 19, 722 4, 20, 5, 21, 6, 22, 7, 23, 723 "punpcklbw"); 724 case X86::BI__builtin_ia32_punpcklwd128: 725 return EmitShuffleVector(Ops[0], Ops[1], 0, 8, 1, 9, 2, 10, 3, 11, 726 "punpcklwd"); 727 case X86::BI__builtin_ia32_movlhps: 728 return EmitShuffleVector(Ops[0], Ops[1], 0, 1, 4, 5, "movlhps"); 729 case X86::BI__builtin_ia32_movhlps: 730 return EmitShuffleVector(Ops[0], Ops[1], 6, 7, 2, 3, "movhlps"); 731 case X86::BI__builtin_ia32_unpckhps: 732 return EmitShuffleVector(Ops[0], Ops[1], 2, 6, 3, 7, "unpckhps"); 733 case X86::BI__builtin_ia32_unpcklps: 734 return EmitShuffleVector(Ops[0], Ops[1], 0, 4, 1, 5, "unpcklps"); 735 case X86::BI__builtin_ia32_movqv4si: { 736 llvm::Type *Ty = llvm::VectorType::get(llvm::Type::Int64Ty, 2); 737 return Builder.CreateBitCast(Ops[0], Ty); 738 } 739 case X86::BI__builtin_ia32_loadlps: 740 case X86::BI__builtin_ia32_loadhps: { 741 // FIXME: This should probably be represented as 742 // shuffle (dst, (v4f32 (insert undef, (load i64), 0)), shuf mask hi/lo) 743 const llvm::Type *EltTy = llvm::Type::DoubleTy; 744 const llvm::Type *VecTy = llvm::VectorType::get(EltTy, 2); 745 const llvm::Type *OrigTy = Ops[0]->getType(); 746 unsigned Index = BuiltinID == X86::BI__builtin_ia32_loadlps ? 0 : 1; 747 llvm::Value *Idx = llvm::ConstantInt::get(llvm::Type::Int32Ty, Index); 748 Ops[1] = Builder.CreateBitCast(Ops[1], llvm::PointerType::getUnqual(EltTy)); 749 Ops[1] = Builder.CreateLoad(Ops[1], "tmp"); 750 Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast"); 751 Ops[0] = Builder.CreateInsertElement(Ops[0], Ops[1], Idx, "loadps"); 752 return Builder.CreateBitCast(Ops[0], OrigTy, "loadps"); 753 } 754 case X86::BI__builtin_ia32_storehps: 755 case X86::BI__builtin_ia32_storelps: { 756 const llvm::Type *EltTy = llvm::Type::Int64Ty; 757 llvm::Type *PtrTy = llvm::PointerType::getUnqual(EltTy); 758 llvm::Type *VecTy = llvm::VectorType::get(EltTy, 2); 759 760 // cast val v2i64 761 Ops[1] = Builder.CreateBitCast(Ops[1], VecTy, "cast"); 762 763 // extract (0, 1) 764 unsigned Index = BuiltinID == X86::BI__builtin_ia32_storelps ? 0 : 1; 765 llvm::Value *Idx = llvm::ConstantInt::get(llvm::Type::Int32Ty, Index); 766 Ops[1] = Builder.CreateExtractElement(Ops[1], Idx, "extract"); 767 768 // cast pointer to i64 & store 769 Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy); 770 return Builder.CreateStore(Ops[1], Ops[0]); 771 } 772 case X86::BI__builtin_ia32_loadlv4si: { 773 // load i64 774 const llvm::Type *EltTy = llvm::Type::Int64Ty; 775 llvm::Type *PtrTy = llvm::PointerType::getUnqual(EltTy); 776 Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy); 777 Ops[0] = Builder.CreateLoad(Ops[0], "load"); 778 779 // scalar to vector: insert i64 into 2 x i64 undef 780 llvm::Type *VecTy = llvm::VectorType::get(EltTy, 2); 781 llvm::Value *Zero = llvm::ConstantInt::get(llvm::Type::Int32Ty, 0); 782 Ops[0] = Builder.CreateInsertElement(llvm::UndefValue::get(VecTy), 783 Ops[0], Zero, "s2v"); 784 785 // shuffle into zero vector. 786 std::vector<llvm::Constant *>Elts; 787 Elts.resize(2, llvm::ConstantInt::get(EltTy, 0)); 788 llvm::Value *ZV = ConstantVector::get(Elts); 789 Ops[0] = EmitShuffleVector(ZV, Ops[0], 2, 1, "loadl"); 790 791 // bitcast to result. 792 return Builder.CreateBitCast(Ops[0], 793 llvm::VectorType::get(llvm::Type::Int32Ty, 4)); 794 } 795 case X86::BI__builtin_ia32_vec_set_v4hi: 796 case X86::BI__builtin_ia32_vec_set_v8hi: 797 return Builder.CreateInsertElement(Ops[0], Ops[1], Ops[2], "pinsrw"); 798 case X86::BI__builtin_ia32_andps: 799 case X86::BI__builtin_ia32_andpd: 800 case X86::BI__builtin_ia32_andnps: 801 case X86::BI__builtin_ia32_andnpd: 802 case X86::BI__builtin_ia32_orps: 803 case X86::BI__builtin_ia32_orpd: 804 case X86::BI__builtin_ia32_xorpd: 805 case X86::BI__builtin_ia32_xorps: { 806 const llvm::Type *ITy = llvm::VectorType::get(llvm::Type::Int32Ty, 4); 807 const llvm::Type *FTy = Ops[0]->getType(); 808 Ops[0] = Builder.CreateBitCast(Ops[0], ITy, "bitcast"); 809 Ops[1] = Builder.CreateBitCast(Ops[1], ITy, "bitcast"); 810 switch (BuiltinID) { 811 case X86::BI__builtin_ia32_andps: 812 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1], "andps"); 813 break; 814 case X86::BI__builtin_ia32_andpd: 815 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1], "andpd"); 816 break; 817 case X86::BI__builtin_ia32_andnps: 818 Ops[0] = Builder.CreateNot(Ops[0], "not"); 819 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1], "andnps"); 820 break; 821 case X86::BI__builtin_ia32_andnpd: 822 Ops[0] = Builder.CreateNot(Ops[0], "not"); 823 Ops[0] = Builder.CreateAnd(Ops[0], Ops[1], "andnpd"); 824 break; 825 case X86::BI__builtin_ia32_orps: 826 Ops[0] = Builder.CreateOr(Ops[0], Ops[1], "orps"); 827 break; 828 case X86::BI__builtin_ia32_orpd: 829 Ops[0] = Builder.CreateOr(Ops[0], Ops[1], "orpd"); 830 break; 831 case X86::BI__builtin_ia32_xorps: 832 Ops[0] = Builder.CreateXor(Ops[0], Ops[1], "xorps"); 833 break; 834 case X86::BI__builtin_ia32_xorpd: 835 Ops[0] = Builder.CreateXor(Ops[0], Ops[1], "xorpd"); 836 break; 837 } 838 return Builder.CreateBitCast(Ops[0], FTy, "bitcast"); 839 } 840 } 841} 842 843Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID, 844 const CallExpr *E) { 845 switch (BuiltinID) { 846 default: return 0; 847 } 848} 849