1 //===-- CommandFlags.cpp - Command Line Flags Interface ---------*- C++ -*-===// 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 codegen-specific flags that are shared between different 10 // command line tools. The tools "llc" and "opt" both use this file to prevent 11 // flag duplication. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "llvm/CodeGen/CommandFlags.h" 16 #include "llvm/IR/Module.h" 17 #include "llvm/MC/SubtargetFeature.h" 18 #include "llvm/Support/CommandLine.h" 19 #include "llvm/Support/Host.h" 20 21 using namespace llvm; 22 23 #define CGOPT(TY, NAME) \ 24 static cl::opt<TY> *NAME##View; \ 25 TY codegen::get##NAME() { \ 26 assert(NAME##View && "RegisterCodeGenFlags not created."); \ 27 return *NAME##View; \ 28 } 29 30 #define CGLIST(TY, NAME) \ 31 static cl::list<TY> *NAME##View; \ 32 std::vector<TY> codegen::get##NAME() { \ 33 assert(NAME##View && "RegisterCodeGenFlags not created."); \ 34 return *NAME##View; \ 35 } 36 37 #define CGOPT_EXP(TY, NAME) \ 38 CGOPT(TY, NAME) \ 39 Optional<TY> codegen::getExplicit##NAME() { \ 40 if (NAME##View->getNumOccurrences()) { \ 41 TY res = *NAME##View; \ 42 return res; \ 43 } \ 44 return None; \ 45 } 46 47 CGOPT(std::string, MArch) 48 CGOPT(std::string, MCPU) 49 CGLIST(std::string, MAttrs) 50 CGOPT_EXP(Reloc::Model, RelocModel) 51 CGOPT(ThreadModel::Model, ThreadModel) 52 CGOPT_EXP(CodeModel::Model, CodeModel) 53 CGOPT(ExceptionHandling, ExceptionModel) 54 CGOPT_EXP(CodeGenFileType, FileType) 55 CGOPT(FramePointer::FP, FramePointerUsage) 56 CGOPT(bool, EnableUnsafeFPMath) 57 CGOPT(bool, EnableNoInfsFPMath) 58 CGOPT(bool, EnableNoNaNsFPMath) 59 CGOPT(bool, EnableNoSignedZerosFPMath) 60 CGOPT(bool, EnableNoTrappingFPMath) 61 CGOPT(DenormalMode::DenormalModeKind, DenormalFPMath) 62 CGOPT(DenormalMode::DenormalModeKind, DenormalFP32Math) 63 CGOPT(bool, EnableHonorSignDependentRoundingFPMath) 64 CGOPT(FloatABI::ABIType, FloatABIForCalls) 65 CGOPT(FPOpFusion::FPOpFusionMode, FuseFPOps) 66 CGOPT(bool, DontPlaceZerosInBSS) 67 CGOPT(bool, EnableGuaranteedTailCallOpt) 68 CGOPT(bool, DisableTailCalls) 69 CGOPT(bool, StackSymbolOrdering) 70 CGOPT(unsigned, OverrideStackAlignment) 71 CGOPT(bool, StackRealign) 72 CGOPT(std::string, TrapFuncName) 73 CGOPT(bool, UseCtors) 74 CGOPT(bool, RelaxELFRelocations) 75 CGOPT_EXP(bool, DataSections) 76 CGOPT_EXP(bool, FunctionSections) 77 CGOPT(std::string, BBSections) 78 CGOPT(unsigned, TLSSize) 79 CGOPT(bool, EmulatedTLS) 80 CGOPT(bool, UniqueSectionNames) 81 CGOPT(bool, UniqueBasicBlockSectionNames) 82 CGOPT(EABI, EABIVersion) 83 CGOPT(DebuggerKind, DebuggerTuningOpt) 84 CGOPT(bool, EnableStackSizeSection) 85 CGOPT(bool, EnableAddrsig) 86 CGOPT(bool, EmitCallSiteInfo) 87 CGOPT(bool, EnableMachineFunctionSplitter) 88 CGOPT(bool, EnableDebugEntryValues) 89 CGOPT(bool, ValueTrackingVariableLocations) 90 CGOPT(bool, ForceDwarfFrameSection) 91 CGOPT(bool, XRayOmitFunctionIndex) 92 93 codegen::RegisterCodeGenFlags::RegisterCodeGenFlags() { 94 #define CGBINDOPT(NAME) \ 95 do { \ 96 NAME##View = std::addressof(NAME); \ 97 } while (0) 98 99 static cl::opt<std::string> MArch( 100 "march", cl::desc("Architecture to generate code for (see --version)")); 101 CGBINDOPT(MArch); 102 103 static cl::opt<std::string> MCPU( 104 "mcpu", cl::desc("Target a specific cpu type (-mcpu=help for details)"), 105 cl::value_desc("cpu-name"), cl::init("")); 106 CGBINDOPT(MCPU); 107 108 static cl::list<std::string> MAttrs( 109 "mattr", cl::CommaSeparated, 110 cl::desc("Target specific attributes (-mattr=help for details)"), 111 cl::value_desc("a1,+a2,-a3,...")); 112 CGBINDOPT(MAttrs); 113 114 static cl::opt<Reloc::Model> RelocModel( 115 "relocation-model", cl::desc("Choose relocation model"), 116 cl::values( 117 clEnumValN(Reloc::Static, "static", "Non-relocatable code"), 118 clEnumValN(Reloc::PIC_, "pic", 119 "Fully relocatable, position independent code"), 120 clEnumValN(Reloc::DynamicNoPIC, "dynamic-no-pic", 121 "Relocatable external references, non-relocatable code"), 122 clEnumValN( 123 Reloc::ROPI, "ropi", 124 "Code and read-only data relocatable, accessed PC-relative"), 125 clEnumValN( 126 Reloc::RWPI, "rwpi", 127 "Read-write data relocatable, accessed relative to static base"), 128 clEnumValN(Reloc::ROPI_RWPI, "ropi-rwpi", 129 "Combination of ropi and rwpi"))); 130 CGBINDOPT(RelocModel); 131 132 static cl::opt<ThreadModel::Model> ThreadModel( 133 "thread-model", cl::desc("Choose threading model"), 134 cl::init(ThreadModel::POSIX), 135 cl::values( 136 clEnumValN(ThreadModel::POSIX, "posix", "POSIX thread model"), 137 clEnumValN(ThreadModel::Single, "single", "Single thread model"))); 138 CGBINDOPT(ThreadModel); 139 140 static cl::opt<CodeModel::Model> CodeModel( 141 "code-model", cl::desc("Choose code model"), 142 cl::values(clEnumValN(CodeModel::Tiny, "tiny", "Tiny code model"), 143 clEnumValN(CodeModel::Small, "small", "Small code model"), 144 clEnumValN(CodeModel::Kernel, "kernel", "Kernel code model"), 145 clEnumValN(CodeModel::Medium, "medium", "Medium code model"), 146 clEnumValN(CodeModel::Large, "large", "Large code model"))); 147 CGBINDOPT(CodeModel); 148 149 static cl::opt<ExceptionHandling> ExceptionModel( 150 "exception-model", cl::desc("exception model"), 151 cl::init(ExceptionHandling::None), 152 cl::values( 153 clEnumValN(ExceptionHandling::None, "default", 154 "default exception handling model"), 155 clEnumValN(ExceptionHandling::DwarfCFI, "dwarf", 156 "DWARF-like CFI based exception handling"), 157 clEnumValN(ExceptionHandling::SjLj, "sjlj", 158 "SjLj exception handling"), 159 clEnumValN(ExceptionHandling::ARM, "arm", "ARM EHABI exceptions"), 160 clEnumValN(ExceptionHandling::WinEH, "wineh", 161 "Windows exception model"), 162 clEnumValN(ExceptionHandling::Wasm, "wasm", 163 "WebAssembly exception handling"))); 164 CGBINDOPT(ExceptionModel); 165 166 static cl::opt<CodeGenFileType> FileType( 167 "filetype", cl::init(CGFT_AssemblyFile), 168 cl::desc( 169 "Choose a file type (not all types are supported by all targets):"), 170 cl::values( 171 clEnumValN(CGFT_AssemblyFile, "asm", "Emit an assembly ('.s') file"), 172 clEnumValN(CGFT_ObjectFile, "obj", 173 "Emit a native object ('.o') file"), 174 clEnumValN(CGFT_Null, "null", 175 "Emit nothing, for performance testing"))); 176 CGBINDOPT(FileType); 177 178 static cl::opt<FramePointer::FP> FramePointerUsage( 179 "frame-pointer", 180 cl::desc("Specify frame pointer elimination optimization"), 181 cl::init(FramePointer::None), 182 cl::values( 183 clEnumValN(FramePointer::All, "all", 184 "Disable frame pointer elimination"), 185 clEnumValN(FramePointer::NonLeaf, "non-leaf", 186 "Disable frame pointer elimination for non-leaf frame"), 187 clEnumValN(FramePointer::None, "none", 188 "Enable frame pointer elimination"))); 189 CGBINDOPT(FramePointerUsage); 190 191 static cl::opt<bool> EnableUnsafeFPMath( 192 "enable-unsafe-fp-math", 193 cl::desc("Enable optimizations that may decrease FP precision"), 194 cl::init(false)); 195 CGBINDOPT(EnableUnsafeFPMath); 196 197 static cl::opt<bool> EnableNoInfsFPMath( 198 "enable-no-infs-fp-math", 199 cl::desc("Enable FP math optimizations that assume no +-Infs"), 200 cl::init(false)); 201 CGBINDOPT(EnableNoInfsFPMath); 202 203 static cl::opt<bool> EnableNoNaNsFPMath( 204 "enable-no-nans-fp-math", 205 cl::desc("Enable FP math optimizations that assume no NaNs"), 206 cl::init(false)); 207 CGBINDOPT(EnableNoNaNsFPMath); 208 209 static cl::opt<bool> EnableNoSignedZerosFPMath( 210 "enable-no-signed-zeros-fp-math", 211 cl::desc("Enable FP math optimizations that assume " 212 "the sign of 0 is insignificant"), 213 cl::init(false)); 214 CGBINDOPT(EnableNoSignedZerosFPMath); 215 216 static cl::opt<bool> EnableNoTrappingFPMath( 217 "enable-no-trapping-fp-math", 218 cl::desc("Enable setting the FP exceptions build " 219 "attribute not to use exceptions"), 220 cl::init(false)); 221 CGBINDOPT(EnableNoTrappingFPMath); 222 223 static const auto DenormFlagEnumOptions = 224 cl::values(clEnumValN(DenormalMode::IEEE, "ieee", 225 "IEEE 754 denormal numbers"), 226 clEnumValN(DenormalMode::PreserveSign, "preserve-sign", 227 "the sign of a flushed-to-zero number is preserved " 228 "in the sign of 0"), 229 clEnumValN(DenormalMode::PositiveZero, "positive-zero", 230 "denormals are flushed to positive zero")); 231 232 // FIXME: Doesn't have way to specify separate input and output modes. 233 static cl::opt<DenormalMode::DenormalModeKind> DenormalFPMath( 234 "denormal-fp-math", 235 cl::desc("Select which denormal numbers the code is permitted to require"), 236 cl::init(DenormalMode::IEEE), 237 DenormFlagEnumOptions); 238 CGBINDOPT(DenormalFPMath); 239 240 static cl::opt<DenormalMode::DenormalModeKind> DenormalFP32Math( 241 "denormal-fp-math-f32", 242 cl::desc("Select which denormal numbers the code is permitted to require for float"), 243 cl::init(DenormalMode::Invalid), 244 DenormFlagEnumOptions); 245 CGBINDOPT(DenormalFP32Math); 246 247 static cl::opt<bool> EnableHonorSignDependentRoundingFPMath( 248 "enable-sign-dependent-rounding-fp-math", cl::Hidden, 249 cl::desc("Force codegen to assume rounding mode can change dynamically"), 250 cl::init(false)); 251 CGBINDOPT(EnableHonorSignDependentRoundingFPMath); 252 253 static cl::opt<FloatABI::ABIType> FloatABIForCalls( 254 "float-abi", cl::desc("Choose float ABI type"), 255 cl::init(FloatABI::Default), 256 cl::values(clEnumValN(FloatABI::Default, "default", 257 "Target default float ABI type"), 258 clEnumValN(FloatABI::Soft, "soft", 259 "Soft float ABI (implied by -soft-float)"), 260 clEnumValN(FloatABI::Hard, "hard", 261 "Hard float ABI (uses FP registers)"))); 262 CGBINDOPT(FloatABIForCalls); 263 264 static cl::opt<FPOpFusion::FPOpFusionMode> FuseFPOps( 265 "fp-contract", cl::desc("Enable aggressive formation of fused FP ops"), 266 cl::init(FPOpFusion::Standard), 267 cl::values( 268 clEnumValN(FPOpFusion::Fast, "fast", 269 "Fuse FP ops whenever profitable"), 270 clEnumValN(FPOpFusion::Standard, "on", "Only fuse 'blessed' FP ops."), 271 clEnumValN(FPOpFusion::Strict, "off", 272 "Only fuse FP ops when the result won't be affected."))); 273 CGBINDOPT(FuseFPOps); 274 275 static cl::opt<bool> DontPlaceZerosInBSS( 276 "nozero-initialized-in-bss", 277 cl::desc("Don't place zero-initialized symbols into bss section"), 278 cl::init(false)); 279 CGBINDOPT(DontPlaceZerosInBSS); 280 281 static cl::opt<bool> EnableGuaranteedTailCallOpt( 282 "tailcallopt", 283 cl::desc( 284 "Turn fastcc calls into tail calls by (potentially) changing ABI."), 285 cl::init(false)); 286 CGBINDOPT(EnableGuaranteedTailCallOpt); 287 288 static cl::opt<bool> DisableTailCalls( 289 "disable-tail-calls", cl::desc("Never emit tail calls"), cl::init(false)); 290 CGBINDOPT(DisableTailCalls); 291 292 static cl::opt<bool> StackSymbolOrdering( 293 "stack-symbol-ordering", cl::desc("Order local stack symbols."), 294 cl::init(true)); 295 CGBINDOPT(StackSymbolOrdering); 296 297 static cl::opt<unsigned> OverrideStackAlignment( 298 "stack-alignment", cl::desc("Override default stack alignment"), 299 cl::init(0)); 300 CGBINDOPT(OverrideStackAlignment); 301 302 static cl::opt<bool> StackRealign( 303 "stackrealign", 304 cl::desc("Force align the stack to the minimum alignment"), 305 cl::init(false)); 306 CGBINDOPT(StackRealign); 307 308 static cl::opt<std::string> TrapFuncName( 309 "trap-func", cl::Hidden, 310 cl::desc("Emit a call to trap function rather than a trap instruction"), 311 cl::init("")); 312 CGBINDOPT(TrapFuncName); 313 314 static cl::opt<bool> UseCtors("use-ctors", 315 cl::desc("Use .ctors instead of .init_array."), 316 cl::init(false)); 317 CGBINDOPT(UseCtors); 318 319 static cl::opt<bool> RelaxELFRelocations( 320 "relax-elf-relocations", 321 cl::desc( 322 "Emit GOTPCRELX/REX_GOTPCRELX instead of GOTPCREL on x86-64 ELF"), 323 cl::init(false)); 324 CGBINDOPT(RelaxELFRelocations); 325 326 static cl::opt<bool> DataSections( 327 "data-sections", cl::desc("Emit data into separate sections"), 328 cl::init(false)); 329 CGBINDOPT(DataSections); 330 331 static cl::opt<bool> FunctionSections( 332 "function-sections", cl::desc("Emit functions into separate sections"), 333 cl::init(false)); 334 CGBINDOPT(FunctionSections); 335 336 static cl::opt<std::string> BBSections( 337 "basic-block-sections", 338 cl::desc("Emit basic blocks into separate sections"), 339 cl::value_desc("all | <function list (file)> | labels | none"), 340 cl::init("none")); 341 CGBINDOPT(BBSections); 342 343 static cl::opt<unsigned> TLSSize( 344 "tls-size", cl::desc("Bit size of immediate TLS offsets"), cl::init(0)); 345 CGBINDOPT(TLSSize); 346 347 static cl::opt<bool> EmulatedTLS( 348 "emulated-tls", cl::desc("Use emulated TLS model"), cl::init(false)); 349 CGBINDOPT(EmulatedTLS); 350 351 static cl::opt<bool> UniqueSectionNames( 352 "unique-section-names", cl::desc("Give unique names to every section"), 353 cl::init(true)); 354 CGBINDOPT(UniqueSectionNames); 355 356 static cl::opt<bool> UniqueBasicBlockSectionNames( 357 "unique-basic-block-section-names", 358 cl::desc("Give unique names to every basic block section"), 359 cl::init(false)); 360 CGBINDOPT(UniqueBasicBlockSectionNames); 361 362 static cl::opt<EABI> EABIVersion( 363 "meabi", cl::desc("Set EABI type (default depends on triple):"), 364 cl::init(EABI::Default), 365 cl::values( 366 clEnumValN(EABI::Default, "default", "Triple default EABI version"), 367 clEnumValN(EABI::EABI4, "4", "EABI version 4"), 368 clEnumValN(EABI::EABI5, "5", "EABI version 5"), 369 clEnumValN(EABI::GNU, "gnu", "EABI GNU"))); 370 CGBINDOPT(EABIVersion); 371 372 static cl::opt<DebuggerKind> DebuggerTuningOpt( 373 "debugger-tune", cl::desc("Tune debug info for a particular debugger"), 374 cl::init(DebuggerKind::Default), 375 cl::values( 376 clEnumValN(DebuggerKind::GDB, "gdb", "gdb"), 377 clEnumValN(DebuggerKind::LLDB, "lldb", "lldb"), 378 clEnumValN(DebuggerKind::SCE, "sce", "SCE targets (e.g. PS4)"))); 379 CGBINDOPT(DebuggerTuningOpt); 380 381 static cl::opt<bool> EnableStackSizeSection( 382 "stack-size-section", 383 cl::desc("Emit a section containing stack size metadata"), 384 cl::init(false)); 385 CGBINDOPT(EnableStackSizeSection); 386 387 static cl::opt<bool> EnableAddrsig( 388 "addrsig", cl::desc("Emit an address-significance table"), 389 cl::init(false)); 390 CGBINDOPT(EnableAddrsig); 391 392 static cl::opt<bool> EmitCallSiteInfo( 393 "emit-call-site-info", 394 cl::desc( 395 "Emit call site debug information, if debug information is enabled."), 396 cl::init(false)); 397 CGBINDOPT(EmitCallSiteInfo); 398 399 static cl::opt<bool> EnableDebugEntryValues( 400 "debug-entry-values", 401 cl::desc("Enable debug info for the debug entry values."), 402 cl::init(false)); 403 CGBINDOPT(EnableDebugEntryValues); 404 405 static cl::opt<bool> ValueTrackingVariableLocations( 406 "experimental-debug-variable-locations", 407 cl::desc("Use experimental new value-tracking variable locations"), 408 cl::init(false)); 409 CGBINDOPT(ValueTrackingVariableLocations); 410 411 static cl::opt<bool> EnableMachineFunctionSplitter( 412 "split-machine-functions", 413 cl::desc("Split out cold basic blocks from machine functions based on " 414 "profile information"), 415 cl::init(false)); 416 CGBINDOPT(EnableMachineFunctionSplitter); 417 418 static cl::opt<bool> ForceDwarfFrameSection( 419 "force-dwarf-frame-section", 420 cl::desc("Always emit a debug frame section."), cl::init(false)); 421 CGBINDOPT(ForceDwarfFrameSection); 422 423 static cl::opt<bool> XRayOmitFunctionIndex( 424 "no-xray-index", cl::desc("Don't emit xray_fn_idx section"), 425 cl::init(false)); 426 CGBINDOPT(XRayOmitFunctionIndex); 427 428 #undef CGBINDOPT 429 430 mc::RegisterMCTargetOptionsFlags(); 431 } 432 433 llvm::BasicBlockSection 434 codegen::getBBSectionsMode(llvm::TargetOptions &Options) { 435 if (getBBSections() == "all") 436 return BasicBlockSection::All; 437 else if (getBBSections() == "labels") 438 return BasicBlockSection::Labels; 439 else if (getBBSections() == "none") 440 return BasicBlockSection::None; 441 else { 442 ErrorOr<std::unique_ptr<MemoryBuffer>> MBOrErr = 443 MemoryBuffer::getFile(getBBSections()); 444 if (!MBOrErr) { 445 errs() << "Error loading basic block sections function list file: " 446 << MBOrErr.getError().message() << "\n"; 447 } else { 448 Options.BBSectionsFuncListBuf = std::move(*MBOrErr); 449 } 450 return BasicBlockSection::List; 451 } 452 } 453 454 // Common utility function tightly tied to the options listed here. Initializes 455 // a TargetOptions object with CodeGen flags and returns it. 456 TargetOptions codegen::InitTargetOptionsFromCodeGenFlags() { 457 TargetOptions Options; 458 Options.AllowFPOpFusion = getFuseFPOps(); 459 Options.UnsafeFPMath = getEnableUnsafeFPMath(); 460 Options.NoInfsFPMath = getEnableNoInfsFPMath(); 461 Options.NoNaNsFPMath = getEnableNoNaNsFPMath(); 462 Options.NoSignedZerosFPMath = getEnableNoSignedZerosFPMath(); 463 Options.NoTrappingFPMath = getEnableNoTrappingFPMath(); 464 465 DenormalMode::DenormalModeKind DenormKind = getDenormalFPMath(); 466 467 // FIXME: Should have separate input and output flags 468 Options.setFPDenormalMode(DenormalMode(DenormKind, DenormKind)); 469 470 Options.HonorSignDependentRoundingFPMathOption = 471 getEnableHonorSignDependentRoundingFPMath(); 472 if (getFloatABIForCalls() != FloatABI::Default) 473 Options.FloatABIType = getFloatABIForCalls(); 474 Options.NoZerosInBSS = getDontPlaceZerosInBSS(); 475 Options.GuaranteedTailCallOpt = getEnableGuaranteedTailCallOpt(); 476 Options.StackAlignmentOverride = getOverrideStackAlignment(); 477 Options.StackSymbolOrdering = getStackSymbolOrdering(); 478 Options.UseInitArray = !getUseCtors(); 479 Options.RelaxELFRelocations = getRelaxELFRelocations(); 480 Options.DataSections = getDataSections(); 481 Options.FunctionSections = getFunctionSections(); 482 Options.BBSections = getBBSectionsMode(Options); 483 Options.UniqueSectionNames = getUniqueSectionNames(); 484 Options.UniqueBasicBlockSectionNames = getUniqueBasicBlockSectionNames(); 485 Options.TLSSize = getTLSSize(); 486 Options.EmulatedTLS = getEmulatedTLS(); 487 Options.ExplicitEmulatedTLS = EmulatedTLSView->getNumOccurrences() > 0; 488 Options.ExceptionModel = getExceptionModel(); 489 Options.EmitStackSizeSection = getEnableStackSizeSection(); 490 Options.EnableMachineFunctionSplitter = getEnableMachineFunctionSplitter(); 491 Options.EmitAddrsig = getEnableAddrsig(); 492 Options.EmitCallSiteInfo = getEmitCallSiteInfo(); 493 Options.EnableDebugEntryValues = getEnableDebugEntryValues(); 494 Options.ValueTrackingVariableLocations = getValueTrackingVariableLocations(); 495 Options.ForceDwarfFrameSection = getForceDwarfFrameSection(); 496 Options.XRayOmitFunctionIndex = getXRayOmitFunctionIndex(); 497 498 Options.MCOptions = mc::InitMCTargetOptionsFromFlags(); 499 500 Options.ThreadModel = getThreadModel(); 501 Options.EABIVersion = getEABIVersion(); 502 Options.DebuggerTuning = getDebuggerTuningOpt(); 503 504 return Options; 505 } 506 507 std::string codegen::getCPUStr() { 508 // If user asked for the 'native' CPU, autodetect here. If autodection fails, 509 // this will set the CPU to an empty string which tells the target to 510 // pick a basic default. 511 if (getMCPU() == "native") 512 return std::string(sys::getHostCPUName()); 513 514 return getMCPU(); 515 } 516 517 std::string codegen::getFeaturesStr() { 518 SubtargetFeatures Features; 519 520 // If user asked for the 'native' CPU, we need to autodetect features. 521 // This is necessary for x86 where the CPU might not support all the 522 // features the autodetected CPU name lists in the target. For example, 523 // not all Sandybridge processors support AVX. 524 if (getMCPU() == "native") { 525 StringMap<bool> HostFeatures; 526 if (sys::getHostCPUFeatures(HostFeatures)) 527 for (auto &F : HostFeatures) 528 Features.AddFeature(F.first(), F.second); 529 } 530 531 for (auto const &MAttr : getMAttrs()) 532 Features.AddFeature(MAttr); 533 534 return Features.getString(); 535 } 536 537 std::vector<std::string> codegen::getFeatureList() { 538 SubtargetFeatures Features; 539 540 // If user asked for the 'native' CPU, we need to autodetect features. 541 // This is necessary for x86 where the CPU might not support all the 542 // features the autodetected CPU name lists in the target. For example, 543 // not all Sandybridge processors support AVX. 544 if (getMCPU() == "native") { 545 StringMap<bool> HostFeatures; 546 if (sys::getHostCPUFeatures(HostFeatures)) 547 for (auto &F : HostFeatures) 548 Features.AddFeature(F.first(), F.second); 549 } 550 551 for (auto const &MAttr : getMAttrs()) 552 Features.AddFeature(MAttr); 553 554 return Features.getFeatures(); 555 } 556 557 void codegen::renderBoolStringAttr(AttrBuilder &B, StringRef Name, bool Val) { 558 B.addAttribute(Name, Val ? "true" : "false"); 559 } 560 561 #define HANDLE_BOOL_ATTR(CL, AttrName) \ 562 do { \ 563 if (CL->getNumOccurrences() > 0 && !F.hasFnAttribute(AttrName)) \ 564 renderBoolStringAttr(NewAttrs, AttrName, *CL); \ 565 } while (0) 566 567 /// Set function attributes of function \p F based on CPU, Features, and command 568 /// line flags. 569 void codegen::setFunctionAttributes(StringRef CPU, StringRef Features, 570 Function &F) { 571 auto &Ctx = F.getContext(); 572 AttributeList Attrs = F.getAttributes(); 573 AttrBuilder NewAttrs; 574 575 if (!CPU.empty() && !F.hasFnAttribute("target-cpu")) 576 NewAttrs.addAttribute("target-cpu", CPU); 577 if (!Features.empty()) { 578 // Append the command line features to any that are already on the function. 579 StringRef OldFeatures = 580 F.getFnAttribute("target-features").getValueAsString(); 581 if (OldFeatures.empty()) 582 NewAttrs.addAttribute("target-features", Features); 583 else { 584 SmallString<256> Appended(OldFeatures); 585 Appended.push_back(','); 586 Appended.append(Features); 587 NewAttrs.addAttribute("target-features", Appended); 588 } 589 } 590 if (FramePointerUsageView->getNumOccurrences() > 0 && 591 !F.hasFnAttribute("frame-pointer")) { 592 if (getFramePointerUsage() == FramePointer::All) 593 NewAttrs.addAttribute("frame-pointer", "all"); 594 else if (getFramePointerUsage() == FramePointer::NonLeaf) 595 NewAttrs.addAttribute("frame-pointer", "non-leaf"); 596 else if (getFramePointerUsage() == FramePointer::None) 597 NewAttrs.addAttribute("frame-pointer", "none"); 598 } 599 if (DisableTailCallsView->getNumOccurrences() > 0) 600 NewAttrs.addAttribute("disable-tail-calls", 601 toStringRef(getDisableTailCalls())); 602 if (getStackRealign()) 603 NewAttrs.addAttribute("stackrealign"); 604 605 HANDLE_BOOL_ATTR(EnableUnsafeFPMathView, "unsafe-fp-math"); 606 HANDLE_BOOL_ATTR(EnableNoInfsFPMathView, "no-infs-fp-math"); 607 HANDLE_BOOL_ATTR(EnableNoNaNsFPMathView, "no-nans-fp-math"); 608 HANDLE_BOOL_ATTR(EnableNoSignedZerosFPMathView, "no-signed-zeros-fp-math"); 609 610 if (DenormalFPMathView->getNumOccurrences() > 0 && 611 !F.hasFnAttribute("denormal-fp-math")) { 612 DenormalMode::DenormalModeKind DenormKind = getDenormalFPMath(); 613 614 // FIXME: Command line flag should expose separate input/output modes. 615 NewAttrs.addAttribute("denormal-fp-math", 616 DenormalMode(DenormKind, DenormKind).str()); 617 } 618 619 if (DenormalFP32MathView->getNumOccurrences() > 0 && 620 !F.hasFnAttribute("denormal-fp-math-f32")) { 621 // FIXME: Command line flag should expose separate input/output modes. 622 DenormalMode::DenormalModeKind DenormKind = getDenormalFP32Math(); 623 624 NewAttrs.addAttribute( 625 "denormal-fp-math-f32", 626 DenormalMode(DenormKind, DenormKind).str()); 627 } 628 629 if (TrapFuncNameView->getNumOccurrences() > 0) 630 for (auto &B : F) 631 for (auto &I : B) 632 if (auto *Call = dyn_cast<CallInst>(&I)) 633 if (const auto *F = Call->getCalledFunction()) 634 if (F->getIntrinsicID() == Intrinsic::debugtrap || 635 F->getIntrinsicID() == Intrinsic::trap) 636 Call->addAttribute( 637 AttributeList::FunctionIndex, 638 Attribute::get(Ctx, "trap-func-name", getTrapFuncName())); 639 640 // Let NewAttrs override Attrs. 641 F.setAttributes( 642 Attrs.addAttributes(Ctx, AttributeList::FunctionIndex, NewAttrs)); 643 } 644 645 /// Set function attributes of functions in Module M based on CPU, 646 /// Features, and command line flags. 647 void codegen::setFunctionAttributes(StringRef CPU, StringRef Features, 648 Module &M) { 649 for (Function &F : M) 650 setFunctionAttributes(CPU, Features, F); 651 } 652