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/ADT/StringExtras.h" 17 #include "llvm/ADT/Triple.h" 18 #include "llvm/IR/Instructions.h" 19 #include "llvm/IR/Intrinsics.h" 20 #include "llvm/IR/Module.h" 21 #include "llvm/MC/MCTargetOptionsCommandFlags.h" 22 #include "llvm/MC/SubtargetFeature.h" 23 #include "llvm/Support/CommandLine.h" 24 #include "llvm/Support/Host.h" 25 #include "llvm/Support/MemoryBuffer.h" 26 #include <optional> 27 28 using namespace llvm; 29 30 #define CGOPT(TY, NAME) \ 31 static cl::opt<TY> *NAME##View; \ 32 TY codegen::get##NAME() { \ 33 assert(NAME##View && "RegisterCodeGenFlags not created."); \ 34 return *NAME##View; \ 35 } 36 37 #define CGLIST(TY, NAME) \ 38 static cl::list<TY> *NAME##View; \ 39 std::vector<TY> codegen::get##NAME() { \ 40 assert(NAME##View && "RegisterCodeGenFlags not created."); \ 41 return *NAME##View; \ 42 } 43 44 // Temporary macro for incremental transition to std::optional. 45 #define CGOPT_EXP(TY, NAME) \ 46 CGOPT(TY, NAME) \ 47 std::optional<TY> codegen::getExplicit##NAME() { \ 48 if (NAME##View->getNumOccurrences()) { \ 49 TY res = *NAME##View; \ 50 return res; \ 51 } \ 52 return std::nullopt; \ 53 } 54 55 CGOPT(std::string, MArch) 56 CGOPT(std::string, MCPU) 57 CGLIST(std::string, MAttrs) 58 CGOPT_EXP(Reloc::Model, RelocModel) 59 CGOPT(ThreadModel::Model, ThreadModel) 60 CGOPT_EXP(CodeModel::Model, CodeModel) 61 CGOPT(ExceptionHandling, ExceptionModel) 62 CGOPT_EXP(CodeGenFileType, FileType) 63 CGOPT(FramePointerKind, FramePointerUsage) 64 CGOPT(bool, EnableUnsafeFPMath) 65 CGOPT(bool, EnableNoInfsFPMath) 66 CGOPT(bool, EnableNoNaNsFPMath) 67 CGOPT(bool, EnableNoSignedZerosFPMath) 68 CGOPT(bool, EnableApproxFuncFPMath) 69 CGOPT(bool, EnableNoTrappingFPMath) 70 CGOPT(bool, EnableAIXExtendedAltivecABI) 71 CGOPT(DenormalMode::DenormalModeKind, DenormalFPMath) 72 CGOPT(DenormalMode::DenormalModeKind, DenormalFP32Math) 73 CGOPT(bool, EnableHonorSignDependentRoundingFPMath) 74 CGOPT(FloatABI::ABIType, FloatABIForCalls) 75 CGOPT(FPOpFusion::FPOpFusionMode, FuseFPOps) 76 CGOPT(SwiftAsyncFramePointerMode, SwiftAsyncFramePointer) 77 CGOPT(bool, DontPlaceZerosInBSS) 78 CGOPT(bool, EnableGuaranteedTailCallOpt) 79 CGOPT(bool, DisableTailCalls) 80 CGOPT(bool, StackSymbolOrdering) 81 CGOPT(bool, StackRealign) 82 CGOPT(std::string, TrapFuncName) 83 CGOPT(bool, UseCtors) 84 CGOPT(bool, LowerGlobalDtorsViaCxaAtExit) 85 CGOPT(bool, RelaxELFRelocations) 86 CGOPT_EXP(bool, DataSections) 87 CGOPT_EXP(bool, FunctionSections) 88 CGOPT(bool, IgnoreXCOFFVisibility) 89 CGOPT(bool, XCOFFTracebackTable) 90 CGOPT(std::string, BBSections) 91 CGOPT(unsigned, TLSSize) 92 CGOPT(bool, EmulatedTLS) 93 CGOPT(bool, UniqueSectionNames) 94 CGOPT(bool, UniqueBasicBlockSectionNames) 95 CGOPT(EABI, EABIVersion) 96 CGOPT(DebuggerKind, DebuggerTuningOpt) 97 CGOPT(bool, EnableStackSizeSection) 98 CGOPT(bool, EnableAddrsig) 99 CGOPT(bool, EmitCallSiteInfo) 100 CGOPT(bool, EnableMachineFunctionSplitter) 101 CGOPT(bool, EnableDebugEntryValues) 102 CGOPT(bool, ForceDwarfFrameSection) 103 CGOPT(bool, XRayOmitFunctionIndex) 104 CGOPT(bool, DebugStrictDwarf) 105 CGOPT(unsigned, AlignLoops) 106 CGOPT(bool, JMCInstrument) 107 108 codegen::RegisterCodeGenFlags::RegisterCodeGenFlags() { 109 #define CGBINDOPT(NAME) \ 110 do { \ 111 NAME##View = std::addressof(NAME); \ 112 } while (0) 113 114 static cl::opt<std::string> MArch( 115 "march", cl::desc("Architecture to generate code for (see --version)")); 116 CGBINDOPT(MArch); 117 118 static cl::opt<std::string> MCPU( 119 "mcpu", cl::desc("Target a specific cpu type (-mcpu=help for details)"), 120 cl::value_desc("cpu-name"), cl::init("")); 121 CGBINDOPT(MCPU); 122 123 static cl::list<std::string> MAttrs( 124 "mattr", cl::CommaSeparated, 125 cl::desc("Target specific attributes (-mattr=help for details)"), 126 cl::value_desc("a1,+a2,-a3,...")); 127 CGBINDOPT(MAttrs); 128 129 static cl::opt<Reloc::Model> RelocModel( 130 "relocation-model", cl::desc("Choose relocation model"), 131 cl::values( 132 clEnumValN(Reloc::Static, "static", "Non-relocatable code"), 133 clEnumValN(Reloc::PIC_, "pic", 134 "Fully relocatable, position independent code"), 135 clEnumValN(Reloc::DynamicNoPIC, "dynamic-no-pic", 136 "Relocatable external references, non-relocatable code"), 137 clEnumValN( 138 Reloc::ROPI, "ropi", 139 "Code and read-only data relocatable, accessed PC-relative"), 140 clEnumValN( 141 Reloc::RWPI, "rwpi", 142 "Read-write data relocatable, accessed relative to static base"), 143 clEnumValN(Reloc::ROPI_RWPI, "ropi-rwpi", 144 "Combination of ropi and rwpi"))); 145 CGBINDOPT(RelocModel); 146 147 static cl::opt<ThreadModel::Model> ThreadModel( 148 "thread-model", cl::desc("Choose threading model"), 149 cl::init(ThreadModel::POSIX), 150 cl::values( 151 clEnumValN(ThreadModel::POSIX, "posix", "POSIX thread model"), 152 clEnumValN(ThreadModel::Single, "single", "Single thread model"))); 153 CGBINDOPT(ThreadModel); 154 155 static cl::opt<CodeModel::Model> CodeModel( 156 "code-model", cl::desc("Choose code model"), 157 cl::values(clEnumValN(CodeModel::Tiny, "tiny", "Tiny code model"), 158 clEnumValN(CodeModel::Small, "small", "Small code model"), 159 clEnumValN(CodeModel::Kernel, "kernel", "Kernel code model"), 160 clEnumValN(CodeModel::Medium, "medium", "Medium code model"), 161 clEnumValN(CodeModel::Large, "large", "Large code model"))); 162 CGBINDOPT(CodeModel); 163 164 static cl::opt<ExceptionHandling> ExceptionModel( 165 "exception-model", cl::desc("exception model"), 166 cl::init(ExceptionHandling::None), 167 cl::values( 168 clEnumValN(ExceptionHandling::None, "default", 169 "default exception handling model"), 170 clEnumValN(ExceptionHandling::DwarfCFI, "dwarf", 171 "DWARF-like CFI based exception handling"), 172 clEnumValN(ExceptionHandling::SjLj, "sjlj", 173 "SjLj exception handling"), 174 clEnumValN(ExceptionHandling::ARM, "arm", "ARM EHABI exceptions"), 175 clEnumValN(ExceptionHandling::WinEH, "wineh", 176 "Windows exception model"), 177 clEnumValN(ExceptionHandling::Wasm, "wasm", 178 "WebAssembly exception handling"))); 179 CGBINDOPT(ExceptionModel); 180 181 static cl::opt<CodeGenFileType> FileType( 182 "filetype", cl::init(CGFT_AssemblyFile), 183 cl::desc( 184 "Choose a file type (not all types are supported by all targets):"), 185 cl::values( 186 clEnumValN(CGFT_AssemblyFile, "asm", "Emit an assembly ('.s') file"), 187 clEnumValN(CGFT_ObjectFile, "obj", 188 "Emit a native object ('.o') file"), 189 clEnumValN(CGFT_Null, "null", 190 "Emit nothing, for performance testing"))); 191 CGBINDOPT(FileType); 192 193 static cl::opt<FramePointerKind> FramePointerUsage( 194 "frame-pointer", 195 cl::desc("Specify frame pointer elimination optimization"), 196 cl::init(FramePointerKind::None), 197 cl::values( 198 clEnumValN(FramePointerKind::All, "all", 199 "Disable frame pointer elimination"), 200 clEnumValN(FramePointerKind::NonLeaf, "non-leaf", 201 "Disable frame pointer elimination for non-leaf frame"), 202 clEnumValN(FramePointerKind::None, "none", 203 "Enable frame pointer elimination"))); 204 CGBINDOPT(FramePointerUsage); 205 206 static cl::opt<bool> EnableUnsafeFPMath( 207 "enable-unsafe-fp-math", 208 cl::desc("Enable optimizations that may decrease FP precision"), 209 cl::init(false)); 210 CGBINDOPT(EnableUnsafeFPMath); 211 212 static cl::opt<bool> EnableNoInfsFPMath( 213 "enable-no-infs-fp-math", 214 cl::desc("Enable FP math optimizations that assume no +-Infs"), 215 cl::init(false)); 216 CGBINDOPT(EnableNoInfsFPMath); 217 218 static cl::opt<bool> EnableNoNaNsFPMath( 219 "enable-no-nans-fp-math", 220 cl::desc("Enable FP math optimizations that assume no NaNs"), 221 cl::init(false)); 222 CGBINDOPT(EnableNoNaNsFPMath); 223 224 static cl::opt<bool> EnableNoSignedZerosFPMath( 225 "enable-no-signed-zeros-fp-math", 226 cl::desc("Enable FP math optimizations that assume " 227 "the sign of 0 is insignificant"), 228 cl::init(false)); 229 CGBINDOPT(EnableNoSignedZerosFPMath); 230 231 static cl::opt<bool> EnableApproxFuncFPMath( 232 "enable-approx-func-fp-math", 233 cl::desc("Enable FP math optimizations that assume approx func"), 234 cl::init(false)); 235 CGBINDOPT(EnableApproxFuncFPMath); 236 237 static cl::opt<bool> EnableNoTrappingFPMath( 238 "enable-no-trapping-fp-math", 239 cl::desc("Enable setting the FP exceptions build " 240 "attribute not to use exceptions"), 241 cl::init(false)); 242 CGBINDOPT(EnableNoTrappingFPMath); 243 244 static const auto DenormFlagEnumOptions = 245 cl::values(clEnumValN(DenormalMode::IEEE, "ieee", 246 "IEEE 754 denormal numbers"), 247 clEnumValN(DenormalMode::PreserveSign, "preserve-sign", 248 "the sign of a flushed-to-zero number is preserved " 249 "in the sign of 0"), 250 clEnumValN(DenormalMode::PositiveZero, "positive-zero", 251 "denormals are flushed to positive zero")); 252 253 // FIXME: Doesn't have way to specify separate input and output modes. 254 static cl::opt<DenormalMode::DenormalModeKind> DenormalFPMath( 255 "denormal-fp-math", 256 cl::desc("Select which denormal numbers the code is permitted to require"), 257 cl::init(DenormalMode::IEEE), 258 DenormFlagEnumOptions); 259 CGBINDOPT(DenormalFPMath); 260 261 static cl::opt<DenormalMode::DenormalModeKind> DenormalFP32Math( 262 "denormal-fp-math-f32", 263 cl::desc("Select which denormal numbers the code is permitted to require for float"), 264 cl::init(DenormalMode::Invalid), 265 DenormFlagEnumOptions); 266 CGBINDOPT(DenormalFP32Math); 267 268 static cl::opt<bool> EnableHonorSignDependentRoundingFPMath( 269 "enable-sign-dependent-rounding-fp-math", cl::Hidden, 270 cl::desc("Force codegen to assume rounding mode can change dynamically"), 271 cl::init(false)); 272 CGBINDOPT(EnableHonorSignDependentRoundingFPMath); 273 274 static cl::opt<FloatABI::ABIType> FloatABIForCalls( 275 "float-abi", cl::desc("Choose float ABI type"), 276 cl::init(FloatABI::Default), 277 cl::values(clEnumValN(FloatABI::Default, "default", 278 "Target default float ABI type"), 279 clEnumValN(FloatABI::Soft, "soft", 280 "Soft float ABI (implied by -soft-float)"), 281 clEnumValN(FloatABI::Hard, "hard", 282 "Hard float ABI (uses FP registers)"))); 283 CGBINDOPT(FloatABIForCalls); 284 285 static cl::opt<FPOpFusion::FPOpFusionMode> FuseFPOps( 286 "fp-contract", cl::desc("Enable aggressive formation of fused FP ops"), 287 cl::init(FPOpFusion::Standard), 288 cl::values( 289 clEnumValN(FPOpFusion::Fast, "fast", 290 "Fuse FP ops whenever profitable"), 291 clEnumValN(FPOpFusion::Standard, "on", "Only fuse 'blessed' FP ops."), 292 clEnumValN(FPOpFusion::Strict, "off", 293 "Only fuse FP ops when the result won't be affected."))); 294 CGBINDOPT(FuseFPOps); 295 296 static cl::opt<SwiftAsyncFramePointerMode> SwiftAsyncFramePointer( 297 "swift-async-fp", 298 cl::desc("Determine when the Swift async frame pointer should be set"), 299 cl::init(SwiftAsyncFramePointerMode::Always), 300 cl::values(clEnumValN(SwiftAsyncFramePointerMode::DeploymentBased, "auto", 301 "Determine based on deployment target"), 302 clEnumValN(SwiftAsyncFramePointerMode::Always, "always", 303 "Always set the bit"), 304 clEnumValN(SwiftAsyncFramePointerMode::Never, "never", 305 "Never set the bit"))); 306 CGBINDOPT(SwiftAsyncFramePointer); 307 308 static cl::opt<bool> DontPlaceZerosInBSS( 309 "nozero-initialized-in-bss", 310 cl::desc("Don't place zero-initialized symbols into bss section"), 311 cl::init(false)); 312 CGBINDOPT(DontPlaceZerosInBSS); 313 314 static cl::opt<bool> EnableAIXExtendedAltivecABI( 315 "vec-extabi", cl::desc("Enable the AIX Extended Altivec ABI."), 316 cl::init(false)); 317 CGBINDOPT(EnableAIXExtendedAltivecABI); 318 319 static cl::opt<bool> EnableGuaranteedTailCallOpt( 320 "tailcallopt", 321 cl::desc( 322 "Turn fastcc calls into tail calls by (potentially) changing ABI."), 323 cl::init(false)); 324 CGBINDOPT(EnableGuaranteedTailCallOpt); 325 326 static cl::opt<bool> DisableTailCalls( 327 "disable-tail-calls", cl::desc("Never emit tail calls"), cl::init(false)); 328 CGBINDOPT(DisableTailCalls); 329 330 static cl::opt<bool> StackSymbolOrdering( 331 "stack-symbol-ordering", cl::desc("Order local stack symbols."), 332 cl::init(true)); 333 CGBINDOPT(StackSymbolOrdering); 334 335 static cl::opt<bool> StackRealign( 336 "stackrealign", 337 cl::desc("Force align the stack to the minimum alignment"), 338 cl::init(false)); 339 CGBINDOPT(StackRealign); 340 341 static cl::opt<std::string> TrapFuncName( 342 "trap-func", cl::Hidden, 343 cl::desc("Emit a call to trap function rather than a trap instruction"), 344 cl::init("")); 345 CGBINDOPT(TrapFuncName); 346 347 static cl::opt<bool> UseCtors("use-ctors", 348 cl::desc("Use .ctors instead of .init_array."), 349 cl::init(false)); 350 CGBINDOPT(UseCtors); 351 352 static cl::opt<bool> LowerGlobalDtorsViaCxaAtExit( 353 "lower-global-dtors-via-cxa-atexit", 354 cl::desc("Lower llvm.global_dtors (global destructors) via __cxa_atexit"), 355 cl::init(true)); 356 CGBINDOPT(LowerGlobalDtorsViaCxaAtExit); 357 358 static cl::opt<bool> RelaxELFRelocations( 359 "relax-elf-relocations", 360 cl::desc( 361 "Emit GOTPCRELX/REX_GOTPCRELX instead of GOTPCREL on x86-64 ELF"), 362 cl::init(true)); 363 CGBINDOPT(RelaxELFRelocations); 364 365 static cl::opt<bool> DataSections( 366 "data-sections", cl::desc("Emit data into separate sections"), 367 cl::init(false)); 368 CGBINDOPT(DataSections); 369 370 static cl::opt<bool> FunctionSections( 371 "function-sections", cl::desc("Emit functions into separate sections"), 372 cl::init(false)); 373 CGBINDOPT(FunctionSections); 374 375 static cl::opt<bool> IgnoreXCOFFVisibility( 376 "ignore-xcoff-visibility", 377 cl::desc("Not emit the visibility attribute for asm in AIX OS or give " 378 "all symbols 'unspecified' visibility in XCOFF object file"), 379 cl::init(false)); 380 CGBINDOPT(IgnoreXCOFFVisibility); 381 382 static cl::opt<bool> XCOFFTracebackTable( 383 "xcoff-traceback-table", cl::desc("Emit the XCOFF traceback table"), 384 cl::init(true)); 385 CGBINDOPT(XCOFFTracebackTable); 386 387 static cl::opt<std::string> BBSections( 388 "basic-block-sections", 389 cl::desc("Emit basic blocks into separate sections"), 390 cl::value_desc("all | <function list (file)> | labels | none"), 391 cl::init("none")); 392 CGBINDOPT(BBSections); 393 394 static cl::opt<unsigned> TLSSize( 395 "tls-size", cl::desc("Bit size of immediate TLS offsets"), cl::init(0)); 396 CGBINDOPT(TLSSize); 397 398 static cl::opt<bool> EmulatedTLS( 399 "emulated-tls", cl::desc("Use emulated TLS model"), cl::init(false)); 400 CGBINDOPT(EmulatedTLS); 401 402 static cl::opt<bool> UniqueSectionNames( 403 "unique-section-names", cl::desc("Give unique names to every section"), 404 cl::init(true)); 405 CGBINDOPT(UniqueSectionNames); 406 407 static cl::opt<bool> UniqueBasicBlockSectionNames( 408 "unique-basic-block-section-names", 409 cl::desc("Give unique names to every basic block section"), 410 cl::init(false)); 411 CGBINDOPT(UniqueBasicBlockSectionNames); 412 413 static cl::opt<EABI> EABIVersion( 414 "meabi", cl::desc("Set EABI type (default depends on triple):"), 415 cl::init(EABI::Default), 416 cl::values( 417 clEnumValN(EABI::Default, "default", "Triple default EABI version"), 418 clEnumValN(EABI::EABI4, "4", "EABI version 4"), 419 clEnumValN(EABI::EABI5, "5", "EABI version 5"), 420 clEnumValN(EABI::GNU, "gnu", "EABI GNU"))); 421 CGBINDOPT(EABIVersion); 422 423 static cl::opt<DebuggerKind> DebuggerTuningOpt( 424 "debugger-tune", cl::desc("Tune debug info for a particular debugger"), 425 cl::init(DebuggerKind::Default), 426 cl::values( 427 clEnumValN(DebuggerKind::GDB, "gdb", "gdb"), 428 clEnumValN(DebuggerKind::LLDB, "lldb", "lldb"), 429 clEnumValN(DebuggerKind::DBX, "dbx", "dbx"), 430 clEnumValN(DebuggerKind::SCE, "sce", "SCE targets (e.g. PS4)"))); 431 CGBINDOPT(DebuggerTuningOpt); 432 433 static cl::opt<bool> EnableStackSizeSection( 434 "stack-size-section", 435 cl::desc("Emit a section containing stack size metadata"), 436 cl::init(false)); 437 CGBINDOPT(EnableStackSizeSection); 438 439 static cl::opt<bool> EnableAddrsig( 440 "addrsig", cl::desc("Emit an address-significance table"), 441 cl::init(false)); 442 CGBINDOPT(EnableAddrsig); 443 444 static cl::opt<bool> EmitCallSiteInfo( 445 "emit-call-site-info", 446 cl::desc( 447 "Emit call site debug information, if debug information is enabled."), 448 cl::init(false)); 449 CGBINDOPT(EmitCallSiteInfo); 450 451 static cl::opt<bool> EnableDebugEntryValues( 452 "debug-entry-values", 453 cl::desc("Enable debug info for the debug entry values."), 454 cl::init(false)); 455 CGBINDOPT(EnableDebugEntryValues); 456 457 static cl::opt<bool> EnableMachineFunctionSplitter( 458 "split-machine-functions", 459 cl::desc("Split out cold basic blocks from machine functions based on " 460 "profile information"), 461 cl::init(false)); 462 CGBINDOPT(EnableMachineFunctionSplitter); 463 464 static cl::opt<bool> ForceDwarfFrameSection( 465 "force-dwarf-frame-section", 466 cl::desc("Always emit a debug frame section."), cl::init(false)); 467 CGBINDOPT(ForceDwarfFrameSection); 468 469 static cl::opt<bool> XRayOmitFunctionIndex( 470 "no-xray-index", cl::desc("Don't emit xray_fn_idx section"), 471 cl::init(false)); 472 CGBINDOPT(XRayOmitFunctionIndex); 473 474 static cl::opt<bool> DebugStrictDwarf( 475 "strict-dwarf", cl::desc("use strict dwarf"), cl::init(false)); 476 CGBINDOPT(DebugStrictDwarf); 477 478 static cl::opt<unsigned> AlignLoops("align-loops", 479 cl::desc("Default alignment for loops")); 480 CGBINDOPT(AlignLoops); 481 482 static cl::opt<bool> JMCInstrument( 483 "enable-jmc-instrument", 484 cl::desc("Instrument functions with a call to __CheckForDebuggerJustMyCode"), 485 cl::init(false)); 486 CGBINDOPT(JMCInstrument); 487 488 #undef CGBINDOPT 489 490 mc::RegisterMCTargetOptionsFlags(); 491 } 492 493 llvm::BasicBlockSection 494 codegen::getBBSectionsMode(llvm::TargetOptions &Options) { 495 if (getBBSections() == "all") 496 return BasicBlockSection::All; 497 else if (getBBSections() == "labels") 498 return BasicBlockSection::Labels; 499 else if (getBBSections() == "none") 500 return BasicBlockSection::None; 501 else { 502 ErrorOr<std::unique_ptr<MemoryBuffer>> MBOrErr = 503 MemoryBuffer::getFile(getBBSections()); 504 if (!MBOrErr) { 505 errs() << "Error loading basic block sections function list file: " 506 << MBOrErr.getError().message() << "\n"; 507 } else { 508 Options.BBSectionsFuncListBuf = std::move(*MBOrErr); 509 } 510 return BasicBlockSection::List; 511 } 512 } 513 514 // Common utility function tightly tied to the options listed here. Initializes 515 // a TargetOptions object with CodeGen flags and returns it. 516 TargetOptions 517 codegen::InitTargetOptionsFromCodeGenFlags(const Triple &TheTriple) { 518 TargetOptions Options; 519 Options.AllowFPOpFusion = getFuseFPOps(); 520 Options.UnsafeFPMath = getEnableUnsafeFPMath(); 521 Options.NoInfsFPMath = getEnableNoInfsFPMath(); 522 Options.NoNaNsFPMath = getEnableNoNaNsFPMath(); 523 Options.NoSignedZerosFPMath = getEnableNoSignedZerosFPMath(); 524 Options.ApproxFuncFPMath = getEnableApproxFuncFPMath(); 525 Options.NoTrappingFPMath = getEnableNoTrappingFPMath(); 526 527 DenormalMode::DenormalModeKind DenormKind = getDenormalFPMath(); 528 529 // FIXME: Should have separate input and output flags 530 Options.setFPDenormalMode(DenormalMode(DenormKind, DenormKind)); 531 532 Options.HonorSignDependentRoundingFPMathOption = 533 getEnableHonorSignDependentRoundingFPMath(); 534 if (getFloatABIForCalls() != FloatABI::Default) 535 Options.FloatABIType = getFloatABIForCalls(); 536 Options.EnableAIXExtendedAltivecABI = getEnableAIXExtendedAltivecABI(); 537 Options.NoZerosInBSS = getDontPlaceZerosInBSS(); 538 Options.GuaranteedTailCallOpt = getEnableGuaranteedTailCallOpt(); 539 Options.StackSymbolOrdering = getStackSymbolOrdering(); 540 Options.UseInitArray = !getUseCtors(); 541 Options.LowerGlobalDtorsViaCxaAtExit = getLowerGlobalDtorsViaCxaAtExit(); 542 Options.RelaxELFRelocations = getRelaxELFRelocations(); 543 Options.DataSections = 544 getExplicitDataSections().value_or(TheTriple.hasDefaultDataSections()); 545 Options.FunctionSections = getFunctionSections(); 546 Options.IgnoreXCOFFVisibility = getIgnoreXCOFFVisibility(); 547 Options.XCOFFTracebackTable = getXCOFFTracebackTable(); 548 Options.BBSections = getBBSectionsMode(Options); 549 Options.UniqueSectionNames = getUniqueSectionNames(); 550 Options.UniqueBasicBlockSectionNames = getUniqueBasicBlockSectionNames(); 551 Options.TLSSize = getTLSSize(); 552 Options.EmulatedTLS = getEmulatedTLS(); 553 Options.ExplicitEmulatedTLS = EmulatedTLSView->getNumOccurrences() > 0; 554 Options.ExceptionModel = getExceptionModel(); 555 Options.EmitStackSizeSection = getEnableStackSizeSection(); 556 Options.EnableMachineFunctionSplitter = getEnableMachineFunctionSplitter(); 557 Options.EmitAddrsig = getEnableAddrsig(); 558 Options.EmitCallSiteInfo = getEmitCallSiteInfo(); 559 Options.EnableDebugEntryValues = getEnableDebugEntryValues(); 560 Options.ForceDwarfFrameSection = getForceDwarfFrameSection(); 561 Options.XRayOmitFunctionIndex = getXRayOmitFunctionIndex(); 562 Options.DebugStrictDwarf = getDebugStrictDwarf(); 563 Options.LoopAlignment = getAlignLoops(); 564 Options.JMCInstrument = getJMCInstrument(); 565 566 Options.MCOptions = mc::InitMCTargetOptionsFromFlags(); 567 568 Options.ThreadModel = getThreadModel(); 569 Options.EABIVersion = getEABIVersion(); 570 Options.DebuggerTuning = getDebuggerTuningOpt(); 571 Options.SwiftAsyncFramePointer = getSwiftAsyncFramePointer(); 572 return Options; 573 } 574 575 std::string codegen::getCPUStr() { 576 // If user asked for the 'native' CPU, autodetect here. If autodection fails, 577 // this will set the CPU to an empty string which tells the target to 578 // pick a basic default. 579 if (getMCPU() == "native") 580 return std::string(sys::getHostCPUName()); 581 582 return getMCPU(); 583 } 584 585 std::string codegen::getFeaturesStr() { 586 SubtargetFeatures Features; 587 588 // If user asked for the 'native' CPU, we need to autodetect features. 589 // This is necessary for x86 where the CPU might not support all the 590 // features the autodetected CPU name lists in the target. For example, 591 // not all Sandybridge processors support AVX. 592 if (getMCPU() == "native") { 593 StringMap<bool> HostFeatures; 594 if (sys::getHostCPUFeatures(HostFeatures)) 595 for (const auto &[Feature, IsEnabled] : HostFeatures) 596 Features.AddFeature(Feature, IsEnabled); 597 } 598 599 for (auto const &MAttr : getMAttrs()) 600 Features.AddFeature(MAttr); 601 602 return Features.getString(); 603 } 604 605 std::vector<std::string> codegen::getFeatureList() { 606 SubtargetFeatures Features; 607 608 // If user asked for the 'native' CPU, we need to autodetect features. 609 // This is necessary for x86 where the CPU might not support all the 610 // features the autodetected CPU name lists in the target. For example, 611 // not all Sandybridge processors support AVX. 612 if (getMCPU() == "native") { 613 StringMap<bool> HostFeatures; 614 if (sys::getHostCPUFeatures(HostFeatures)) 615 for (const auto &[Feature, IsEnabled] : HostFeatures) 616 Features.AddFeature(Feature, IsEnabled); 617 } 618 619 for (auto const &MAttr : getMAttrs()) 620 Features.AddFeature(MAttr); 621 622 return Features.getFeatures(); 623 } 624 625 void codegen::renderBoolStringAttr(AttrBuilder &B, StringRef Name, bool Val) { 626 B.addAttribute(Name, Val ? "true" : "false"); 627 } 628 629 #define HANDLE_BOOL_ATTR(CL, AttrName) \ 630 do { \ 631 if (CL->getNumOccurrences() > 0 && !F.hasFnAttribute(AttrName)) \ 632 renderBoolStringAttr(NewAttrs, AttrName, *CL); \ 633 } while (0) 634 635 /// Set function attributes of function \p F based on CPU, Features, and command 636 /// line flags. 637 void codegen::setFunctionAttributes(StringRef CPU, StringRef Features, 638 Function &F) { 639 auto &Ctx = F.getContext(); 640 AttributeList Attrs = F.getAttributes(); 641 AttrBuilder NewAttrs(Ctx); 642 643 if (!CPU.empty() && !F.hasFnAttribute("target-cpu")) 644 NewAttrs.addAttribute("target-cpu", CPU); 645 if (!Features.empty()) { 646 // Append the command line features to any that are already on the function. 647 StringRef OldFeatures = 648 F.getFnAttribute("target-features").getValueAsString(); 649 if (OldFeatures.empty()) 650 NewAttrs.addAttribute("target-features", Features); 651 else { 652 SmallString<256> Appended(OldFeatures); 653 Appended.push_back(','); 654 Appended.append(Features); 655 NewAttrs.addAttribute("target-features", Appended); 656 } 657 } 658 if (FramePointerUsageView->getNumOccurrences() > 0 && 659 !F.hasFnAttribute("frame-pointer")) { 660 if (getFramePointerUsage() == FramePointerKind::All) 661 NewAttrs.addAttribute("frame-pointer", "all"); 662 else if (getFramePointerUsage() == FramePointerKind::NonLeaf) 663 NewAttrs.addAttribute("frame-pointer", "non-leaf"); 664 else if (getFramePointerUsage() == FramePointerKind::None) 665 NewAttrs.addAttribute("frame-pointer", "none"); 666 } 667 if (DisableTailCallsView->getNumOccurrences() > 0) 668 NewAttrs.addAttribute("disable-tail-calls", 669 toStringRef(getDisableTailCalls())); 670 if (getStackRealign()) 671 NewAttrs.addAttribute("stackrealign"); 672 673 HANDLE_BOOL_ATTR(EnableUnsafeFPMathView, "unsafe-fp-math"); 674 HANDLE_BOOL_ATTR(EnableNoInfsFPMathView, "no-infs-fp-math"); 675 HANDLE_BOOL_ATTR(EnableNoNaNsFPMathView, "no-nans-fp-math"); 676 HANDLE_BOOL_ATTR(EnableNoSignedZerosFPMathView, "no-signed-zeros-fp-math"); 677 HANDLE_BOOL_ATTR(EnableApproxFuncFPMathView, "approx-func-fp-math"); 678 679 if (DenormalFPMathView->getNumOccurrences() > 0 && 680 !F.hasFnAttribute("denormal-fp-math")) { 681 DenormalMode::DenormalModeKind DenormKind = getDenormalFPMath(); 682 683 // FIXME: Command line flag should expose separate input/output modes. 684 NewAttrs.addAttribute("denormal-fp-math", 685 DenormalMode(DenormKind, DenormKind).str()); 686 } 687 688 if (DenormalFP32MathView->getNumOccurrences() > 0 && 689 !F.hasFnAttribute("denormal-fp-math-f32")) { 690 // FIXME: Command line flag should expose separate input/output modes. 691 DenormalMode::DenormalModeKind DenormKind = getDenormalFP32Math(); 692 693 NewAttrs.addAttribute( 694 "denormal-fp-math-f32", 695 DenormalMode(DenormKind, DenormKind).str()); 696 } 697 698 if (TrapFuncNameView->getNumOccurrences() > 0) 699 for (auto &B : F) 700 for (auto &I : B) 701 if (auto *Call = dyn_cast<CallInst>(&I)) 702 if (const auto *F = Call->getCalledFunction()) 703 if (F->getIntrinsicID() == Intrinsic::debugtrap || 704 F->getIntrinsicID() == Intrinsic::trap) 705 Call->addFnAttr( 706 Attribute::get(Ctx, "trap-func-name", getTrapFuncName())); 707 708 // Let NewAttrs override Attrs. 709 F.setAttributes(Attrs.addFnAttributes(Ctx, NewAttrs)); 710 } 711 712 /// Set function attributes of functions in Module M based on CPU, 713 /// Features, and command line flags. 714 void codegen::setFunctionAttributes(StringRef CPU, StringRef Features, 715 Module &M) { 716 for (Function &F : M) 717 setFunctionAttributes(CPU, Features, F); 718 } 719