1 //===--- HIPAMD.cpp - HIP Tool and ToolChain Implementations ----*- 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 #include "HIPAMD.h" 10 #include "AMDGPU.h" 11 #include "CommonArgs.h" 12 #include "HIPUtility.h" 13 #include "SPIRV.h" 14 #include "clang/Basic/Cuda.h" 15 #include "clang/Basic/TargetID.h" 16 #include "clang/Driver/Compilation.h" 17 #include "clang/Driver/Driver.h" 18 #include "clang/Driver/DriverDiagnostic.h" 19 #include "clang/Driver/InputInfo.h" 20 #include "clang/Driver/Options.h" 21 #include "clang/Driver/SanitizerArgs.h" 22 #include "llvm/Support/Alignment.h" 23 #include "llvm/Support/FileSystem.h" 24 #include "llvm/Support/Path.h" 25 #include "llvm/TargetParser/TargetParser.h" 26 27 using namespace clang::driver; 28 using namespace clang::driver::toolchains; 29 using namespace clang::driver::tools; 30 using namespace clang; 31 using namespace llvm::opt; 32 33 #if defined(_WIN32) || defined(_WIN64) 34 #define NULL_FILE "nul" 35 #else 36 #define NULL_FILE "/dev/null" 37 #endif 38 39 void AMDGCN::Linker::constructLlvmLinkCommand(Compilation &C, 40 const JobAction &JA, 41 const InputInfoList &Inputs, 42 const InputInfo &Output, 43 const llvm::opt::ArgList &Args) const { 44 // Construct llvm-link command. 45 // The output from llvm-link is a bitcode file. 46 ArgStringList LlvmLinkArgs; 47 48 assert(!Inputs.empty() && "Must have at least one input."); 49 50 LlvmLinkArgs.append({"-o", Output.getFilename()}); 51 for (auto Input : Inputs) 52 LlvmLinkArgs.push_back(Input.getFilename()); 53 54 // Look for archive of bundled bitcode in arguments, and add temporary files 55 // for the extracted archive of bitcode to inputs. 56 auto TargetID = Args.getLastArgValue(options::OPT_mcpu_EQ); 57 AddStaticDeviceLibsLinking(C, *this, JA, Inputs, Args, LlvmLinkArgs, "amdgcn", 58 TargetID, /*IsBitCodeSDL=*/true); 59 60 const char *LlvmLink = 61 Args.MakeArgString(getToolChain().GetProgramPath("llvm-link")); 62 C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(), 63 LlvmLink, LlvmLinkArgs, Inputs, 64 Output)); 65 } 66 67 void AMDGCN::Linker::constructLldCommand(Compilation &C, const JobAction &JA, 68 const InputInfoList &Inputs, 69 const InputInfo &Output, 70 const llvm::opt::ArgList &Args) const { 71 // Construct lld command. 72 // The output from ld.lld is an HSA code object file. 73 ArgStringList LldArgs{"-flavor", 74 "gnu", 75 "-m", 76 "elf64_amdgpu", 77 "--no-undefined", 78 "-shared", 79 "-plugin-opt=-amdgpu-internalize-symbols"}; 80 if (Args.hasArg(options::OPT_hipstdpar)) 81 LldArgs.push_back("-plugin-opt=-amdgpu-enable-hipstdpar"); 82 83 auto &TC = getToolChain(); 84 auto &D = TC.getDriver(); 85 assert(!Inputs.empty() && "Must have at least one input."); 86 bool IsThinLTO = D.getOffloadLTOMode() == LTOK_Thin; 87 addLTOOptions(TC, Args, LldArgs, Output, Inputs[0], IsThinLTO); 88 89 // Extract all the -m options 90 std::vector<llvm::StringRef> Features; 91 amdgpu::getAMDGPUTargetFeatures(D, TC.getTriple(), Args, Features); 92 93 // Add features to mattr such as cumode 94 std::string MAttrString = "-plugin-opt=-mattr="; 95 for (auto OneFeature : unifyTargetFeatures(Features)) { 96 MAttrString.append(Args.MakeArgString(OneFeature)); 97 if (OneFeature != Features.back()) 98 MAttrString.append(","); 99 } 100 if (!Features.empty()) 101 LldArgs.push_back(Args.MakeArgString(MAttrString)); 102 103 // ToDo: Remove this option after AMDGPU backend supports ISA-level linking. 104 // Since AMDGPU backend currently does not support ISA-level linking, all 105 // called functions need to be imported. 106 if (IsThinLTO) 107 LldArgs.push_back(Args.MakeArgString("-plugin-opt=-force-import-all")); 108 109 for (const Arg *A : Args.filtered(options::OPT_mllvm)) { 110 LldArgs.push_back( 111 Args.MakeArgString(Twine("-plugin-opt=") + A->getValue(0))); 112 } 113 114 if (C.getDriver().isSaveTempsEnabled()) 115 LldArgs.push_back("-save-temps"); 116 117 addLinkerCompressDebugSectionsOption(TC, Args, LldArgs); 118 119 // Given that host and device linking happen in separate processes, the device 120 // linker doesn't always have the visibility as to which device symbols are 121 // needed by a program, especially for the device symbol dependencies that are 122 // introduced through the host symbol resolution. 123 // For example: host_A() (A.obj) --> host_B(B.obj) --> device_kernel_B() 124 // (B.obj) In this case, the device linker doesn't know that A.obj actually 125 // depends on the kernel functions in B.obj. When linking to static device 126 // library, the device linker may drop some of the device global symbols if 127 // they aren't referenced. As a workaround, we are adding to the 128 // --whole-archive flag such that all global symbols would be linked in. 129 LldArgs.push_back("--whole-archive"); 130 131 for (auto *Arg : Args.filtered(options::OPT_Xoffload_linker)) { 132 StringRef ArgVal = Arg->getValue(1); 133 auto SplitArg = ArgVal.split("-mllvm="); 134 if (!SplitArg.second.empty()) { 135 LldArgs.push_back( 136 Args.MakeArgString(Twine("-plugin-opt=") + SplitArg.second)); 137 } else { 138 LldArgs.push_back(Args.MakeArgString(ArgVal)); 139 } 140 Arg->claim(); 141 } 142 143 LldArgs.append({"-o", Output.getFilename()}); 144 for (auto Input : Inputs) 145 LldArgs.push_back(Input.getFilename()); 146 147 // Look for archive of bundled bitcode in arguments, and add temporary files 148 // for the extracted archive of bitcode to inputs. 149 auto TargetID = Args.getLastArgValue(options::OPT_mcpu_EQ); 150 AddStaticDeviceLibsLinking(C, *this, JA, Inputs, Args, LldArgs, "amdgcn", 151 TargetID, /*IsBitCodeSDL=*/true); 152 153 LldArgs.push_back("--no-whole-archive"); 154 155 const char *Lld = Args.MakeArgString(getToolChain().GetProgramPath("lld")); 156 C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(), 157 Lld, LldArgs, Inputs, Output)); 158 } 159 160 // For SPIR-V the inputs for the job are device AMDGCN SPIR-V flavoured bitcode 161 // and the output is either a compiled SPIR-V binary or bitcode (-emit-llvm). It 162 // calls llvm-link and then the llvm-spirv translator. Once the SPIR-V BE will 163 // be promoted from experimental, we will switch to using that. TODO: consider 164 // if we want to run any targeted optimisations over IR here, over generic 165 // SPIR-V. 166 void AMDGCN::Linker::constructLinkAndEmitSpirvCommand( 167 Compilation &C, const JobAction &JA, const InputInfoList &Inputs, 168 const InputInfo &Output, const llvm::opt::ArgList &Args) const { 169 assert(!Inputs.empty() && "Must have at least one input."); 170 171 constructLlvmLinkCommand(C, JA, Inputs, Output, Args); 172 173 // Linked BC is now in Output 174 175 // Emit SPIR-V binary. 176 llvm::opt::ArgStringList TrArgs{ 177 "--spirv-max-version=1.6", 178 "--spirv-ext=+all", 179 "--spirv-allow-unknown-intrinsics", 180 "--spirv-lower-const-expr", 181 "--spirv-preserve-auxdata", 182 "--spirv-debug-info-version=nonsemantic-shader-200"}; 183 SPIRV::constructTranslateCommand(C, *this, JA, Output, Output, TrArgs); 184 } 185 186 // For amdgcn the inputs of the linker job are device bitcode and output is 187 // either an object file or bitcode (-emit-llvm). It calls llvm-link, opt, 188 // llc, then lld steps. 189 void AMDGCN::Linker::ConstructJob(Compilation &C, const JobAction &JA, 190 const InputInfo &Output, 191 const InputInfoList &Inputs, 192 const ArgList &Args, 193 const char *LinkingOutput) const { 194 if (Inputs.size() > 0 && 195 Inputs[0].getType() == types::TY_Image && 196 JA.getType() == types::TY_Object) 197 return HIP::constructGenerateObjFileFromHIPFatBinary(C, Output, Inputs, 198 Args, JA, *this); 199 200 if (JA.getType() == types::TY_HIP_FATBIN) 201 return HIP::constructHIPFatbinCommand(C, JA, Output.getFilename(), Inputs, 202 Args, *this); 203 204 if (JA.getType() == types::TY_LLVM_BC) 205 return constructLlvmLinkCommand(C, JA, Inputs, Output, Args); 206 207 if (getToolChain().getEffectiveTriple().isSPIRV()) 208 return constructLinkAndEmitSpirvCommand(C, JA, Inputs, Output, Args); 209 210 return constructLldCommand(C, JA, Inputs, Output, Args); 211 } 212 213 HIPAMDToolChain::HIPAMDToolChain(const Driver &D, const llvm::Triple &Triple, 214 const ToolChain &HostTC, const ArgList &Args) 215 : ROCMToolChain(D, Triple, Args), HostTC(HostTC) { 216 // Lookup binaries into the driver directory, this is used to 217 // discover the clang-offload-bundler executable. 218 getProgramPaths().push_back(getDriver().Dir); 219 220 // Diagnose unsupported sanitizer options only once. 221 if (!Args.hasFlag(options::OPT_fgpu_sanitize, options::OPT_fno_gpu_sanitize, 222 true)) 223 return; 224 for (auto *A : Args.filtered(options::OPT_fsanitize_EQ)) { 225 SanitizerMask K = parseSanitizerValue(A->getValue(), /*AllowGroups=*/false); 226 if (K != SanitizerKind::Address) 227 D.getDiags().Report(clang::diag::warn_drv_unsupported_option_for_target) 228 << A->getAsString(Args) << getTriple().str(); 229 } 230 } 231 232 void HIPAMDToolChain::addClangTargetOptions( 233 const llvm::opt::ArgList &DriverArgs, llvm::opt::ArgStringList &CC1Args, 234 Action::OffloadKind DeviceOffloadingKind) const { 235 HostTC.addClangTargetOptions(DriverArgs, CC1Args, DeviceOffloadingKind); 236 237 assert(DeviceOffloadingKind == Action::OFK_HIP && 238 "Only HIP offloading kinds are supported for GPUs."); 239 240 CC1Args.append({"-fcuda-is-device", "-fno-threadsafe-statics"}); 241 242 if (!DriverArgs.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc, 243 false)) 244 CC1Args.append({"-mllvm", "-amdgpu-internalize-symbols"}); 245 if (DriverArgs.hasArgNoClaim(options::OPT_hipstdpar)) 246 CC1Args.append({"-mllvm", "-amdgpu-enable-hipstdpar"}); 247 248 StringRef MaxThreadsPerBlock = 249 DriverArgs.getLastArgValue(options::OPT_gpu_max_threads_per_block_EQ); 250 if (!MaxThreadsPerBlock.empty()) { 251 std::string ArgStr = 252 (Twine("--gpu-max-threads-per-block=") + MaxThreadsPerBlock).str(); 253 CC1Args.push_back(DriverArgs.MakeArgStringRef(ArgStr)); 254 } 255 256 CC1Args.push_back("-fcuda-allow-variadic-functions"); 257 258 // Default to "hidden" visibility, as object level linking will not be 259 // supported for the foreseeable future. 260 if (!DriverArgs.hasArg(options::OPT_fvisibility_EQ, 261 options::OPT_fvisibility_ms_compat)) { 262 CC1Args.append({"-fvisibility=hidden"}); 263 CC1Args.push_back("-fapply-global-visibility-to-externs"); 264 } 265 266 if (getEffectiveTriple().isSPIRV()) { 267 // For SPIR-V we embed the command-line into the generated binary, in order 268 // to retrieve it at JIT time and be able to do target specific compilation 269 // with options that match the user-supplied ones. 270 if (!DriverArgs.hasArg(options::OPT_fembed_bitcode_marker)) 271 CC1Args.push_back("-fembed-bitcode=marker"); 272 return; // No DeviceLibs for SPIR-V. 273 } 274 275 for (auto BCFile : getDeviceLibs(DriverArgs)) { 276 CC1Args.push_back(BCFile.ShouldInternalize ? "-mlink-builtin-bitcode" 277 : "-mlink-bitcode-file"); 278 CC1Args.push_back(DriverArgs.MakeArgString(BCFile.Path)); 279 } 280 } 281 282 llvm::opt::DerivedArgList * 283 HIPAMDToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args, 284 StringRef BoundArch, 285 Action::OffloadKind DeviceOffloadKind) const { 286 DerivedArgList *DAL = 287 HostTC.TranslateArgs(Args, BoundArch, DeviceOffloadKind); 288 if (!DAL) 289 DAL = new DerivedArgList(Args.getBaseArgs()); 290 291 const OptTable &Opts = getDriver().getOpts(); 292 293 for (Arg *A : Args) { 294 if (!shouldSkipSanitizeOption(*this, Args, BoundArch, A)) 295 DAL->append(A); 296 } 297 298 if (!BoundArch.empty()) { 299 DAL->eraseArg(options::OPT_mcpu_EQ); 300 DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_mcpu_EQ), BoundArch); 301 checkTargetID(*DAL); 302 } 303 304 return DAL; 305 } 306 307 Tool *HIPAMDToolChain::buildLinker() const { 308 assert(getTriple().getArch() == llvm::Triple::amdgcn || 309 getTriple().getArch() == llvm::Triple::spirv64); 310 return new tools::AMDGCN::Linker(*this); 311 } 312 313 void HIPAMDToolChain::addClangWarningOptions(ArgStringList &CC1Args) const { 314 AMDGPUToolChain::addClangWarningOptions(CC1Args); 315 HostTC.addClangWarningOptions(CC1Args); 316 } 317 318 ToolChain::CXXStdlibType 319 HIPAMDToolChain::GetCXXStdlibType(const ArgList &Args) const { 320 return HostTC.GetCXXStdlibType(Args); 321 } 322 323 void HIPAMDToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs, 324 ArgStringList &CC1Args) const { 325 HostTC.AddClangSystemIncludeArgs(DriverArgs, CC1Args); 326 } 327 328 void HIPAMDToolChain::AddClangCXXStdlibIncludeArgs( 329 const ArgList &Args, ArgStringList &CC1Args) const { 330 HostTC.AddClangCXXStdlibIncludeArgs(Args, CC1Args); 331 } 332 333 void HIPAMDToolChain::AddIAMCUIncludeArgs(const ArgList &Args, 334 ArgStringList &CC1Args) const { 335 HostTC.AddIAMCUIncludeArgs(Args, CC1Args); 336 } 337 338 void HIPAMDToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs, 339 ArgStringList &CC1Args) const { 340 RocmInstallation->AddHIPIncludeArgs(DriverArgs, CC1Args); 341 } 342 343 SanitizerMask HIPAMDToolChain::getSupportedSanitizers() const { 344 // The HIPAMDToolChain only supports sanitizers in the sense that it allows 345 // sanitizer arguments on the command line if they are supported by the host 346 // toolchain. The HIPAMDToolChain will actually ignore any command line 347 // arguments for any of these "supported" sanitizers. That means that no 348 // sanitization of device code is actually supported at this time. 349 // 350 // This behavior is necessary because the host and device toolchains 351 // invocations often share the command line, so the device toolchain must 352 // tolerate flags meant only for the host toolchain. 353 return HostTC.getSupportedSanitizers(); 354 } 355 356 VersionTuple HIPAMDToolChain::computeMSVCVersion(const Driver *D, 357 const ArgList &Args) const { 358 return HostTC.computeMSVCVersion(D, Args); 359 } 360 361 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12> 362 HIPAMDToolChain::getDeviceLibs(const llvm::opt::ArgList &DriverArgs) const { 363 llvm::SmallVector<BitCodeLibraryInfo, 12> BCLibs; 364 if (DriverArgs.hasArg(options::OPT_nogpulib) || 365 getGPUArch(DriverArgs) == "amdgcnspirv") 366 return {}; 367 ArgStringList LibraryPaths; 368 369 // Find in --hip-device-lib-path and HIP_LIBRARY_PATH. 370 for (StringRef Path : RocmInstallation->getRocmDeviceLibPathArg()) 371 LibraryPaths.push_back(DriverArgs.MakeArgString(Path)); 372 373 addDirectoryList(DriverArgs, LibraryPaths, "", "HIP_DEVICE_LIB_PATH"); 374 375 // Maintain compatability with --hip-device-lib. 376 auto BCLibArgs = DriverArgs.getAllArgValues(options::OPT_hip_device_lib_EQ); 377 if (!BCLibArgs.empty()) { 378 llvm::for_each(BCLibArgs, [&](StringRef BCName) { 379 StringRef FullName; 380 for (StringRef LibraryPath : LibraryPaths) { 381 SmallString<128> Path(LibraryPath); 382 llvm::sys::path::append(Path, BCName); 383 FullName = Path; 384 if (llvm::sys::fs::exists(FullName)) { 385 BCLibs.push_back(FullName); 386 return; 387 } 388 } 389 getDriver().Diag(diag::err_drv_no_such_file) << BCName; 390 }); 391 } else { 392 if (!RocmInstallation->hasDeviceLibrary()) { 393 getDriver().Diag(diag::err_drv_no_rocm_device_lib) << 0; 394 return {}; 395 } 396 StringRef GpuArch = getGPUArch(DriverArgs); 397 assert(!GpuArch.empty() && "Must have an explicit GPU arch."); 398 399 // If --hip-device-lib is not set, add the default bitcode libraries. 400 if (DriverArgs.hasFlag(options::OPT_fgpu_sanitize, 401 options::OPT_fno_gpu_sanitize, true) && 402 getSanitizerArgs(DriverArgs).needsAsanRt()) { 403 auto AsanRTL = RocmInstallation->getAsanRTLPath(); 404 if (AsanRTL.empty()) { 405 unsigned DiagID = getDriver().getDiags().getCustomDiagID( 406 DiagnosticsEngine::Error, 407 "AMDGPU address sanitizer runtime library (asanrtl) is not found. " 408 "Please install ROCm device library which supports address " 409 "sanitizer"); 410 getDriver().Diag(DiagID); 411 return {}; 412 } else 413 BCLibs.emplace_back(AsanRTL, /*ShouldInternalize=*/false); 414 } 415 416 // Add the HIP specific bitcode library. 417 BCLibs.push_back(RocmInstallation->getHIPPath()); 418 419 // Add common device libraries like ocml etc. 420 for (StringRef N : getCommonDeviceLibNames(DriverArgs, GpuArch.str())) 421 BCLibs.emplace_back(N); 422 423 // Add instrument lib. 424 auto InstLib = 425 DriverArgs.getLastArgValue(options::OPT_gpu_instrument_lib_EQ); 426 if (InstLib.empty()) 427 return BCLibs; 428 if (llvm::sys::fs::exists(InstLib)) 429 BCLibs.push_back(InstLib); 430 else 431 getDriver().Diag(diag::err_drv_no_such_file) << InstLib; 432 } 433 434 return BCLibs; 435 } 436 437 void HIPAMDToolChain::checkTargetID( 438 const llvm::opt::ArgList &DriverArgs) const { 439 auto PTID = getParsedTargetID(DriverArgs); 440 if (PTID.OptionalTargetID && !PTID.OptionalGPUArch && 441 PTID.OptionalTargetID != "amdgcnspirv") 442 getDriver().Diag(clang::diag::err_drv_bad_target_id) 443 << *PTID.OptionalTargetID; 444 } 445