1 //===--- AMDGPU.cpp - AMDGPU 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 "AMDGPU.h"
10 #include "CommonArgs.h"
11 #include "clang/Basic/TargetID.h"
12 #include "clang/Config/config.h"
13 #include "clang/Driver/Compilation.h"
14 #include "clang/Driver/DriverDiagnostic.h"
15 #include "clang/Driver/InputInfo.h"
16 #include "clang/Driver/Options.h"
17 #include "clang/Driver/SanitizerArgs.h"
18 #include "llvm/ADT/StringExtras.h"
19 #include "llvm/Option/ArgList.h"
20 #include "llvm/Support/Error.h"
21 #include "llvm/Support/LineIterator.h"
22 #include "llvm/Support/Path.h"
23 #include "llvm/Support/Process.h"
24 #include "llvm/Support/VirtualFileSystem.h"
25 #include "llvm/TargetParser/Host.h"
26 #include <optional>
27 #include <system_error>
28
29 using namespace clang::driver;
30 using namespace clang::driver::tools;
31 using namespace clang::driver::toolchains;
32 using namespace clang;
33 using namespace llvm::opt;
34
35 // Look for sub-directory starts with PackageName under ROCm candidate path.
36 // If there is one and only one matching sub-directory found, append the
37 // sub-directory to Path. If there is no matching sub-directory or there are
38 // more than one matching sub-directories, diagnose them. Returns the full
39 // path of the package if there is only one matching sub-directory, otherwise
40 // returns an empty string.
41 llvm::SmallString<0>
findSPACKPackage(const Candidate & Cand,StringRef PackageName)42 RocmInstallationDetector::findSPACKPackage(const Candidate &Cand,
43 StringRef PackageName) {
44 if (!Cand.isSPACK())
45 return {};
46 std::error_code EC;
47 std::string Prefix = Twine(PackageName + "-" + Cand.SPACKReleaseStr).str();
48 llvm::SmallVector<llvm::SmallString<0>> SubDirs;
49 for (llvm::vfs::directory_iterator File = D.getVFS().dir_begin(Cand.Path, EC),
50 FileEnd;
51 File != FileEnd && !EC; File.increment(EC)) {
52 llvm::StringRef FileName = llvm::sys::path::filename(File->path());
53 if (FileName.starts_with(Prefix)) {
54 SubDirs.push_back(FileName);
55 if (SubDirs.size() > 1)
56 break;
57 }
58 }
59 if (SubDirs.size() == 1) {
60 auto PackagePath = Cand.Path;
61 llvm::sys::path::append(PackagePath, SubDirs[0]);
62 return PackagePath;
63 }
64 if (SubDirs.size() == 0 && Verbose) {
65 llvm::errs() << "SPACK package " << Prefix << " not found at " << Cand.Path
66 << '\n';
67 return {};
68 }
69
70 if (SubDirs.size() > 1 && Verbose) {
71 llvm::errs() << "Cannot use SPACK package " << Prefix << " at " << Cand.Path
72 << " due to multiple installations for the same version\n";
73 }
74 return {};
75 }
76
scanLibDevicePath(llvm::StringRef Path)77 void RocmInstallationDetector::scanLibDevicePath(llvm::StringRef Path) {
78 assert(!Path.empty());
79
80 const StringRef Suffix(".bc");
81 const StringRef Suffix2(".amdgcn.bc");
82
83 std::error_code EC;
84 for (llvm::vfs::directory_iterator LI = D.getVFS().dir_begin(Path, EC), LE;
85 !EC && LI != LE; LI = LI.increment(EC)) {
86 StringRef FilePath = LI->path();
87 StringRef FileName = llvm::sys::path::filename(FilePath);
88 if (!FileName.ends_with(Suffix))
89 continue;
90
91 StringRef BaseName;
92 if (FileName.ends_with(Suffix2))
93 BaseName = FileName.drop_back(Suffix2.size());
94 else if (FileName.ends_with(Suffix))
95 BaseName = FileName.drop_back(Suffix.size());
96
97 const StringRef ABIVersionPrefix = "oclc_abi_version_";
98 if (BaseName == "ocml") {
99 OCML = FilePath;
100 } else if (BaseName == "ockl") {
101 OCKL = FilePath;
102 } else if (BaseName == "opencl") {
103 OpenCL = FilePath;
104 } else if (BaseName == "hip") {
105 HIP = FilePath;
106 } else if (BaseName == "asanrtl") {
107 AsanRTL = FilePath;
108 } else if (BaseName == "oclc_finite_only_off") {
109 FiniteOnly.Off = FilePath;
110 } else if (BaseName == "oclc_finite_only_on") {
111 FiniteOnly.On = FilePath;
112 } else if (BaseName == "oclc_daz_opt_on") {
113 DenormalsAreZero.On = FilePath;
114 } else if (BaseName == "oclc_daz_opt_off") {
115 DenormalsAreZero.Off = FilePath;
116 } else if (BaseName == "oclc_correctly_rounded_sqrt_on") {
117 CorrectlyRoundedSqrt.On = FilePath;
118 } else if (BaseName == "oclc_correctly_rounded_sqrt_off") {
119 CorrectlyRoundedSqrt.Off = FilePath;
120 } else if (BaseName == "oclc_unsafe_math_on") {
121 UnsafeMath.On = FilePath;
122 } else if (BaseName == "oclc_unsafe_math_off") {
123 UnsafeMath.Off = FilePath;
124 } else if (BaseName == "oclc_wavefrontsize64_on") {
125 WavefrontSize64.On = FilePath;
126 } else if (BaseName == "oclc_wavefrontsize64_off") {
127 WavefrontSize64.Off = FilePath;
128 } else if (BaseName.starts_with(ABIVersionPrefix)) {
129 unsigned ABIVersionNumber;
130 if (BaseName.drop_front(ABIVersionPrefix.size())
131 .getAsInteger(/*Redex=*/0, ABIVersionNumber))
132 continue;
133 ABIVersionMap[ABIVersionNumber] = FilePath.str();
134 } else {
135 // Process all bitcode filenames that look like
136 // ocl_isa_version_XXX.amdgcn.bc
137 const StringRef DeviceLibPrefix = "oclc_isa_version_";
138 if (!BaseName.starts_with(DeviceLibPrefix))
139 continue;
140
141 StringRef IsaVersionNumber =
142 BaseName.drop_front(DeviceLibPrefix.size());
143
144 llvm::Twine GfxName = Twine("gfx") + IsaVersionNumber;
145 SmallString<8> Tmp;
146 LibDeviceMap.insert(
147 std::make_pair(GfxName.toStringRef(Tmp), FilePath.str()));
148 }
149 }
150 }
151
152 // Parse and extract version numbers from `.hipVersion`. Return `true` if
153 // the parsing fails.
parseHIPVersionFile(llvm::StringRef V)154 bool RocmInstallationDetector::parseHIPVersionFile(llvm::StringRef V) {
155 SmallVector<StringRef, 4> VersionParts;
156 V.split(VersionParts, '\n');
157 unsigned Major = ~0U;
158 unsigned Minor = ~0U;
159 for (auto Part : VersionParts) {
160 auto Splits = Part.rtrim().split('=');
161 if (Splits.first == "HIP_VERSION_MAJOR") {
162 if (Splits.second.getAsInteger(0, Major))
163 return true;
164 } else if (Splits.first == "HIP_VERSION_MINOR") {
165 if (Splits.second.getAsInteger(0, Minor))
166 return true;
167 } else if (Splits.first == "HIP_VERSION_PATCH")
168 VersionPatch = Splits.second.str();
169 }
170 if (Major == ~0U || Minor == ~0U)
171 return true;
172 VersionMajorMinor = llvm::VersionTuple(Major, Minor);
173 DetectedVersion =
174 (Twine(Major) + "." + Twine(Minor) + "." + VersionPatch).str();
175 return false;
176 }
177
178 /// \returns a list of candidate directories for ROCm installation, which is
179 /// cached and populated only once.
180 const SmallVectorImpl<RocmInstallationDetector::Candidate> &
getInstallationPathCandidates()181 RocmInstallationDetector::getInstallationPathCandidates() {
182
183 // Return the cached candidate list if it has already been populated.
184 if (!ROCmSearchDirs.empty())
185 return ROCmSearchDirs;
186
187 auto DoPrintROCmSearchDirs = [&]() {
188 if (PrintROCmSearchDirs)
189 for (auto Cand : ROCmSearchDirs) {
190 llvm::errs() << "ROCm installation search path";
191 if (Cand.isSPACK())
192 llvm::errs() << " (Spack " << Cand.SPACKReleaseStr << ")";
193 llvm::errs() << ": " << Cand.Path << '\n';
194 }
195 };
196
197 // For candidate specified by --rocm-path we do not do strict check, i.e.,
198 // checking existence of HIP version file and device library files.
199 if (!RocmPathArg.empty()) {
200 ROCmSearchDirs.emplace_back(RocmPathArg.str());
201 DoPrintROCmSearchDirs();
202 return ROCmSearchDirs;
203 } else if (std::optional<std::string> RocmPathEnv =
204 llvm::sys::Process::GetEnv("ROCM_PATH")) {
205 if (!RocmPathEnv->empty()) {
206 ROCmSearchDirs.emplace_back(std::move(*RocmPathEnv));
207 DoPrintROCmSearchDirs();
208 return ROCmSearchDirs;
209 }
210 }
211
212 // Try to find relative to the compiler binary.
213 StringRef InstallDir = D.Dir;
214
215 // Check both a normal Unix prefix position of the clang binary, as well as
216 // the Windows-esque layout the ROCm packages use with the host architecture
217 // subdirectory of bin.
218 auto DeduceROCmPath = [](StringRef ClangPath) {
219 // Strip off directory (usually bin)
220 StringRef ParentDir = llvm::sys::path::parent_path(ClangPath);
221 StringRef ParentName = llvm::sys::path::filename(ParentDir);
222
223 // Some builds use bin/{host arch}, so go up again.
224 if (ParentName == "bin") {
225 ParentDir = llvm::sys::path::parent_path(ParentDir);
226 ParentName = llvm::sys::path::filename(ParentDir);
227 }
228
229 // Detect ROCm packages built with SPACK.
230 // clang is installed at
231 // <rocm_root>/llvm-amdgpu-<rocm_release_string>-<hash>/bin directory.
232 // We only consider the parent directory of llvm-amdgpu package as ROCm
233 // installation candidate for SPACK.
234 if (ParentName.starts_with("llvm-amdgpu-")) {
235 auto SPACKPostfix =
236 ParentName.drop_front(strlen("llvm-amdgpu-")).split('-');
237 auto SPACKReleaseStr = SPACKPostfix.first;
238 if (!SPACKReleaseStr.empty()) {
239 ParentDir = llvm::sys::path::parent_path(ParentDir);
240 return Candidate(ParentDir.str(), /*StrictChecking=*/true,
241 SPACKReleaseStr);
242 }
243 }
244
245 // Some versions of the rocm llvm package install to /opt/rocm/llvm/bin
246 // Some versions of the aomp package install to /opt/rocm/aomp/bin
247 if (ParentName == "llvm" || ParentName.starts_with("aomp"))
248 ParentDir = llvm::sys::path::parent_path(ParentDir);
249
250 return Candidate(ParentDir.str(), /*StrictChecking=*/true);
251 };
252
253 // Deduce ROCm path by the path used to invoke clang. Do not resolve symbolic
254 // link of clang itself.
255 ROCmSearchDirs.emplace_back(DeduceROCmPath(InstallDir));
256
257 // Deduce ROCm path by the real path of the invoked clang, resolving symbolic
258 // link of clang itself.
259 llvm::SmallString<256> RealClangPath;
260 llvm::sys::fs::real_path(D.getClangProgramPath(), RealClangPath);
261 auto ParentPath = llvm::sys::path::parent_path(RealClangPath);
262 if (ParentPath != InstallDir)
263 ROCmSearchDirs.emplace_back(DeduceROCmPath(ParentPath));
264
265 // Device library may be installed in clang or resource directory.
266 auto ClangRoot = llvm::sys::path::parent_path(InstallDir);
267 auto RealClangRoot = llvm::sys::path::parent_path(ParentPath);
268 ROCmSearchDirs.emplace_back(ClangRoot.str(), /*StrictChecking=*/true);
269 if (RealClangRoot != ClangRoot)
270 ROCmSearchDirs.emplace_back(RealClangRoot.str(), /*StrictChecking=*/true);
271 ROCmSearchDirs.emplace_back(D.ResourceDir,
272 /*StrictChecking=*/true);
273
274 ROCmSearchDirs.emplace_back(D.SysRoot + "/opt/rocm",
275 /*StrictChecking=*/true);
276
277 // Find the latest /opt/rocm-{release} directory.
278 std::error_code EC;
279 std::string LatestROCm;
280 llvm::VersionTuple LatestVer;
281 // Get ROCm version from ROCm directory name.
282 auto GetROCmVersion = [](StringRef DirName) {
283 llvm::VersionTuple V;
284 std::string VerStr = DirName.drop_front(strlen("rocm-")).str();
285 // The ROCm directory name follows the format of
286 // rocm-{major}.{minor}.{subMinor}[-{build}]
287 std::replace(VerStr.begin(), VerStr.end(), '-', '.');
288 V.tryParse(VerStr);
289 return V;
290 };
291 for (llvm::vfs::directory_iterator
292 File = D.getVFS().dir_begin(D.SysRoot + "/opt", EC),
293 FileEnd;
294 File != FileEnd && !EC; File.increment(EC)) {
295 llvm::StringRef FileName = llvm::sys::path::filename(File->path());
296 if (!FileName.starts_with("rocm-"))
297 continue;
298 if (LatestROCm.empty()) {
299 LatestROCm = FileName.str();
300 LatestVer = GetROCmVersion(LatestROCm);
301 continue;
302 }
303 auto Ver = GetROCmVersion(FileName);
304 if (LatestVer < Ver) {
305 LatestROCm = FileName.str();
306 LatestVer = Ver;
307 }
308 }
309 if (!LatestROCm.empty())
310 ROCmSearchDirs.emplace_back(D.SysRoot + "/opt/" + LatestROCm,
311 /*StrictChecking=*/true);
312
313 ROCmSearchDirs.emplace_back(D.SysRoot + "/usr/local",
314 /*StrictChecking=*/true);
315 ROCmSearchDirs.emplace_back(D.SysRoot + "/usr",
316 /*StrictChecking=*/true);
317
318 DoPrintROCmSearchDirs();
319 return ROCmSearchDirs;
320 }
321
RocmInstallationDetector(const Driver & D,const llvm::Triple & HostTriple,const llvm::opt::ArgList & Args,bool DetectHIPRuntime,bool DetectDeviceLib)322 RocmInstallationDetector::RocmInstallationDetector(
323 const Driver &D, const llvm::Triple &HostTriple,
324 const llvm::opt::ArgList &Args, bool DetectHIPRuntime, bool DetectDeviceLib)
325 : D(D) {
326 Verbose = Args.hasArg(options::OPT_v);
327 RocmPathArg = Args.getLastArgValue(clang::driver::options::OPT_rocm_path_EQ);
328 PrintROCmSearchDirs =
329 Args.hasArg(clang::driver::options::OPT_print_rocm_search_dirs);
330 RocmDeviceLibPathArg =
331 Args.getAllArgValues(clang::driver::options::OPT_rocm_device_lib_path_EQ);
332 HIPPathArg = Args.getLastArgValue(clang::driver::options::OPT_hip_path_EQ);
333 HIPStdParPathArg =
334 Args.getLastArgValue(clang::driver::options::OPT_hipstdpar_path_EQ);
335 HasHIPStdParLibrary =
336 !HIPStdParPathArg.empty() && D.getVFS().exists(HIPStdParPathArg +
337 "/hipstdpar_lib.hpp");
338 HIPRocThrustPathArg =
339 Args.getLastArgValue(clang::driver::options::OPT_hipstdpar_thrust_path_EQ);
340 HasRocThrustLibrary = !HIPRocThrustPathArg.empty() &&
341 D.getVFS().exists(HIPRocThrustPathArg + "/thrust");
342 HIPRocPrimPathArg =
343 Args.getLastArgValue(clang::driver::options::OPT_hipstdpar_prim_path_EQ);
344 HasRocPrimLibrary = !HIPRocPrimPathArg.empty() &&
345 D.getVFS().exists(HIPRocPrimPathArg + "/rocprim");
346
347 if (auto *A = Args.getLastArg(clang::driver::options::OPT_hip_version_EQ)) {
348 HIPVersionArg = A->getValue();
349 unsigned Major = ~0U;
350 unsigned Minor = ~0U;
351 SmallVector<StringRef, 3> Parts;
352 HIPVersionArg.split(Parts, '.');
353 if (Parts.size())
354 Parts[0].getAsInteger(0, Major);
355 if (Parts.size() > 1)
356 Parts[1].getAsInteger(0, Minor);
357 if (Parts.size() > 2)
358 VersionPatch = Parts[2].str();
359 if (VersionPatch.empty())
360 VersionPatch = "0";
361 if (Major != ~0U && Minor == ~0U)
362 Minor = 0;
363 if (Major == ~0U || Minor == ~0U)
364 D.Diag(diag::err_drv_invalid_value)
365 << A->getAsString(Args) << HIPVersionArg;
366
367 VersionMajorMinor = llvm::VersionTuple(Major, Minor);
368 DetectedVersion =
369 (Twine(Major) + "." + Twine(Minor) + "." + VersionPatch).str();
370 } else {
371 VersionPatch = DefaultVersionPatch;
372 VersionMajorMinor =
373 llvm::VersionTuple(DefaultVersionMajor, DefaultVersionMinor);
374 DetectedVersion = (Twine(DefaultVersionMajor) + "." +
375 Twine(DefaultVersionMinor) + "." + VersionPatch)
376 .str();
377 }
378
379 if (DetectHIPRuntime)
380 detectHIPRuntime();
381 if (DetectDeviceLib)
382 detectDeviceLibrary();
383 }
384
detectDeviceLibrary()385 void RocmInstallationDetector::detectDeviceLibrary() {
386 assert(LibDevicePath.empty());
387
388 if (!RocmDeviceLibPathArg.empty())
389 LibDevicePath = RocmDeviceLibPathArg[RocmDeviceLibPathArg.size() - 1];
390 else if (std::optional<std::string> LibPathEnv =
391 llvm::sys::Process::GetEnv("HIP_DEVICE_LIB_PATH"))
392 LibDevicePath = std::move(*LibPathEnv);
393
394 auto &FS = D.getVFS();
395 if (!LibDevicePath.empty()) {
396 // Maintain compatability with HIP flag/envvar pointing directly at the
397 // bitcode library directory. This points directly at the library path instead
398 // of the rocm root installation.
399 if (!FS.exists(LibDevicePath))
400 return;
401
402 scanLibDevicePath(LibDevicePath);
403 HasDeviceLibrary = allGenericLibsValid() && !LibDeviceMap.empty();
404 return;
405 }
406
407 // Check device library exists at the given path.
408 auto CheckDeviceLib = [&](StringRef Path, bool StrictChecking) {
409 bool CheckLibDevice = (!NoBuiltinLibs || StrictChecking);
410 if (CheckLibDevice && !FS.exists(Path))
411 return false;
412
413 scanLibDevicePath(Path);
414
415 if (!NoBuiltinLibs) {
416 // Check that the required non-target libraries are all available.
417 if (!allGenericLibsValid())
418 return false;
419
420 // Check that we have found at least one libdevice that we can link in
421 // if -nobuiltinlib hasn't been specified.
422 if (LibDeviceMap.empty())
423 return false;
424 }
425 return true;
426 };
427
428 // Find device libraries in <LLVM_DIR>/lib/clang/<ver>/lib/amdgcn/bitcode
429 LibDevicePath = D.ResourceDir;
430 llvm::sys::path::append(LibDevicePath, CLANG_INSTALL_LIBDIR_BASENAME,
431 "amdgcn", "bitcode");
432 HasDeviceLibrary = CheckDeviceLib(LibDevicePath, true);
433 if (HasDeviceLibrary)
434 return;
435
436 // Find device libraries in a legacy ROCm directory structure
437 // ${ROCM_ROOT}/amdgcn/bitcode/*
438 auto &ROCmDirs = getInstallationPathCandidates();
439 for (const auto &Candidate : ROCmDirs) {
440 LibDevicePath = Candidate.Path;
441 llvm::sys::path::append(LibDevicePath, "amdgcn", "bitcode");
442 HasDeviceLibrary = CheckDeviceLib(LibDevicePath, Candidate.StrictChecking);
443 if (HasDeviceLibrary)
444 return;
445 }
446 }
447
detectHIPRuntime()448 void RocmInstallationDetector::detectHIPRuntime() {
449 SmallVector<Candidate, 4> HIPSearchDirs;
450 if (!HIPPathArg.empty())
451 HIPSearchDirs.emplace_back(HIPPathArg.str());
452 else if (std::optional<std::string> HIPPathEnv =
453 llvm::sys::Process::GetEnv("HIP_PATH")) {
454 if (!HIPPathEnv->empty())
455 HIPSearchDirs.emplace_back(std::move(*HIPPathEnv));
456 }
457 if (HIPSearchDirs.empty())
458 HIPSearchDirs.append(getInstallationPathCandidates());
459 auto &FS = D.getVFS();
460
461 for (const auto &Candidate : HIPSearchDirs) {
462 InstallPath = Candidate.Path;
463 if (InstallPath.empty() || !FS.exists(InstallPath))
464 continue;
465 // HIP runtime built by SPACK is installed to
466 // <rocm_root>/hip-<rocm_release_string>-<hash> directory.
467 auto SPACKPath = findSPACKPackage(Candidate, "hip");
468 InstallPath = SPACKPath.empty() ? InstallPath : SPACKPath;
469
470 BinPath = InstallPath;
471 llvm::sys::path::append(BinPath, "bin");
472 IncludePath = InstallPath;
473 llvm::sys::path::append(IncludePath, "include");
474 LibPath = InstallPath;
475 llvm::sys::path::append(LibPath, "lib");
476 SharePath = InstallPath;
477 llvm::sys::path::append(SharePath, "share");
478
479 // Get parent of InstallPath and append "share"
480 SmallString<0> ParentSharePath = llvm::sys::path::parent_path(InstallPath);
481 llvm::sys::path::append(ParentSharePath, "share");
482
483 auto Append = [](SmallString<0> &path, const Twine &a, const Twine &b = "",
484 const Twine &c = "", const Twine &d = "") {
485 SmallString<0> newpath = path;
486 llvm::sys::path::append(newpath, a, b, c, d);
487 return newpath;
488 };
489 // If HIP version file can be found and parsed, use HIP version from there.
490 std::vector<SmallString<0>> VersionFilePaths = {
491 Append(SharePath, "hip", "version"),
492 InstallPath != D.SysRoot + "/usr/local"
493 ? Append(ParentSharePath, "hip", "version")
494 : SmallString<0>(),
495 Append(BinPath, ".hipVersion")};
496
497 for (const auto &VersionFilePath : VersionFilePaths) {
498 if (VersionFilePath.empty())
499 continue;
500 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> VersionFile =
501 FS.getBufferForFile(VersionFilePath);
502 if (!VersionFile)
503 continue;
504 if (HIPVersionArg.empty() && VersionFile)
505 if (parseHIPVersionFile((*VersionFile)->getBuffer()))
506 continue;
507
508 HasHIPRuntime = true;
509 return;
510 }
511 // Otherwise, if -rocm-path is specified (no strict checking), use the
512 // default HIP version or specified by --hip-version.
513 if (!Candidate.StrictChecking) {
514 HasHIPRuntime = true;
515 return;
516 }
517 }
518 HasHIPRuntime = false;
519 }
520
print(raw_ostream & OS) const521 void RocmInstallationDetector::print(raw_ostream &OS) const {
522 if (hasHIPRuntime())
523 OS << "Found HIP installation: " << InstallPath << ", version "
524 << DetectedVersion << '\n';
525 }
526
AddHIPIncludeArgs(const ArgList & DriverArgs,ArgStringList & CC1Args) const527 void RocmInstallationDetector::AddHIPIncludeArgs(const ArgList &DriverArgs,
528 ArgStringList &CC1Args) const {
529 bool UsesRuntimeWrapper = VersionMajorMinor > llvm::VersionTuple(3, 5) &&
530 !DriverArgs.hasArg(options::OPT_nohipwrapperinc);
531 bool HasHipStdPar = DriverArgs.hasArg(options::OPT_hipstdpar);
532
533 if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
534 // HIP header includes standard library wrapper headers under clang
535 // cuda_wrappers directory. Since these wrapper headers include_next
536 // standard C++ headers, whereas libc++ headers include_next other clang
537 // headers. The include paths have to follow this order:
538 // - wrapper include path
539 // - standard C++ include path
540 // - other clang include path
541 // Since standard C++ and other clang include paths are added in other
542 // places after this function, here we only need to make sure wrapper
543 // include path is added.
544 //
545 // ROCm 3.5 does not fully support the wrapper headers. Therefore it needs
546 // a workaround.
547 SmallString<128> P(D.ResourceDir);
548 if (UsesRuntimeWrapper)
549 llvm::sys::path::append(P, "include", "cuda_wrappers");
550 CC1Args.push_back("-internal-isystem");
551 CC1Args.push_back(DriverArgs.MakeArgString(P));
552 }
553
554 const auto HandleHipStdPar = [=, &DriverArgs, &CC1Args]() {
555 StringRef Inc = getIncludePath();
556 auto &FS = D.getVFS();
557
558 if (!hasHIPStdParLibrary())
559 if (!HIPStdParPathArg.empty() ||
560 !FS.exists(Inc + "/thrust/system/hip/hipstdpar/hipstdpar_lib.hpp")) {
561 D.Diag(diag::err_drv_no_hipstdpar_lib);
562 return;
563 }
564 if (!HasRocThrustLibrary && !FS.exists(Inc + "/thrust")) {
565 D.Diag(diag::err_drv_no_hipstdpar_thrust_lib);
566 return;
567 }
568 if (!HasRocPrimLibrary && !FS.exists(Inc + "/rocprim")) {
569 D.Diag(diag::err_drv_no_hipstdpar_prim_lib);
570 return;
571 }
572 const char *ThrustPath;
573 if (HasRocThrustLibrary)
574 ThrustPath = DriverArgs.MakeArgString(HIPRocThrustPathArg);
575 else
576 ThrustPath = DriverArgs.MakeArgString(Inc + "/thrust");
577
578 const char *HIPStdParPath;
579 if (hasHIPStdParLibrary())
580 HIPStdParPath = DriverArgs.MakeArgString(HIPStdParPathArg);
581 else
582 HIPStdParPath = DriverArgs.MakeArgString(StringRef(ThrustPath) +
583 "/system/hip/hipstdpar");
584
585 const char *PrimPath;
586 if (HasRocPrimLibrary)
587 PrimPath = DriverArgs.MakeArgString(HIPRocPrimPathArg);
588 else
589 PrimPath = DriverArgs.MakeArgString(getIncludePath() + "/rocprim");
590
591 CC1Args.append({"-idirafter", ThrustPath, "-idirafter", PrimPath,
592 "-idirafter", HIPStdParPath, "-include",
593 "hipstdpar_lib.hpp"});
594 };
595
596 if (DriverArgs.hasArg(options::OPT_nogpuinc)) {
597 if (HasHipStdPar)
598 HandleHipStdPar();
599
600 return;
601 }
602
603 if (!hasHIPRuntime()) {
604 D.Diag(diag::err_drv_no_hip_runtime);
605 return;
606 }
607
608 CC1Args.push_back("-idirafter");
609 CC1Args.push_back(DriverArgs.MakeArgString(getIncludePath()));
610 if (UsesRuntimeWrapper)
611 CC1Args.append({"-include", "__clang_hip_runtime_wrapper.h"});
612 if (HasHipStdPar)
613 HandleHipStdPar();
614 }
615
ConstructJob(Compilation & C,const JobAction & JA,const InputInfo & Output,const InputInfoList & Inputs,const ArgList & Args,const char * LinkingOutput) const616 void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
617 const InputInfo &Output,
618 const InputInfoList &Inputs,
619 const ArgList &Args,
620 const char *LinkingOutput) const {
621 std::string Linker = getToolChain().GetLinkerPath();
622 ArgStringList CmdArgs;
623 CmdArgs.push_back("--no-undefined");
624 CmdArgs.push_back("-shared");
625
626 addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
627 Args.AddAllArgs(CmdArgs, options::OPT_L);
628 getToolChain().AddFilePathLibArgs(Args, CmdArgs);
629 AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs, JA);
630 if (C.getDriver().isUsingLTO())
631 addLTOOptions(getToolChain(), Args, CmdArgs, Output, Inputs[0],
632 C.getDriver().getLTOMode() == LTOK_Thin);
633 else if (Args.hasArg(options::OPT_mcpu_EQ))
634 CmdArgs.push_back(Args.MakeArgString(
635 "-plugin-opt=mcpu=" + Args.getLastArgValue(options::OPT_mcpu_EQ)));
636 CmdArgs.push_back("-o");
637 CmdArgs.push_back(Output.getFilename());
638 C.addCommand(std::make_unique<Command>(
639 JA, *this, ResponseFileSupport::AtFileCurCP(), Args.MakeArgString(Linker),
640 CmdArgs, Inputs, Output));
641 }
642
getAMDGPUTargetFeatures(const Driver & D,const llvm::Triple & Triple,const llvm::opt::ArgList & Args,std::vector<StringRef> & Features)643 void amdgpu::getAMDGPUTargetFeatures(const Driver &D,
644 const llvm::Triple &Triple,
645 const llvm::opt::ArgList &Args,
646 std::vector<StringRef> &Features) {
647 // Add target ID features to -target-feature options. No diagnostics should
648 // be emitted here since invalid target ID is diagnosed at other places.
649 StringRef TargetID;
650 if (Args.hasArg(options::OPT_mcpu_EQ))
651 TargetID = Args.getLastArgValue(options::OPT_mcpu_EQ);
652 else if (Args.hasArg(options::OPT_march_EQ))
653 TargetID = Args.getLastArgValue(options::OPT_march_EQ);
654 if (!TargetID.empty()) {
655 llvm::StringMap<bool> FeatureMap;
656 auto OptionalGpuArch = parseTargetID(Triple, TargetID, &FeatureMap);
657 if (OptionalGpuArch) {
658 StringRef GpuArch = *OptionalGpuArch;
659 // Iterate through all possible target ID features for the given GPU.
660 // If it is mapped to true, add +feature.
661 // If it is mapped to false, add -feature.
662 // If it is not in the map (default), do not add it
663 for (auto &&Feature : getAllPossibleTargetIDFeatures(Triple, GpuArch)) {
664 auto Pos = FeatureMap.find(Feature);
665 if (Pos == FeatureMap.end())
666 continue;
667 Features.push_back(Args.MakeArgStringRef(
668 (Twine(Pos->second ? "+" : "-") + Feature).str()));
669 }
670 }
671 }
672
673 if (Args.hasFlag(options::OPT_mwavefrontsize64,
674 options::OPT_mno_wavefrontsize64, false))
675 Features.push_back("+wavefrontsize64");
676
677 if (Args.hasFlag(options::OPT_mamdgpu_precise_memory_op,
678 options::OPT_mno_amdgpu_precise_memory_op, false))
679 Features.push_back("+precise-memory");
680
681 handleTargetFeaturesGroup(D, Triple, Args, Features,
682 options::OPT_m_amdgpu_Features_Group);
683 }
684
685 /// AMDGPU Toolchain
AMDGPUToolChain(const Driver & D,const llvm::Triple & Triple,const ArgList & Args)686 AMDGPUToolChain::AMDGPUToolChain(const Driver &D, const llvm::Triple &Triple,
687 const ArgList &Args)
688 : Generic_ELF(D, Triple, Args),
689 OptionsDefault(
690 {{options::OPT_O, "3"}, {options::OPT_cl_std_EQ, "CL1.2"}}) {
691 // Check code object version options. Emit warnings for legacy options
692 // and errors for the last invalid code object version options.
693 // It is done here to avoid repeated warning or error messages for
694 // each tool invocation.
695 checkAMDGPUCodeObjectVersion(D, Args);
696 }
697
buildLinker() const698 Tool *AMDGPUToolChain::buildLinker() const {
699 return new tools::amdgpu::Linker(*this);
700 }
701
702 DerivedArgList *
TranslateArgs(const DerivedArgList & Args,StringRef BoundArch,Action::OffloadKind DeviceOffloadKind) const703 AMDGPUToolChain::TranslateArgs(const DerivedArgList &Args, StringRef BoundArch,
704 Action::OffloadKind DeviceOffloadKind) const {
705
706 DerivedArgList *DAL =
707 Generic_ELF::TranslateArgs(Args, BoundArch, DeviceOffloadKind);
708
709 const OptTable &Opts = getDriver().getOpts();
710
711 if (!DAL)
712 DAL = new DerivedArgList(Args.getBaseArgs());
713
714 for (Arg *A : Args)
715 DAL->append(A);
716
717 // Replace -mcpu=native with detected GPU.
718 Arg *LastMCPUArg = DAL->getLastArg(options::OPT_mcpu_EQ);
719 if (LastMCPUArg && StringRef(LastMCPUArg->getValue()) == "native") {
720 DAL->eraseArg(options::OPT_mcpu_EQ);
721 auto GPUsOrErr = getSystemGPUArchs(Args);
722 if (!GPUsOrErr) {
723 getDriver().Diag(diag::err_drv_undetermined_gpu_arch)
724 << llvm::Triple::getArchTypeName(getArch())
725 << llvm::toString(GPUsOrErr.takeError()) << "-mcpu";
726 } else {
727 auto &GPUs = *GPUsOrErr;
728 if (GPUs.size() > 1) {
729 getDriver().Diag(diag::warn_drv_multi_gpu_arch)
730 << llvm::Triple::getArchTypeName(getArch())
731 << llvm::join(GPUs, ", ") << "-mcpu";
732 }
733 DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_mcpu_EQ),
734 Args.MakeArgString(GPUs.front()));
735 }
736 }
737
738 checkTargetID(*DAL);
739
740 if (Args.getLastArgValue(options::OPT_x) != "cl")
741 return DAL;
742
743 // Phase 1 (.cl -> .bc)
744 if (Args.hasArg(options::OPT_c) && Args.hasArg(options::OPT_emit_llvm)) {
745 DAL->AddFlagArg(nullptr, Opts.getOption(getTriple().isArch64Bit()
746 ? options::OPT_m64
747 : options::OPT_m32));
748
749 // Have to check OPT_O4, OPT_O0 & OPT_Ofast separately
750 // as they defined that way in Options.td
751 if (!Args.hasArg(options::OPT_O, options::OPT_O0, options::OPT_O4,
752 options::OPT_Ofast))
753 DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_O),
754 getOptionDefault(options::OPT_O));
755 }
756
757 return DAL;
758 }
759
getDefaultDenormsAreZeroForTarget(llvm::AMDGPU::GPUKind Kind)760 bool AMDGPUToolChain::getDefaultDenormsAreZeroForTarget(
761 llvm::AMDGPU::GPUKind Kind) {
762
763 // Assume nothing without a specific target.
764 if (Kind == llvm::AMDGPU::GK_NONE)
765 return false;
766
767 const unsigned ArchAttr = llvm::AMDGPU::getArchAttrAMDGCN(Kind);
768
769 // Default to enabling f32 denormals by default on subtargets where fma is
770 // fast with denormals
771 const bool BothDenormAndFMAFast =
772 (ArchAttr & llvm::AMDGPU::FEATURE_FAST_FMA_F32) &&
773 (ArchAttr & llvm::AMDGPU::FEATURE_FAST_DENORMAL_F32);
774 return !BothDenormAndFMAFast;
775 }
776
getDefaultDenormalModeForType(const llvm::opt::ArgList & DriverArgs,const JobAction & JA,const llvm::fltSemantics * FPType) const777 llvm::DenormalMode AMDGPUToolChain::getDefaultDenormalModeForType(
778 const llvm::opt::ArgList &DriverArgs, const JobAction &JA,
779 const llvm::fltSemantics *FPType) const {
780 // Denormals should always be enabled for f16 and f64.
781 if (!FPType || FPType != &llvm::APFloat::IEEEsingle())
782 return llvm::DenormalMode::getIEEE();
783
784 if (JA.getOffloadingDeviceKind() == Action::OFK_HIP ||
785 JA.getOffloadingDeviceKind() == Action::OFK_Cuda) {
786 auto Arch = getProcessorFromTargetID(getTriple(), JA.getOffloadingArch());
787 auto Kind = llvm::AMDGPU::parseArchAMDGCN(Arch);
788 if (FPType && FPType == &llvm::APFloat::IEEEsingle() &&
789 DriverArgs.hasFlag(options::OPT_fgpu_flush_denormals_to_zero,
790 options::OPT_fno_gpu_flush_denormals_to_zero,
791 getDefaultDenormsAreZeroForTarget(Kind)))
792 return llvm::DenormalMode::getPreserveSign();
793
794 return llvm::DenormalMode::getIEEE();
795 }
796
797 const StringRef GpuArch = getGPUArch(DriverArgs);
798 auto Kind = llvm::AMDGPU::parseArchAMDGCN(GpuArch);
799
800 // TODO: There are way too many flags that change this. Do we need to check
801 // them all?
802 bool DAZ = DriverArgs.hasArg(options::OPT_cl_denorms_are_zero) ||
803 getDefaultDenormsAreZeroForTarget(Kind);
804
805 // Outputs are flushed to zero (FTZ), preserving sign. Denormal inputs are
806 // also implicit treated as zero (DAZ).
807 return DAZ ? llvm::DenormalMode::getPreserveSign() :
808 llvm::DenormalMode::getIEEE();
809 }
810
isWave64(const llvm::opt::ArgList & DriverArgs,llvm::AMDGPU::GPUKind Kind)811 bool AMDGPUToolChain::isWave64(const llvm::opt::ArgList &DriverArgs,
812 llvm::AMDGPU::GPUKind Kind) {
813 const unsigned ArchAttr = llvm::AMDGPU::getArchAttrAMDGCN(Kind);
814 bool HasWave32 = (ArchAttr & llvm::AMDGPU::FEATURE_WAVE32);
815
816 return !HasWave32 || DriverArgs.hasFlag(
817 options::OPT_mwavefrontsize64, options::OPT_mno_wavefrontsize64, false);
818 }
819
820
821 /// ROCM Toolchain
ROCMToolChain(const Driver & D,const llvm::Triple & Triple,const ArgList & Args)822 ROCMToolChain::ROCMToolChain(const Driver &D, const llvm::Triple &Triple,
823 const ArgList &Args)
824 : AMDGPUToolChain(D, Triple, Args) {
825 RocmInstallation->detectDeviceLibrary();
826 }
827
addClangTargetOptions(const llvm::opt::ArgList & DriverArgs,llvm::opt::ArgStringList & CC1Args,Action::OffloadKind DeviceOffloadingKind) const828 void AMDGPUToolChain::addClangTargetOptions(
829 const llvm::opt::ArgList &DriverArgs,
830 llvm::opt::ArgStringList &CC1Args,
831 Action::OffloadKind DeviceOffloadingKind) const {
832 // Default to "hidden" visibility, as object level linking will not be
833 // supported for the foreseeable future.
834 if (!DriverArgs.hasArg(options::OPT_fvisibility_EQ,
835 options::OPT_fvisibility_ms_compat)) {
836 CC1Args.push_back("-fvisibility=hidden");
837 CC1Args.push_back("-fapply-global-visibility-to-externs");
838 }
839 }
840
addClangWarningOptions(ArgStringList & CC1Args) const841 void AMDGPUToolChain::addClangWarningOptions(ArgStringList &CC1Args) const {
842 // AMDGPU does not support atomic lib call. Treat atomic alignment
843 // warnings as errors.
844 CC1Args.push_back("-Werror=atomic-alignment");
845 }
846
847 StringRef
getGPUArch(const llvm::opt::ArgList & DriverArgs) const848 AMDGPUToolChain::getGPUArch(const llvm::opt::ArgList &DriverArgs) const {
849 return getProcessorFromTargetID(
850 getTriple(), DriverArgs.getLastArgValue(options::OPT_mcpu_EQ));
851 }
852
853 AMDGPUToolChain::ParsedTargetIDType
getParsedTargetID(const llvm::opt::ArgList & DriverArgs) const854 AMDGPUToolChain::getParsedTargetID(const llvm::opt::ArgList &DriverArgs) const {
855 StringRef TargetID = DriverArgs.getLastArgValue(options::OPT_mcpu_EQ);
856 if (TargetID.empty())
857 return {std::nullopt, std::nullopt, std::nullopt};
858
859 llvm::StringMap<bool> FeatureMap;
860 auto OptionalGpuArch = parseTargetID(getTriple(), TargetID, &FeatureMap);
861 if (!OptionalGpuArch)
862 return {TargetID.str(), std::nullopt, std::nullopt};
863
864 return {TargetID.str(), OptionalGpuArch->str(), FeatureMap};
865 }
866
checkTargetID(const llvm::opt::ArgList & DriverArgs) const867 void AMDGPUToolChain::checkTargetID(
868 const llvm::opt::ArgList &DriverArgs) const {
869 auto PTID = getParsedTargetID(DriverArgs);
870 if (PTID.OptionalTargetID && !PTID.OptionalGPUArch) {
871 getDriver().Diag(clang::diag::err_drv_bad_target_id)
872 << *PTID.OptionalTargetID;
873 }
874 }
875
876 Expected<SmallVector<std::string>>
getSystemGPUArchs(const ArgList & Args) const877 AMDGPUToolChain::getSystemGPUArchs(const ArgList &Args) const {
878 // Detect AMD GPUs availible on the system.
879 std::string Program;
880 if (Arg *A = Args.getLastArg(options::OPT_amdgpu_arch_tool_EQ))
881 Program = A->getValue();
882 else
883 Program = GetProgramPath("amdgpu-arch");
884
885 auto StdoutOrErr = executeToolChainProgram(Program, /*SecondsToWait=*/10);
886 if (!StdoutOrErr)
887 return StdoutOrErr.takeError();
888
889 SmallVector<std::string, 1> GPUArchs;
890 for (StringRef Arch : llvm::split((*StdoutOrErr)->getBuffer(), "\n"))
891 if (!Arch.empty())
892 GPUArchs.push_back(Arch.str());
893
894 if (GPUArchs.empty())
895 return llvm::createStringError(std::error_code(),
896 "No AMD GPU detected in the system");
897
898 return std::move(GPUArchs);
899 }
900
addClangTargetOptions(const llvm::opt::ArgList & DriverArgs,llvm::opt::ArgStringList & CC1Args,Action::OffloadKind DeviceOffloadingKind) const901 void ROCMToolChain::addClangTargetOptions(
902 const llvm::opt::ArgList &DriverArgs, llvm::opt::ArgStringList &CC1Args,
903 Action::OffloadKind DeviceOffloadingKind) const {
904 AMDGPUToolChain::addClangTargetOptions(DriverArgs, CC1Args,
905 DeviceOffloadingKind);
906
907 // For the OpenCL case where there is no offload target, accept -nostdlib to
908 // disable bitcode linking.
909 if (DeviceOffloadingKind == Action::OFK_None &&
910 DriverArgs.hasArg(options::OPT_nostdlib))
911 return;
912
913 if (DriverArgs.hasArg(options::OPT_nogpulib))
914 return;
915
916 // Get the device name and canonicalize it
917 const StringRef GpuArch = getGPUArch(DriverArgs);
918 auto Kind = llvm::AMDGPU::parseArchAMDGCN(GpuArch);
919 const StringRef CanonArch = llvm::AMDGPU::getArchNameAMDGCN(Kind);
920 StringRef LibDeviceFile = RocmInstallation->getLibDeviceFile(CanonArch);
921 auto ABIVer = DeviceLibABIVersion::fromCodeObjectVersion(
922 getAMDGPUCodeObjectVersion(getDriver(), DriverArgs));
923 if (!RocmInstallation->checkCommonBitcodeLibs(CanonArch, LibDeviceFile,
924 ABIVer))
925 return;
926
927 bool Wave64 = isWave64(DriverArgs, Kind);
928
929 // TODO: There are way too many flags that change this. Do we need to check
930 // them all?
931 bool DAZ = DriverArgs.hasArg(options::OPT_cl_denorms_are_zero) ||
932 getDefaultDenormsAreZeroForTarget(Kind);
933 bool FiniteOnly = DriverArgs.hasArg(options::OPT_cl_finite_math_only);
934
935 bool UnsafeMathOpt =
936 DriverArgs.hasArg(options::OPT_cl_unsafe_math_optimizations);
937 bool FastRelaxedMath = DriverArgs.hasArg(options::OPT_cl_fast_relaxed_math);
938 bool CorrectSqrt =
939 DriverArgs.hasArg(options::OPT_cl_fp32_correctly_rounded_divide_sqrt);
940
941 // Add the OpenCL specific bitcode library.
942 llvm::SmallVector<std::string, 12> BCLibs;
943 BCLibs.push_back(RocmInstallation->getOpenCLPath().str());
944
945 // Add the generic set of libraries.
946 BCLibs.append(RocmInstallation->getCommonBitcodeLibs(
947 DriverArgs, LibDeviceFile, Wave64, DAZ, FiniteOnly, UnsafeMathOpt,
948 FastRelaxedMath, CorrectSqrt, ABIVer, false));
949
950 if (getSanitizerArgs(DriverArgs).needsAsanRt()) {
951 CC1Args.push_back("-mlink-bitcode-file");
952 CC1Args.push_back(
953 DriverArgs.MakeArgString(RocmInstallation->getAsanRTLPath()));
954 }
955 for (StringRef BCFile : BCLibs) {
956 CC1Args.push_back("-mlink-builtin-bitcode");
957 CC1Args.push_back(DriverArgs.MakeArgString(BCFile));
958 }
959 }
960
checkCommonBitcodeLibs(StringRef GPUArch,StringRef LibDeviceFile,DeviceLibABIVersion ABIVer) const961 bool RocmInstallationDetector::checkCommonBitcodeLibs(
962 StringRef GPUArch, StringRef LibDeviceFile,
963 DeviceLibABIVersion ABIVer) const {
964 if (!hasDeviceLibrary()) {
965 D.Diag(diag::err_drv_no_rocm_device_lib) << 0;
966 return false;
967 }
968 if (LibDeviceFile.empty()) {
969 D.Diag(diag::err_drv_no_rocm_device_lib) << 1 << GPUArch;
970 return false;
971 }
972 if (ABIVer.requiresLibrary() && getABIVersionPath(ABIVer).empty()) {
973 D.Diag(diag::err_drv_no_rocm_device_lib) << 2 << ABIVer.toString();
974 return false;
975 }
976 return true;
977 }
978
979 llvm::SmallVector<std::string, 12>
getCommonBitcodeLibs(const llvm::opt::ArgList & DriverArgs,StringRef LibDeviceFile,bool Wave64,bool DAZ,bool FiniteOnly,bool UnsafeMathOpt,bool FastRelaxedMath,bool CorrectSqrt,DeviceLibABIVersion ABIVer,bool isOpenMP=false) const980 RocmInstallationDetector::getCommonBitcodeLibs(
981 const llvm::opt::ArgList &DriverArgs, StringRef LibDeviceFile, bool Wave64,
982 bool DAZ, bool FiniteOnly, bool UnsafeMathOpt, bool FastRelaxedMath,
983 bool CorrectSqrt, DeviceLibABIVersion ABIVer, bool isOpenMP = false) const {
984 llvm::SmallVector<std::string, 12> BCLibs;
985
986 auto AddBCLib = [&](StringRef BCFile) { BCLibs.push_back(BCFile.str()); };
987
988 AddBCLib(getOCMLPath());
989 if (!isOpenMP)
990 AddBCLib(getOCKLPath());
991 AddBCLib(getDenormalsAreZeroPath(DAZ));
992 AddBCLib(getUnsafeMathPath(UnsafeMathOpt || FastRelaxedMath));
993 AddBCLib(getFiniteOnlyPath(FiniteOnly || FastRelaxedMath));
994 AddBCLib(getCorrectlyRoundedSqrtPath(CorrectSqrt));
995 AddBCLib(getWavefrontSize64Path(Wave64));
996 AddBCLib(LibDeviceFile);
997 auto ABIVerPath = getABIVersionPath(ABIVer);
998 if (!ABIVerPath.empty())
999 AddBCLib(ABIVerPath);
1000
1001 return BCLibs;
1002 }
1003
1004 llvm::SmallVector<std::string, 12>
getCommonDeviceLibNames(const llvm::opt::ArgList & DriverArgs,const std::string & GPUArch,bool isOpenMP) const1005 ROCMToolChain::getCommonDeviceLibNames(const llvm::opt::ArgList &DriverArgs,
1006 const std::string &GPUArch,
1007 bool isOpenMP) const {
1008 auto Kind = llvm::AMDGPU::parseArchAMDGCN(GPUArch);
1009 const StringRef CanonArch = llvm::AMDGPU::getArchNameAMDGCN(Kind);
1010
1011 StringRef LibDeviceFile = RocmInstallation->getLibDeviceFile(CanonArch);
1012 auto ABIVer = DeviceLibABIVersion::fromCodeObjectVersion(
1013 getAMDGPUCodeObjectVersion(getDriver(), DriverArgs));
1014 if (!RocmInstallation->checkCommonBitcodeLibs(CanonArch, LibDeviceFile,
1015 ABIVer))
1016 return {};
1017
1018 // If --hip-device-lib is not set, add the default bitcode libraries.
1019 // TODO: There are way too many flags that change this. Do we need to check
1020 // them all?
1021 bool DAZ = DriverArgs.hasFlag(options::OPT_fgpu_flush_denormals_to_zero,
1022 options::OPT_fno_gpu_flush_denormals_to_zero,
1023 getDefaultDenormsAreZeroForTarget(Kind));
1024 bool FiniteOnly = DriverArgs.hasFlag(
1025 options::OPT_ffinite_math_only, options::OPT_fno_finite_math_only, false);
1026 bool UnsafeMathOpt =
1027 DriverArgs.hasFlag(options::OPT_funsafe_math_optimizations,
1028 options::OPT_fno_unsafe_math_optimizations, false);
1029 bool FastRelaxedMath = DriverArgs.hasFlag(options::OPT_ffast_math,
1030 options::OPT_fno_fast_math, false);
1031 bool CorrectSqrt = DriverArgs.hasFlag(
1032 options::OPT_fhip_fp32_correctly_rounded_divide_sqrt,
1033 options::OPT_fno_hip_fp32_correctly_rounded_divide_sqrt, true);
1034 bool Wave64 = isWave64(DriverArgs, Kind);
1035
1036 return RocmInstallation->getCommonBitcodeLibs(
1037 DriverArgs, LibDeviceFile, Wave64, DAZ, FiniteOnly, UnsafeMathOpt,
1038 FastRelaxedMath, CorrectSqrt, ABIVer, isOpenMP);
1039 }
1040