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