1 //===- AMDGPUOpenMP.cpp - AMDGPUOpenMP ToolChain Implementation -*- 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 "AMDGPUOpenMP.h"
10 #include "AMDGPU.h"
11 #include "CommonArgs.h"
12 #include "ToolChains/ROCm.h"
13 #include "clang/Basic/DiagnosticDriver.h"
14 #include "clang/Driver/Compilation.h"
15 #include "clang/Driver/Driver.h"
16 #include "clang/Driver/DriverDiagnostic.h"
17 #include "clang/Driver/InputInfo.h"
18 #include "clang/Driver/Options.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/Support/FileSystem.h"
21 #include "llvm/Support/FormatAdapters.h"
22 #include "llvm/Support/FormatVariadic.h"
23 #include "llvm/Support/Path.h"
24 
25 using namespace clang::driver;
26 using namespace clang::driver::toolchains;
27 using namespace clang::driver::tools;
28 using namespace clang;
29 using namespace llvm::opt;
30 
31 namespace {
32 
33 static const char *getOutputFileName(Compilation &C, StringRef Base,
34                                      const char *Postfix,
35                                      const char *Extension) {
36   const char *OutputFileName;
37   if (C.getDriver().isSaveTempsEnabled()) {
38     OutputFileName =
39         C.getArgs().MakeArgString(Base.str() + Postfix + "." + Extension);
40   } else {
41     std::string TmpName =
42         C.getDriver().GetTemporaryPath(Base.str() + Postfix, Extension);
43     OutputFileName = C.addTempFile(C.getArgs().MakeArgString(TmpName));
44   }
45   return OutputFileName;
46 }
47 
48 static void addLLCOptArg(const llvm::opt::ArgList &Args,
49                          llvm::opt::ArgStringList &CmdArgs) {
50   if (Arg *A = Args.getLastArg(options::OPT_O_Group)) {
51     StringRef OOpt = "0";
52     if (A->getOption().matches(options::OPT_O4) ||
53         A->getOption().matches(options::OPT_Ofast))
54       OOpt = "3";
55     else if (A->getOption().matches(options::OPT_O0))
56       OOpt = "0";
57     else if (A->getOption().matches(options::OPT_O)) {
58       // Clang and opt support -Os/-Oz; llc only supports -O0, -O1, -O2 and -O3
59       // so we map -Os/-Oz to -O2.
60       // Only clang supports -Og, and maps it to -O1.
61       // We map anything else to -O2.
62       OOpt = llvm::StringSwitch<const char *>(A->getValue())
63                  .Case("1", "1")
64                  .Case("2", "2")
65                  .Case("3", "3")
66                  .Case("s", "2")
67                  .Case("z", "2")
68                  .Case("g", "1")
69                  .Default("0");
70     }
71     CmdArgs.push_back(Args.MakeArgString("-O" + OOpt));
72   }
73 }
74 
75 static bool checkSystemForAMDGPU(const ArgList &Args, const AMDGPUToolChain &TC,
76                                  std::string &GPUArch) {
77   if (auto Err = TC.getSystemGPUArch(Args, GPUArch)) {
78     std::string ErrMsg =
79         llvm::formatv("{0}", llvm::fmt_consume(std::move(Err)));
80     TC.getDriver().Diag(diag::err_drv_undetermined_amdgpu_arch) << ErrMsg;
81     return false;
82   }
83 
84   return true;
85 }
86 } // namespace
87 
88 const char *AMDGCN::OpenMPLinker::constructLLVMLinkCommand(
89     const toolchains::AMDGPUOpenMPToolChain &AMDGPUOpenMPTC, Compilation &C,
90     const JobAction &JA, const InputInfoList &Inputs, const ArgList &Args,
91     StringRef SubArchName, StringRef OutputFilePrefix) const {
92   ArgStringList CmdArgs;
93 
94   for (const auto &II : Inputs)
95     if (II.isFilename())
96       CmdArgs.push_back(II.getFilename());
97 
98   if (Args.hasArg(options::OPT_l)) {
99     auto Lm = Args.getAllArgValues(options::OPT_l);
100     bool HasLibm = false;
101     for (auto &Lib : Lm) {
102       if (Lib == "m") {
103         HasLibm = true;
104         break;
105       }
106     }
107 
108     if (HasLibm) {
109       SmallVector<std::string, 12> BCLibs =
110           AMDGPUOpenMPTC.getCommonDeviceLibNames(Args, SubArchName.str());
111       llvm::for_each(BCLibs, [&](StringRef BCFile) {
112         CmdArgs.push_back(Args.MakeArgString(BCFile));
113       });
114     }
115   }
116 
117   AddStaticDeviceLibsLinking(C, *this, JA, Inputs, Args, CmdArgs, "amdgcn",
118                       SubArchName,
119                       /* bitcode SDL?*/ true,
120                       /* PostClang Link? */ false);
121   // Add an intermediate output file.
122   CmdArgs.push_back("-o");
123   const char *OutputFileName =
124       getOutputFileName(C, OutputFilePrefix, "-linked", "bc");
125   CmdArgs.push_back(OutputFileName);
126   const char *Exec =
127       Args.MakeArgString(getToolChain().GetProgramPath("llvm-link"));
128   C.addCommand(std::make_unique<Command>(
129       JA, *this, ResponseFileSupport::AtFileCurCP(), Exec, CmdArgs, Inputs,
130       InputInfo(&JA, Args.MakeArgString(OutputFileName))));
131   return OutputFileName;
132 }
133 
134 const char *AMDGCN::OpenMPLinker::constructLlcCommand(
135     Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
136     const llvm::opt::ArgList &Args, llvm::StringRef SubArchName,
137     llvm::StringRef OutputFilePrefix, const char *InputFileName,
138     bool OutputIsAsm) const {
139   // Construct llc command.
140   ArgStringList LlcArgs;
141   // The input to llc is the output from opt.
142   LlcArgs.push_back(InputFileName);
143   // Pass optimization arg to llc.
144   addLLCOptArg(Args, LlcArgs);
145   LlcArgs.push_back("-mtriple=amdgcn-amd-amdhsa");
146   LlcArgs.push_back(Args.MakeArgString("-mcpu=" + SubArchName));
147   LlcArgs.push_back(
148       Args.MakeArgString(Twine("-filetype=") + (OutputIsAsm ? "asm" : "obj")));
149 
150   for (const Arg *A : Args.filtered(options::OPT_mllvm)) {
151     LlcArgs.push_back(A->getValue(0));
152   }
153 
154   // Add output filename
155   LlcArgs.push_back("-o");
156   const char *LlcOutputFile =
157       getOutputFileName(C, OutputFilePrefix, "", OutputIsAsm ? "s" : "o");
158   LlcArgs.push_back(LlcOutputFile);
159   const char *Llc = Args.MakeArgString(getToolChain().GetProgramPath("llc"));
160   C.addCommand(std::make_unique<Command>(
161       JA, *this, ResponseFileSupport::AtFileCurCP(), Llc, LlcArgs, Inputs,
162       InputInfo(&JA, Args.MakeArgString(LlcOutputFile))));
163   return LlcOutputFile;
164 }
165 
166 void AMDGCN::OpenMPLinker::constructLldCommand(
167     Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
168     const InputInfo &Output, const llvm::opt::ArgList &Args,
169     const char *InputFileName) const {
170   // Construct lld command.
171   // The output from ld.lld is an HSA code object file.
172   ArgStringList LldArgs{"-flavor",    "gnu", "--no-undefined",
173                         "-shared",    "-o",  Output.getFilename(),
174                         InputFileName};
175 
176   const char *Lld = Args.MakeArgString(getToolChain().GetProgramPath("lld"));
177   C.addCommand(std::make_unique<Command>(
178       JA, *this, ResponseFileSupport::AtFileCurCP(), Lld, LldArgs, Inputs,
179       InputInfo(&JA, Args.MakeArgString(Output.getFilename()))));
180 }
181 
182 // For amdgcn the inputs of the linker job are device bitcode and output is
183 // object file. It calls llvm-link, opt, llc, then lld steps.
184 void AMDGCN::OpenMPLinker::ConstructJob(Compilation &C, const JobAction &JA,
185                                         const InputInfo &Output,
186                                         const InputInfoList &Inputs,
187                                         const ArgList &Args,
188                                         const char *LinkingOutput) const {
189   const ToolChain &TC = getToolChain();
190   assert(getToolChain().getTriple().isAMDGCN() && "Unsupported target");
191 
192   const toolchains::AMDGPUOpenMPToolChain &AMDGPUOpenMPTC =
193       static_cast<const toolchains::AMDGPUOpenMPToolChain &>(TC);
194 
195   std::string GPUArch = Args.getLastArgValue(options::OPT_march_EQ).str();
196   if (GPUArch.empty()) {
197     if (!checkSystemForAMDGPU(Args, AMDGPUOpenMPTC, GPUArch))
198       return;
199   }
200 
201   // Prefix for temporary file name.
202   std::string Prefix;
203   for (const auto &II : Inputs)
204     if (II.isFilename())
205       Prefix = llvm::sys::path::stem(II.getFilename()).str() + "-" + GPUArch;
206   assert(Prefix.length() && "no linker inputs are files ");
207 
208   // Each command outputs different files.
209   const char *LLVMLinkCommand = constructLLVMLinkCommand(
210       AMDGPUOpenMPTC, C, JA, Inputs, Args, GPUArch, Prefix);
211 
212   // Produce readable assembly if save-temps is enabled.
213   if (C.getDriver().isSaveTempsEnabled())
214     constructLlcCommand(C, JA, Inputs, Args, GPUArch, Prefix, LLVMLinkCommand,
215                         /*OutputIsAsm=*/true);
216   const char *LlcCommand = constructLlcCommand(C, JA, Inputs, Args, GPUArch,
217                                                Prefix, LLVMLinkCommand);
218   constructLldCommand(C, JA, Inputs, Output, Args, LlcCommand);
219 }
220 
221 AMDGPUOpenMPToolChain::AMDGPUOpenMPToolChain(const Driver &D,
222                                              const llvm::Triple &Triple,
223                                              const ToolChain &HostTC,
224                                              const ArgList &Args)
225     : ROCMToolChain(D, Triple, Args), HostTC(HostTC) {
226   // Lookup binaries into the driver directory, this is used to
227   // discover the clang-offload-bundler executable.
228   getProgramPaths().push_back(getDriver().Dir);
229 }
230 
231 void AMDGPUOpenMPToolChain::addClangTargetOptions(
232     const llvm::opt::ArgList &DriverArgs, llvm::opt::ArgStringList &CC1Args,
233     Action::OffloadKind DeviceOffloadingKind) const {
234   HostTC.addClangTargetOptions(DriverArgs, CC1Args, DeviceOffloadingKind);
235 
236   std::string GPUArch = DriverArgs.getLastArgValue(options::OPT_march_EQ).str();
237   if (GPUArch.empty()) {
238     if (!checkSystemForAMDGPU(DriverArgs, *this, GPUArch))
239       return;
240   }
241 
242   assert(DeviceOffloadingKind == Action::OFK_OpenMP &&
243          "Only OpenMP offloading kinds are supported.");
244 
245   CC1Args.push_back("-target-cpu");
246   CC1Args.push_back(DriverArgs.MakeArgStringRef(GPUArch));
247   CC1Args.push_back("-fcuda-is-device");
248 
249   if (DriverArgs.hasArg(options::OPT_nogpulib))
250     return;
251 
252   std::string BitcodeSuffix;
253   if (DriverArgs.hasFlag(options::OPT_fopenmp_target_new_runtime,
254                          options::OPT_fno_openmp_target_new_runtime, false))
255     BitcodeSuffix = "new-amdgcn-" + GPUArch;
256   else
257     BitcodeSuffix = "amdgcn-" + GPUArch;
258 
259   addOpenMPDeviceRTL(getDriver(), DriverArgs, CC1Args, BitcodeSuffix,
260                      getTriple());
261 }
262 
263 llvm::opt::DerivedArgList *AMDGPUOpenMPToolChain::TranslateArgs(
264     const llvm::opt::DerivedArgList &Args, StringRef BoundArch,
265     Action::OffloadKind DeviceOffloadKind) const {
266   DerivedArgList *DAL =
267       HostTC.TranslateArgs(Args, BoundArch, DeviceOffloadKind);
268   if (!DAL)
269     DAL = new DerivedArgList(Args.getBaseArgs());
270 
271   const OptTable &Opts = getDriver().getOpts();
272 
273   if (DeviceOffloadKind != Action::OFK_OpenMP) {
274     for (Arg *A : Args) {
275       DAL->append(A);
276     }
277   }
278 
279   if (!BoundArch.empty()) {
280     DAL->eraseArg(options::OPT_march_EQ);
281     DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_march_EQ),
282                       BoundArch);
283   }
284 
285   return DAL;
286 }
287 
288 Tool *AMDGPUOpenMPToolChain::buildLinker() const {
289   assert(getTriple().isAMDGCN());
290   return new tools::AMDGCN::OpenMPLinker(*this);
291 }
292 
293 void AMDGPUOpenMPToolChain::addClangWarningOptions(
294     ArgStringList &CC1Args) const {
295   HostTC.addClangWarningOptions(CC1Args);
296 }
297 
298 ToolChain::CXXStdlibType
299 AMDGPUOpenMPToolChain::GetCXXStdlibType(const ArgList &Args) const {
300   return HostTC.GetCXXStdlibType(Args);
301 }
302 
303 void AMDGPUOpenMPToolChain::AddClangSystemIncludeArgs(
304     const ArgList &DriverArgs, ArgStringList &CC1Args) const {
305   HostTC.AddClangSystemIncludeArgs(DriverArgs, CC1Args);
306 }
307 
308 void AMDGPUOpenMPToolChain::AddIAMCUIncludeArgs(const ArgList &Args,
309                                                 ArgStringList &CC1Args) const {
310   HostTC.AddIAMCUIncludeArgs(Args, CC1Args);
311 }
312 
313 SanitizerMask AMDGPUOpenMPToolChain::getSupportedSanitizers() const {
314   // The AMDGPUOpenMPToolChain only supports sanitizers in the sense that it
315   // allows sanitizer arguments on the command line if they are supported by the
316   // host toolchain. The AMDGPUOpenMPToolChain will actually ignore any command
317   // line arguments for any of these "supported" sanitizers. That means that no
318   // sanitization of device code is actually supported at this time.
319   //
320   // This behavior is necessary because the host and device toolchains
321   // invocations often share the command line, so the device toolchain must
322   // tolerate flags meant only for the host toolchain.
323   return HostTC.getSupportedSanitizers();
324 }
325 
326 VersionTuple
327 AMDGPUOpenMPToolChain::computeMSVCVersion(const Driver *D,
328                                           const ArgList &Args) const {
329   return HostTC.computeMSVCVersion(D, Args);
330 }
331