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     Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
90     const ArgList &Args, StringRef SubArchName,
91     StringRef OutputFilePrefix) const {
92   ArgStringList CmdArgs;
93 
94   for (const auto &II : Inputs)
95     if (II.isFilename())
96       CmdArgs.push_back(II.getFilename());
97   // Add an intermediate output file.
98   CmdArgs.push_back("-o");
99   const char *OutputFileName =
100       getOutputFileName(C, OutputFilePrefix, "-linked", "bc");
101   CmdArgs.push_back(OutputFileName);
102   const char *Exec =
103       Args.MakeArgString(getToolChain().GetProgramPath("llvm-link"));
104   C.addCommand(std::make_unique<Command>(
105       JA, *this, ResponseFileSupport::AtFileCurCP(), Exec, CmdArgs, Inputs,
106       InputInfo(&JA, Args.MakeArgString(OutputFileName))));
107   return OutputFileName;
108 }
109 
110 const char *AMDGCN::OpenMPLinker::constructLlcCommand(
111     Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
112     const llvm::opt::ArgList &Args, llvm::StringRef SubArchName,
113     llvm::StringRef OutputFilePrefix, const char *InputFileName,
114     bool OutputIsAsm) const {
115   // Construct llc command.
116   ArgStringList LlcArgs;
117   // The input to llc is the output from opt.
118   LlcArgs.push_back(InputFileName);
119   // Pass optimization arg to llc.
120   addLLCOptArg(Args, LlcArgs);
121   LlcArgs.push_back("-mtriple=amdgcn-amd-amdhsa");
122   LlcArgs.push_back(Args.MakeArgString("-mcpu=" + SubArchName));
123   LlcArgs.push_back(
124       Args.MakeArgString(Twine("-filetype=") + (OutputIsAsm ? "asm" : "obj")));
125 
126   for (const Arg *A : Args.filtered(options::OPT_mllvm)) {
127     LlcArgs.push_back(A->getValue(0));
128   }
129 
130   // Add output filename
131   LlcArgs.push_back("-o");
132   const char *LlcOutputFile =
133       getOutputFileName(C, OutputFilePrefix, "", OutputIsAsm ? "s" : "o");
134   LlcArgs.push_back(LlcOutputFile);
135   const char *Llc = Args.MakeArgString(getToolChain().GetProgramPath("llc"));
136   C.addCommand(std::make_unique<Command>(
137       JA, *this, ResponseFileSupport::AtFileCurCP(), Llc, LlcArgs, Inputs,
138       InputInfo(&JA, Args.MakeArgString(LlcOutputFile))));
139   return LlcOutputFile;
140 }
141 
142 void AMDGCN::OpenMPLinker::constructLldCommand(
143     Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
144     const InputInfo &Output, const llvm::opt::ArgList &Args,
145     const char *InputFileName) const {
146   // Construct lld command.
147   // The output from ld.lld is an HSA code object file.
148   ArgStringList LldArgs{"-flavor",    "gnu", "--no-undefined",
149                         "-shared",    "-o",  Output.getFilename(),
150                         InputFileName};
151 
152   const char *Lld = Args.MakeArgString(getToolChain().GetProgramPath("lld"));
153   C.addCommand(std::make_unique<Command>(
154       JA, *this, ResponseFileSupport::AtFileCurCP(), Lld, LldArgs, Inputs,
155       InputInfo(&JA, Args.MakeArgString(Output.getFilename()))));
156 }
157 
158 // For amdgcn the inputs of the linker job are device bitcode and output is
159 // object file. It calls llvm-link, opt, llc, then lld steps.
160 void AMDGCN::OpenMPLinker::ConstructJob(Compilation &C, const JobAction &JA,
161                                         const InputInfo &Output,
162                                         const InputInfoList &Inputs,
163                                         const ArgList &Args,
164                                         const char *LinkingOutput) const {
165   const ToolChain &TC = getToolChain();
166   assert(getToolChain().getTriple().isAMDGCN() && "Unsupported target");
167 
168   const toolchains::AMDGPUOpenMPToolChain &AMDGPUOpenMPTC =
169       static_cast<const toolchains::AMDGPUOpenMPToolChain &>(TC);
170 
171   std::string GPUArch = Args.getLastArgValue(options::OPT_march_EQ).str();
172   if (GPUArch.empty()) {
173     if (!checkSystemForAMDGPU(Args, AMDGPUOpenMPTC, GPUArch))
174       return;
175   }
176 
177   // Prefix for temporary file name.
178   std::string Prefix;
179   for (const auto &II : Inputs)
180     if (II.isFilename())
181       Prefix = llvm::sys::path::stem(II.getFilename()).str() + "-" + GPUArch;
182   assert(Prefix.length() && "no linker inputs are files ");
183 
184   // Each command outputs different files.
185   const char *LLVMLinkCommand =
186       constructLLVMLinkCommand(C, JA, Inputs, Args, GPUArch, Prefix);
187 
188   // Produce readable assembly if save-temps is enabled.
189   if (C.getDriver().isSaveTempsEnabled())
190     constructLlcCommand(C, JA, Inputs, Args, GPUArch, Prefix, LLVMLinkCommand,
191                         /*OutputIsAsm=*/true);
192   const char *LlcCommand = constructLlcCommand(C, JA, Inputs, Args, GPUArch,
193                                                Prefix, LLVMLinkCommand);
194   constructLldCommand(C, JA, Inputs, Output, Args, LlcCommand);
195 }
196 
197 AMDGPUOpenMPToolChain::AMDGPUOpenMPToolChain(const Driver &D,
198                                              const llvm::Triple &Triple,
199                                              const ToolChain &HostTC,
200                                              const ArgList &Args)
201     : ROCMToolChain(D, Triple, Args), HostTC(HostTC) {
202   // Lookup binaries into the driver directory, this is used to
203   // discover the clang-offload-bundler executable.
204   getProgramPaths().push_back(getDriver().Dir);
205 }
206 
207 void AMDGPUOpenMPToolChain::addClangTargetOptions(
208     const llvm::opt::ArgList &DriverArgs, llvm::opt::ArgStringList &CC1Args,
209     Action::OffloadKind DeviceOffloadingKind) const {
210   HostTC.addClangTargetOptions(DriverArgs, CC1Args, DeviceOffloadingKind);
211 
212   std::string GPUArch = DriverArgs.getLastArgValue(options::OPT_march_EQ).str();
213   if (GPUArch.empty()) {
214     if (!checkSystemForAMDGPU(DriverArgs, *this, GPUArch))
215       return;
216   }
217 
218   assert(DeviceOffloadingKind == Action::OFK_OpenMP &&
219          "Only OpenMP offloading kinds are supported.");
220 
221   CC1Args.push_back("-target-cpu");
222   CC1Args.push_back(DriverArgs.MakeArgStringRef(GPUArch));
223   CC1Args.push_back("-fcuda-is-device");
224 
225   if (DriverArgs.hasArg(options::OPT_nogpulib))
226     return;
227 
228   std::string BitcodeSuffix;
229   if (DriverArgs.hasFlag(options::OPT_fopenmp_target_new_runtime,
230                          options::OPT_fno_openmp_target_new_runtime, false))
231     BitcodeSuffix = "new-amdgcn-" + GPUArch;
232   else
233     BitcodeSuffix = "amdgcn-" + GPUArch;
234 
235   addOpenMPDeviceRTL(getDriver(), DriverArgs, CC1Args, BitcodeSuffix,
236                      getTriple());
237 
238   if (!DriverArgs.hasArg(options::OPT_l))
239     return;
240 
241   auto Lm = DriverArgs.getAllArgValues(options::OPT_l);
242   bool HasLibm = false;
243   for (auto &Lib : Lm) {
244     if (Lib == "m") {
245       HasLibm = true;
246       break;
247     }
248   }
249 
250   if (HasLibm) {
251     SmallVector<std::string, 12> BCLibs =
252         getCommonDeviceLibNames(DriverArgs, GPUArch);
253     llvm::for_each(BCLibs, [&](StringRef BCFile) {
254       CC1Args.push_back("-mlink-builtin-bitcode");
255       CC1Args.push_back(DriverArgs.MakeArgString(BCFile));
256     });
257   }
258 }
259 
260 llvm::opt::DerivedArgList *AMDGPUOpenMPToolChain::TranslateArgs(
261     const llvm::opt::DerivedArgList &Args, StringRef BoundArch,
262     Action::OffloadKind DeviceOffloadKind) const {
263   DerivedArgList *DAL =
264       HostTC.TranslateArgs(Args, BoundArch, DeviceOffloadKind);
265   if (!DAL)
266     DAL = new DerivedArgList(Args.getBaseArgs());
267 
268   const OptTable &Opts = getDriver().getOpts();
269 
270   if (DeviceOffloadKind != Action::OFK_OpenMP) {
271     for (Arg *A : Args) {
272       DAL->append(A);
273     }
274   }
275 
276   if (!BoundArch.empty()) {
277     DAL->eraseArg(options::OPT_march_EQ);
278     DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_march_EQ),
279                       BoundArch);
280   }
281 
282   return DAL;
283 }
284 
285 Tool *AMDGPUOpenMPToolChain::buildLinker() const {
286   assert(getTriple().isAMDGCN());
287   return new tools::AMDGCN::OpenMPLinker(*this);
288 }
289 
290 void AMDGPUOpenMPToolChain::addClangWarningOptions(
291     ArgStringList &CC1Args) const {
292   HostTC.addClangWarningOptions(CC1Args);
293 }
294 
295 ToolChain::CXXStdlibType
296 AMDGPUOpenMPToolChain::GetCXXStdlibType(const ArgList &Args) const {
297   return HostTC.GetCXXStdlibType(Args);
298 }
299 
300 void AMDGPUOpenMPToolChain::AddClangSystemIncludeArgs(
301     const ArgList &DriverArgs, ArgStringList &CC1Args) const {
302   HostTC.AddClangSystemIncludeArgs(DriverArgs, CC1Args);
303 }
304 
305 void AMDGPUOpenMPToolChain::AddIAMCUIncludeArgs(const ArgList &Args,
306                                                 ArgStringList &CC1Args) const {
307   HostTC.AddIAMCUIncludeArgs(Args, CC1Args);
308 }
309 
310 SanitizerMask AMDGPUOpenMPToolChain::getSupportedSanitizers() const {
311   // The AMDGPUOpenMPToolChain only supports sanitizers in the sense that it
312   // allows sanitizer arguments on the command line if they are supported by the
313   // host toolchain. The AMDGPUOpenMPToolChain will actually ignore any command
314   // line arguments for any of these "supported" sanitizers. That means that no
315   // sanitization of device code is actually supported at this time.
316   //
317   // This behavior is necessary because the host and device toolchains
318   // invocations often share the command line, so the device toolchain must
319   // tolerate flags meant only for the host toolchain.
320   return HostTC.getSupportedSanitizers();
321 }
322 
323 VersionTuple
324 AMDGPUOpenMPToolChain::computeMSVCVersion(const Driver *D,
325                                           const ArgList &Args) const {
326   return HostTC.computeMSVCVersion(D, Args);
327 }
328