1 //===--- HIPUtility.cpp - Common HIP Tool Chain Utilities -------*- 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 "HIPUtility.h"
10 #include "clang/Driver/CommonArgs.h"
11 #include "clang/Driver/Compilation.h"
12 #include "clang/Driver/Options.h"
13 #include "llvm/ADT/StringExtras.h"
14 #include "llvm/ADT/StringRef.h"
15 #include "llvm/Object/Archive.h"
16 #include "llvm/Object/ObjectFile.h"
17 #include "llvm/Support/MD5.h"
18 #include "llvm/Support/MemoryBuffer.h"
19 #include "llvm/Support/Path.h"
20 #include "llvm/Support/raw_ostream.h"
21 #include "llvm/TargetParser/Triple.h"
22 #include <deque>
23 #include <set>
24
25 using namespace clang;
26 using namespace clang::driver;
27 using namespace clang::driver::tools;
28 using namespace llvm::opt;
29 using llvm::dyn_cast;
30
31 #if defined(_WIN32) || defined(_WIN64)
32 #define NULL_FILE "nul"
33 #else
34 #define NULL_FILE "/dev/null"
35 #endif
36
37 namespace {
38 const unsigned HIPCodeObjectAlign = 4096;
39 } // namespace
40
41 // Constructs a triple string for clang offload bundler.
normalizeForBundler(const llvm::Triple & T,bool HasTargetID)42 static std::string normalizeForBundler(const llvm::Triple &T,
43 bool HasTargetID) {
44 return HasTargetID ? (T.getArchName() + "-" + T.getVendorName() + "-" +
45 T.getOSName() + "-" + T.getEnvironmentName())
46 .str()
47 : T.normalize(llvm::Triple::CanonicalForm::FOUR_IDENT);
48 }
49
50 // Collect undefined __hip_fatbin* and __hip_gpubin_handle* symbols from all
51 // input object or archive files.
52 class HIPUndefinedFatBinSymbols {
53 public:
HIPUndefinedFatBinSymbols(const Compilation & C,const llvm::opt::ArgList & Args_)54 HIPUndefinedFatBinSymbols(const Compilation &C,
55 const llvm::opt::ArgList &Args_)
56 : C(C), Args(Args_),
57 DiagID(C.getDriver().getDiags().getCustomDiagID(
58 DiagnosticsEngine::Error,
59 "Error collecting HIP undefined fatbin symbols: %0")),
60 Quiet(C.getArgs().hasArg(options::OPT__HASH_HASH_HASH)),
61 Verbose(C.getArgs().hasArg(options::OPT_v)) {
62 populateSymbols();
63 processStaticLibraries();
64 if (Verbose) {
65 for (const auto &Name : FatBinSymbols)
66 llvm::errs() << "Found undefined HIP fatbin symbol: " << Name << "\n";
67 for (const auto &Name : GPUBinHandleSymbols)
68 llvm::errs() << "Found undefined HIP gpubin handle symbol: " << Name
69 << "\n";
70 }
71 }
72
getFatBinSymbols() const73 const std::set<std::string> &getFatBinSymbols() const {
74 return FatBinSymbols;
75 }
76
getGPUBinHandleSymbols() const77 const std::set<std::string> &getGPUBinHandleSymbols() const {
78 return GPUBinHandleSymbols;
79 }
80
81 // Collect symbols from static libraries specified by -l options.
processStaticLibraries()82 void processStaticLibraries() {
83 llvm::SmallVector<llvm::StringRef, 16> LibNames;
84 llvm::SmallVector<llvm::StringRef, 16> LibPaths;
85 llvm::SmallVector<llvm::StringRef, 16> ExactLibNames;
86 llvm::Triple Triple(C.getDriver().getTargetTriple());
87 bool IsMSVC = Triple.isWindowsMSVCEnvironment();
88 llvm::StringRef Ext = IsMSVC ? ".lib" : ".a";
89
90 for (const auto *Arg : Args.filtered(options::OPT_l)) {
91 llvm::StringRef Value = Arg->getValue();
92 if (Value.starts_with(":"))
93 ExactLibNames.push_back(Value.drop_front());
94 else
95 LibNames.push_back(Value);
96 }
97 for (const auto *Arg : Args.filtered(options::OPT_L)) {
98 auto Path = Arg->getValue();
99 LibPaths.push_back(Path);
100 if (Verbose)
101 llvm::errs() << "HIP fatbin symbol search uses library path: " << Path
102 << "\n";
103 }
104
105 auto ProcessLib = [&](llvm::StringRef LibName, bool IsExact) {
106 llvm::SmallString<256> FullLibName(
107 IsExact ? Twine(LibName).str()
108 : IsMSVC ? (Twine(LibName) + Ext).str()
109 : (Twine("lib") + LibName + Ext).str());
110
111 bool Found = false;
112 for (const auto Path : LibPaths) {
113 llvm::SmallString<256> FullPath = Path;
114 llvm::sys::path::append(FullPath, FullLibName);
115
116 if (llvm::sys::fs::exists(FullPath)) {
117 if (Verbose)
118 llvm::errs() << "HIP fatbin symbol search found library: "
119 << FullPath << "\n";
120 auto BufferOrErr = llvm::MemoryBuffer::getFile(FullPath);
121 if (!BufferOrErr) {
122 errorHandler(llvm::errorCodeToError(BufferOrErr.getError()));
123 continue;
124 }
125 processInput(BufferOrErr.get()->getMemBufferRef());
126 Found = true;
127 break;
128 }
129 }
130 if (!Found && Verbose)
131 llvm::errs() << "HIP fatbin symbol search could not find library: "
132 << FullLibName << "\n";
133 };
134
135 for (const auto LibName : ExactLibNames)
136 ProcessLib(LibName, true);
137
138 for (const auto LibName : LibNames)
139 ProcessLib(LibName, false);
140 }
141
142 private:
143 const Compilation &C;
144 const llvm::opt::ArgList &Args;
145 unsigned DiagID;
146 bool Quiet;
147 bool Verbose;
148 std::set<std::string> FatBinSymbols;
149 std::set<std::string> GPUBinHandleSymbols;
150 std::set<std::string, std::less<>> DefinedFatBinSymbols;
151 std::set<std::string, std::less<>> DefinedGPUBinHandleSymbols;
152 const std::string FatBinPrefix = "__hip_fatbin";
153 const std::string GPUBinHandlePrefix = "__hip_gpubin_handle";
154
populateSymbols()155 void populateSymbols() {
156 std::deque<const Action *> WorkList;
157 std::set<const Action *> Visited;
158
159 for (const auto &Action : C.getActions())
160 WorkList.push_back(Action);
161
162 while (!WorkList.empty()) {
163 const Action *CurrentAction = WorkList.front();
164 WorkList.pop_front();
165
166 if (!CurrentAction || !Visited.insert(CurrentAction).second)
167 continue;
168
169 if (const auto *IA = dyn_cast<InputAction>(CurrentAction)) {
170 std::string ID = IA->getId().str();
171 if (!ID.empty()) {
172 ID = llvm::utohexstr(llvm::MD5Hash(ID), /*LowerCase=*/true);
173 FatBinSymbols.insert((FatBinPrefix + Twine('_') + ID).str());
174 GPUBinHandleSymbols.insert(
175 (GPUBinHandlePrefix + Twine('_') + ID).str());
176 continue;
177 }
178 if (IA->getInputArg().getNumValues() == 0)
179 continue;
180 const char *Filename = IA->getInputArg().getValue();
181 if (!Filename)
182 continue;
183 auto BufferOrErr = llvm::MemoryBuffer::getFile(Filename);
184 // Input action could be options to linker, therefore, ignore it
185 // if cannot read it. If it turns out to be a file that cannot be read,
186 // the error will be caught by the linker.
187 if (!BufferOrErr)
188 continue;
189
190 processInput(BufferOrErr.get()->getMemBufferRef());
191 } else
192 llvm::append_range(WorkList, CurrentAction->getInputs());
193 }
194 }
195
processInput(const llvm::MemoryBufferRef & Buffer)196 void processInput(const llvm::MemoryBufferRef &Buffer) {
197 // Try processing as object file first.
198 auto ObjFileOrErr = llvm::object::ObjectFile::createObjectFile(Buffer);
199 if (ObjFileOrErr) {
200 processSymbols(**ObjFileOrErr);
201 return;
202 }
203
204 // Then try processing as archive files.
205 llvm::consumeError(ObjFileOrErr.takeError());
206 auto ArchiveOrErr = llvm::object::Archive::create(Buffer);
207 if (ArchiveOrErr) {
208 llvm::Error Err = llvm::Error::success();
209 llvm::object::Archive &Archive = *ArchiveOrErr.get();
210 for (auto &Child : Archive.children(Err)) {
211 auto ChildBufOrErr = Child.getMemoryBufferRef();
212 if (ChildBufOrErr)
213 processInput(*ChildBufOrErr);
214 else
215 errorHandler(ChildBufOrErr.takeError());
216 }
217
218 if (Err)
219 errorHandler(std::move(Err));
220 return;
221 }
222
223 // Ignore other files.
224 llvm::consumeError(ArchiveOrErr.takeError());
225 }
226
processSymbols(const llvm::object::ObjectFile & Obj)227 void processSymbols(const llvm::object::ObjectFile &Obj) {
228 for (const auto &Symbol : Obj.symbols()) {
229 auto FlagOrErr = Symbol.getFlags();
230 if (!FlagOrErr) {
231 errorHandler(FlagOrErr.takeError());
232 continue;
233 }
234
235 auto NameOrErr = Symbol.getName();
236 if (!NameOrErr) {
237 errorHandler(NameOrErr.takeError());
238 continue;
239 }
240 llvm::StringRef Name = *NameOrErr;
241
242 bool isUndefined =
243 FlagOrErr.get() & llvm::object::SymbolRef::SF_Undefined;
244 bool isFatBinSymbol = Name.starts_with(FatBinPrefix);
245 bool isGPUBinHandleSymbol = Name.starts_with(GPUBinHandlePrefix);
246
247 // Handling for defined symbols
248 if (!isUndefined) {
249 if (isFatBinSymbol) {
250 DefinedFatBinSymbols.insert(Name.str());
251 FatBinSymbols.erase(Name.str());
252 } else if (isGPUBinHandleSymbol) {
253 DefinedGPUBinHandleSymbols.insert(Name.str());
254 GPUBinHandleSymbols.erase(Name.str());
255 }
256 continue;
257 }
258
259 // Add undefined symbols if they are not in the defined sets
260 if (isFatBinSymbol &&
261 DefinedFatBinSymbols.find(Name) == DefinedFatBinSymbols.end())
262 FatBinSymbols.insert(Name.str());
263 else if (isGPUBinHandleSymbol && DefinedGPUBinHandleSymbols.find(Name) ==
264 DefinedGPUBinHandleSymbols.end())
265 GPUBinHandleSymbols.insert(Name.str());
266 }
267 }
268
errorHandler(llvm::Error Err)269 void errorHandler(llvm::Error Err) {
270 if (Quiet)
271 return;
272 C.getDriver().Diag(DiagID) << llvm::toString(std::move(Err));
273 }
274 };
275
276 // Construct a clang-offload-bundler command to bundle code objects for
277 // different devices into a HIP fat binary.
constructHIPFatbinCommand(Compilation & C,const JobAction & JA,llvm::StringRef OutputFileName,const InputInfoList & Inputs,const llvm::opt::ArgList & Args,const Tool & T)278 void HIP::constructHIPFatbinCommand(Compilation &C, const JobAction &JA,
279 llvm::StringRef OutputFileName,
280 const InputInfoList &Inputs,
281 const llvm::opt::ArgList &Args,
282 const Tool &T) {
283 // Construct clang-offload-bundler command to bundle object files for
284 // for different GPU archs.
285 ArgStringList BundlerArgs;
286 BundlerArgs.push_back(Args.MakeArgString("-type=o"));
287 BundlerArgs.push_back(
288 Args.MakeArgString("-bundle-align=" + Twine(HIPCodeObjectAlign)));
289
290 // ToDo: Remove the dummy host binary entry which is required by
291 // clang-offload-bundler.
292 std::string BundlerTargetArg = "-targets=host-x86_64-unknown-linux-gnu";
293 // AMDGCN:
294 // For code object version 2 and 3, the offload kind in bundle ID is 'hip'
295 // for backward compatibility. For code object version 4 and greater, the
296 // offload kind in bundle ID is 'hipv4'.
297 std::string OffloadKind = "hip";
298 auto &TT = T.getToolChain().getTriple();
299 if (TT.isAMDGCN() && getAMDGPUCodeObjectVersion(C.getDriver(), Args) >= 4)
300 OffloadKind = OffloadKind + "v4";
301 for (const auto &II : Inputs) {
302 const auto *A = II.getAction();
303 auto ArchStr = llvm::StringRef(A->getOffloadingArch());
304 BundlerTargetArg += ',' + OffloadKind + '-';
305 if (ArchStr == "amdgcnspirv")
306 BundlerTargetArg +=
307 normalizeForBundler(llvm::Triple("spirv64-amd-amdhsa"), true);
308 else
309 BundlerTargetArg += normalizeForBundler(TT, !ArchStr.empty());
310 if (!ArchStr.empty())
311 BundlerTargetArg += '-' + ArchStr.str();
312 }
313 BundlerArgs.push_back(Args.MakeArgString(BundlerTargetArg));
314
315 // Use a NULL file as input for the dummy host binary entry
316 std::string BundlerInputArg = "-input=" NULL_FILE;
317 BundlerArgs.push_back(Args.MakeArgString(BundlerInputArg));
318 for (const auto &II : Inputs) {
319 BundlerInputArg = std::string("-input=") + II.getFilename();
320 BundlerArgs.push_back(Args.MakeArgString(BundlerInputArg));
321 }
322
323 std::string Output = std::string(OutputFileName);
324 auto *BundlerOutputArg =
325 Args.MakeArgString(std::string("-output=").append(Output));
326 BundlerArgs.push_back(BundlerOutputArg);
327
328 addOffloadCompressArgs(Args, BundlerArgs);
329
330 const char *Bundler = Args.MakeArgString(
331 T.getToolChain().GetProgramPath("clang-offload-bundler"));
332 C.addCommand(std::make_unique<Command>(
333 JA, T, ResponseFileSupport::None(), Bundler, BundlerArgs, Inputs,
334 InputInfo(&JA, Args.MakeArgString(Output))));
335 }
336
337 /// Add Generated HIP Object File which has device images embedded into the
338 /// host to the argument list for linking. Using MC directives, embed the
339 /// device code and also define symbols required by the code generation so that
340 /// the image can be retrieved at runtime.
constructGenerateObjFileFromHIPFatBinary(Compilation & C,const InputInfo & Output,const InputInfoList & Inputs,const ArgList & Args,const JobAction & JA,const Tool & T)341 void HIP::constructGenerateObjFileFromHIPFatBinary(
342 Compilation &C, const InputInfo &Output, const InputInfoList &Inputs,
343 const ArgList &Args, const JobAction &JA, const Tool &T) {
344 const Driver &D = C.getDriver();
345 std::string Name = std::string(llvm::sys::path::stem(Output.getFilename()));
346
347 // Create Temp Object File Generator,
348 // Offload Bundled file and Bundled Object file.
349 // Keep them if save-temps is enabled.
350 const char *ObjinFile;
351 const char *BundleFile;
352 if (D.isSaveTempsEnabled()) {
353 ObjinFile = C.getArgs().MakeArgString(Name + ".mcin");
354 BundleFile = C.getArgs().MakeArgString(Name + ".hipfb");
355 } else {
356 auto TmpNameMcin = D.GetTemporaryPath(Name, "mcin");
357 ObjinFile = C.addTempFile(C.getArgs().MakeArgString(TmpNameMcin));
358 auto TmpNameFb = D.GetTemporaryPath(Name, "hipfb");
359 BundleFile = C.addTempFile(C.getArgs().MakeArgString(TmpNameFb));
360 }
361 HIP::constructHIPFatbinCommand(C, JA, BundleFile, Inputs, Args, T);
362
363 // Create a buffer to write the contents of the temp obj generator.
364 std::string ObjBuffer;
365 llvm::raw_string_ostream ObjStream(ObjBuffer);
366
367 auto HostTriple =
368 C.getSingleOffloadToolChain<Action::OFK_Host>()->getTriple();
369
370 HIPUndefinedFatBinSymbols Symbols(C, Args);
371
372 std::string PrimaryHipFatbinSymbol;
373 std::string PrimaryGpuBinHandleSymbol;
374 bool FoundPrimaryHipFatbinSymbol = false;
375 bool FoundPrimaryGpuBinHandleSymbol = false;
376
377 std::vector<std::string> AliasHipFatbinSymbols;
378 std::vector<std::string> AliasGpuBinHandleSymbols;
379
380 // Iterate through symbols to find the primary ones and collect others for
381 // aliasing
382 for (const auto &Symbol : Symbols.getFatBinSymbols()) {
383 if (!FoundPrimaryHipFatbinSymbol) {
384 PrimaryHipFatbinSymbol = Symbol;
385 FoundPrimaryHipFatbinSymbol = true;
386 } else
387 AliasHipFatbinSymbols.push_back(Symbol);
388 }
389
390 for (const auto &Symbol : Symbols.getGPUBinHandleSymbols()) {
391 if (!FoundPrimaryGpuBinHandleSymbol) {
392 PrimaryGpuBinHandleSymbol = Symbol;
393 FoundPrimaryGpuBinHandleSymbol = true;
394 } else
395 AliasGpuBinHandleSymbols.push_back(Symbol);
396 }
397
398 // Add MC directives to embed target binaries. We ensure that each
399 // section and image is 16-byte aligned. This is not mandatory, but
400 // increases the likelihood of data to be aligned with a cache block
401 // in several main host machines.
402 ObjStream << "# HIP Object Generator\n";
403 ObjStream << "# *** Automatically generated by Clang ***\n";
404 if (FoundPrimaryGpuBinHandleSymbol) {
405 // Define the first gpubin handle symbol
406 if (HostTriple.isWindowsMSVCEnvironment())
407 ObjStream << " .section .hip_gpubin_handle,\"dw\"\n";
408 else {
409 ObjStream << " .protected " << PrimaryGpuBinHandleSymbol << "\n";
410 ObjStream << " .type " << PrimaryGpuBinHandleSymbol << ",@object\n";
411 ObjStream << " .section .hip_gpubin_handle,\"aw\"\n";
412 }
413 ObjStream << " .globl " << PrimaryGpuBinHandleSymbol << "\n";
414 ObjStream << " .p2align 3\n"; // Align 8
415 ObjStream << PrimaryGpuBinHandleSymbol << ":\n";
416 ObjStream << " .zero 8\n"; // Size 8
417
418 // Generate alias directives for other gpubin handle symbols
419 for (const auto &AliasSymbol : AliasGpuBinHandleSymbols) {
420 ObjStream << " .globl " << AliasSymbol << "\n";
421 ObjStream << " .set " << AliasSymbol << "," << PrimaryGpuBinHandleSymbol
422 << "\n";
423 }
424 }
425 if (FoundPrimaryHipFatbinSymbol) {
426 // Define the first fatbin symbol
427 if (HostTriple.isWindowsMSVCEnvironment())
428 ObjStream << " .section .hip_fatbin,\"dw\"\n";
429 else {
430 ObjStream << " .protected " << PrimaryHipFatbinSymbol << "\n";
431 ObjStream << " .type " << PrimaryHipFatbinSymbol << ",@object\n";
432 ObjStream << " .section .hip_fatbin,\"a\",@progbits\n";
433 }
434 ObjStream << " .globl " << PrimaryHipFatbinSymbol << "\n";
435 ObjStream << " .p2align " << llvm::Log2(llvm::Align(HIPCodeObjectAlign))
436 << "\n";
437 // Generate alias directives for other fatbin symbols
438 for (const auto &AliasSymbol : AliasHipFatbinSymbols) {
439 ObjStream << " .globl " << AliasSymbol << "\n";
440 ObjStream << " .set " << AliasSymbol << "," << PrimaryHipFatbinSymbol
441 << "\n";
442 }
443 ObjStream << PrimaryHipFatbinSymbol << ":\n";
444 ObjStream << " .incbin ";
445 llvm::sys::printArg(ObjStream, BundleFile, /*Quote=*/true);
446 ObjStream << "\n";
447 }
448 if (HostTriple.isOSLinux() && HostTriple.isOSBinFormatELF())
449 ObjStream << " .section .note.GNU-stack, \"\", @progbits\n";
450
451 // Dump the contents of the temp object file gen if the user requested that.
452 // We support this option to enable testing of behavior with -###.
453 if (C.getArgs().hasArg(options::OPT_fhip_dump_offload_linker_script))
454 llvm::errs() << ObjBuffer;
455
456 // Open script file and write the contents.
457 std::error_code EC;
458 llvm::raw_fd_ostream Objf(ObjinFile, EC, llvm::sys::fs::OF_None);
459
460 if (EC) {
461 D.Diag(clang::diag::err_unable_to_make_temp) << EC.message();
462 return;
463 }
464
465 Objf << ObjBuffer;
466
467 ArgStringList ClangArgs{"-target", Args.MakeArgString(HostTriple.normalize()),
468 "-o", Output.getFilename(),
469 "-x", "assembler",
470 ObjinFile, "-c"};
471 C.addCommand(std::make_unique<Command>(JA, T, ResponseFileSupport::None(),
472 D.getClangProgramPath(), ClangArgs,
473 Inputs, Output, D.getPrependArg()));
474 }
475