10eae32dcSDimitry Andric //===--- HIPUtility.cpp - Common HIP Tool Chain Utilities -------*- C++ -*-===//
20eae32dcSDimitry Andric //
30eae32dcSDimitry Andric // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
40eae32dcSDimitry Andric // See https://llvm.org/LICENSE.txt for license information.
50eae32dcSDimitry Andric // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
60eae32dcSDimitry Andric //
70eae32dcSDimitry Andric //===----------------------------------------------------------------------===//
80eae32dcSDimitry Andric
90eae32dcSDimitry Andric #include "HIPUtility.h"
10*0fca6ea1SDimitry Andric #include "Clang.h"
110eae32dcSDimitry Andric #include "CommonArgs.h"
120eae32dcSDimitry Andric #include "clang/Driver/Compilation.h"
13*0fca6ea1SDimitry Andric #include "clang/Driver/Options.h"
14*0fca6ea1SDimitry Andric #include "llvm/ADT/StringExtras.h"
150eae32dcSDimitry Andric #include "llvm/ADT/StringRef.h"
16*0fca6ea1SDimitry Andric #include "llvm/Object/Archive.h"
17*0fca6ea1SDimitry Andric #include "llvm/Object/ObjectFile.h"
18*0fca6ea1SDimitry Andric #include "llvm/Support/MD5.h"
19*0fca6ea1SDimitry Andric #include "llvm/Support/MemoryBuffer.h"
200eae32dcSDimitry Andric #include "llvm/Support/Path.h"
21*0fca6ea1SDimitry Andric #include "llvm/Support/raw_ostream.h"
2206c3fb27SDimitry Andric #include "llvm/TargetParser/Triple.h"
23*0fca6ea1SDimitry Andric #include <deque>
24*0fca6ea1SDimitry Andric #include <set>
250eae32dcSDimitry Andric
26*0fca6ea1SDimitry Andric using namespace clang;
270eae32dcSDimitry Andric using namespace clang::driver;
280eae32dcSDimitry Andric using namespace clang::driver::tools;
290eae32dcSDimitry Andric using namespace llvm::opt;
30*0fca6ea1SDimitry Andric using llvm::dyn_cast;
310eae32dcSDimitry Andric
320eae32dcSDimitry Andric #if defined(_WIN32) || defined(_WIN64)
330eae32dcSDimitry Andric #define NULL_FILE "nul"
340eae32dcSDimitry Andric #else
350eae32dcSDimitry Andric #define NULL_FILE "/dev/null"
360eae32dcSDimitry Andric #endif
370eae32dcSDimitry Andric
380eae32dcSDimitry Andric namespace {
390eae32dcSDimitry Andric const unsigned HIPCodeObjectAlign = 4096;
400eae32dcSDimitry Andric } // namespace
410eae32dcSDimitry Andric
420eae32dcSDimitry Andric // Constructs a triple string for clang offload bundler.
normalizeForBundler(const llvm::Triple & T,bool HasTargetID)430eae32dcSDimitry Andric static std::string normalizeForBundler(const llvm::Triple &T,
440eae32dcSDimitry Andric bool HasTargetID) {
450eae32dcSDimitry Andric return HasTargetID ? (T.getArchName() + "-" + T.getVendorName() + "-" +
460eae32dcSDimitry Andric T.getOSName() + "-" + T.getEnvironmentName())
470eae32dcSDimitry Andric .str()
480eae32dcSDimitry Andric : T.normalize();
490eae32dcSDimitry Andric }
500eae32dcSDimitry Andric
51*0fca6ea1SDimitry Andric // Collect undefined __hip_fatbin* and __hip_gpubin_handle* symbols from all
52*0fca6ea1SDimitry Andric // input object or archive files.
53*0fca6ea1SDimitry Andric class HIPUndefinedFatBinSymbols {
54*0fca6ea1SDimitry Andric public:
HIPUndefinedFatBinSymbols(const Compilation & C)55*0fca6ea1SDimitry Andric HIPUndefinedFatBinSymbols(const Compilation &C)
56*0fca6ea1SDimitry Andric : C(C), DiagID(C.getDriver().getDiags().getCustomDiagID(
57*0fca6ea1SDimitry Andric DiagnosticsEngine::Error,
58*0fca6ea1SDimitry Andric "Error collecting HIP undefined fatbin symbols: %0")),
59*0fca6ea1SDimitry Andric Quiet(C.getArgs().hasArg(options::OPT__HASH_HASH_HASH)),
60*0fca6ea1SDimitry Andric Verbose(C.getArgs().hasArg(options::OPT_v)) {
61*0fca6ea1SDimitry Andric populateSymbols();
62*0fca6ea1SDimitry Andric if (Verbose) {
63*0fca6ea1SDimitry Andric for (const auto &Name : FatBinSymbols)
64*0fca6ea1SDimitry Andric llvm::errs() << "Found undefined HIP fatbin symbol: " << Name << "\n";
65*0fca6ea1SDimitry Andric for (const auto &Name : GPUBinHandleSymbols)
66*0fca6ea1SDimitry Andric llvm::errs() << "Found undefined HIP gpubin handle symbol: " << Name
67*0fca6ea1SDimitry Andric << "\n";
68*0fca6ea1SDimitry Andric }
69*0fca6ea1SDimitry Andric }
70*0fca6ea1SDimitry Andric
getFatBinSymbols() const71*0fca6ea1SDimitry Andric const std::set<std::string> &getFatBinSymbols() const {
72*0fca6ea1SDimitry Andric return FatBinSymbols;
73*0fca6ea1SDimitry Andric }
74*0fca6ea1SDimitry Andric
getGPUBinHandleSymbols() const75*0fca6ea1SDimitry Andric const std::set<std::string> &getGPUBinHandleSymbols() const {
76*0fca6ea1SDimitry Andric return GPUBinHandleSymbols;
77*0fca6ea1SDimitry Andric }
78*0fca6ea1SDimitry Andric
79*0fca6ea1SDimitry Andric private:
80*0fca6ea1SDimitry Andric const Compilation &C;
81*0fca6ea1SDimitry Andric unsigned DiagID;
82*0fca6ea1SDimitry Andric bool Quiet;
83*0fca6ea1SDimitry Andric bool Verbose;
84*0fca6ea1SDimitry Andric std::set<std::string> FatBinSymbols;
85*0fca6ea1SDimitry Andric std::set<std::string> GPUBinHandleSymbols;
86*0fca6ea1SDimitry Andric std::set<std::string> DefinedFatBinSymbols;
87*0fca6ea1SDimitry Andric std::set<std::string> DefinedGPUBinHandleSymbols;
88*0fca6ea1SDimitry Andric const std::string FatBinPrefix = "__hip_fatbin";
89*0fca6ea1SDimitry Andric const std::string GPUBinHandlePrefix = "__hip_gpubin_handle";
90*0fca6ea1SDimitry Andric
populateSymbols()91*0fca6ea1SDimitry Andric void populateSymbols() {
92*0fca6ea1SDimitry Andric std::deque<const Action *> WorkList;
93*0fca6ea1SDimitry Andric std::set<const Action *> Visited;
94*0fca6ea1SDimitry Andric
95*0fca6ea1SDimitry Andric for (const auto &Action : C.getActions())
96*0fca6ea1SDimitry Andric WorkList.push_back(Action);
97*0fca6ea1SDimitry Andric
98*0fca6ea1SDimitry Andric while (!WorkList.empty()) {
99*0fca6ea1SDimitry Andric const Action *CurrentAction = WorkList.front();
100*0fca6ea1SDimitry Andric WorkList.pop_front();
101*0fca6ea1SDimitry Andric
102*0fca6ea1SDimitry Andric if (!CurrentAction || !Visited.insert(CurrentAction).second)
103*0fca6ea1SDimitry Andric continue;
104*0fca6ea1SDimitry Andric
105*0fca6ea1SDimitry Andric if (const auto *IA = dyn_cast<InputAction>(CurrentAction)) {
106*0fca6ea1SDimitry Andric std::string ID = IA->getId().str();
107*0fca6ea1SDimitry Andric if (!ID.empty()) {
108*0fca6ea1SDimitry Andric ID = llvm::utohexstr(llvm::MD5Hash(ID), /*LowerCase=*/true);
109*0fca6ea1SDimitry Andric FatBinSymbols.insert((FatBinPrefix + Twine('_') + ID).str());
110*0fca6ea1SDimitry Andric GPUBinHandleSymbols.insert(
111*0fca6ea1SDimitry Andric (GPUBinHandlePrefix + Twine('_') + ID).str());
112*0fca6ea1SDimitry Andric continue;
113*0fca6ea1SDimitry Andric }
114*0fca6ea1SDimitry Andric if (IA->getInputArg().getNumValues() == 0)
115*0fca6ea1SDimitry Andric continue;
116*0fca6ea1SDimitry Andric const char *Filename = IA->getInputArg().getValue();
117*0fca6ea1SDimitry Andric if (!Filename)
118*0fca6ea1SDimitry Andric continue;
119*0fca6ea1SDimitry Andric auto BufferOrErr = llvm::MemoryBuffer::getFile(Filename);
120*0fca6ea1SDimitry Andric // Input action could be options to linker, therefore, ignore it
121*0fca6ea1SDimitry Andric // if cannot read it. If it turns out to be a file that cannot be read,
122*0fca6ea1SDimitry Andric // the error will be caught by the linker.
123*0fca6ea1SDimitry Andric if (!BufferOrErr)
124*0fca6ea1SDimitry Andric continue;
125*0fca6ea1SDimitry Andric
126*0fca6ea1SDimitry Andric processInput(BufferOrErr.get()->getMemBufferRef());
127*0fca6ea1SDimitry Andric } else
128*0fca6ea1SDimitry Andric WorkList.insert(WorkList.end(), CurrentAction->getInputs().begin(),
129*0fca6ea1SDimitry Andric CurrentAction->getInputs().end());
130*0fca6ea1SDimitry Andric }
131*0fca6ea1SDimitry Andric }
132*0fca6ea1SDimitry Andric
processInput(const llvm::MemoryBufferRef & Buffer)133*0fca6ea1SDimitry Andric void processInput(const llvm::MemoryBufferRef &Buffer) {
134*0fca6ea1SDimitry Andric // Try processing as object file first.
135*0fca6ea1SDimitry Andric auto ObjFileOrErr = llvm::object::ObjectFile::createObjectFile(Buffer);
136*0fca6ea1SDimitry Andric if (ObjFileOrErr) {
137*0fca6ea1SDimitry Andric processSymbols(**ObjFileOrErr);
138*0fca6ea1SDimitry Andric return;
139*0fca6ea1SDimitry Andric }
140*0fca6ea1SDimitry Andric
141*0fca6ea1SDimitry Andric // Then try processing as archive files.
142*0fca6ea1SDimitry Andric llvm::consumeError(ObjFileOrErr.takeError());
143*0fca6ea1SDimitry Andric auto ArchiveOrErr = llvm::object::Archive::create(Buffer);
144*0fca6ea1SDimitry Andric if (ArchiveOrErr) {
145*0fca6ea1SDimitry Andric llvm::Error Err = llvm::Error::success();
146*0fca6ea1SDimitry Andric llvm::object::Archive &Archive = *ArchiveOrErr.get();
147*0fca6ea1SDimitry Andric for (auto &Child : Archive.children(Err)) {
148*0fca6ea1SDimitry Andric auto ChildBufOrErr = Child.getMemoryBufferRef();
149*0fca6ea1SDimitry Andric if (ChildBufOrErr)
150*0fca6ea1SDimitry Andric processInput(*ChildBufOrErr);
151*0fca6ea1SDimitry Andric else
152*0fca6ea1SDimitry Andric errorHandler(ChildBufOrErr.takeError());
153*0fca6ea1SDimitry Andric }
154*0fca6ea1SDimitry Andric
155*0fca6ea1SDimitry Andric if (Err)
156*0fca6ea1SDimitry Andric errorHandler(std::move(Err));
157*0fca6ea1SDimitry Andric return;
158*0fca6ea1SDimitry Andric }
159*0fca6ea1SDimitry Andric
160*0fca6ea1SDimitry Andric // Ignore other files.
161*0fca6ea1SDimitry Andric llvm::consumeError(ArchiveOrErr.takeError());
162*0fca6ea1SDimitry Andric }
163*0fca6ea1SDimitry Andric
processSymbols(const llvm::object::ObjectFile & Obj)164*0fca6ea1SDimitry Andric void processSymbols(const llvm::object::ObjectFile &Obj) {
165*0fca6ea1SDimitry Andric for (const auto &Symbol : Obj.symbols()) {
166*0fca6ea1SDimitry Andric auto FlagOrErr = Symbol.getFlags();
167*0fca6ea1SDimitry Andric if (!FlagOrErr) {
168*0fca6ea1SDimitry Andric errorHandler(FlagOrErr.takeError());
169*0fca6ea1SDimitry Andric continue;
170*0fca6ea1SDimitry Andric }
171*0fca6ea1SDimitry Andric
172*0fca6ea1SDimitry Andric auto NameOrErr = Symbol.getName();
173*0fca6ea1SDimitry Andric if (!NameOrErr) {
174*0fca6ea1SDimitry Andric errorHandler(NameOrErr.takeError());
175*0fca6ea1SDimitry Andric continue;
176*0fca6ea1SDimitry Andric }
177*0fca6ea1SDimitry Andric llvm::StringRef Name = *NameOrErr;
178*0fca6ea1SDimitry Andric
179*0fca6ea1SDimitry Andric bool isUndefined =
180*0fca6ea1SDimitry Andric FlagOrErr.get() & llvm::object::SymbolRef::SF_Undefined;
181*0fca6ea1SDimitry Andric bool isFatBinSymbol = Name.starts_with(FatBinPrefix);
182*0fca6ea1SDimitry Andric bool isGPUBinHandleSymbol = Name.starts_with(GPUBinHandlePrefix);
183*0fca6ea1SDimitry Andric
184*0fca6ea1SDimitry Andric // Handling for defined symbols
185*0fca6ea1SDimitry Andric if (!isUndefined) {
186*0fca6ea1SDimitry Andric if (isFatBinSymbol) {
187*0fca6ea1SDimitry Andric DefinedFatBinSymbols.insert(Name.str());
188*0fca6ea1SDimitry Andric FatBinSymbols.erase(Name.str());
189*0fca6ea1SDimitry Andric } else if (isGPUBinHandleSymbol) {
190*0fca6ea1SDimitry Andric DefinedGPUBinHandleSymbols.insert(Name.str());
191*0fca6ea1SDimitry Andric GPUBinHandleSymbols.erase(Name.str());
192*0fca6ea1SDimitry Andric }
193*0fca6ea1SDimitry Andric continue;
194*0fca6ea1SDimitry Andric }
195*0fca6ea1SDimitry Andric
196*0fca6ea1SDimitry Andric // Add undefined symbols if they are not in the defined sets
197*0fca6ea1SDimitry Andric if (isFatBinSymbol &&
198*0fca6ea1SDimitry Andric DefinedFatBinSymbols.find(Name.str()) == DefinedFatBinSymbols.end())
199*0fca6ea1SDimitry Andric FatBinSymbols.insert(Name.str());
200*0fca6ea1SDimitry Andric else if (isGPUBinHandleSymbol &&
201*0fca6ea1SDimitry Andric DefinedGPUBinHandleSymbols.find(Name.str()) ==
202*0fca6ea1SDimitry Andric DefinedGPUBinHandleSymbols.end())
203*0fca6ea1SDimitry Andric GPUBinHandleSymbols.insert(Name.str());
204*0fca6ea1SDimitry Andric }
205*0fca6ea1SDimitry Andric }
206*0fca6ea1SDimitry Andric
errorHandler(llvm::Error Err)207*0fca6ea1SDimitry Andric void errorHandler(llvm::Error Err) {
208*0fca6ea1SDimitry Andric if (Quiet)
209*0fca6ea1SDimitry Andric return;
210*0fca6ea1SDimitry Andric C.getDriver().Diag(DiagID) << llvm::toString(std::move(Err));
211*0fca6ea1SDimitry Andric }
212*0fca6ea1SDimitry Andric };
213*0fca6ea1SDimitry Andric
2140eae32dcSDimitry Andric // Construct a clang-offload-bundler command to bundle code objects for
2150eae32dcSDimitry Andric // 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)2160eae32dcSDimitry Andric void HIP::constructHIPFatbinCommand(Compilation &C, const JobAction &JA,
2170eae32dcSDimitry Andric llvm::StringRef OutputFileName,
2180eae32dcSDimitry Andric const InputInfoList &Inputs,
2190eae32dcSDimitry Andric const llvm::opt::ArgList &Args,
2200eae32dcSDimitry Andric const Tool &T) {
2210eae32dcSDimitry Andric // Construct clang-offload-bundler command to bundle object files for
2220eae32dcSDimitry Andric // for different GPU archs.
2230eae32dcSDimitry Andric ArgStringList BundlerArgs;
2240eae32dcSDimitry Andric BundlerArgs.push_back(Args.MakeArgString("-type=o"));
2250eae32dcSDimitry Andric BundlerArgs.push_back(
2260eae32dcSDimitry Andric Args.MakeArgString("-bundle-align=" + Twine(HIPCodeObjectAlign)));
2270eae32dcSDimitry Andric
2280eae32dcSDimitry Andric // ToDo: Remove the dummy host binary entry which is required by
2290eae32dcSDimitry Andric // clang-offload-bundler.
2300eae32dcSDimitry Andric std::string BundlerTargetArg = "-targets=host-x86_64-unknown-linux";
2310eae32dcSDimitry Andric // AMDGCN:
2320eae32dcSDimitry Andric // For code object version 2 and 3, the offload kind in bundle ID is 'hip'
2330eae32dcSDimitry Andric // for backward compatibility. For code object version 4 and greater, the
2340eae32dcSDimitry Andric // offload kind in bundle ID is 'hipv4'.
2350eae32dcSDimitry Andric std::string OffloadKind = "hip";
2360eae32dcSDimitry Andric auto &TT = T.getToolChain().getTriple();
2370eae32dcSDimitry Andric if (TT.isAMDGCN() && getAMDGPUCodeObjectVersion(C.getDriver(), Args) >= 4)
2380eae32dcSDimitry Andric OffloadKind = OffloadKind + "v4";
2390eae32dcSDimitry Andric for (const auto &II : Inputs) {
2400eae32dcSDimitry Andric const auto *A = II.getAction();
2410eae32dcSDimitry Andric auto ArchStr = llvm::StringRef(A->getOffloadingArch());
2420eae32dcSDimitry Andric BundlerTargetArg +=
2430eae32dcSDimitry Andric "," + OffloadKind + "-" + normalizeForBundler(TT, !ArchStr.empty());
2440eae32dcSDimitry Andric if (!ArchStr.empty())
2450eae32dcSDimitry Andric BundlerTargetArg += "-" + ArchStr.str();
2460eae32dcSDimitry Andric }
2470eae32dcSDimitry Andric BundlerArgs.push_back(Args.MakeArgString(BundlerTargetArg));
24881ad6265SDimitry Andric
24981ad6265SDimitry Andric // Use a NULL file as input for the dummy host binary entry
25081ad6265SDimitry Andric std::string BundlerInputArg = "-input=" NULL_FILE;
2510eae32dcSDimitry Andric BundlerArgs.push_back(Args.MakeArgString(BundlerInputArg));
25281ad6265SDimitry Andric for (const auto &II : Inputs) {
25381ad6265SDimitry Andric BundlerInputArg = std::string("-input=") + II.getFilename();
25481ad6265SDimitry Andric BundlerArgs.push_back(Args.MakeArgString(BundlerInputArg));
25581ad6265SDimitry Andric }
2560eae32dcSDimitry Andric
2570eae32dcSDimitry Andric std::string Output = std::string(OutputFileName);
2580eae32dcSDimitry Andric auto *BundlerOutputArg =
25981ad6265SDimitry Andric Args.MakeArgString(std::string("-output=").append(Output));
2600eae32dcSDimitry Andric BundlerArgs.push_back(BundlerOutputArg);
2610eae32dcSDimitry Andric
262*0fca6ea1SDimitry Andric addOffloadCompressArgs(Args, BundlerArgs);
2635f757f3fSDimitry Andric
2640eae32dcSDimitry Andric const char *Bundler = Args.MakeArgString(
2650eae32dcSDimitry Andric T.getToolChain().GetProgramPath("clang-offload-bundler"));
2660eae32dcSDimitry Andric C.addCommand(std::make_unique<Command>(
2670eae32dcSDimitry Andric JA, T, ResponseFileSupport::None(), Bundler, BundlerArgs, Inputs,
2680eae32dcSDimitry Andric InputInfo(&JA, Args.MakeArgString(Output))));
2690eae32dcSDimitry Andric }
2700eae32dcSDimitry Andric
2710eae32dcSDimitry Andric /// Add Generated HIP Object File which has device images embedded into the
2720eae32dcSDimitry Andric /// host to the argument list for linking. Using MC directives, embed the
2730eae32dcSDimitry Andric /// device code and also define symbols required by the code generation so that
2740eae32dcSDimitry Andric /// 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)2750eae32dcSDimitry Andric void HIP::constructGenerateObjFileFromHIPFatBinary(
2760eae32dcSDimitry Andric Compilation &C, const InputInfo &Output, const InputInfoList &Inputs,
2770eae32dcSDimitry Andric const ArgList &Args, const JobAction &JA, const Tool &T) {
2780eae32dcSDimitry Andric const ToolChain &TC = T.getToolChain();
2790eae32dcSDimitry Andric std::string Name = std::string(llvm::sys::path::stem(Output.getFilename()));
2800eae32dcSDimitry Andric
2810eae32dcSDimitry Andric // Create Temp Object File Generator,
2820eae32dcSDimitry Andric // Offload Bundled file and Bundled Object file.
2830eae32dcSDimitry Andric // Keep them if save-temps is enabled.
2840eae32dcSDimitry Andric const char *McinFile;
2850eae32dcSDimitry Andric const char *BundleFile;
2860eae32dcSDimitry Andric if (C.getDriver().isSaveTempsEnabled()) {
2870eae32dcSDimitry Andric McinFile = C.getArgs().MakeArgString(Name + ".mcin");
2880eae32dcSDimitry Andric BundleFile = C.getArgs().MakeArgString(Name + ".hipfb");
2890eae32dcSDimitry Andric } else {
2900eae32dcSDimitry Andric auto TmpNameMcin = C.getDriver().GetTemporaryPath(Name, "mcin");
2910eae32dcSDimitry Andric McinFile = C.addTempFile(C.getArgs().MakeArgString(TmpNameMcin));
2920eae32dcSDimitry Andric auto TmpNameFb = C.getDriver().GetTemporaryPath(Name, "hipfb");
2930eae32dcSDimitry Andric BundleFile = C.addTempFile(C.getArgs().MakeArgString(TmpNameFb));
2940eae32dcSDimitry Andric }
2950eae32dcSDimitry Andric HIP::constructHIPFatbinCommand(C, JA, BundleFile, Inputs, Args, T);
2960eae32dcSDimitry Andric
2970eae32dcSDimitry Andric // Create a buffer to write the contents of the temp obj generator.
2980eae32dcSDimitry Andric std::string ObjBuffer;
2990eae32dcSDimitry Andric llvm::raw_string_ostream ObjStream(ObjBuffer);
3000eae32dcSDimitry Andric
3010eae32dcSDimitry Andric auto HostTriple =
3020eae32dcSDimitry Andric C.getSingleOffloadToolChain<Action::OFK_Host>()->getTriple();
3030eae32dcSDimitry Andric
304*0fca6ea1SDimitry Andric HIPUndefinedFatBinSymbols Symbols(C);
305*0fca6ea1SDimitry Andric
306*0fca6ea1SDimitry Andric std::string PrimaryHipFatbinSymbol;
307*0fca6ea1SDimitry Andric std::string PrimaryGpuBinHandleSymbol;
308*0fca6ea1SDimitry Andric bool FoundPrimaryHipFatbinSymbol = false;
309*0fca6ea1SDimitry Andric bool FoundPrimaryGpuBinHandleSymbol = false;
310*0fca6ea1SDimitry Andric
311*0fca6ea1SDimitry Andric std::vector<std::string> AliasHipFatbinSymbols;
312*0fca6ea1SDimitry Andric std::vector<std::string> AliasGpuBinHandleSymbols;
313*0fca6ea1SDimitry Andric
314*0fca6ea1SDimitry Andric // Iterate through symbols to find the primary ones and collect others for
315*0fca6ea1SDimitry Andric // aliasing
316*0fca6ea1SDimitry Andric for (const auto &Symbol : Symbols.getFatBinSymbols()) {
317*0fca6ea1SDimitry Andric if (!FoundPrimaryHipFatbinSymbol) {
318*0fca6ea1SDimitry Andric PrimaryHipFatbinSymbol = Symbol;
319*0fca6ea1SDimitry Andric FoundPrimaryHipFatbinSymbol = true;
320*0fca6ea1SDimitry Andric } else
321*0fca6ea1SDimitry Andric AliasHipFatbinSymbols.push_back(Symbol);
322*0fca6ea1SDimitry Andric }
323*0fca6ea1SDimitry Andric
324*0fca6ea1SDimitry Andric for (const auto &Symbol : Symbols.getGPUBinHandleSymbols()) {
325*0fca6ea1SDimitry Andric if (!FoundPrimaryGpuBinHandleSymbol) {
326*0fca6ea1SDimitry Andric PrimaryGpuBinHandleSymbol = Symbol;
327*0fca6ea1SDimitry Andric FoundPrimaryGpuBinHandleSymbol = true;
328*0fca6ea1SDimitry Andric } else
329*0fca6ea1SDimitry Andric AliasGpuBinHandleSymbols.push_back(Symbol);
330*0fca6ea1SDimitry Andric }
331*0fca6ea1SDimitry Andric
3320eae32dcSDimitry Andric // Add MC directives to embed target binaries. We ensure that each
3330eae32dcSDimitry Andric // section and image is 16-byte aligned. This is not mandatory, but
3340eae32dcSDimitry Andric // increases the likelihood of data to be aligned with a cache block
3350eae32dcSDimitry Andric // in several main host machines.
3360eae32dcSDimitry Andric ObjStream << "# HIP Object Generator\n";
3370eae32dcSDimitry Andric ObjStream << "# *** Automatically generated by Clang ***\n";
338*0fca6ea1SDimitry Andric if (FoundPrimaryGpuBinHandleSymbol) {
339*0fca6ea1SDimitry Andric // Define the first gpubin handle symbol
340*0fca6ea1SDimitry Andric if (HostTriple.isWindowsMSVCEnvironment())
341*0fca6ea1SDimitry Andric ObjStream << " .section .hip_gpubin_handle,\"dw\"\n";
342*0fca6ea1SDimitry Andric else {
343*0fca6ea1SDimitry Andric ObjStream << " .protected " << PrimaryGpuBinHandleSymbol << "\n";
344*0fca6ea1SDimitry Andric ObjStream << " .type " << PrimaryGpuBinHandleSymbol << ",@object\n";
345*0fca6ea1SDimitry Andric ObjStream << " .section .hip_gpubin_handle,\"aw\"\n";
346*0fca6ea1SDimitry Andric }
347*0fca6ea1SDimitry Andric ObjStream << " .globl " << PrimaryGpuBinHandleSymbol << "\n";
348*0fca6ea1SDimitry Andric ObjStream << " .p2align 3\n"; // Align 8
349*0fca6ea1SDimitry Andric ObjStream << PrimaryGpuBinHandleSymbol << ":\n";
350*0fca6ea1SDimitry Andric ObjStream << " .zero 8\n"; // Size 8
351*0fca6ea1SDimitry Andric
352*0fca6ea1SDimitry Andric // Generate alias directives for other gpubin handle symbols
353*0fca6ea1SDimitry Andric for (const auto &AliasSymbol : AliasGpuBinHandleSymbols) {
354*0fca6ea1SDimitry Andric ObjStream << " .globl " << AliasSymbol << "\n";
355*0fca6ea1SDimitry Andric ObjStream << " .set " << AliasSymbol << "," << PrimaryGpuBinHandleSymbol
356*0fca6ea1SDimitry Andric << "\n";
357*0fca6ea1SDimitry Andric }
358*0fca6ea1SDimitry Andric }
359*0fca6ea1SDimitry Andric if (FoundPrimaryHipFatbinSymbol) {
360*0fca6ea1SDimitry Andric // Define the first fatbin symbol
361*0fca6ea1SDimitry Andric if (HostTriple.isWindowsMSVCEnvironment())
3620eae32dcSDimitry Andric ObjStream << " .section .hip_fatbin,\"dw\"\n";
363*0fca6ea1SDimitry Andric else {
364*0fca6ea1SDimitry Andric ObjStream << " .protected " << PrimaryHipFatbinSymbol << "\n";
365*0fca6ea1SDimitry Andric ObjStream << " .type " << PrimaryHipFatbinSymbol << ",@object\n";
3660eae32dcSDimitry Andric ObjStream << " .section .hip_fatbin,\"a\",@progbits\n";
3670eae32dcSDimitry Andric }
368*0fca6ea1SDimitry Andric ObjStream << " .globl " << PrimaryHipFatbinSymbol << "\n";
3690eae32dcSDimitry Andric ObjStream << " .p2align " << llvm::Log2(llvm::Align(HIPCodeObjectAlign))
3700eae32dcSDimitry Andric << "\n";
371*0fca6ea1SDimitry Andric // Generate alias directives for other fatbin symbols
372*0fca6ea1SDimitry Andric for (const auto &AliasSymbol : AliasHipFatbinSymbols) {
373*0fca6ea1SDimitry Andric ObjStream << " .globl " << AliasSymbol << "\n";
374*0fca6ea1SDimitry Andric ObjStream << " .set " << AliasSymbol << "," << PrimaryHipFatbinSymbol
375*0fca6ea1SDimitry Andric << "\n";
376*0fca6ea1SDimitry Andric }
377*0fca6ea1SDimitry Andric ObjStream << PrimaryHipFatbinSymbol << ":\n";
3780eae32dcSDimitry Andric ObjStream << " .incbin ";
3790eae32dcSDimitry Andric llvm::sys::printArg(ObjStream, BundleFile, /*Quote=*/true);
3800eae32dcSDimitry Andric ObjStream << "\n";
381*0fca6ea1SDimitry Andric }
3825f757f3fSDimitry Andric if (HostTriple.isOSLinux() && HostTriple.isOSBinFormatELF())
3835f757f3fSDimitry Andric ObjStream << " .section .note.GNU-stack, \"\", @progbits\n";
3840eae32dcSDimitry Andric ObjStream.flush();
3850eae32dcSDimitry Andric
3860eae32dcSDimitry Andric // Dump the contents of the temp object file gen if the user requested that.
3870eae32dcSDimitry Andric // We support this option to enable testing of behavior with -###.
3880eae32dcSDimitry Andric if (C.getArgs().hasArg(options::OPT_fhip_dump_offload_linker_script))
3890eae32dcSDimitry Andric llvm::errs() << ObjBuffer;
3900eae32dcSDimitry Andric
3910eae32dcSDimitry Andric // Open script file and write the contents.
3920eae32dcSDimitry Andric std::error_code EC;
3930eae32dcSDimitry Andric llvm::raw_fd_ostream Objf(McinFile, EC, llvm::sys::fs::OF_None);
3940eae32dcSDimitry Andric
3950eae32dcSDimitry Andric if (EC) {
3960eae32dcSDimitry Andric C.getDriver().Diag(clang::diag::err_unable_to_make_temp) << EC.message();
3970eae32dcSDimitry Andric return;
3980eae32dcSDimitry Andric }
3990eae32dcSDimitry Andric
4000eae32dcSDimitry Andric Objf << ObjBuffer;
4010eae32dcSDimitry Andric
4020eae32dcSDimitry Andric ArgStringList McArgs{"-triple", Args.MakeArgString(HostTriple.normalize()),
4030eae32dcSDimitry Andric "-o", Output.getFilename(),
4040eae32dcSDimitry Andric McinFile, "--filetype=obj"};
4050eae32dcSDimitry Andric const char *Mc = Args.MakeArgString(TC.GetProgramPath("llvm-mc"));
4060eae32dcSDimitry Andric C.addCommand(std::make_unique<Command>(JA, T, ResponseFileSupport::None(), Mc,
4070eae32dcSDimitry Andric McArgs, Inputs, Output));
4080eae32dcSDimitry Andric }
409