hanchenye-llvm-project/lld/COFF/SymbolTable.cpp

445 lines
13 KiB
C++
Raw Normal View History

//===- SymbolTable.cpp ----------------------------------------------------===//
//
// The LLVM Linker
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "Config.h"
#include "Driver.h"
#include "Error.h"
#include "SymbolTable.h"
#include "Symbols.h"
#include "lld/Core/Parallel.h"
#include "llvm/LTO/LTOCodeGenerator.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <utility>
using namespace llvm;
namespace lld {
namespace coff {
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
void SymbolTable::addFile(std::unique_ptr<InputFile> FileP) {
#if LLVM_ENABLE_THREADS
std::launch Policy = std::launch::async;
#else
std::launch Policy = std::launch::deferred;
#endif
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
InputFile *File = FileP.get();
Files.push_back(std::move(FileP));
if (auto *F = dyn_cast<ArchiveFile>(File)) {
ArchiveQueue.push_back(
std::async(Policy, [=]() { F->parse(); return F; }));
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
return;
}
ObjectQueue.push_back(
std::async(Policy, [=]() { File->parse(); return File; }));
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
if (auto *F = dyn_cast<ObjectFile>(File)) {
ObjectFiles.push_back(F);
} else if (auto *F = dyn_cast<BitcodeFile>(File)) {
BitcodeFiles.push_back(F);
} else {
ImportFiles.push_back(cast<ImportFile>(File));
}
}
void SymbolTable::step() {
if (queueEmpty())
return;
readObjects();
readArchives();
}
void SymbolTable::run() {
while (!queueEmpty())
step();
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
}
void SymbolTable::readArchives() {
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
if (ArchiveQueue.empty())
return;
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
// Add lazy symbols to the symbol table. Lazy symbols that conflict
// with existing undefined symbols are accumulated in LazySyms.
std::vector<Symbol *> LazySyms;
for (std::future<ArchiveFile *> &Future : ArchiveQueue) {
ArchiveFile *File = Future.get();
if (Config->Verbose)
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
llvm::outs() << "Reading " << File->getShortName() << "\n";
for (Lazy &Sym : File->getLazySymbols())
addLazy(&Sym, &LazySyms);
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
}
ArchiveQueue.clear();
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
// Add archive member files to ObjectQueue that should resolve
// existing undefined symbols.
for (Symbol *Sym : LazySyms)
addMemberFile(cast<Lazy>(Sym->Body));
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
}
void SymbolTable::readObjects() {
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
if (ObjectQueue.empty())
return;
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
// Add defined and undefined symbols to the symbol table.
std::vector<StringRef> Directives;
for (size_t I = 0; I < ObjectQueue.size(); ++I) {
InputFile *File = ObjectQueue[I].get();
if (Config->Verbose)
llvm::outs() << "Reading " << File->getShortName() << "\n";
// Adding symbols may add more files to ObjectQueue
// (but not to ArchiveQueue).
for (SymbolBody *Sym : File->getSymbols())
if (Sym->isExternal())
addSymbol(Sym);
StringRef S = File->getDirectives();
if (!S.empty()) {
Directives.push_back(S);
if (Config->Verbose)
llvm::outs() << "Directives: " << File->getShortName()
<< ": " << S << "\n";
}
}
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
ObjectQueue.clear();
// Parse directive sections. This may add files to
// ArchiveQueue and ObjectQueue.
for (StringRef S : Directives)
Driver->parseDirectives(S);
}
bool SymbolTable::queueEmpty() {
return ArchiveQueue.empty() && ObjectQueue.empty();
}
void SymbolTable::reportRemainingUndefines(bool Resolve) {
llvm::SmallPtrSet<SymbolBody *, 8> Undefs;
for (auto &I : Symtab) {
Symbol *Sym = I.second;
auto *Undef = dyn_cast<Undefined>(Sym->Body);
if (!Undef)
continue;
StringRef Name = Undef->getName();
// A weak alias may have been resolved, so check for that.
if (Defined *D = Undef->getWeakAlias()) {
if (Resolve)
Sym->Body = D;
continue;
}
// If we can resolve a symbol by removing __imp_ prefix, do that.
// This odd rule is for compatibility with MSVC linker.
if (Name.startswith("__imp_")) {
Symbol *Imp = find(Name.substr(strlen("__imp_")));
if (Imp && isa<Defined>(Imp->Body)) {
if (!Resolve)
continue;
auto *D = cast<Defined>(Imp->Body);
auto *S = new (Alloc) DefinedLocalImport(Name, D);
LocalImportChunks.push_back(S->getChunk());
Sym->Body = S;
continue;
}
}
// Remaining undefined symbols are not fatal if /force is specified.
// They are replaced with dummy defined symbols.
if (Config->Force && Resolve)
Sym->Body = new (Alloc) DefinedAbsolute(Name, 0);
Undefs.insert(Sym->Body);
}
if (Undefs.empty())
return;
for (Undefined *U : Config->GCRoot)
if (Undefs.count(U->repl()))
llvm::errs() << "<root>: undefined symbol: " << U->getName() << "\n";
for (std::unique_ptr<InputFile> &File : Files)
if (!isa<ArchiveFile>(File.get()))
for (SymbolBody *Sym : File->getSymbols())
if (Undefs.count(Sym->repl()))
llvm::errs() << File->getShortName() << ": undefined symbol: "
<< Sym->getName() << "\n";
if (!Config->Force)
error("Link failed");
}
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
void SymbolTable::addLazy(Lazy *New, std::vector<Symbol *> *Accum) {
Symbol *Sym = insert(New);
if (Sym->Body == New)
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
return;
SymbolBody *Existing = Sym->Body;
if (isa<Defined>(Existing))
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
return;
if (Lazy *L = dyn_cast<Lazy>(Existing))
if (L->getFileIndex() < New->getFileIndex())
return;
Sym->Body = New;
New->setBackref(Sym);
if (isa<Undefined>(Existing))
Accum->push_back(Sym);
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
}
void SymbolTable::addSymbol(SymbolBody *New) {
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
// Find an existing symbol or create and insert a new one.
assert(isa<Defined>(New) || isa<Undefined>(New));
Symbol *Sym = insert(New);
if (Sym->Body == New)
return;
SymbolBody *Existing = Sym->Body;
// If we have an undefined symbol and a lazy symbol,
// let the lazy symbol to read a member file.
if (auto *L = dyn_cast<Lazy>(Existing)) {
// Undefined symbols with weak aliases need not to be resolved,
// since they would be replaced with weak aliases if they remain
// undefined.
if (auto *U = dyn_cast<Undefined>(New)) {
if (!U->WeakAlias) {
addMemberFile(L);
return;
}
}
Sym->Body = New;
return;
}
// compare() returns -1, 0, or 1 if the lhs symbol is less preferable,
// equivalent (conflicting), or more preferable, respectively.
int Comp = Existing->compare(New);
if (Comp == 0)
error(Twine("duplicate symbol: ") + Existing->getDebugName() + " and " +
New->getDebugName());
if (Comp < 0)
Sym->Body = New;
}
Symbol *SymbolTable::insert(SymbolBody *New) {
Symbol *&Sym = Symtab[New->getName()];
if (Sym) {
New->setBackref(Sym);
return Sym;
}
Sym = new (Alloc) Symbol(New);
New->setBackref(Sym);
return Sym;
}
// Reads an archive member file pointed by a given symbol.
void SymbolTable::addMemberFile(Lazy *Body) {
std::unique_ptr<InputFile> File = Body->getMember();
// getMember returns an empty buffer if the member was already
// read from the library.
if (!File)
return;
if (Config->Verbose)
llvm::outs() << "Loaded " << File->getShortName() << " for "
<< Body->getName() << "\n";
addFile(std::move(File));
}
std::vector<Chunk *> SymbolTable::getChunks() {
std::vector<Chunk *> Res;
for (ObjectFile *File : ObjectFiles) {
std::vector<Chunk *> &V = File->getChunks();
Res.insert(Res.end(), V.begin(), V.end());
}
return Res;
}
Symbol *SymbolTable::find(StringRef Name) {
auto It = Symtab.find(Name);
if (It == Symtab.end())
return nullptr;
return It->second;
}
Symbol *SymbolTable::findUnderscore(StringRef Name) {
if (Config->Machine == I386)
return find(("_" + Name).str());
return find(Name);
}
StringRef SymbolTable::findByPrefix(StringRef Prefix) {
for (auto Pair : Symtab) {
StringRef Name = Pair.first;
if (Name.startswith(Prefix))
return Name;
}
return "";
}
StringRef SymbolTable::findMangle(StringRef Name) {
if (Symbol *Sym = find(Name))
if (!isa<Undefined>(Sym->Body))
return Name;
if (Config->Machine != I386)
return findByPrefix(("?" + Name + "@@Y").str());
if (!Name.startswith("_"))
return "";
// Search for x86 C function.
StringRef S = findByPrefix((Name + "@").str());
if (!S.empty())
return S;
// Search for x86 C++ non-member function.
return findByPrefix(("?" + Name.substr(1) + "@@Y").str());
}
void SymbolTable::mangleMaybe(Undefined *U) {
if (U->WeakAlias)
return;
if (!isa<Undefined>(U->repl()))
return;
StringRef Alias = findMangle(U->getName());
if (!Alias.empty())
U->WeakAlias = addUndefined(Alias);
}
Undefined *SymbolTable::addUndefined(StringRef Name) {
auto *New = new (Alloc) Undefined(Name);
addSymbol(New);
if (auto *U = dyn_cast<Undefined>(New->repl()))
return U;
return New;
}
DefinedRelative *SymbolTable::addRelative(StringRef Name, uint64_t VA) {
auto *New = new (Alloc) DefinedRelative(Name, VA);
addSymbol(New);
return New;
}
DefinedAbsolute *SymbolTable::addAbsolute(StringRef Name, uint64_t VA) {
auto *New = new (Alloc) DefinedAbsolute(Name, VA);
addSymbol(New);
return New;
}
void SymbolTable::printMap(llvm::raw_ostream &OS) {
for (ObjectFile *File : ObjectFiles) {
OS << File->getShortName() << ":\n";
for (SymbolBody *Body : File->getSymbols())
if (auto *R = dyn_cast<DefinedRegular>(Body))
if (R->getChunk()->isLive())
OS << Twine::utohexstr(Config->ImageBase + R->getRVA())
<< " " << R->getName() << "\n";
}
}
void SymbolTable::addCombinedLTOObject(ObjectFile *Obj) {
for (SymbolBody *Body : Obj->getSymbols()) {
if (!Body->isExternal())
continue;
// We should not see any new undefined symbols at this point, but we'll
// diagnose them later in reportRemainingUndefines().
StringRef Name = Body->getName();
Symbol *Sym = insert(Body);
if (isa<DefinedBitcode>(Sym->Body)) {
Sym->Body = Body;
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
continue;
}
if (auto *L = dyn_cast<Lazy>(Sym->Body)) {
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
// We may see new references to runtime library symbols such as __chkstk
// here. These symbols must be wholly defined in non-bitcode files.
addMemberFile(L);
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
continue;
}
SymbolBody *Existing = Sym->Body;
int Comp = Existing->compare(Body);
if (Comp == 0)
error(Twine("LTO: unexpected duplicate symbol: ") + Name);
COFF: Change the order of adding symbols to the symbol table. Previously, the order of adding symbols to the symbol table was simple. We have a list of all input files. We read each file from beginning of the list and add all symbols in it to the symbol table. This patch changes that order. Now all archive files are added to the symbol table first, and then all the other object files are added. This shouldn't change the behavior in single-threading, and make room to parallelize in multi-threading. In the first step, only lazy symbols are added to the symbol table because archives contain only Lazy symbols. Member object files found to be necessary are queued. In the second step, defined and undefined symbols are added from object files. Adding an undefined symbol to the symbol table may cause more member files to be added to the queue. We simply continue reading all object files until the queue is empty. Finally, new archive or object files may be added to the queues by object files' directive sections (which contain new command line options). The above process is repeated until we get no new files. Symbols defined both in object files and in archives can make results undeterministic. If an archive is read before an object, a new member file gets linked, while in the other way, no new file would be added. That is the most popular cause of an undeterministic result or linking failure as I observed. Separating phases of adding lazy symbols and undefined symbols makes that deterministic. Adding symbols in each phase should be parallelizable. llvm-svn: 241107
2015-07-01 03:35:21 +08:00
if (Comp < 0)
Sym->Body = Body;
}
}
void SymbolTable::addCombinedLTOObjects() {
if (BitcodeFiles.empty())
return;
// Diagnose any undefined symbols early, but do not resolve weak externals,
// as resolution breaks the invariant that each Symbol points to a unique
// SymbolBody, which we rely on to replace DefinedBitcode symbols correctly.
reportRemainingUndefines(/*Resolve=*/false);
// Create an object file and add it to the symbol table by replacing any
// DefinedBitcode symbols with the definitions in the object file.
LTOCodeGenerator CG;
CG.setOptLevel(Config->LTOOptLevel);
std::vector<ObjectFile *> Objs = createLTOObjects(&CG);
for (ObjectFile *Obj : Objs)
addCombinedLTOObject(Obj);
size_t NumBitcodeFiles = BitcodeFiles.size();
run();
if (BitcodeFiles.size() != NumBitcodeFiles)
error("LTO: late loaded symbol created new bitcode reference");
}
// Combine and compile bitcode files and then return the result
// as a vector of regular COFF object files.
std::vector<ObjectFile *> SymbolTable::createLTOObjects(LTOCodeGenerator *CG) {
// All symbols referenced by non-bitcode objects must be preserved.
for (ObjectFile *File : ObjectFiles)
for (SymbolBody *Body : File->getSymbols())
if (auto *S = dyn_cast<DefinedBitcode>(Body->repl()))
CG->addMustPreserveSymbol(S->getName());
// Likewise for bitcode symbols which we initially resolved to non-bitcode.
for (BitcodeFile *File : BitcodeFiles)
for (SymbolBody *Body : File->getSymbols())
if (isa<DefinedBitcode>(Body) && !isa<DefinedBitcode>(Body->repl()))
CG->addMustPreserveSymbol(Body->getName());
// Likewise for other symbols that must be preserved.
for (Undefined *U : Config->GCRoot) {
if (auto *S = dyn_cast<DefinedBitcode>(U->repl()))
CG->addMustPreserveSymbol(S->getName());
else if (auto *S = dyn_cast_or_null<DefinedBitcode>(U->getWeakAlias()))
CG->addMustPreserveSymbol(S->getName());
}
CG->setModule(BitcodeFiles[0]->takeModule());
for (unsigned I = 1, E = BitcodeFiles.size(); I != E; ++I)
CG->addModule(BitcodeFiles[I]->getModule());
bool DisableVerify = true;
#ifdef NDEBUG
DisableVerify = false;
#endif
if (!CG->optimize(DisableVerify, false, false, false))
error(""); // optimize() should have emitted any error message.
Objs.resize(Config->LTOJobs);
// Use std::list to avoid invalidation of pointers in OSPtrs.
std::list<raw_svector_ostream> OSs;
std::vector<raw_pwrite_stream *> OSPtrs;
for (SmallVector<char, 0> &Obj : Objs) {
OSs.emplace_back(Obj);
OSPtrs.push_back(&OSs.back());
}
if (!CG->compileOptimized(OSPtrs))
error(""); // compileOptimized() should have emitted any error message.
std::vector<ObjectFile *> ObjFiles;
for (SmallVector<char, 0> &Obj : Objs) {
auto *ObjFile = new ObjectFile(
MemoryBufferRef(StringRef(Obj.data(), Obj.size()), "<LTO object>"));
Files.emplace_back(ObjFile);
ObjectFiles.push_back(ObjFile);
ObjFile->parse();
ObjFiles.push_back(ObjFile);
}
return ObjFiles;
}
} // namespace coff
} // namespace lld