llvm-project/lld/ELF/ICF.cpp

350 lines
12 KiB
C++

//===- ICF.cpp ------------------------------------------------------------===//
//
// The LLVM Linker
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// Identical Code Folding is a feature to merge sections not by name (which
// is regular comdat handling) but by contents. If two non-writable sections
// have the same data, relocations, attributes, etc., then the two
// are considered identical and merged by the linker. This optimization
// makes outputs smaller.
//
// ICF is theoretically a problem of reducing graphs by merging as many
// identical subgraphs as possible if we consider sections as vertices and
// relocations as edges. It may sound simple, but it is a bit more
// complicated than you might think. The order of processing sections
// matters because merging two sections can make other sections, whose
// relocations now point to the same section, mergeable. Graphs may contain
// cycles. We need a sophisticated algorithm to do this properly and
// efficiently.
//
// What we do in this file is this. We split sections into groups. Sections
// in the same group are considered identical.
//
// We begin by optimistically putting all sections into a single equivalence
// class. Then we apply a series of checks that split this initial
// equivalence class into more and more refined equivalence classes based on
// the properties by which a section can be distinguished.
//
// We begin by checking that the section contents and flags are the
// same. This only needs to be done once since these properties don't depend
// on the current equivalence class assignment.
//
// Then we split the equivalence classes based on checking that their
// relocations are the same, where relocation targets are compared by their
// equivalence class, not the concrete section. This may need to be done
// multiple times because as the equivalence classes are refined, two
// sections that had a relocation target in the same equivalence class may
// now target different equivalence classes, and hence these two sections
// must be put in different equivalence classes (whereas in the previous
// iteration they were not since the relocation target was the same.)
//
// Our algorithm is smart enough to merge the following mutually-recursive
// functions.
//
// void foo() { bar(); }
// void bar() { foo(); }
//
// This algorithm is so-called "optimistic" algorithm described in
// http://research.google.com/pubs/pub36912.html. (Note that what GNU
// gold implemented is different from the optimistic algorithm.)
//
//===----------------------------------------------------------------------===//
#include "ICF.h"
#include "Config.h"
#include "OutputSections.h"
#include "SymbolTable.h"
#include "llvm/ADT/Hashing.h"
#include "llvm/Object/ELF.h"
#include "llvm/Support/ELF.h"
#include "llvm/Support/raw_ostream.h"
using namespace lld;
using namespace lld::elf;
using namespace llvm;
using namespace llvm::ELF;
using namespace llvm::object;
namespace lld {
namespace elf {
template <class ELFT> class ICF {
typedef typename ELFT::Shdr Elf_Shdr;
typedef typename ELFT::Sym Elf_Sym;
typedef typename ELFT::uint uintX_t;
typedef Elf_Rel_Impl<ELFT, false> Elf_Rel;
using Comparator = std::function<bool(const InputSection<ELFT> *,
const InputSection<ELFT> *)>;
public:
void run();
private:
uint64_t NextId = 1;
static void setLive(SymbolTable<ELFT> *S);
static uint64_t relSize(InputSection<ELFT> *S);
static uint64_t getHash(InputSection<ELFT> *S);
static bool isEligible(InputSectionBase<ELFT> *Sec);
static std::vector<InputSection<ELFT> *> getSections();
void segregate(InputSection<ELFT> **Begin, InputSection<ELFT> **End,
Comparator Eq);
void forEachGroup(std::vector<InputSection<ELFT> *> &V, Comparator Eq);
template <class RelTy>
static bool relocationEq(ArrayRef<RelTy> RA, ArrayRef<RelTy> RB);
template <class RelTy>
static bool variableEq(const InputSection<ELFT> *A,
const InputSection<ELFT> *B, ArrayRef<RelTy> RA,
ArrayRef<RelTy> RB);
static bool equalsConstant(const InputSection<ELFT> *A,
const InputSection<ELFT> *B);
static bool equalsVariable(const InputSection<ELFT> *A,
const InputSection<ELFT> *B);
};
}
}
// Returns a hash value for S. Note that the information about
// relocation targets is not included in the hash value.
template <class ELFT> uint64_t ICF<ELFT>::getHash(InputSection<ELFT> *S) {
uint64_t Flags = S->getSectionHdr()->sh_flags;
uint64_t H = hash_combine(Flags, S->getSize());
for (const Elf_Shdr *Rel : S->RelocSections)
H = hash_combine(H, (uint64_t)Rel->sh_size);
return H;
}
// Returns true if Sec is subject of ICF.
template <class ELFT> bool ICF<ELFT>::isEligible(InputSectionBase<ELFT> *Sec) {
if (!Sec || Sec == &InputSection<ELFT>::Discarded || !Sec->Live)
return false;
auto *S = dyn_cast<InputSection<ELFT>>(Sec);
if (!S)
return false;
// .init and .fini contains instructions that must be executed to
// initialize and finalize the process. They cannot and should not
// be merged.
StringRef Name = S->getSectionName();
if (Name == ".init" || Name == ".fini")
return false;
const Elf_Shdr &H = *S->getSectionHdr();
return (H.sh_flags & SHF_ALLOC) && (~H.sh_flags & SHF_WRITE);
}
template <class ELFT>
std::vector<InputSection<ELFT> *> ICF<ELFT>::getSections() {
std::vector<InputSection<ELFT> *> V;
for (const std::unique_ptr<ObjectFile<ELFT>> &F :
Symtab<ELFT>::X->getObjectFiles())
for (InputSectionBase<ELFT> *S : F->getSections())
if (isEligible(S))
V.push_back(cast<InputSection<ELFT>>(S));
return V;
}
// All sections between Begin and End must have the same group ID before
// you call this function. This function compare sections between Begin
// and End using Eq and assign new group IDs for new groups.
template <class ELFT>
void ICF<ELFT>::segregate(InputSection<ELFT> **Begin, InputSection<ELFT> **End,
Comparator Eq) {
// This loop rearranges [Begin, End) so that all sections that are
// equal in terms of Eq are contiguous. The algorithm is quadratic in
// the worst case, but that is not an issue in practice because the
// number of distinct sections in [Begin, End) is usually very small.
InputSection<ELFT> **I = Begin;
for (;;) {
InputSection<ELFT> *Head = *I;
auto Bound = std::stable_partition(
I + 1, End, [&](InputSection<ELFT> *S) { return Eq(Head, S); });
if (Bound == End)
return;
uint64_t Id = NextId++;
for (; I != Bound; ++I)
(*I)->GroupId = Id;
}
}
template <class ELFT>
void ICF<ELFT>::forEachGroup(std::vector<InputSection<ELFT> *> &V,
Comparator Eq) {
for (InputSection<ELFT> **I = V.data(), **E = I + V.size(); I != E;) {
InputSection<ELFT> *Head = *I;
auto Bound = std::find_if(I + 1, E, [&](InputSection<ELFT> *S) {
return S->GroupId != Head->GroupId;
});
segregate(I, Bound, Eq);
I = Bound;
}
}
// Compare two lists of relocations.
template <class ELFT>
template <class RelTy>
bool ICF<ELFT>::relocationEq(ArrayRef<RelTy> RelsA, ArrayRef<RelTy> RelsB) {
const RelTy *IA = RelsA.begin();
const RelTy *EA = RelsA.end();
const RelTy *IB = RelsB.begin();
const RelTy *EB = RelsB.end();
if (EA - IA != EB - IB)
return false;
for (; IA != EA; ++IA, ++IB)
if (IA->r_offset != IB->r_offset ||
IA->getType(Config->Mips64EL) != IB->getType(Config->Mips64EL) ||
getAddend<ELFT>(*IA) != getAddend<ELFT>(*IB))
return false;
return true;
}
// Compare "non-moving" part of two InputSections, namely everything
// except relocation targets.
template <class ELFT>
bool ICF<ELFT>::equalsConstant(const InputSection<ELFT> *A,
const InputSection<ELFT> *B) {
if (A->RelocSections.size() != B->RelocSections.size())
return false;
for (size_t I = 0, E = A->RelocSections.size(); I != E; ++I) {
const Elf_Shdr *RA = A->RelocSections[I];
const Elf_Shdr *RB = B->RelocSections[I];
ELFFile<ELFT> &FileA = A->File->getObj();
ELFFile<ELFT> &FileB = B->File->getObj();
if (RA->sh_type == SHT_RELA) {
if (!relocationEq(FileA.relas(RA), FileB.relas(RB)))
return false;
} else {
if (!relocationEq(FileA.rels(RA), FileB.rels(RB)))
return false;
}
}
return A->getSectionHdr()->sh_flags == B->getSectionHdr()->sh_flags &&
A->getSize() == B->getSize() &&
A->getSectionData() == B->getSectionData();
}
template <class ELFT>
template <class RelTy>
bool ICF<ELFT>::variableEq(const InputSection<ELFT> *A,
const InputSection<ELFT> *B, ArrayRef<RelTy> RelsA,
ArrayRef<RelTy> RelsB) {
const RelTy *IA = RelsA.begin();
const RelTy *EA = RelsA.end();
const RelTy *IB = RelsB.begin();
for (; IA != EA; ++IA, ++IB) {
SymbolBody &SA = A->File->getRelocTargetSym(*IA);
SymbolBody &SB = B->File->getRelocTargetSym(*IB);
if (&SA == &SB)
continue;
// Or, the symbols should be pointing to the same section
// in terms of the group ID.
auto *DA = dyn_cast<DefinedRegular<ELFT>>(&SA);
auto *DB = dyn_cast<DefinedRegular<ELFT>>(&SB);
if (!DA || !DB)
return false;
if (DA->Value != DB->Value)
return false;
InputSection<ELFT> *X = dyn_cast<InputSection<ELFT>>(DA->Section);
InputSection<ELFT> *Y = dyn_cast<InputSection<ELFT>>(DB->Section);
if (X && Y && X->GroupId && X->GroupId == Y->GroupId)
continue;
return false;
}
return true;
}
// Compare "moving" part of two InputSections, namely relocation targets.
template <class ELFT>
bool ICF<ELFT>::equalsVariable(const InputSection<ELFT> *A,
const InputSection<ELFT> *B) {
for (size_t I = 0, E = A->RelocSections.size(); I != E; ++I) {
const Elf_Shdr *RA = A->RelocSections[I];
const Elf_Shdr *RB = B->RelocSections[I];
ELFFile<ELFT> &FileA = A->File->getObj();
ELFFile<ELFT> &FileB = B->File->getObj();
if (RA->sh_type == SHT_RELA) {
if (!variableEq(A, B, FileA.relas(RA), FileB.relas(RB)))
return false;
} else {
if (!variableEq(A, B, FileA.rels(RA), FileB.rels(RB)))
return false;
}
}
return true;
}
// The main function of ICF.
template <class ELFT> void ICF<ELFT>::run() {
// Initially, we use hash values as section group IDs. Therefore,
// if two sections have the same ID, they are likely (but not
// guaranteed) to have the same static contents in terms of ICF.
std::vector<InputSection<ELFT> *> V = getSections();
for (InputSection<ELFT> *S : V)
// Set MSB on to avoid collisions with serial group IDs
S->GroupId = getHash(S) | (uint64_t(1) << 63);
// From now on, sections in V are ordered so that sections in
// the same group are consecutive in the vector.
std::stable_sort(V.begin(), V.end(),
[](InputSection<ELFT> *A, InputSection<ELFT> *B) {
if (A->GroupId != B->GroupId)
return A->GroupId < B->GroupId;
// Within a group, put the highest alignment
// requirement first, so that's the one we'll keep.
return B->Alignment < A->Alignment;
});
// Compare static contents and assign unique IDs for each static content.
forEachGroup(V, equalsConstant);
// Split groups by comparing relocations until we get a convergence.
int Cnt = 1;
for (;;) {
++Cnt;
uint64_t Id = NextId;
forEachGroup(V, equalsVariable);
if (Id == NextId)
break;
}
log("ICF needed " + Twine(Cnt) + " iterations.");
// Merge sections in the same group.
for (auto I = V.begin(), E = V.end(); I != E;) {
InputSection<ELFT> *Head = *I++;
auto Bound = std::find_if(I, E, [&](InputSection<ELFT> *S) {
return Head->GroupId != S->GroupId;
});
if (I == Bound)
continue;
log("selected " + Head->getSectionName());
while (I != Bound) {
InputSection<ELFT> *S = *I++;
log(" removed " + S->getSectionName());
Head->replace(S);
}
}
}
// ICF entry point function.
template <class ELFT> void elf::doIcf() { ICF<ELFT>().run(); }
template void elf::doIcf<ELF32LE>();
template void elf::doIcf<ELF32BE>();
template void elf::doIcf<ELF64LE>();
template void elf::doIcf<ELF64BE>();