llvm-project/lld/COFF/Symbols.h

419 lines
13 KiB
C
Raw Normal View History

//===- Symbols.h ------------------------------------------------*- C++ -*-===//
//
// The LLVM Linker
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#ifndef LLD_COFF_SYMBOLS_H
#define LLD_COFF_SYMBOLS_H
#include "Chunks.h"
#include "Config.h"
#include "Memory.h"
#include "lld/Core/LLVM.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/Object/Archive.h"
#include "llvm/Object/COFF.h"
#include <atomic>
#include <memory>
#include <vector>
namespace lld {
namespace coff {
using llvm::object::Archive;
using llvm::object::COFFSymbolRef;
using llvm::object::coff_import_header;
using llvm::object::coff_symbol_generic;
class ArchiveFile;
class InputFile;
class ObjectFile;
struct Symbol;
class SymbolTable;
// The base class for real symbol classes.
class SymbolBody {
public:
enum Kind {
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
// The order of these is significant. We start with the regular defined
// symbols as those are the most prevelant and the zero tag is the cheapest
// to set. Among the defined kinds, the lower the kind is preferred over
// the higher kind when testing wether one symbol should take precedence
// over another.
DefinedRegularKind = 0,
DefinedCommonKind,
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
DefinedLocalImportKind,
DefinedImportThunkKind,
DefinedImportDataKind,
DefinedAbsoluteKind,
DefinedRelativeKind,
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
UndefinedKind,
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
LazyKind,
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
LastDefinedCOFFKind = DefinedCommonKind,
LastDefinedKind = DefinedRelativeKind,
};
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
Kind kind() const { return static_cast<Kind>(SymbolKind); }
// Returns true if this is an external symbol.
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
bool isExternal() { return IsExternal; }
// Returns the symbol name.
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
StringRef getName();
// Returns the file from which this symbol was created.
InputFile *getFile();
Symbol *symbol();
const Symbol *symbol() const {
return const_cast<SymbolBody *>(this)->symbol();
}
protected:
friend SymbolTable;
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
explicit SymbolBody(Kind K, StringRef N = "")
: SymbolKind(K), IsExternal(true), IsCOMDAT(false),
WrittenToSymtab(false), Name(N) {}
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
const unsigned SymbolKind : 8;
unsigned IsExternal : 1;
// This bit is used by the \c DefinedRegular subclass.
unsigned IsCOMDAT : 1;
public:
// This bit is used by Writer::createSymbolAndStringTable() to prevent
// symbols from being written to the symbol table more than once.
unsigned WrittenToSymtab : 1;
protected:
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
StringRef Name;
};
// The base class for any defined symbols, including absolute symbols,
// etc.
class Defined : public SymbolBody {
public:
Defined(Kind K, StringRef N) : SymbolBody(K, N) {}
static bool classof(const SymbolBody *S) {
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
return S->kind() <= LastDefinedKind;
}
// Returns the RVA (relative virtual address) of this symbol. The
// writer sets and uses RVAs.
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
uint64_t getRVA();
// Returns the RVA relative to the beginning of the output section.
// Used to implement SECREL relocation type.
uint64_t getSecrel();
// Returns the output section index.
// Used to implement SECTION relocation type.
uint64_t getSectionIndex();
// Returns true if this symbol points to an executable (e.g. .text) section.
// Used to implement ARM relocations.
bool isExecutable();
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
};
// Symbols defined via a COFF object file or bitcode file. For COFF files, this
// stores a coff_symbol_generic*, and names of internal symbols are lazily
// loaded through that. For bitcode files, Sym is nullptr and the name is stored
// as a StringRef.
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
class DefinedCOFF : public Defined {
friend SymbolBody;
public:
DefinedCOFF(Kind K, InputFile *F, StringRef N, const coff_symbol_generic *S)
: Defined(K, N), File(F), Sym(S) {}
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
static bool classof(const SymbolBody *S) {
return S->kind() <= LastDefinedCOFFKind;
}
InputFile *getFile() { return File; }
COFFSymbolRef getCOFFSymbol();
InputFile *File;
protected:
const coff_symbol_generic *Sym;
};
// Regular defined symbols read from object file symbol tables.
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
class DefinedRegular : public DefinedCOFF {
public:
DefinedRegular(InputFile *F, StringRef N, bool IsCOMDAT,
bool IsExternal = false,
const coff_symbol_generic *S = nullptr,
SectionChunk *C = nullptr)
: DefinedCOFF(DefinedRegularKind, F, N, S), Data(C ? &C->Repl : nullptr) {
this->IsExternal = IsExternal;
this->IsCOMDAT = IsCOMDAT;
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedRegularKind;
}
uint64_t getRVA() { return (*Data)->getRVA() + Sym->Value; }
bool isCOMDAT() { return IsCOMDAT; }
[opt] Replace the recursive walk for GC with a worklist algorithm. This flattens the entire liveness walk from a recursive mark approach to a worklist approach. It also sinks the worklist management completely out of the SectionChunk and into the Writer by exposing the ability to iterato over children of a chunk and over the symbol bodies of relocated symbols. I'm not 100% happy with the API names, so suggestions welcome there. This allows us to use a single worklist for the entire recursive walk and would also be a natural place to take advantage of parallelism at some future point. With this, we completely inline away the GC walk into the Writer::markLive function and it makes it very easy to profile what is slow. Currently, time is being wasted checking whether a Chunk isa SectionChunk (it essentially always is), finding (or skipping) a replacement for a symbol, and chasing pointers between symbols and their chunks. There are a bunch of things we can do to fix this, and its easier to do them after this change IMO. This change alone saves 1-2% of the time for my self-link of lld.exe (which I'm running and benchmarking on Linux ironically). Perhaps more notably, we'll no longer blow out the stack for large links. =] Just as an FYI, at this point, I/O is starting to really dominate the profile. Well over 10% of the time appears to be inside the kernel doing page table silliness. I think a decent chunk of this can be nuked as well, but it's a little odd as cross-linking in this way isn't really the primary goal here. Differential Revision: http://reviews.llvm.org/D10790 llvm-svn: 240995
2015-06-30 05:12:49 +08:00
SectionChunk *getChunk() { return *Data; }
uint32_t getValue() { return Sym->Value; }
private:
SectionChunk **Data;
};
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
class DefinedCommon : public DefinedCOFF {
public:
DefinedCommon(InputFile *F, StringRef N, uint64_t Size,
const coff_symbol_generic *S = nullptr,
CommonChunk *C = nullptr)
: DefinedCOFF(DefinedCommonKind, F, N, S), Data(C), Size(Size) {
this->IsExternal = true;
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedCommonKind;
}
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
uint64_t getRVA() { return Data->getRVA(); }
private:
friend SymbolTable;
uint64_t getSize() const { return Size; }
CommonChunk *Data;
uint64_t Size;
};
// Absolute symbols.
class DefinedAbsolute : public Defined {
public:
DefinedAbsolute(StringRef N, COFFSymbolRef S)
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
: Defined(DefinedAbsoluteKind, N), VA(S.getValue()) {
IsExternal = S.isExternal();
}
DefinedAbsolute(StringRef N, uint64_t V)
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
: Defined(DefinedAbsoluteKind, N), VA(V) {}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedAbsoluteKind;
}
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
uint64_t getRVA() { return VA - Config->ImageBase; }
void setVA(uint64_t V) { VA = V; }
private:
uint64_t VA;
};
// This is a kind of absolute symbol but relative to the image base.
// Unlike absolute symbols, relocations referring this kind of symbols
// are subject of the base relocation. This type is used rarely --
// mainly for __ImageBase.
class DefinedRelative : public Defined {
public:
explicit DefinedRelative(StringRef Name, uint64_t V = 0)
: Defined(DefinedRelativeKind, Name), RVA(V) {}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedRelativeKind;
}
uint64_t getRVA() { return RVA; }
void setRVA(uint64_t V) { RVA = V; }
private:
uint64_t RVA;
};
// This class represents a symbol defined in an archive file. It is
// created from an archive file header, and it knows how to load an
// object file from an archive to replace itself with a defined
// symbol. If the resolver finds both Undefined and Lazy for
// the same name, it will ask the Lazy to load a file.
class Lazy : public SymbolBody {
public:
Lazy(ArchiveFile *F, const Archive::Symbol S)
[opt] Devirtualize the SymbolBody type hierarchy and start compacting its members into the base class. First, to help motivate this kind of change, understand that in a self-link, LLD creates 5.5 million defined regular symbol bodies (and 6 million symbol bodies total). A significant portion of its time is spent allocating the memory for these symbols, and befor ethis patch the defined regular symbol body objects alone consumed some 420mb of memory during the self link. As a consequence, I think it is worth expending considerable effort to make these objects as memory efficient as possible. This is the first of several components of that. This change starts with the goal of removing the virtual functins from SymbolBody so that it can avoid having a vptr embedded in it when it already contains a "kind" member, and that member can be much more compact than a vptr. The primary way of doing this is to sink as much of the logic that we would have to dispatch for into data in the base class. As part of this, I made the various flags bits that will pack into a bitfield with the kind tag. I also sank the Name down to eliminate the dispatch for that, and used LLVM's RTTI-style dispatch for everything else (most of which is cold and so doesn't matter terribly if we get minutely worse lowering than a vtable dispatch). As I was doing this, I wanted to make the RTTI-dispatch (which would become much hotter than before) as efficient as possible, so I've re-organized the tags somewhat. Notably, the common case (regular defined symbols) is now zero which we can test for faster. I also needed to rewrite the comparison routine used during resolving symbols. This proved to be quite complex as the semantics of the existing one were very subtle due to the back-and-forth virtual dispatch caused by re-dispatching with reversed operands. I've consolidated it to a single function and tried to comment it quite a bit more to help explain what is going on. However, this may need more comments or other explanations. It at least passes all the regression tests. I'm not working on Windows, so I can't fully test it. With all of these changes, the size of a DefinedRegular symbol on a 64-bit build goes from 80 bytes to 64 bytes, and we save approximately 84mb or 20% of the memory consumed by these symbol bodies during the link. The link time appears marginally faster as well, and the profile hotness of the memory allocation subsystem got a bit better, but there is still a lot of allocation traffic. Differential Revision: http://reviews.llvm.org/D10792 llvm-svn: 241001
2015-06-30 05:35:48 +08:00
: SymbolBody(LazyKind, S.getName()), File(F), Sym(S) {}
static bool classof(const SymbolBody *S) { return S->kind() == LazyKind; }
ArchiveFile *File;
private:
friend SymbolTable;
private:
const Archive::Symbol Sym;
};
// Undefined symbols.
class Undefined : public SymbolBody {
public:
explicit Undefined(StringRef N) : SymbolBody(UndefinedKind, N) {}
static bool classof(const SymbolBody *S) {
return S->kind() == UndefinedKind;
}
// An undefined symbol can have a fallback symbol which gives an
// undefined symbol a second chance if it would remain undefined.
// If it remains undefined, it'll be replaced with whatever the
// Alias pointer points to.
SymbolBody *WeakAlias = nullptr;
// If this symbol is external weak, try to resolve it to a defined
// symbol by searching the chain of fallback symbols. Returns the symbol if
// successful, otherwise returns null.
Defined *getWeakAlias();
};
// Windows-specific classes.
// This class represents a symbol imported from a DLL. This has two
// names for internal use and external use. The former is used for
// name resolution, and the latter is used for the import descriptor
// table in an output. The former has "__imp_" prefix.
class DefinedImportData : public Defined {
public:
DefinedImportData(StringRef N, ImportFile *F)
: Defined(DefinedImportDataKind, N), File(F) {
}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedImportDataKind;
}
uint64_t getRVA() { return File->Location->getRVA(); }
StringRef getDLLName() { return File->DLLName; }
StringRef getExternalName() { return File->ExternalName; }
void setLocation(Chunk *AddressTable) { File->Location = AddressTable; }
uint16_t getOrdinal() { return File->Hdr->OrdinalHint; }
private:
ImportFile *File;
};
// This class represents a symbol for a jump table entry which jumps
// to a function in a DLL. Linker are supposed to create such symbols
// without "__imp_" prefix for all function symbols exported from
// DLLs, so that you can call DLL functions as regular functions with
// a regular name. A function pointer is given as a DefinedImportData.
class DefinedImportThunk : public Defined {
public:
DefinedImportThunk(StringRef Name, DefinedImportData *S, uint16_t Machine);
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedImportThunkKind;
}
uint64_t getRVA() { return Data->getRVA(); }
Chunk *getChunk() { return Data; }
private:
Chunk *Data;
};
// If you have a symbol "__imp_foo" in your object file, a symbol name
// "foo" becomes automatically available as a pointer to "__imp_foo".
// This class is for such automatically-created symbols.
// Yes, this is an odd feature. We didn't intend to implement that.
// This is here just for compatibility with MSVC.
class DefinedLocalImport : public Defined {
public:
DefinedLocalImport(StringRef N, Defined *S)
: Defined(DefinedLocalImportKind, N), Data(make<LocalImportChunk>(S)) {}
static bool classof(const SymbolBody *S) {
return S->kind() == DefinedLocalImportKind;
}
uint64_t getRVA() { return Data->getRVA(); }
Chunk *getChunk() { return Data; }
private:
LocalImportChunk *Data;
};
inline uint64_t Defined::getRVA() {
switch (kind()) {
case DefinedAbsoluteKind:
return cast<DefinedAbsolute>(this)->getRVA();
case DefinedRelativeKind:
return cast<DefinedRelative>(this)->getRVA();
case DefinedImportDataKind:
return cast<DefinedImportData>(this)->getRVA();
case DefinedImportThunkKind:
return cast<DefinedImportThunk>(this)->getRVA();
case DefinedLocalImportKind:
return cast<DefinedLocalImport>(this)->getRVA();
case DefinedCommonKind:
return cast<DefinedCommon>(this)->getRVA();
case DefinedRegularKind:
return cast<DefinedRegular>(this)->getRVA();
case LazyKind:
case UndefinedKind:
llvm_unreachable("Cannot get the address for an undefined symbol.");
}
llvm_unreachable("unknown symbol kind");
}
// A real symbol object, SymbolBody, is usually stored within a Symbol. There's
// always one Symbol for each symbol name. The resolver updates the SymbolBody
// stored in the Body field of this object as it resolves symbols. Symbol also
// holds computed properties of symbol names.
struct Symbol {
// True if this symbol was referenced by a regular (non-bitcode) object.
unsigned IsUsedInRegularObj : 1;
// True if we've seen both a lazy and an undefined symbol with this symbol
// name, which means that we have enqueued an archive member load and should
// not load any more archive members to resolve the same symbol.
unsigned PendingArchiveLoad : 1;
// This field is used to store the Symbol's SymbolBody. This instantiation of
// AlignedCharArrayUnion gives us a struct with a char array field that is
// large and aligned enough to store any derived class of SymbolBody.
llvm::AlignedCharArrayUnion<
DefinedRegular, DefinedCommon, DefinedAbsolute, DefinedRelative, Lazy,
Undefined, DefinedImportData, DefinedImportThunk, DefinedLocalImport>
Body;
SymbolBody *body() {
return reinterpret_cast<SymbolBody *>(Body.buffer);
}
const SymbolBody *body() const { return const_cast<Symbol *>(this)->body(); }
};
template <typename T, typename... ArgT>
void replaceBody(Symbol *S, ArgT &&... Arg) {
static_assert(sizeof(T) <= sizeof(S->Body), "Body too small");
static_assert(alignof(T) <= alignof(decltype(S->Body)),
"Body not aligned enough");
assert(static_cast<SymbolBody *>(static_cast<T *>(nullptr)) == nullptr &&
"Not a SymbolBody");
new (S->Body.buffer) T(std::forward<ArgT>(Arg)...);
}
inline Symbol *SymbolBody::symbol() {
assert(isExternal());
return reinterpret_cast<Symbol *>(reinterpret_cast<char *>(this) -
offsetof(Symbol, Body));
}
} // namespace coff
std::string toString(coff::SymbolBody &B);
} // namespace lld
#endif