1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/lib/IR/LLVMContextImpl.cpp
Jianzhou Zhao b6143710cb Reduce dropTriviallyDeadConstantArrays cumulative time percentage from 17% to 4%
The history of dropTriviallyDeadConstantArrays is like this. Because the appending linkage uses too much memory (http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20150105/251381.html), dropTriviallyDeadConstantArrays was introduced (https://reviews.llvm.org/rG81f385b0c6ea37dd7195a65be162c75bbdef29d2) to release unused constant arrays. Recently, dropTriviallyDeadConstantArrays was improved (https://reviews.llvm.org/rG81f385b0c6ea37dd7195a65be162c75bbdef29d2) to reduce its quadratic cost.

Our recent LTO profiling shows that when a target is large, 15-20% of time cost is from the SetVector::insert called by dropTriviallyDeadConstantArrays.

A large application has hundreds or thousands of modules; each module calls dropTriviallyDeadConstantArrays once for cleaning up tens of thousands of ConstantArrays a module has. In those ConstantArrays, usually around 5 can be deleted; a very very few deleted ConstantArrays reference other ConstantArrays: less than 10 out of millions.

Given this, the cost of SetVector::insert is mainly from the construction of WorkList from ArrayConstants. This motivated the fix that iterates ArrayConstants directly, and uses WorkList only when necessary.

Our evaluation shows that
1) The cumulative time percentage of dropTriviallyDeadConstantArrays is reduced from 15-17% to 4-6%.
2) For targets with LTO time > 20min, the time reduction is about 20%.
3) No observable performance impact for build without using LTO.

{F12506218}
{F12506221}

Reviewed By: mehdi_amini, tejohnson, jdoerfert

Differential Revision: https://reviews.llvm.org/D85379
2020-08-07 11:36:30 -07:00

247 lines
7.9 KiB
C++

//===- LLVMContextImpl.cpp - Implement LLVMContextImpl --------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements the opaque LLVMContextImpl.
//
//===----------------------------------------------------------------------===//
#include "LLVMContextImpl.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/OptBisect.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/ManagedStatic.h"
#include <cassert>
#include <utility>
using namespace llvm;
LLVMContextImpl::LLVMContextImpl(LLVMContext &C)
: DiagHandler(std::make_unique<DiagnosticHandler>()),
VoidTy(C, Type::VoidTyID),
LabelTy(C, Type::LabelTyID),
HalfTy(C, Type::HalfTyID),
BFloatTy(C, Type::BFloatTyID),
FloatTy(C, Type::FloatTyID),
DoubleTy(C, Type::DoubleTyID),
MetadataTy(C, Type::MetadataTyID),
TokenTy(C, Type::TokenTyID),
X86_FP80Ty(C, Type::X86_FP80TyID),
FP128Ty(C, Type::FP128TyID),
PPC_FP128Ty(C, Type::PPC_FP128TyID),
X86_MMXTy(C, Type::X86_MMXTyID),
Int1Ty(C, 1),
Int8Ty(C, 8),
Int16Ty(C, 16),
Int32Ty(C, 32),
Int64Ty(C, 64),
Int128Ty(C, 128) {}
LLVMContextImpl::~LLVMContextImpl() {
// NOTE: We need to delete the contents of OwnedModules, but Module's dtor
// will call LLVMContextImpl::removeModule, thus invalidating iterators into
// the container. Avoid iterators during this operation:
while (!OwnedModules.empty())
delete *OwnedModules.begin();
#ifndef NDEBUG
// Check for metadata references from leaked Instructions.
for (auto &Pair : InstructionMetadata)
Pair.first->dump();
assert(InstructionMetadata.empty() &&
"Instructions with metadata have been leaked");
#endif
// Drop references for MDNodes. Do this before Values get deleted to avoid
// unnecessary RAUW when nodes are still unresolved.
for (auto *I : DistinctMDNodes)
I->dropAllReferences();
#define HANDLE_MDNODE_LEAF_UNIQUABLE(CLASS) \
for (auto *I : CLASS##s) \
I->dropAllReferences();
#include "llvm/IR/Metadata.def"
// Also drop references that come from the Value bridges.
for (auto &Pair : ValuesAsMetadata)
Pair.second->dropUsers();
for (auto &Pair : MetadataAsValues)
Pair.second->dropUse();
// Destroy MDNodes.
for (MDNode *I : DistinctMDNodes)
I->deleteAsSubclass();
#define HANDLE_MDNODE_LEAF_UNIQUABLE(CLASS) \
for (CLASS * I : CLASS##s) \
delete I;
#include "llvm/IR/Metadata.def"
// Free the constants.
for (auto *I : ExprConstants)
I->dropAllReferences();
for (auto *I : ArrayConstants)
I->dropAllReferences();
for (auto *I : StructConstants)
I->dropAllReferences();
for (auto *I : VectorConstants)
I->dropAllReferences();
ExprConstants.freeConstants();
ArrayConstants.freeConstants();
StructConstants.freeConstants();
VectorConstants.freeConstants();
InlineAsms.freeConstants();
CAZConstants.clear();
CPNConstants.clear();
UVConstants.clear();
IntConstants.clear();
FPConstants.clear();
for (auto &CDSConstant : CDSConstants)
delete CDSConstant.second;
CDSConstants.clear();
// Destroy attribute node lists.
for (FoldingSetIterator<AttributeSetNode> I = AttrsSetNodes.begin(),
E = AttrsSetNodes.end(); I != E; ) {
FoldingSetIterator<AttributeSetNode> Elem = I++;
delete &*Elem;
}
// Destroy MetadataAsValues.
{
SmallVector<MetadataAsValue *, 8> MDVs;
MDVs.reserve(MetadataAsValues.size());
for (auto &Pair : MetadataAsValues)
MDVs.push_back(Pair.second);
MetadataAsValues.clear();
for (auto *V : MDVs)
delete V;
}
// Destroy ValuesAsMetadata.
for (auto &Pair : ValuesAsMetadata)
delete Pair.second;
}
void LLVMContextImpl::dropTriviallyDeadConstantArrays() {
SmallSetVector<ConstantArray *, 4> WorkList;
// When ArrayConstants are of substantial size and only a few in them are
// dead, starting WorkList with all elements of ArrayConstants can be
// wasteful. Instead, starting WorkList with only elements that have empty
// uses.
for (ConstantArray *C : ArrayConstants)
if (C->use_empty())
WorkList.insert(C);
while (!WorkList.empty()) {
ConstantArray *C = WorkList.pop_back_val();
if (C->use_empty()) {
for (const Use &Op : C->operands()) {
if (auto *COp = dyn_cast<ConstantArray>(Op))
WorkList.insert(COp);
}
C->destroyConstant();
}
}
}
void Module::dropTriviallyDeadConstantArrays() {
Context.pImpl->dropTriviallyDeadConstantArrays();
}
namespace llvm {
/// Make MDOperand transparent for hashing.
///
/// This overload of an implementation detail of the hashing library makes
/// MDOperand hash to the same value as a \a Metadata pointer.
///
/// Note that overloading \a hash_value() as follows:
///
/// \code
/// size_t hash_value(const MDOperand &X) { return hash_value(X.get()); }
/// \endcode
///
/// does not cause MDOperand to be transparent. In particular, a bare pointer
/// doesn't get hashed before it's combined, whereas \a MDOperand would.
static const Metadata *get_hashable_data(const MDOperand &X) { return X.get(); }
} // end namespace llvm
unsigned MDNodeOpsKey::calculateHash(MDNode *N, unsigned Offset) {
unsigned Hash = hash_combine_range(N->op_begin() + Offset, N->op_end());
#ifndef NDEBUG
{
SmallVector<Metadata *, 8> MDs(N->op_begin() + Offset, N->op_end());
unsigned RawHash = calculateHash(MDs);
assert(Hash == RawHash &&
"Expected hash of MDOperand to equal hash of Metadata*");
}
#endif
return Hash;
}
unsigned MDNodeOpsKey::calculateHash(ArrayRef<Metadata *> Ops) {
return hash_combine_range(Ops.begin(), Ops.end());
}
StringMapEntry<uint32_t> *LLVMContextImpl::getOrInsertBundleTag(StringRef Tag) {
uint32_t NewIdx = BundleTagCache.size();
return &*(BundleTagCache.insert(std::make_pair(Tag, NewIdx)).first);
}
void LLVMContextImpl::getOperandBundleTags(SmallVectorImpl<StringRef> &Tags) const {
Tags.resize(BundleTagCache.size());
for (const auto &T : BundleTagCache)
Tags[T.second] = T.first();
}
uint32_t LLVMContextImpl::getOperandBundleTagID(StringRef Tag) const {
auto I = BundleTagCache.find(Tag);
assert(I != BundleTagCache.end() && "Unknown tag!");
return I->second;
}
SyncScope::ID LLVMContextImpl::getOrInsertSyncScopeID(StringRef SSN) {
auto NewSSID = SSC.size();
assert(NewSSID < std::numeric_limits<SyncScope::ID>::max() &&
"Hit the maximum number of synchronization scopes allowed!");
return SSC.insert(std::make_pair(SSN, SyncScope::ID(NewSSID))).first->second;
}
void LLVMContextImpl::getSyncScopeNames(
SmallVectorImpl<StringRef> &SSNs) const {
SSNs.resize(SSC.size());
for (const auto &SSE : SSC)
SSNs[SSE.second] = SSE.first();
}
/// Singleton instance of the OptBisect class.
///
/// This singleton is accessed via the LLVMContext::getOptPassGate() function.
/// It provides a mechanism to disable passes and individual optimizations at
/// compile time based on a command line option (-opt-bisect-limit) in order to
/// perform a bisecting search for optimization-related problems.
///
/// Even if multiple LLVMContext objects are created, they will all return the
/// same instance of OptBisect in order to provide a single bisect count. Any
/// code that uses the OptBisect object should be serialized when bisection is
/// enabled in order to enable a consistent bisect count.
static ManagedStatic<OptBisect> OptBisector;
OptPassGate &LLVMContextImpl::getOptPassGate() const {
if (!OPG)
OPG = &(*OptBisector);
return *OPG;
}
void LLVMContextImpl::setOptPassGate(OptPassGate& OPG) {
this->OPG = &OPG;
}