llvm-project/llvm/lib/Bitcode/Reader/ValueList.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

226 lines
7.1 KiB
C++
Raw Normal View History

//===- ValueList.cpp - Internal BitcodeReader implementation --------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
#include "ValueList.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/IR/ValueHandle.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/ErrorHandling.h"
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <limits>
#include <utility>
using namespace llvm;
namespace llvm {
namespace {
/// A class for maintaining the slot number definition
/// as a placeholder for the actual definition for forward constants defs.
class ConstantPlaceHolder : public ConstantExpr {
public:
explicit ConstantPlaceHolder(Type *Ty, LLVMContext &Context)
: ConstantExpr(Ty, Instruction::UserOp1, &Op<0>(), 1) {
Op<0>() = UndefValue::get(Type::getInt32Ty(Context));
}
ConstantPlaceHolder &operator=(const ConstantPlaceHolder &) = delete;
// allocate space for exactly one operand
void *operator new(size_t s) { return User::operator new(s, 1); }
/// Methods to support type inquiry through isa, cast, and dyn_cast.
static bool classof(const Value *V) {
return isa<ConstantExpr>(V) &&
cast<ConstantExpr>(V)->getOpcode() == Instruction::UserOp1;
}
/// Provide fast operand accessors
DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value);
};
} // end anonymous namespace
// FIXME: can we inherit this from ConstantExpr?
template <>
struct OperandTraits<ConstantPlaceHolder>
: public FixedNumOperandTraits<ConstantPlaceHolder, 1> {};
DEFINE_TRANSPARENT_OPERAND_ACCESSORS(ConstantPlaceHolder, Value)
} // end namespace llvm
void BitcodeReaderValueList::assignValue(Value *V, unsigned Idx, Type *FullTy) {
if (Idx == size()) {
push_back(V, FullTy);
return;
}
if (Idx >= size())
resize(Idx + 1);
assert(FullTypes[Idx] == nullptr || FullTypes[Idx] == FullTy);
FullTypes[Idx] = FullTy;
WeakTrackingVH &OldV = ValuePtrs[Idx];
if (!OldV) {
OldV = V;
return;
}
// Handle constants and non-constants (e.g. instrs) differently for
// efficiency.
if (Constant *PHC = dyn_cast<Constant>(&*OldV)) {
ResolveConstants.push_back(std::make_pair(PHC, Idx));
OldV = V;
} else {
// If there was a forward reference to this value, replace it.
Value *PrevVal = OldV;
OldV->replaceAllUsesWith(V);
[IR] De-virtualize ~Value to save a vptr Summary: Implements PR889 Removing the virtual table pointer from Value saves 1% of RSS when doing LTO of llc on Linux. The impact on time was positive, but too noisy to conclusively say that performance improved. Here is a link to the spreadsheet with the original data: https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing This change makes it invalid to directly delete a Value, User, or Instruction pointer. Instead, such code can be rewritten to a null check and a call Value::deleteValue(). Value objects tend to have their lifetimes managed through iplist, so for the most part, this isn't a big deal. However, there are some places where LLVM deletes values, and those places had to be migrated to deleteValue. I have also created llvm::unique_value, which has a custom deleter, so it can be used in place of std::unique_ptr<Value>. I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which derives from User outside of lib/IR. Code in IR cannot include MemorySSA headers or call the MemoryAccess object destructors without introducing a circular dependency, so we need some level of indirection. Unfortunately, no class derived from User may have any virtual methods, because adding a virtual method would break User::getHungOffOperands(), which assumes that it can find the use list immediately prior to the User object. I've added a static_assert to the appropriate OperandTraits templates to help people avoid this trap. Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv Reviewed By: chandlerc Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits Differential Revision: https://reviews.llvm.org/D31261 llvm-svn: 303362
2017-05-19 01:24:10 +08:00
PrevVal->deleteValue();
}
}
Constant *BitcodeReaderValueList::getConstantFwdRef(unsigned Idx, Type *Ty) {
// Bail out for a clearly invalid value.
if (Idx >= RefsUpperBound)
return nullptr;
if (Idx >= size())
resize(Idx + 1);
if (Value *V = ValuePtrs[Idx]) {
if (Ty != V->getType())
report_fatal_error("Type mismatch in constant table!");
return cast<Constant>(V);
}
// Create and return a placeholder, which will later be RAUW'd.
Constant *C = new ConstantPlaceHolder(Ty, Context);
ValuePtrs[Idx] = C;
return C;
}
Value *BitcodeReaderValueList::getValueFwdRef(unsigned Idx, Type *Ty,
Type **FullTy) {
// Bail out for a clearly invalid value.
if (Idx >= RefsUpperBound)
return nullptr;
if (Idx >= size())
resize(Idx + 1);
if (Value *V = ValuePtrs[Idx]) {
// If the types don't match, it's invalid.
if (Ty && Ty != V->getType())
return nullptr;
if (FullTy)
*FullTy = FullTypes[Idx];
return V;
}
// No type specified, must be invalid reference.
if (!Ty)
return nullptr;
// Create and return a placeholder, which will later be RAUW'd.
Value *V = new Argument(Ty);
ValuePtrs[Idx] = V;
return V;
}
/// Once all constants are read, this method bulk resolves any forward
/// references. The idea behind this is that we sometimes get constants (such
/// as large arrays) which reference *many* forward ref constants. Replacing
/// each of these causes a lot of thrashing when building/reuniquing the
/// constant. Instead of doing this, we look at all the uses and rewrite all
/// the place holders at once for any constant that uses a placeholder.
void BitcodeReaderValueList::resolveConstantForwardRefs() {
// Sort the values by-pointer so that they are efficient to look up with a
// binary search.
llvm::sort(ResolveConstants);
SmallVector<Constant *, 64> NewOps;
while (!ResolveConstants.empty()) {
Value *RealVal = operator[](ResolveConstants.back().second);
Constant *Placeholder = ResolveConstants.back().first;
ResolveConstants.pop_back();
// Loop over all users of the placeholder, updating them to reference the
// new value. If they reference more than one placeholder, update them all
// at once.
while (!Placeholder->use_empty()) {
auto UI = Placeholder->user_begin();
User *U = *UI;
// If the using object isn't uniqued, just update the operands. This
// handles instructions and initializers for global variables.
if (!isa<Constant>(U) || isa<GlobalValue>(U)) {
UI.getUse().set(RealVal);
continue;
}
// Otherwise, we have a constant that uses the placeholder. Replace that
// constant with a new constant that has *all* placeholder uses updated.
Constant *UserC = cast<Constant>(U);
for (User::op_iterator I = UserC->op_begin(), E = UserC->op_end(); I != E;
++I) {
Value *NewOp;
if (!isa<ConstantPlaceHolder>(*I)) {
// Not a placeholder reference.
NewOp = *I;
} else if (*I == Placeholder) {
// Common case is that it just references this one placeholder.
NewOp = RealVal;
} else {
// Otherwise, look up the placeholder in ResolveConstants.
ResolveConstantsTy::iterator It = llvm::lower_bound(
ResolveConstants,
std::pair<Constant *, unsigned>(cast<Constant>(*I), 0));
assert(It != ResolveConstants.end() && It->first == *I);
NewOp = operator[](It->second);
}
NewOps.push_back(cast<Constant>(NewOp));
}
// Make the new constant.
Constant *NewC;
if (ConstantArray *UserCA = dyn_cast<ConstantArray>(UserC)) {
NewC = ConstantArray::get(UserCA->getType(), NewOps);
} else if (ConstantStruct *UserCS = dyn_cast<ConstantStruct>(UserC)) {
NewC = ConstantStruct::get(UserCS->getType(), NewOps);
} else if (isa<ConstantVector>(UserC)) {
NewC = ConstantVector::get(NewOps);
} else {
assert(isa<ConstantExpr>(UserC) && "Must be a ConstantExpr.");
NewC = cast<ConstantExpr>(UserC)->getWithOperands(NewOps);
}
UserC->replaceAllUsesWith(NewC);
UserC->destroyConstant();
NewOps.clear();
}
// Update all ValueHandles, they should be the only users at this point.
Placeholder->replaceAllUsesWith(RealVal);
[IR] De-virtualize ~Value to save a vptr Summary: Implements PR889 Removing the virtual table pointer from Value saves 1% of RSS when doing LTO of llc on Linux. The impact on time was positive, but too noisy to conclusively say that performance improved. Here is a link to the spreadsheet with the original data: https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing This change makes it invalid to directly delete a Value, User, or Instruction pointer. Instead, such code can be rewritten to a null check and a call Value::deleteValue(). Value objects tend to have their lifetimes managed through iplist, so for the most part, this isn't a big deal. However, there are some places where LLVM deletes values, and those places had to be migrated to deleteValue. I have also created llvm::unique_value, which has a custom deleter, so it can be used in place of std::unique_ptr<Value>. I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which derives from User outside of lib/IR. Code in IR cannot include MemorySSA headers or call the MemoryAccess object destructors without introducing a circular dependency, so we need some level of indirection. Unfortunately, no class derived from User may have any virtual methods, because adding a virtual method would break User::getHungOffOperands(), which assumes that it can find the use list immediately prior to the User object. I've added a static_assert to the appropriate OperandTraits templates to help people avoid this trap. Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv Reviewed By: chandlerc Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits Differential Revision: https://reviews.llvm.org/D31261 llvm-svn: 303362
2017-05-19 01:24:10 +08:00
Placeholder->deleteValue();
}
}