llvm-project/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp

501 lines
17 KiB
C++
Raw Normal View History

//===- DeadStoreElimination.cpp - Fast Dead Store Elimination -------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements a trivial dead store elimination that only considers
// basic-block local redundant stores.
//
// FIXME: This should eventually be extended to be a post-dominator tree
// traversal. Doing so would be pretty trivial.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "dse"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Pass.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Support/Compiler.h"
using namespace llvm;
STATISTIC(NumFastStores, "Number of stores deleted");
STATISTIC(NumFastOther , "Number of other instrs removed");
namespace {
struct VISIBILITY_HIDDEN DSE : public FunctionPass {
static char ID; // Pass identification, replacement for typeid
DSE() : FunctionPass(&ID) {}
virtual bool runOnFunction(Function &F) {
bool Changed = false;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
Changed |= runOnBasicBlock(*I);
return Changed;
}
bool runOnBasicBlock(BasicBlock &BB);
bool handleFreeWithNonTrivialDependency(FreeInst *F, Instruction *Dep);
bool handleEndBlock(BasicBlock &BB);
bool RemoveUndeadPointers(Value* pointer, uint64_t killPointerSize,
BasicBlock::iterator& BBI,
SmallPtrSet<Value*, 64>& deadPointers);
void DeleteDeadInstruction(Instruction *I,
SmallPtrSet<Value*, 64> *deadPointers = 0);
// getAnalysisUsage - We require post dominance frontiers (aka Control
// Dependence Graph)
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesCFG();
AU.addRequired<DominatorTree>();
AU.addRequired<TargetData>();
AU.addRequired<AliasAnalysis>();
AU.addRequired<MemoryDependenceAnalysis>();
AU.addPreserved<DominatorTree>();
AU.addPreserved<AliasAnalysis>();
AU.addPreserved<MemoryDependenceAnalysis>();
}
};
}
char DSE::ID = 0;
static RegisterPass<DSE> X("dse", "Dead Store Elimination");
FunctionPass *llvm::createDeadStoreEliminationPass() { return new DSE(); }
bool DSE::runOnBasicBlock(BasicBlock &BB) {
MemoryDependenceAnalysis& MD = getAnalysis<MemoryDependenceAnalysis>();
TargetData &TD = getAnalysis<TargetData>();
// Record the last-seen store to this pointer
DenseMap<Value*, StoreInst*> lastStore;
bool MadeChange = false;
// Do a top-down walk on the BB
for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) {
Instruction *Inst = BBI++;
// If we find a store or a free...
if (!isa<StoreInst>(Inst) && !isa<FreeInst>(Inst))
continue;
Value* pointer = 0;
if (StoreInst* S = dyn_cast<StoreInst>(Inst)) {
if (S->isVolatile())
continue;
pointer = S->getPointerOperand();
} else {
pointer = cast<FreeInst>(Inst)->getPointerOperand();
}
pointer = pointer->stripPointerCasts();
StoreInst *&last = lastStore[pointer];
// ... to a pointer that has been stored to before...
if (last) {
Instruction* dep = MD.getDependency(Inst);
bool deletedStore = false;
// ... and no other memory dependencies are between them....
while (dep != MemoryDependenceAnalysis::None &&
dep != MemoryDependenceAnalysis::NonLocal &&
isa<StoreInst>(dep)) {
if (dep != last ||
Executive summary: getTypeSize -> getTypeStoreSize / getABITypeSize. The meaning of getTypeSize was not clear - clarifying it is important now that we have x86 long double and arbitrary precision integers. The issue with long double is that it requires 80 bits, and this is not a multiple of its alignment. This gives a primitive type for which getTypeSize differed from getABITypeSize. For arbitrary precision integers it is even worse: there is the minimum number of bits needed to hold the type (eg: 36 for an i36), the maximum number of bits that will be overwriten when storing the type (40 bits for i36) and the ABI size (i.e. the storage size rounded up to a multiple of the alignment; 64 bits for i36). This patch removes getTypeSize (not really - it is still there but deprecated to allow for a gradual transition). Instead there is: (1) getTypeSizeInBits - a number of bits that suffices to hold all values of the type. For a primitive type, this is the minimum number of bits. For an i36 this is 36 bits. For x86 long double it is 80. This corresponds to gcc's TYPE_PRECISION. (2) getTypeStoreSizeInBits - the maximum number of bits that is written when storing the type (or read when reading it). For an i36 this is 40 bits, for an x86 long double it is 80 bits. This is the size alias analysis is interested in (getTypeStoreSize returns the number of bytes). There doesn't seem to be anything corresponding to this in gcc. (3) getABITypeSizeInBits - this is getTypeStoreSizeInBits rounded up to a multiple of the alignment. For an i36 this is 64, for an x86 long double this is 96 or 128 depending on the OS. This is the spacing between consecutive elements when you form an array out of this type (getABITypeSize returns the number of bytes). This is TYPE_SIZE in gcc. Since successive elements in a SequentialType (arrays, pointers and vectors) need to be aligned, the spacing between them will be given by getABITypeSize. This means that the size of an array is the length times the getABITypeSize. It also means that GEP computations need to use getABITypeSize when computing offsets. Furthermore, if an alloca allocates several elements at once then these too need to be aligned, so the size of the alloca has to be the number of elements multiplied by getABITypeSize. Logically speaking this doesn't have to be the case when allocating just one element, but it is simpler to also use getABITypeSize in this case. So alloca's and mallocs should use getABITypeSize. Finally, since gcc's only notion of size is that given by getABITypeSize, if you want to output assembler etc the same as gcc then getABITypeSize is the size you want. Since a store will overwrite no more than getTypeStoreSize bytes, and a read will read no more than that many bytes, this is the notion of size appropriate for alias analysis calculations. In this patch I have corrected all type size uses except some of those in ScalarReplAggregates, lib/Codegen, lib/Target (the hard cases). I will get around to auditing these too at some point, but I could do with some help. Finally, I made one change which I think wise but others might consider pointless and suboptimal: in an unpacked struct the amount of space allocated for a field is now given by the ABI size rather than getTypeStoreSize. I did this because every other place that reserves memory for a type (eg: alloca) now uses getABITypeSize, and I didn't want to make an exception for unpacked structs, i.e. I did it to make things more uniform. This only effects structs containing long doubles and arbitrary precision integers. If someone wants to pack these types more tightly they can always use a packed struct. llvm-svn: 43620
2007-11-02 04:53:16 +08:00
TD.getTypeStoreSize(last->getOperand(0)->getType()) >
TD.getTypeStoreSize(Inst->getOperand(0)->getType())) {
dep = MD.getDependency(Inst, dep);
continue;
}
// Delete the store and now-dead instructions that feed it.
DeleteDeadInstruction(last);
NumFastStores++;
deletedStore = true;
MadeChange = true;
break;
}
// If we deleted a store, reinvestigate this instruction.
if (deletedStore) {
--BBI;
continue;
}
}
// Handle frees whose dependencies are non-trivial.
if (FreeInst* F = dyn_cast<FreeInst>(Inst)) {
MadeChange |= handleFreeWithNonTrivialDependency(F, MD.getDependency(F));
// No known stores after the free.
last = 0;
} else {
StoreInst* S = cast<StoreInst>(Inst);
// If we're storing the same value back to a pointer that we just
// loaded from, then the store can be removed;
if (LoadInst* L = dyn_cast<LoadInst>(S->getOperand(0))) {
// FIXME: Don't do dep query if Parents don't match and other stuff!
Instruction* dep = MD.getDependency(S);
DominatorTree& DT = getAnalysis<DominatorTree>();
if (!S->isVolatile() && S->getParent() == L->getParent() &&
S->getPointerOperand() == L->getPointerOperand() &&
(dep == MemoryDependenceAnalysis::None ||
dep == MemoryDependenceAnalysis::NonLocal ||
DT.dominates(dep, L))) {
DeleteDeadInstruction(S);
--BBI;
NumFastStores++;
MadeChange = true;
} else
// Update our most-recent-store map.
last = S;
} else
// Update our most-recent-store map.
last = S;
}
}
// If this block ends in a return, unwind, or unreachable, all allocas are
// dead at its end, which means stores to them are also dead.
if (BB.getTerminator()->getNumSuccessors() == 0)
MadeChange |= handleEndBlock(BB);
return MadeChange;
}
/// handleFreeWithNonTrivialDependency - Handle frees of entire structures whose
/// dependency is a store to a field of that structure.
bool DSE::handleFreeWithNonTrivialDependency(FreeInst* F, Instruction* dep) {
TargetData &TD = getAnalysis<TargetData>();
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
if (dep == MemoryDependenceAnalysis::None ||
dep == MemoryDependenceAnalysis::NonLocal)
return false;
StoreInst* dependency = dyn_cast<StoreInst>(dep);
if (!dependency)
return false;
else if (dependency->isVolatile())
return false;
Value* depPointer = dependency->getPointerOperand();
const Type* depType = dependency->getOperand(0)->getType();
Executive summary: getTypeSize -> getTypeStoreSize / getABITypeSize. The meaning of getTypeSize was not clear - clarifying it is important now that we have x86 long double and arbitrary precision integers. The issue with long double is that it requires 80 bits, and this is not a multiple of its alignment. This gives a primitive type for which getTypeSize differed from getABITypeSize. For arbitrary precision integers it is even worse: there is the minimum number of bits needed to hold the type (eg: 36 for an i36), the maximum number of bits that will be overwriten when storing the type (40 bits for i36) and the ABI size (i.e. the storage size rounded up to a multiple of the alignment; 64 bits for i36). This patch removes getTypeSize (not really - it is still there but deprecated to allow for a gradual transition). Instead there is: (1) getTypeSizeInBits - a number of bits that suffices to hold all values of the type. For a primitive type, this is the minimum number of bits. For an i36 this is 36 bits. For x86 long double it is 80. This corresponds to gcc's TYPE_PRECISION. (2) getTypeStoreSizeInBits - the maximum number of bits that is written when storing the type (or read when reading it). For an i36 this is 40 bits, for an x86 long double it is 80 bits. This is the size alias analysis is interested in (getTypeStoreSize returns the number of bytes). There doesn't seem to be anything corresponding to this in gcc. (3) getABITypeSizeInBits - this is getTypeStoreSizeInBits rounded up to a multiple of the alignment. For an i36 this is 64, for an x86 long double this is 96 or 128 depending on the OS. This is the spacing between consecutive elements when you form an array out of this type (getABITypeSize returns the number of bytes). This is TYPE_SIZE in gcc. Since successive elements in a SequentialType (arrays, pointers and vectors) need to be aligned, the spacing between them will be given by getABITypeSize. This means that the size of an array is the length times the getABITypeSize. It also means that GEP computations need to use getABITypeSize when computing offsets. Furthermore, if an alloca allocates several elements at once then these too need to be aligned, so the size of the alloca has to be the number of elements multiplied by getABITypeSize. Logically speaking this doesn't have to be the case when allocating just one element, but it is simpler to also use getABITypeSize in this case. So alloca's and mallocs should use getABITypeSize. Finally, since gcc's only notion of size is that given by getABITypeSize, if you want to output assembler etc the same as gcc then getABITypeSize is the size you want. Since a store will overwrite no more than getTypeStoreSize bytes, and a read will read no more than that many bytes, this is the notion of size appropriate for alias analysis calculations. In this patch I have corrected all type size uses except some of those in ScalarReplAggregates, lib/Codegen, lib/Target (the hard cases). I will get around to auditing these too at some point, but I could do with some help. Finally, I made one change which I think wise but others might consider pointless and suboptimal: in an unpacked struct the amount of space allocated for a field is now given by the ABI size rather than getTypeStoreSize. I did this because every other place that reserves memory for a type (eg: alloca) now uses getABITypeSize, and I didn't want to make an exception for unpacked structs, i.e. I did it to make things more uniform. This only effects structs containing long doubles and arbitrary precision integers. If someone wants to pack these types more tightly they can always use a packed struct. llvm-svn: 43620
2007-11-02 04:53:16 +08:00
unsigned depPointerSize = TD.getTypeStoreSize(depType);
// Check for aliasing
AliasAnalysis::AliasResult A = AA.alias(F->getPointerOperand(), ~0U,
depPointer, depPointerSize);
if (A != AliasAnalysis::MustAlias)
return false;
// DCE instructions only used to calculate that store
DeleteDeadInstruction(dependency);
NumFastStores++;
return true;
}
/// handleEndBlock - Remove dead stores to stack-allocated locations in the
/// function end block. Ex:
/// %A = alloca i32
/// ...
/// store i32 1, i32* %A
/// ret void
bool DSE::handleEndBlock(BasicBlock &BB) {
TargetData &TD = getAnalysis<TargetData>();
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
bool MadeChange = false;
// Pointers alloca'd in this function are dead in the end block
SmallPtrSet<Value*, 64> deadPointers;
// Find all of the alloca'd pointers in the entry block.
BasicBlock *Entry = BB.getParent()->begin();
for (BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I)
if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
deadPointers.insert(AI);
// Treat byval arguments the same, stores to them are dead at the end of the
// function.
for (Function::arg_iterator AI = BB.getParent()->arg_begin(),
AE = BB.getParent()->arg_end(); AI != AE; ++AI)
if (AI->hasByValAttr())
deadPointers.insert(AI);
// Scan the basic block backwards
for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){
--BBI;
// If we find a store whose pointer is dead.
if (StoreInst* S = dyn_cast<StoreInst>(BBI)) {
if (!S->isVolatile()) {
// See through pointer-to-pointer bitcasts
Value* pointerOperand = S->getPointerOperand()->getUnderlyingObject();
// Alloca'd pointers or byval arguments (which are functionally like
// alloca's) are valid candidates for removal.
if (deadPointers.count(pointerOperand)) {
// DCE instructions only used to calculate that store.
BBI++;
DeleteDeadInstruction(S, &deadPointers);
NumFastStores++;
MadeChange = true;
}
}
continue;
}
// We can also remove memcpy's to local variables at the end of a function.
if (MemCpyInst *M = dyn_cast<MemCpyInst>(BBI)) {
Value *dest = M->getDest()->getUnderlyingObject();
if (deadPointers.count(dest)) {
BBI++;
DeleteDeadInstruction(M, &deadPointers);
NumFastOther++;
MadeChange = true;
continue;
}
// Because a memcpy is also a load, we can't skip it if we didn't remove
// it.
}
Value* killPointer = 0;
uint64_t killPointerSize = ~0UL;
// If we encounter a use of the pointer, it is no longer considered dead
if (LoadInst *L = dyn_cast<LoadInst>(BBI)) {
2008-05-13 09:48:26 +08:00
// However, if this load is unused and not volatile, we can go ahead and
// remove it, and not have to worry about it making our pointer undead!
if (L->use_empty() && !L->isVolatile()) {
BBI++;
DeleteDeadInstruction(L, &deadPointers);
NumFastOther++;
MadeChange = true;
continue;
}
killPointer = L->getPointerOperand();
} else if (VAArgInst* V = dyn_cast<VAArgInst>(BBI)) {
killPointer = V->getOperand(0);
} else if (isa<MemCpyInst>(BBI) &&
isa<ConstantInt>(cast<MemCpyInst>(BBI)->getLength())) {
killPointer = cast<MemCpyInst>(BBI)->getSource();
killPointerSize = cast<ConstantInt>(
cast<MemCpyInst>(BBI)->getLength())->getZExtValue();
} else if (AllocaInst* A = dyn_cast<AllocaInst>(BBI)) {
deadPointers.erase(A);
// Dead alloca's can be DCE'd when we reach them
if (A->use_empty()) {
BBI++;
DeleteDeadInstruction(A, &deadPointers);
NumFastOther++;
MadeChange = true;
}
continue;
} else if (CallSite::get(BBI).getInstruction() != 0) {
// If this call does not access memory, it can't
// be undeadifying any of our pointers.
CallSite CS = CallSite::get(BBI);
if (AA.doesNotAccessMemory(CS))
continue;
unsigned modRef = 0;
unsigned other = 0;
// Remove any pointers made undead by the call from the dead set
std::vector<Value*> dead;
for (SmallPtrSet<Value*, 64>::iterator I = deadPointers.begin(),
E = deadPointers.end(); I != E; ++I) {
// HACK: if we detect that our AA is imprecise, it's not
// worth it to scan the rest of the deadPointers set. Just
// assume that the AA will return ModRef for everything, and
// go ahead and bail.
if (modRef >= 16 && other == 0) {
deadPointers.clear();
return MadeChange;
}
// Get size information for the alloca
unsigned pointerSize = ~0U;
if (AllocaInst* A = dyn_cast<AllocaInst>(*I)) {
if (ConstantInt* C = dyn_cast<ConstantInt>(A->getArraySize()))
pointerSize = C->getZExtValue() *
TD.getABITypeSize(A->getAllocatedType());
} else {
const PointerType* PT = cast<PointerType>(
cast<Argument>(*I)->getType());
pointerSize = TD.getABITypeSize(PT->getElementType());
}
// See if the call site touches it
AliasAnalysis::ModRefResult A = AA.getModRefInfo(CS, *I, pointerSize);
if (A == AliasAnalysis::ModRef)
modRef++;
else
other++;
if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref)
dead.push_back(*I);
}
for (std::vector<Value*>::iterator I = dead.begin(), E = dead.end();
I != E; ++I)
deadPointers.erase(*I);
continue;
} else if (isInstructionTriviallyDead(BBI)) {
// For any non-memory-affecting non-terminators, DCE them as we reach them
Instruction *Inst = BBI;
BBI++;
DeleteDeadInstruction(Inst, &deadPointers);
NumFastOther++;
MadeChange = true;
continue;
}
if (!killPointer)
continue;
killPointer = killPointer->getUnderlyingObject();
// Deal with undead pointers
MadeChange |= RemoveUndeadPointers(killPointer, killPointerSize, BBI,
deadPointers);
}
return MadeChange;
}
/// RemoveUndeadPointers - check for uses of a pointer that make it
/// undead when scanning for dead stores to alloca's.
bool DSE::RemoveUndeadPointers(Value* killPointer, uint64_t killPointerSize,
BasicBlock::iterator &BBI,
SmallPtrSet<Value*, 64>& deadPointers) {
TargetData &TD = getAnalysis<TargetData>();
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
// If the kill pointer can be easily reduced to an alloca,
// don't bother doing extraneous AA queries.
if (deadPointers.count(killPointer)) {
deadPointers.erase(killPointer);
return false;
}
// A global can't be in the dead pointer set.
if (isa<GlobalValue>(killPointer))
return false;
bool MadeChange = false;
SmallVector<Value*, 16> undead;
for (SmallPtrSet<Value*, 64>::iterator I = deadPointers.begin(),
E = deadPointers.end(); I != E; ++I) {
// Get size information for the alloca.
unsigned pointerSize = ~0U;
if (AllocaInst* A = dyn_cast<AllocaInst>(*I)) {
if (ConstantInt* C = dyn_cast<ConstantInt>(A->getArraySize()))
pointerSize = C->getZExtValue() *
TD.getABITypeSize(A->getAllocatedType());
} else {
const PointerType* PT = cast<PointerType>(cast<Argument>(*I)->getType());
pointerSize = TD.getABITypeSize(PT->getElementType());
}
// See if this pointer could alias it
AliasAnalysis::AliasResult A = AA.alias(*I, pointerSize,
killPointer, killPointerSize);
// If it must-alias and a store, we can delete it
if (isa<StoreInst>(BBI) && A == AliasAnalysis::MustAlias) {
StoreInst* S = cast<StoreInst>(BBI);
// Remove it!
BBI++;
DeleteDeadInstruction(S, &deadPointers);
NumFastStores++;
MadeChange = true;
continue;
// Otherwise, it is undead
} else if (A != AliasAnalysis::NoAlias)
undead.push_back(*I);
}
for (SmallVector<Value*, 16>::iterator I = undead.begin(), E = undead.end();
I != E; ++I)
deadPointers.erase(*I);
return MadeChange;
}
/// DeleteDeadInstruction - Delete this instruction. Before we do, go through
/// and zero out all the operands of this instruction. If any of them become
/// dead, delete them and the computation tree that feeds them.
///
/// If ValueSet is non-null, remove any deleted instructions from it as well.
///
void DSE::DeleteDeadInstruction(Instruction *I,
SmallPtrSet<Value*, 64> *ValueSet) {
SmallVector<Instruction*, 32> NowDeadInsts;
NowDeadInsts.push_back(I);
--NumFastOther;
// Before we touch this instruction, remove it from memdep!
MemoryDependenceAnalysis &MDA = getAnalysis<MemoryDependenceAnalysis>();
while (!NowDeadInsts.empty()) {
Instruction *DeadInst = NowDeadInsts.back();
NowDeadInsts.pop_back();
++NumFastOther;
// This instruction is dead, zap it, in stages. Start by removing it from
// MemDep, which needs to know the operands and needs it to be in the
// function.
MDA.removeInstruction(DeadInst);
for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) {
Value *Op = DeadInst->getOperand(op);
DeadInst->setOperand(op, 0);
// If this operand just became dead, add it to the NowDeadInsts list.
if (!Op->use_empty()) continue;
if (Instruction *OpI = dyn_cast<Instruction>(Op))
if (isInstructionTriviallyDead(OpI))
NowDeadInsts.push_back(OpI);
}
DeadInst->eraseFromParent();
if (ValueSet) ValueSet->erase(DeadInst);
}
}