llvm-6502/lib/Analysis/MemoryDependenceAnalysis.cpp
Chris Lattner 125ce36269 remove a bit of incorrect code that tried to be tricky about speeding up
dependencies.  The basic situation was this: consider if we had:

  store1
  ...
  store2
  ...
  store3

Where memdep thinks that store3 depends on store2 and store2 depends 
on store1.  The problem happens when we delete store2: The code in 
question was updating dep info for store3 to be store1.  This is a
spiffy optimization, but is not safe at all, because aliasing isn't
transitive.  This bug isn't exposed today with DSE because DSE will only
zap store2 if it is identifical to store 3, and in this case, it is 
safe to update it to depend on store1.  However, memcpyopt is not so
fortunate, which is presumably why the "dropInstruction" code used to
exist.

Since this doesn't actually provide a speedup in practice, just rip the
code out.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@60263 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-30 01:09:30 +00:00

465 lines
18 KiB
C++

//===- MemoryDependenceAnalysis.cpp - Mem Deps Implementation --*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements an analysis that determines, for a given memory
// operation, what preceding memory operations it depends on. It builds on
// alias analysis information, and tries to provide a lazy, caching interface to
// a common kind of alias information query.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "memdep"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Constants.h"
#include "llvm/Instructions.h"
#include "llvm/Function.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/CFG.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/TargetData.h"
using namespace llvm;
STATISTIC(NumCacheNonLocal, "Number of cached non-local responses");
STATISTIC(NumUncacheNonLocal, "Number of uncached non-local responses");
char MemoryDependenceAnalysis::ID = 0;
// Register this pass...
static RegisterPass<MemoryDependenceAnalysis> X("memdep",
"Memory Dependence Analysis", false, true);
/// getAnalysisUsage - Does not modify anything. It uses Alias Analysis.
///
void MemoryDependenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
AU.addRequiredTransitive<AliasAnalysis>();
AU.addRequiredTransitive<TargetData>();
}
/// getCallSiteDependency - Private helper for finding the local dependencies
/// of a call site.
MemDepResult MemoryDependenceAnalysis::
getCallSiteDependency(CallSite C, BasicBlock::iterator ScanIt,
BasicBlock *BB) {
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
TargetData &TD = getAnalysis<TargetData>();
// Walk backwards through the block, looking for dependencies
while (ScanIt != BB->begin()) {
Instruction *Inst = --ScanIt;
// If this inst is a memory op, get the pointer it accessed
Value *Pointer = 0;
uint64_t PointerSize = 0;
if (StoreInst *S = dyn_cast<StoreInst>(Inst)) {
Pointer = S->getPointerOperand();
PointerSize = TD.getTypeStoreSize(S->getOperand(0)->getType());
} else if (AllocationInst *AI = dyn_cast<AllocationInst>(Inst)) {
Pointer = AI;
if (ConstantInt *C = dyn_cast<ConstantInt>(AI->getArraySize()))
// Use ABI size (size between elements), not store size (size of one
// element without padding).
PointerSize = C->getZExtValue() *
TD.getABITypeSize(AI->getAllocatedType());
else
PointerSize = ~0UL;
} else if (VAArgInst *V = dyn_cast<VAArgInst>(Inst)) {
Pointer = V->getOperand(0);
PointerSize = TD.getTypeStoreSize(V->getType());
} else if (FreeInst *F = dyn_cast<FreeInst>(Inst)) {
Pointer = F->getPointerOperand();
// FreeInsts erase the entire structure
PointerSize = ~0UL;
} else if (isa<CallInst>(Inst) || isa<InvokeInst>(Inst)) {
if (AA.getModRefBehavior(CallSite::get(Inst)) ==
AliasAnalysis::DoesNotAccessMemory)
continue;
return MemDepResult::get(Inst);
} else
continue;
if (AA.getModRefInfo(C, Pointer, PointerSize) != AliasAnalysis::NoModRef)
return MemDepResult::get(Inst);
}
// No dependence found.
return MemDepResult::getNonLocal();
}
/// getNonLocalDependency - Perform a full dependency query for the
/// specified instruction, returning the set of blocks that the value is
/// potentially live across. The returned set of results will include a
/// "NonLocal" result for all blocks where the value is live across.
///
/// This method assumes the instruction returns a "nonlocal" dependency
/// within its own block.
///
void MemoryDependenceAnalysis::
getNonLocalDependency(Instruction *QueryInst,
SmallVectorImpl<std::pair<BasicBlock*,
MemDepResult> > &Result) {
assert(getDependency(QueryInst).isNonLocal() &&
"getNonLocalDependency should only be used on insts with non-local deps!");
DenseMap<BasicBlock*, DepResultTy> &Cache = NonLocalDeps[QueryInst];
/// DirtyBlocks - This is the set of blocks that need to be recomputed. In
/// the cached case, this can happen due to instructions being deleted etc. In
/// the uncached case, this starts out as the set of predecessors we care
/// about.
SmallVector<BasicBlock*, 32> DirtyBlocks;
if (!Cache.empty()) {
// If we already have a partially computed set of results, scan them to
// determine what is dirty, seeding our initial DirtyBlocks worklist.
// FIXME: In the "don't need to be updated" case, this is expensive, why not
// have a per-"cache" flag saying it is undirty?
for (DenseMap<BasicBlock*, DepResultTy>::iterator I = Cache.begin(),
E = Cache.end(); I != E; ++I)
if (I->second.getInt() == Dirty)
DirtyBlocks.push_back(I->first);
NumCacheNonLocal++;
//cerr << "CACHED CASE: " << DirtyBlocks.size() << " dirty: "
// << Cache.size() << " cached: " << *QueryInst;
} else {
// Seed DirtyBlocks with each of the preds of QueryInst's block.
BasicBlock *QueryBB = QueryInst->getParent();
DirtyBlocks.append(pred_begin(QueryBB), pred_end(QueryBB));
NumUncacheNonLocal++;
}
// Iterate while we still have blocks to update.
while (!DirtyBlocks.empty()) {
BasicBlock *DirtyBB = DirtyBlocks.back();
DirtyBlocks.pop_back();
// Get the entry for this block. Note that this relies on DepResultTy
// default initializing to Dirty.
DepResultTy &DirtyBBEntry = Cache[DirtyBB];
// If DirtyBBEntry isn't dirty, it ended up on the worklist multiple times.
if (DirtyBBEntry.getInt() != Dirty) continue;
// Find out if this block has a local dependency for QueryInst.
// FIXME: If the dirty entry has an instruction pointer, scan from it!
// FIXME: Don't convert back and forth for MemDepResult <-> DepResultTy.
// If the dirty entry has a pointer, start scanning from it so we don't have
// to rescan the entire block.
BasicBlock::iterator ScanPos = DirtyBB->end();
if (Instruction *Inst = DirtyBBEntry.getPointer())
ScanPos = Inst;
DirtyBBEntry = ConvFromResult(getDependencyFrom(QueryInst, ScanPos,
DirtyBB));
// If the block has a dependency (i.e. it isn't completely transparent to
// the value), remember it!
if (DirtyBBEntry.getInt() != NonLocal) {
// Keep the ReverseNonLocalDeps map up to date so we can efficiently
// update this when we remove instructions.
if (Instruction *Inst = DirtyBBEntry.getPointer())
ReverseNonLocalDeps[Inst].insert(QueryInst);
continue;
}
// If the block *is* completely transparent to the load, we need to check
// the predecessors of this block. Add them to our worklist.
DirtyBlocks.append(pred_begin(DirtyBB), pred_end(DirtyBB));
}
// Copy the result into the output set.
for (DenseMap<BasicBlock*, DepResultTy>::iterator I = Cache.begin(),
E = Cache.end(); I != E; ++I)
Result.push_back(std::make_pair(I->first, ConvToResult(I->second)));
}
/// getDependency - Return the instruction on which a memory operation
/// depends. The local parameter indicates if the query should only
/// evaluate dependencies within the same basic block.
MemDepResult MemoryDependenceAnalysis::
getDependencyFrom(Instruction *QueryInst, BasicBlock::iterator ScanIt,
BasicBlock *BB) {
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
TargetData &TD = getAnalysis<TargetData>();
// Get the pointer value for which dependence will be determined
Value *MemPtr = 0;
uint64_t MemSize = 0;
bool MemVolatile = false;
if (StoreInst* S = dyn_cast<StoreInst>(QueryInst)) {
MemPtr = S->getPointerOperand();
MemSize = TD.getTypeStoreSize(S->getOperand(0)->getType());
MemVolatile = S->isVolatile();
} else if (LoadInst* L = dyn_cast<LoadInst>(QueryInst)) {
MemPtr = L->getPointerOperand();
MemSize = TD.getTypeStoreSize(L->getType());
MemVolatile = L->isVolatile();
} else if (VAArgInst* V = dyn_cast<VAArgInst>(QueryInst)) {
MemPtr = V->getOperand(0);
MemSize = TD.getTypeStoreSize(V->getType());
} else if (FreeInst* F = dyn_cast<FreeInst>(QueryInst)) {
MemPtr = F->getPointerOperand();
// FreeInsts erase the entire structure, not just a field.
MemSize = ~0UL;
} else if (isa<CallInst>(QueryInst) || isa<InvokeInst>(QueryInst))
return getCallSiteDependency(CallSite::get(QueryInst), ScanIt, BB);
else // Non-memory instructions depend on nothing.
return MemDepResult::getNone();
// Walk backwards through the basic block, looking for dependencies
while (ScanIt != BB->begin()) {
Instruction *Inst = --ScanIt;
// If the access is volatile and this is a volatile load/store, return a
// dependence.
if (MemVolatile &&
((isa<LoadInst>(Inst) && cast<LoadInst>(Inst)->isVolatile()) ||
(isa<StoreInst>(Inst) && cast<StoreInst>(Inst)->isVolatile())))
return MemDepResult::get(Inst);
// MemDep is broken w.r.t. loads: it says that two loads of the same pointer
// depend on each other. :(
// FIXME: ELIMINATE THIS!
if (LoadInst *L = dyn_cast<LoadInst>(Inst)) {
Value *Pointer = L->getPointerOperand();
uint64_t PointerSize = TD.getTypeStoreSize(L->getType());
// If we found a pointer, check if it could be the same as our pointer
AliasAnalysis::AliasResult R =
AA.alias(Pointer, PointerSize, MemPtr, MemSize);
if (R == AliasAnalysis::NoAlias)
continue;
// May-alias loads don't depend on each other without a dependence.
if (isa<LoadInst>(QueryInst) && R == AliasAnalysis::MayAlias)
continue;
return MemDepResult::get(Inst);
}
// FIXME: This claims that an access depends on the allocation. This may
// make sense, but is dubious at best. It would be better to fix GVN to
// handle a 'None' Query.
if (AllocationInst *AI = dyn_cast<AllocationInst>(Inst)) {
Value *Pointer = AI;
uint64_t PointerSize;
if (ConstantInt *C = dyn_cast<ConstantInt>(AI->getArraySize()))
// Use ABI size (size between elements), not store size (size of one
// element without padding).
PointerSize = C->getZExtValue() *
TD.getABITypeSize(AI->getAllocatedType());
else
PointerSize = ~0UL;
AliasAnalysis::AliasResult R =
AA.alias(Pointer, PointerSize, MemPtr, MemSize);
if (R == AliasAnalysis::NoAlias)
continue;
return MemDepResult::get(Inst);
}
// See if this instruction mod/ref's the pointer.
AliasAnalysis::ModRefResult MRR = AA.getModRefInfo(Inst, MemPtr, MemSize);
if (MRR == AliasAnalysis::NoModRef)
continue;
// Loads don't depend on read-only instructions.
if (isa<LoadInst>(QueryInst) && MRR == AliasAnalysis::Ref)
continue;
// Otherwise, there is a dependence.
return MemDepResult::get(Inst);
}
// If we found nothing, return the non-local flag.
return MemDepResult::getNonLocal();
}
/// getDependency - Return the instruction on which a memory operation
/// depends.
MemDepResult MemoryDependenceAnalysis::getDependency(Instruction *QueryInst) {
Instruction *ScanPos = QueryInst;
// Check for a cached result
DepResultTy &LocalCache = LocalDeps[QueryInst];
// If the cached entry is non-dirty, just return it. Note that this depends
// on DepResultTy's default constructing to 'dirty'.
if (LocalCache.getInt() != Dirty)
return ConvToResult(LocalCache);
// Otherwise, if we have a dirty entry, we know we can start the scan at that
// instruction, which may save us some work.
if (Instruction *Inst = LocalCache.getPointer())
ScanPos = Inst;
// Do the scan.
MemDepResult Res =
getDependencyFrom(QueryInst, ScanPos, QueryInst->getParent());
// Remember the result!
// FIXME: Don't convert back and forth! Make a shared helper function.
LocalCache = ConvFromResult(Res);
if (Instruction *I = Res.getInst())
ReverseLocalDeps[I].insert(QueryInst);
return Res;
}
/// removeInstruction - Remove an instruction from the dependence analysis,
/// updating the dependence of instructions that previously depended on it.
/// This method attempts to keep the cache coherent using the reverse map.
void MemoryDependenceAnalysis::removeInstruction(Instruction *RemInst) {
// Walk through the Non-local dependencies, removing this one as the value
// for any cached queries.
for (DenseMap<BasicBlock*, DepResultTy>::iterator DI =
NonLocalDeps[RemInst].begin(), DE = NonLocalDeps[RemInst].end();
DI != DE; ++DI)
if (Instruction *Inst = DI->second.getPointer())
ReverseNonLocalDeps[Inst].erase(RemInst);
// If we have a cached local dependence query for this instruction, remove it.
//
LocalDepMapType::iterator LocalDepEntry = LocalDeps.find(RemInst);
if (LocalDepEntry != LocalDeps.end()) {
// Remove us from DepInst's reverse set now that the local dep info is gone.
if (Instruction *Inst = LocalDepEntry->second.getPointer()) {
SmallPtrSet<Instruction*, 4> &RLD = ReverseLocalDeps[Inst];
RLD.erase(RemInst);
if (RLD.empty())
ReverseLocalDeps.erase(Inst);
}
// Remove this local dependency info.
LocalDeps.erase(LocalDepEntry);
}
// Loop over all of the things that depend on the instruction we're removing.
//
SmallVector<std::pair<Instruction*, Instruction*>, 8> ReverseDepsToAdd;
ReverseDepMapType::iterator ReverseDepIt = ReverseLocalDeps.find(RemInst);
if (ReverseDepIt != ReverseLocalDeps.end()) {
SmallPtrSet<Instruction*, 4> &ReverseDeps = ReverseDepIt->second;
// RemInst can't be the terminator if it has stuff depending on it.
assert(!ReverseDeps.empty() && !isa<TerminatorInst>(RemInst) &&
"Nothing can locally depend on a terminator");
// Anything that was locally dependent on RemInst is now going to be
// dependent on the instruction after RemInst. It will have the dirty flag
// set so it will rescan. This saves having to scan the entire block to get
// to this point.
Instruction *NewDepInst = next(BasicBlock::iterator(RemInst));
for (SmallPtrSet<Instruction*, 4>::iterator I = ReverseDeps.begin(),
E = ReverseDeps.end(); I != E; ++I) {
Instruction *InstDependingOnRemInst = *I;
// If we thought the instruction depended on itself (possible for
// unconfirmed dependencies) ignore the update.
if (InstDependingOnRemInst == RemInst) continue;
LocalDeps[InstDependingOnRemInst] = DepResultTy(NewDepInst, Dirty);
// Make sure to remember that new things depend on NewDepInst.
ReverseDepsToAdd.push_back(std::make_pair(NewDepInst,
InstDependingOnRemInst));
}
ReverseLocalDeps.erase(ReverseDepIt);
// Add new reverse deps after scanning the set, to avoid invalidating the
// 'ReverseDeps' reference.
while (!ReverseDepsToAdd.empty()) {
ReverseLocalDeps[ReverseDepsToAdd.back().first]
.insert(ReverseDepsToAdd.back().second);
ReverseDepsToAdd.pop_back();
}
}
ReverseDepIt = ReverseNonLocalDeps.find(RemInst);
if (ReverseDepIt != ReverseNonLocalDeps.end()) {
SmallPtrSet<Instruction*, 4>& set = ReverseDepIt->second;
for (SmallPtrSet<Instruction*, 4>::iterator I = set.begin(), E = set.end();
I != E; ++I)
for (DenseMap<BasicBlock*, DepResultTy>::iterator
DI = NonLocalDeps[*I].begin(), DE = NonLocalDeps[*I].end();
DI != DE; ++DI)
if (DI->second.getPointer() == RemInst) {
// Convert to a dirty entry for the subsequent instruction.
DI->second.setInt(Dirty);
if (RemInst->isTerminator())
DI->second.setPointer(0);
else {
Instruction *NextI = next(BasicBlock::iterator(RemInst));
DI->second.setPointer(NextI);
assert(NextI != RemInst);
ReverseDepsToAdd.push_back(std::make_pair(NextI, *I));
}
}
ReverseNonLocalDeps.erase(ReverseDepIt);
// Add new reverse deps after scanning the set, to avoid invalidating 'Set'
while (!ReverseDepsToAdd.empty()) {
ReverseNonLocalDeps[ReverseDepsToAdd.back().first]
.insert(ReverseDepsToAdd.back().second);
ReverseDepsToAdd.pop_back();
}
}
NonLocalDeps.erase(RemInst);
getAnalysis<AliasAnalysis>().deleteValue(RemInst);
DEBUG(verifyRemoved(RemInst));
}
/// verifyRemoved - Verify that the specified instruction does not occur
/// in our internal data structures.
void MemoryDependenceAnalysis::verifyRemoved(Instruction *D) const {
for (LocalDepMapType::const_iterator I = LocalDeps.begin(),
E = LocalDeps.end(); I != E; ++I) {
assert(I->first != D && "Inst occurs in data structures");
assert(I->second.getPointer() != D &&
"Inst occurs in data structures");
}
for (NonLocalDepMapType::const_iterator I = NonLocalDeps.begin(),
E = NonLocalDeps.end(); I != E; ++I) {
assert(I->first != D && "Inst occurs in data structures");
for (DenseMap<BasicBlock*, DepResultTy>::iterator II = I->second.begin(),
EE = I->second.end(); II != EE; ++II)
assert(II->second.getPointer() != D && "Inst occurs in data structures");
}
for (ReverseDepMapType::const_iterator I = ReverseLocalDeps.begin(),
E = ReverseLocalDeps.end(); I != E; ++I)
for (SmallPtrSet<Instruction*, 4>::const_iterator II = I->second.begin(),
EE = I->second.end(); II != EE; ++II)
assert(*II != D && "Inst occurs in data structures");
for (ReverseDepMapType::const_iterator I = ReverseNonLocalDeps.begin(),
E = ReverseNonLocalDeps.end();
I != E; ++I)
for (SmallPtrSet<Instruction*, 4>::const_iterator II = I->second.begin(),
EE = I->second.end(); II != EE; ++II)
assert(*II != D && "Inst occurs in data structures");
}