/// Many allocas are only used within a single basic block. If this is the
/// case, avoid traversing the CFG and inserting a lot of potentially useless
/// PHI nodes by just performing a single linear pass over the basic block
/// using the Alloca.
///
/// If we cannot promote this alloca (because it is read before it is written),
/// return false. This is necessary in cases where, due to control flow, the
/// alloca is undefined only on some control flow paths. e.g. code like
/// this is correct in LLVM IR:
/// // A is an alloca with no stores so far
/// for (...) {
/// int t = *A;
/// if (!first_iteration)
/// use(t);
/// *A = 42;
/// }
static bool promoteSingleBlockAlloca(AllocaInst *AI, const AllocaInfo &Info,
LargeBlockInfo &LBI,
const DataLayout &DL,
DominatorTree &DT,
AssumptionCache *AC) {
// The trickiest case to handle is when we have large blocks. Because of this,
// this code is optimized assuming that large blocks happen. This does not
// significantly pessimize the small block case. This uses LargeBlockInfo to
// make it efficient to get the index of various operations in the block.
// Walk the use-def list of the alloca, getting the locations of all stores.
using StoresByIndexTy = SmallVector<std::pair<unsigned, StoreInst *>, 64>;
StoresByIndexTy StoresByIndex;
for (User *U : AI->users())
if (StoreInst *SI = dyn_cast<StoreInst>(U))
StoresByIndex.push_back(std::make_pair(LBI.getInstructionIndex(SI), SI));
// Sort the stores by their index, making it efficient to do a lookup with a
// binary search.
llvm::sort(StoresByIndex, less_first());
// Walk all of the loads from this alloca, replacing them with the nearest
// store above them, if any.
for (auto UI = AI->user_begin(), E = AI->user_end(); UI != E;) {
LoadInst *LI = dyn_cast<LoadInst>(*UI++);
if (!LI)
continue;
unsigned LoadIdx = LBI.getInstructionIndex(LI);
// Find the nearest store that has a lower index than this load.
StoresByIndexTy::iterator I =
std::lower_bound(StoresByIndex.begin(), StoresByIndex.end(),
std::make_pair(LoadIdx,
static_cast<StoreInst *>(nullptr)),
less_first());
if (I == StoresByIndex.begin()) {
if (StoresByIndex.empty())
// If there are no stores, the load takes the undef value.
LI->replaceAllUsesWith(UndefValue::get(LI->getType()));
else
// There is no store before this load, bail out (load may be affected
// by the following stores - see main comment).
return false;
} else {
// Otherwise, there was a store before this load, the load takes its value.
// Note, if the load was marked as nonnull we don't want to lose that
// information when we erase it. So we preserve it with an assume.
Value *ReplVal = std::prev(I)->second->getOperand(0);
if (AC && LI->getMetadata(LLVMContext::MD_nonnull) &&
!isKnownNonZero(ReplVal, DL, 0, AC, LI, &DT))
addAssumeNonNull(AC, LI);
// If the replacement value is the load, this must occur in unreachable
// code.
if (ReplVal == LI)
ReplVal = UndefValue::get(LI->getType());
LI->replaceAllUsesWith(ReplVal);
}
LI->eraseFromParent();
LBI.deleteValue(LI);
}
// Remove the (now dead) stores and alloca.
while (!AI->use_empty()) {
StoreInst *SI = cast<StoreInst>(AI->user_back());
// Record debuginfo for the store before removing it.
for (DbgVariableIntrinsic *DII : Info.DbgDeclares) {
DIBuilder DIB(*AI->getModule(), /*AllowUnresolved*/ false);
ConvertDebugDeclareToDebugValue(DII, SI, DIB);
}
SI->eraseFromParent();
LBI.deleteValue(SI);
}
AI->eraseFromParent();
LBI.deleteValue(AI);
// The alloca's debuginfo can be removed as well.
for (DbgVariableIntrinsic *DII : Info.DbgDeclares) {
DII->eraseFromParent();
LBI.deleteValue(DII);
}
++NumLocalPromoted;
return true;
}
/// PromoteSingleBlockAlloca - Many allocas are only used within a single basic
/// block. If this is the case, avoid traversing the CFG and inserting a lot of
/// potentially useless PHI nodes by just performing a single linear pass over
/// the basic block using the Alloca.
///
/// If we cannot promote this alloca (because it is read before it is written),
/// return true. This is necessary in cases where, due to control flow, the
/// alloca is potentially undefined on some control flow paths. e.g. code like
/// this is potentially correct:
///
/// for (...) { if (c) { A = undef; undef = B; } }
///
/// ... so long as A is not used before undef is set.
///
void PromoteMem2Reg::PromoteSingleBlockAlloca(AllocaInst *AI, AllocaInfo &Info,
LargeBlockInfo &LBI) {
// The trickiest case to handle is when we have large blocks. Because of this,
// this code is optimized assuming that large blocks happen. This does not
// significantly pessimize the small block case. This uses LargeBlockInfo to
// make it efficient to get the index of various operations in the block.
// Clear out UsingBlocks. We will reconstruct it here if needed.
Info.UsingBlocks.clear();
// Walk the use-def list of the alloca, getting the locations of all stores.
typedef SmallVector<std::pair<unsigned, StoreInst*>, 64> StoresByIndexTy;
StoresByIndexTy StoresByIndex;
for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end();
UI != E; ++UI)
if (StoreInst *SI = dyn_cast<StoreInst>(*UI))
StoresByIndex.push_back(std::make_pair(LBI.getInstructionIndex(SI), SI));
// If there are no stores to the alloca, just replace any loads with undef.
if (StoresByIndex.empty()) {
for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;)
if (LoadInst *LI = dyn_cast<LoadInst>(*UI++)) {
LI->replaceAllUsesWith(UndefValue::get(LI->getType()));
if (AST && LI->getType()->isPointerTy())
AST->deleteValue(LI);
LBI.deleteValue(LI);
LI->eraseFromParent();
}
return;
}
// Sort the stores by their index, making it efficient to do a lookup with a
// binary search.
std::sort(StoresByIndex.begin(), StoresByIndex.end());
// Walk all of the loads from this alloca, replacing them with the nearest
// store above them, if any.
for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;) {
LoadInst *LI = dyn_cast<LoadInst>(*UI++);
if (!LI) continue;
unsigned LoadIdx = LBI.getInstructionIndex(LI);
// Find the nearest store that has a lower than this load.
StoresByIndexTy::iterator I =
std::lower_bound(StoresByIndex.begin(), StoresByIndex.end(),
std::pair<unsigned, StoreInst*>(LoadIdx, static_cast<StoreInst*>(0)),
StoreIndexSearchPredicate());
// If there is no store before this load, then we can't promote this load.
if (I == StoresByIndex.begin()) {
// Can't handle this load, bail out.
Info.UsingBlocks.push_back(LI->getParent());
continue;
}
// Otherwise, there was a store before this load, the load takes its value.
--I;
LI->replaceAllUsesWith(I->second->getOperand(0));
if (AST && LI->getType()->isPointerTy())
AST->deleteValue(LI);
LI->eraseFromParent();
LBI.deleteValue(LI);
}
}
/// Many allocas are only used within a single basic block. If this is the
/// case, avoid traversing the CFG and inserting a lot of potentially useless
/// PHI nodes by just performing a single linear pass over the basic block
/// using the Alloca.
///
/// If we cannot promote this alloca (because it is read before it is written),
/// return false. This is necessary in cases where, due to control flow, the
/// alloca is undefined only on some control flow paths. e.g. code like
/// this is correct in LLVM IR:
/// // A is an alloca with no stores so far
/// for (...) {
/// int t = *A;
/// if (!first_iteration)
/// use(t);
/// *A = 42;
/// }
static bool promoteSingleBlockAlloca(AllocaInst *AI, const AllocaInfo &Info,
LargeBlockInfo &LBI,
AliasSetTracker *AST) {
// The trickiest case to handle is when we have large blocks. Because of this,
// this code is optimized assuming that large blocks happen. This does not
// significantly pessimize the small block case. This uses LargeBlockInfo to
// make it efficient to get the index of various operations in the block.
// Walk the use-def list of the alloca, getting the locations of all stores.
typedef SmallVector<std::pair<unsigned, StoreInst *>, 64> StoresByIndexTy;
StoresByIndexTy StoresByIndex;
for (User *U : AI->users())
if (StoreInst *SI = dyn_cast<StoreInst>(U))
StoresByIndex.push_back(std::make_pair(LBI.getInstructionIndex(SI), SI));
// Sort the stores by their index, making it efficient to do a lookup with a
// binary search.
std::sort(StoresByIndex.begin(), StoresByIndex.end(), less_first());
// Walk all of the loads from this alloca, replacing them with the nearest
// store above them, if any.
for (auto UI = AI->user_begin(), E = AI->user_end(); UI != E;) {
LoadInst *LI = dyn_cast<LoadInst>(*UI++);
if (!LI)
continue;
unsigned LoadIdx = LBI.getInstructionIndex(LI);
// Find the nearest store that has a lower index than this load.
StoresByIndexTy::iterator I =
std::lower_bound(StoresByIndex.begin(), StoresByIndex.end(),
std::make_pair(LoadIdx,
static_cast<StoreInst *>(nullptr)),
less_first());
if (I == StoresByIndex.begin()) {
if (StoresByIndex.empty())
// If there are no stores, the load takes the undef value.
LI->replaceAllUsesWith(UndefValue::get(LI->getType()));
else
// There is no store before this load, bail out (load may be affected
// by the following stores - see main comment).
return false;
}
else
// Otherwise, there was a store before this load, the load takes its value.
LI->replaceAllUsesWith(std::prev(I)->second->getOperand(0));
if (AST && LI->getType()->isPointerTy())
AST->deleteValue(LI);
LI->eraseFromParent();
LBI.deleteValue(LI);
}
// Remove the (now dead) stores and alloca.
while (!AI->use_empty()) {
StoreInst *SI = cast<StoreInst>(AI->user_back());
// Record debuginfo for the store before removing it.
if (DbgDeclareInst *DDI = Info.DbgDeclare) {
DIBuilder DIB(*AI->getParent()->getParent()->getParent(),
/*AllowUnresolved*/ false);
ConvertDebugDeclareToDebugValue(DDI, SI, DIB);
}
SI->eraseFromParent();
LBI.deleteValue(SI);
}
if (AST)
AST->deleteValue(AI);
AI->eraseFromParent();
LBI.deleteValue(AI);
// The alloca's debuginfo can be removed as well.
if (DbgDeclareInst *DDI = Info.DbgDeclare) {
DDI->eraseFromParent();
LBI.deleteValue(DDI);
}
++NumLocalPromoted;
return true;
}