[llvm-commits] CVS: llvm/lib/Analysis/LoadValueNumbering.cpp
Chris Lattner
lattner at cs.uiuc.edu
Thu Feb 5 11:21:02 PST 2004
Changes in directory llvm/lib/Analysis:
LoadValueNumbering.cpp updated: 1.14 -> 1.15
---
Log message:
In a "seeing the forest through the trees" kinda situation, I realized that a
complete rewrite of load-vn will make it a bit faster. This changes speeds up
the gcse pass (which uses load-vn) from 25.45s to 0.42s on the testcase in
PR209.
I've also verified that this gives the exact same results as the old one.
---
Diffs of the changes: (+230 -227)
Index: llvm/lib/Analysis/LoadValueNumbering.cpp
diff -u llvm/lib/Analysis/LoadValueNumbering.cpp:1.14 llvm/lib/Analysis/LoadValueNumbering.cpp:1.15
--- llvm/lib/Analysis/LoadValueNumbering.cpp:1.14 Wed Feb 4 23:56:23 2004
+++ llvm/lib/Analysis/LoadValueNumbering.cpp Thu Feb 5 11:20:00 2004
@@ -50,17 +50,6 @@
///
virtual void getEqualNumberNodes(Value *V1,
std::vector<Value*> &RetVals) const;
- private:
- /// haveEqualValueNumber - Given two load instructions, determine if they
- /// both produce the same value on every execution of the program, assuming
- /// that their source operands always give the same value. This uses the
- /// AliasAnalysis implementation to invalidate loads when stores or function
- /// calls occur that could modify the value produced by the load.
- ///
- bool haveEqualValueNumber(LoadInst *LI, LoadInst *LI2, AliasAnalysis &AA,
- DominatorSet &DomSetInfo) const;
- bool haveEqualValueNumber(LoadInst *LI, StoreInst *SI, AliasAnalysis &AA,
- DominatorSet &DomSetInfo) const;
};
// Register this pass...
@@ -84,6 +73,43 @@
AU.addRequired<TargetData>();
}
+static bool isPathTransparentTo(BasicBlock *CurBlock, BasicBlock *Dom,
+ Value *Ptr, unsigned Size, AliasAnalysis &AA,
+ std::set<BasicBlock*> &Visited,
+ std::map<BasicBlock*, bool> &TransparentBlocks){
+ // If we have already checked out this path, or if we reached our destination,
+ // stop searching, returning success.
+ if (CurBlock == Dom || !Visited.insert(CurBlock).second)
+ return true;
+
+ // Check whether this block is known transparent or not.
+ std::map<BasicBlock*, bool>::iterator TBI =
+ TransparentBlocks.lower_bound(CurBlock);
+
+ if (TBI == TransparentBlocks.end() || TBI->first != CurBlock) {
+ // If this basic block can modify the memory location, then the path is not
+ // transparent!
+ if (AA.canBasicBlockModify(*CurBlock, Ptr, Size)) {
+ TransparentBlocks.insert(TBI, std::make_pair(CurBlock, false));
+ return false;
+ }
+ TransparentBlocks.insert(TBI, std::make_pair(CurBlock, true));
+ } else if (!TBI->second)
+ // This block is known non-transparent, so that path can't be either.
+ return false;
+
+ // The current block is known to be transparent. The entire path is
+ // transparent if all of the predecessors paths to the parent is also
+ // transparent to the memory location.
+ for (pred_iterator PI = pred_begin(CurBlock), E = pred_end(CurBlock);
+ PI != E; ++PI)
+ if (!isPathTransparentTo(*PI, Dom, Ptr, Size, AA, Visited,
+ TransparentBlocks))
+ return false;
+ return true;
+}
+
+
// getEqualNumberNodes - Return nodes with the same value number as the
// specified Value. This fills in the argument vector with any equal values.
//
@@ -120,14 +146,15 @@
getEqualNumberNodes(LI->getOperand(0), PointerSources);
PointerSources.push_back(LI->getOperand(0));
- Function *F = LI->getParent()->getParent();
+ BasicBlock *LoadBB = LI->getParent();
+ Function *F = LoadBB->getParent();
// Now that we know the set of equivalent source pointers for the load
// instruction, look to see if there are any load or store candidates that are
// identical.
//
- std::vector<LoadInst*> CandidateLoads;
- std::vector<StoreInst*> CandidateStores;
+ std::map<BasicBlock*, std::vector<LoadInst*> > CandidateLoads;
+ std::map<BasicBlock*, std::vector<StoreInst*> > CandidateStores;
while (!PointerSources.empty()) {
Value *Source = PointerSources.back();
@@ -138,239 +165,215 @@
if (LoadInst *Cand = dyn_cast<LoadInst>(*UI)) {// Is a load of source?
if (Cand->getParent()->getParent() == F && // In the same function?
Cand != LI && !Cand->isVolatile()) // Not LI itself?
- CandidateLoads.push_back(Cand); // Got one...
+ CandidateLoads[Cand->getParent()].push_back(Cand); // Got one...
} else if (StoreInst *Cand = dyn_cast<StoreInst>(*UI)) {
if (Cand->getParent()->getParent() == F && !Cand->isVolatile() &&
Cand->getOperand(1) == Source) // It's a store THROUGH the ptr...
- CandidateStores.push_back(Cand);
+ CandidateStores[Cand->getParent()].push_back(Cand);
}
}
- // Get Alias Analysis...
+ // Get alias analysis & dominators.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
DominatorSet &DomSetInfo = getAnalysis<DominatorSet>();
-
- // Loop over all of the candidate loads. If they are not invalidated by
- // stores or calls between execution of them and LI, then add them to RetVals.
- for (unsigned i = 0, e = CandidateLoads.size(); i != e; ++i)
- if (haveEqualValueNumber(LI, CandidateLoads[i], AA, DomSetInfo))
- RetVals.push_back(CandidateLoads[i]);
- for (unsigned i = 0, e = CandidateStores.size(); i != e; ++i)
- if (haveEqualValueNumber(LI, CandidateStores[i], AA, DomSetInfo))
- RetVals.push_back(CandidateStores[i]->getOperand(0));
-
-}
-
-// CheckForInvalidatingInst - Return true if BB or any of the predecessors of BB
-// (until DestBB) contain an instruction that might invalidate Ptr.
-//
-static bool CheckForInvalidatingInst(BasicBlock *BB, BasicBlock *DestBB,
- Value *Ptr, unsigned Size,
- AliasAnalysis &AA,
- std::set<BasicBlock*> &VisitedSet) {
- // Found the termination point!
- if (BB == DestBB || VisitedSet.count(BB)) return false;
-
- // Avoid infinite recursion!
- VisitedSet.insert(BB);
-
- // Can this basic block modify Ptr?
- if (AA.canBasicBlockModify(*BB, Ptr, Size))
- return true;
-
- // Check all of our predecessor blocks...
- for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE; ++PI)
- if (CheckForInvalidatingInst(*PI, DestBB, Ptr, Size, AA, VisitedSet))
- return true;
-
- // None of our predecessor blocks contain an invalidating instruction, and we
- // don't either!
- return false;
-}
-
-
-/// haveEqualValueNumber - Given two load instructions, determine if they both
-/// produce the same value on every execution of the program, assuming that
-/// their source operands always give the same value. This uses the
-/// AliasAnalysis implementation to invalidate loads when stores or function
-/// calls occur that could modify the value produced by the load.
-///
-bool LoadVN::haveEqualValueNumber(LoadInst *L1, LoadInst *L2,
- AliasAnalysis &AA,
- DominatorSet &DomSetInfo) const {
- assert(L1 != L2 && "haveEqualValueNumber assumes differing loads!");
- assert(L1->getType() == L2->getType() &&
- "How could the same source pointer return different types?");
- Value *LoadAddress = L1->getOperand(0);
-
+ Value *LoadPtr = LI->getOperand(0);
// Find out how many bytes of memory are loaded by the load instruction...
- unsigned LoadSize = getAnalysis<TargetData>().getTypeSize(L1->getType());
+ unsigned LoadSize = getAnalysis<TargetData>().getTypeSize(LI->getType());
- // If the two loads are in the same basic block, just do a local analysis.
- if (L1->getParent() == L2->getParent()) {
- // It can be _very_ expensive to determine which instruction occurs first in
- // the basic block if the block is large (see PR209). For this reason,
- // instead of figuring out which block is first, then scanning all of the
- // instructions, we scan the instructions both ways from L1 until we find
- // L2. Along the way if we find a potentially modifying instruction, we
- // kill the search. This helps in cases where we have large blocks the have
- // potentially modifying instructions in them which stop the search.
-
- BasicBlock *BB = L1->getParent();
- BasicBlock::iterator UpIt = L1, DownIt = L1; ++DownIt;
- bool NoModifiesUp = true, NoModifiesDown = true;
-
- // Scan up and down looking for L2, a modifying instruction, or the end of a
- // basic block.
- while (UpIt != BB->begin() && DownIt != BB->end()) {
- // Scan up...
- --UpIt;
- if (&*UpIt == L2)
- return NoModifiesUp; // No instructions invalidate the loads!
- if (NoModifiesUp)
- NoModifiesUp &=
- !(AA.getModRefInfo(UpIt, LoadAddress, LoadSize) & AliasAnalysis::Mod);
-
- if (&*DownIt == L2)
- return NoModifiesDown;
- if (NoModifiesDown)
- NoModifiesDown &=
- !(AA.getModRefInfo(DownIt, LoadAddress, LoadSize)
- & AliasAnalysis::Mod);
- ++DownIt;
+ // Find all of the candidate loads and stores that are in the same block as
+ // the defining instruction.
+ std::set<Instruction*> Instrs;
+ Instrs.insert(CandidateLoads[LoadBB].begin(), CandidateLoads[LoadBB].end());
+ CandidateLoads.erase(LoadBB);
+ Instrs.insert(CandidateStores[LoadBB].begin(), CandidateStores[LoadBB].end());
+ CandidateStores.erase(LoadBB);
+
+ // Figure out if the load is invalidated from the entry of the block it is in
+ // until the actual instruction. This scans the block backwards from LI. If
+ // we see any candidate load or store instructions, then we know that the
+ // candidates have the same value # as LI.
+ bool LoadInvalidatedInBBBefore = false;
+ for (BasicBlock::iterator I = LI; I != LoadBB->begin(); ) {
+ --I;
+ // If this instruction is a candidate load before LI, we know there are no
+ // invalidating instructions between it and LI, so they have the same value
+ // number.
+ if (isa<LoadInst>(I) && Instrs.count(I)) {
+ RetVals.push_back(I);
+ Instrs.erase(I);
}
- // If we got here, we ran into one end of the basic block or the other.
- if (UpIt != BB->begin()) {
- // If we know that the upward scan found a modifier, return false.
- if (!NoModifiesUp) return false;
-
- // Otherwise, continue the scan looking for a modifier or L2.
- for (--UpIt; &*UpIt != L2; --UpIt)
- if (AA.getModRefInfo(UpIt, LoadAddress, LoadSize) & AliasAnalysis::Mod)
- return false;
- return true;
- } else {
- // If we know that the downward scan found a modifier, return false.
- assert(DownIt != BB->end() && "Didn't find instructions??");
- if (!NoModifiesDown) return false;
-
- // Otherwise, continue the scan looking for a modifier or L2.
- for (; &*DownIt != L2; ++DownIt) {
- if (AA.getModRefInfo(DownIt, LoadAddress, LoadSize) &AliasAnalysis::Mod)
- return false;
+ if (AA.getModRefInfo(I, LoadPtr, LoadSize) & AliasAnalysis::Mod) {
+ // If the invalidating instruction is a store, and its in our candidate
+ // set, then we can do store-load forwarding: the load has the same value
+ // # as the stored value.
+ if (isa<StoreInst>(I) && Instrs.count(I)) {
+ Instrs.erase(I);
+ RetVals.push_back(I->getOperand(0));
}
- return true;
- }
- } else {
- // Figure out which load dominates the other one. If neither dominates the
- // other we cannot eliminate them.
- //
- // FIXME: This could be enhanced greatly!
- //
- if (DomSetInfo.dominates(L2, L1))
- std::swap(L1, L2); // Make L1 dominate L2
- else if (!DomSetInfo.dominates(L1, L2))
- return false; // Neither instruction dominates the other one...
-
- BasicBlock *BB1 = L1->getParent(), *BB2 = L2->getParent();
-
- // L1 now dominates L2. Check to see if the intervening instructions
- // between the two loads might modify the loaded location.
- // Make sure that there are no modifying instructions between L1 and the end
- // of its basic block.
- //
- if (AA.canInstructionRangeModify(*L1, *BB1->getTerminator(), LoadAddress,
- LoadSize))
- return false; // Cannot eliminate load
-
- // Make sure that there are no modifying instructions between the start of
- // BB2 and the second load instruction.
- //
- if (AA.canInstructionRangeModify(BB2->front(), *L2, LoadAddress, LoadSize))
- return false; // Cannot eliminate load
-
- // Do a depth first traversal of the inverse CFG starting at L2's block,
- // looking for L1's block. The inverse CFG is made up of the predecessor
- // nodes of a block... so all of the edges in the graph are "backward".
- //
- std::set<BasicBlock*> VisitedSet;
- for (pred_iterator PI = pred_begin(BB2), PE = pred_end(BB2); PI != PE; ++PI)
- if (CheckForInvalidatingInst(*PI, BB1, LoadAddress, LoadSize, AA,
- VisitedSet))
- return false;
-
- // If we passed all of these checks then we are sure that the two loads
- // produce the same value.
- return true;
+ LoadInvalidatedInBBBefore = true;
+ break;
+ }
}
-}
+ // Figure out if the load is invalidated between the load and the exit of the
+ // block it is defined in. While we are scanning the current basic block, if
+ // we see any candidate loads, then we know they have the same value # as LI.
+ //
+ bool LoadInvalidatedInBBAfter = false;
+ for (BasicBlock::iterator I = LI->getNext(); I != LoadBB->end(); ++I) {
+ // If this instruction is a load, then this instruction returns the same
+ // value as LI.
+ if (isa<LoadInst>(I) && Instrs.count(I)) {
+ RetVals.push_back(I);
+ Instrs.erase(I);
+ }
-/// haveEqualValueNumber - Given a load instruction and a store instruction,
-/// determine if the stored value reaches the loaded value unambiguously on
-/// every execution of the program. This uses the AliasAnalysis implementation
-/// to invalidate the stored value when stores or function calls occur that
-/// could modify the value produced by the load.
-///
-bool LoadVN::haveEqualValueNumber(LoadInst *Load, StoreInst *Store,
- AliasAnalysis &AA,
- DominatorSet &DomSetInfo) const {
- // If the store does not dominate the load, we cannot do anything...
- if (!DomSetInfo.dominates(Store, Load))
- return false;
-
- BasicBlock *BB1 = Store->getParent(), *BB2 = Load->getParent();
- Value *LoadAddress = Load->getOperand(0);
-
- assert(LoadAddress->getType() == Store->getOperand(1)->getType() &&
- "How could the same source pointer return different types?");
-
- // Find out how many bytes of memory are loaded by the load instruction...
- unsigned LoadSize = getAnalysis<TargetData>().getTypeSize(Load->getType());
+ if (AA.getModRefInfo(I, LoadPtr, LoadSize) & AliasAnalysis::Mod) {
+ LoadInvalidatedInBBAfter = true;
+ break;
+ }
+ }
- // Compute a basic block iterator pointing to the instruction after the store.
- BasicBlock::iterator StoreIt = Store; ++StoreIt;
+ // If there is anything left in the Instrs set, it could not possibly equal
+ // LI.
+ Instrs.clear();
+
+ // TransparentBlocks - For each basic block the load/store is alive across,
+ // figure out if the pointer is invalidated or not. If it is invalidated, the
+ // boolean is set to false, if it's not it is set to true. If we don't know
+ // yet, the entry is not in the map.
+ std::map<BasicBlock*, bool> TransparentBlocks;
+
+ // Loop over all of the basic blocks that also load the value. If the value
+ // is live across the CFG from the source to destination blocks, and if the
+ // value is not invalidated in either the source or destination blocks, add it
+ // to the equivalence sets.
+ for (std::map<BasicBlock*, std::vector<LoadInst*> >::iterator
+ I = CandidateLoads.begin(), E = CandidateLoads.end(); I != E; ++I) {
+ bool CantEqual = false;
+
+ // Right now we only can handle cases where one load dominates the other.
+ // FIXME: generalize this!
+ BasicBlock *BB1 = I->first, *BB2 = LoadBB;
+ if (DomSetInfo.dominates(BB1, BB2)) {
+ // The other load dominates LI. If the loaded value is killed entering
+ // the LoadBB block, we know the load is not live.
+ if (LoadInvalidatedInBBBefore)
+ CantEqual = true;
+ } else if (DomSetInfo.dominates(BB2, BB1)) {
+ std::swap(BB1, BB2); // Canonicalize
+ // LI dominates the other load. If the loaded value is killed exiting
+ // the LoadBB block, we know the load is not live.
+ if (LoadInvalidatedInBBAfter)
+ CantEqual = true;
+ } else {
+ // None of these loads can VN the same.
+ CantEqual = true;
+ }
- // Check to see if the intervening instructions between the two store and load
- // include a store or call...
- //
- if (BB1 == BB2) { // In same basic block?
- // In this degenerate case, no checking of global basic blocks has to occur
- // just check the instructions BETWEEN Store & Load...
- //
- if (AA.canInstructionRangeModify(*StoreIt, *Load, LoadAddress, LoadSize))
- return false; // Cannot eliminate load
+ if (!CantEqual) {
+ // Ok, at this point, we know that BB1 dominates BB2, and that there is
+ // nothing in the LI block that kills the loaded value. Check to see if
+ // the value is live across the CFG.
+ std::set<BasicBlock*> Visited;
+ for (pred_iterator PI = pred_begin(BB2), E = pred_end(BB2); PI!=E; ++PI)
+ if (!isPathTransparentTo(*PI, BB1, LoadPtr, LoadSize, AA,
+ Visited, TransparentBlocks)) {
+ // None of these loads can VN the same.
+ CantEqual = true;
+ break;
+ }
+ }
- // No instructions invalidate the stored value, they produce the same value!
- return true;
- } else {
- // Make sure that there are no store instructions between the Store and the
- // end of its basic block...
- //
- if (AA.canInstructionRangeModify(*StoreIt, *BB1->getTerminator(),
- LoadAddress, LoadSize))
- return false; // Cannot eliminate load
-
- // Make sure that there are no store instructions between the start of BB2
- // and the second load instruction...
- //
- if (AA.canInstructionRangeModify(BB2->front(), *Load, LoadAddress,LoadSize))
- return false; // Cannot eliminate load
-
- // Do a depth first traversal of the inverse CFG starting at L2's block,
- // looking for L1's block. The inverse CFG is made up of the predecessor
- // nodes of a block... so all of the edges in the graph are "backward".
- //
- std::set<BasicBlock*> VisitedSet;
- for (pred_iterator PI = pred_begin(BB2), PE = pred_end(BB2); PI != PE; ++PI)
- if (CheckForInvalidatingInst(*PI, BB1, LoadAddress, LoadSize, AA,
- VisitedSet))
- return false;
+ // If the loads can equal so far, scan the basic block that contains the
+ // loads under consideration to see if they are invalidated in the block.
+ // For any loads that are not invalidated, add them to the equivalence
+ // set!
+ if (!CantEqual) {
+ Instrs.insert(I->second.begin(), I->second.end());
+ if (BB1 == LoadBB) {
+ // If LI dominates the block in question, check to see if any of the
+ // loads in this block are invalidated before they are reached.
+ for (BasicBlock::iterator BBI = I->first->begin(); ; ++BBI) {
+ if (isa<LoadInst>(BBI) && Instrs.count(BBI)) {
+ // The load is in the set!
+ RetVals.push_back(BBI);
+ Instrs.erase(BBI);
+ if (Instrs.empty()) break;
+ } else if (AA.getModRefInfo(BBI, LoadPtr, LoadSize)
+ & AliasAnalysis::Mod) {
+ // If there is a modifying instruction, nothing below it will value
+ // # the same.
+ break;
+ }
+ }
+ } else {
+ // If the block dominates LI, make sure that the loads in the block are
+ // not invalidated before the block ends.
+ BasicBlock::iterator BBI = I->first->end();
+ while (1) {
+ --BBI;
+ if (isa<LoadInst>(BBI) && Instrs.count(BBI)) {
+ // The load is in the set!
+ RetVals.push_back(BBI);
+ Instrs.erase(BBI);
+ if (Instrs.empty()) break;
+ } else if (AA.getModRefInfo(BBI, LoadPtr, LoadSize)
+ & AliasAnalysis::Mod) {
+ // If there is a modifying instruction, nothing above it will value
+ // # the same.
+ break;
+ }
+ }
+ }
- // If we passed all of these checks then we are sure that the two loads
- // produce the same value.
- return true;
+ Instrs.clear();
+ }
}
+
+ // Handle candidate stores. If the loaded location is clobbered on entrance
+ // to the LoadBB, no store outside of the LoadBB can value number equal, so
+ // quick exit.
+ if (LoadInvalidatedInBBBefore)
+ return;
+
+ for (std::map<BasicBlock*, std::vector<StoreInst*> >::iterator
+ I = CandidateStores.begin(), E = CandidateStores.end(); I != E; ++I)
+ if (DomSetInfo.dominates(I->first, LoadBB)) {
+ // Check to see if the path from the store to the load is transparent
+ // w.r.t. the memory location.
+ bool CantEqual = false;
+ std::set<BasicBlock*> Visited;
+ for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB);
+ PI != E; ++PI)
+ if (!isPathTransparentTo(*PI, I->first, LoadPtr, LoadSize, AA,
+ Visited, TransparentBlocks)) {
+ // None of these stores can VN the same.
+ CantEqual = true;
+ break;
+ }
+ Visited.clear();
+ if (!CantEqual) {
+ // Okay, the path from the store block to the load block is clear, and
+ // we know that there are no invalidating instructions from the start
+ // of the load block to the load itself. Now we just scan the store
+ // block.
+
+ BasicBlock::iterator BBI = I->first->end();
+ while (1) {
+ --BBI;
+ if (AA.getModRefInfo(BBI, LoadPtr, LoadSize)& AliasAnalysis::Mod){
+ // If the invalidating instruction is one of the candidates,
+ // then it provides the value the load loads.
+ if (StoreInst *SI = dyn_cast<StoreInst>(BBI))
+ if (std::find(I->second.begin(), I->second.end(), SI) !=
+ I->second.end())
+ RetVals.push_back(SI->getOperand(0));
+ break;
+ }
+ }
+ }
+ }
}
More information about the llvm-commits
mailing list