//
// The LLVM Compiler Infrastructure
//
-// This file was developed by Nick Lewycky and is distributed under the
-// University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//
// The ValueRanges class stores the known integer bounds of a Value. When we
// encounter i8 %a u< %b, the ValueRanges stores that %a = [1, 255] and
-// %b = [0, 254]. Because we store these by Value*, you should always
-// canonicalize through the InequalityGraph first.
+// %b = [0, 254].
//
// It never stores an empty range, because that means that the code is
// unreachable. It never stores a single-element range since that's an equality
-// relationship and better stored in the InequalityGraph.
+// relationship and better stored in the InequalityGraph, nor an empty range
+// since that is better stored in UnreachableBlocks.
//
//===----------------------------------------------------------------------===//
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Analysis/Dominators.h"
-#include "llvm/Analysis/ET-Forest.h"
+#include "llvm/Assembly/Writer.h"
#include "llvm/Support/CFG.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ConstantRange.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/InstVisitor.h"
+#include "llvm/Target/TargetData.h"
#include "llvm/Transforms/Utils/Local.h"
#include <algorithm>
#include <deque>
-#include <sstream>
+#include <stack>
using namespace llvm;
STATISTIC(NumVarsReplaced, "Number of argument substitutions");
STATISTIC(NumSnuggle , "Number of comparisons snuggled");
namespace {
+ class DomTreeDFS {
+ public:
+ class Node {
+ friend class DomTreeDFS;
+ public:
+ typedef std::vector<Node *>::iterator iterator;
+ typedef std::vector<Node *>::const_iterator const_iterator;
+
+ unsigned getDFSNumIn() const { return DFSin; }
+ unsigned getDFSNumOut() const { return DFSout; }
+
+ BasicBlock *getBlock() const { return BB; }
+
+ iterator begin() { return Children.begin(); }
+ iterator end() { return Children.end(); }
+
+ const_iterator begin() const { return Children.begin(); }
+ const_iterator end() const { return Children.end(); }
+
+ bool dominates(const Node *N) const {
+ return DFSin <= N->DFSin && DFSout >= N->DFSout;
+ }
+
+ bool DominatedBy(const Node *N) const {
+ return N->dominates(this);
+ }
+
+ /// Sorts by the number of descendants. With this, you can iterate
+ /// through a sorted list and the first matching entry is the most
+ /// specific match for your basic block. The order provided is stable;
+ /// DomTreeDFS::Nodes with the same number of descendants are sorted by
+ /// DFS in number.
+ bool operator<(const Node &N) const {
+ unsigned spread = DFSout - DFSin;
+ unsigned N_spread = N.DFSout - N.DFSin;
+ if (spread == N_spread) return DFSin < N.DFSin;
+ return spread < N_spread;
+ }
+ bool operator>(const Node &N) const { return N < *this; }
+
+ private:
+ unsigned DFSin, DFSout;
+ BasicBlock *BB;
+
+ std::vector<Node *> Children;
+ };
+
+ // XXX: this may be slow. Instead of using "new" for each node, consider
+ // putting them in a vector to keep them contiguous.
+ explicit DomTreeDFS(DominatorTree *DT) {
+ std::stack<std::pair<Node *, DomTreeNode *> > S;
+
+ Entry = new Node;
+ Entry->BB = DT->getRootNode()->getBlock();
+ S.push(std::make_pair(Entry, DT->getRootNode()));
+
+ NodeMap[Entry->BB] = Entry;
+
+ while (!S.empty()) {
+ std::pair<Node *, DomTreeNode *> &Pair = S.top();
+ Node *N = Pair.first;
+ DomTreeNode *DTNode = Pair.second;
+ S.pop();
+
+ for (DomTreeNode::iterator I = DTNode->begin(), E = DTNode->end();
+ I != E; ++I) {
+ Node *NewNode = new Node;
+ NewNode->BB = (*I)->getBlock();
+ N->Children.push_back(NewNode);
+ S.push(std::make_pair(NewNode, *I));
+
+ NodeMap[NewNode->BB] = NewNode;
+ }
+ }
+
+ renumber();
+
+#ifndef NDEBUG
+ DEBUG(dump());
+#endif
+ }
+
+#ifndef NDEBUG
+ virtual
+#endif
+ ~DomTreeDFS() {
+ std::stack<Node *> S;
+
+ S.push(Entry);
+ while (!S.empty()) {
+ Node *N = S.top(); S.pop();
+
+ for (Node::iterator I = N->begin(), E = N->end(); I != E; ++I)
+ S.push(*I);
+
+ delete N;
+ }
+ }
+
+ /// getRootNode - This returns the entry node for the CFG of the function.
+ Node *getRootNode() const { return Entry; }
+
+ /// getNodeForBlock - return the node for the specified basic block.
+ Node *getNodeForBlock(BasicBlock *BB) const {
+ if (!NodeMap.count(BB)) return 0;
+ return const_cast<DomTreeDFS*>(this)->NodeMap[BB];
+ }
+
+ /// dominates - returns true if the basic block for I1 dominates that of
+ /// the basic block for I2. If the instructions belong to the same basic
+ /// block, the instruction first instruction sequentially in the block is
+ /// considered dominating.
+ bool dominates(Instruction *I1, Instruction *I2) {
+ BasicBlock *BB1 = I1->getParent(),
+ *BB2 = I2->getParent();
+ if (BB1 == BB2) {
+ if (isa<TerminatorInst>(I1)) return false;
+ if (isa<TerminatorInst>(I2)) return true;
+ if ( isa<PHINode>(I1) && !isa<PHINode>(I2)) return true;
+ if (!isa<PHINode>(I1) && isa<PHINode>(I2)) return false;
+
+ for (BasicBlock::const_iterator I = BB2->begin(), E = BB2->end();
+ I != E; ++I) {
+ if (&*I == I1) return true;
+ else if (&*I == I2) return false;
+ }
+ assert(!"Instructions not found in parent BasicBlock?");
+ } else {
+ Node *Node1 = getNodeForBlock(BB1),
+ *Node2 = getNodeForBlock(BB2);
+ return Node1 && Node2 && Node1->dominates(Node2);
+ }
+ return false; // Not reached
+ }
+
+ private:
+ /// renumber - calculates the depth first search numberings and applies
+ /// them onto the nodes.
+ void renumber() {
+ std::stack<std::pair<Node *, Node::iterator> > S;
+ unsigned n = 0;
+
+ Entry->DFSin = ++n;
+ S.push(std::make_pair(Entry, Entry->begin()));
+
+ while (!S.empty()) {
+ std::pair<Node *, Node::iterator> &Pair = S.top();
+ Node *N = Pair.first;
+ Node::iterator &I = Pair.second;
+
+ if (I == N->end()) {
+ N->DFSout = ++n;
+ S.pop();
+ } else {
+ Node *Next = *I++;
+ Next->DFSin = ++n;
+ S.push(std::make_pair(Next, Next->begin()));
+ }
+ }
+ }
+
+#ifndef NDEBUG
+ virtual void dump() const {
+ dump(*cerr.stream());
+ }
+
+ void dump(std::ostream &os) const {
+ os << "Predicate simplifier DomTreeDFS: \n";
+ dump(Entry, 0, os);
+ os << "\n\n";
+ }
+
+ void dump(Node *N, int depth, std::ostream &os) const {
+ ++depth;
+ for (int i = 0; i < depth; ++i) { os << " "; }
+ os << "[" << depth << "] ";
+
+ os << N->getBlock()->getName() << " (" << N->getDFSNumIn()
+ << ", " << N->getDFSNumOut() << ")\n";
+
+ for (Node::iterator I = N->begin(), E = N->end(); I != E; ++I)
+ dump(*I, depth, os);
+ }
+#endif
+
+ Node *Entry;
+ std::map<BasicBlock *, Node *> NodeMap;
+ };
+
// SLT SGT ULT UGT EQ
// 0 1 0 1 0 -- GT 10
// 0 1 0 1 1 -- GE 11
UGE = UGT | EQ_BIT
};
+ /// validPredicate - determines whether a given value is actually a lattice
+ /// value. Only used in assertions or debugging.
static bool validPredicate(LatticeVal LV) {
switch (LV) {
case GT: case GE: case LT: case LE: case NE:
return Rev;
}
- /// This is a StrictWeakOrdering predicate that sorts ETNodes by how many
- /// descendants they have. With this, you can iterate through a list sorted
- /// by this operation and the first matching entry is the most specific
- /// match for your basic block. The order provided is stable; ETNodes with
- /// the same number of children are sorted by pointer address.
- struct VISIBILITY_HIDDEN OrderByDominance {
- bool operator()(const ETNode *LHS, const ETNode *RHS) const {
- unsigned LHS_spread = LHS->getDFSNumOut() - LHS->getDFSNumIn();
- unsigned RHS_spread = RHS->getDFSNumOut() - RHS->getDFSNumIn();
- if (LHS_spread != RHS_spread) return LHS_spread < RHS_spread;
- else return LHS < RHS;
+ /// ValueNumbering stores the scope-specific value numbers for a given Value.
+ class VISIBILITY_HIDDEN ValueNumbering {
+
+ /// VNPair is a tuple of {Value, index number, DomTreeDFS::Node}. It
+ /// includes the comparison operators necessary to allow you to store it
+ /// in a sorted vector.
+ class VISIBILITY_HIDDEN VNPair {
+ public:
+ Value *V;
+ unsigned index;
+ DomTreeDFS::Node *Subtree;
+
+ VNPair(Value *V, unsigned index, DomTreeDFS::Node *Subtree)
+ : V(V), index(index), Subtree(Subtree) {}
+
+ bool operator==(const VNPair &RHS) const {
+ return V == RHS.V && Subtree == RHS.Subtree;
+ }
+
+ bool operator<(const VNPair &RHS) const {
+ if (V != RHS.V) return V < RHS.V;
+ return *Subtree < *RHS.Subtree;
+ }
+
+ bool operator<(Value *RHS) const {
+ return V < RHS;
+ }
+
+ bool operator>(Value *RHS) const {
+ return V > RHS;
+ }
+
+ friend bool operator<(Value *RHS, const VNPair &pair) {
+ return pair.operator>(RHS);
+ }
+ };
+
+ typedef std::vector<VNPair> VNMapType;
+ VNMapType VNMap;
+
+ /// The canonical choice for value number at index.
+ std::vector<Value *> Values;
+
+ DomTreeDFS *DTDFS;
+
+ public:
+#ifndef NDEBUG
+ virtual ~ValueNumbering() {}
+ virtual void dump() {
+ dump(*cerr.stream());
+ }
+
+ void dump(std::ostream &os) {
+ for (unsigned i = 1; i <= Values.size(); ++i) {
+ os << i << " = ";
+ WriteAsOperand(os, Values[i-1]);
+ os << " {";
+ for (unsigned j = 0; j < VNMap.size(); ++j) {
+ if (VNMap[j].index == i) {
+ WriteAsOperand(os, VNMap[j].V);
+ os << " (" << VNMap[j].Subtree->getDFSNumIn() << ") ";
+ }
+ }
+ os << "}\n";
+ }
+ }
+#endif
+
+ /// compare - returns true if V1 is a better canonical value than V2.
+ bool compare(Value *V1, Value *V2) const {
+ if (isa<Constant>(V1))
+ return !isa<Constant>(V2);
+ else if (isa<Constant>(V2))
+ return false;
+ else if (isa<Argument>(V1))
+ return !isa<Argument>(V2);
+ else if (isa<Argument>(V2))
+ return false;
+
+ Instruction *I1 = dyn_cast<Instruction>(V1);
+ Instruction *I2 = dyn_cast<Instruction>(V2);
+
+ if (!I1 || !I2)
+ return V1->getNumUses() < V2->getNumUses();
+
+ return DTDFS->dominates(I1, I2);
+ }
+
+ ValueNumbering(DomTreeDFS *DTDFS) : DTDFS(DTDFS) {}
+
+ /// valueNumber - finds the value number for V under the Subtree. If
+ /// there is no value number, returns zero.
+ unsigned valueNumber(Value *V, DomTreeDFS::Node *Subtree) {
+ if (!(isa<Constant>(V) || isa<Argument>(V) || isa<Instruction>(V))
+ || V->getType() == Type::VoidTy) return 0;
+
+ VNMapType::iterator E = VNMap.end();
+ VNPair pair(V, 0, Subtree);
+ VNMapType::iterator I = std::lower_bound(VNMap.begin(), E, pair);
+ while (I != E && I->V == V) {
+ if (I->Subtree->dominates(Subtree))
+ return I->index;
+ ++I;
+ }
+ return 0;
+ }
+
+ /// getOrInsertVN - always returns a value number, creating it if necessary.
+ unsigned getOrInsertVN(Value *V, DomTreeDFS::Node *Subtree) {
+ if (unsigned n = valueNumber(V, Subtree))
+ return n;
+ else
+ return newVN(V);
+ }
+
+ /// newVN - creates a new value number. Value V must not already have a
+ /// value number assigned.
+ unsigned newVN(Value *V) {
+ assert((isa<Constant>(V) || isa<Argument>(V) || isa<Instruction>(V)) &&
+ "Bad Value for value numbering.");
+ assert(V->getType() != Type::VoidTy && "Won't value number a void value");
+
+ Values.push_back(V);
+
+ VNPair pair = VNPair(V, Values.size(), DTDFS->getRootNode());
+ VNMapType::iterator I = std::lower_bound(VNMap.begin(), VNMap.end(), pair);
+ assert((I == VNMap.end() || value(I->index) != V) &&
+ "Attempt to create a duplicate value number.");
+ VNMap.insert(I, pair);
+
+ return Values.size();
+ }
+
+ /// value - returns the Value associated with a value number.
+ Value *value(unsigned index) const {
+ assert(index != 0 && "Zero index is reserved for not found.");
+ assert(index <= Values.size() && "Index out of range.");
+ return Values[index-1];
+ }
+
+ /// canonicalize - return a Value that is equal to V under Subtree.
+ Value *canonicalize(Value *V, DomTreeDFS::Node *Subtree) {
+ if (isa<Constant>(V)) return V;
+
+ if (unsigned n = valueNumber(V, Subtree))
+ return value(n);
+ else
+ return V;
+ }
+
+ /// addEquality - adds that value V belongs to the set of equivalent
+ /// values defined by value number n under Subtree.
+ void addEquality(unsigned n, Value *V, DomTreeDFS::Node *Subtree) {
+ assert(canonicalize(value(n), Subtree) == value(n) &&
+ "Node's 'canonical' choice isn't best within this subtree.");
+
+ // Suppose that we are given "%x -> node #1 (%y)". The problem is that
+ // we may already have "%z -> node #2 (%x)" somewhere above us in the
+ // graph. We need to find those edges and add "%z -> node #1 (%y)"
+ // to keep the lookups canonical.
+
+ std::vector<Value *> ToRepoint(1, V);
+
+ if (unsigned Conflict = valueNumber(V, Subtree)) {
+ for (VNMapType::iterator I = VNMap.begin(), E = VNMap.end();
+ I != E; ++I) {
+ if (I->index == Conflict && I->Subtree->dominates(Subtree))
+ ToRepoint.push_back(I->V);
+ }
+ }
+
+ for (std::vector<Value *>::iterator VI = ToRepoint.begin(),
+ VE = ToRepoint.end(); VI != VE; ++VI) {
+ Value *V = *VI;
+
+ VNPair pair(V, n, Subtree);
+ VNMapType::iterator B = VNMap.begin(), E = VNMap.end();
+ VNMapType::iterator I = std::lower_bound(B, E, pair);
+ if (I != E && I->V == V && I->Subtree == Subtree)
+ I->index = n; // Update best choice
+ else
+ VNMap.insert(I, pair); // New Value
+
+ // XXX: we currently don't have to worry about updating values with
+ // more specific Subtrees, but we will need to for PHI node support.
+
+#ifndef NDEBUG
+ Value *V_n = value(n);
+ if (isa<Constant>(V) && isa<Constant>(V_n)) {
+ assert(V == V_n && "Constant equals different constant?");
+ }
+#endif
+ }
+ }
+
+ /// remove - removes all references to value V.
+ void remove(Value *V) {
+ VNMapType::iterator B = VNMap.begin(), E = VNMap.end();
+ VNPair pair(V, 0, DTDFS->getRootNode());
+ VNMapType::iterator J = std::upper_bound(B, E, pair);
+ VNMapType::iterator I = J;
+
+ while (I != B && (I == E || I->V == V)) --I;
+
+ VNMap.erase(I, J);
}
};
/// The InequalityGraph class may invalidate Node*s after any mutator call.
/// @brief The InequalityGraph stores the relationships between values.
class VISIBILITY_HIDDEN InequalityGraph {
- ETNode *TreeRoot;
+ ValueNumbering &VN;
+ DomTreeDFS::Node *TreeRoot;
InequalityGraph(); // DO NOT IMPLEMENT
InequalityGraph(InequalityGraph &); // DO NOT IMPLEMENT
public:
- explicit InequalityGraph(ETNode *TreeRoot) : TreeRoot(TreeRoot) {}
+ InequalityGraph(ValueNumbering &VN, DomTreeDFS::Node *TreeRoot)
+ : VN(VN), TreeRoot(TreeRoot) {}
class Node;
/// An Edge is contained inside a Node making one end of the edge implicit
/// and contains a pointer to the other end. The edge contains a lattice
- /// value specifying the relationship and an ETNode specifying the root
- /// in the dominator tree to which this edge applies.
+ /// value specifying the relationship and an DomTreeDFS::Node specifying
+ /// the root in the dominator tree to which this edge applies.
class VISIBILITY_HIDDEN Edge {
public:
- Edge(unsigned T, LatticeVal V, ETNode *ST)
+ Edge(unsigned T, LatticeVal V, DomTreeDFS::Node *ST)
: To(T), LV(V), Subtree(ST) {}
unsigned To;
LatticeVal LV;
- ETNode *Subtree;
+ DomTreeDFS::Node *Subtree;
bool operator<(const Edge &edge) const {
if (To != edge.To) return To < edge.To;
- else return OrderByDominance()(Subtree, edge.Subtree);
+ return *Subtree < *edge.Subtree;
}
+
bool operator<(unsigned to) const {
return To < to;
}
+
+ bool operator>(unsigned to) const {
+ return To > to;
+ }
+
+ friend bool operator<(unsigned to, const Edge &edge) {
+ return edge.operator>(to);
+ }
};
/// A single node in the InequalityGraph. This stores the canonical Value
typedef SmallVector<Edge, 4> RelationsType;
RelationsType Relations;
- Value *Canonical;
-
// TODO: can this idea improve performance?
//friend class std::vector<Node>;
//Node(Node &N) { RelationsType.swap(N.RelationsType); }
typedef RelationsType::iterator iterator;
typedef RelationsType::const_iterator const_iterator;
- Node(Value *V) : Canonical(V) {}
-
- private:
#ifndef NDEBUG
- public:
virtual ~Node() {}
virtual void dump() const {
dump(*cerr.stream());
}
private:
- void dump(std::ostream &os) const {
- os << *getValue() << ":\n";
+ void dump(std::ostream &os) const {
+ static const std::string names[32] =
+ { "000000", "000001", "000002", "000003", "000004", "000005",
+ "000006", "000007", "000008", "000009", " >", " >=",
+ " s>u<", "s>=u<=", " s>", " s>=", "000016", "000017",
+ " s<u>", "s<=u>=", " <", " <=", " s<", " s<=",
+ "000024", "000025", " u>", " u>=", " u<", " u<=",
+ " !=", "000031" };
for (Node::const_iterator NI = begin(), NE = end(); NI != NE; ++NI) {
- static const std::string names[32] =
- { "000000", "000001", "000002", "000003", "000004", "000005",
- "000006", "000007", "000008", "000009", " >", " >=",
- " s>u<", "s>=u<=", " s>", " s>=", "000016", "000017",
- " s<u>", "s<=u>=", " <", " <=", " s<", " s<=",
- "000024", "000025", " u>", " u>=", " u<", " u<=",
- " !=", "000031" };
- os << " " << names[NI->LV] << " " << NI->To
- << " (" << NI->Subtree->getDFSNumIn() << ")\n";
+ os << names[NI->LV] << " " << NI->To
+ << " (" << NI->Subtree->getDFSNumIn() << "), ";
}
}
+ public:
#endif
- public:
iterator begin() { return Relations.begin(); }
iterator end() { return Relations.end(); }
const_iterator begin() const { return Relations.begin(); }
const_iterator end() const { return Relations.end(); }
- iterator find(unsigned n, ETNode *Subtree) {
+ iterator find(unsigned n, DomTreeDFS::Node *Subtree) {
iterator E = end();
for (iterator I = std::lower_bound(begin(), E, n);
I != E && I->To == n; ++I) {
return E;
}
- const_iterator find(unsigned n, ETNode *Subtree) const {
+ const_iterator find(unsigned n, DomTreeDFS::Node *Subtree) const {
const_iterator E = end();
for (const_iterator I = std::lower_bound(begin(), E, n);
I != E && I->To == n; ++I) {
return E;
}
- Value *getValue() const
- {
- return Canonical;
- }
-
- /// Updates the lattice value for a given node. Create a new entry if
- /// one doesn't exist, otherwise it merges the values. The new lattice
- /// value must not be inconsistent with any previously existing value.
- void update(unsigned n, LatticeVal R, ETNode *Subtree) {
+ /// update - updates the lattice value for a given node, creating a new
+ /// entry if one doesn't exist. The new lattice value must not be
+ /// inconsistent with any previously existing value.
+ void update(unsigned n, LatticeVal R, DomTreeDFS::Node *Subtree) {
assert(validPredicate(R) && "Invalid predicate.");
- iterator I = find(n, Subtree);
- if (I == end()) {
- Edge edge(n, R, Subtree);
- iterator Insert = std::lower_bound(begin(), end(), edge);
- Relations.insert(Insert, edge);
- } else {
- LatticeVal LV = static_cast<LatticeVal>(I->LV & R);
- assert(validPredicate(LV) && "Invalid union of lattice values.");
- if (LV != I->LV) {
- if (Subtree != I->Subtree) {
- assert(Subtree->DominatedBy(I->Subtree) &&
- "Find returned subtree that doesn't apply.");
-
- Edge edge(n, R, Subtree);
- iterator Insert = std::lower_bound(begin(), end(), edge);
- Relations.insert(Insert, edge); // invalidates I
- I = find(n, Subtree);
- }
- // Also, we have to tighten any edge that Subtree dominates.
- for (iterator B = begin(); I->To == n; --I) {
- if (I->Subtree->DominatedBy(Subtree)) {
- LatticeVal LV = static_cast<LatticeVal>(I->LV & R);
- assert(validPredicate(LV) && "Invalid union of lattice values.");
- I->LV = LV;
- }
- if (I == B) break;
- }
- }
+ Edge edge(n, R, Subtree);
+ iterator B = begin(), E = end();
+ iterator I = std::lower_bound(B, E, edge);
+
+ iterator J = I;
+ while (J != E && J->To == n) {
+ if (Subtree->DominatedBy(J->Subtree))
+ break;
+ ++J;
}
- }
- };
- private:
- struct VISIBILITY_HIDDEN NodeMapEdge {
- Value *V;
- unsigned index;
- ETNode *Subtree;
+ if (J != E && J->To == n) {
+ edge.LV = static_cast<LatticeVal>(J->LV & R);
+ assert(validPredicate(edge.LV) && "Invalid union of lattice values.");
- NodeMapEdge(Value *V, unsigned index, ETNode *Subtree)
- : V(V), index(index), Subtree(Subtree) {}
-
- bool operator==(const NodeMapEdge &RHS) const {
- return V == RHS.V &&
- Subtree == RHS.Subtree;
- }
+ if (edge.LV == J->LV)
+ return; // This update adds nothing new.
+ }
- bool operator<(const NodeMapEdge &RHS) const {
- if (V != RHS.V) return V < RHS.V;
- return OrderByDominance()(Subtree, RHS.Subtree);
- }
+ if (I != B) {
+ // We also have to tighten any edge beneath our update.
+ for (iterator K = I - 1; K->To == n; --K) {
+ if (K->Subtree->DominatedBy(Subtree)) {
+ LatticeVal LV = static_cast<LatticeVal>(K->LV & edge.LV);
+ assert(validPredicate(LV) && "Invalid union of lattice values");
+ K->LV = LV;
+ }
+ if (K == B) break;
+ }
+ }
- bool operator<(Value *RHS) const {
- return V < RHS;
+ // Insert new edge at Subtree if it isn't already there.
+ if (I == E || I->To != n || Subtree != I->Subtree)
+ Relations.insert(I, edge);
}
};
- typedef std::vector<NodeMapEdge> NodeMapType;
- NodeMapType NodeMap;
+ private:
std::vector<Node> Nodes;
public:
- /// node - returns the node object at a given index retrieved from getNode.
- /// Index zero is reserved and may not be passed in here. The pointer
- /// returned is valid until the next call to newNode or getOrInsertNode.
+ /// node - returns the node object at a given value number. The pointer
+ /// returned may be invalidated on the next call to node().
Node *node(unsigned index) {
- assert(index != 0 && "Zero index is reserved for not found.");
- assert(index <= Nodes.size() && "Index out of range.");
+ assert(VN.value(index)); // This triggers the necessary checks.
+ if (Nodes.size() < index) Nodes.resize(index);
return &Nodes[index-1];
}
- /// Returns the node currently representing Value V, or zero if no such
- /// node exists.
- unsigned getNode(Value *V, ETNode *Subtree) {
- NodeMapType::iterator E = NodeMap.end();
- NodeMapEdge Edge(V, 0, Subtree);
- NodeMapType::iterator I = std::lower_bound(NodeMap.begin(), E, Edge);
- while (I != E && I->V == V) {
- if (Subtree->DominatedBy(I->Subtree))
- return I->index;
- ++I;
- }
- return 0;
- }
-
- /// getOrInsertNode - always returns a valid node index, creating a node
- /// to match the Value if needed.
- unsigned getOrInsertNode(Value *V, ETNode *Subtree) {
- if (unsigned n = getNode(V, Subtree))
- return n;
- else
- return newNode(V);
- }
-
- /// newNode - creates a new node for a given Value and returns the index.
- unsigned newNode(Value *V) {
- Nodes.push_back(Node(V));
-
- NodeMapEdge MapEntry = NodeMapEdge(V, Nodes.size(), TreeRoot);
- assert(!std::binary_search(NodeMap.begin(), NodeMap.end(), MapEntry) &&
- "Attempt to create a duplicate Node.");
- NodeMap.insert(std::lower_bound(NodeMap.begin(), NodeMap.end(),
- MapEntry), MapEntry);
- return MapEntry.index;
- }
-
- /// If the Value is in the graph, return the canonical form. Otherwise,
- /// return the original Value.
- Value *canonicalize(Value *V, ETNode *Subtree) {
- if (isa<Constant>(V)) return V;
-
- if (unsigned n = getNode(V, Subtree))
- return node(n)->getValue();
- else
- return V;
- }
-
/// isRelatedBy - true iff n1 op n2
- bool isRelatedBy(unsigned n1, unsigned n2, ETNode *Subtree, LatticeVal LV) {
+ bool isRelatedBy(unsigned n1, unsigned n2, DomTreeDFS::Node *Subtree,
+ LatticeVal LV) {
if (n1 == n2) return LV & EQ_BIT;
Node *N1 = node(n1);
// The add* methods assume that your input is logically valid and may
// assertion-fail or infinitely loop if you attempt a contradiction.
- void addEquality(unsigned n, Value *V, ETNode *Subtree) {
- assert(canonicalize(node(n)->getValue(), Subtree) == node(n)->getValue()
- && "Node's 'canonical' choice isn't best within this subtree.");
-
- // Suppose that we are given "%x -> node #1 (%y)". The problem is that
- // we may already have "%z -> node #2 (%x)" somewhere above us in the
- // graph. We need to find those edges and add "%z -> node #1 (%y)"
- // to keep the lookups canonical.
-
- std::vector<Value *> ToRepoint;
- ToRepoint.push_back(V);
-
- if (unsigned Conflict = getNode(V, Subtree)) {
- // XXX: NodeMap.size() exceeds 68,000 entries compiling kimwitu++!
- for (NodeMapType::iterator I = NodeMap.begin(), E = NodeMap.end();
- I != E; ++I) {
- if (I->index == Conflict && Subtree->DominatedBy(I->Subtree))
- ToRepoint.push_back(I->V);
- }
- }
-
- for (std::vector<Value *>::iterator VI = ToRepoint.begin(),
- VE = ToRepoint.end(); VI != VE; ++VI) {
- Value *V = *VI;
-
- // XXX: review this code. This may be doing too many insertions.
- NodeMapEdge Edge(V, n, Subtree);
- NodeMapType::iterator E = NodeMap.end();
- NodeMapType::iterator I = std::lower_bound(NodeMap.begin(), E, Edge);
- if (I == E || I->V != V || I->Subtree != Subtree) {
- // New Value
- NodeMap.insert(I, Edge);
- } else if (I != E && I->V == V && I->Subtree == Subtree) {
- // Update best choice
- I->index = n;
- }
-
-#ifndef NDEBUG
- Node *N = node(n);
- if (isa<Constant>(V)) {
- if (isa<Constant>(N->getValue())) {
- assert(V == N->getValue() && "Constant equals different constant?");
- }
- }
-#endif
- }
- }
-
/// addInequality - Sets n1 op n2.
/// It is also an error to call this on an inequality that is already true.
- void addInequality(unsigned n1, unsigned n2, ETNode *Subtree,
+ void addInequality(unsigned n1, unsigned n2, DomTreeDFS::Node *Subtree,
LatticeVal LV1) {
assert(n1 != n2 && "A node can't be inequal to itself.");
assert(!isRelatedBy(n1, n2, Subtree, reversePredicate(LV1)) &&
"Contradictory inequality.");
- Node *N1 = node(n1);
- Node *N2 = node(n2);
-
// Suppose we're adding %n1 < %n2. Find all the %a < %n1 and
// add %a < %n2 too. This keeps the graph fully connected.
if (LV1 != NE) {
- // Someone with a head for this sort of logic, please review this.
- // Given that %x SLTUGT %y and %a SLE %x, what is the relationship
- // between %a and %y? I believe the below code is correct, but I don't
- // think it's the most efficient solution.
+ // Break up the relationship into signed and unsigned comparison parts.
+ // If the signed parts of %a op1 %n1 match that of %n1 op2 %n2, and
+ // op1 and op2 aren't NE, then add %a op3 %n2. The new relationship
+ // should have the EQ_BIT iff it's set for both op1 and op2.
unsigned LV1_s = LV1 & (SLT_BIT|SGT_BIT);
unsigned LV1_u = LV1 & (ULT_BIT|UGT_BIT);
- for (Node::iterator I = N1->begin(), E = N1->end(); I != E; ++I) {
+
+ for (Node::iterator I = node(n1)->begin(), E = node(n1)->end(); I != E; ++I) {
if (I->LV != NE && I->To != n2) {
- ETNode *Local_Subtree = NULL;
+
+ DomTreeDFS::Node *Local_Subtree = NULL;
if (Subtree->DominatedBy(I->Subtree))
Local_Subtree = Subtree;
else if (I->Subtree->DominatedBy(Subtree))
if (LV1_s != (SLT_BIT|SGT_BIT) && ILV_s == LV1_s)
new_relationship |= ILV_s;
-
if (LV1_u != (ULT_BIT|UGT_BIT) && ILV_u == LV1_u)
new_relationship |= ILV_u;
LatticeVal NewLV = static_cast<LatticeVal>(new_relationship);
node(I->To)->update(n2, NewLV, Local_Subtree);
- N2->update(I->To, reversePredicate(NewLV), Local_Subtree);
+ node(n2)->update(I->To, reversePredicate(NewLV), Local_Subtree);
}
}
}
}
- for (Node::iterator I = N2->begin(), E = N2->end(); I != E; ++I) {
+ for (Node::iterator I = node(n2)->begin(), E = node(n2)->end(); I != E; ++I) {
if (I->LV != NE && I->To != n1) {
- ETNode *Local_Subtree = NULL;
+ DomTreeDFS::Node *Local_Subtree = NULL;
if (Subtree->DominatedBy(I->Subtree))
Local_Subtree = Subtree;
else if (I->Subtree->DominatedBy(Subtree))
LatticeVal NewLV = static_cast<LatticeVal>(new_relationship);
- N1->update(I->To, NewLV, Local_Subtree);
+ node(n1)->update(I->To, NewLV, Local_Subtree);
node(I->To)->update(n1, reversePredicate(NewLV), Local_Subtree);
}
}
}
}
- N1->update(n2, LV1, Subtree);
- N2->update(n1, reversePredicate(LV1), Subtree);
+ node(n1)->update(n2, LV1, Subtree);
+ node(n2)->update(n1, reversePredicate(LV1), Subtree);
}
- /// remove - Removes a Value from the graph. If the value is the canonical
- /// choice for a Node, destroys the Node from the graph deleting all edges
- /// to and from it. This method does not renumber the nodes.
- void remove(Value *V) {
- for (unsigned i = 0; i < NodeMap.size();) {
- NodeMapType::iterator I = NodeMap.begin()+i;
- if (I->V == V) {
- Node *N = node(I->index);
- if (node(I->index)->getValue() == V) {
- for (Node::iterator NI = N->begin(), NE = N->end(); NI != NE; ++NI){
- Node::iterator Iter = node(NI->To)->find(I->index, TreeRoot);
- do {
- node(NI->To)->Relations.erase(Iter);
- Iter = node(NI->To)->find(I->index, TreeRoot);
- } while (Iter != node(NI->To)->end());
- }
- N->Canonical = NULL;
- }
- N->Relations.clear();
- NodeMap.erase(I);
- } else ++i;
+ /// remove - removes a node from the graph by removing all references to
+ /// and from it.
+ void remove(unsigned n) {
+ Node *N = node(n);
+ for (Node::iterator NI = N->begin(), NE = N->end(); NI != NE; ++NI) {
+ Node::iterator Iter = node(NI->To)->find(n, TreeRoot);
+ do {
+ node(NI->To)->Relations.erase(Iter);
+ Iter = node(NI->To)->find(n, TreeRoot);
+ } while (Iter != node(NI->To)->end());
}
+ N->Relations.clear();
}
#ifndef NDEBUG
}
void dump(std::ostream &os) {
- std::set<Node *> VisitedNodes;
- for (NodeMapType::const_iterator I = NodeMap.begin(), E = NodeMap.end();
- I != E; ++I) {
- Node *N = node(I->index);
- os << *I->V << " == " << I->index
- << "(" << I->Subtree->getDFSNumIn() << ")\n";
- if (VisitedNodes.insert(N).second) {
- os << I->index << ". ";
- if (!N->getValue()) os << "(deleted node)\n";
- else N->dump(os);
+ for (unsigned i = 1; i <= Nodes.size(); ++i) {
+ os << i << " = {";
+ node(i)->dump(os);
+ os << "}\n";
}
}
- }
#endif
};
/// ValueRanges tracks the known integer ranges and anti-ranges of the nodes
/// in the InequalityGraph.
class VISIBILITY_HIDDEN ValueRanges {
+ ValueNumbering &VN;
+ TargetData *TD;
- /// A ScopedRange ties an InequalityGraph node with a ConstantRange under
- /// the scope of a rooted subtree in the dominator tree.
class VISIBILITY_HIDDEN ScopedRange {
+ typedef std::vector<std::pair<DomTreeDFS::Node *, ConstantRange> >
+ RangeListType;
+ RangeListType RangeList;
+
+ static bool swo(const std::pair<DomTreeDFS::Node *, ConstantRange> &LHS,
+ const std::pair<DomTreeDFS::Node *, ConstantRange> &RHS) {
+ return *LHS.first < *RHS.first;
+ }
+
public:
- ScopedRange(Value *V, ConstantRange CR, ETNode *ST)
- : V(V), CR(CR), Subtree(ST) {}
+#ifndef NDEBUG
+ virtual ~ScopedRange() {}
+ virtual void dump() const {
+ dump(*cerr.stream());
+ }
- Value *V;
- ConstantRange CR;
- ETNode *Subtree;
+ void dump(std::ostream &os) const {
+ os << "{";
+ for (const_iterator I = begin(), E = end(); I != E; ++I) {
+ os << I->second << " (" << I->first->getDFSNumIn() << "), ";
+ }
+ os << "}";
+ }
+#endif
+
+ typedef RangeListType::iterator iterator;
+ typedef RangeListType::const_iterator const_iterator;
+
+ iterator begin() { return RangeList.begin(); }
+ iterator end() { return RangeList.end(); }
+ const_iterator begin() const { return RangeList.begin(); }
+ const_iterator end() const { return RangeList.end(); }
+
+ iterator find(DomTreeDFS::Node *Subtree) {
+ static ConstantRange empty(1, false);
+ iterator E = end();
+ iterator I = std::lower_bound(begin(), E,
+ std::make_pair(Subtree, empty), swo);
- bool operator<(const ScopedRange &range) const {
- if (V != range.V) return V < range.V;
- else return OrderByDominance()(Subtree, range.Subtree);
+ while (I != E && !I->first->dominates(Subtree)) ++I;
+ return I;
}
- bool operator<(const Value *value) const {
- return V < value;
+ const_iterator find(DomTreeDFS::Node *Subtree) const {
+ static const ConstantRange empty(1, false);
+ const_iterator E = end();
+ const_iterator I = std::lower_bound(begin(), E,
+ std::make_pair(Subtree, empty), swo);
+
+ while (I != E && !I->first->dominates(Subtree)) ++I;
+ return I;
+ }
+
+ void update(const ConstantRange &CR, DomTreeDFS::Node *Subtree) {
+ assert(!CR.isEmptySet() && "Empty ConstantRange.");
+ assert(!CR.isSingleElement() && "Refusing to store single element.");
+
+ static ConstantRange empty(1, false);
+ iterator E = end();
+ iterator I =
+ std::lower_bound(begin(), E, std::make_pair(Subtree, empty), swo);
+
+ if (I != end() && I->first == Subtree) {
+ ConstantRange CR2 = I->second.maximalIntersectWith(CR);
+ assert(!CR2.isEmptySet() && !CR2.isSingleElement() &&
+ "Invalid union of ranges.");
+ I->second = CR2;
+ } else
+ RangeList.insert(I, std::make_pair(Subtree, CR));
}
};
std::vector<ScopedRange> Ranges;
- typedef std::vector<ScopedRange>::iterator iterator;
- // XXX: this is a copy of the code in InequalityGraph::Node. Perhaps a
- // intrusive domtree-scoped container is in order?
+ void update(unsigned n, const ConstantRange &CR, DomTreeDFS::Node *Subtree){
+ if (CR.isFullSet()) return;
+ if (Ranges.size() < n) Ranges.resize(n);
+ Ranges[n-1].update(CR, Subtree);
+ }
- iterator begin() { return Ranges.begin(); }
- iterator end() { return Ranges.end(); }
+ /// create - Creates a ConstantRange that matches the given LatticeVal
+ /// relation with a given integer.
+ ConstantRange create(LatticeVal LV, const ConstantRange &CR) {
+ assert(!CR.isEmptySet() && "Can't deal with empty set.");
- iterator find(Value *V, ETNode *Subtree) {
- iterator E = end();
- for (iterator I = std::lower_bound(begin(), E, V);
- I != E && I->V == V; ++I) {
- if (Subtree->DominatedBy(I->Subtree))
- return I;
- }
- return E;
- }
+ if (LV == NE)
+ return makeConstantRange(ICmpInst::ICMP_NE, CR);
- void update(Value *V, ConstantRange CR, ETNode *Subtree) {
- assert(!CR.isEmptySet() && "Empty ConstantRange!");
- if (CR.isFullSet()) return;
+ unsigned LV_s = LV & (SGT_BIT|SLT_BIT);
+ unsigned LV_u = LV & (UGT_BIT|ULT_BIT);
+ bool hasEQ = LV & EQ_BIT;
- iterator I = find(V, Subtree);
- if (I == end()) {
- ScopedRange range(V, CR, Subtree);
- iterator Insert = std::lower_bound(begin(), end(), range);
- Ranges.insert(Insert, range);
- } else {
- CR = CR.intersectWith(I->CR);
- assert(!CR.isEmptySet() && "Empty intersection of ConstantRanges!");
-
- if (CR != I->CR) {
- if (Subtree != I->Subtree) {
- assert(Subtree->DominatedBy(I->Subtree) &&
- "Find returned subtree that doesn't apply.");
-
- ScopedRange range(V, CR, Subtree);
- iterator Insert = std::lower_bound(begin(), end(), range);
- Ranges.insert(Insert, range); // invalidates I
- I = find(V, Subtree);
- }
+ ConstantRange Range(CR.getBitWidth());
- // Also, we have to tighten any edge that Subtree dominates.
- for (iterator B = begin(); I->V == V; --I) {
- if (I->Subtree->DominatedBy(Subtree)) {
- CR = CR.intersectWith(I->CR);
- assert(!CR.isEmptySet() &&
- "Empty intersection of ConstantRanges!");
- I->CR = CR;
- }
- if (I == B) break;
- }
- }
+ if (LV_s == SGT_BIT) {
+ Range = Range.maximalIntersectWith(makeConstantRange(
+ hasEQ ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_SGT, CR));
+ } else if (LV_s == SLT_BIT) {
+ Range = Range.maximalIntersectWith(makeConstantRange(
+ hasEQ ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_SLT, CR));
}
+
+ if (LV_u == UGT_BIT) {
+ Range = Range.maximalIntersectWith(makeConstantRange(
+ hasEQ ? ICmpInst::ICMP_UGE : ICmpInst::ICMP_UGT, CR));
+ } else if (LV_u == ULT_BIT) {
+ Range = Range.maximalIntersectWith(makeConstantRange(
+ hasEQ ? ICmpInst::ICMP_ULE : ICmpInst::ICMP_ULT, CR));
+ }
+
+ return Range;
}
- /// range - Creates a ConstantRange representing the set of all values
- /// that match the ICmpInst::Predicate with any of the values in CR.
- ConstantRange range(ICmpInst::Predicate ICmpOpcode,
- const ConstantRange &CR) {
+ /// makeConstantRange - Creates a ConstantRange representing the set of all
+ /// value that match the ICmpInst::Predicate with any of the values in CR.
+ ConstantRange makeConstantRange(ICmpInst::Predicate ICmpOpcode,
+ const ConstantRange &CR) {
uint32_t W = CR.getBitWidth();
switch (ICmpOpcode) {
- default: assert(!"Invalid ICmp opcode to range()");
+ default: assert(!"Invalid ICmp opcode to makeConstantRange()");
case ICmpInst::ICMP_EQ:
return ConstantRange(CR.getLower(), CR.getUpper());
case ICmpInst::ICMP_NE:
case ICmpInst::ICMP_SLT:
return ConstantRange(APInt::getSignedMinValue(W), CR.getSignedMax());
case ICmpInst::ICMP_ULE: {
- APInt UMax = CR.getUnsignedMax();
- if (UMax == APInt::getMaxValue(W))
+ APInt UMax(CR.getUnsignedMax());
+ if (UMax.isMaxValue())
return ConstantRange(W);
return ConstantRange(APInt::getMinValue(W), UMax + 1);
}
case ICmpInst::ICMP_SLE: {
- APInt SMax = CR.getSignedMax();
- if (SMax == APInt::getSignedMaxValue(W) ||
- SMax + 1 == APInt::getSignedMaxValue(W))
+ APInt SMax(CR.getSignedMax());
+ if (SMax.isMaxSignedValue() || (SMax+1).isMaxSignedValue())
return ConstantRange(W);
return ConstantRange(APInt::getSignedMinValue(W), SMax + 1);
}
case ICmpInst::ICMP_UGT:
- return ConstantRange(CR.getUnsignedMin() + 1,
- APInt::getMaxValue(W) + 1);
+ return ConstantRange(CR.getUnsignedMin() + 1, APInt::getNullValue(W));
case ICmpInst::ICMP_SGT:
return ConstantRange(CR.getSignedMin() + 1,
- APInt::getSignedMaxValue(W) + 1);
+ APInt::getSignedMinValue(W));
case ICmpInst::ICMP_UGE: {
- APInt UMin = CR.getUnsignedMin();
- if (UMin == APInt::getMinValue(W))
+ APInt UMin(CR.getUnsignedMin());
+ if (UMin.isMinValue())
return ConstantRange(W);
- return ConstantRange(UMin, APInt::getMaxValue(W) + 1);
+ return ConstantRange(UMin, APInt::getNullValue(W));
}
case ICmpInst::ICMP_SGE: {
- APInt SMin = CR.getSignedMin();
- if (SMin == APInt::getSignedMinValue(W))
+ APInt SMin(CR.getSignedMin());
+ if (SMin.isMinSignedValue())
return ConstantRange(W);
- return ConstantRange(SMin, APInt::getSignedMaxValue(W) + 1);
+ return ConstantRange(SMin, APInt::getSignedMinValue(W));
}
}
}
- /// create - Creates a ConstantRange that matches the given LatticeVal
- /// relation with a given integer.
- ConstantRange create(LatticeVal LV, const ConstantRange &CR) {
- assert(!CR.isEmptySet() && "Can't deal with empty set.");
-
- if (LV == NE)
- return range(ICmpInst::ICMP_NE, CR);
-
- unsigned LV_s = LV & (SGT_BIT|SLT_BIT);
- unsigned LV_u = LV & (UGT_BIT|ULT_BIT);
- bool hasEQ = LV & EQ_BIT;
+#ifndef NDEBUG
+ bool isCanonical(Value *V, DomTreeDFS::Node *Subtree) {
+ return V == VN.canonicalize(V, Subtree);
+ }
+#endif
- ConstantRange Range(CR.getBitWidth());
+ public:
- if (LV_s == SGT_BIT) {
- Range = Range.intersectWith(range(
- hasEQ ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_SGT, CR));
- } else if (LV_s == SLT_BIT) {
- Range = Range.intersectWith(range(
- hasEQ ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_SLT, CR));
- }
+ ValueRanges(ValueNumbering &VN, TargetData *TD) : VN(VN), TD(TD) {}
- if (LV_u == UGT_BIT) {
- Range = Range.intersectWith(range(
- hasEQ ? ICmpInst::ICMP_UGE : ICmpInst::ICMP_UGT, CR));
- } else if (LV_u == ULT_BIT) {
- Range = Range.intersectWith(range(
- hasEQ ? ICmpInst::ICMP_ULE : ICmpInst::ICMP_ULT, CR));
- }
+#ifndef NDEBUG
+ virtual ~ValueRanges() {}
- return Range;
+ virtual void dump() const {
+ dump(*cerr.stream());
}
- ConstantRange rangeFromValue(Value *V, ETNode *Subtree, uint32_t W) {
- ConstantInt *C = dyn_cast<ConstantInt>(V);
- if (C) {
- return ConstantRange(C->getValue());
- } else {
- iterator I = find(V, Subtree);
- if (I != end())
- return I->CR;
+ void dump(std::ostream &os) const {
+ for (unsigned i = 0, e = Ranges.size(); i != e; ++i) {
+ os << (i+1) << " = ";
+ Ranges[i].dump(os);
+ os << "\n";
}
- return ConstantRange(W);
}
+#endif
- static uint32_t widthOfValue(Value *V) {
- const Type *Ty = V->getType();
- if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty))
- return ITy->getBitWidth();
+ /// range - looks up the ConstantRange associated with a value number.
+ ConstantRange range(unsigned n, DomTreeDFS::Node *Subtree) {
+ assert(VN.value(n)); // performs range checks
- // XXX: I'd like to transform T* into the appropriate integer by
- // bit length, however that data may not be available.
+ if (n <= Ranges.size()) {
+ ScopedRange::iterator I = Ranges[n-1].find(Subtree);
+ if (I != Ranges[n-1].end()) return I->second;
+ }
- return 0;
+ Value *V = VN.value(n);
+ ConstantRange CR = range(V);
+ return CR;
}
-#ifndef NDEBUG
- bool isCanonical(Value *V, ETNode *Subtree, VRPSolver *VRP);
-#endif
-
- public:
-
- bool isRelatedBy(Value *V1, Value *V2, ETNode *Subtree, LatticeVal LV) {
- uint32_t W = widthOfValue(V1);
- if (!W) return false;
+ /// range - determine a range from a Value without performing any lookups.
+ ConstantRange range(Value *V) const {
+ if (ConstantInt *C = dyn_cast<ConstantInt>(V))
+ return ConstantRange(C->getValue());
+ else if (isa<ConstantPointerNull>(V))
+ return ConstantRange(APInt::getNullValue(typeToWidth(V->getType())));
+ else
+ return ConstantRange(typeToWidth(V->getType()));
+ }
- ConstantRange CR1 = rangeFromValue(V1, Subtree, W);
- ConstantRange CR2 = rangeFromValue(V2, Subtree, W);
+ // typeToWidth - returns the number of bits necessary to store a value of
+ // this type, or zero if unknown.
+ uint32_t typeToWidth(const Type *Ty) const {
+ if (TD)
+ return TD->getTypeSizeInBits(Ty);
+ else
+ return Ty->getPrimitiveSizeInBits();
+ }
- // True iff all values in CR1 are LV to all values in CR2.
+ static bool isRelatedBy(const ConstantRange &CR1, const ConstantRange &CR2,
+ LatticeVal LV) {
switch (LV) {
default: assert(!"Impossible lattice value!");
case NE:
- return CR1.intersectWith(CR2).isEmptySet();
+ return CR1.maximalIntersectWith(CR2).isEmptySet();
case ULT:
return CR1.getUnsignedMax().ult(CR2.getUnsignedMin());
case ULE:
}
}
- void addToWorklist(Value *V, const APInt *I, ICmpInst::Predicate Pred,
- VRPSolver *VRP);
+ bool isRelatedBy(unsigned n1, unsigned n2, DomTreeDFS::Node *Subtree,
+ LatticeVal LV) {
+ ConstantRange CR1 = range(n1, Subtree);
+ ConstantRange CR2 = range(n2, Subtree);
- void mergeInto(Value **I, unsigned n, Value *New, ETNode *Subtree,
- VRPSolver *VRP) {
- assert(isCanonical(New, Subtree, VRP) && "Best choice not canonical?");
+ // True iff all values in CR1 are LV to all values in CR2.
+ return isRelatedBy(CR1, CR2, LV);
+ }
- uint32_t W = widthOfValue(New);
- if (!W) return;
+ void addToWorklist(Value *V, Constant *C, ICmpInst::Predicate Pred,
+ VRPSolver *VRP);
+ void markBlock(VRPSolver *VRP);
- ConstantRange CR_New = rangeFromValue(New, Subtree, W);
+ void mergeInto(Value **I, unsigned n, unsigned New,
+ DomTreeDFS::Node *Subtree, VRPSolver *VRP) {
+ ConstantRange CR_New = range(New, Subtree);
ConstantRange Merged = CR_New;
for (; n != 0; ++I, --n) {
- ConstantRange CR_Kill = rangeFromValue(*I, Subtree, W);
+ unsigned i = VN.valueNumber(*I, Subtree);
+ ConstantRange CR_Kill = i ? range(i, Subtree) : range(*I);
if (CR_Kill.isFullSet()) continue;
- Merged = Merged.intersectWith(CR_Kill);
+ Merged = Merged.maximalIntersectWith(CR_Kill);
}
if (Merged.isFullSet() || Merged == CR_New) return;
- if (Merged.isSingleElement())
- addToWorklist(New, Merged.getSingleElement(),
- ICmpInst::ICMP_EQ, VRP);
- else
- update(New, Merged, Subtree);
+ applyRange(New, Merged, Subtree, VRP);
}
- void addInequality(Value *V1, Value *V2, ETNode *Subtree, LatticeVal LV,
- VRPSolver *VRP) {
- assert(!isRelatedBy(V1, V2, Subtree, LV) && "Asked to do useless work.");
+ void applyRange(unsigned n, const ConstantRange &CR,
+ DomTreeDFS::Node *Subtree, VRPSolver *VRP) {
+ ConstantRange Merged = CR.maximalIntersectWith(range(n, Subtree));
+ if (Merged.isEmptySet()) {
+ markBlock(VRP);
+ return;
+ }
- assert(isCanonical(V1, Subtree, VRP) && "Value not canonical.");
- assert(isCanonical(V2, Subtree, VRP) && "Value not canonical.");
+ if (const APInt *I = Merged.getSingleElement()) {
+ Value *V = VN.value(n); // XXX: redesign worklist.
+ const Type *Ty = V->getType();
+ if (Ty->isInteger()) {
+ addToWorklist(V, ConstantInt::get(*I), ICmpInst::ICMP_EQ, VRP);
+ return;
+ } else if (const PointerType *PTy = dyn_cast<PointerType>(Ty)) {
+ assert(*I == 0 && "Pointer is null but not zero?");
+ addToWorklist(V, ConstantPointerNull::get(PTy),
+ ICmpInst::ICMP_EQ, VRP);
+ return;
+ }
+ }
- if (LV == NE) return; // we can't represent those.
- // XXX: except in the case where isSingleElement and equal to either
- // Lower or Upper. That's probably not profitable. (Type::Int1Ty?)
+ update(n, Merged, Subtree);
+ }
- uint32_t W = widthOfValue(V1);
- if (!W) return;
+ void addNotEquals(unsigned n1, unsigned n2, DomTreeDFS::Node *Subtree,
+ VRPSolver *VRP) {
+ ConstantRange CR1 = range(n1, Subtree);
+ ConstantRange CR2 = range(n2, Subtree);
+
+ uint32_t W = CR1.getBitWidth();
+
+ if (const APInt *I = CR1.getSingleElement()) {
+ if (CR2.isFullSet()) {
+ ConstantRange NewCR2(CR1.getUpper(), CR1.getLower());
+ applyRange(n2, NewCR2, Subtree, VRP);
+ } else if (*I == CR2.getLower()) {
+ APInt NewLower(CR2.getLower() + 1),
+ NewUpper(CR2.getUpper());
+ if (NewLower == NewUpper)
+ NewLower = NewUpper = APInt::getMinValue(W);
+
+ ConstantRange NewCR2(NewLower, NewUpper);
+ applyRange(n2, NewCR2, Subtree, VRP);
+ } else if (*I == CR2.getUpper() - 1) {
+ APInt NewLower(CR2.getLower()),
+ NewUpper(CR2.getUpper() - 1);
+ if (NewLower == NewUpper)
+ NewLower = NewUpper = APInt::getMinValue(W);
+
+ ConstantRange NewCR2(NewLower, NewUpper);
+ applyRange(n2, NewCR2, Subtree, VRP);
+ }
+ }
- ConstantRange CR1 = rangeFromValue(V1, Subtree, W);
- ConstantRange CR2 = rangeFromValue(V2, Subtree, W);
+ if (const APInt *I = CR2.getSingleElement()) {
+ if (CR1.isFullSet()) {
+ ConstantRange NewCR1(CR2.getUpper(), CR2.getLower());
+ applyRange(n1, NewCR1, Subtree, VRP);
+ } else if (*I == CR1.getLower()) {
+ APInt NewLower(CR1.getLower() + 1),
+ NewUpper(CR1.getUpper());
+ if (NewLower == NewUpper)
+ NewLower = NewUpper = APInt::getMinValue(W);
+
+ ConstantRange NewCR1(NewLower, NewUpper);
+ applyRange(n1, NewCR1, Subtree, VRP);
+ } else if (*I == CR1.getUpper() - 1) {
+ APInt NewLower(CR1.getLower()),
+ NewUpper(CR1.getUpper() - 1);
+ if (NewLower == NewUpper)
+ NewLower = NewUpper = APInt::getMinValue(W);
+
+ ConstantRange NewCR1(NewLower, NewUpper);
+ applyRange(n1, NewCR1, Subtree, VRP);
+ }
+ }
+ }
+
+ void addInequality(unsigned n1, unsigned n2, DomTreeDFS::Node *Subtree,
+ LatticeVal LV, VRPSolver *VRP) {
+ assert(!isRelatedBy(n1, n2, Subtree, LV) && "Asked to do useless work.");
+
+ if (LV == NE) {
+ addNotEquals(n1, n2, Subtree, VRP);
+ return;
+ }
+
+ ConstantRange CR1 = range(n1, Subtree);
+ ConstantRange CR2 = range(n2, Subtree);
if (!CR1.isSingleElement()) {
- ConstantRange NewCR1 = CR1.intersectWith(create(LV, CR2));
- if (NewCR1 != CR1) {
- if (NewCR1.isSingleElement())
- addToWorklist(V1, NewCR1.getSingleElement(),
- ICmpInst::ICMP_EQ, VRP);
- else
- update(V1, NewCR1, Subtree);
- }
+ ConstantRange NewCR1 = CR1.maximalIntersectWith(create(LV, CR2));
+ if (NewCR1 != CR1)
+ applyRange(n1, NewCR1, Subtree, VRP);
}
if (!CR2.isSingleElement()) {
- ConstantRange NewCR2 = CR2.intersectWith(create(reversePredicate(LV),
- CR1));
- if (NewCR2 != CR2) {
- if (NewCR2.isSingleElement())
- addToWorklist(V2, NewCR2.getSingleElement(),
- ICmpInst::ICMP_EQ, VRP);
- else
- update(V2, NewCR2, Subtree);
- }
+ ConstantRange NewCR2 = CR2.maximalIntersectWith(
+ create(reversePredicate(LV), CR1));
+ if (NewCR2 != CR2)
+ applyRange(n2, NewCR2, Subtree, VRP);
}
}
};
Value *LHS, *RHS;
ICmpInst::Predicate Op;
- BasicBlock *ContextBB;
+ BasicBlock *ContextBB; // XXX use a DomTreeDFS::Node instead
Instruction *ContextInst;
};
std::deque<Operation> WorkList;
+ ValueNumbering &VN;
InequalityGraph &IG;
UnreachableBlocks &UB;
ValueRanges &VR;
-
- ETForest *Forest;
- ETNode *Top;
+ DomTreeDFS *DTDFS;
+ DomTreeDFS::Node *Top;
BasicBlock *TopBB;
Instruction *TopInst;
bool &modified;
typedef InequalityGraph::Node Node;
- /// IdomI - Determines whether one Instruction dominates another.
- bool IdomI(Instruction *I1, Instruction *I2) const {
- BasicBlock *BB1 = I1->getParent(),
- *BB2 = I2->getParent();
- if (BB1 == BB2) {
- if (isa<TerminatorInst>(I1)) return false;
- if (isa<TerminatorInst>(I2)) return true;
- if (isa<PHINode>(I1) && !isa<PHINode>(I2)) return true;
- if (!isa<PHINode>(I1) && isa<PHINode>(I2)) return false;
-
- for (BasicBlock::const_iterator I = BB1->begin(), E = BB1->end();
- I != E; ++I) {
- if (&*I == I1) return true;
- if (&*I == I2) return false;
+ // below - true if the Instruction is dominated by the current context
+ // block or instruction
+ bool below(Instruction *I) {
+ BasicBlock *BB = I->getParent();
+ if (TopInst && TopInst->getParent() == BB) {
+ if (isa<TerminatorInst>(TopInst)) return false;
+ if (isa<TerminatorInst>(I)) return true;
+ if ( isa<PHINode>(TopInst) && !isa<PHINode>(I)) return true;
+ if (!isa<PHINode>(TopInst) && isa<PHINode>(I)) return false;
+
+ for (BasicBlock::const_iterator Iter = BB->begin(), E = BB->end();
+ Iter != E; ++Iter) {
+ if (&*Iter == TopInst) return true;
+ else if (&*Iter == I) return false;
}
assert(!"Instructions not found in parent BasicBlock?");
} else {
- return Forest->properlyDominates(BB1, BB2);
+ DomTreeDFS::Node *Node = DTDFS->getNodeForBlock(BB);
+ if (!Node) return false;
+ return Top->dominates(Node);
}
- return false;
+ return false; // Not reached
}
- /// Returns true if V1 is a better canonical value than V2.
- bool compare(Value *V1, Value *V2) const {
- if (isa<Constant>(V1))
- return !isa<Constant>(V2);
- else if (isa<Constant>(V2))
- return false;
- else if (isa<Argument>(V1))
- return !isa<Argument>(V2);
- else if (isa<Argument>(V2))
- return false;
-
- Instruction *I1 = dyn_cast<Instruction>(V1);
- Instruction *I2 = dyn_cast<Instruction>(V2);
+ // aboveOrBelow - true if the Instruction either dominates or is dominated
+ // by the current context block or instruction
+ bool aboveOrBelow(Instruction *I) {
+ BasicBlock *BB = I->getParent();
+ DomTreeDFS::Node *Node = DTDFS->getNodeForBlock(BB);
+ if (!Node) return false;
- if (!I1 || !I2)
- return V1->getNumUses() < V2->getNumUses();
-
- return IdomI(I1, I2);
- }
-
- // below - true if the Instruction is dominated by the current context
- // block or instruction
- bool below(Instruction *I) {
- if (TopInst)
- return IdomI(TopInst, I);
- else {
- ETNode *Node = Forest->getNodeForBlock(I->getParent());
- return Node->DominatedBy(Top);
- }
+ return Top == Node || Top->dominates(Node) || Node->dominates(Top);
}
bool makeEqual(Value *V1, Value *V2) {
DOUT << "makeEqual(" << *V1 << ", " << *V2 << ")\n";
+ DOUT << "context is ";
+ if (TopInst) DOUT << "I: " << *TopInst << "\n";
+ else DOUT << "BB: " << TopBB->getName()
+ << "(" << Top->getDFSNumIn() << ")\n";
+
+ assert(V1->getType() == V2->getType() &&
+ "Can't make two values with different types equal.");
if (V1 == V2) return true;
if (isa<Constant>(V1) && isa<Constant>(V2))
return false;
- unsigned n1 = IG.getNode(V1, Top), n2 = IG.getNode(V2, Top);
+ unsigned n1 = VN.valueNumber(V1, Top), n2 = VN.valueNumber(V2, Top);
if (n1 && n2) {
if (n1 == n2) return true;
if (IG.isRelatedBy(n1, n2, Top, NE)) return false;
}
- if (n1) assert(V1 == IG.node(n1)->getValue() && "Value isn't canonical.");
- if (n2) assert(V2 == IG.node(n2)->getValue() && "Value isn't canonical.");
+ if (n1) assert(V1 == VN.value(n1) && "Value isn't canonical.");
+ if (n2) assert(V2 == VN.value(n2) && "Value isn't canonical.");
- assert(!compare(V2, V1) && "Please order parameters to makeEqual.");
+ assert(!VN.compare(V2, V1) && "Please order parameters to makeEqual.");
assert(!isa<Constant>(V2) && "Tried to remove a constant.");
// be EQ and that's invalid. What we're doing is looking for any nodes
// %z such that %x <= %z and %y >= %z, and vice versa.
- Node *N1 = IG.node(n1);
- Node *N2 = IG.node(n2);
- Node::iterator end = N2->end();
+ Node::iterator end = IG.node(n2)->end();
// Find the intersection between N1 and N2 which is dominated by
// Top. If we find %x where N1 <= %x <= N2 (or >=) then add %x to
// Remove.
- for (Node::iterator I = N1->begin(), E = N1->end(); I != E; ++I) {
+ for (Node::iterator I = IG.node(n1)->begin(), E = IG.node(n1)->end();
+ I != E; ++I) {
if (!(I->LV & EQ_BIT) || !Top->DominatedBy(I->Subtree)) continue;
unsigned ILV_s = I->LV & (SLT_BIT|SGT_BIT);
unsigned ILV_u = I->LV & (ULT_BIT|UGT_BIT);
- Node::iterator NI = N2->find(I->To, Top);
+ Node::iterator NI = IG.node(n2)->find(I->To, Top);
if (NI != end) {
LatticeVal NILV = reversePredicate(NI->LV);
unsigned NILV_s = NILV & (SLT_BIT|SGT_BIT);
for (SetVector<unsigned>::iterator I = Remove.begin()+1 /* skip n2 */,
E = Remove.end(); I != E; ++I) {
unsigned n = *I;
- Value *V = IG.node(n)->getValue();
- if (compare(V, V1)) {
+ Value *V = VN.value(n);
+ if (VN.compare(V, V1)) {
V1 = V;
n1 = n;
DontRemove = I;
}
// We'd like to allow makeEqual on two values to perform a simple
- // substitution without every creating nodes in the IG whenever possible.
+ // substitution without creating nodes in the IG whenever possible.
//
// The first iteration through this loop operates on V2 before going
// through the Remove list and operating on those too. If all of the
bool mergeIGNode = false;
unsigned i = 0;
for (Value *R = V2; i == 0 || i < Remove.size(); ++i) {
- if (i) R = IG.node(Remove[i])->getValue(); // skip n2.
+ if (i) R = VN.value(Remove[i]); // skip n2.
// Try to replace the whole instruction. If we can, we're done.
Instruction *I2 = dyn_cast<Instruction>(R);
if (!isa<Constant>(V1)) {
if (Remove.empty()) {
- VR.mergeInto(&V2, 1, V1, Top, this);
+ VR.mergeInto(&V2, 1, VN.getOrInsertVN(V1, Top), Top, this);
} else {
std::vector<Value*> RemoveVals;
RemoveVals.reserve(Remove.size());
for (SetVector<unsigned>::iterator I = Remove.begin(),
E = Remove.end(); I != E; ++I) {
- Value *V = IG.node(*I)->getValue();
+ Value *V = VN.value(*I);
if (!V->use_empty())
RemoveVals.push_back(V);
}
- VR.mergeInto(&RemoveVals[0], RemoveVals.size(), V1, Top, this);
+ VR.mergeInto(&RemoveVals[0], RemoveVals.size(),
+ VN.getOrInsertVN(V1, Top), Top, this);
}
}
if (mergeIGNode) {
// Create N1.
- if (!n1) n1 = IG.newNode(V1);
+ if (!n1) n1 = VN.getOrInsertVN(V1, Top);
+ IG.node(n1); // Ensure that IG.Nodes won't get resized
// Migrate relationships from removed nodes to N1.
- Node *N1 = IG.node(n1);
for (SetVector<unsigned>::iterator I = Remove.begin(), E = Remove.end();
I != E; ++I) {
unsigned n = *I;
- Node *N = IG.node(n);
- for (Node::iterator NI = N->begin(), NE = N->end(); NI != NE; ++NI) {
+ for (Node::iterator NI = IG.node(n)->begin(), NE = IG.node(n)->end();
+ NI != NE; ++NI) {
if (NI->Subtree->DominatedBy(Top)) {
if (NI->To == n1) {
assert((NI->LV & EQ_BIT) && "Node inequal to itself.");
continue;
IG.node(NI->To)->update(n1, reversePredicate(NI->LV), Top);
- N1->update(NI->To, NI->LV, Top);
+ IG.node(n1)->update(NI->To, NI->LV, Top);
}
}
}
// Point V2 (and all items in Remove) to N1.
if (!n2)
- IG.addEquality(n1, V2, Top);
+ VN.addEquality(n1, V2, Top);
else {
for (SetVector<unsigned>::iterator I = Remove.begin(),
E = Remove.end(); I != E; ++I) {
- IG.addEquality(n1, IG.node(*I)->getValue(), Top);
+ VN.addEquality(n1, VN.value(*I), Top);
}
}
// Even when Remove is empty, we still want to process V2.
i = 0;
for (Value *R = V2; i == 0 || i < Remove.size(); ++i) {
- if (i) R = IG.node(Remove[i])->getValue(); // skip n2.
+ if (i) R = VN.value(Remove[i]); // skip n2.
if (Instruction *I2 = dyn_cast<Instruction>(R)) {
- if (below(I2) ||
- Top->DominatedBy(Forest->getNodeForBlock(I2->getParent())))
+ if (aboveOrBelow(I2))
defToOps(I2);
}
for (Value::use_iterator UI = V2->use_begin(), UE = V2->use_end();
Use &TheUse = UI.getUse();
++UI;
if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser())) {
- if (below(I) ||
- Top->DominatedBy(Forest->getNodeForBlock(I->getParent())))
+ if (aboveOrBelow(I))
opsToDef(I);
}
}
Value *V = TheUse.getUser();
if (!V->use_empty()) {
if (Instruction *Inst = dyn_cast<Instruction>(V)) {
- if (below(Inst) ||
- Top->DominatedBy(Forest->getNodeForBlock(Inst->getParent())))
+ if (aboveOrBelow(Inst))
opsToDef(Inst);
}
}
}
public:
- VRPSolver(InequalityGraph &IG, UnreachableBlocks &UB, ValueRanges &VR,
- ETForest *Forest, bool &modified, BasicBlock *TopBB)
- : IG(IG),
+ VRPSolver(ValueNumbering &VN, InequalityGraph &IG, UnreachableBlocks &UB,
+ ValueRanges &VR, DomTreeDFS *DTDFS, bool &modified,
+ BasicBlock *TopBB)
+ : VN(VN),
+ IG(IG),
UB(UB),
VR(VR),
- Forest(Forest),
- Top(Forest->getNodeForBlock(TopBB)),
+ DTDFS(DTDFS),
+ Top(DTDFS->getNodeForBlock(TopBB)),
TopBB(TopBB),
TopInst(NULL),
- modified(modified) {}
+ modified(modified)
+ {
+ assert(Top && "VRPSolver created for unreachable basic block.");
+ }
- VRPSolver(InequalityGraph &IG, UnreachableBlocks &UB, ValueRanges &VR,
- ETForest *Forest, bool &modified, Instruction *TopInst)
- : IG(IG),
+ VRPSolver(ValueNumbering &VN, InequalityGraph &IG, UnreachableBlocks &UB,
+ ValueRanges &VR, DomTreeDFS *DTDFS, bool &modified,
+ Instruction *TopInst)
+ : VN(VN),
+ IG(IG),
UB(UB),
VR(VR),
- Forest(Forest),
+ DTDFS(DTDFS),
+ Top(DTDFS->getNodeForBlock(TopInst->getParent())),
+ TopBB(TopInst->getParent()),
TopInst(TopInst),
modified(modified)
{
- TopBB = TopInst->getParent();
- Top = Forest->getNodeForBlock(TopBB);
+ assert(Top && "VRPSolver created for unreachable basic block.");
+ assert(Top->getBlock() == TopInst->getParent() && "Context mismatch.");
}
bool isRelatedBy(Value *V1, Value *V2, ICmpInst::Predicate Pred) const {
return ConstantExpr::getCompare(Pred, C1, C2) ==
ConstantInt::getTrue();
- if (unsigned n1 = IG.getNode(V1, Top))
- if (unsigned n2 = IG.getNode(V2, Top)) {
- if (n1 == n2) return Pred == ICmpInst::ICMP_EQ ||
- Pred == ICmpInst::ICMP_ULE ||
- Pred == ICmpInst::ICMP_UGE ||
- Pred == ICmpInst::ICMP_SLE ||
- Pred == ICmpInst::ICMP_SGE;
- if (Pred == ICmpInst::ICMP_EQ) return false;
- if (IG.isRelatedBy(n1, n2, Top, cmpInstToLattice(Pred))) return true;
- }
+ unsigned n1 = VN.valueNumber(V1, Top);
+ unsigned n2 = VN.valueNumber(V2, Top);
+
+ if (n1 && n2) {
+ if (n1 == n2) return Pred == ICmpInst::ICMP_EQ ||
+ Pred == ICmpInst::ICMP_ULE ||
+ Pred == ICmpInst::ICMP_UGE ||
+ Pred == ICmpInst::ICMP_SLE ||
+ Pred == ICmpInst::ICMP_SGE;
+ if (Pred == ICmpInst::ICMP_EQ) return false;
+ if (IG.isRelatedBy(n1, n2, Top, cmpInstToLattice(Pred))) return true;
+ if (VR.isRelatedBy(n1, n2, Top, cmpInstToLattice(Pred))) return true;
+ }
+
+ if ((n1 && !n2 && isa<Constant>(V2)) ||
+ (n2 && !n1 && isa<Constant>(V1))) {
+ ConstantRange CR1 = n1 ? VR.range(n1, Top) : VR.range(V1);
+ ConstantRange CR2 = n2 ? VR.range(n2, Top) : VR.range(V2);
+ if (Pred == ICmpInst::ICMP_EQ)
+ return CR1.isSingleElement() &&
+ CR1.getSingleElement() == CR2.getSingleElement();
+
+ return VR.isRelatedBy(CR1, CR2, cmpInstToLattice(Pred));
+ }
if (Pred == ICmpInst::ICMP_EQ) return V1 == V2;
- return VR.isRelatedBy(V1, V2, Top, cmpInstToLattice(Pred));
+ return false;
}
/// add - adds a new property to the work queue
Instruction *I = NULL) {
DOUT << "adding " << *V1 << " " << Pred << " " << *V2;
if (I) DOUT << " context: " << *I;
- else DOUT << " default context";
+ else DOUT << " default context (" << Top->getDFSNumIn() << ")";
DOUT << "\n";
+ assert(V1->getType() == V2->getType() &&
+ "Can't relate two values with different types.");
+
WorkList.push_back(Operation());
Operation &O = WorkList.back();
O.LHS = V1, O.RHS = V2, O.Op = Pred, O.ContextInst = I;
/// new about, find any new relationships between its operands.
void defToOps(Instruction *I) {
Instruction *NewContext = below(I) ? I : TopInst;
- Value *Canonical = IG.canonicalize(I, Top);
+ Value *Canonical = VN.canonicalize(I, Top);
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
const Type *Ty = BO->getType();
assert(!Ty->isFPOrFPVector() && "Float in work queue!");
- Value *Op0 = IG.canonicalize(BO->getOperand(0), Top);
- Value *Op1 = IG.canonicalize(BO->getOperand(1), Top);
+ Value *Op0 = VN.canonicalize(BO->getOperand(0), Top);
+ Value *Op1 = VN.canonicalize(BO->getOperand(1), Top);
// TODO: "and i32 -1, %x" EQ %y then %x EQ %y.
Value *True = SI->getTrueValue();
Value *False = SI->getFalseValue();
if (isRelatedBy(True, False, ICmpInst::ICMP_NE)) {
- if (Canonical == IG.canonicalize(True, Top) ||
+ if (Canonical == VN.canonicalize(True, Top) ||
isRelatedBy(Canonical, False, ICmpInst::ICMP_NE))
add(SI->getCondition(), ConstantInt::getTrue(),
ICmpInst::ICMP_EQ, NewContext);
- else if (Canonical == IG.canonicalize(False, Top) ||
+ else if (Canonical == VN.canonicalize(False, Top) ||
isRelatedBy(Canonical, True, ICmpInst::ICMP_NE))
add(SI->getCondition(), ConstantInt::getFalse(),
ICmpInst::ICMP_EQ, NewContext);
} else if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I)) {
for (GetElementPtrInst::op_iterator OI = GEPI->idx_begin(),
OE = GEPI->idx_end(); OI != OE; ++OI) {
- ConstantInt *Op = dyn_cast<ConstantInt>(IG.canonicalize(*OI, Top));
+ ConstantInt *Op = dyn_cast<ConstantInt>(VN.canonicalize(*OI, Top));
if (!Op || !Op->isZero()) return;
}
// TODO: The GEPI indices are all zero. Copy from definition to operand,
add(Ptr, Constant::getNullValue(Ptr->getType()), ICmpInst::ICMP_NE,
NewContext);
}
+ } else if (CastInst *CI = dyn_cast<CastInst>(I)) {
+ const Type *SrcTy = CI->getSrcTy();
+
+ unsigned ci = VN.getOrInsertVN(CI, Top);
+ uint32_t W = VR.typeToWidth(SrcTy);
+ if (!W) return;
+ ConstantRange CR = VR.range(ci, Top);
+
+ if (CR.isFullSet()) return;
+
+ switch (CI->getOpcode()) {
+ default: break;
+ case Instruction::ZExt:
+ case Instruction::SExt:
+ VR.applyRange(VN.getOrInsertVN(CI->getOperand(0), Top),
+ CR.truncate(W), Top, this);
+ break;
+ case Instruction::BitCast:
+ VR.applyRange(VN.getOrInsertVN(CI->getOperand(0), Top),
+ CR, Top, this);
+ break;
+ }
}
- // TODO: CastInst "%a = cast ... %b" where %a is EQ or NE a constant.
}
/// opsToDef - A new relationship was discovered involving one of this
Instruction *NewContext = below(I) ? I : TopInst;
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
- Value *Op0 = IG.canonicalize(BO->getOperand(0), Top);
- Value *Op1 = IG.canonicalize(BO->getOperand(1), Top);
+ Value *Op0 = VN.canonicalize(BO->getOperand(0), Top);
+ Value *Op1 = VN.canonicalize(BO->getOperand(1), Top);
if (ConstantInt *CI0 = dyn_cast<ConstantInt>(Op0))
if (ConstantInt *CI1 = dyn_cast<ConstantInt>(Op1)) {
assert(!Ty->isFPOrFPVector() && "Float in work queue!");
Constant *Zero = Constant::getNullValue(Ty);
- Constant *AllOnes = ConstantInt::getAllOnesValue(Ty);
+ ConstantInt *AllOnes = ConstantInt::getAllOnesValue(Ty);
switch (Opcode) {
default: break;
// "%x = udiv i32 %y, %z" and %x EQ %y then %z EQ 1
Value *Known = Op0, *Unknown = Op1,
- *TheBO = IG.canonicalize(BO, Top);
+ *TheBO = VN.canonicalize(BO, Top);
if (Known != TheBO) std::swap(Known, Unknown);
if (Known == TheBO) {
switch (Opcode) {
if (!isRelatedBy(Known, Zero, ICmpInst::ICMP_NE)) break;
// otherwise, fall-through.
case Instruction::Sub:
- if (Unknown == Op1) break;
+ if (Unknown == Op0) break;
// otherwise, fall-through.
case Instruction::Xor:
case Instruction::Add:
// "%a = icmp ult i32 %b, %c" and %b u>= %c then %a EQ false
// etc.
- Value *Op0 = IG.canonicalize(IC->getOperand(0), Top);
- Value *Op1 = IG.canonicalize(IC->getOperand(1), Top);
+ Value *Op0 = VN.canonicalize(IC->getOperand(0), Top);
+ Value *Op1 = VN.canonicalize(IC->getOperand(1), Top);
ICmpInst::Predicate Pred = IC->getPredicate();
- if (isRelatedBy(Op0, Op1, Pred)) {
+ if (isRelatedBy(Op0, Op1, Pred))
add(IC, ConstantInt::getTrue(), ICmpInst::ICMP_EQ, NewContext);
- } else if (isRelatedBy(Op0, Op1, ICmpInst::getInversePredicate(Pred))) {
+ else if (isRelatedBy(Op0, Op1, ICmpInst::getInversePredicate(Pred)))
add(IC, ConstantInt::getFalse(), ICmpInst::ICMP_EQ, NewContext);
- }
} else if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
if (I->getType()->isFPOrFPVector()) return;
// %x EQ false then %a EQ %c
// %b EQ %c then %a EQ %b
- Value *Canonical = IG.canonicalize(SI->getCondition(), Top);
+ Value *Canonical = VN.canonicalize(SI->getCondition(), Top);
if (Canonical == ConstantInt::getTrue()) {
add(SI, SI->getTrueValue(), ICmpInst::ICMP_EQ, NewContext);
} else if (Canonical == ConstantInt::getFalse()) {
add(SI, SI->getFalseValue(), ICmpInst::ICMP_EQ, NewContext);
- } else if (IG.canonicalize(SI->getTrueValue(), Top) ==
- IG.canonicalize(SI->getFalseValue(), Top)) {
+ } else if (VN.canonicalize(SI->getTrueValue(), Top) ==
+ VN.canonicalize(SI->getFalseValue(), Top)) {
add(SI, SI->getTrueValue(), ICmpInst::ICMP_EQ, NewContext);
}
} else if (CastInst *CI = dyn_cast<CastInst>(I)) {
- const Type *Ty = CI->getDestTy();
- if (Ty->isFPOrFPVector()) return;
+ const Type *DestTy = CI->getDestTy();
+ if (DestTy->isFPOrFPVector()) return;
- if (Constant *C = dyn_cast<Constant>(
- IG.canonicalize(CI->getOperand(0), Top))) {
- add(CI, ConstantExpr::getCast(CI->getOpcode(), C, Ty),
+ Value *Op = VN.canonicalize(CI->getOperand(0), Top);
+ Instruction::CastOps Opcode = CI->getOpcode();
+
+ if (Constant *C = dyn_cast<Constant>(Op)) {
+ add(CI, ConstantExpr::getCast(Opcode, C, DestTy),
ICmpInst::ICMP_EQ, NewContext);
}
- // TODO: "%a = cast ... %b" where %b is NE/LT/GT a constant.
+ uint32_t W = VR.typeToWidth(DestTy);
+ unsigned ci = VN.getOrInsertVN(CI, Top);
+ ConstantRange CR = VR.range(VN.getOrInsertVN(Op, Top), Top);
+
+ if (!CR.isFullSet()) {
+ switch (Opcode) {
+ default: break;
+ case Instruction::ZExt:
+ VR.applyRange(ci, CR.zeroExtend(W), Top, this);
+ break;
+ case Instruction::SExt:
+ VR.applyRange(ci, CR.signExtend(W), Top, this);
+ break;
+ case Instruction::Trunc: {
+ ConstantRange Result = CR.truncate(W);
+ if (!Result.isFullSet())
+ VR.applyRange(ci, Result, Top, this);
+ } break;
+ case Instruction::BitCast:
+ VR.applyRange(ci, CR, Top, this);
+ break;
+ // TODO: other casts?
+ }
+ }
} else if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I)) {
for (GetElementPtrInst::op_iterator OI = GEPI->idx_begin(),
OE = GEPI->idx_end(); OI != OE; ++OI) {
- ConstantInt *Op = dyn_cast<ConstantInt>(IG.canonicalize(*OI, Top));
+ ConstantInt *Op = dyn_cast<ConstantInt>(VN.canonicalize(*OI, Top));
if (!Op || !Op->isZero()) return;
}
// TODO: The GEPI indices are all zero. Copy from operand to definition,
}
/// solve - process the work queue
- /// Return false if a logical contradiction occurs.
void solve() {
//DOUT << "WorkList entry, size: " << WorkList.size() << "\n";
while (!WorkList.empty()) {
Operation &O = WorkList.front();
TopInst = O.ContextInst;
TopBB = O.ContextBB;
- Top = Forest->getNodeForBlock(TopBB);
+ Top = DTDFS->getNodeForBlock(TopBB); // XXX move this into Context
- O.LHS = IG.canonicalize(O.LHS, Top);
- O.RHS = IG.canonicalize(O.RHS, Top);
+ O.LHS = VN.canonicalize(O.LHS, Top);
+ O.RHS = VN.canonicalize(O.RHS, Top);
- assert(O.LHS == IG.canonicalize(O.LHS, Top) && "Canonicalize isn't.");
- assert(O.RHS == IG.canonicalize(O.RHS, Top) && "Canonicalize isn't.");
+ assert(O.LHS == VN.canonicalize(O.LHS, Top) && "Canonicalize isn't.");
+ assert(O.RHS == VN.canonicalize(O.RHS, Top) && "Canonicalize isn't.");
DOUT << "solving " << *O.LHS << " " << O.Op << " " << *O.RHS;
if (O.ContextInst) DOUT << " context inst: " << *O.ContextInst;
else DOUT << " context block: " << O.ContextBB->getName();
DOUT << "\n";
+ DEBUG(VN.dump());
DEBUG(IG.dump());
+ DEBUG(VR.dump());
// If they're both Constant, skip it. Check for contradiction and mark
// the BB as unreachable if so.
}
}
- if (compare(O.LHS, O.RHS)) {
+ if (VN.compare(O.LHS, O.RHS)) {
std::swap(O.LHS, O.RHS);
O.Op = ICmpInst::getSwappedPredicate(O.Op);
}
continue;
}
- unsigned n1 = IG.getNode(O.LHS, Top);
- unsigned n2 = IG.getNode(O.RHS, Top);
+ unsigned n1 = VN.getOrInsertVN(O.LHS, Top);
+ unsigned n2 = VN.getOrInsertVN(O.RHS, Top);
- if (n1 && n1 == n2) {
+ if (n1 == n2) {
if (O.Op != ICmpInst::ICMP_UGE && O.Op != ICmpInst::ICMP_ULE &&
O.Op != ICmpInst::ICMP_SGE && O.Op != ICmpInst::ICMP_SLE)
UB.mark(TopBB);
continue;
}
- if (VR.isRelatedBy(O.LHS, O.RHS, Top, LV) ||
- (n1 && n2 && IG.isRelatedBy(n1, n2, Top, LV))) {
+ if (VR.isRelatedBy(n1, n2, Top, LV) ||
+ IG.isRelatedBy(n1, n2, Top, LV)) {
WorkList.pop_front();
continue;
}
- VR.addInequality(O.LHS, O.RHS, Top, LV, this);
+ VR.addInequality(n1, n2, Top, LV, this);
if ((!isa<ConstantInt>(O.RHS) && !isa<ConstantInt>(O.LHS)) ||
- LV == NE) {
- if (!n1) n1 = IG.newNode(O.LHS);
- if (!n2) n2 = IG.newNode(O.RHS);
+ LV == NE)
IG.addInequality(n1, n2, Top, LV);
- }
if (Instruction *I1 = dyn_cast<Instruction>(O.LHS)) {
- if (below(I1) ||
- Top->DominatedBy(Forest->getNodeForBlock(I1->getParent())))
+ if (aboveOrBelow(I1))
defToOps(I1);
}
if (isa<Instruction>(O.LHS) || isa<Argument>(O.LHS)) {
Use &TheUse = UI.getUse();
++UI;
if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser())) {
- if (below(I) ||
- Top->DominatedBy(Forest->getNodeForBlock(I->getParent())))
+ if (aboveOrBelow(I))
opsToDef(I);
}
}
}
if (Instruction *I2 = dyn_cast<Instruction>(O.RHS)) {
- if (below(I2) ||
- Top->DominatedBy(Forest->getNodeForBlock(I2->getParent())))
+ if (aboveOrBelow(I2))
defToOps(I2);
}
if (isa<Instruction>(O.RHS) || isa<Argument>(O.RHS)) {
Use &TheUse = UI.getUse();
++UI;
if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser())) {
- if (below(I) ||
- Top->DominatedBy(Forest->getNodeForBlock(I->getParent())))
-
+ if (aboveOrBelow(I))
opsToDef(I);
}
}
}
};
- void ValueRanges::addToWorklist(Value *V, const APInt *I,
+ void ValueRanges::addToWorklist(Value *V, Constant *C,
ICmpInst::Predicate Pred, VRPSolver *VRP) {
- VRP->add(V, ConstantInt::get(*I), Pred, VRP->TopInst);
+ VRP->add(V, C, Pred, VRP->TopInst);
}
-#ifndef NDEBUG
- bool ValueRanges::isCanonical(Value *V, ETNode *Subtree, VRPSolver *VRP) {
- return V == VRP->IG.canonicalize(V, Subtree);
+ void ValueRanges::markBlock(VRPSolver *VRP) {
+ VRP->UB.mark(VRP->TopBB);
}
-#endif
/// PredicateSimplifier - This class is a simplifier that replaces
/// one equivalent variable with another. It also tracks what
/// can't be equal and will solve setcc instructions when possible.
/// @brief Root of the predicate simplifier optimization.
class VISIBILITY_HIDDEN PredicateSimplifier : public FunctionPass {
- DominatorTree *DT;
- ETForest *Forest;
+ DomTreeDFS *DTDFS;
bool modified;
+ ValueNumbering *VN;
InequalityGraph *IG;
UnreachableBlocks UB;
ValueRanges *VR;
- std::vector<DominatorTree::Node *> WorkList;
+ std::vector<DomTreeDFS::Node *> WorkList;
public:
+ static char ID; // Pass identification, replacement for typeid
+ PredicateSimplifier() : FunctionPass((intptr_t)&ID) {}
+
bool runOnFunction(Function &F);
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequiredID(BreakCriticalEdgesID);
AU.addRequired<DominatorTree>();
- AU.addRequired<ETForest>();
+ AU.addRequired<TargetData>();
+ AU.addPreserved<TargetData>();
}
private:
- /// Forwards - Adds new properties into PropertySet and uses them to
+ /// Forwards - Adds new properties to VRPSolver and uses them to
/// simplify instructions. Because new properties sometimes apply to
/// a transition from one BasicBlock to another, this will use the
/// PredicateSimplifier::proceedToSuccessor(s) interface to enter the
- /// basic block with the new PropertySet.
+ /// basic block.
/// @brief Performs abstract execution of the program.
class VISIBILITY_HIDDEN Forwards : public InstVisitor<Forwards> {
friend class InstVisitor<Forwards>;
PredicateSimplifier *PS;
- DominatorTree::Node *DTNode;
+ DomTreeDFS::Node *DTNode;
public:
+ ValueNumbering &VN;
InequalityGraph &IG;
UnreachableBlocks &UB;
ValueRanges &VR;
- Forwards(PredicateSimplifier *PS, DominatorTree::Node *DTNode)
- : PS(PS), DTNode(DTNode), IG(*PS->IG), UB(PS->UB), VR(*PS->VR) {}
+ Forwards(PredicateSimplifier *PS, DomTreeDFS::Node *DTNode)
+ : PS(PS), DTNode(DTNode), VN(*PS->VN), IG(*PS->IG), UB(PS->UB),
+ VR(*PS->VR) {}
void visitTerminatorInst(TerminatorInst &TI);
void visitBranchInst(BranchInst &BI);
// Used by terminator instructions to proceed from the current basic
// block to the next. Verifies that "current" dominates "next",
// then calls visitBasicBlock.
- void proceedToSuccessors(DominatorTree::Node *Current) {
- for (DominatorTree::Node::iterator I = Current->begin(),
+ void proceedToSuccessors(DomTreeDFS::Node *Current) {
+ for (DomTreeDFS::Node::iterator I = Current->begin(),
E = Current->end(); I != E; ++I) {
WorkList.push_back(*I);
}
}
- void proceedToSuccessor(DominatorTree::Node *Next) {
+ void proceedToSuccessor(DomTreeDFS::Node *Next) {
WorkList.push_back(Next);
}
// Visits each instruction in the basic block.
- void visitBasicBlock(DominatorTree::Node *Node) {
+ void visitBasicBlock(DomTreeDFS::Node *Node) {
BasicBlock *BB = Node->getBlock();
- ETNode *ET = Forest->getNodeForBlock(BB);
DOUT << "Entering Basic Block: " << BB->getName()
- << " (" << ET->getDFSNumIn() << ")\n";
+ << " (" << Node->getDFSNumIn() << ")\n";
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) {
- visitInstruction(I++, Node, ET);
+ visitInstruction(I++, Node);
}
}
- // Tries to simplify each Instruction and add new properties to
- // the PropertySet.
- void visitInstruction(Instruction *I, DominatorTree::Node *DT, ETNode *ET) {
+ // Tries to simplify each Instruction and add new properties.
+ void visitInstruction(Instruction *I, DomTreeDFS::Node *DT) {
DOUT << "Considering instruction " << *I << "\n";
+ DEBUG(VN->dump());
DEBUG(IG->dump());
+ DEBUG(VR->dump());
// Sometimes instructions are killed in earlier analysis.
if (isInstructionTriviallyDead(I)) {
++NumSimple;
modified = true;
- IG->remove(I);
+ if (unsigned n = VN->valueNumber(I, DTDFS->getRootNode()))
+ if (VN->value(n) == I) IG->remove(n);
+ VN->remove(I);
I->eraseFromParent();
return;
}
#ifndef NDEBUG
// Try to replace the whole instruction.
- Value *V = IG->canonicalize(I, ET);
+ Value *V = VN->canonicalize(I, DT);
assert(V == I && "Late instruction canonicalization.");
if (V != I) {
modified = true;
++NumInstruction;
DOUT << "Removing " << *I << ", replacing with " << *V << "\n";
- IG->remove(I);
+ if (unsigned n = VN->valueNumber(I, DTDFS->getRootNode()))
+ if (VN->value(n) == I) IG->remove(n);
+ VN->remove(I);
I->replaceAllUsesWith(V);
I->eraseFromParent();
return;
// Try to substitute operands.
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
Value *Oper = I->getOperand(i);
- Value *V = IG->canonicalize(Oper, ET);
+ Value *V = VN->canonicalize(Oper, DT);
assert(V == Oper && "Late operand canonicalization.");
if (V != Oper) {
modified = true;
};
bool PredicateSimplifier::runOnFunction(Function &F) {
- DT = &getAnalysis<DominatorTree>();
- Forest = &getAnalysis<ETForest>();
-
- Forest->updateDFSNumbers(); // XXX: should only act when numbers are out of date
+ DominatorTree *DT = &getAnalysis<DominatorTree>();
+ DTDFS = new DomTreeDFS(DT);
+ TargetData *TD = &getAnalysis<TargetData>();
DOUT << "Entering Function: " << F.getName() << "\n";
modified = false;
- BasicBlock *RootBlock = &F.getEntryBlock();
- IG = new InequalityGraph(Forest->getNodeForBlock(RootBlock));
- VR = new ValueRanges();
- WorkList.push_back(DT->getRootNode());
+ DomTreeDFS::Node *Root = DTDFS->getRootNode();
+ VN = new ValueNumbering(DTDFS);
+ IG = new InequalityGraph(*VN, Root);
+ VR = new ValueRanges(*VN, TD);
+ WorkList.push_back(Root);
do {
- DominatorTree::Node *DTNode = WorkList.back();
+ DomTreeDFS::Node *DTNode = WorkList.back();
WorkList.pop_back();
if (!UB.isDead(DTNode->getBlock())) visitBasicBlock(DTNode);
} while (!WorkList.empty());
+ delete DTDFS;
delete VR;
delete IG;
return;
}
- for (DominatorTree::Node::iterator I = DTNode->begin(), E = DTNode->end();
+ for (DomTreeDFS::Node::iterator I = DTNode->begin(), E = DTNode->end();
I != E; ++I) {
BasicBlock *Dest = (*I)->getBlock();
DOUT << "Branch thinking about %" << Dest->getName()
- << "(" << PS->Forest->getNodeForBlock(Dest)->getDFSNumIn() << ")\n";
+ << "(" << PS->DTDFS->getNodeForBlock(Dest)->getDFSNumIn() << ")\n";
if (Dest == TrueDest) {
DOUT << "(" << DTNode->getBlock()->getName() << ") true set:\n";
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, Dest);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, Dest);
VRP.add(ConstantInt::getTrue(), Condition, ICmpInst::ICMP_EQ);
VRP.solve();
+ DEBUG(VN.dump());
DEBUG(IG.dump());
+ DEBUG(VR.dump());
} else if (Dest == FalseDest) {
DOUT << "(" << DTNode->getBlock()->getName() << ") false set:\n";
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, Dest);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, Dest);
VRP.add(ConstantInt::getFalse(), Condition, ICmpInst::ICMP_EQ);
VRP.solve();
+ DEBUG(VN.dump());
DEBUG(IG.dump());
+ DEBUG(VR.dump());
}
PS->proceedToSuccessor(*I);
// Set the EQProperty in each of the cases BBs, and the NEProperties
// in the default BB.
- for (DominatorTree::Node::iterator I = DTNode->begin(), E = DTNode->end();
+ for (DomTreeDFS::Node::iterator I = DTNode->begin(), E = DTNode->end();
I != E; ++I) {
BasicBlock *BB = (*I)->getBlock();
DOUT << "Switch thinking about BB %" << BB->getName()
- << "(" << PS->Forest->getNodeForBlock(BB)->getDFSNumIn() << ")\n";
+ << "(" << PS->DTDFS->getNodeForBlock(BB)->getDFSNumIn() << ")\n";
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, BB);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, BB);
if (BB == SI.getDefaultDest()) {
for (unsigned i = 1, e = SI.getNumCases(); i < e; ++i)
if (SI.getSuccessor(i) != BB)
}
void PredicateSimplifier::Forwards::visitAllocaInst(AllocaInst &AI) {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &AI);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &AI);
VRP.add(Constant::getNullValue(AI.getType()), &AI, ICmpInst::ICMP_NE);
VRP.solve();
}
// avoid "load uint* null" -> null NE null.
if (isa<Constant>(Ptr)) return;
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &LI);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &LI);
VRP.add(Constant::getNullValue(Ptr->getType()), Ptr, ICmpInst::ICMP_NE);
VRP.solve();
}
Value *Ptr = SI.getPointerOperand();
if (isa<Constant>(Ptr)) return;
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &SI);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &SI);
VRP.add(Constant::getNullValue(Ptr->getType()), Ptr, ICmpInst::ICMP_NE);
VRP.solve();
}
void PredicateSimplifier::Forwards::visitSExtInst(SExtInst &SI) {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &SI);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &SI);
uint32_t SrcBitWidth = cast<IntegerType>(SI.getSrcTy())->getBitWidth();
uint32_t DstBitWidth = cast<IntegerType>(SI.getDestTy())->getBitWidth();
- APInt Min(APInt::getSignedMinValue(SrcBitWidth));
- APInt Max(APInt::getSignedMaxValue(SrcBitWidth));
- Min.sext(DstBitWidth);
- Max.sext(DstBitWidth);
+ APInt Min(APInt::getHighBitsSet(DstBitWidth, DstBitWidth-SrcBitWidth+1));
+ APInt Max(APInt::getLowBitsSet(DstBitWidth, SrcBitWidth-1));
VRP.add(ConstantInt::get(Min), &SI, ICmpInst::ICMP_SLE);
VRP.add(ConstantInt::get(Max), &SI, ICmpInst::ICMP_SGE);
VRP.solve();
}
void PredicateSimplifier::Forwards::visitZExtInst(ZExtInst &ZI) {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &ZI);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &ZI);
uint32_t SrcBitWidth = cast<IntegerType>(ZI.getSrcTy())->getBitWidth();
uint32_t DstBitWidth = cast<IntegerType>(ZI.getDestTy())->getBitWidth();
- APInt Max(APInt::getMaxValue(SrcBitWidth));
- Max.zext(DstBitWidth);
+ APInt Max(APInt::getLowBitsSet(DstBitWidth, SrcBitWidth));
VRP.add(ConstantInt::get(Max), &ZI, ICmpInst::ICMP_UGE);
VRP.solve();
}
case Instruction::UDiv:
case Instruction::SDiv: {
Value *Divisor = BO.getOperand(1);
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(Constant::getNullValue(Divisor->getType()), Divisor,
ICmpInst::ICMP_NE);
VRP.solve();
switch (ops) {
default: break;
case Instruction::Shl: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(0), ICmpInst::ICMP_UGE);
VRP.solve();
} break;
case Instruction::AShr: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(0), ICmpInst::ICMP_SLE);
VRP.solve();
} break;
case Instruction::LShr:
case Instruction::UDiv: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(0), ICmpInst::ICMP_ULE);
VRP.solve();
} break;
case Instruction::URem: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(1), ICmpInst::ICMP_ULE);
VRP.solve();
} break;
case Instruction::And: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(0), ICmpInst::ICMP_ULE);
VRP.add(&BO, BO.getOperand(1), ICmpInst::ICMP_ULE);
VRP.solve();
} break;
case Instruction::Or: {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &BO);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &BO);
VRP.add(&BO, BO.getOperand(0), ICmpInst::ICMP_UGE);
VRP.add(&BO, BO.getOperand(1), ICmpInst::ICMP_UGE);
VRP.solve();
// Eg., if x = [0, 4) and we're being asked icmp uge %x, 3 then change
// the predicate to eq.
+ // XXX: once we do full PHI handling, modifying the instruction in the
+ // Forwards visitor will cause missed optimizations.
+
ICmpInst::Predicate Pred = IC.getPredicate();
+ switch (Pred) {
+ default: break;
+ case ICmpInst::ICMP_ULE: Pred = ICmpInst::ICMP_ULT; break;
+ case ICmpInst::ICMP_UGE: Pred = ICmpInst::ICMP_UGT; break;
+ case ICmpInst::ICMP_SLE: Pred = ICmpInst::ICMP_SLT; break;
+ case ICmpInst::ICMP_SGE: Pred = ICmpInst::ICMP_SGT; break;
+ }
+ if (Pred != IC.getPredicate()) {
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &IC);
+ if (VRP.isRelatedBy(IC.getOperand(1), IC.getOperand(0),
+ ICmpInst::ICMP_NE)) {
+ ++NumSnuggle;
+ PS->modified = true;
+ IC.setPredicate(Pred);
+ }
+ }
+
+ Pred = IC.getPredicate();
+
if (ConstantInt *Op1 = dyn_cast<ConstantInt>(IC.getOperand(1))) {
ConstantInt *NextVal = 0;
- switch(Pred) {
+ switch (Pred) {
default: break;
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_ULT:
if (Op1->getValue() != 0)
- NextVal = cast<ConstantInt>(ConstantExpr::getSub(
- Op1, ConstantInt::get(Op1->getType(), 1)));
+ NextVal = ConstantInt::get(Op1->getValue()-1);
break;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_UGT:
if (!Op1->getValue().isAllOnesValue())
- NextVal = cast<ConstantInt>(ConstantExpr::getAdd(
- Op1, ConstantInt::get(Op1->getType(), 1)));
+ NextVal = ConstantInt::get(Op1->getValue()+1);
break;
}
if (NextVal) {
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &IC);
+ VRPSolver VRP(VN, IG, UB, VR, PS->DTDFS, PS->modified, &IC);
if (VRP.isRelatedBy(IC.getOperand(0), NextVal,
ICmpInst::getInversePredicate(Pred))) {
ICmpInst *NewIC = new ICmpInst(ICmpInst::ICMP_EQ, IC.getOperand(0),
NextVal, "", &IC);
NewIC->takeName(&IC);
IC.replaceAllUsesWith(NewIC);
- IG.remove(&IC); // XXX: prove this isn't necessary
+
+ // XXX: prove this isn't necessary
+ if (unsigned n = VN.valueNumber(&IC, PS->DTDFS->getRootNode()))
+ if (VN.value(n) == &IC) IG.remove(n);
+ VN.remove(&IC);
+
IC.eraseFromParent();
++NumSnuggle;
PS->modified = true;
- return;
}
}
}
-
- switch(Pred) {
- default: return;
- case ICmpInst::ICMP_ULE: Pred = ICmpInst::ICMP_ULT; break;
- case ICmpInst::ICMP_UGE: Pred = ICmpInst::ICMP_UGT; break;
- case ICmpInst::ICMP_SLE: Pred = ICmpInst::ICMP_SLT; break;
- case ICmpInst::ICMP_SGE: Pred = ICmpInst::ICMP_SGT; break;
- }
- VRPSolver VRP(IG, UB, VR, PS->Forest, PS->modified, &IC);
- if (VRP.isRelatedBy(IC.getOperand(1), IC.getOperand(0), Pred)) {
- ++NumSnuggle;
- PS->modified = true;
- IC.setPredicate(Pred);
- }
}
-
- RegisterPass<PredicateSimplifier> X("predsimplify",
- "Predicate Simplifier");
}
+char PredicateSimplifier::ID = 0;
+static RegisterPass<PredicateSimplifier>
+X("predsimplify", "Predicate Simplifier");
+
FunctionPass *llvm::createPredicateSimplifierPass() {
return new PredicateSimplifier();
}