//
// This pass looks for equivalent functions that are mergable and folds them.
//
-// A hash is computed from the function, based on its type and number of
-// basic blocks.
+// Order relation is defined on set of functions. It was made through
+// special function comparison procedure that returns
+// 0 when functions are equal,
+// -1 when Left function is less than right function, and
+// 1 for opposite case. We need total-ordering, so we need to maintain
+// four properties on the functions set:
+// a <= a (reflexivity)
+// if a <= b and b <= a then a = b (antisymmetry)
+// if a <= b and b <= c then a <= c (transitivity).
+// for all a and b: a <= b or b <= a (totality).
//
-// Once all hashes are computed, we perform an expensive equality comparison
-// on each function pair. This takes n^2/2 comparisons per bucket, so it's
-// important that the hash function be high quality. The equality comparison
-// iterates through each instruction in each basic block.
+// Comparison iterates through each instruction in each basic block.
+// Functions are kept on binary tree. For each new function F we perform
+// lookup in binary tree.
+// In practice it works the following way:
+// -- We define Function* container class with custom "operator<" (FunctionPtr).
+// -- "FunctionPtr" instances are stored in std::set collection, so every
+// std::set::insert operation will give you result in log(N) time.
//
// When a match is found the functions are folded. If both functions are
// overridable, we move the functionality into a new internal function and
//
// Many functions have their address taken by the virtual function table for
// the object they belong to. However, as long as it's only used for a lookup
-// and call, this is irrelevant, and we'd like to fold such implementations.
+// and call, this is irrelevant, and we'd like to fold such functions.
//
-// * switch from n^2 pair-wise comparisons to an n-way comparison for each
-// bucket.
-//
-// * be smarter about bitcast.
+// * be smarter about bitcasts.
//
// In order to fold functions, we will sometimes add either bitcast instructions
// or bitcast constant expressions. Unfortunately, this can confound further
// analysis since the two functions differ where one has a bitcast and the
-// other doesn't. We should learn to peer through bitcasts without imposing bad
-// performance properties.
+// other doesn't. We should learn to look through bitcasts.
+//
+// * Compare complex types with pointer types inside.
+// * Compare cross-reference cases.
+// * Compare complex expressions.
+//
+// All the three issues above could be described as ability to prove that
+// fA == fB == fC == fE == fF == fG in example below:
+//
+// void fA() {
+// fB();
+// }
+// void fB() {
+// fA();
+// }
//
-// * emit aliases for ELF
+// void fE() {
+// fF();
+// }
+// void fF() {
+// fG();
+// }
+// void fG() {
+// fE();
+// }
//
-// ELF supports symbol aliases which are represented with GlobalAlias in the
-// Module, and we could emit them in the case that the addresses don't need to
-// be distinct. The problem is that not all object formats support equivalent
-// functionality. There's a few approaches to this problem;
-// a) teach codegen to lower global aliases to thunks on platforms which don't
-// support them.
-// b) always emit thunks, and create a separate thunk-to-alias pass which
-// runs on ELF systems. This has the added benefit of transforming other
-// thunks such as those produced by a C++ frontend into aliases when legal
-// to do so.
+// Simplest cross-reference case (fA <--> fB) was implemented in previous
+// versions of MergeFunctions, though it presented only in two function pairs
+// in test-suite (that counts >50k functions)
+// Though possibility to detect complex cross-referencing (e.g.: A->B->C->D->A)
+// could cover much more cases.
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "mergefunc"
#include "llvm/Transforms/IPO.h"
-#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/FoldingSet.h"
+#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
-#include "llvm/Constants.h"
-#include "llvm/InlineAsm.h"
-#include "llvm/Instructions.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Module.h"
+#include "llvm/IR/CallSite.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
+#include "llvm/IR/ValueHandle.h"
#include "llvm/Pass.h"
-#include "llvm/Support/CallSite.h"
+#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
-#include <map>
#include <vector>
using namespace llvm;
+#define DEBUG_TYPE "mergefunc"
+
STATISTIC(NumFunctionsMerged, "Number of functions merged");
+STATISTIC(NumThunksWritten, "Number of thunks generated");
+STATISTIC(NumAliasesWritten, "Number of aliases generated");
+STATISTIC(NumDoubleWeak, "Number of new functions created");
+
+static cl::opt<unsigned> NumFunctionsForSanityCheck(
+ "mergefunc-sanity",
+ cl::desc("How many functions in module could be used for "
+ "MergeFunctions pass sanity check. "
+ "'0' disables this check. Works only with '-debug' key."),
+ cl::init(0), cl::Hidden);
namespace {
- /// MergeFunctions finds functions which will generate identical machine code,
- /// by considering all pointer types to be equivalent. Once identified,
- /// MergeFunctions will fold them by replacing a call to one to a call to a
- /// bitcast of the other.
+
+/// FunctionComparator - Compares two functions to determine whether or not
+/// they will generate machine code with the same behaviour. DataLayout is
+/// used if available. The comparator always fails conservatively (erring on the
+/// side of claiming that two functions are different).
+class FunctionComparator {
+public:
+ FunctionComparator(const DataLayout *DL, const Function *F1,
+ const Function *F2)
+ : FnL(F1), FnR(F2), DL(DL) {}
+
+ /// Test whether the two functions have equivalent behaviour.
+ int compare();
+
+private:
+ /// Test whether two basic blocks have equivalent behaviour.
+ int compare(const BasicBlock *BBL, const BasicBlock *BBR);
+
+ /// Constants comparison.
+ /// Its analog to lexicographical comparison between hypothetical numbers
+ /// of next format:
+ /// <bitcastability-trait><raw-bit-contents>
+ ///
+ /// 1. Bitcastability.
+ /// Check whether L's type could be losslessly bitcasted to R's type.
+ /// On this stage method, in case when lossless bitcast is not possible
+ /// method returns -1 or 1, thus also defining which type is greater in
+ /// context of bitcastability.
+ /// Stage 0: If types are equal in terms of cmpTypes, then we can go straight
+ /// to the contents comparison.
+ /// If types differ, remember types comparison result and check
+ /// whether we still can bitcast types.
+ /// Stage 1: Types that satisfies isFirstClassType conditions are always
+ /// greater then others.
+ /// Stage 2: Vector is greater then non-vector.
+ /// If both types are vectors, then vector with greater bitwidth is
+ /// greater.
+ /// If both types are vectors with the same bitwidth, then types
+ /// are bitcastable, and we can skip other stages, and go to contents
+ /// comparison.
+ /// Stage 3: Pointer types are greater than non-pointers. If both types are
+ /// pointers of the same address space - go to contents comparison.
+ /// Different address spaces: pointer with greater address space is
+ /// greater.
+ /// Stage 4: Types are neither vectors, nor pointers. And they differ.
+ /// We don't know how to bitcast them. So, we better don't do it,
+ /// and return types comparison result (so it determines the
+ /// relationship among constants we don't know how to bitcast).
///
- struct MergeFunctions : public ModulePass {
- static char ID; // Pass identification, replacement for typeid
- MergeFunctions() : ModulePass(&ID) {}
+ /// Just for clearance, let's see how the set of constants could look
+ /// on single dimension axis:
+ ///
+ /// [NFCT], [FCT, "others"], [FCT, pointers], [FCT, vectors]
+ /// Where: NFCT - Not a FirstClassType
+ /// FCT - FirstClassTyp:
+ ///
+ /// 2. Compare raw contents.
+ /// It ignores types on this stage and only compares bits from L and R.
+ /// Returns 0, if L and R has equivalent contents.
+ /// -1 or 1 if values are different.
+ /// Pretty trivial:
+ /// 2.1. If contents are numbers, compare numbers.
+ /// Ints with greater bitwidth are greater. Ints with same bitwidths
+ /// compared by their contents.
+ /// 2.2. "And so on". Just to avoid discrepancies with comments
+ /// perhaps it would be better to read the implementation itself.
+ /// 3. And again about overall picture. Let's look back at how the ordered set
+ /// of constants will look like:
+ /// [NFCT], [FCT, "others"], [FCT, pointers], [FCT, vectors]
+ ///
+ /// Now look, what could be inside [FCT, "others"], for example:
+ /// [FCT, "others"] =
+ /// [
+ /// [double 0.1], [double 1.23],
+ /// [i32 1], [i32 2],
+ /// { double 1.0 }, ; StructTyID, NumElements = 1
+ /// { i32 1 }, ; StructTyID, NumElements = 1
+ /// { double 1, i32 1 }, ; StructTyID, NumElements = 2
+ /// { i32 1, double 1 } ; StructTyID, NumElements = 2
+ /// ]
+ ///
+ /// Let's explain the order. Float numbers will be less than integers, just
+ /// because of cmpType terms: FloatTyID < IntegerTyID.
+ /// Floats (with same fltSemantics) are sorted according to their value.
+ /// Then you can see integers, and they are, like a floats,
+ /// could be easy sorted among each others.
+ /// The structures. Structures are grouped at the tail, again because of their
+ /// TypeID: StructTyID > IntegerTyID > FloatTyID.
+ /// Structures with greater number of elements are greater. Structures with
+ /// greater elements going first are greater.
+ /// The same logic with vectors, arrays and other possible complex types.
+ ///
+ /// Bitcastable constants.
+ /// Let's assume, that some constant, belongs to some group of
+ /// "so-called-equal" values with different types, and at the same time
+ /// belongs to another group of constants with equal types
+ /// and "really" equal values.
+ ///
+ /// Now, prove that this is impossible:
+ ///
+ /// If constant A with type TyA is bitcastable to B with type TyB, then:
+ /// 1. All constants with equal types to TyA, are bitcastable to B. Since
+ /// those should be vectors (if TyA is vector), pointers
+ /// (if TyA is pointer), or else (if TyA equal to TyB), those types should
+ /// be equal to TyB.
+ /// 2. All constants with non-equal, but bitcastable types to TyA, are
+ /// bitcastable to B.
+ /// Once again, just because we allow it to vectors and pointers only.
+ /// This statement could be expanded as below:
+ /// 2.1. All vectors with equal bitwidth to vector A, has equal bitwidth to
+ /// vector B, and thus bitcastable to B as well.
+ /// 2.2. All pointers of the same address space, no matter what they point to,
+ /// bitcastable. So if C is pointer, it could be bitcasted to A and to B.
+ /// So any constant equal or bitcastable to A is equal or bitcastable to B.
+ /// QED.
+ ///
+ /// In another words, for pointers and vectors, we ignore top-level type and
+ /// look at their particular properties (bit-width for vectors, and
+ /// address space for pointers).
+ /// If these properties are equal - compare their contents.
+ int cmpConstants(const Constant *L, const Constant *R);
+
+ /// Assign or look up previously assigned numbers for the two values, and
+ /// return whether the numbers are equal. Numbers are assigned in the order
+ /// visited.
+ /// Comparison order:
+ /// Stage 0: Value that is function itself is always greater then others.
+ /// If left and right values are references to their functions, then
+ /// they are equal.
+ /// Stage 1: Constants are greater than non-constants.
+ /// If both left and right are constants, then the result of
+ /// cmpConstants is used as cmpValues result.
+ /// Stage 2: InlineAsm instances are greater than others. If both left and
+ /// right are InlineAsm instances, InlineAsm* pointers casted to
+ /// integers and compared as numbers.
+ /// Stage 3: For all other cases we compare order we meet these values in
+ /// their functions. If right value was met first during scanning,
+ /// then left value is greater.
+ /// In another words, we compare serial numbers, for more details
+ /// see comments for sn_mapL and sn_mapR.
+ int cmpValues(const Value *L, const Value *R);
+
+ /// Compare two Instructions for equivalence, similar to
+ /// Instruction::isSameOperationAs but with modifications to the type
+ /// comparison.
+ /// Stages are listed in "most significant stage first" order:
+ /// On each stage below, we do comparison between some left and right
+ /// operation parts. If parts are non-equal, we assign parts comparison
+ /// result to the operation comparison result and exit from method.
+ /// Otherwise we proceed to the next stage.
+ /// Stages:
+ /// 1. Operations opcodes. Compared as numbers.
+ /// 2. Number of operands.
+ /// 3. Operation types. Compared with cmpType method.
+ /// 4. Compare operation subclass optional data as stream of bytes:
+ /// just convert it to integers and call cmpNumbers.
+ /// 5. Compare in operation operand types with cmpType in
+ /// most significant operand first order.
+ /// 6. Last stage. Check operations for some specific attributes.
+ /// For example, for Load it would be:
+ /// 6.1.Load: volatile (as boolean flag)
+ /// 6.2.Load: alignment (as integer numbers)
+ /// 6.3.Load: synch-scope (as integer numbers)
+ /// 6.4.Load: range metadata (as integer numbers)
+ /// On this stage its better to see the code, since its not more than 10-15
+ /// strings for particular instruction, and could change sometimes.
+ int cmpOperations(const Instruction *L, const Instruction *R) const;
+
+ /// Compare two GEPs for equivalent pointer arithmetic.
+ /// Parts to be compared for each comparison stage,
+ /// most significant stage first:
+ /// 1. Address space. As numbers.
+ /// 2. Constant offset, (if "DataLayout *DL" field is not NULL,
+ /// using GEPOperator::accumulateConstantOffset method).
+ /// 3. Pointer operand type (using cmpType method).
+ /// 4. Number of operands.
+ /// 5. Compare operands, using cmpValues method.
+ int cmpGEPs(const GEPOperator *GEPL, const GEPOperator *GEPR);
+ int cmpGEPs(const GetElementPtrInst *GEPL, const GetElementPtrInst *GEPR) {
+ return cmpGEPs(cast<GEPOperator>(GEPL), cast<GEPOperator>(GEPR));
+ }
- bool runOnModule(Module &M);
- };
-}
+ /// cmpType - compares two types,
+ /// defines total ordering among the types set.
+ ///
+ /// Return values:
+ /// 0 if types are equal,
+ /// -1 if Left is less than Right,
+ /// +1 if Left is greater than Right.
+ ///
+ /// Description:
+ /// Comparison is broken onto stages. Like in lexicographical comparison
+ /// stage coming first has higher priority.
+ /// On each explanation stage keep in mind total ordering properties.
+ ///
+ /// 0. Before comparison we coerce pointer types of 0 address space to
+ /// integer.
+ /// We also don't bother with same type at left and right, so
+ /// just return 0 in this case.
+ ///
+ /// 1. If types are of different kind (different type IDs).
+ /// Return result of type IDs comparison, treating them as numbers.
+ /// 2. If types are vectors or integers, compare Type* values as numbers.
+ /// 3. Types has same ID, so check whether they belongs to the next group:
+ /// * Void
+ /// * Float
+ /// * Double
+ /// * X86_FP80
+ /// * FP128
+ /// * PPC_FP128
+ /// * Label
+ /// * Metadata
+ /// If so - return 0, yes - we can treat these types as equal only because
+ /// their IDs are same.
+ /// 4. If Left and Right are pointers, return result of address space
+ /// comparison (numbers comparison). We can treat pointer types of same
+ /// address space as equal.
+ /// 5. If types are complex.
+ /// Then both Left and Right are to be expanded and their element types will
+ /// be checked with the same way. If we get Res != 0 on some stage, return it.
+ /// Otherwise return 0.
+ /// 6. For all other cases put llvm_unreachable.
+ int cmpTypes(Type *TyL, Type *TyR) const;
+
+ int cmpNumbers(uint64_t L, uint64_t R) const;
+
+ int cmpAPInt(const APInt &L, const APInt &R) const;
+ int cmpAPFloat(const APFloat &L, const APFloat &R) const;
+ int cmpStrings(StringRef L, StringRef R) const;
+ int cmpAttrs(const AttributeSet L, const AttributeSet R) const;
-char MergeFunctions::ID = 0;
-INITIALIZE_PASS(MergeFunctions, "mergefunc", "Merge Functions", false, false);
+ // The two functions undergoing comparison.
+ const Function *FnL, *FnR;
+
+ const DataLayout *DL;
+
+ /// Assign serial numbers to values from left function, and values from
+ /// right function.
+ /// Explanation:
+ /// Being comparing functions we need to compare values we meet at left and
+ /// right sides.
+ /// Its easy to sort things out for external values. It just should be
+ /// the same value at left and right.
+ /// But for local values (those were introduced inside function body)
+ /// we have to ensure they were introduced at exactly the same place,
+ /// and plays the same role.
+ /// Let's assign serial number to each value when we meet it first time.
+ /// Values that were met at same place will be with same serial numbers.
+ /// In this case it would be good to explain few points about values assigned
+ /// to BBs and other ways of implementation (see below).
+ ///
+ /// 1. Safety of BB reordering.
+ /// It's safe to change the order of BasicBlocks in function.
+ /// Relationship with other functions and serial numbering will not be
+ /// changed in this case.
+ /// As follows from FunctionComparator::compare(), we do CFG walk: we start
+ /// from the entry, and then take each terminator. So it doesn't matter how in
+ /// fact BBs are ordered in function. And since cmpValues are called during
+ /// this walk, the numbering depends only on how BBs located inside the CFG.
+ /// So the answer is - yes. We will get the same numbering.
+ ///
+ /// 2. Impossibility to use dominance properties of values.
+ /// If we compare two instruction operands: first is usage of local
+ /// variable AL from function FL, and second is usage of local variable AR
+ /// from FR, we could compare their origins and check whether they are
+ /// defined at the same place.
+ /// But, we are still not able to compare operands of PHI nodes, since those
+ /// could be operands from further BBs we didn't scan yet.
+ /// So it's impossible to use dominance properties in general.
+ DenseMap<const Value*, int> sn_mapL, sn_mapR;
+};
-ModulePass *llvm::createMergeFunctionsPass() {
- return new MergeFunctions();
-}
+class FunctionPtr {
+ AssertingVH<Function> F;
+ const DataLayout *DL;
-// ===----------------------------------------------------------------------===
-// Comparison of functions
-// ===----------------------------------------------------------------------===
-namespace {
-class FunctionComparator {
public:
- FunctionComparator(TargetData *TD, Function *F1, Function *F2)
- : F1(F1), F2(F2), TD(TD) {}
+ FunctionPtr(Function *F, const DataLayout *DL) : F(F), DL(DL) {}
+ Function *getFunc() const { return F; }
+ void release() { F = 0; }
+ bool operator<(const FunctionPtr &RHS) const {
+ return (FunctionComparator(DL, F, RHS.getFunc()).compare()) == -1;
+ }
+};
+}
- // Compare - test whether the two functions have equivalent behaviour.
- bool Compare();
+int FunctionComparator::cmpNumbers(uint64_t L, uint64_t R) const {
+ if (L < R) return -1;
+ if (L > R) return 1;
+ return 0;
+}
-private:
- // Compare - test whether two basic blocks have equivalent behaviour.
- bool Compare(const BasicBlock *BB1, const BasicBlock *BB2);
+int FunctionComparator::cmpAPInt(const APInt &L, const APInt &R) const {
+ if (int Res = cmpNumbers(L.getBitWidth(), R.getBitWidth()))
+ return Res;
+ if (L.ugt(R)) return 1;
+ if (R.ugt(L)) return -1;
+ return 0;
+}
- // getDomain - a value's domain is its parent function if it is specific to a
- // function, or NULL otherwise.
- const Function *getDomain(const Value *V) const;
+int FunctionComparator::cmpAPFloat(const APFloat &L, const APFloat &R) const {
+ if (int Res = cmpNumbers((uint64_t)&L.getSemantics(),
+ (uint64_t)&R.getSemantics()))
+ return Res;
+ return cmpAPInt(L.bitcastToAPInt(), R.bitcastToAPInt());
+}
- // Enumerate - Assign or look up previously assigned numbers for the two
- // values, and return whether the numbers are equal. Numbers are assigned in
- // the order visited.
- bool Enumerate(const Value *V1, const Value *V2);
+int FunctionComparator::cmpStrings(StringRef L, StringRef R) const {
+ // Prevent heavy comparison, compare sizes first.
+ if (int Res = cmpNumbers(L.size(), R.size()))
+ return Res;
- // isEquivalentOperation - Compare two Instructions for equivalence, similar
- // to Instruction::isSameOperationAs but with modifications to the type
- // comparison.
- bool isEquivalentOperation(const Instruction *I1,
- const Instruction *I2) const;
+ // Compare strings lexicographically only when it is necessary: only when
+ // strings are equal in size.
+ return L.compare(R);
+}
- // isEquivalentGEP - Compare two GEPs for equivalent pointer arithmetic.
- bool isEquivalentGEP(const GEPOperator *GEP1, const GEPOperator *GEP2);
- bool isEquivalentGEP(const GetElementPtrInst *GEP1,
- const GetElementPtrInst *GEP2) {
- return isEquivalentGEP(cast<GEPOperator>(GEP1), cast<GEPOperator>(GEP2));
+int FunctionComparator::cmpAttrs(const AttributeSet L,
+ const AttributeSet R) const {
+ if (int Res = cmpNumbers(L.getNumSlots(), R.getNumSlots()))
+ return Res;
+
+ for (unsigned i = 0, e = L.getNumSlots(); i != e; ++i) {
+ AttributeSet::iterator LI = L.begin(i), LE = L.end(i), RI = R.begin(i),
+ RE = R.end(i);
+ for (; LI != LE && RI != RE; ++LI, ++RI) {
+ Attribute LA = *LI;
+ Attribute RA = *RI;
+ if (LA < RA)
+ return -1;
+ if (RA < LA)
+ return 1;
+ }
+ if (LI != LE)
+ return 1;
+ if (RI != RE)
+ return -1;
}
+ return 0;
+}
- // isEquivalentType - Compare two Types, treating all pointer types as equal.
- bool isEquivalentType(const Type *Ty1, const Type *Ty2) const;
+/// Constants comparison:
+/// 1. Check whether type of L constant could be losslessly bitcasted to R
+/// type.
+/// 2. Compare constant contents.
+/// For more details see declaration comments.
+int FunctionComparator::cmpConstants(const Constant *L, const Constant *R) {
+
+ Type *TyL = L->getType();
+ Type *TyR = R->getType();
+
+ // Check whether types are bitcastable. This part is just re-factored
+ // Type::canLosslesslyBitCastTo method, but instead of returning true/false,
+ // we also pack into result which type is "less" for us.
+ int TypesRes = cmpTypes(TyL, TyR);
+ if (TypesRes != 0) {
+ // Types are different, but check whether we can bitcast them.
+ if (!TyL->isFirstClassType()) {
+ if (TyR->isFirstClassType())
+ return -1;
+ // Neither TyL nor TyR are values of first class type. Return the result
+ // of comparing the types
+ return TypesRes;
+ }
+ if (!TyR->isFirstClassType()) {
+ if (TyL->isFirstClassType())
+ return 1;
+ return TypesRes;
+ }
- // The two functions undergoing comparison.
- Function *F1, *F2;
+ // Vector -> Vector conversions are always lossless if the two vector types
+ // have the same size, otherwise not.
+ unsigned TyLWidth = 0;
+ unsigned TyRWidth = 0;
+
+ if (const VectorType *VecTyL = dyn_cast<VectorType>(TyL))
+ TyLWidth = VecTyL->getBitWidth();
+ if (const VectorType *VecTyR = dyn_cast<VectorType>(TyR))
+ TyRWidth = VecTyR->getBitWidth();
+
+ if (TyLWidth != TyRWidth)
+ return cmpNumbers(TyLWidth, TyRWidth);
+
+ // Zero bit-width means neither TyL nor TyR are vectors.
+ if (!TyLWidth) {
+ PointerType *PTyL = dyn_cast<PointerType>(TyL);
+ PointerType *PTyR = dyn_cast<PointerType>(TyR);
+ if (PTyL && PTyR) {
+ unsigned AddrSpaceL = PTyL->getAddressSpace();
+ unsigned AddrSpaceR = PTyR->getAddressSpace();
+ if (int Res = cmpNumbers(AddrSpaceL, AddrSpaceR))
+ return Res;
+ }
+ if (PTyL)
+ return 1;
+ if (PTyR)
+ return -1;
+
+ // TyL and TyR aren't vectors, nor pointers. We don't know how to
+ // bitcast them.
+ return TypesRes;
+ }
+ }
- TargetData *TD;
+ // OK, types are bitcastable, now check constant contents.
- typedef DenseMap<const Value *, unsigned long> IDMap;
- IDMap Map;
- DenseMap<const Function *, IDMap> Domains;
- DenseMap<const Function *, unsigned long> DomainCount;
-};
-}
+ if (L->isNullValue() && R->isNullValue())
+ return TypesRes;
+ if (L->isNullValue() && !R->isNullValue())
+ return 1;
+ if (!L->isNullValue() && R->isNullValue())
+ return -1;
-/// Compute a number which is guaranteed to be equal for two equivalent
-/// functions, but is very likely to be different for different functions. This
-/// needs to be computed as efficiently as possible.
-static unsigned long ProfileFunction(const Function *F) {
- const FunctionType *FTy = F->getFunctionType();
-
- FoldingSetNodeID ID;
- ID.AddInteger(F->size());
- ID.AddInteger(F->getCallingConv());
- ID.AddBoolean(F->hasGC());
- ID.AddBoolean(FTy->isVarArg());
- ID.AddInteger(FTy->getReturnType()->getTypeID());
- for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
- ID.AddInteger(FTy->getParamType(i)->getTypeID());
- return ID.ComputeHash();
+ if (int Res = cmpNumbers(L->getValueID(), R->getValueID()))
+ return Res;
+
+ switch (L->getValueID()) {
+ case Value::UndefValueVal: return TypesRes;
+ case Value::ConstantIntVal: {
+ const APInt &LInt = cast<ConstantInt>(L)->getValue();
+ const APInt &RInt = cast<ConstantInt>(R)->getValue();
+ return cmpAPInt(LInt, RInt);
+ }
+ case Value::ConstantFPVal: {
+ const APFloat &LAPF = cast<ConstantFP>(L)->getValueAPF();
+ const APFloat &RAPF = cast<ConstantFP>(R)->getValueAPF();
+ return cmpAPFloat(LAPF, RAPF);
+ }
+ case Value::ConstantArrayVal: {
+ const ConstantArray *LA = cast<ConstantArray>(L);
+ const ConstantArray *RA = cast<ConstantArray>(R);
+ uint64_t NumElementsL = cast<ArrayType>(TyL)->getNumElements();
+ uint64_t NumElementsR = cast<ArrayType>(TyR)->getNumElements();
+ if (int Res = cmpNumbers(NumElementsL, NumElementsR))
+ return Res;
+ for (uint64_t i = 0; i < NumElementsL; ++i) {
+ if (int Res = cmpConstants(cast<Constant>(LA->getOperand(i)),
+ cast<Constant>(RA->getOperand(i))))
+ return Res;
+ }
+ return 0;
+ }
+ case Value::ConstantStructVal: {
+ const ConstantStruct *LS = cast<ConstantStruct>(L);
+ const ConstantStruct *RS = cast<ConstantStruct>(R);
+ unsigned NumElementsL = cast<StructType>(TyL)->getNumElements();
+ unsigned NumElementsR = cast<StructType>(TyR)->getNumElements();
+ if (int Res = cmpNumbers(NumElementsL, NumElementsR))
+ return Res;
+ for (unsigned i = 0; i != NumElementsL; ++i) {
+ if (int Res = cmpConstants(cast<Constant>(LS->getOperand(i)),
+ cast<Constant>(RS->getOperand(i))))
+ return Res;
+ }
+ return 0;
+ }
+ case Value::ConstantVectorVal: {
+ const ConstantVector *LV = cast<ConstantVector>(L);
+ const ConstantVector *RV = cast<ConstantVector>(R);
+ unsigned NumElementsL = cast<VectorType>(TyL)->getNumElements();
+ unsigned NumElementsR = cast<VectorType>(TyR)->getNumElements();
+ if (int Res = cmpNumbers(NumElementsL, NumElementsR))
+ return Res;
+ for (uint64_t i = 0; i < NumElementsL; ++i) {
+ if (int Res = cmpConstants(cast<Constant>(LV->getOperand(i)),
+ cast<Constant>(RV->getOperand(i))))
+ return Res;
+ }
+ return 0;
+ }
+ case Value::ConstantExprVal: {
+ const ConstantExpr *LE = cast<ConstantExpr>(L);
+ const ConstantExpr *RE = cast<ConstantExpr>(R);
+ unsigned NumOperandsL = LE->getNumOperands();
+ unsigned NumOperandsR = RE->getNumOperands();
+ if (int Res = cmpNumbers(NumOperandsL, NumOperandsR))
+ return Res;
+ for (unsigned i = 0; i < NumOperandsL; ++i) {
+ if (int Res = cmpConstants(cast<Constant>(LE->getOperand(i)),
+ cast<Constant>(RE->getOperand(i))))
+ return Res;
+ }
+ return 0;
+ }
+ case Value::FunctionVal:
+ case Value::GlobalVariableVal:
+ case Value::GlobalAliasVal:
+ default: // Unknown constant, cast L and R pointers to numbers and compare.
+ return cmpNumbers((uint64_t)L, (uint64_t)R);
+ }
}
-/// isEquivalentType - any two pointers are equivalent. Otherwise, standard
-/// type equivalence rules apply.
-bool FunctionComparator::isEquivalentType(const Type *Ty1,
- const Type *Ty2) const {
- if (Ty1 == Ty2)
- return true;
- if (Ty1->getTypeID() != Ty2->getTypeID())
- return false;
+/// cmpType - compares two types,
+/// defines total ordering among the types set.
+/// See method declaration comments for more details.
+int FunctionComparator::cmpTypes(Type *TyL, Type *TyR) const {
+
+ PointerType *PTyL = dyn_cast<PointerType>(TyL);
+ PointerType *PTyR = dyn_cast<PointerType>(TyR);
- switch(Ty1->getTypeID()) {
+ if (DL) {
+ if (PTyL && PTyL->getAddressSpace() == 0) TyL = DL->getIntPtrType(TyL);
+ if (PTyR && PTyR->getAddressSpace() == 0) TyR = DL->getIntPtrType(TyR);
+ }
+
+ if (TyL == TyR)
+ return 0;
+
+ if (int Res = cmpNumbers(TyL->getTypeID(), TyR->getTypeID()))
+ return Res;
+
+ switch (TyL->getTypeID()) {
default:
llvm_unreachable("Unknown type!");
// Fall through in Release mode.
case Type::IntegerTyID:
- case Type::OpaqueTyID:
- // Ty1 == Ty2 would have returned true earlier.
- return false;
+ case Type::VectorTyID:
+ // TyL == TyR would have returned true earlier.
+ return cmpNumbers((uint64_t)TyL, (uint64_t)TyR);
case Type::VoidTyID:
case Type::FloatTyID:
case Type::PPC_FP128TyID:
case Type::LabelTyID:
case Type::MetadataTyID:
- return true;
+ return 0;
case Type::PointerTyID: {
- const PointerType *PTy1 = cast<PointerType>(Ty1);
- const PointerType *PTy2 = cast<PointerType>(Ty2);
- return PTy1->getAddressSpace() == PTy2->getAddressSpace();
+ assert(PTyL && PTyR && "Both types must be pointers here.");
+ return cmpNumbers(PTyL->getAddressSpace(), PTyR->getAddressSpace());
}
case Type::StructTyID: {
- const StructType *STy1 = cast<StructType>(Ty1);
- const StructType *STy2 = cast<StructType>(Ty2);
- if (STy1->getNumElements() != STy2->getNumElements())
- return false;
-
- if (STy1->isPacked() != STy2->isPacked())
- return false;
+ StructType *STyL = cast<StructType>(TyL);
+ StructType *STyR = cast<StructType>(TyR);
+ if (STyL->getNumElements() != STyR->getNumElements())
+ return cmpNumbers(STyL->getNumElements(), STyR->getNumElements());
- for (unsigned i = 0, e = STy1->getNumElements(); i != e; ++i) {
- if (!isEquivalentType(STy1->getElementType(i), STy2->getElementType(i)))
- return false;
- }
- return true;
- }
-
- case Type::UnionTyID: {
- const UnionType *UTy1 = cast<UnionType>(Ty1);
- const UnionType *UTy2 = cast<UnionType>(Ty2);
-
- // TODO: we could be fancy with union(A, union(A, B)) === union(A, B), etc.
- if (UTy1->getNumElements() != UTy2->getNumElements())
- return false;
+ if (STyL->isPacked() != STyR->isPacked())
+ return cmpNumbers(STyL->isPacked(), STyR->isPacked());
- for (unsigned i = 0, e = UTy1->getNumElements(); i != e; ++i) {
- if (!isEquivalentType(UTy1->getElementType(i), UTy2->getElementType(i)))
- return false;
+ for (unsigned i = 0, e = STyL->getNumElements(); i != e; ++i) {
+ if (int Res = cmpTypes(STyL->getElementType(i), STyR->getElementType(i)))
+ return Res;
}
- return true;
+ return 0;
}
case Type::FunctionTyID: {
- const FunctionType *FTy1 = cast<FunctionType>(Ty1);
- const FunctionType *FTy2 = cast<FunctionType>(Ty2);
- if (FTy1->getNumParams() != FTy2->getNumParams() ||
- FTy1->isVarArg() != FTy2->isVarArg())
- return false;
+ FunctionType *FTyL = cast<FunctionType>(TyL);
+ FunctionType *FTyR = cast<FunctionType>(TyR);
+ if (FTyL->getNumParams() != FTyR->getNumParams())
+ return cmpNumbers(FTyL->getNumParams(), FTyR->getNumParams());
- if (!isEquivalentType(FTy1->getReturnType(), FTy2->getReturnType()))
- return false;
+ if (FTyL->isVarArg() != FTyR->isVarArg())
+ return cmpNumbers(FTyL->isVarArg(), FTyR->isVarArg());
+
+ if (int Res = cmpTypes(FTyL->getReturnType(), FTyR->getReturnType()))
+ return Res;
- for (unsigned i = 0, e = FTy1->getNumParams(); i != e; ++i) {
- if (!isEquivalentType(FTy1->getParamType(i), FTy2->getParamType(i)))
- return false;
+ for (unsigned i = 0, e = FTyL->getNumParams(); i != e; ++i) {
+ if (int Res = cmpTypes(FTyL->getParamType(i), FTyR->getParamType(i)))
+ return Res;
}
- return true;
+ return 0;
}
case Type::ArrayTyID: {
- const ArrayType *ATy1 = cast<ArrayType>(Ty1);
- const ArrayType *ATy2 = cast<ArrayType>(Ty2);
- return ATy1->getNumElements() == ATy2->getNumElements() &&
- isEquivalentType(ATy1->getElementType(), ATy2->getElementType());
- }
- case Type::VectorTyID: {
- const VectorType *VTy1 = cast<VectorType>(Ty1);
- const VectorType *VTy2 = cast<VectorType>(Ty2);
- return VTy1->getNumElements() == VTy2->getNumElements() &&
- isEquivalentType(VTy1->getElementType(), VTy2->getElementType());
+ ArrayType *ATyL = cast<ArrayType>(TyL);
+ ArrayType *ATyR = cast<ArrayType>(TyR);
+ if (ATyL->getNumElements() != ATyR->getNumElements())
+ return cmpNumbers(ATyL->getNumElements(), ATyR->getNumElements());
+ return cmpTypes(ATyL->getElementType(), ATyR->getElementType());
}
}
}
-/// isEquivalentOperation - determine whether the two operations are the same
-/// except that pointer-to-A and pointer-to-B are equivalent. This should be
-/// kept in sync with Instruction::isSameOperationAs.
-bool FunctionComparator::isEquivalentOperation(const Instruction *I1,
- const Instruction *I2) const {
- if (I1->getOpcode() != I2->getOpcode() ||
- I1->getNumOperands() != I2->getNumOperands() ||
- !isEquivalentType(I1->getType(), I2->getType()) ||
- !I1->hasSameSubclassOptionalData(I2))
- return false;
+// Determine whether the two operations are the same except that pointer-to-A
+// and pointer-to-B are equivalent. This should be kept in sync with
+// Instruction::isSameOperationAs.
+// Read method declaration comments for more details.
+int FunctionComparator::cmpOperations(const Instruction *L,
+ const Instruction *R) const {
+ // Differences from Instruction::isSameOperationAs:
+ // * replace type comparison with calls to isEquivalentType.
+ // * we test for I->hasSameSubclassOptionalData (nuw/nsw/tail) at the top
+ // * because of the above, we don't test for the tail bit on calls later on
+ if (int Res = cmpNumbers(L->getOpcode(), R->getOpcode()))
+ return Res;
+
+ if (int Res = cmpNumbers(L->getNumOperands(), R->getNumOperands()))
+ return Res;
+
+ if (int Res = cmpTypes(L->getType(), R->getType()))
+ return Res;
+
+ if (int Res = cmpNumbers(L->getRawSubclassOptionalData(),
+ R->getRawSubclassOptionalData()))
+ return Res;
// We have two instructions of identical opcode and #operands. Check to see
// if all operands are the same type
- for (unsigned i = 0, e = I1->getNumOperands(); i != e; ++i)
- if (!isEquivalentType(I1->getOperand(i)->getType(),
- I2->getOperand(i)->getType()))
- return false;
+ for (unsigned i = 0, e = L->getNumOperands(); i != e; ++i) {
+ if (int Res =
+ cmpTypes(L->getOperand(i)->getType(), R->getOperand(i)->getType()))
+ return Res;
+ }
// Check special state that is a part of some instructions.
- if (const LoadInst *LI = dyn_cast<LoadInst>(I1))
- return LI->isVolatile() == cast<LoadInst>(I2)->isVolatile() &&
- LI->getAlignment() == cast<LoadInst>(I2)->getAlignment();
- if (const StoreInst *SI = dyn_cast<StoreInst>(I1))
- return SI->isVolatile() == cast<StoreInst>(I2)->isVolatile() &&
- SI->getAlignment() == cast<StoreInst>(I2)->getAlignment();
- if (const CmpInst *CI = dyn_cast<CmpInst>(I1))
- return CI->getPredicate() == cast<CmpInst>(I2)->getPredicate();
- if (const CallInst *CI = dyn_cast<CallInst>(I1))
- return CI->isTailCall() == cast<CallInst>(I2)->isTailCall() &&
- CI->getCallingConv() == cast<CallInst>(I2)->getCallingConv() &&
- CI->getAttributes().getRawPointer() ==
- cast<CallInst>(I2)->getAttributes().getRawPointer();
- if (const InvokeInst *CI = dyn_cast<InvokeInst>(I1))
- return CI->getCallingConv() == cast<InvokeInst>(I2)->getCallingConv() &&
- CI->getAttributes().getRawPointer() ==
- cast<InvokeInst>(I2)->getAttributes().getRawPointer();
- if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(I1)) {
- if (IVI->getNumIndices() != cast<InsertValueInst>(I2)->getNumIndices())
- return false;
- for (unsigned i = 0, e = IVI->getNumIndices(); i != e; ++i)
- if (IVI->idx_begin()[i] != cast<InsertValueInst>(I2)->idx_begin()[i])
- return false;
- return true;
+ if (const LoadInst *LI = dyn_cast<LoadInst>(L)) {
+ if (int Res = cmpNumbers(LI->isVolatile(), cast<LoadInst>(R)->isVolatile()))
+ return Res;
+ if (int Res =
+ cmpNumbers(LI->getAlignment(), cast<LoadInst>(R)->getAlignment()))
+ return Res;
+ if (int Res =
+ cmpNumbers(LI->getOrdering(), cast<LoadInst>(R)->getOrdering()))
+ return Res;
+ if (int Res =
+ cmpNumbers(LI->getSynchScope(), cast<LoadInst>(R)->getSynchScope()))
+ return Res;
+ return cmpNumbers((uint64_t)LI->getMetadata(LLVMContext::MD_range),
+ (uint64_t)cast<LoadInst>(R)->getMetadata(LLVMContext::MD_range));
}
- if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I1)) {
- if (EVI->getNumIndices() != cast<ExtractValueInst>(I2)->getNumIndices())
- return false;
- for (unsigned i = 0, e = EVI->getNumIndices(); i != e; ++i)
- if (EVI->idx_begin()[i] != cast<ExtractValueInst>(I2)->idx_begin()[i])
- return false;
- return true;
+ if (const StoreInst *SI = dyn_cast<StoreInst>(L)) {
+ if (int Res =
+ cmpNumbers(SI->isVolatile(), cast<StoreInst>(R)->isVolatile()))
+ return Res;
+ if (int Res =
+ cmpNumbers(SI->getAlignment(), cast<StoreInst>(R)->getAlignment()))
+ return Res;
+ if (int Res =
+ cmpNumbers(SI->getOrdering(), cast<StoreInst>(R)->getOrdering()))
+ return Res;
+ return cmpNumbers(SI->getSynchScope(), cast<StoreInst>(R)->getSynchScope());
+ }
+ if (const CmpInst *CI = dyn_cast<CmpInst>(L))
+ return cmpNumbers(CI->getPredicate(), cast<CmpInst>(R)->getPredicate());
+ if (const CallInst *CI = dyn_cast<CallInst>(L)) {
+ if (int Res = cmpNumbers(CI->getCallingConv(),
+ cast<CallInst>(R)->getCallingConv()))
+ return Res;
+ if (int Res =
+ cmpAttrs(CI->getAttributes(), cast<CallInst>(R)->getAttributes()))
+ return Res;
+ return cmpNumbers(
+ (uint64_t)CI->getMetadata(LLVMContext::MD_range),
+ (uint64_t)cast<CallInst>(R)->getMetadata(LLVMContext::MD_range));
+ }
+ if (const InvokeInst *CI = dyn_cast<InvokeInst>(L)) {
+ if (int Res = cmpNumbers(CI->getCallingConv(),
+ cast<InvokeInst>(R)->getCallingConv()))
+ return Res;
+ if (int Res =
+ cmpAttrs(CI->getAttributes(), cast<InvokeInst>(R)->getAttributes()))
+ return Res;
+ return cmpNumbers(
+ (uint64_t)CI->getMetadata(LLVMContext::MD_range),
+ (uint64_t)cast<InvokeInst>(R)->getMetadata(LLVMContext::MD_range));
+ }
+ if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(L)) {
+ ArrayRef<unsigned> LIndices = IVI->getIndices();
+ ArrayRef<unsigned> RIndices = cast<InsertValueInst>(R)->getIndices();
+ if (int Res = cmpNumbers(LIndices.size(), RIndices.size()))
+ return Res;
+ for (size_t i = 0, e = LIndices.size(); i != e; ++i) {
+ if (int Res = cmpNumbers(LIndices[i], RIndices[i]))
+ return Res;
+ }
+ }
+ if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(L)) {
+ ArrayRef<unsigned> LIndices = EVI->getIndices();
+ ArrayRef<unsigned> RIndices = cast<ExtractValueInst>(R)->getIndices();
+ if (int Res = cmpNumbers(LIndices.size(), RIndices.size()))
+ return Res;
+ for (size_t i = 0, e = LIndices.size(); i != e; ++i) {
+ if (int Res = cmpNumbers(LIndices[i], RIndices[i]))
+ return Res;
+ }
+ }
+ if (const FenceInst *FI = dyn_cast<FenceInst>(L)) {
+ if (int Res =
+ cmpNumbers(FI->getOrdering(), cast<FenceInst>(R)->getOrdering()))
+ return Res;
+ return cmpNumbers(FI->getSynchScope(), cast<FenceInst>(R)->getSynchScope());
}
- return true;
+ if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(L)) {
+ if (int Res = cmpNumbers(CXI->isVolatile(),
+ cast<AtomicCmpXchgInst>(R)->isVolatile()))
+ return Res;
+ if (int Res = cmpNumbers(CXI->isWeak(),
+ cast<AtomicCmpXchgInst>(R)->isWeak()))
+ return Res;
+ if (int Res = cmpNumbers(CXI->getSuccessOrdering(),
+ cast<AtomicCmpXchgInst>(R)->getSuccessOrdering()))
+ return Res;
+ if (int Res = cmpNumbers(CXI->getFailureOrdering(),
+ cast<AtomicCmpXchgInst>(R)->getFailureOrdering()))
+ return Res;
+ return cmpNumbers(CXI->getSynchScope(),
+ cast<AtomicCmpXchgInst>(R)->getSynchScope());
+ }
+ if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(L)) {
+ if (int Res = cmpNumbers(RMWI->getOperation(),
+ cast<AtomicRMWInst>(R)->getOperation()))
+ return Res;
+ if (int Res = cmpNumbers(RMWI->isVolatile(),
+ cast<AtomicRMWInst>(R)->isVolatile()))
+ return Res;
+ if (int Res = cmpNumbers(RMWI->getOrdering(),
+ cast<AtomicRMWInst>(R)->getOrdering()))
+ return Res;
+ return cmpNumbers(RMWI->getSynchScope(),
+ cast<AtomicRMWInst>(R)->getSynchScope());
+ }
+ return 0;
}
-/// isEquivalentGEP - determine whether two GEP operations perform the same
-/// underlying arithmetic.
-bool FunctionComparator::isEquivalentGEP(const GEPOperator *GEP1,
- const GEPOperator *GEP2) {
- // When we have target data, we can reduce the GEP down to the value in bytes
- // added to the address.
- if (TD && GEP1->hasAllConstantIndices() && GEP2->hasAllConstantIndices()) {
- SmallVector<Value *, 8> Indices1(GEP1->idx_begin(), GEP1->idx_end());
- SmallVector<Value *, 8> Indices2(GEP2->idx_begin(), GEP2->idx_end());
- uint64_t Offset1 = TD->getIndexedOffset(GEP1->getPointerOperandType(),
- Indices1.data(), Indices1.size());
- uint64_t Offset2 = TD->getIndexedOffset(GEP2->getPointerOperandType(),
- Indices2.data(), Indices2.size());
- return Offset1 == Offset2;
- }
-
- if (GEP1->getPointerOperand()->getType() !=
- GEP2->getPointerOperand()->getType())
- return false;
-
- if (GEP1->getNumOperands() != GEP2->getNumOperands())
- return false;
+// Determine whether two GEP operations perform the same underlying arithmetic.
+// Read method declaration comments for more details.
+int FunctionComparator::cmpGEPs(const GEPOperator *GEPL,
+ const GEPOperator *GEPR) {
- for (unsigned i = 0, e = GEP1->getNumOperands(); i != e; ++i) {
- if (!Enumerate(GEP1->getOperand(i), GEP2->getOperand(i)))
- return false;
- }
+ unsigned int ASL = GEPL->getPointerAddressSpace();
+ unsigned int ASR = GEPR->getPointerAddressSpace();
- return true;
-}
+ if (int Res = cmpNumbers(ASL, ASR))
+ return Res;
-/// getDomain - a value's domain is its parent function if it is specific to a
-/// function, or NULL otherwise.
-const Function *FunctionComparator::getDomain(const Value *V) const {
- if (const Argument *A = dyn_cast<Argument>(V)) {
- return A->getParent();
- } else if (const BasicBlock *BB = dyn_cast<BasicBlock>(V)) {
- return BB->getParent();
- } else if (const Instruction *I = dyn_cast<Instruction>(V)) {
- return I->getParent()->getParent();
- }
- return NULL;
-}
+ // When we have target data, we can reduce the GEP down to the value in bytes
+ // added to the address.
+ if (DL) {
+ unsigned BitWidth = DL->getPointerSizeInBits(ASL);
+ APInt OffsetL(BitWidth, 0), OffsetR(BitWidth, 0);
+ if (GEPL->accumulateConstantOffset(*DL, OffsetL) &&
+ GEPR->accumulateConstantOffset(*DL, OffsetR))
+ return cmpAPInt(OffsetL, OffsetR);
+ }
-/// Enumerate - Compare two values used by the two functions under pair-wise
-/// comparison. If this is the first time the values are seen, they're added to
-/// the mapping so that we will detect mismatches on next use.
-bool FunctionComparator::Enumerate(const Value *V1, const Value *V2) {
- // Check for function @f1 referring to itself and function @f2 referring to
- // itself, or referring to each other, or both referring to either of them.
- // They're all equivalent if the two functions are otherwise equivalent.
- if (V1 == F1 || V1 == F2)
- if (V2 == F1 || V2 == F2)
- return true;
-
- // TODO: constant expressions with GEP or references to F1 or F2.
- if (isa<Constant>(V1))
- return V1 == V2;
-
- if (isa<InlineAsm>(V1) && isa<InlineAsm>(V2)) {
- const InlineAsm *IA1 = cast<InlineAsm>(V1);
- const InlineAsm *IA2 = cast<InlineAsm>(V2);
- return IA1->getAsmString() == IA2->getAsmString() &&
- IA1->getConstraintString() == IA2->getConstraintString();
- }
-
- // We enumerate constants globally and arguments, basic blocks or
- // instructions within the function they belong to.
- const Function *Domain1 = getDomain(V1);
- const Function *Domain2 = getDomain(V2);
-
- // The domains have to either be both NULL, or F1, F2.
- if (Domain1 != Domain2)
- if (Domain1 != F1 && Domain1 != F2)
- return false;
+ if (int Res = cmpNumbers((uint64_t)GEPL->getPointerOperand()->getType(),
+ (uint64_t)GEPR->getPointerOperand()->getType()))
+ return Res;
- IDMap &Map1 = Domains[Domain1];
- unsigned long &ID1 = Map1[V1];
- if (!ID1)
- ID1 = ++DomainCount[Domain1];
+ if (int Res = cmpNumbers(GEPL->getNumOperands(), GEPR->getNumOperands()))
+ return Res;
- IDMap &Map2 = Domains[Domain2];
- unsigned long &ID2 = Map2[V2];
- if (!ID2)
- ID2 = ++DomainCount[Domain2];
+ for (unsigned i = 0, e = GEPL->getNumOperands(); i != e; ++i) {
+ if (int Res = cmpValues(GEPL->getOperand(i), GEPR->getOperand(i)))
+ return Res;
+ }
- return ID1 == ID2;
+ return 0;
}
-// Compare - test whether two basic blocks have equivalent behaviour.
-bool FunctionComparator::Compare(const BasicBlock *BB1, const BasicBlock *BB2) {
- BasicBlock::const_iterator F1I = BB1->begin(), F1E = BB1->end();
- BasicBlock::const_iterator F2I = BB2->begin(), F2E = BB2->end();
+/// Compare two values used by the two functions under pair-wise comparison. If
+/// this is the first time the values are seen, they're added to the mapping so
+/// that we will detect mismatches on next use.
+/// See comments in declaration for more details.
+int FunctionComparator::cmpValues(const Value *L, const Value *R) {
+ // Catch self-reference case.
+ if (L == FnL) {
+ if (R == FnR)
+ return 0;
+ return -1;
+ }
+ if (R == FnR) {
+ if (L == FnL)
+ return 0;
+ return 1;
+ }
- do {
- if (!Enumerate(F1I, F2I))
- return false;
+ const Constant *ConstL = dyn_cast<Constant>(L);
+ const Constant *ConstR = dyn_cast<Constant>(R);
+ if (ConstL && ConstR) {
+ if (L == R)
+ return 0;
+ return cmpConstants(ConstL, ConstR);
+ }
- if (const GetElementPtrInst *GEP1 = dyn_cast<GetElementPtrInst>(F1I)) {
- const GetElementPtrInst *GEP2 = dyn_cast<GetElementPtrInst>(F2I);
- if (!GEP2)
- return false;
+ if (ConstL)
+ return 1;
+ if (ConstR)
+ return -1;
- if (!Enumerate(GEP1->getPointerOperand(), GEP2->getPointerOperand()))
- return false;
+ const InlineAsm *InlineAsmL = dyn_cast<InlineAsm>(L);
+ const InlineAsm *InlineAsmR = dyn_cast<InlineAsm>(R);
- if (!isEquivalentGEP(GEP1, GEP2))
- return false;
- } else {
- if (!isEquivalentOperation(F1I, F2I))
- return false;
+ if (InlineAsmL && InlineAsmR)
+ return cmpNumbers((uint64_t)L, (uint64_t)R);
+ if (InlineAsmL)
+ return 1;
+ if (InlineAsmR)
+ return -1;
- assert(F1I->getNumOperands() == F2I->getNumOperands());
- for (unsigned i = 0, e = F1I->getNumOperands(); i != e; ++i) {
- Value *OpF1 = F1I->getOperand(i);
- Value *OpF2 = F2I->getOperand(i);
+ auto LeftSN = sn_mapL.insert(std::make_pair(L, sn_mapL.size())),
+ RightSN = sn_mapR.insert(std::make_pair(R, sn_mapR.size()));
- if (!Enumerate(OpF1, OpF2))
- return false;
+ return cmpNumbers(LeftSN.first->second, RightSN.first->second);
+}
+// Test whether two basic blocks have equivalent behaviour.
+int FunctionComparator::compare(const BasicBlock *BBL, const BasicBlock *BBR) {
+ BasicBlock::const_iterator InstL = BBL->begin(), InstLE = BBL->end();
+ BasicBlock::const_iterator InstR = BBR->begin(), InstRE = BBR->end();
- if (OpF1->getValueID() != OpF2->getValueID() ||
- !isEquivalentType(OpF1->getType(), OpF2->getType()))
- return false;
+ do {
+ if (int Res = cmpValues(InstL, InstR))
+ return Res;
+
+ const GetElementPtrInst *GEPL = dyn_cast<GetElementPtrInst>(InstL);
+ const GetElementPtrInst *GEPR = dyn_cast<GetElementPtrInst>(InstR);
+
+ if (GEPL && !GEPR)
+ return 1;
+ if (GEPR && !GEPL)
+ return -1;
+
+ if (GEPL && GEPR) {
+ if (int Res =
+ cmpValues(GEPL->getPointerOperand(), GEPR->getPointerOperand()))
+ return Res;
+ if (int Res = cmpGEPs(GEPL, GEPR))
+ return Res;
+ } else {
+ if (int Res = cmpOperations(InstL, InstR))
+ return Res;
+ assert(InstL->getNumOperands() == InstR->getNumOperands());
+
+ for (unsigned i = 0, e = InstL->getNumOperands(); i != e; ++i) {
+ Value *OpL = InstL->getOperand(i);
+ Value *OpR = InstR->getOperand(i);
+ if (int Res = cmpValues(OpL, OpR))
+ return Res;
+ if (int Res = cmpNumbers(OpL->getValueID(), OpR->getValueID()))
+ return Res;
+ // TODO: Already checked in cmpOperation
+ if (int Res = cmpTypes(OpL->getType(), OpR->getType()))
+ return Res;
}
}
- ++F1I, ++F2I;
- } while (F1I != F1E && F2I != F2E);
+ ++InstL, ++InstR;
+ } while (InstL != InstLE && InstR != InstRE);
- return F1I == F1E && F2I == F2E;
+ if (InstL != InstLE && InstR == InstRE)
+ return 1;
+ if (InstL == InstLE && InstR != InstRE)
+ return -1;
+ return 0;
}
-bool FunctionComparator::Compare() {
- // We need to recheck everything, but check the things that weren't included
- // in the hash first.
+// Test whether the two functions have equivalent behaviour.
+int FunctionComparator::compare() {
- if (F1->getAttributes() != F2->getAttributes())
- return false;
+ sn_mapL.clear();
+ sn_mapR.clear();
- if (F1->hasGC() != F2->hasGC())
- return false;
+ if (int Res = cmpAttrs(FnL->getAttributes(), FnR->getAttributes()))
+ return Res;
- if (F1->hasGC() && F1->getGC() != F2->getGC())
- return false;
+ if (int Res = cmpNumbers(FnL->hasGC(), FnR->hasGC()))
+ return Res;
- if (F1->hasSection() != F2->hasSection())
- return false;
+ if (FnL->hasGC()) {
+ if (int Res = cmpNumbers((uint64_t)FnL->getGC(), (uint64_t)FnR->getGC()))
+ return Res;
+ }
- if (F1->hasSection() && F1->getSection() != F2->getSection())
- return false;
+ if (int Res = cmpNumbers(FnL->hasSection(), FnR->hasSection()))
+ return Res;
- if (F1->isVarArg() != F2->isVarArg())
- return false;
+ if (FnL->hasSection()) {
+ if (int Res = cmpStrings(FnL->getSection(), FnR->getSection()))
+ return Res;
+ }
+
+ if (int Res = cmpNumbers(FnL->isVarArg(), FnR->isVarArg()))
+ return Res;
// TODO: if it's internal and only used in direct calls, we could handle this
// case too.
- if (F1->getCallingConv() != F2->getCallingConv())
- return false;
+ if (int Res = cmpNumbers(FnL->getCallingConv(), FnR->getCallingConv()))
+ return Res;
- if (!isEquivalentType(F1->getFunctionType(), F2->getFunctionType()))
- return false;
+ if (int Res = cmpTypes(FnL->getFunctionType(), FnR->getFunctionType()))
+ return Res;
- assert(F1->arg_size() == F2->arg_size() &&
- "Identical functions have a different number of args.");
+ assert(FnL->arg_size() == FnR->arg_size() &&
+ "Identically typed functions have different numbers of args!");
// Visit the arguments so that they get enumerated in the order they're
// passed in.
- for (Function::const_arg_iterator f1i = F1->arg_begin(),
- f2i = F2->arg_begin(), f1e = F1->arg_end(); f1i != f1e; ++f1i, ++f2i) {
- if (!Enumerate(f1i, f2i))
- llvm_unreachable("Arguments repeat");
+ for (Function::const_arg_iterator ArgLI = FnL->arg_begin(),
+ ArgRI = FnR->arg_begin(),
+ ArgLE = FnL->arg_end();
+ ArgLI != ArgLE; ++ArgLI, ++ArgRI) {
+ if (cmpValues(ArgLI, ArgRI) != 0)
+ llvm_unreachable("Arguments repeat!");
}
- // We need to do an ordered walk since the actual ordering of the blocks in
- // the linked list is immaterial. Our walk starts at the entry block for both
+ // We do a CFG-ordered walk since the actual ordering of the blocks in the
+ // linked list is immaterial. Our walk starts at the entry block for both
// functions, then takes each block from each terminator in order. As an
// artifact, this also means that unreachable blocks are ignored.
- SmallVector<const BasicBlock *, 8> F1BBs, F2BBs;
+ SmallVector<const BasicBlock *, 8> FnLBBs, FnRBBs;
SmallSet<const BasicBlock *, 128> VisitedBBs; // in terms of F1.
- F1BBs.push_back(&F1->getEntryBlock());
- F2BBs.push_back(&F2->getEntryBlock());
- VisitedBBs.insert(F1BBs[0]);
- while (!F1BBs.empty()) {
- const BasicBlock *F1BB = F1BBs.pop_back_val();
- const BasicBlock *F2BB = F2BBs.pop_back_val();
- if (!Enumerate(F1BB, F2BB) || !Compare(F1BB, F2BB))
- return false;
- const TerminatorInst *F1TI = F1BB->getTerminator();
- const TerminatorInst *F2TI = F2BB->getTerminator();
- assert(F1TI->getNumSuccessors() == F2TI->getNumSuccessors());
- for (unsigned i = 0, e = F1TI->getNumSuccessors(); i != e; ++i) {
- if (!VisitedBBs.insert(F1TI->getSuccessor(i)))
+
+ FnLBBs.push_back(&FnL->getEntryBlock());
+ FnRBBs.push_back(&FnR->getEntryBlock());
+
+ VisitedBBs.insert(FnLBBs[0]);
+ while (!FnLBBs.empty()) {
+ const BasicBlock *BBL = FnLBBs.pop_back_val();
+ const BasicBlock *BBR = FnRBBs.pop_back_val();
+
+ if (int Res = cmpValues(BBL, BBR))
+ return Res;
+
+ if (int Res = compare(BBL, BBR))
+ return Res;
+
+ const TerminatorInst *TermL = BBL->getTerminator();
+ const TerminatorInst *TermR = BBR->getTerminator();
+
+ assert(TermL->getNumSuccessors() == TermR->getNumSuccessors());
+ for (unsigned i = 0, e = TermL->getNumSuccessors(); i != e; ++i) {
+ if (!VisitedBBs.insert(TermL->getSuccessor(i)))
continue;
- F1BBs.push_back(F1TI->getSuccessor(i));
- F2BBs.push_back(F2TI->getSuccessor(i));
+
+ FnLBBs.push_back(TermL->getSuccessor(i));
+ FnRBBs.push_back(TermR->getSuccessor(i));
}
}
- return true;
+ return 0;
}
-// ===----------------------------------------------------------------------===
-// Folding of functions
-// ===----------------------------------------------------------------------===
-
-// Cases:
-// * F is external strong, G is external strong:
-// turn G into a thunk to F (1)
-// * F is external strong, G is external weak:
-// turn G into a thunk to F (1)
-// * F is external weak, G is external weak:
-// unfoldable
-// * F is external strong, G is internal:
-// address of G taken:
-// turn G into a thunk to F (1)
-// address of G not taken:
-// make G an alias to F (2)
-// * F is internal, G is external weak
-// address of F is taken:
-// turn G into a thunk to F (1)
-// address of F is not taken:
-// make G an alias of F (2)
-// * F is internal, G is internal:
-// address of F and G are taken:
-// turn G into a thunk to F (1)
-// address of G is not taken:
-// make G an alias to F (2)
-//
-// alias requires linkage == (external,local,weak) fallback to creating a thunk
-// external means 'externally visible' linkage != (internal,private)
-// internal means linkage == (internal,private)
-// weak means linkage mayBeOverridable
-// being external implies that the address is taken
-//
-// 1. turn G into a thunk to F
-// 2. make G an alias to F
+namespace {
+
+/// MergeFunctions finds functions which will generate identical machine code,
+/// by considering all pointer types to be equivalent. Once identified,
+/// MergeFunctions will fold them by replacing a call to one to a call to a
+/// bitcast of the other.
+///
+class MergeFunctions : public ModulePass {
+public:
+ static char ID;
+ MergeFunctions()
+ : ModulePass(ID), HasGlobalAliases(false) {
+ initializeMergeFunctionsPass(*PassRegistry::getPassRegistry());
+ }
+
+ bool runOnModule(Module &M) override;
+
+private:
+ typedef std::set<FunctionPtr> FnTreeType;
+
+ /// A work queue of functions that may have been modified and should be
+ /// analyzed again.
+ std::vector<WeakVH> Deferred;
+
+ /// Checks the rules of order relation introduced among functions set.
+ /// Returns true, if sanity check has been passed, and false if failed.
+ bool doSanityCheck(std::vector<WeakVH> &Worklist);
+
+ /// Insert a ComparableFunction into the FnTree, or merge it away if it's
+ /// equal to one that's already present.
+ bool insert(Function *NewFunction);
-enum LinkageCategory {
- ExternalStrong,
- ExternalWeak,
- Internal
+ /// Remove a Function from the FnTree and queue it up for a second sweep of
+ /// analysis.
+ void remove(Function *F);
+
+ /// Find the functions that use this Value and remove them from FnTree and
+ /// queue the functions.
+ void removeUsers(Value *V);
+
+ /// Replace all direct calls of Old with calls of New. Will bitcast New if
+ /// necessary to make types match.
+ void replaceDirectCallers(Function *Old, Function *New);
+
+ /// Merge two equivalent functions. Upon completion, G may be deleted, or may
+ /// be converted into a thunk. In either case, it should never be visited
+ /// again.
+ void mergeTwoFunctions(Function *F, Function *G);
+
+ /// Replace G with a thunk or an alias to F. Deletes G.
+ void writeThunkOrAlias(Function *F, Function *G);
+
+ /// Replace G with a simple tail call to bitcast(F). Also replace direct uses
+ /// of G with bitcast(F). Deletes G.
+ void writeThunk(Function *F, Function *G);
+
+ /// Replace G with an alias to F. Deletes G.
+ void writeAlias(Function *F, Function *G);
+
+ /// The set of all distinct functions. Use the insert() and remove() methods
+ /// to modify it.
+ FnTreeType FnTree;
+
+ /// DataLayout for more accurate GEP comparisons. May be NULL.
+ const DataLayout *DL;
+
+ /// Whether or not the target supports global aliases.
+ bool HasGlobalAliases;
};
-static LinkageCategory categorize(const Function *F) {
- switch (F->getLinkage()) {
- case GlobalValue::InternalLinkage:
- case GlobalValue::PrivateLinkage:
- case GlobalValue::LinkerPrivateLinkage:
- return Internal;
-
- case GlobalValue::WeakAnyLinkage:
- case GlobalValue::WeakODRLinkage:
- case GlobalValue::ExternalWeakLinkage:
- case GlobalValue::LinkerPrivateWeakLinkage:
- return ExternalWeak;
-
- case GlobalValue::ExternalLinkage:
- case GlobalValue::AvailableExternallyLinkage:
- case GlobalValue::LinkOnceAnyLinkage:
- case GlobalValue::LinkOnceODRLinkage:
- case GlobalValue::AppendingLinkage:
- case GlobalValue::DLLImportLinkage:
- case GlobalValue::DLLExportLinkage:
- case GlobalValue::CommonLinkage:
- return ExternalStrong;
- }
-
- llvm_unreachable("Unknown LinkageType.");
- return ExternalWeak;
+} // end anonymous namespace
+
+char MergeFunctions::ID = 0;
+INITIALIZE_PASS(MergeFunctions, "mergefunc", "Merge Functions", false, false)
+
+ModulePass *llvm::createMergeFunctionsPass() {
+ return new MergeFunctions();
+}
+
+bool MergeFunctions::doSanityCheck(std::vector<WeakVH> &Worklist) {
+ if (const unsigned Max = NumFunctionsForSanityCheck) {
+ unsigned TripleNumber = 0;
+ bool Valid = true;
+
+ dbgs() << "MERGEFUNC-SANITY: Started for first " << Max << " functions.\n";
+
+ unsigned i = 0;
+ for (std::vector<WeakVH>::iterator I = Worklist.begin(), E = Worklist.end();
+ I != E && i < Max; ++I, ++i) {
+ unsigned j = i;
+ for (std::vector<WeakVH>::iterator J = I; J != E && j < Max; ++J, ++j) {
+ Function *F1 = cast<Function>(*I);
+ Function *F2 = cast<Function>(*J);
+ int Res1 = FunctionComparator(DL, F1, F2).compare();
+ int Res2 = FunctionComparator(DL, F2, F1).compare();
+
+ // If F1 <= F2, then F2 >= F1, otherwise report failure.
+ if (Res1 != -Res2) {
+ dbgs() << "MERGEFUNC-SANITY: Non-symmetric; triple: " << TripleNumber
+ << "\n";
+ F1->dump();
+ F2->dump();
+ Valid = false;
+ }
+
+ if (Res1 == 0)
+ continue;
+
+ unsigned k = j;
+ for (std::vector<WeakVH>::iterator K = J; K != E && k < Max;
+ ++k, ++K, ++TripleNumber) {
+ if (K == J)
+ continue;
+
+ Function *F3 = cast<Function>(*K);
+ int Res3 = FunctionComparator(DL, F1, F3).compare();
+ int Res4 = FunctionComparator(DL, F2, F3).compare();
+
+ bool Transitive = true;
+
+ if (Res1 != 0 && Res1 == Res4) {
+ // F1 > F2, F2 > F3 => F1 > F3
+ Transitive = Res3 == Res1;
+ } else if (Res3 != 0 && Res3 == -Res4) {
+ // F1 > F3, F3 > F2 => F1 > F2
+ Transitive = Res3 == Res1;
+ } else if (Res4 != 0 && -Res3 == Res4) {
+ // F2 > F3, F3 > F1 => F2 > F1
+ Transitive = Res4 == -Res1;
+ }
+
+ if (!Transitive) {
+ dbgs() << "MERGEFUNC-SANITY: Non-transitive; triple: "
+ << TripleNumber << "\n";
+ dbgs() << "Res1, Res3, Res4: " << Res1 << ", " << Res3 << ", "
+ << Res4 << "\n";
+ F1->dump();
+ F2->dump();
+ F3->dump();
+ Valid = false;
+ }
+ }
+ }
+ }
+
+ dbgs() << "MERGEFUNC-SANITY: " << (Valid ? "Passed." : "Failed.") << "\n";
+ return Valid;
+ }
+ return true;
}
-static void ThunkGToF(Function *F, Function *G) {
+bool MergeFunctions::runOnModule(Module &M) {
+ bool Changed = false;
+ DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
+ DL = DLP ? &DLP->getDataLayout() : nullptr;
+
+ for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
+ if (!I->isDeclaration() && !I->hasAvailableExternallyLinkage())
+ Deferred.push_back(WeakVH(I));
+ }
+
+ do {
+ std::vector<WeakVH> Worklist;
+ Deferred.swap(Worklist);
+
+ DEBUG(doSanityCheck(Worklist));
+
+ DEBUG(dbgs() << "size of module: " << M.size() << '\n');
+ DEBUG(dbgs() << "size of worklist: " << Worklist.size() << '\n');
+
+ // Insert only strong functions and merge them. Strong function merging
+ // always deletes one of them.
+ for (std::vector<WeakVH>::iterator I = Worklist.begin(),
+ E = Worklist.end(); I != E; ++I) {
+ if (!*I) continue;
+ Function *F = cast<Function>(*I);
+ if (!F->isDeclaration() && !F->hasAvailableExternallyLinkage() &&
+ !F->mayBeOverridden()) {
+ Changed |= insert(F);
+ }
+ }
+
+ // Insert only weak functions and merge them. By doing these second we
+ // create thunks to the strong function when possible. When two weak
+ // functions are identical, we create a new strong function with two weak
+ // weak thunks to it which are identical but not mergable.
+ for (std::vector<WeakVH>::iterator I = Worklist.begin(),
+ E = Worklist.end(); I != E; ++I) {
+ if (!*I) continue;
+ Function *F = cast<Function>(*I);
+ if (!F->isDeclaration() && !F->hasAvailableExternallyLinkage() &&
+ F->mayBeOverridden()) {
+ Changed |= insert(F);
+ }
+ }
+ DEBUG(dbgs() << "size of FnTree: " << FnTree.size() << '\n');
+ } while (!Deferred.empty());
+
+ FnTree.clear();
+
+ return Changed;
+}
+
+// Replace direct callers of Old with New.
+void MergeFunctions::replaceDirectCallers(Function *Old, Function *New) {
+ Constant *BitcastNew = ConstantExpr::getBitCast(New, Old->getType());
+ for (auto UI = Old->use_begin(), UE = Old->use_end(); UI != UE;) {
+ Use *U = &*UI;
+ ++UI;
+ CallSite CS(U->getUser());
+ if (CS && CS.isCallee(U)) {
+ remove(CS.getInstruction()->getParent()->getParent());
+ U->set(BitcastNew);
+ }
+ }
+}
+
+// Replace G with an alias to F if possible, or else a thunk to F. Deletes G.
+void MergeFunctions::writeThunkOrAlias(Function *F, Function *G) {
+ if (HasGlobalAliases && G->hasUnnamedAddr()) {
+ if (G->hasExternalLinkage() || G->hasLocalLinkage() ||
+ G->hasWeakLinkage()) {
+ writeAlias(F, G);
+ return;
+ }
+ }
+
+ writeThunk(F, G);
+}
+
+// Helper for writeThunk,
+// Selects proper bitcast operation,
+// but a bit simpler then CastInst::getCastOpcode.
+static Value *createCast(IRBuilder<false> &Builder, Value *V, Type *DestTy) {
+ Type *SrcTy = V->getType();
+ if (SrcTy->isStructTy()) {
+ assert(DestTy->isStructTy());
+ assert(SrcTy->getStructNumElements() == DestTy->getStructNumElements());
+ Value *Result = UndefValue::get(DestTy);
+ for (unsigned int I = 0, E = SrcTy->getStructNumElements(); I < E; ++I) {
+ Value *Element = createCast(
+ Builder, Builder.CreateExtractValue(V, ArrayRef<unsigned int>(I)),
+ DestTy->getStructElementType(I));
+
+ Result =
+ Builder.CreateInsertValue(Result, Element, ArrayRef<unsigned int>(I));
+ }
+ return Result;
+ }
+ assert(!DestTy->isStructTy());
+ if (SrcTy->isIntegerTy() && DestTy->isPointerTy())
+ return Builder.CreateIntToPtr(V, DestTy);
+ else if (SrcTy->isPointerTy() && DestTy->isIntegerTy())
+ return Builder.CreatePtrToInt(V, DestTy);
+ else
+ return Builder.CreateBitCast(V, DestTy);
+}
+
+// Replace G with a simple tail call to bitcast(F). Also replace direct uses
+// of G with bitcast(F). Deletes G.
+void MergeFunctions::writeThunk(Function *F, Function *G) {
if (!G->mayBeOverridden()) {
// Redirect direct callers of G to F.
- Constant *BitcastF = ConstantExpr::getBitCast(F, G->getType());
- for (Value::use_iterator UI = G->use_begin(), UE = G->use_end();
- UI != UE;) {
- Value::use_iterator TheIter = UI;
- ++UI;
- CallSite CS(*TheIter);
- if (CS && CS.isCallee(TheIter))
- TheIter.getUse().set(BitcastF);
- }
+ replaceDirectCallers(G, F);
+ }
+
+ // If G was internal then we may have replaced all uses of G with F. If so,
+ // stop here and delete G. There's no need for a thunk.
+ if (G->hasLocalLinkage() && G->use_empty()) {
+ G->eraseFromParent();
+ return;
}
Function *NewG = Function::Create(G->getFunctionType(), G->getLinkage(), "",
G->getParent());
BasicBlock *BB = BasicBlock::Create(F->getContext(), "", NewG);
+ IRBuilder<false> Builder(BB);
SmallVector<Value *, 16> Args;
unsigned i = 0;
- const FunctionType *FFTy = F->getFunctionType();
+ FunctionType *FFTy = F->getFunctionType();
for (Function::arg_iterator AI = NewG->arg_begin(), AE = NewG->arg_end();
AI != AE; ++AI) {
- if (FFTy->getParamType(i) == AI->getType()) {
- Args.push_back(AI);
- } else {
- Args.push_back(new BitCastInst(AI, FFTy->getParamType(i), "", BB));
- }
+ Args.push_back(createCast(Builder, (Value*)AI, FFTy->getParamType(i)));
++i;
}
- CallInst *CI = CallInst::Create(F, Args.begin(), Args.end(), "", BB);
+ CallInst *CI = Builder.CreateCall(F, Args);
CI->setTailCall();
CI->setCallingConv(F->getCallingConv());
if (NewG->getReturnType()->isVoidTy()) {
- ReturnInst::Create(F->getContext(), BB);
- } else if (CI->getType() != NewG->getReturnType()) {
- Value *BCI = new BitCastInst(CI, NewG->getReturnType(), "", BB);
- ReturnInst::Create(F->getContext(), BCI, BB);
+ Builder.CreateRetVoid();
} else {
- ReturnInst::Create(F->getContext(), CI, BB);
+ Builder.CreateRet(createCast(Builder, CI, NewG->getReturnType()));
}
NewG->copyAttributesFrom(G);
NewG->takeName(G);
+ removeUsers(G);
G->replaceAllUsesWith(NewG);
G->eraseFromParent();
-}
-static void AliasGToF(Function *F, Function *G) {
- // Darwin will trigger llvm_unreachable if asked to codegen an alias.
- return ThunkGToF(F, G);
-
-#if 0
- if (!G->hasExternalLinkage() && !G->hasLocalLinkage() && !G->hasWeakLinkage())
- return ThunkGToF(F, G);
+ DEBUG(dbgs() << "writeThunk: " << NewG->getName() << '\n');
+ ++NumThunksWritten;
+}
- GlobalAlias *GA = new GlobalAlias(
- G->getType(), G->getLinkage(), "",
- ConstantExpr::getBitCast(F, G->getType()), G->getParent());
+// Replace G with an alias to F and delete G.
+void MergeFunctions::writeAlias(Function *F, Function *G) {
+ PointerType *PTy = G->getType();
+ auto *GA = GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(),
+ G->getLinkage(), "", F);
F->setAlignment(std::max(F->getAlignment(), G->getAlignment()));
GA->takeName(G);
GA->setVisibility(G->getVisibility());
+ removeUsers(G);
G->replaceAllUsesWith(GA);
G->eraseFromParent();
-#endif
+
+ DEBUG(dbgs() << "writeAlias: " << GA->getName() << '\n');
+ ++NumAliasesWritten;
}
-static bool fold(std::vector<Function *> &FnVec, unsigned i, unsigned j) {
- Function *F = FnVec[i];
- Function *G = FnVec[j];
-
- LinkageCategory catF = categorize(F);
- LinkageCategory catG = categorize(G);
-
- if (catF == ExternalWeak || (catF == Internal && catG == ExternalStrong)) {
- std::swap(FnVec[i], FnVec[j]);
- std::swap(F, G);
- std::swap(catF, catG);
- }
-
- switch (catF) {
- case ExternalStrong:
- switch (catG) {
- case ExternalStrong:
- case ExternalWeak:
- ThunkGToF(F, G);
- break;
- case Internal:
- if (G->hasAddressTaken())
- ThunkGToF(F, G);
- else
- AliasGToF(F, G);
- break;
+// Merge two equivalent functions. Upon completion, Function G is deleted.
+void MergeFunctions::mergeTwoFunctions(Function *F, Function *G) {
+ if (F->mayBeOverridden()) {
+ assert(G->mayBeOverridden());
+
+ if (HasGlobalAliases) {
+ // Make them both thunks to the same internal function.
+ Function *H = Function::Create(F->getFunctionType(), F->getLinkage(), "",
+ F->getParent());
+ H->copyAttributesFrom(F);
+ H->takeName(F);
+ removeUsers(F);
+ F->replaceAllUsesWith(H);
+
+ unsigned MaxAlignment = std::max(G->getAlignment(), H->getAlignment());
+
+ writeAlias(F, G);
+ writeAlias(F, H);
+
+ F->setAlignment(MaxAlignment);
+ F->setLinkage(GlobalValue::PrivateLinkage);
+ } else {
+ // We can't merge them. Instead, pick one and update all direct callers
+ // to call it and hope that we improve the instruction cache hit rate.
+ replaceDirectCallers(G, F);
}
- break;
-
- case ExternalWeak: {
- assert(catG == ExternalWeak);
-
- // Make them both thunks to the same internal function.
- F->setAlignment(std::max(F->getAlignment(), G->getAlignment()));
- Function *H = Function::Create(F->getFunctionType(), F->getLinkage(), "",
- F->getParent());
- H->copyAttributesFrom(F);
- H->takeName(F);
- F->replaceAllUsesWith(H);
-
- ThunkGToF(F, G);
- ThunkGToF(F, H);
-
- F->setLinkage(GlobalValue::InternalLinkage);
- } break;
-
- case Internal:
- switch (catG) {
- case ExternalStrong:
- llvm_unreachable(0);
- // fall-through
- case ExternalWeak:
- if (F->hasAddressTaken())
- ThunkGToF(F, G);
- else
- AliasGToF(F, G);
- break;
- case Internal: {
- bool addrTakenF = F->hasAddressTaken();
- bool addrTakenG = G->hasAddressTaken();
- if (!addrTakenF && addrTakenG) {
- std::swap(FnVec[i], FnVec[j]);
- std::swap(F, G);
- std::swap(addrTakenF, addrTakenG);
- }
- if (addrTakenF && addrTakenG) {
- ThunkGToF(F, G);
- } else {
- assert(!addrTakenG);
- AliasGToF(F, G);
- }
- } break;
- } break;
+ ++NumDoubleWeak;
+ } else {
+ writeThunkOrAlias(F, G);
}
++NumFunctionsMerged;
- return true;
}
-// ===----------------------------------------------------------------------===
-// Pass definition
-// ===----------------------------------------------------------------------===
+// Insert a ComparableFunction into the FnTree, or merge it away if equal to one
+// that was already inserted.
+bool MergeFunctions::insert(Function *NewFunction) {
+ std::pair<FnTreeType::iterator, bool> Result =
+ FnTree.insert(FunctionPtr(NewFunction, DL));
-bool MergeFunctions::runOnModule(Module &M) {
- bool Changed = false;
+ if (Result.second) {
+ DEBUG(dbgs() << "Inserting as unique: " << NewFunction->getName() << '\n');
+ return false;
+ }
- std::map<unsigned long, std::vector<Function *> > FnMap;
+ const FunctionPtr &OldF = *Result.first;
- for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
- if (F->isDeclaration())
- continue;
+ // Don't merge tiny functions, since it can just end up making the function
+ // larger.
+ // FIXME: Should still merge them if they are unnamed_addr and produce an
+ // alias.
+ if (NewFunction->size() == 1) {
+ if (NewFunction->front().size() <= 2) {
+ DEBUG(dbgs() << NewFunction->getName()
+ << " is to small to bother merging\n");
+ return false;
+ }
+ }
+
+ // Never thunk a strong function to a weak function.
+ assert(!OldF.getFunc()->mayBeOverridden() || NewFunction->mayBeOverridden());
+
+ DEBUG(dbgs() << " " << OldF.getFunc()->getName()
+ << " == " << NewFunction->getName() << '\n');
+
+ Function *DeleteF = NewFunction;
+ mergeTwoFunctions(OldF.getFunc(), DeleteF);
+ return true;
+}
- FnMap[ProfileFunction(F)].push_back(F);
+// Remove a function from FnTree. If it was already in FnTree, add
+// it to Deferred so that we'll look at it in the next round.
+void MergeFunctions::remove(Function *F) {
+ // We need to make sure we remove F, not a function "equal" to F per the
+ // function equality comparator.
+ FnTreeType::iterator found = FnTree.find(FunctionPtr(F, DL));
+ size_t Erased = 0;
+ if (found != FnTree.end() && found->getFunc() == F) {
+ Erased = 1;
+ FnTree.erase(found);
}
- TargetData *TD = getAnalysisIfAvailable<TargetData>();
+ if (Erased) {
+ DEBUG(dbgs() << "Removed " << F->getName()
+ << " from set and deferred it.\n");
+ Deferred.push_back(F);
+ }
+}
- bool LocalChanged;
- do {
- LocalChanged = false;
- DEBUG(dbgs() << "size: " << FnMap.size() << "\n");
- for (std::map<unsigned long, std::vector<Function *> >::iterator
- I = FnMap.begin(), E = FnMap.end(); I != E; ++I) {
- std::vector<Function *> &FnVec = I->second;
- DEBUG(dbgs() << "hash (" << I->first << "): " << FnVec.size() << "\n");
-
- for (int i = 0, e = FnVec.size(); i != e; ++i) {
- for (int j = i + 1; j != e; ++j) {
- bool isEqual = FunctionComparator(TD, FnVec[i], FnVec[j]).Compare();
-
- DEBUG(dbgs() << " " << FnVec[i]->getName()
- << (isEqual ? " == " : " != ")
- << FnVec[j]->getName() << "\n");
-
- if (isEqual) {
- if (fold(FnVec, i, j)) {
- LocalChanged = true;
- FnVec.erase(FnVec.begin() + j);
- --j, --e;
- }
- }
- }
+// For each instruction used by the value, remove() the function that contains
+// the instruction. This should happen right before a call to RAUW.
+void MergeFunctions::removeUsers(Value *V) {
+ std::vector<Value *> Worklist;
+ Worklist.push_back(V);
+ while (!Worklist.empty()) {
+ Value *V = Worklist.back();
+ Worklist.pop_back();
+
+ for (User *U : V->users()) {
+ if (Instruction *I = dyn_cast<Instruction>(U)) {
+ remove(I->getParent()->getParent());
+ } else if (isa<GlobalValue>(U)) {
+ // do nothing
+ } else if (Constant *C = dyn_cast<Constant>(U)) {
+ for (User *UU : C->users())
+ Worklist.push_back(UU);
}
-
}
- Changed |= LocalChanged;
- } while (LocalChanged);
-
- return Changed;
+ }
}