From 18c7f80b3e83ab584bd8572695a3cde8bafd9d3c Mon Sep 17 00:00:00 2001 From: Chris Lattner Date: Sun, 5 Feb 2012 02:29:43 +0000 Subject: reapply the patches reverted in r149470 that reenable ConstantDataArray, but with a critical fix to the SelectionDAG code that optimizes copies from strings into immediate stores: the previous code was stopping reading string data at the first nul. Address this by adding a new argument to llvm::getConstantStringInfo, preserving the behavior before the patch. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149800 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Analysis/ConstantFolding.cpp | 6 +- lib/Analysis/ValueTracking.cpp | 180 ++++++++++----------------------------- 2 files changed, 47 insertions(+), 139 deletions(-) (limited to 'lib/Analysis') diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp index 121e334..48e75a1 100644 --- a/lib/Analysis/ConstantFolding.cpp +++ b/lib/Analysis/ConstantFolding.cpp @@ -476,9 +476,9 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, // Instead of loading constant c string, use corresponding integer value // directly if string length is small enough. - std::string Str; - if (TD && GetConstantStringInfo(CE, Str) && !Str.empty()) { - unsigned StrLen = Str.length(); + StringRef Str; + if (TD && getConstantStringInfo(CE, Str) && !Str.empty()) { + unsigned StrLen = Str.size(); Type *Ty = cast(CE->getType())->getElementType(); unsigned NumBits = Ty->getPrimitiveSizeInBits(); // Replace load with immediate integer if the result is an integer or fp diff --git a/lib/Analysis/ValueTracking.cpp b/lib/Analysis/ValueTracking.cpp index 6403f03..f0ed1b6 100644 --- a/lib/Analysis/ValueTracking.cpp +++ b/lib/Analysis/ValueTracking.cpp @@ -1369,25 +1369,21 @@ Value *llvm::isBytewiseValue(Value *V) { } } - // A ConstantArray is splatable if all its members are equal and also - // splatable. - if (ConstantArray *CA = dyn_cast(V)) { - if (CA->getNumOperands() == 0) - return 0; - - Value *Val = isBytewiseValue(CA->getOperand(0)); + // A ConstantDataArray/Vector is splatable if all its members are equal and + // also splatable. + if (ConstantDataSequential *CA = dyn_cast(V)) { + Value *Elt = CA->getElementAsConstant(0); + Value *Val = isBytewiseValue(Elt); if (!Val) return 0; - for (unsigned I = 1, E = CA->getNumOperands(); I != E; ++I) - if (CA->getOperand(I-1) != CA->getOperand(I)) + for (unsigned I = 1, E = CA->getNumElements(); I != E; ++I) + if (CA->getElementAsConstant(I) != Elt) return 0; return Val; } - // FIXME: Vector types (e.g., <4 x i32> ). - // Conceptually, we could handle things like: // %a = zext i8 %X to i16 // %b = shl i16 %a, 8 @@ -1607,33 +1603,19 @@ Value *llvm::GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset, } -/// GetConstantStringInfo - This function computes the length of a +/// getConstantStringInfo - This function computes the length of a /// null-terminated C string pointed to by V. If successful, it returns true /// and returns the string in Str. If unsuccessful, it returns false. -bool llvm::GetConstantStringInfo(const Value *V, std::string &Str, - uint64_t Offset, bool StopAtNul) { - // If V is NULL then return false; - if (V == NULL) return false; - - // Look through bitcast instructions. - if (const BitCastInst *BCI = dyn_cast(V)) - return GetConstantStringInfo(BCI->getOperand(0), Str, Offset, StopAtNul); - - // If the value is not a GEP instruction nor a constant expression with a - // GEP instruction, then return false because ConstantArray can't occur - // any other way. - const User *GEP = 0; - if (const GetElementPtrInst *GEPI = dyn_cast(V)) { - GEP = GEPI; - } else if (const ConstantExpr *CE = dyn_cast(V)) { - if (CE->getOpcode() == Instruction::BitCast) - return GetConstantStringInfo(CE->getOperand(0), Str, Offset, StopAtNul); - if (CE->getOpcode() != Instruction::GetElementPtr) - return false; - GEP = CE; - } +bool llvm::getConstantStringInfo(const Value *V, StringRef &Str, + uint64_t Offset, bool TrimAtNul) { + assert(V); + + // Look through bitcast instructions and geps. + V = V->stripPointerCasts(); - if (GEP) { + // If the value is a GEP instructionor constant expression, treat it as an + // offset. + if (const GEPOperator *GEP = dyn_cast(V)) { // Make sure the GEP has exactly three arguments. if (GEP->getNumOperands() != 3) return false; @@ -1658,51 +1640,48 @@ bool llvm::GetConstantStringInfo(const Value *V, std::string &Str, StartIdx = CI->getZExtValue(); else return false; - return GetConstantStringInfo(GEP->getOperand(0), Str, StartIdx+Offset, - StopAtNul); + return getConstantStringInfo(GEP->getOperand(0), Str, StartIdx+Offset); } // The GEP instruction, constant or instruction, must reference a global // variable that is a constant and is initialized. The referenced constant // initializer is the array that we'll use for optimization. - const GlobalVariable* GV = dyn_cast(V); + const GlobalVariable *GV = dyn_cast(V); if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer()) return false; - const Constant *GlobalInit = GV->getInitializer(); - + // Handle the all-zeros case - if (GlobalInit->isNullValue()) { + if (GV->getInitializer()->isNullValue()) { // This is a degenerate case. The initializer is constant zero so the // length of the string must be zero. - Str.clear(); + Str = ""; return true; } // Must be a Constant Array - const ConstantArray *Array = dyn_cast(GlobalInit); - if (Array == 0 || !Array->getType()->getElementType()->isIntegerTy(8)) + const ConstantDataArray *Array = + dyn_cast(GV->getInitializer()); + if (Array == 0 || !Array->isString()) return false; // Get the number of elements in the array - uint64_t NumElts = Array->getType()->getNumElements(); - + uint64_t NumElts = Array->getType()->getArrayNumElements(); + + // Start out with the entire array in the StringRef. + Str = Array->getAsString(); + if (Offset > NumElts) return false; - // Traverse the constant array from 'Offset' which is the place the GEP refers - // to in the array. - Str.reserve(NumElts-Offset); - for (unsigned i = Offset; i != NumElts; ++i) { - const Constant *Elt = Array->getOperand(i); - const ConstantInt *CI = dyn_cast(Elt); - if (!CI) // This array isn't suitable, non-int initializer. - return false; - if (StopAtNul && CI->isZero()) - return true; // we found end of string, success! - Str += (char)CI->getZExtValue(); - } + // Skip over 'offset' bytes. + Str = Str.substr(Offset); - // The array isn't null terminated, but maybe this is a memcpy, not a strcpy. + if (TrimAtNul) { + // Trim off the \0 and anything after it. If the array is not nul + // terminated, we just return the whole end of string. The client may know + // some other way that the string is length-bound. + Str = Str.substr(0, Str.find('\0')); + } return true; } @@ -1714,8 +1693,7 @@ bool llvm::GetConstantStringInfo(const Value *V, std::string &Str, /// the specified pointer, return 'len+1'. If we can't, return 0. static uint64_t GetStringLengthH(Value *V, SmallPtrSet &PHIs) { // Look through noop bitcast instructions. - if (BitCastInst *BCI = dyn_cast(V)) - return GetStringLengthH(BCI->getOperand(0), PHIs); + V = V->stripPointerCasts(); // If this is a PHI node, there are two cases: either we have already seen it // or we haven't. @@ -1751,83 +1729,13 @@ static uint64_t GetStringLengthH(Value *V, SmallPtrSet &PHIs) { if (Len1 != Len2) return 0; return Len1; } - - // As a special-case, "@string = constant i8 0" is also a string with zero - // length, not wrapped in a bitcast or GEP. - if (GlobalVariable *GV = dyn_cast(V)) { - if (GV->isConstant() && GV->hasDefinitiveInitializer()) - if (GV->getInitializer()->isNullValue()) return 1; - return 0; - } - - // If the value is not a GEP instruction nor a constant expression with a - // GEP instruction, then return unknown. - User *GEP = 0; - if (GetElementPtrInst *GEPI = dyn_cast(V)) { - GEP = GEPI; - } else if (ConstantExpr *CE = dyn_cast(V)) { - if (CE->getOpcode() != Instruction::GetElementPtr) - return 0; - GEP = CE; - } else { - return 0; - } - - // Make sure the GEP has exactly three arguments. - if (GEP->getNumOperands() != 3) - return 0; - - // Check to make sure that the first operand of the GEP is an integer and - // has value 0 so that we are sure we're indexing into the initializer. - if (ConstantInt *Idx = dyn_cast(GEP->getOperand(1))) { - if (!Idx->isZero()) - return 0; - } else - return 0; - - // If the second index isn't a ConstantInt, then this is a variable index - // into the array. If this occurs, we can't say anything meaningful about - // the string. - uint64_t StartIdx = 0; - if (ConstantInt *CI = dyn_cast(GEP->getOperand(2))) - StartIdx = CI->getZExtValue(); - else - return 0; - - // The GEP instruction, constant or instruction, must reference a global - // variable that is a constant and is initialized. The referenced constant - // initializer is the array that we'll use for optimization. - GlobalVariable* GV = dyn_cast(GEP->getOperand(0)); - if (!GV || !GV->isConstant() || !GV->hasInitializer() || - GV->mayBeOverridden()) + + // Otherwise, see if we can read the string. + StringRef StrData; + if (!getConstantStringInfo(V, StrData)) return 0; - Constant *GlobalInit = GV->getInitializer(); - - // Handle the ConstantAggregateZero case, which is a degenerate case. The - // initializer is constant zero so the length of the string must be zero. - if (isa(GlobalInit)) - return 1; // Len = 0 offset by 1. - - // Must be a Constant Array - ConstantArray *Array = dyn_cast(GlobalInit); - if (!Array || !Array->getType()->getElementType()->isIntegerTy(8)) - return false; - - // Get the number of elements in the array - uint64_t NumElts = Array->getType()->getNumElements(); - - // Traverse the constant array from StartIdx (derived above) which is - // the place the GEP refers to in the array. - for (unsigned i = StartIdx; i != NumElts; ++i) { - Constant *Elt = Array->getOperand(i); - ConstantInt *CI = dyn_cast(Elt); - if (!CI) // This array isn't suitable, non-int initializer. - return 0; - if (CI->isZero()) - return i-StartIdx+1; // We found end of string, success! - } - return 0; // The array isn't null terminated, conservatively return 'unknown'. + return StrData.size()+1; } /// GetStringLength - If we can compute the length of the string pointed to by -- cgit v1.1