diff options
-rw-r--r-- | lib/Analysis/ConstantFolding.cpp | 26 | ||||
-rw-r--r-- | test/Transforms/InstCombine/constant-fold-gep-overindex.ll | 22 |
2 files changed, 34 insertions, 14 deletions
diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp index 2dab8dd..19ea603 100644 --- a/lib/Analysis/ConstantFolding.cpp +++ b/lib/Analysis/ConstantFolding.cpp @@ -129,8 +129,9 @@ static Constant *SymbolicallyEvaluateGEP(Constant* const* Ops, unsigned NumOps, Constant *Ptr = Ops[0]; if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized()) return 0; - - uint64_t BasePtr = 0; + + unsigned BitWidth = TD->getTypeSizeInBits(TD->getIntPtrType(Context)); + APInt BasePtr(BitWidth, 0); bool BaseIsInt = true; if (!Ptr->isNullValue()) { // If this is a inttoptr from a constant int, we can fold this as the base, @@ -138,7 +139,7 @@ static Constant *SymbolicallyEvaluateGEP(Constant* const* Ops, unsigned NumOps, if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr)) if (CE->getOpcode() == Instruction::IntToPtr) if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0))) - BasePtr = Base->getZExtValue(); + BasePtr = Base->getValue(); if (BasePtr == 0) BaseIsInt = false; @@ -150,12 +151,13 @@ static Constant *SymbolicallyEvaluateGEP(Constant* const* Ops, unsigned NumOps, if (!isa<ConstantInt>(Ops[i])) return 0; - uint64_t Offset = TD->getIndexedOffset(Ptr->getType(), - (Value**)Ops+1, NumOps-1); + APInt Offset = APInt(BitWidth, + TD->getIndexedOffset(Ptr->getType(), + (Value**)Ops+1, NumOps-1)); // If the base value for this address is a literal integer value, fold the // getelementptr to the resulting integer value casted to the pointer type. if (BaseIsInt) { - Constant *C = ConstantInt::get(TD->getIntPtrType(Context), Offset+BasePtr); + Constant *C = ConstantInt::get(Context, Offset+BasePtr); return ConstantExpr::getIntToPtr(C, ResultTy); } @@ -171,19 +173,21 @@ static Constant *SymbolicallyEvaluateGEP(Constant* const* Ops, unsigned NumOps, if (isa<PointerType>(ATy) && ATy != Ptr->getType()) break; // Determine which element of the array the offset points into. - uint64_t ElemSize = TD->getTypeAllocSize(ATy->getElementType()); + APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType())); if (ElemSize == 0) return 0; - uint64_t NewIdx = Offset / ElemSize; + APInt NewIdx = Offset.udiv(ElemSize); Offset -= NewIdx * ElemSize; NewIdxs.push_back(ConstantInt::get(TD->getIntPtrType(Context), NewIdx)); Ty = ATy->getElementType(); } else if (const StructType *STy = dyn_cast<StructType>(Ty)) { - // Determine which field of the struct the offset points into. + // Determine which field of the struct the offset points into. The + // getZExtValue is at least as safe as the StructLayout API because we + // know the offset is within the struct at this point. const StructLayout &SL = *TD->getStructLayout(STy); - unsigned ElIdx = SL.getElementContainingOffset(Offset); + unsigned ElIdx = SL.getElementContainingOffset(Offset.getZExtValue()); NewIdxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), ElIdx)); - Offset -= SL.getElementOffset(ElIdx); + Offset -= APInt(BitWidth, SL.getElementOffset(ElIdx)); Ty = STy->getTypeAtIndex(ElIdx); } else { // We've reached some non-indexable type. diff --git a/test/Transforms/InstCombine/constant-fold-gep-overindex.ll b/test/Transforms/InstCombine/constant-fold-gep-overindex.ll index 15bdab4..954cccb 100644 --- a/test/Transforms/InstCombine/constant-fold-gep-overindex.ll +++ b/test/Transforms/InstCombine/constant-fold-gep-overindex.ll @@ -1,17 +1,33 @@ -; RUN: llvm-as < %s | opt -instcombine +; RUN: llvm-as < %s | opt -instcombine | llvm-dis | FileCheck %s + target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" target triple = "i386-apple-darwin10.0" %0 = type { i8*, [19 x i8] } %1 = type { i8*, [0 x i8] } +@array = external global [11 x i8] + @s = external global %0 ; <%0*> [#uses=1] @"\01LC8" = external constant [17 x i8] ; <[17 x i8]*> [#uses=1] +; Instcombine should be able to fold this getelementptr. + define i32 @main() nounwind { -entry: - %0 = call i32 (i8*, ...)* @printf(i8* getelementptr ([17 x i8]* @"\01LC8", i32 0, i32 0), i8* undef, i8* getelementptr (%1* bitcast (%0* @s to %1*), i32 0, i32 1, i32 0)) nounwind ; <i32> [#uses=0] +; CHECK: call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([17 x i8]* @"\01LC8", i32 0, i32 0), i8* undef, i8* bitcast (i8** getelementptr (%1* bitcast (%0* @s to %1*), i32 1, i32 0) to i8*)) nounwind + + call i32 (i8*, ...)* @printf(i8* getelementptr ([17 x i8]* @"\01LC8", i32 0, i32 0), i8* undef, i8* getelementptr (%1* bitcast (%0* @s to %1*), i32 0, i32 1, i32 0)) nounwind ; <i32> [#uses=0] ret i32 0 } +; Instcombine should constant-fold the GEP so that indices that have +; static array extents are within bounds of those array extents. +; In the below, -1 is not in the range [0,11). After the transformation, +; the same address is computed, but 3 is in the range of [0,11). + +define i8* @foo() nounwind { +; CHECK: ret i8* getelementptr ([11 x i8]* @array, i32 390451572, i32 3) + ret i8* getelementptr ([11 x i8]* @array, i32 0, i64 -1) +} + declare i32 @printf(i8*, ...) nounwind |