X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FLocalStackSlotAllocation.cpp;h=26a117652b08248dd8b4b8f941028d9063c6a6e9;hb=200bb6df7ff84e2129bcc0f4f79d6d96d47eb645;hp=dabfb469c34f5d05c5d0db8c6347b8a3294dfc2b;hpb=3d72367d30c9ce6f387764a028763f7a366cc443;p=oota-llvm.git diff --git a/lib/CodeGen/LocalStackSlotAllocation.cpp b/lib/CodeGen/LocalStackSlotAllocation.cpp index dabfb469c34..26a117652b0 100644 --- a/lib/CodeGen/LocalStackSlotAllocation.cpp +++ b/lib/CodeGen/LocalStackSlotAllocation.cpp @@ -9,40 +9,62 @@ // // This pass assigns local frame indices to stack slots relative to one another // and allocates additional base registers to access them when the target -// estimates the are likely to be out of range of stack pointer and frame +// estimates they are likely to be out of range of stack pointer and frame // pointer relative addressing. // //===----------------------------------------------------------------------===// #define DEBUG_TYPE "localstackalloc" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/Intrinsics.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" +#include "llvm/CodeGen/Passes.h" +#include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/Statistic.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineFunctionPass.h" -#include "llvm/CodeGen/Passes.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetFrameLowering.h" #include "llvm/Target/TargetRegisterInfo.h" -#include "llvm/Target/TargetFrameInfo.h" using namespace llvm; -STATISTIC(NumAllocations, "Number of frame indices processed"); +STATISTIC(NumAllocations, "Number of frame indices allocated into local block"); +STATISTIC(NumBaseRegisters, "Number of virtual frame base registers allocated"); +STATISTIC(NumReplacements, "Number of frame indices references replaced"); namespace { + class FrameRef { + MachineBasicBlock::iterator MI; // Instr referencing the frame + int64_t LocalOffset; // Local offset of the frame idx referenced + int FrameIdx; // The frame index + public: + FrameRef(MachineBasicBlock::iterator I, int64_t Offset, int Idx) : + MI(I), LocalOffset(Offset), FrameIdx(Idx) {} + bool operator<(const FrameRef &RHS) const { + return LocalOffset < RHS.LocalOffset; + } + MachineBasicBlock::iterator getMachineInstr() const { return MI; } + int64_t getLocalOffset() const { return LocalOffset; } + int getFrameIndex() const { return FrameIdx; } + }; + class LocalStackSlotPass: public MachineFunctionPass { - int64_t LocalStackSize; + SmallVector LocalOffsets; + void AdjustStackOffset(MachineFrameInfo *MFI, int FrameIdx, int64_t &Offset, + bool StackGrowsDown, unsigned &MaxAlign); void calculateFrameObjectOffsets(MachineFunction &Fn); + bool insertFrameReferenceRegisters(MachineFunction &Fn); public: static char ID; // Pass identification, replacement for typeid explicit LocalStackSlotPass() : MachineFunctionPass(ID) { } @@ -52,30 +74,54 @@ namespace { AU.setPreservesCFG(); MachineFunctionPass::getAnalysisUsage(AU); } - const char *getPassName() const { - return "Local Stack Slot Allocation"; - } private: }; } // end anonymous namespace char LocalStackSlotPass::ID = 0; - -FunctionPass *llvm::createLocalStackSlotAllocationPass() { - return new LocalStackSlotPass(); -} +char &llvm::LocalStackSlotAllocationID = LocalStackSlotPass::ID; +INITIALIZE_PASS(LocalStackSlotPass, "localstackalloc", + "Local Stack Slot Allocation", false, false) bool LocalStackSlotPass::runOnMachineFunction(MachineFunction &MF) { + MachineFrameInfo *MFI = MF.getFrameInfo(); + const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo(); + unsigned LocalObjectCount = MFI->getObjectIndexEnd(); + + // If the target doesn't want/need this pass, or if there are no locals + // to consider, early exit. + if (!TRI->requiresVirtualBaseRegisters(MF) || LocalObjectCount == 0) + return true; + + // Make sure we have enough space to store the local offsets. + LocalOffsets.resize(MFI->getObjectIndexEnd()); + + // Lay out the local blob. calculateFrameObjectOffsets(MF); - DEBUG(dbgs() << LocalStackSize << " bytes of local storage pre-allocated\n"); + + // Insert virtual base registers to resolve frame index references. + bool UsedBaseRegs = insertFrameReferenceRegisters(MF); + + // Tell MFI whether any base registers were allocated. PEI will only + // want to use the local block allocations from this pass if there were any. + // Otherwise, PEI can do a bit better job of getting the alignment right + // without a hole at the start since it knows the alignment of the stack + // at the start of local allocation, and this pass doesn't. + MFI->setUseLocalStackAllocationBlock(UsedBaseRegs); + return true; } /// AdjustStackOffset - Helper function used to adjust the stack frame offset. -static inline void -AdjustStackOffset(MachineFrameInfo *MFI, int FrameIdx, int64_t &Offset, - unsigned &MaxAlign) { +void LocalStackSlotPass::AdjustStackOffset(MachineFrameInfo *MFI, + int FrameIdx, int64_t &Offset, + bool StackGrowsDown, + unsigned &MaxAlign) { + // If the stack grows down, add the object size to find the lowest address. + if (StackGrowsDown) + Offset += MFI->getObjectSize(FrameIdx); + unsigned Align = MFI->getObjectAlignment(FrameIdx); // If the alignment of this object is greater than that of the stack, then @@ -85,10 +131,16 @@ AdjustStackOffset(MachineFrameInfo *MFI, int FrameIdx, int64_t &Offset, // Adjust to alignment boundary. Offset = (Offset + Align - 1) / Align * Align; + int64_t LocalOffset = StackGrowsDown ? -Offset : Offset; DEBUG(dbgs() << "Allocate FI(" << FrameIdx << ") to local offset " - << Offset << "\n"); - MFI->mapLocalFrameObject(FrameIdx, Offset); - Offset += MFI->getObjectSize(FrameIdx); + << LocalOffset << "\n"); + // Keep the offset available for base register allocation + LocalOffsets[FrameIdx] = LocalOffset; + // And tell MFI about it for PEI to use later + MFI->mapLocalFrameObject(FrameIdx, LocalOffset); + + if (!StackGrowsDown) + Offset += MFI->getObjectSize(FrameIdx); ++NumAllocations; } @@ -97,18 +149,20 @@ AdjustStackOffset(MachineFrameInfo *MFI, int FrameIdx, int64_t &Offset, /// abstract stack objects. /// void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) { - const TargetFrameInfo &TFI = *Fn.getTarget().getFrameInfo(); - // Loop over all of the stack objects, assigning sequential addresses... MachineFrameInfo *MFI = Fn.getFrameInfo(); + const TargetFrameLowering &TFI = *Fn.getTarget().getFrameLowering(); + bool StackGrowsDown = + TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown; int64_t Offset = 0; - unsigned MaxAlign = MFI->getMaxAlignment(); + unsigned MaxAlign = 0; // Make sure that the stack protector comes before the local variables on the // stack. SmallSet LargeStackObjs; if (MFI->getStackProtectorIndex() >= 0) { - AdjustStackOffset(MFI, MFI->getStackProtectorIndex(), Offset, MaxAlign); + AdjustStackOffset(MFI, MFI->getStackProtectorIndex(), Offset, + StackGrowsDown, MaxAlign); // Assign large stack objects first. for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) { @@ -119,7 +173,7 @@ void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) { if (!MFI->MayNeedStackProtector(i)) continue; - AdjustStackOffset(MFI, i, Offset, MaxAlign); + AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign); LargeStackObjs.insert(i); } } @@ -134,36 +188,186 @@ void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) { if (LargeStackObjs.count(i)) continue; - AdjustStackOffset(MFI, i, Offset, MaxAlign); + AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign); } - const TargetRegisterInfo *RegInfo = Fn.getTarget().getRegisterInfo(); - if (!RegInfo->targetHandlesStackFrameRounding()) { - // If we have reserved argument space for call sites in the function - // immediately on entry to the current function, count it as part of the - // overall stack size. - if (MFI->adjustsStack() && RegInfo->hasReservedCallFrame(Fn)) - Offset += MFI->getMaxCallFrameSize(); - - // Round up the size to a multiple of the alignment. If the function has - // any calls or alloca's, align to the target's StackAlignment value to - // ensure that the callee's frame or the alloca data is suitably aligned; - // otherwise, for leaf functions, align to the TransientStackAlignment - // value. - unsigned StackAlign; - if (MFI->adjustsStack() || MFI->hasVarSizedObjects() || - (RegInfo->needsStackRealignment(Fn) && MFI->getObjectIndexEnd() != 0)) - StackAlign = TFI.getStackAlignment(); - else - StackAlign = TFI.getTransientStackAlignment(); - - // If the frame pointer is eliminated, all frame offsets will be relative to - // SP not FP. Align to MaxAlign so this works. - StackAlign = std::max(StackAlign, MaxAlign); - unsigned AlignMask = StackAlign - 1; - Offset = (Offset + AlignMask) & ~uint64_t(AlignMask); + // Remember how big this blob of stack space is + MFI->setLocalFrameSize(Offset); + MFI->setLocalFrameMaxAlign(MaxAlign); +} + +static inline bool +lookupCandidateBaseReg(int64_t BaseOffset, + int64_t FrameSizeAdjust, + int64_t LocalFrameOffset, + const MachineInstr *MI, + const TargetRegisterInfo *TRI) { + // Check if the relative offset from the where the base register references + // to the target address is in range for the instruction. + int64_t Offset = FrameSizeAdjust + LocalFrameOffset - BaseOffset; + return TRI->isFrameOffsetLegal(MI, Offset); +} + +bool LocalStackSlotPass::insertFrameReferenceRegisters(MachineFunction &Fn) { + // Scan the function's instructions looking for frame index references. + // For each, ask the target if it wants a virtual base register for it + // based on what we can tell it about where the local will end up in the + // stack frame. If it wants one, re-use a suitable one we've previously + // allocated, or if there isn't one that fits the bill, allocate a new one + // and ask the target to create a defining instruction for it. + bool UsedBaseReg = false; + + MachineFrameInfo *MFI = Fn.getFrameInfo(); + const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo(); + const TargetFrameLowering &TFI = *Fn.getTarget().getFrameLowering(); + bool StackGrowsDown = + TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown; + + // Collect all of the instructions in the block that reference + // a frame index. Also store the frame index referenced to ease later + // lookup. (For any insn that has more than one FI reference, we arbitrarily + // choose the first one). + SmallVector FrameReferenceInsns; + + for (MachineFunction::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) { + for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) { + MachineInstr *MI = I; + + // Debug value instructions can't be out of range, so they don't need + // any updates. + if (MI->isDebugValue()) + continue; + + // For now, allocate the base register(s) within the basic block + // where they're used, and don't try to keep them around outside + // of that. It may be beneficial to try sharing them more broadly + // than that, but the increased register pressure makes that a + // tricky thing to balance. Investigate if re-materializing these + // becomes an issue. + for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { + // Consider replacing all frame index operands that reference + // an object allocated in the local block. + if (MI->getOperand(i).isFI()) { + // Don't try this with values not in the local block. + if (!MFI->isObjectPreAllocated(MI->getOperand(i).getIndex())) + break; + int Idx = MI->getOperand(i).getIndex(); + int64_t LocalOffset = LocalOffsets[Idx]; + if (!TRI->needsFrameBaseReg(MI, LocalOffset)) + break; + FrameReferenceInsns. + push_back(FrameRef(MI, LocalOffset, Idx)); + break; + } + } + } } - // Remember how big this blob of stack space is - LocalStackSize = Offset; + // Sort the frame references by local offset + array_pod_sort(FrameReferenceInsns.begin(), FrameReferenceInsns.end()); + + MachineBasicBlock *Entry = Fn.begin(); + + unsigned BaseReg = 0; + int64_t BaseOffset = 0; + + // Loop through the frame references and allocate for them as necessary. + for (int ref = 0, e = FrameReferenceInsns.size(); ref < e ; ++ref) { + FrameRef &FR = FrameReferenceInsns[ref]; + MachineBasicBlock::iterator I = FR.getMachineInstr(); + MachineInstr *MI = I; + int64_t LocalOffset = FR.getLocalOffset(); + int FrameIdx = FR.getFrameIndex(); + assert(MFI->isObjectPreAllocated(FrameIdx) && + "Only pre-allocated locals expected!"); + + DEBUG(dbgs() << "Considering: " << *MI); + + unsigned idx = 0; + for (unsigned f = MI->getNumOperands(); idx != f; ++idx) { + if (!MI->getOperand(idx).isFI()) + continue; + + if (FrameIdx == I->getOperand(idx).getIndex()) + break; + } + + assert(idx < MI->getNumOperands() && "Cannot find FI operand"); + + int64_t Offset = 0; + int64_t FrameSizeAdjust = StackGrowsDown ? MFI->getLocalFrameSize() : 0; + + DEBUG(dbgs() << " Replacing FI in: " << *MI); + + // If we have a suitable base register available, use it; otherwise + // create a new one. Note that any offset encoded in the + // instruction itself will be taken into account by the target, + // so we don't have to adjust for it here when reusing a base + // register. + if (UsedBaseReg && lookupCandidateBaseReg(BaseOffset, FrameSizeAdjust, + LocalOffset, MI, TRI)) { + DEBUG(dbgs() << " Reusing base register " << BaseReg << "\n"); + // We found a register to reuse. + Offset = FrameSizeAdjust + LocalOffset - BaseOffset; + } else { + // No previously defined register was in range, so create a // new one. + + int64_t InstrOffset = TRI->getFrameIndexInstrOffset(MI, idx); + + int64_t PrevBaseOffset = BaseOffset; + BaseOffset = FrameSizeAdjust + LocalOffset + InstrOffset; + + // We'd like to avoid creating single-use virtual base registers. + // Because the FrameRefs are in sorted order, and we've already + // processed all FrameRefs before this one, just check whether or not + // the next FrameRef will be able to reuse this new register. If not, + // then don't bother creating it. + bool CanReuse = false; + for (int refn = ref + 1; refn < e; ++refn) { + FrameRef &FRN = FrameReferenceInsns[refn]; + MachineBasicBlock::iterator J = FRN.getMachineInstr(); + MachineInstr *MIN = J; + + CanReuse = lookupCandidateBaseReg(BaseOffset, FrameSizeAdjust, + FRN.getLocalOffset(), MIN, TRI); + break; + } + + if (!CanReuse) { + BaseOffset = PrevBaseOffset; + continue; + } + + const MachineFunction *MF = MI->getParent()->getParent(); + const TargetRegisterClass *RC = TRI->getPointerRegClass(*MF); + BaseReg = Fn.getRegInfo().createVirtualRegister(RC); + + DEBUG(dbgs() << " Materializing base register " << BaseReg << + " at frame local offset " << LocalOffset + InstrOffset << "\n"); + + // Tell the target to insert the instruction to initialize + // the base register. + // MachineBasicBlock::iterator InsertionPt = Entry->begin(); + TRI->materializeFrameBaseRegister(Entry, BaseReg, FrameIdx, + InstrOffset); + + // The base register already includes any offset specified + // by the instruction, so account for that so it doesn't get + // applied twice. + Offset = -InstrOffset; + + ++NumBaseRegisters; + UsedBaseReg = true; + } + assert(BaseReg != 0 && "Unable to allocate virtual base register!"); + + // Modify the instruction to use the new base register rather + // than the frame index operand. + TRI->resolveFrameIndex(I, BaseReg, Offset); + DEBUG(dbgs() << "Resolved: " << *MI); + + ++NumReplacements; + } + + return UsedBaseReg; }