lib/CodeGen/LocalStackSlotAllocation.cpp

   1 //===- LocalStackSlotAllocation.cpp - Pre-allocate locals to stack slots --===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This pass assigns local frame indices to stack slots relative to one another
  10 // and allocates additional base registers to access them when the target
  11 // estimates they are likely to be out of range of stack pointer and frame
  12 // pointer relative addressing.
  13 //
  14 //===----------------------------------------------------------------------===//
  15
  16 #include "llvm/ADT/SetVector.h"
  17 #include "llvm/ADT/SmallSet.h"
  18 #include "llvm/ADT/SmallVector.h"
  19 #include "llvm/ADT/Statistic.h"
  20 #include "llvm/CodeGen/MachineBasicBlock.h"
  21 #include "llvm/CodeGen/MachineFrameInfo.h"
  22 #include "llvm/CodeGen/MachineFunction.h"
  23 #include "llvm/CodeGen/MachineFunctionPass.h"
  24 #include "llvm/CodeGen/MachineInstr.h"
  25 #include "llvm/CodeGen/MachineOperand.h"
  26 #include "llvm/CodeGen/MachineRegisterInfo.h"
  27 #include "llvm/CodeGen/TargetFrameLowering.h"
  28 #include "llvm/CodeGen/TargetOpcodes.h"
  29 #include "llvm/CodeGen/TargetRegisterInfo.h"
  30 #include "llvm/CodeGen/TargetSubtargetInfo.h"
  31 #include "llvm/Pass.h"
  32 #include "llvm/Support/Debug.h"
  33 #include "llvm/Support/ErrorHandling.h"
  34 #include "llvm/Support/raw_ostream.h"
  35 #include <algorithm>
  36 #include <cassert>
  37 #include <cstdint>
  38 #include <tuple>
  39
  40 using namespace llvm;
  41
  42 #define DEBUG_TYPE "localstackalloc"
  43
  44 STATISTIC(NumAllocations, "Number of frame indices allocated into local block");
  45 STATISTIC(NumBaseRegisters, "Number of virtual frame base registers allocated");
  46 STATISTIC(NumReplacements, "Number of frame indices references replaced");
  47
  48 namespace {
  49
  50   class FrameRef {
  51     MachineBasicBlock::iterator MI; // Instr referencing the frame
  52     int64_t LocalOffset;            // Local offset of the frame idx referenced
  53     int FrameIdx;                   // The frame index
  54
  55     // Order reference instruction appears in program. Used to ensure
  56     // deterministic order when multiple instructions may reference the same
  57     // location.
  58     unsigned Order;
  59
  60   public:
  61     FrameRef(MachineInstr *I, int64_t Offset, int Idx, unsigned Ord) :
  62       MI(I), LocalOffset(Offset), FrameIdx(Idx), Order(Ord) {}
  63
  64     bool operator<(const FrameRef &RHS) const {
  65       return std::tie(LocalOffset, FrameIdx, Order) <
  66              std::tie(RHS.LocalOffset, RHS.FrameIdx, RHS.Order);
  67     }
  68
  69     MachineBasicBlock::iterator getMachineInstr() const { return MI; }
  70     int64_t getLocalOffset() const { return LocalOffset; }
  71     int getFrameIndex() const { return FrameIdx; }
  72   };
  73
  74   class LocalStackSlotPass: public MachineFunctionPass {
  75     SmallVector<int64_t, 16> LocalOffsets;
  76
  77     /// StackObjSet - A set of stack object indexes
  78     using StackObjSet = SmallSetVector<int, 8>;
  79
  80     void AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx, int64_t &Offset,
  81                            bool StackGrowsDown, unsigned &MaxAlign);
  82     void AssignProtectedObjSet(const StackObjSet &UnassignedObjs,
  83                                SmallSet<int, 16> &ProtectedObjs,
  84                                MachineFrameInfo &MFI, bool StackGrowsDown,
  85                                int64_t &Offset, unsigned &MaxAlign);
  86     void calculateFrameObjectOffsets(MachineFunction &Fn);
  87     bool insertFrameReferenceRegisters(MachineFunction &Fn);
  88
  89   public:
  90     static char ID; // Pass identification, replacement for typeid
  91
  92     explicit LocalStackSlotPass() : MachineFunctionPass(ID) {
  93       initializeLocalStackSlotPassPass(*PassRegistry::getPassRegistry());
  94     }
  95
  96     bool runOnMachineFunction(MachineFunction &MF) override;
  97
  98     void getAnalysisUsage(AnalysisUsage &AU) const override {
  99       AU.setPreservesCFG();
 100       MachineFunctionPass::getAnalysisUsage(AU);
 101     }
 102   };
 103
 104 } // end anonymous namespace
 105
 106 char LocalStackSlotPass::ID = 0;
 107
 108 char &llvm::LocalStackSlotAllocationID = LocalStackSlotPass::ID;
 109 INITIALIZE_PASS(LocalStackSlotPass, DEBUG_TYPE,
 110                 "Local Stack Slot Allocation", false, false)
 111
 112 bool LocalStackSlotPass::runOnMachineFunction(MachineFunction &MF) {
 113   MachineFrameInfo &MFI = MF.getFrameInfo();
 114   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
 115   unsigned LocalObjectCount = MFI.getObjectIndexEnd();
 116
 117   // If the target doesn't want/need this pass, or if there are no locals
 118   // to consider, early exit.
 119   if (!TRI->requiresVirtualBaseRegisters(MF) || LocalObjectCount == 0)
 120     return true;
 121
 122   // Make sure we have enough space to store the local offsets.
 123   LocalOffsets.resize(MFI.getObjectIndexEnd());
 124
 125   // Lay out the local blob.
 126   calculateFrameObjectOffsets(MF);
 127
 128   // Insert virtual base registers to resolve frame index references.
 129   bool UsedBaseRegs = insertFrameReferenceRegisters(MF);
 130
 131   // Tell MFI whether any base registers were allocated. PEI will only
 132   // want to use the local block allocations from this pass if there were any.
 133   // Otherwise, PEI can do a bit better job of getting the alignment right
 134   // without a hole at the start since it knows the alignment of the stack
 135   // at the start of local allocation, and this pass doesn't.
 136   MFI.setUseLocalStackAllocationBlock(UsedBaseRegs);
 137
 138   return true;
 139 }
 140
 141 /// AdjustStackOffset - Helper function used to adjust the stack frame offset.
 142 void LocalStackSlotPass::AdjustStackOffset(MachineFrameInfo &MFI,
 143                                            int FrameIdx, int64_t &Offset,
 144                                            bool StackGrowsDown,
 145                                            unsigned &MaxAlign) {
 146   // If the stack grows down, add the object size to find the lowest address.
 147   if (StackGrowsDown)
 148     Offset += MFI.getObjectSize(FrameIdx);
 149
 150   unsigned Align = MFI.getObjectAlignment(FrameIdx);
 151
 152   // If the alignment of this object is greater than that of the stack, then
 153   // increase the stack alignment to match.
 154   MaxAlign = std::max(MaxAlign, Align);
 155
 156   // Adjust to alignment boundary.
 157   Offset = (Offset + Align - 1) / Align * Align;
 158
 159   int64_t LocalOffset = StackGrowsDown ? -Offset : Offset;
 160   LLVM_DEBUG(dbgs() << "Allocate FI(" << FrameIdx << ") to local offset "
 161                     << LocalOffset << "\n");
 162   // Keep the offset available for base register allocation
 163   LocalOffsets[FrameIdx] = LocalOffset;
 164   // And tell MFI about it for PEI to use later
 165   MFI.mapLocalFrameObject(FrameIdx, LocalOffset);
 166
 167   if (!StackGrowsDown)
 168     Offset += MFI.getObjectSize(FrameIdx);
 169
 170   ++NumAllocations;
 171 }
 172
 173 /// AssignProtectedObjSet - Helper function to assign large stack objects (i.e.,
 174 /// those required to be close to the Stack Protector) to stack offsets.
 175 void LocalStackSlotPass::AssignProtectedObjSet(const StackObjSet &UnassignedObjs,
 176                                            SmallSet<int, 16> &ProtectedObjs,
 177                                            MachineFrameInfo &MFI,
 178                                            bool StackGrowsDown, int64_t &Offset,
 179                                            unsigned &MaxAlign) {
 180   for (StackObjSet::const_iterator I = UnassignedObjs.begin(),
 181         E = UnassignedObjs.end(); I != E; ++I) {
 182     int i = *I;
 183     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 184     ProtectedObjs.insert(i);
 185   }
 186 }
 187
 188 /// calculateFrameObjectOffsets - Calculate actual frame offsets for all of the
 189 /// abstract stack objects.
 190 void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
 191   // Loop over all of the stack objects, assigning sequential addresses...
 192   MachineFrameInfo &MFI = Fn.getFrameInfo();
 193   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 194   bool StackGrowsDown =
 195     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 196   int64_t Offset = 0;
 197   unsigned MaxAlign = 0;
 198
 199   // Make sure that the stack protector comes before the local variables on the
 200   // stack.
 201   SmallSet<int, 16> ProtectedObjs;
 202   if (MFI.hasStackProtectorIndex()) {
 203     int StackProtectorFI = MFI.getStackProtectorIndex();
 204
 205     // We need to make sure we didn't pre-allocate the stack protector when
 206     // doing this.
 207     // If we already have a stack protector, this will re-assign it to a slot
 208     // that is **not** covering the protected objects.
 209     assert(!MFI.isObjectPreAllocated(StackProtectorFI) &&
 210            "Stack protector pre-allocated in LocalStackSlotAllocation");
 211
 212     StackObjSet LargeArrayObjs;
 213     StackObjSet SmallArrayObjs;
 214     StackObjSet AddrOfObjs;
 215
 216     AdjustStackOffset(MFI, StackProtectorFI, Offset, StackGrowsDown, MaxAlign);
 217
 218     // Assign large stack objects first.
 219     for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 220       if (MFI.isDeadObjectIndex(i))
 221         continue;
 222       if (StackProtectorFI == (int)i)
 223         continue;
 224
 225       switch (MFI.getObjectSSPLayout(i)) {
 226       case MachineFrameInfo::SSPLK_None:
 227         continue;
 228       case MachineFrameInfo::SSPLK_SmallArray:
 229         SmallArrayObjs.insert(i);
 230         continue;
 231       case MachineFrameInfo::SSPLK_AddrOf:
 232         AddrOfObjs.insert(i);
 233         continue;
 234       case MachineFrameInfo::SSPLK_LargeArray:
 235         LargeArrayObjs.insert(i);
 236         continue;
 237       }
 238       llvm_unreachable("Unexpected SSPLayoutKind.");
 239     }
 240
 241     AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 242                           Offset, MaxAlign);
 243     AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 244                           Offset, MaxAlign);
 245     AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
 246                           Offset, MaxAlign);
 247   }
 248
 249   // Then assign frame offsets to stack objects that are not used to spill
 250   // callee saved registers.
 251   for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 252     if (MFI.isDeadObjectIndex(i))
 253       continue;
 254     if (MFI.getStackProtectorIndex() == (int)i)
 255       continue;
 256     if (ProtectedObjs.count(i))
 257       continue;
 258
 259     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 260   }
 261
 262   // Remember how big this blob of stack space is
 263   MFI.setLocalFrameSize(Offset);
 264   MFI.setLocalFrameMaxAlign(MaxAlign);
 265 }
 266
 267 static inline bool
 268 lookupCandidateBaseReg(unsigned BaseReg,
 269                        int64_t BaseOffset,
 270                        int64_t FrameSizeAdjust,
 271                        int64_t LocalFrameOffset,
 272                        const MachineInstr &MI,
 273                        const TargetRegisterInfo *TRI) {
 274   // Check if the relative offset from the where the base register references
 275   // to the target address is in range for the instruction.
 276   int64_t Offset = FrameSizeAdjust + LocalFrameOffset - BaseOffset;
 277   return TRI->isFrameOffsetLegal(&MI, BaseReg, Offset);
 278 }
 279
 280 bool LocalStackSlotPass::insertFrameReferenceRegisters(MachineFunction &Fn) {
 281   // Scan the function's instructions looking for frame index references.
 282   // For each, ask the target if it wants a virtual base register for it
 283   // based on what we can tell it about where the local will end up in the
 284   // stack frame. If it wants one, re-use a suitable one we've previously
 285   // allocated, or if there isn't one that fits the bill, allocate a new one
 286   // and ask the target to create a defining instruction for it.
 287   bool UsedBaseReg = false;
 288
 289   MachineFrameInfo &MFI = Fn.getFrameInfo();
 290   const TargetRegisterInfo *TRI = Fn.getSubtarget().getRegisterInfo();
 291   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 292   bool StackGrowsDown =
 293     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 294
 295   // Collect all of the instructions in the block that reference
 296   // a frame index. Also store the frame index referenced to ease later
 297   // lookup. (For any insn that has more than one FI reference, we arbitrarily
 298   // choose the first one).
 299   SmallVector<FrameRef, 64> FrameReferenceInsns;
 300
 301   unsigned Order = 0;
 302
 303   for (MachineBasicBlock &BB : Fn) {
 304     for (MachineInstr &MI : BB) {
 305       // Debug value, stackmap and patchpoint instructions can't be out of
 306       // range, so they don't need any updates.
 307       if (MI.isDebugInstr() || MI.getOpcode() == TargetOpcode::STATEPOINT ||
 308           MI.getOpcode() == TargetOpcode::STACKMAP ||
 309           MI.getOpcode() == TargetOpcode::PATCHPOINT)
 310         continue;
 311
 312       // For now, allocate the base register(s) within the basic block
 313       // where they're used, and don't try to keep them around outside
 314       // of that. It may be beneficial to try sharing them more broadly
 315       // than that, but the increased register pressure makes that a
 316       // tricky thing to balance. Investigate if re-materializing these
 317       // becomes an issue.
 318       for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
 319         // Consider replacing all frame index operands that reference
 320         // an object allocated in the local block.
 321         if (MI.getOperand(i).isFI()) {
 322           // Don't try this with values not in the local block.
 323           if (!MFI.isObjectPreAllocated(MI.getOperand(i).getIndex()))
 324             break;
 325           int Idx = MI.getOperand(i).getIndex();
 326           int64_t LocalOffset = LocalOffsets[Idx];
 327           if (!TRI->needsFrameBaseReg(&MI, LocalOffset))
 328             break;
 329           FrameReferenceInsns.push_back(FrameRef(&MI, LocalOffset, Idx, Order++));
 330           break;
 331         }
 332       }
 333     }
 334   }
 335
 336   // Sort the frame references by local offset.
 337   // Use frame index as a tie-breaker in case MI's have the same offset.
 338   llvm::sort(FrameReferenceInsns);
 339
 340   MachineBasicBlock *Entry = &Fn.front();
 341
 342   unsigned BaseReg = 0;
 343   int64_t BaseOffset = 0;
 344
 345   // Loop through the frame references and allocate for them as necessary.
 346   for (int ref = 0, e = FrameReferenceInsns.size(); ref < e ; ++ref) {
 347     FrameRef &FR = FrameReferenceInsns[ref];
 348     MachineInstr &MI = *FR.getMachineInstr();
 349     int64_t LocalOffset = FR.getLocalOffset();
 350     int FrameIdx = FR.getFrameIndex();
 351     assert(MFI.isObjectPreAllocated(FrameIdx) &&
 352            "Only pre-allocated locals expected!");
 353
 354     // We need to keep the references to the stack protector slot through frame
 355     // index operands so that it gets resolved by PEI rather than this pass.
 356     // This avoids accesses to the stack protector though virtual base
 357     // registers, and forces PEI to address it using fp/sp/bp.
 358     if (MFI.hasStackProtectorIndex() &&
 359         FrameIdx == MFI.getStackProtectorIndex())
 360       continue;
 361
 362     LLVM_DEBUG(dbgs() << "Considering: " << MI);
 363
 364     unsigned idx = 0;
 365     for (unsigned f = MI.getNumOperands(); idx != f; ++idx) {
 366       if (!MI.getOperand(idx).isFI())
 367         continue;
 368
 369       if (FrameIdx == MI.getOperand(idx).getIndex())
 370         break;
 371     }
 372
 373     assert(idx < MI.getNumOperands() && "Cannot find FI operand");
 374
 375     int64_t Offset = 0;
 376     int64_t FrameSizeAdjust = StackGrowsDown ? MFI.getLocalFrameSize() : 0;
 377
 378     LLVM_DEBUG(dbgs() << "  Replacing FI in: " << MI);
 379
 380     // If we have a suitable base register available, use it; otherwise
 381     // create a new one. Note that any offset encoded in the
 382     // instruction itself will be taken into account by the target,
 383     // so we don't have to adjust for it here when reusing a base
 384     // register.
 385     if (UsedBaseReg &&
 386         lookupCandidateBaseReg(BaseReg, BaseOffset, FrameSizeAdjust,
 387                                LocalOffset, MI, TRI)) {
 388       LLVM_DEBUG(dbgs() << "  Reusing base register " << BaseReg << "\n");
 389       // We found a register to reuse.
 390       Offset = FrameSizeAdjust + LocalOffset - BaseOffset;
 391     } else {
 392       // No previously defined register was in range, so create a new one.
 393       int64_t InstrOffset = TRI->getFrameIndexInstrOffset(&MI, idx);
 394
 395       int64_t PrevBaseOffset = BaseOffset;
 396       BaseOffset = FrameSizeAdjust + LocalOffset + InstrOffset;
 397
 398       // We'd like to avoid creating single-use virtual base registers.
 399       // Because the FrameRefs are in sorted order, and we've already
 400       // processed all FrameRefs before this one, just check whether or not
 401       // the next FrameRef will be able to reuse this new register. If not,
 402       // then don't bother creating it.
 403       if (ref + 1 >= e ||
 404           !lookupCandidateBaseReg(
 405               BaseReg, BaseOffset, FrameSizeAdjust,
 406               FrameReferenceInsns[ref + 1].getLocalOffset(),
 407               *FrameReferenceInsns[ref + 1].getMachineInstr(), TRI)) {
 408         BaseOffset = PrevBaseOffset;
 409         continue;
 410       }
 411
 412       const MachineFunction *MF = MI.getMF();
 413       const TargetRegisterClass *RC = TRI->getPointerRegClass(*MF);
 414       BaseReg = Fn.getRegInfo().createVirtualRegister(RC);
 415
 416       LLVM_DEBUG(dbgs() << "  Materializing base register " << BaseReg
 417                         << " at frame local offset "
 418                         << LocalOffset + InstrOffset << "\n");
 419
 420       // Tell the target to insert the instruction to initialize
 421       // the base register.
 422       //            MachineBasicBlock::iterator InsertionPt = Entry->begin();
 423       TRI->materializeFrameBaseRegister(Entry, BaseReg, FrameIdx,
 424                                         InstrOffset);
 425
 426       // The base register already includes any offset specified
 427       // by the instruction, so account for that so it doesn't get
 428       // applied twice.
 429       Offset = -InstrOffset;
 430
 431       ++NumBaseRegisters;
 432       UsedBaseReg = true;
 433     }
 434     assert(BaseReg != 0 && "Unable to allocate virtual base register!");
 435
 436     // Modify the instruction to use the new base register rather
 437     // than the frame index operand.
 438     TRI->resolveFrameIndex(MI, BaseReg, Offset);
 439     LLVM_DEBUG(dbgs() << "Resolved: " << MI);
 440
 441     ++NumReplacements;
 442   }
 443
 444   return UsedBaseReg;
 445 }