llvm/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/InitializePasses.h"
  32 #include "llvm/Pass.h"
  33 #include "llvm/Support/Debug.h"
  34 #include "llvm/Support/ErrorHandling.h"
  35 #include "llvm/Support/raw_ostream.h"
  36 #include <algorithm>
  37 #include <cassert>
  38 #include <cstdint>
  39 #include <tuple>
  40
  41 using namespace llvm;
  42
  43 #define DEBUG_TYPE "localstackalloc"
  44
  45 STATISTIC(NumAllocations, "Number of frame indices allocated into local block");
  46 STATISTIC(NumBaseRegisters, "Number of virtual frame base registers allocated");
  47 STATISTIC(NumReplacements, "Number of frame indices references replaced");
  48
  49 namespace {
  50
  51   class FrameRef {
  52     MachineBasicBlock::iterator MI; // Instr referencing the frame
  53     int64_t LocalOffset;            // Local offset of the frame idx referenced
  54     int FrameIdx;                   // The frame index
  55
  56     // Order reference instruction appears in program. Used to ensure
  57     // deterministic order when multiple instructions may reference the same
  58     // location.
  59     unsigned Order;
  60
  61   public:
  62     FrameRef(MachineInstr *I, int64_t Offset, int Idx, unsigned Ord) :
  63       MI(I), LocalOffset(Offset), FrameIdx(Idx), Order(Ord) {}
  64
  65     bool operator<(const FrameRef &RHS) const {
  66       return std::tie(LocalOffset, FrameIdx, Order) <
  67              std::tie(RHS.LocalOffset, RHS.FrameIdx, RHS.Order);
  68     }
  69
  70     MachineBasicBlock::iterator getMachineInstr() const { return MI; }
  71     int64_t getLocalOffset() const { return LocalOffset; }
  72     int getFrameIndex() const { return FrameIdx; }
  73   };
  74
  75   class LocalStackSlotPass: public MachineFunctionPass {
  76     SmallVector<int64_t, 16> LocalOffsets;
  77
  78     /// StackObjSet - A set of stack object indexes
  79     using StackObjSet = SmallSetVector<int, 8>;
  80
  81     void AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx, int64_t &Offset,
  82                            bool StackGrowsDown, Align &MaxAlign);
  83     void AssignProtectedObjSet(const StackObjSet &UnassignedObjs,
  84                                SmallSet<int, 16> &ProtectedObjs,
  85                                MachineFrameInfo &MFI, bool StackGrowsDown,
  86                                int64_t &Offset, Align &MaxAlign);
  87     void calculateFrameObjectOffsets(MachineFunction &Fn);
  88     bool insertFrameReferenceRegisters(MachineFunction &Fn);
  89
  90   public:
  91     static char ID; // Pass identification, replacement for typeid
  92
  93     explicit LocalStackSlotPass() : MachineFunctionPass(ID) {
  94       initializeLocalStackSlotPassPass(*PassRegistry::getPassRegistry());
  95     }
  96
  97     bool runOnMachineFunction(MachineFunction &MF) override;
  98
  99     void getAnalysisUsage(AnalysisUsage &AU) const override {
 100       AU.setPreservesCFG();
 101       MachineFunctionPass::getAnalysisUsage(AU);
 102     }
 103   };
 104
 105 } // end anonymous namespace
 106
 107 char LocalStackSlotPass::ID = 0;
 108
 109 char &llvm::LocalStackSlotAllocationID = LocalStackSlotPass::ID;
 110 INITIALIZE_PASS(LocalStackSlotPass, DEBUG_TYPE,
 111                 "Local Stack Slot Allocation", false, false)
 112
 113 bool LocalStackSlotPass::runOnMachineFunction(MachineFunction &MF) {
 114   MachineFrameInfo &MFI = MF.getFrameInfo();
 115   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
 116   unsigned LocalObjectCount = MFI.getObjectIndexEnd();
 117
 118   // If the target doesn't want/need this pass, or if there are no locals
 119   // to consider, early exit.
 120   if (LocalObjectCount == 0 || !TRI->requiresVirtualBaseRegisters(MF))
 121     return true;
 122
 123   // Make sure we have enough space to store the local offsets.
 124   LocalOffsets.resize(MFI.getObjectIndexEnd());
 125
 126   // Lay out the local blob.
 127   calculateFrameObjectOffsets(MF);
 128
 129   // Insert virtual base registers to resolve frame index references.
 130   bool UsedBaseRegs = insertFrameReferenceRegisters(MF);
 131
 132   // Tell MFI whether any base registers were allocated. PEI will only
 133   // want to use the local block allocations from this pass if there were any.
 134   // Otherwise, PEI can do a bit better job of getting the alignment right
 135   // without a hole at the start since it knows the alignment of the stack
 136   // at the start of local allocation, and this pass doesn't.
 137   MFI.setUseLocalStackAllocationBlock(UsedBaseRegs);
 138
 139   return true;
 140 }
 141
 142 /// AdjustStackOffset - Helper function used to adjust the stack frame offset.
 143 void LocalStackSlotPass::AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx,
 144                                            int64_t &Offset, bool StackGrowsDown,
 145                                            Align &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   Align Alignment = MFI.getObjectAlign(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, Alignment);
 155
 156   // Adjust to alignment boundary.
 157   Offset = alignTo(Offset, Alignment);
 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(
 176     const StackObjSet &UnassignedObjs, SmallSet<int, 16> &ProtectedObjs,
 177     MachineFrameInfo &MFI, bool StackGrowsDown, int64_t &Offset,
 178     Align &MaxAlign) {
 179   for (int i : UnassignedObjs) {
 180     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 181     ProtectedObjs.insert(i);
 182   }
 183 }
 184
 185 /// calculateFrameObjectOffsets - Calculate actual frame offsets for all of the
 186 /// abstract stack objects.
 187 void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
 188   // Loop over all of the stack objects, assigning sequential addresses...
 189   MachineFrameInfo &MFI = Fn.getFrameInfo();
 190   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 191   bool StackGrowsDown =
 192     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 193   int64_t Offset = 0;
 194   Align MaxAlign;
 195
 196   // Make sure that the stack protector comes before the local variables on the
 197   // stack.
 198   SmallSet<int, 16> ProtectedObjs;
 199   if (MFI.hasStackProtectorIndex()) {
 200     int StackProtectorFI = MFI.getStackProtectorIndex();
 201
 202     // We need to make sure we didn't pre-allocate the stack protector when
 203     // doing this.
 204     // If we already have a stack protector, this will re-assign it to a slot
 205     // that is **not** covering the protected objects.
 206     assert(!MFI.isObjectPreAllocated(StackProtectorFI) &&
 207            "Stack protector pre-allocated in LocalStackSlotAllocation");
 208
 209     StackObjSet LargeArrayObjs;
 210     StackObjSet SmallArrayObjs;
 211     StackObjSet AddrOfObjs;
 212
 213     AdjustStackOffset(MFI, StackProtectorFI, Offset, StackGrowsDown, MaxAlign);
 214
 215     // Assign large stack objects first.
 216     for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 217       if (MFI.isDeadObjectIndex(i))
 218         continue;
 219       if (StackProtectorFI == (int)i)
 220         continue;
 221       if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
 222         continue;
 223
 224       switch (MFI.getObjectSSPLayout(i)) {
 225       case MachineFrameInfo::SSPLK_None:
 226         continue;
 227       case MachineFrameInfo::SSPLK_SmallArray:
 228         SmallArrayObjs.insert(i);
 229         continue;
 230       case MachineFrameInfo::SSPLK_AddrOf:
 231         AddrOfObjs.insert(i);
 232         continue;
 233       case MachineFrameInfo::SSPLK_LargeArray:
 234         LargeArrayObjs.insert(i);
 235         continue;
 236       }
 237       llvm_unreachable("Unexpected SSPLayoutKind.");
 238     }
 239
 240     AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 241                           Offset, MaxAlign);
 242     AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 243                           Offset, MaxAlign);
 244     AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
 245                           Offset, MaxAlign);
 246   }
 247
 248   // Then assign frame offsets to stack objects that are not used to spill
 249   // callee saved registers.
 250   for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 251     if (MFI.isDeadObjectIndex(i))
 252       continue;
 253     if (MFI.getStackProtectorIndex() == (int)i)
 254       continue;
 255     if (ProtectedObjs.count(i))
 256       continue;
 257     if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
 258       continue;
 259
 260     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 261   }
 262
 263   // Remember how big this blob of stack space is
 264   MFI.setLocalFrameSize(Offset);
 265   MFI.setLocalFrameMaxAlign(MaxAlign);
 266 }
 267
 268 static inline bool
 269 lookupCandidateBaseReg(unsigned BaseReg,
 270                        int64_t BaseOffset,
 271                        int64_t FrameSizeAdjust,
 272                        int64_t LocalFrameOffset,
 273                        const MachineInstr &MI,
 274                        const TargetRegisterInfo *TRI) {
 275   // Check if the relative offset from the where the base register references
 276   // to the target address is in range for the instruction.
 277   int64_t Offset = FrameSizeAdjust + LocalFrameOffset - BaseOffset;
 278   return TRI->isFrameOffsetLegal(&MI, BaseReg, Offset);
 279 }
 280
 281 bool LocalStackSlotPass::insertFrameReferenceRegisters(MachineFunction &Fn) {
 282   // Scan the function's instructions looking for frame index references.
 283   // For each, ask the target if it wants a virtual base register for it
 284   // based on what we can tell it about where the local will end up in the
 285   // stack frame. If it wants one, re-use a suitable one we've previously
 286   // allocated, or if there isn't one that fits the bill, allocate a new one
 287   // and ask the target to create a defining instruction for it.
 288   bool UsedBaseReg = false;
 289
 290   MachineFrameInfo &MFI = Fn.getFrameInfo();
 291   const TargetRegisterInfo *TRI = Fn.getSubtarget().getRegisterInfo();
 292   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 293   bool StackGrowsDown =
 294     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 295
 296   // Collect all of the instructions in the block that reference
 297   // a frame index. Also store the frame index referenced to ease later
 298   // lookup. (For any insn that has more than one FI reference, we arbitrarily
 299   // choose the first one).
 300   SmallVector<FrameRef, 64> FrameReferenceInsns;
 301
 302   unsigned Order = 0;
 303
 304   for (MachineBasicBlock &BB : Fn) {
 305     for (MachineInstr &MI : BB) {
 306       // Debug value, stackmap and patchpoint instructions can't be out of
 307       // range, so they don't need any updates.
 308       if (MI.isDebugInstr() || MI.getOpcode() == TargetOpcode::STATEPOINT ||
 309           MI.getOpcode() == TargetOpcode::STACKMAP ||
 310           MI.getOpcode() == TargetOpcode::PATCHPOINT)
 311         continue;
 312
 313       // For now, allocate the base register(s) within the basic block
 314       // where they're used, and don't try to keep them around outside
 315       // of that. It may be beneficial to try sharing them more broadly
 316       // than that, but the increased register pressure makes that a
 317       // tricky thing to balance. Investigate if re-materializing these
 318       // becomes an issue.
 319       for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
 320         // Consider replacing all frame index operands that reference
 321         // an object allocated in the local block.
 322         if (MI.getOperand(i).isFI()) {
 323           // Don't try this with values not in the local block.
 324           if (!MFI.isObjectPreAllocated(MI.getOperand(i).getIndex()))
 325             break;
 326           int Idx = MI.getOperand(i).getIndex();
 327           int64_t LocalOffset = LocalOffsets[Idx];
 328           if (!TRI->needsFrameBaseReg(&MI, LocalOffset))
 329             break;
 330           FrameReferenceInsns.push_back(FrameRef(&MI, LocalOffset, Idx, Order++));
 331           break;
 332         }
 333       }
 334     }
 335   }
 336
 337   // Sort the frame references by local offset.
 338   // Use frame index as a tie-breaker in case MI's have the same offset.
 339   llvm::sort(FrameReferenceInsns);
 340
 341   MachineBasicBlock *Entry = &Fn.front();
 342
 343   unsigned BaseReg = 0;
 344   int64_t BaseOffset = 0;
 345
 346   // Loop through the frame references and allocate for them as necessary.
 347   for (int ref = 0, e = FrameReferenceInsns.size(); ref < e ; ++ref) {
 348     FrameRef &FR = FrameReferenceInsns[ref];
 349     MachineInstr &MI = *FR.getMachineInstr();
 350     int64_t LocalOffset = FR.getLocalOffset();
 351     int FrameIdx = FR.getFrameIndex();
 352     assert(MFI.isObjectPreAllocated(FrameIdx) &&
 353            "Only pre-allocated locals expected!");
 354
 355     // We need to keep the references to the stack protector slot through frame
 356     // index operands so that it gets resolved by PEI rather than this pass.
 357     // This avoids accesses to the stack protector though virtual base
 358     // registers, and forces PEI to address it using fp/sp/bp.
 359     if (MFI.hasStackProtectorIndex() &&
 360         FrameIdx == MFI.getStackProtectorIndex())
 361       continue;
 362
 363     LLVM_DEBUG(dbgs() << "Considering: " << MI);
 364
 365     unsigned idx = 0;
 366     for (unsigned f = MI.getNumOperands(); idx != f; ++idx) {
 367       if (!MI.getOperand(idx).isFI())
 368         continue;
 369
 370       if (FrameIdx == MI.getOperand(idx).getIndex())
 371         break;
 372     }
 373
 374     assert(idx < MI.getNumOperands() && "Cannot find FI operand");
 375
 376     int64_t Offset = 0;
 377     int64_t FrameSizeAdjust = StackGrowsDown ? MFI.getLocalFrameSize() : 0;
 378
 379     LLVM_DEBUG(dbgs() << "  Replacing FI in: " << MI);
 380
 381     // If we have a suitable base register available, use it; otherwise
 382     // create a new one. Note that any offset encoded in the
 383     // instruction itself will be taken into account by the target,
 384     // so we don't have to adjust for it here when reusing a base
 385     // register.
 386     if (UsedBaseReg &&
 387         lookupCandidateBaseReg(BaseReg, BaseOffset, FrameSizeAdjust,
 388                                LocalOffset, MI, TRI)) {
 389       LLVM_DEBUG(dbgs() << "  Reusing base register " << BaseReg << "\n");
 390       // We found a register to reuse.
 391       Offset = FrameSizeAdjust + LocalOffset - BaseOffset;
 392     } else {
 393       // No previously defined register was in range, so create a new one.
 394       int64_t InstrOffset = TRI->getFrameIndexInstrOffset(&MI, idx);
 395
 396       int64_t PrevBaseOffset = BaseOffset;
 397       BaseOffset = FrameSizeAdjust + LocalOffset + InstrOffset;
 398
 399       // We'd like to avoid creating single-use virtual base registers.
 400       // Because the FrameRefs are in sorted order, and we've already
 401       // processed all FrameRefs before this one, just check whether or not
 402       // the next FrameRef will be able to reuse this new register. If not,
 403       // then don't bother creating it.
 404       if (ref + 1 >= e ||
 405           !lookupCandidateBaseReg(
 406               BaseReg, BaseOffset, FrameSizeAdjust,
 407               FrameReferenceInsns[ref + 1].getLocalOffset(),
 408               *FrameReferenceInsns[ref + 1].getMachineInstr(), TRI)) {
 409         BaseOffset = PrevBaseOffset;
 410         continue;
 411       }
 412
 413       const MachineFunction *MF = MI.getMF();
 414       const TargetRegisterClass *RC = TRI->getPointerRegClass(*MF);
 415       BaseReg = Fn.getRegInfo().createVirtualRegister(RC);
 416
 417       LLVM_DEBUG(dbgs() << "  Materializing base register"
 418                         << " at frame local offset "
 419                         << LocalOffset + InstrOffset);
 420
 421       // Tell the target to insert the instruction to initialize
 422       // the base register.
 423       //            MachineBasicBlock::iterator InsertionPt = Entry->begin();
 424       BaseReg = TRI->materializeFrameBaseRegister(Entry, FrameIdx, InstrOffset);
 425
 426       LLVM_DEBUG(dbgs() << " into " << printReg(BaseReg, TRI) << '\n');
 427
 428       // The base register already includes any offset specified
 429       // by the instruction, so account for that so it doesn't get
 430       // applied twice.
 431       Offset = -InstrOffset;
 432
 433       ++NumBaseRegisters;
 434       UsedBaseReg = true;
 435     }
 436     assert(BaseReg != 0 && "Unable to allocate virtual base register!");
 437
 438     // Modify the instruction to use the new base register rather
 439     // than the frame index operand.
 440     TRI->resolveFrameIndex(MI, BaseReg, Offset);
 441     LLVM_DEBUG(dbgs() << "Resolved: " << MI);
 442
 443     ++NumReplacements;
 444   }
 445
 446   return UsedBaseReg;
 447 }