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/CodeGen/LocalStackSlotAllocation.h"
  17 #include "llvm/ADT/SetVector.h"
  18 #include "llvm/ADT/SmallSet.h"
  19 #include "llvm/ADT/SmallVector.h"
  20 #include "llvm/ADT/Statistic.h"
  21 #include "llvm/CodeGen/MachineBasicBlock.h"
  22 #include "llvm/CodeGen/MachineFrameInfo.h"
  23 #include "llvm/CodeGen/MachineFunction.h"
  24 #include "llvm/CodeGen/MachineFunctionPass.h"
  25 #include "llvm/CodeGen/MachineInstr.h"
  26 #include "llvm/CodeGen/MachineOperand.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 LocalStackSlotImpl {
  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     bool runOnMachineFunction(MachineFunction &MF);
  92   };
  93
  94   class LocalStackSlotPass : public MachineFunctionPass {
  95   public:
  96     static char ID; // Pass identification, replacement for typeid
  97
  98     explicit LocalStackSlotPass() : MachineFunctionPass(ID) {
  99       initializeLocalStackSlotPassPass(*PassRegistry::getPassRegistry());
 100     }
 101
 102     bool runOnMachineFunction(MachineFunction &MF) override {
 103       return LocalStackSlotImpl().runOnMachineFunction(MF);
 104     }
 105
 106     void getAnalysisUsage(AnalysisUsage &AU) const override {
 107       AU.setPreservesCFG();
 108       MachineFunctionPass::getAnalysisUsage(AU);
 109     }
 110   };
 111
 112 } // end anonymous namespace
 113
 114 PreservedAnalyses
 115 LocalStackSlotAllocationPass::run(MachineFunction &MF,
 116                                   MachineFunctionAnalysisManager &) {
 117   bool Changed = LocalStackSlotImpl().runOnMachineFunction(MF);
 118   if (!Changed)
 119     return PreservedAnalyses::all();
 120   auto PA = getMachineFunctionPassPreservedAnalyses();
 121   PA.preserveSet<CFGAnalyses>();
 122   return PA;
 123 }
 124
 125 char LocalStackSlotPass::ID = 0;
 126
 127 char &llvm::LocalStackSlotAllocationID = LocalStackSlotPass::ID;
 128 INITIALIZE_PASS(LocalStackSlotPass, DEBUG_TYPE,
 129                 "Local Stack Slot Allocation", false, false)
 130
 131 bool LocalStackSlotImpl::runOnMachineFunction(MachineFunction &MF) {
 132   MachineFrameInfo &MFI = MF.getFrameInfo();
 133   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
 134   unsigned LocalObjectCount = MFI.getObjectIndexEnd();
 135
 136   // If the target doesn't want/need this pass, or if there are no locals
 137   // to consider, early exit.
 138   if (LocalObjectCount == 0 || !TRI->requiresVirtualBaseRegisters(MF))
 139     return false;
 140
 141   // Make sure we have enough space to store the local offsets.
 142   LocalOffsets.resize(MFI.getObjectIndexEnd());
 143
 144   // Lay out the local blob.
 145   calculateFrameObjectOffsets(MF);
 146
 147   // Insert virtual base registers to resolve frame index references.
 148   bool UsedBaseRegs = insertFrameReferenceRegisters(MF);
 149
 150   // Tell MFI whether any base registers were allocated. PEI will only
 151   // want to use the local block allocations from this pass if there were any.
 152   // Otherwise, PEI can do a bit better job of getting the alignment right
 153   // without a hole at the start since it knows the alignment of the stack
 154   // at the start of local allocation, and this pass doesn't.
 155   MFI.setUseLocalStackAllocationBlock(UsedBaseRegs);
 156
 157   return true;
 158 }
 159
 160 /// AdjustStackOffset - Helper function used to adjust the stack frame offset.
 161 void LocalStackSlotImpl::AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx,
 162                                            int64_t &Offset, bool StackGrowsDown,
 163                                            Align &MaxAlign) {
 164   // If the stack grows down, add the object size to find the lowest address.
 165   if (StackGrowsDown)
 166     Offset += MFI.getObjectSize(FrameIdx);
 167
 168   Align Alignment = MFI.getObjectAlign(FrameIdx);
 169
 170   // If the alignment of this object is greater than that of the stack, then
 171   // increase the stack alignment to match.
 172   MaxAlign = std::max(MaxAlign, Alignment);
 173
 174   // Adjust to alignment boundary.
 175   Offset = alignTo(Offset, Alignment);
 176
 177   int64_t LocalOffset = StackGrowsDown ? -Offset : Offset;
 178   LLVM_DEBUG(dbgs() << "Allocate FI(" << FrameIdx << ") to local offset "
 179                     << LocalOffset << "\n");
 180   // Keep the offset available for base register allocation
 181   LocalOffsets[FrameIdx] = LocalOffset;
 182   // And tell MFI about it for PEI to use later
 183   MFI.mapLocalFrameObject(FrameIdx, LocalOffset);
 184
 185   if (!StackGrowsDown)
 186     Offset += MFI.getObjectSize(FrameIdx);
 187
 188   ++NumAllocations;
 189 }
 190
 191 /// AssignProtectedObjSet - Helper function to assign large stack objects (i.e.,
 192 /// those required to be close to the Stack Protector) to stack offsets.
 193 void LocalStackSlotImpl::AssignProtectedObjSet(
 194     const StackObjSet &UnassignedObjs, SmallSet<int, 16> &ProtectedObjs,
 195     MachineFrameInfo &MFI, bool StackGrowsDown, int64_t &Offset,
 196     Align &MaxAlign) {
 197   for (int i : UnassignedObjs) {
 198     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 199     ProtectedObjs.insert(i);
 200   }
 201 }
 202
 203 /// calculateFrameObjectOffsets - Calculate actual frame offsets for all of the
 204 /// abstract stack objects.
 205 void LocalStackSlotImpl::calculateFrameObjectOffsets(MachineFunction &Fn) {
 206   // Loop over all of the stack objects, assigning sequential addresses...
 207   MachineFrameInfo &MFI = Fn.getFrameInfo();
 208   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 209   bool StackGrowsDown =
 210     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 211   int64_t Offset = 0;
 212   Align MaxAlign;
 213
 214   // Make sure that the stack protector comes before the local variables on the
 215   // stack.
 216   SmallSet<int, 16> ProtectedObjs;
 217   if (MFI.hasStackProtectorIndex()) {
 218     int StackProtectorFI = MFI.getStackProtectorIndex();
 219
 220     // We need to make sure we didn't pre-allocate the stack protector when
 221     // doing this.
 222     // If we already have a stack protector, this will re-assign it to a slot
 223     // that is **not** covering the protected objects.
 224     assert(!MFI.isObjectPreAllocated(StackProtectorFI) &&
 225            "Stack protector pre-allocated in LocalStackSlotAllocation");
 226
 227     StackObjSet LargeArrayObjs;
 228     StackObjSet SmallArrayObjs;
 229     StackObjSet AddrOfObjs;
 230
 231     // Only place the stack protector in the local stack area if the target
 232     // allows it.
 233     if (TFI.isStackIdSafeForLocalArea(MFI.getStackID(StackProtectorFI)))
 234       AdjustStackOffset(MFI, StackProtectorFI, Offset, StackGrowsDown,
 235                         MaxAlign);
 236
 237     // Assign large stack objects first.
 238     for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 239       if (MFI.isDeadObjectIndex(i))
 240         continue;
 241       if (StackProtectorFI == (int)i)
 242         continue;
 243       if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
 244         continue;
 245
 246       switch (MFI.getObjectSSPLayout(i)) {
 247       case MachineFrameInfo::SSPLK_None:
 248         continue;
 249       case MachineFrameInfo::SSPLK_SmallArray:
 250         SmallArrayObjs.insert(i);
 251         continue;
 252       case MachineFrameInfo::SSPLK_AddrOf:
 253         AddrOfObjs.insert(i);
 254         continue;
 255       case MachineFrameInfo::SSPLK_LargeArray:
 256         LargeArrayObjs.insert(i);
 257         continue;
 258       }
 259       llvm_unreachable("Unexpected SSPLayoutKind.");
 260     }
 261
 262     AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 263                           Offset, MaxAlign);
 264     AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
 265                           Offset, MaxAlign);
 266     AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
 267                           Offset, MaxAlign);
 268   }
 269
 270   // Then assign frame offsets to stack objects that are not used to spill
 271   // callee saved registers.
 272   for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 273     if (MFI.isDeadObjectIndex(i))
 274       continue;
 275     if (MFI.getStackProtectorIndex() == (int)i)
 276       continue;
 277     if (ProtectedObjs.count(i))
 278       continue;
 279     if (!TFI.isStackIdSafeForLocalArea(MFI.getStackID(i)))
 280       continue;
 281
 282     AdjustStackOffset(MFI, i, Offset, StackGrowsDown, MaxAlign);
 283   }
 284
 285   // Remember how big this blob of stack space is
 286   MFI.setLocalFrameSize(Offset);
 287   MFI.setLocalFrameMaxAlign(MaxAlign);
 288 }
 289
 290 static inline bool
 291 lookupCandidateBaseReg(unsigned BaseReg,
 292                        int64_t BaseOffset,
 293                        int64_t FrameSizeAdjust,
 294                        int64_t LocalFrameOffset,
 295                        const MachineInstr &MI,
 296                        const TargetRegisterInfo *TRI) {
 297   // Check if the relative offset from the where the base register references
 298   // to the target address is in range for the instruction.
 299   int64_t Offset = FrameSizeAdjust + LocalFrameOffset - BaseOffset;
 300   return TRI->isFrameOffsetLegal(&MI, BaseReg, Offset);
 301 }
 302
 303 bool LocalStackSlotImpl::insertFrameReferenceRegisters(MachineFunction &Fn) {
 304   // Scan the function's instructions looking for frame index references.
 305   // For each, ask the target if it wants a virtual base register for it
 306   // based on what we can tell it about where the local will end up in the
 307   // stack frame. If it wants one, re-use a suitable one we've previously
 308   // allocated, or if there isn't one that fits the bill, allocate a new one
 309   // and ask the target to create a defining instruction for it.
 310
 311   MachineFrameInfo &MFI = Fn.getFrameInfo();
 312   const TargetRegisterInfo *TRI = Fn.getSubtarget().getRegisterInfo();
 313   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 314   bool StackGrowsDown =
 315     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 316
 317   // Collect all of the instructions in the block that reference
 318   // a frame index. Also store the frame index referenced to ease later
 319   // lookup. (For any insn that has more than one FI reference, we arbitrarily
 320   // choose the first one).
 321   SmallVector<FrameRef, 64> FrameReferenceInsns;
 322
 323   unsigned Order = 0;
 324
 325   for (MachineBasicBlock &BB : Fn) {
 326     for (MachineInstr &MI : BB) {
 327       // Debug value, stackmap and patchpoint instructions can't be out of
 328       // range, so they don't need any updates.
 329       if (MI.isDebugInstr() || MI.getOpcode() == TargetOpcode::STATEPOINT ||
 330           MI.getOpcode() == TargetOpcode::STACKMAP ||
 331           MI.getOpcode() == TargetOpcode::PATCHPOINT)
 332         continue;
 333
 334       // For now, allocate the base register(s) within the basic block
 335       // where they're used, and don't try to keep them around outside
 336       // of that. It may be beneficial to try sharing them more broadly
 337       // than that, but the increased register pressure makes that a
 338       // tricky thing to balance. Investigate if re-materializing these
 339       // becomes an issue.
 340       for (const MachineOperand &MO : MI.operands()) {
 341         // Consider replacing all frame index operands that reference
 342         // an object allocated in the local block.
 343         if (MO.isFI()) {
 344           // Don't try this with values not in the local block.
 345           if (!MFI.isObjectPreAllocated(MO.getIndex()))
 346             break;
 347           int Idx = MO.getIndex();
 348           int64_t LocalOffset = LocalOffsets[Idx];
 349           if (!TRI->needsFrameBaseReg(&MI, LocalOffset))
 350             break;
 351           FrameReferenceInsns.push_back(FrameRef(&MI, LocalOffset, Idx, Order++));
 352           break;
 353         }
 354       }
 355     }
 356   }
 357
 358   // Sort the frame references by local offset.
 359   // Use frame index as a tie-breaker in case MI's have the same offset.
 360   llvm::sort(FrameReferenceInsns);
 361
 362   MachineBasicBlock *Entry = &Fn.front();
 363
 364   Register BaseReg;
 365   int64_t BaseOffset = 0;
 366
 367   // Loop through the frame references and allocate for them as necessary.
 368   for (int ref = 0, e = FrameReferenceInsns.size(); ref < e ; ++ref) {
 369     FrameRef &FR = FrameReferenceInsns[ref];
 370     MachineInstr &MI = *FR.getMachineInstr();
 371     int64_t LocalOffset = FR.getLocalOffset();
 372     int FrameIdx = FR.getFrameIndex();
 373     assert(MFI.isObjectPreAllocated(FrameIdx) &&
 374            "Only pre-allocated locals expected!");
 375
 376     // We need to keep the references to the stack protector slot through frame
 377     // index operands so that it gets resolved by PEI rather than this pass.
 378     // This avoids accesses to the stack protector though virtual base
 379     // registers, and forces PEI to address it using fp/sp/bp.
 380     if (MFI.hasStackProtectorIndex() &&
 381         FrameIdx == MFI.getStackProtectorIndex())
 382       continue;
 383
 384     LLVM_DEBUG(dbgs() << "Considering: " << MI);
 385
 386     unsigned idx = 0;
 387     for (unsigned f = MI.getNumOperands(); idx != f; ++idx) {
 388       if (!MI.getOperand(idx).isFI())
 389         continue;
 390
 391       if (FrameIdx == MI.getOperand(idx).getIndex())
 392         break;
 393     }
 394
 395     assert(idx < MI.getNumOperands() && "Cannot find FI operand");
 396
 397     int64_t Offset = 0;
 398     int64_t FrameSizeAdjust = StackGrowsDown ? MFI.getLocalFrameSize() : 0;
 399
 400     LLVM_DEBUG(dbgs() << "  Replacing FI in: " << MI);
 401
 402     // If we have a suitable base register available, use it; otherwise
 403     // create a new one. Note that any offset encoded in the
 404     // instruction itself will be taken into account by the target,
 405     // so we don't have to adjust for it here when reusing a base
 406     // register.
 407     if (BaseReg.isValid() &&
 408         lookupCandidateBaseReg(BaseReg, BaseOffset, FrameSizeAdjust,
 409                                LocalOffset, MI, TRI)) {
 410       LLVM_DEBUG(dbgs() << "  Reusing base register " << BaseReg << "\n");
 411       // We found a register to reuse.
 412       Offset = FrameSizeAdjust + LocalOffset - BaseOffset;
 413     } else {
 414       // No previously defined register was in range, so create a new one.
 415       int64_t InstrOffset = TRI->getFrameIndexInstrOffset(&MI, idx);
 416
 417       int64_t CandBaseOffset = FrameSizeAdjust + LocalOffset + InstrOffset;
 418
 419       // We'd like to avoid creating single-use virtual base registers.
 420       // Because the FrameRefs are in sorted order, and we've already
 421       // processed all FrameRefs before this one, just check whether or not
 422       // the next FrameRef will be able to reuse this new register. If not,
 423       // then don't bother creating it.
 424       if (ref + 1 >= e ||
 425           !lookupCandidateBaseReg(
 426               BaseReg, CandBaseOffset, FrameSizeAdjust,
 427               FrameReferenceInsns[ref + 1].getLocalOffset(),
 428               *FrameReferenceInsns[ref + 1].getMachineInstr(), TRI))
 429         continue;
 430
 431       // Save the base offset.
 432       BaseOffset = CandBaseOffset;
 433
 434       // Tell the target to insert the instruction to initialize
 435       // the base register.
 436       //            MachineBasicBlock::iterator InsertionPt = Entry->begin();
 437       BaseReg = TRI->materializeFrameBaseRegister(Entry, FrameIdx, InstrOffset);
 438
 439       LLVM_DEBUG(dbgs() << "  Materialized base register at frame local offset "
 440                         << LocalOffset + InstrOffset
 441                         << " into " << printReg(BaseReg, TRI) << '\n');
 442
 443       // The base register already includes any offset specified
 444       // by the instruction, so account for that so it doesn't get
 445       // applied twice.
 446       Offset = -InstrOffset;
 447
 448       ++NumBaseRegisters;
 449     }
 450     assert(BaseReg && "Unable to allocate virtual base register!");
 451
 452     // Modify the instruction to use the new base register rather
 453     // than the frame index operand.
 454     TRI->resolveFrameIndex(MI, BaseReg, Offset);
 455     LLVM_DEBUG(dbgs() << "Resolved: " << MI);
 456
 457     ++NumReplacements;
 458   }
 459
 460   return BaseReg.isValid();
 461 }