| 1 | //=- AArch64MachineFunctionInfo.h - AArch64 machine function info -*- C++ -*-=// |
| 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 file declares AArch64-specific per-machine-function information. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #ifndef LLVM_LIB_TARGET_AARCH64_AARCH64MACHINEFUNCTIONINFO_H |
| 14 | #define LLVM_LIB_TARGET_AARCH64_AARCH64MACHINEFUNCTIONINFO_H |
| 15 | |
| 16 | #include "AArch64SMEAttributes.h" |
| 17 | #include "AArch64Subtarget.h" |
| 18 | #include "llvm/ADT/ArrayRef.h" |
| 19 | #include "llvm/ADT/SmallPtrSet.h" |
| 20 | #include "llvm/ADT/SmallVector.h" |
| 21 | #include "llvm/CodeGen/CallingConvLower.h" |
| 22 | #include "llvm/CodeGen/MIRYamlMapping.h" |
| 23 | #include "llvm/CodeGen/MachineFrameInfo.h" |
| 24 | #include "llvm/CodeGen/MachineFunction.h" |
| 25 | #include "llvm/IR/Function.h" |
| 26 | #include "llvm/MC/MCLinkerOptimizationHint.h" |
| 27 | #include "llvm/MC/MCSymbol.h" |
| 28 | #include <cassert> |
| 29 | #include <optional> |
| 30 | |
| 31 | namespace llvm { |
| 32 | |
| 33 | namespace yaml { |
| 34 | struct AArch64FunctionInfo; |
| 35 | } // end namespace yaml |
| 36 | |
| 37 | class AArch64Subtarget; |
| 38 | class MachineInstr; |
| 39 | |
| 40 | struct TPIDR2Object { |
| 41 | int FrameIndex = std::numeric_limits<int>::max(); |
| 42 | unsigned Uses = 0; |
| 43 | }; |
| 44 | |
| 45 | /// Condition of signing the return address in a function. |
| 46 | /// |
| 47 | /// Corresponds to possible values of "sign-return-address" function attribute. |
| 48 | enum class SignReturnAddress { |
| 49 | None, |
| 50 | NonLeaf, |
| 51 | All, |
| 52 | }; |
| 53 | |
| 54 | /// AArch64FunctionInfo - This class is derived from MachineFunctionInfo and |
| 55 | /// contains private AArch64-specific information for each MachineFunction. |
| 56 | class AArch64FunctionInfo final : public MachineFunctionInfo { |
| 57 | /// Number of bytes of arguments this function has on the stack. If the callee |
| 58 | /// is expected to restore the argument stack this should be a multiple of 16, |
| 59 | /// all usable during a tail call. |
| 60 | /// |
| 61 | /// The alternative would forbid tail call optimisation in some cases: if we |
| 62 | /// want to transfer control from a function with 8-bytes of stack-argument |
| 63 | /// space to a function with 16-bytes then misalignment of this value would |
| 64 | /// make a stack adjustment necessary, which could not be undone by the |
| 65 | /// callee. |
| 66 | unsigned BytesInStackArgArea = 0; |
| 67 | |
| 68 | /// The number of bytes to restore to deallocate space for incoming |
| 69 | /// arguments. Canonically 0 in the C calling convention, but non-zero when |
| 70 | /// callee is expected to pop the args. |
| 71 | unsigned ArgumentStackToRestore = 0; |
| 72 | |
| 73 | /// Space just below incoming stack pointer reserved for arguments being |
| 74 | /// passed on the stack during a tail call. This will be the difference |
| 75 | /// between the largest tail call argument space needed in this function and |
| 76 | /// what's already available by reusing space of incoming arguments. |
| 77 | unsigned TailCallReservedStack = 0; |
| 78 | |
| 79 | /// HasStackFrame - True if this function has a stack frame. Set by |
| 80 | /// determineCalleeSaves(). |
| 81 | bool HasStackFrame = false; |
| 82 | |
| 83 | /// Amount of stack frame size, not including callee-saved registers. |
| 84 | uint64_t LocalStackSize = 0; |
| 85 | |
| 86 | /// Amount of stack frame size used for saving callee-saved registers. |
| 87 | unsigned CalleeSavedStackSize = 0; |
| 88 | unsigned ZPRCalleeSavedStackSize = 0; |
| 89 | unsigned PPRCalleeSavedStackSize = 0; |
| 90 | bool HasCalleeSavedStackSize = false; |
| 91 | bool HasSVECalleeSavedStackSize = false; |
| 92 | |
| 93 | /// Number of TLS accesses using the special (combinable) |
| 94 | /// _TLS_MODULE_BASE_ symbol. |
| 95 | unsigned NumLocalDynamicTLSAccesses = 0; |
| 96 | |
| 97 | /// FrameIndex for start of varargs area for arguments passed on the |
| 98 | /// stack. |
| 99 | int VarArgsStackIndex = 0; |
| 100 | |
| 101 | /// Offset of start of varargs area for arguments passed on the stack. |
| 102 | unsigned VarArgsStackOffset = 0; |
| 103 | |
| 104 | /// FrameIndex for start of varargs area for arguments passed in |
| 105 | /// general purpose registers. |
| 106 | int VarArgsGPRIndex = 0; |
| 107 | |
| 108 | /// Size of the varargs area for arguments passed in general purpose |
| 109 | /// registers. |
| 110 | unsigned VarArgsGPRSize = 0; |
| 111 | |
| 112 | /// FrameIndex for start of varargs area for arguments passed in |
| 113 | /// floating-point registers. |
| 114 | int VarArgsFPRIndex = 0; |
| 115 | |
| 116 | /// Size of the varargs area for arguments passed in floating-point |
| 117 | /// registers. |
| 118 | unsigned VarArgsFPRSize = 0; |
| 119 | |
| 120 | /// The stack slots used to add space between FPR and GPR accesses when using |
| 121 | /// hazard padding. StackHazardCSRSlotIndex is added between GPR and FPR CSRs. |
| 122 | /// StackHazardSlotIndex is added between (sorted) stack objects. |
| 123 | int StackHazardSlotIndex = std::numeric_limits<int>::max(); |
| 124 | int StackHazardCSRSlotIndex = std::numeric_limits<int>::max(); |
| 125 | |
| 126 | /// True if this function has a subset of CSRs that is handled explicitly via |
| 127 | /// copies. |
| 128 | bool IsSplitCSR = false; |
| 129 | |
| 130 | /// True when the stack gets realigned dynamically because the size of stack |
| 131 | /// frame is unknown at compile time. e.g., in case of VLAs. |
| 132 | bool StackRealigned = false; |
| 133 | |
| 134 | /// True when the callee-save stack area has unused gaps that may be used for |
| 135 | /// other stack allocations. |
| 136 | bool CalleeSaveStackHasFreeSpace = false; |
| 137 | |
| 138 | /// SRetReturnReg - sret lowering includes returning the value of the |
| 139 | /// returned struct in a register. This field holds the virtual register into |
| 140 | /// which the sret argument is passed. |
| 141 | Register SRetReturnReg; |
| 142 | |
| 143 | /// SVE stack size (for predicates and data vectors) are maintained here |
| 144 | /// rather than in FrameInfo, as the placement and Stack IDs are target |
| 145 | /// specific. |
| 146 | uint64_t StackSizeZPR = 0; |
| 147 | uint64_t StackSizePPR = 0; |
| 148 | |
| 149 | /// Are SVE objects (vectors and predicates) split into separate regions on |
| 150 | /// the stack. |
| 151 | bool SplitSVEObjects = false; |
| 152 | |
| 153 | /// HasCalculatedStackSizeSVE indicates whether StackSizeZPR/PPR is valid. |
| 154 | bool HasCalculatedStackSizeSVE = false; |
| 155 | |
| 156 | /// Has a value when it is known whether or not the function uses a |
| 157 | /// redzone, and no value otherwise. |
| 158 | /// Initialized during frame lowering, unless the function has the noredzone |
| 159 | /// attribute, in which case it is set to false at construction. |
| 160 | std::optional<bool> HasRedZone; |
| 161 | |
| 162 | /// ForwardedMustTailRegParms - A list of virtual and physical registers |
| 163 | /// that must be forwarded to every musttail call. |
| 164 | SmallVector<ForwardedRegister, 1> ForwardedMustTailRegParms; |
| 165 | |
| 166 | /// FrameIndex for the tagged base pointer. |
| 167 | std::optional<int> TaggedBasePointerIndex; |
| 168 | |
| 169 | /// Offset from SP-at-entry to the tagged base pointer. |
| 170 | /// Tagged base pointer is set up to point to the first (lowest address) |
| 171 | /// tagged stack slot. |
| 172 | unsigned TaggedBasePointerOffset; |
| 173 | |
| 174 | /// OutliningStyle denotes, if a function was outined, how it was outlined, |
| 175 | /// e.g. Tail Call, Thunk, or Function if none apply. |
| 176 | std::optional<std::string> OutliningStyle; |
| 177 | |
| 178 | // Offset from SP-after-callee-saved-spills (i.e. SP-at-entry minus |
| 179 | // CalleeSavedStackSize) to the address of the frame record. |
| 180 | int CalleeSaveBaseToFrameRecordOffset = 0; |
| 181 | |
| 182 | /// SignCondition controls when PAC-RET protection should be used. |
| 183 | SignReturnAddress SignCondition = SignReturnAddress::None; |
| 184 | |
| 185 | /// SignWithBKey modifies the default PAC-RET mode to signing with the B key. |
| 186 | bool SignWithBKey = false; |
| 187 | |
| 188 | /// HasELFSignedGOT is true if the target binary format is ELF and the IR |
| 189 | /// module containing the corresponding function has "ptrauth-elf-got" flag |
| 190 | /// set to 1. |
| 191 | bool HasELFSignedGOT = false; |
| 192 | |
| 193 | /// SigningInstrOffset captures the offset of the PAC-RET signing instruction |
| 194 | /// within the prologue, so it can be re-used for authentication in the |
| 195 | /// epilogue when using PC as a second salt (FEAT_PAuth_LR) |
| 196 | MCSymbol *SignInstrLabel = nullptr; |
| 197 | |
| 198 | /// BranchTargetEnforcement enables placing BTI instructions at potential |
| 199 | /// indirect branch destinations. |
| 200 | bool BranchTargetEnforcement = false; |
| 201 | |
| 202 | /// Indicates that SP signing should be diversified with PC as-per PAuthLR. |
| 203 | /// This is set by -mbranch-protection and will emit NOP instructions unless |
| 204 | /// the subtarget feature +pauthlr is also used (in which case non-NOP |
| 205 | /// instructions are emitted). |
| 206 | bool BranchProtectionPAuthLR = false; |
| 207 | |
| 208 | /// Whether this function has an extended frame record [Ctx, FP, LR]. If so, |
| 209 | /// bit 60 of the in-memory FP will be 1 to enable other tools to detect the |
| 210 | /// extended record. |
| 211 | bool HasSwiftAsyncContext = false; |
| 212 | |
| 213 | /// The stack slot where the Swift asynchronous context is stored. |
| 214 | int SwiftAsyncContextFrameIdx = std::numeric_limits<int>::max(); |
| 215 | |
| 216 | bool IsMTETagged = false; |
| 217 | |
| 218 | /// The function has Scalable Vector or Scalable Predicate register argument |
| 219 | /// or return type |
| 220 | bool IsSVECC = false; |
| 221 | |
| 222 | /// Whether this function changes streaming mode within the function. |
| 223 | bool HasStreamingModeChanges = false; |
| 224 | |
| 225 | /// True if the function need unwind information. |
| 226 | mutable std::optional<bool> NeedsDwarfUnwindInfo; |
| 227 | |
| 228 | /// True if the function need asynchronous unwind information. |
| 229 | mutable std::optional<bool> NeedsAsyncDwarfUnwindInfo; |
| 230 | |
| 231 | int64_t StackProbeSize = 0; |
| 232 | |
| 233 | // Holds a register containing pstate.sm. This is set |
| 234 | // on function entry to record the initial pstate of a function. |
| 235 | Register PStateSMReg = MCRegister::NoRegister; |
| 236 | |
| 237 | // Has the PNReg used to build PTRUE instruction. |
| 238 | // The PTRUE is used for the LD/ST of ZReg pairs in save and restore. |
| 239 | unsigned PredicateRegForFillSpill = 0; |
| 240 | |
| 241 | // Holds the SME function attributes (streaming mode, ZA/ZT0 state). |
| 242 | SMEAttrs SMEFnAttrs; |
| 243 | |
| 244 | // Holds the TPIDR2 block if allocated early (for Windows/stack probes |
| 245 | // support). |
| 246 | Register EarlyAllocSMESaveBuffer = AArch64::NoRegister; |
| 247 | |
| 248 | // Holds the spill slot for ZT0. |
| 249 | int ZT0SpillSlotIndex = std::numeric_limits<int>::max(); |
| 250 | |
| 251 | // Note: The following properties are only used for the old SME ABI lowering: |
| 252 | /// The frame-index for the TPIDR2 object used for lazy saves. |
| 253 | TPIDR2Object TPIDR2; |
| 254 | // Holds a pointer to a buffer that is large enough to represent |
| 255 | // all SME ZA state and any additional state required by the |
| 256 | // __arm_sme_save/restore support routines. |
| 257 | Register SMESaveBufferAddr = MCRegister::NoRegister; |
| 258 | // true if SMESaveBufferAddr is used. |
| 259 | bool SMESaveBufferUsed = false; |
| 260 | |
| 261 | public: |
| 262 | AArch64FunctionInfo(const Function &F, const AArch64Subtarget *STI); |
| 263 | |
| 264 | MachineFunctionInfo * |
| 265 | clone(BumpPtrAllocator &Allocator, MachineFunction &DestMF, |
| 266 | const DenseMap<MachineBasicBlock *, MachineBasicBlock *> &Src2DstMBB) |
| 267 | const override; |
| 268 | |
| 269 | void setEarlyAllocSMESaveBuffer(Register Ptr) { |
| 270 | EarlyAllocSMESaveBuffer = Ptr; |
| 271 | } |
| 272 | |
| 273 | Register getEarlyAllocSMESaveBuffer() const { |
| 274 | return EarlyAllocSMESaveBuffer; |
| 275 | } |
| 276 | |
| 277 | void setZT0SpillSlotIndex(int FI) { ZT0SpillSlotIndex = FI; } |
| 278 | int getZT0SpillSlotIndex() const { |
| 279 | assert(hasZT0SpillSlotIndex() && "ZT0 spill slot index not set!" ); |
| 280 | return ZT0SpillSlotIndex; |
| 281 | } |
| 282 | bool hasZT0SpillSlotIndex() const { |
| 283 | return ZT0SpillSlotIndex != std::numeric_limits<int>::max(); |
| 284 | } |
| 285 | |
| 286 | // Old SME ABI lowering state getters/setters: |
| 287 | Register getSMESaveBufferAddr() const { return SMESaveBufferAddr; }; |
| 288 | void setSMESaveBufferAddr(Register Reg) { SMESaveBufferAddr = Reg; }; |
| 289 | unsigned isSMESaveBufferUsed() const { return SMESaveBufferUsed; }; |
| 290 | void setSMESaveBufferUsed(bool Used = true) { SMESaveBufferUsed = Used; }; |
| 291 | TPIDR2Object &getTPIDR2Obj() { return TPIDR2; } |
| 292 | |
| 293 | void setPredicateRegForFillSpill(unsigned Reg) { |
| 294 | PredicateRegForFillSpill = Reg; |
| 295 | } |
| 296 | unsigned getPredicateRegForFillSpill() const { |
| 297 | return PredicateRegForFillSpill; |
| 298 | } |
| 299 | |
| 300 | Register getPStateSMReg() const { return PStateSMReg; }; |
| 301 | void setPStateSMReg(Register Reg) { PStateSMReg = Reg; }; |
| 302 | |
| 303 | bool isSVECC() const { return IsSVECC; }; |
| 304 | void setIsSVECC(bool s) { IsSVECC = s; }; |
| 305 | |
| 306 | void initializeBaseYamlFields(const yaml::AArch64FunctionInfo &YamlMFI); |
| 307 | |
| 308 | unsigned getBytesInStackArgArea() const { return BytesInStackArgArea; } |
| 309 | void setBytesInStackArgArea(unsigned bytes) { BytesInStackArgArea = bytes; } |
| 310 | |
| 311 | unsigned getArgumentStackToRestore() const { return ArgumentStackToRestore; } |
| 312 | void setArgumentStackToRestore(unsigned bytes) { |
| 313 | ArgumentStackToRestore = bytes; |
| 314 | } |
| 315 | |
| 316 | unsigned getTailCallReservedStack() const { return TailCallReservedStack; } |
| 317 | void setTailCallReservedStack(unsigned bytes) { |
| 318 | TailCallReservedStack = bytes; |
| 319 | } |
| 320 | |
| 321 | void setStackSizeSVE(uint64_t ZPR, uint64_t PPR) { |
| 322 | assert(isAligned(Align(16), ZPR) && isAligned(Align(16), PPR) && |
| 323 | "expected SVE stack sizes to be aligned to 16-bytes" ); |
| 324 | StackSizeZPR = ZPR; |
| 325 | StackSizePPR = PPR; |
| 326 | HasCalculatedStackSizeSVE = true; |
| 327 | } |
| 328 | |
| 329 | uint64_t getStackSizeZPR() const { |
| 330 | assert(hasCalculatedStackSizeSVE()); |
| 331 | return StackSizeZPR; |
| 332 | } |
| 333 | uint64_t getStackSizePPR() const { |
| 334 | assert(hasCalculatedStackSizeSVE()); |
| 335 | return StackSizePPR; |
| 336 | } |
| 337 | |
| 338 | bool hasCalculatedStackSizeSVE() const { return HasCalculatedStackSizeSVE; } |
| 339 | |
| 340 | bool hasSVEStackSize() const { |
| 341 | return getStackSizeZPR() > 0 || getStackSizePPR() > 0; |
| 342 | } |
| 343 | |
| 344 | bool hasStackFrame() const { return HasStackFrame; } |
| 345 | void setHasStackFrame(bool s) { HasStackFrame = s; } |
| 346 | |
| 347 | bool isStackRealigned() const { return StackRealigned; } |
| 348 | void setStackRealigned(bool s) { StackRealigned = s; } |
| 349 | bool hasCalleeSaveStackFreeSpace() const { |
| 350 | return CalleeSaveStackHasFreeSpace; |
| 351 | } |
| 352 | void setCalleeSaveStackHasFreeSpace(bool s) { |
| 353 | CalleeSaveStackHasFreeSpace = s; |
| 354 | } |
| 355 | bool isSplitCSR() const { return IsSplitCSR; } |
| 356 | void setIsSplitCSR(bool s) { IsSplitCSR = s; } |
| 357 | |
| 358 | void setLocalStackSize(uint64_t Size) { LocalStackSize = Size; } |
| 359 | uint64_t getLocalStackSize() const { return LocalStackSize; } |
| 360 | |
| 361 | void setOutliningStyle(const std::string &Style) { OutliningStyle = Style; } |
| 362 | std::optional<std::string> getOutliningStyle() const { |
| 363 | return OutliningStyle; |
| 364 | } |
| 365 | |
| 366 | void setCalleeSavedStackSize(unsigned Size) { |
| 367 | CalleeSavedStackSize = Size; |
| 368 | HasCalleeSavedStackSize = true; |
| 369 | } |
| 370 | |
| 371 | // When CalleeSavedStackSize has not been set (for example when |
| 372 | // some MachineIR pass is run in isolation), then recalculate |
| 373 | // the CalleeSavedStackSize directly from the CalleeSavedInfo. |
| 374 | // Note: This information can only be recalculated after PEI |
| 375 | // has assigned offsets to the callee save objects. |
| 376 | unsigned getCalleeSavedStackSize(const MachineFrameInfo &MFI) const { |
| 377 | bool ValidateCalleeSavedStackSize = false; |
| 378 | |
| 379 | #ifndef NDEBUG |
| 380 | // Make sure the calculated size derived from the CalleeSavedInfo |
| 381 | // equals the cached size that was calculated elsewhere (e.g. in |
| 382 | // determineCalleeSaves). |
| 383 | ValidateCalleeSavedStackSize = HasCalleeSavedStackSize; |
| 384 | #endif |
| 385 | |
| 386 | if (!HasCalleeSavedStackSize || ValidateCalleeSavedStackSize) { |
| 387 | assert(MFI.isCalleeSavedInfoValid() && "CalleeSavedInfo not calculated" ); |
| 388 | if (MFI.getCalleeSavedInfo().empty()) |
| 389 | return 0; |
| 390 | |
| 391 | int64_t MinOffset = std::numeric_limits<int64_t>::max(); |
| 392 | int64_t MaxOffset = std::numeric_limits<int64_t>::min(); |
| 393 | for (const auto &Info : MFI.getCalleeSavedInfo()) { |
| 394 | int FrameIdx = Info.getFrameIdx(); |
| 395 | if (MFI.getStackID(ObjectIdx: FrameIdx) != TargetStackID::Default) |
| 396 | continue; |
| 397 | int64_t Offset = MFI.getObjectOffset(ObjectIdx: FrameIdx); |
| 398 | int64_t ObjSize = MFI.getObjectSize(ObjectIdx: FrameIdx); |
| 399 | MinOffset = std::min<int64_t>(a: Offset, b: MinOffset); |
| 400 | MaxOffset = std::max<int64_t>(a: Offset + ObjSize, b: MaxOffset); |
| 401 | } |
| 402 | |
| 403 | if (SwiftAsyncContextFrameIdx != std::numeric_limits<int>::max()) { |
| 404 | int64_t Offset = MFI.getObjectOffset(ObjectIdx: getSwiftAsyncContextFrameIdx()); |
| 405 | int64_t ObjSize = MFI.getObjectSize(ObjectIdx: getSwiftAsyncContextFrameIdx()); |
| 406 | MinOffset = std::min<int64_t>(a: Offset, b: MinOffset); |
| 407 | MaxOffset = std::max<int64_t>(a: Offset + ObjSize, b: MaxOffset); |
| 408 | } |
| 409 | |
| 410 | if (StackHazardCSRSlotIndex != std::numeric_limits<int>::max()) { |
| 411 | int64_t Offset = MFI.getObjectOffset(ObjectIdx: StackHazardCSRSlotIndex); |
| 412 | int64_t ObjSize = MFI.getObjectSize(ObjectIdx: StackHazardCSRSlotIndex); |
| 413 | MinOffset = std::min<int64_t>(a: Offset, b: MinOffset); |
| 414 | MaxOffset = std::max<int64_t>(a: Offset + ObjSize, b: MaxOffset); |
| 415 | } |
| 416 | |
| 417 | unsigned Size = alignTo(Value: MaxOffset - MinOffset, Align: 16); |
| 418 | assert((!HasCalleeSavedStackSize || getCalleeSavedStackSize() == Size) && |
| 419 | "Invalid size calculated for callee saves" ); |
| 420 | return Size; |
| 421 | } |
| 422 | |
| 423 | return getCalleeSavedStackSize(); |
| 424 | } |
| 425 | |
| 426 | unsigned getCalleeSavedStackSize() const { |
| 427 | assert(HasCalleeSavedStackSize && |
| 428 | "CalleeSavedStackSize has not been calculated" ); |
| 429 | return CalleeSavedStackSize; |
| 430 | } |
| 431 | |
| 432 | // Saves the CalleeSavedStackSize for SVE vectors in 'scalable bytes' |
| 433 | void setSVECalleeSavedStackSize(unsigned ZPR, unsigned PPR) { |
| 434 | assert(isAligned(Align(16), ZPR) && isAligned(Align(16), PPR) && |
| 435 | "expected SVE callee-save sizes to be aligned to 16-bytes" ); |
| 436 | ZPRCalleeSavedStackSize = ZPR; |
| 437 | PPRCalleeSavedStackSize = PPR; |
| 438 | HasSVECalleeSavedStackSize = true; |
| 439 | } |
| 440 | unsigned getZPRCalleeSavedStackSize() const { |
| 441 | assert(HasSVECalleeSavedStackSize && |
| 442 | "ZPRCalleeSavedStackSize has not been calculated" ); |
| 443 | return ZPRCalleeSavedStackSize; |
| 444 | } |
| 445 | unsigned getPPRCalleeSavedStackSize() const { |
| 446 | assert(HasSVECalleeSavedStackSize && |
| 447 | "PPRCalleeSavedStackSize has not been calculated" ); |
| 448 | return PPRCalleeSavedStackSize; |
| 449 | } |
| 450 | |
| 451 | unsigned getSVECalleeSavedStackSize() const { |
| 452 | assert(!hasSplitSVEObjects() && |
| 453 | "ZPRs and PPRs are split. Use get[ZPR|PPR]CalleeSavedStackSize()" ); |
| 454 | return getZPRCalleeSavedStackSize() + getPPRCalleeSavedStackSize(); |
| 455 | } |
| 456 | |
| 457 | void incNumLocalDynamicTLSAccesses() { ++NumLocalDynamicTLSAccesses; } |
| 458 | unsigned getNumLocalDynamicTLSAccesses() const { |
| 459 | return NumLocalDynamicTLSAccesses; |
| 460 | } |
| 461 | |
| 462 | bool isStackHazardIncludedInCalleeSaveArea() const { |
| 463 | return hasStackHazardSlotIndex() && !hasSplitSVEObjects(); |
| 464 | } |
| 465 | |
| 466 | std::optional<bool> hasRedZone() const { return HasRedZone; } |
| 467 | void setHasRedZone(bool s) { HasRedZone = s; } |
| 468 | |
| 469 | int getVarArgsStackIndex() const { return VarArgsStackIndex; } |
| 470 | void setVarArgsStackIndex(int Index) { VarArgsStackIndex = Index; } |
| 471 | |
| 472 | unsigned getVarArgsStackOffset() const { return VarArgsStackOffset; } |
| 473 | void setVarArgsStackOffset(unsigned Offset) { VarArgsStackOffset = Offset; } |
| 474 | |
| 475 | int getVarArgsGPRIndex() const { return VarArgsGPRIndex; } |
| 476 | void setVarArgsGPRIndex(int Index) { VarArgsGPRIndex = Index; } |
| 477 | |
| 478 | unsigned getVarArgsGPRSize() const { return VarArgsGPRSize; } |
| 479 | void setVarArgsGPRSize(unsigned Size) { VarArgsGPRSize = Size; } |
| 480 | |
| 481 | int getVarArgsFPRIndex() const { return VarArgsFPRIndex; } |
| 482 | void setVarArgsFPRIndex(int Index) { VarArgsFPRIndex = Index; } |
| 483 | |
| 484 | unsigned getVarArgsFPRSize() const { return VarArgsFPRSize; } |
| 485 | void setVarArgsFPRSize(unsigned Size) { VarArgsFPRSize = Size; } |
| 486 | |
| 487 | bool hasStackHazardSlotIndex() const { |
| 488 | return StackHazardSlotIndex != std::numeric_limits<int>::max(); |
| 489 | } |
| 490 | int getStackHazardSlotIndex() const { return StackHazardSlotIndex; } |
| 491 | void setStackHazardSlotIndex(int Index) { |
| 492 | assert(StackHazardSlotIndex == std::numeric_limits<int>::max()); |
| 493 | StackHazardSlotIndex = Index; |
| 494 | } |
| 495 | int getStackHazardCSRSlotIndex() const { return StackHazardCSRSlotIndex; } |
| 496 | void setStackHazardCSRSlotIndex(int Index) { |
| 497 | assert(StackHazardCSRSlotIndex == std::numeric_limits<int>::max()); |
| 498 | StackHazardCSRSlotIndex = Index; |
| 499 | } |
| 500 | |
| 501 | bool hasSplitSVEObjects() const { return SplitSVEObjects; } |
| 502 | void setSplitSVEObjects(bool s) { SplitSVEObjects = s; } |
| 503 | |
| 504 | bool hasSVE_AAPCS(const MachineFunction &MF) const { |
| 505 | return hasSplitSVEObjects() || isSVECC() || |
| 506 | MF.getFunction().getCallingConv() == |
| 507 | CallingConv::AArch64_SVE_VectorCall; |
| 508 | } |
| 509 | |
| 510 | SMEAttrs getSMEFnAttrs() const { return SMEFnAttrs; } |
| 511 | |
| 512 | unsigned getSRetReturnReg() const { return SRetReturnReg; } |
| 513 | void setSRetReturnReg(unsigned Reg) { SRetReturnReg = Reg; } |
| 514 | |
| 515 | unsigned getJumpTableEntrySize(int Idx) const { |
| 516 | return JumpTableEntryInfo[Idx].first; |
| 517 | } |
| 518 | MCSymbol *getJumpTableEntryPCRelSymbol(int Idx) const { |
| 519 | return JumpTableEntryInfo[Idx].second; |
| 520 | } |
| 521 | void setJumpTableEntryInfo(int Idx, unsigned Size, MCSymbol *PCRelSym) { |
| 522 | if ((unsigned)Idx >= JumpTableEntryInfo.size()) |
| 523 | JumpTableEntryInfo.resize(N: Idx+1); |
| 524 | JumpTableEntryInfo[Idx] = std::make_pair(x&: Size, y&: PCRelSym); |
| 525 | } |
| 526 | |
| 527 | using SetOfInstructions = SmallPtrSet<const MachineInstr *, 16>; |
| 528 | |
| 529 | const SetOfInstructions &getLOHRelated() const { return LOHRelated; } |
| 530 | |
| 531 | // Shortcuts for LOH related types. |
| 532 | class MILOHDirective { |
| 533 | MCLOHType Kind; |
| 534 | |
| 535 | /// Arguments of this directive. Order matters. |
| 536 | SmallVector<const MachineInstr *, 3> Args; |
| 537 | |
| 538 | public: |
| 539 | using LOHArgs = ArrayRef<const MachineInstr *>; |
| 540 | |
| 541 | MILOHDirective(MCLOHType Kind, LOHArgs Args) |
| 542 | : Kind(Kind), Args(Args.begin(), Args.end()) { |
| 543 | assert(isValidMCLOHType(Kind) && "Invalid LOH directive type!" ); |
| 544 | } |
| 545 | |
| 546 | MCLOHType getKind() const { return Kind; } |
| 547 | LOHArgs getArgs() const { return Args; } |
| 548 | }; |
| 549 | |
| 550 | using MILOHArgs = MILOHDirective::LOHArgs; |
| 551 | using MILOHContainer = SmallVector<MILOHDirective, 32>; |
| 552 | |
| 553 | const MILOHContainer &getLOHContainer() const { return LOHContainerSet; } |
| 554 | |
| 555 | /// Add a LOH directive of this @p Kind and this @p Args. |
| 556 | void addLOHDirective(MCLOHType Kind, MILOHArgs Args) { |
| 557 | LOHContainerSet.push_back(Elt: MILOHDirective(Kind, Args)); |
| 558 | LOHRelated.insert_range(R&: Args); |
| 559 | } |
| 560 | |
| 561 | size_t |
| 562 | clearLinkerOptimizationHints(const SmallPtrSetImpl<MachineInstr *> &MIs) { |
| 563 | size_t InitialSize = LOHContainerSet.size(); |
| 564 | erase_if(C&: LOHContainerSet, P: [&](const auto &D) { |
| 565 | return any_of(D.getArgs(), [&](auto *Arg) { return MIs.contains(Ptr: Arg); }); |
| 566 | }); |
| 567 | // In theory there could be an LOH with one label in MIs and another label |
| 568 | // outside MIs, however we don't know if the label outside MIs is used in |
| 569 | // any other LOHs, so we can't remove them from LOHRelated. In that case, we |
| 570 | // might produce a few extra labels, but it won't break anything. |
| 571 | LOHRelated.remove_if(P: [&](auto *MI) { return MIs.contains(Ptr: MI); }); |
| 572 | return InitialSize - LOHContainerSet.size(); |
| 573 | }; |
| 574 | |
| 575 | SmallVectorImpl<ForwardedRegister> &getForwardedMustTailRegParms() { |
| 576 | return ForwardedMustTailRegParms; |
| 577 | } |
| 578 | |
| 579 | std::optional<int> getTaggedBasePointerIndex() const { |
| 580 | return TaggedBasePointerIndex; |
| 581 | } |
| 582 | void setTaggedBasePointerIndex(int Index) { TaggedBasePointerIndex = Index; } |
| 583 | |
| 584 | unsigned getTaggedBasePointerOffset() const { |
| 585 | return TaggedBasePointerOffset; |
| 586 | } |
| 587 | void setTaggedBasePointerOffset(unsigned Offset) { |
| 588 | TaggedBasePointerOffset = Offset; |
| 589 | } |
| 590 | |
| 591 | int getCalleeSaveBaseToFrameRecordOffset() const { |
| 592 | return CalleeSaveBaseToFrameRecordOffset; |
| 593 | } |
| 594 | void setCalleeSaveBaseToFrameRecordOffset(int Offset) { |
| 595 | CalleeSaveBaseToFrameRecordOffset = Offset; |
| 596 | } |
| 597 | |
| 598 | static bool shouldSignReturnAddress(SignReturnAddress Condition, |
| 599 | bool IsLRSpilled); |
| 600 | |
| 601 | bool shouldSignReturnAddress(const MachineFunction &MF) const; |
| 602 | |
| 603 | SignReturnAddress getSignReturnAddressCondition() const { |
| 604 | return SignCondition; |
| 605 | } |
| 606 | |
| 607 | bool needsShadowCallStackPrologueEpilogue(MachineFunction &MF) const; |
| 608 | |
| 609 | bool shouldSignWithBKey() const { return SignWithBKey; } |
| 610 | |
| 611 | bool hasELFSignedGOT() const { return HasELFSignedGOT; } |
| 612 | |
| 613 | MCSymbol *getSigningInstrLabel() const { return SignInstrLabel; } |
| 614 | void setSigningInstrLabel(MCSymbol *Label) { SignInstrLabel = Label; } |
| 615 | |
| 616 | bool isMTETagged() const { return IsMTETagged; } |
| 617 | |
| 618 | bool branchTargetEnforcement() const { return BranchTargetEnforcement; } |
| 619 | |
| 620 | bool branchProtectionPAuthLR() const { return BranchProtectionPAuthLR; } |
| 621 | |
| 622 | void setHasSwiftAsyncContext(bool HasContext) { |
| 623 | HasSwiftAsyncContext = HasContext; |
| 624 | } |
| 625 | bool hasSwiftAsyncContext() const { return HasSwiftAsyncContext; } |
| 626 | |
| 627 | void setSwiftAsyncContextFrameIdx(int FI) { |
| 628 | SwiftAsyncContextFrameIdx = FI; |
| 629 | } |
| 630 | int getSwiftAsyncContextFrameIdx() const { return SwiftAsyncContextFrameIdx; } |
| 631 | |
| 632 | bool needsDwarfUnwindInfo(const MachineFunction &MF) const; |
| 633 | bool needsAsyncDwarfUnwindInfo(const MachineFunction &MF) const; |
| 634 | |
| 635 | bool hasStreamingModeChanges() const { return HasStreamingModeChanges; } |
| 636 | void setHasStreamingModeChanges(bool HasChanges) { |
| 637 | HasStreamingModeChanges = HasChanges; |
| 638 | } |
| 639 | |
| 640 | bool hasStackProbing() const { return StackProbeSize != 0; } |
| 641 | |
| 642 | int64_t getStackProbeSize() const { return StackProbeSize; } |
| 643 | |
| 644 | private: |
| 645 | // Hold the lists of LOHs. |
| 646 | MILOHContainer LOHContainerSet; |
| 647 | SetOfInstructions LOHRelated; |
| 648 | |
| 649 | SmallVector<std::pair<unsigned, MCSymbol *>, 2> JumpTableEntryInfo; |
| 650 | }; |
| 651 | |
| 652 | namespace yaml { |
| 653 | struct AArch64FunctionInfo final : public yaml::MachineFunctionInfo { |
| 654 | std::optional<bool> HasRedZone; |
| 655 | std::optional<uint64_t> StackSizeZPR; |
| 656 | std::optional<uint64_t> StackSizePPR; |
| 657 | std::optional<bool> HasStackFrame; |
| 658 | std::optional<bool> HasStreamingModeChanges; |
| 659 | |
| 660 | AArch64FunctionInfo() = default; |
| 661 | AArch64FunctionInfo(const llvm::AArch64FunctionInfo &MFI); |
| 662 | |
| 663 | void mappingImpl(yaml::IO &YamlIO) override; |
| 664 | ~AArch64FunctionInfo() override = default; |
| 665 | }; |
| 666 | |
| 667 | template <> struct MappingTraits<AArch64FunctionInfo> { |
| 668 | static void mapping(IO &YamlIO, AArch64FunctionInfo &MFI) { |
| 669 | YamlIO.mapOptional(Key: "hasRedZone" , Val&: MFI.HasRedZone); |
| 670 | YamlIO.mapOptional(Key: "stackSizeZPR" , Val&: MFI.StackSizeZPR); |
| 671 | YamlIO.mapOptional(Key: "stackSizePPR" , Val&: MFI.StackSizePPR); |
| 672 | YamlIO.mapOptional(Key: "hasStackFrame" , Val&: MFI.HasStackFrame); |
| 673 | YamlIO.mapOptional(Key: "hasStreamingModeChanges" , Val&: MFI.HasStreamingModeChanges); |
| 674 | } |
| 675 | }; |
| 676 | |
| 677 | } // end namespace yaml |
| 678 | |
| 679 | } // end namespace llvm |
| 680 | |
| 681 | #endif // LLVM_LIB_TARGET_AARCH64_AARCH64MACHINEFUNCTIONINFO_H |
| 682 | |