File: | llvm/lib/Target/AArch64/AArch64FrameLowering.cpp |
Warning: | line 2577, column 3 1st function call argument is an uninitialized value |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | //===- AArch64FrameLowering.cpp - AArch64 Frame Lowering -------*- 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 contains the AArch64 implementation of TargetFrameLowering class. | |||
10 | // | |||
11 | // On AArch64, stack frames are structured as follows: | |||
12 | // | |||
13 | // The stack grows downward. | |||
14 | // | |||
15 | // All of the individual frame areas on the frame below are optional, i.e. it's | |||
16 | // possible to create a function so that the particular area isn't present | |||
17 | // in the frame. | |||
18 | // | |||
19 | // At function entry, the "frame" looks as follows: | |||
20 | // | |||
21 | // | | Higher address | |||
22 | // |-----------------------------------| | |||
23 | // | | | |||
24 | // | arguments passed on the stack | | |||
25 | // | | | |||
26 | // |-----------------------------------| <- sp | |||
27 | // | | Lower address | |||
28 | // | |||
29 | // | |||
30 | // After the prologue has run, the frame has the following general structure. | |||
31 | // Note that this doesn't depict the case where a red-zone is used. Also, | |||
32 | // technically the last frame area (VLAs) doesn't get created until in the | |||
33 | // main function body, after the prologue is run. However, it's depicted here | |||
34 | // for completeness. | |||
35 | // | |||
36 | // | | Higher address | |||
37 | // |-----------------------------------| | |||
38 | // | | | |||
39 | // | arguments passed on the stack | | |||
40 | // | | | |||
41 | // |-----------------------------------| | |||
42 | // | | | |||
43 | // | (Win64 only) varargs from reg | | |||
44 | // | | | |||
45 | // |-----------------------------------| | |||
46 | // | | | |||
47 | // | callee-saved gpr registers | <--. | |||
48 | // | | | On Darwin platforms these | |||
49 | // |- - - - - - - - - - - - - - - - - -| | callee saves are swapped, | |||
50 | // | | | (frame record first) | |||
51 | // | prev_fp, prev_lr | <--' | |||
52 | // | (a.k.a. "frame record") | | |||
53 | // |-----------------------------------| <- fp(=x29) | |||
54 | // | | | |||
55 | // | callee-saved fp/simd/SVE regs | | |||
56 | // | | | |||
57 | // |-----------------------------------| | |||
58 | // | | | |||
59 | // | SVE stack objects | | |||
60 | // | | | |||
61 | // |-----------------------------------| | |||
62 | // |.empty.space.to.make.part.below....| | |||
63 | // |.aligned.in.case.it.needs.more.than| (size of this area is unknown at | |||
64 | // |.the.standard.16-byte.alignment....| compile time; if present) | |||
65 | // |-----------------------------------| | |||
66 | // | | | |||
67 | // | local variables of fixed size | | |||
68 | // | including spill slots | | |||
69 | // |-----------------------------------| <- bp(not defined by ABI, | |||
70 | // |.variable-sized.local.variables....| LLVM chooses X19) | |||
71 | // |.(VLAs)............................| (size of this area is unknown at | |||
72 | // |...................................| compile time) | |||
73 | // |-----------------------------------| <- sp | |||
74 | // | | Lower address | |||
75 | // | |||
76 | // | |||
77 | // To access the data in a frame, at-compile time, a constant offset must be | |||
78 | // computable from one of the pointers (fp, bp, sp) to access it. The size | |||
79 | // of the areas with a dotted background cannot be computed at compile-time | |||
80 | // if they are present, making it required to have all three of fp, bp and | |||
81 | // sp to be set up to be able to access all contents in the frame areas, | |||
82 | // assuming all of the frame areas are non-empty. | |||
83 | // | |||
84 | // For most functions, some of the frame areas are empty. For those functions, | |||
85 | // it may not be necessary to set up fp or bp: | |||
86 | // * A base pointer is definitely needed when there are both VLAs and local | |||
87 | // variables with more-than-default alignment requirements. | |||
88 | // * A frame pointer is definitely needed when there are local variables with | |||
89 | // more-than-default alignment requirements. | |||
90 | // | |||
91 | // For Darwin platforms the frame-record (fp, lr) is stored at the top of the | |||
92 | // callee-saved area, since the unwind encoding does not allow for encoding | |||
93 | // this dynamically and existing tools depend on this layout. For other | |||
94 | // platforms, the frame-record is stored at the bottom of the (gpr) callee-saved | |||
95 | // area to allow SVE stack objects (allocated directly below the callee-saves, | |||
96 | // if available) to be accessed directly from the framepointer. | |||
97 | // The SVE spill/fill instructions have VL-scaled addressing modes such | |||
98 | // as: | |||
99 | // ldr z8, [fp, #-7 mul vl] | |||
100 | // For SVE the size of the vector length (VL) is not known at compile-time, so | |||
101 | // '#-7 mul vl' is an offset that can only be evaluated at runtime. With this | |||
102 | // layout, we don't need to add an unscaled offset to the framepointer before | |||
103 | // accessing the SVE object in the frame. | |||
104 | // | |||
105 | // In some cases when a base pointer is not strictly needed, it is generated | |||
106 | // anyway when offsets from the frame pointer to access local variables become | |||
107 | // so large that the offset can't be encoded in the immediate fields of loads | |||
108 | // or stores. | |||
109 | // | |||
110 | // FIXME: also explain the redzone concept. | |||
111 | // FIXME: also explain the concept of reserved call frames. | |||
112 | // | |||
113 | //===----------------------------------------------------------------------===// | |||
114 | ||||
115 | #include "AArch64FrameLowering.h" | |||
116 | #include "AArch64InstrInfo.h" | |||
117 | #include "AArch64MachineFunctionInfo.h" | |||
118 | #include "AArch64RegisterInfo.h" | |||
119 | #include "AArch64StackOffset.h" | |||
120 | #include "AArch64Subtarget.h" | |||
121 | #include "AArch64TargetMachine.h" | |||
122 | #include "MCTargetDesc/AArch64AddressingModes.h" | |||
123 | #include "llvm/ADT/ScopeExit.h" | |||
124 | #include "llvm/ADT/SmallVector.h" | |||
125 | #include "llvm/ADT/Statistic.h" | |||
126 | #include "llvm/CodeGen/LivePhysRegs.h" | |||
127 | #include "llvm/CodeGen/MachineBasicBlock.h" | |||
128 | #include "llvm/CodeGen/MachineFrameInfo.h" | |||
129 | #include "llvm/CodeGen/MachineFunction.h" | |||
130 | #include "llvm/CodeGen/MachineInstr.h" | |||
131 | #include "llvm/CodeGen/MachineInstrBuilder.h" | |||
132 | #include "llvm/CodeGen/MachineMemOperand.h" | |||
133 | #include "llvm/CodeGen/MachineModuleInfo.h" | |||
134 | #include "llvm/CodeGen/MachineOperand.h" | |||
135 | #include "llvm/CodeGen/MachineRegisterInfo.h" | |||
136 | #include "llvm/CodeGen/RegisterScavenging.h" | |||
137 | #include "llvm/CodeGen/TargetInstrInfo.h" | |||
138 | #include "llvm/CodeGen/TargetRegisterInfo.h" | |||
139 | #include "llvm/CodeGen/TargetSubtargetInfo.h" | |||
140 | #include "llvm/CodeGen/WinEHFuncInfo.h" | |||
141 | #include "llvm/IR/Attributes.h" | |||
142 | #include "llvm/IR/CallingConv.h" | |||
143 | #include "llvm/IR/DataLayout.h" | |||
144 | #include "llvm/IR/DebugLoc.h" | |||
145 | #include "llvm/IR/Function.h" | |||
146 | #include "llvm/MC/MCAsmInfo.h" | |||
147 | #include "llvm/MC/MCDwarf.h" | |||
148 | #include "llvm/Support/CommandLine.h" | |||
149 | #include "llvm/Support/Debug.h" | |||
150 | #include "llvm/Support/ErrorHandling.h" | |||
151 | #include "llvm/Support/MathExtras.h" | |||
152 | #include "llvm/Support/raw_ostream.h" | |||
153 | #include "llvm/Target/TargetMachine.h" | |||
154 | #include "llvm/Target/TargetOptions.h" | |||
155 | #include <cassert> | |||
156 | #include <cstdint> | |||
157 | #include <iterator> | |||
158 | #include <vector> | |||
159 | ||||
160 | using namespace llvm; | |||
161 | ||||
162 | #define DEBUG_TYPE"frame-info" "frame-info" | |||
163 | ||||
164 | static cl::opt<bool> EnableRedZone("aarch64-redzone", | |||
165 | cl::desc("enable use of redzone on AArch64"), | |||
166 | cl::init(false), cl::Hidden); | |||
167 | ||||
168 | static cl::opt<bool> | |||
169 | ReverseCSRRestoreSeq("reverse-csr-restore-seq", | |||
170 | cl::desc("reverse the CSR restore sequence"), | |||
171 | cl::init(false), cl::Hidden); | |||
172 | ||||
173 | STATISTIC(NumRedZoneFunctions, "Number of functions using red zone")static llvm::Statistic NumRedZoneFunctions = {"frame-info", "NumRedZoneFunctions" , "Number of functions using red zone"}; | |||
174 | ||||
175 | /// This is the biggest offset to the stack pointer we can encode in aarch64 | |||
176 | /// instructions (without using a separate calculation and a temp register). | |||
177 | /// Note that the exception here are vector stores/loads which cannot encode any | |||
178 | /// displacements (see estimateRSStackSizeLimit(), isAArch64FrameOffsetLegal()). | |||
179 | static const unsigned DefaultSafeSPDisplacement = 255; | |||
180 | ||||
181 | /// Look at each instruction that references stack frames and return the stack | |||
182 | /// size limit beyond which some of these instructions will require a scratch | |||
183 | /// register during their expansion later. | |||
184 | static unsigned estimateRSStackSizeLimit(MachineFunction &MF) { | |||
185 | // FIXME: For now, just conservatively guestimate based on unscaled indexing | |||
186 | // range. We'll end up allocating an unnecessary spill slot a lot, but | |||
187 | // realistically that's not a big deal at this stage of the game. | |||
188 | for (MachineBasicBlock &MBB : MF) { | |||
189 | for (MachineInstr &MI : MBB) { | |||
190 | if (MI.isDebugInstr() || MI.isPseudo() || | |||
191 | MI.getOpcode() == AArch64::ADDXri || | |||
192 | MI.getOpcode() == AArch64::ADDSXri) | |||
193 | continue; | |||
194 | ||||
195 | for (const MachineOperand &MO : MI.operands()) { | |||
196 | if (!MO.isFI()) | |||
197 | continue; | |||
198 | ||||
199 | StackOffset Offset; | |||
200 | if (isAArch64FrameOffsetLegal(MI, Offset, nullptr, nullptr, nullptr) == | |||
201 | AArch64FrameOffsetCannotUpdate) | |||
202 | return 0; | |||
203 | } | |||
204 | } | |||
205 | } | |||
206 | return DefaultSafeSPDisplacement; | |||
207 | } | |||
208 | ||||
209 | /// Returns the size of the entire SVE stackframe (calleesaves + spills). | |||
210 | static StackOffset getSVEStackSize(const MachineFunction &MF) { | |||
211 | const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
212 | return {(int64_t)AFI->getStackSizeSVE(), MVT::nxv1i8}; | |||
213 | } | |||
214 | ||||
215 | bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const { | |||
216 | if (!EnableRedZone) | |||
217 | return false; | |||
218 | // Don't use the red zone if the function explicitly asks us not to. | |||
219 | // This is typically used for kernel code. | |||
220 | if (MF.getFunction().hasFnAttribute(Attribute::NoRedZone)) | |||
221 | return false; | |||
222 | ||||
223 | const MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
224 | const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
225 | unsigned NumBytes = AFI->getLocalStackSize(); | |||
226 | ||||
227 | return !(MFI.hasCalls() || hasFP(MF) || NumBytes > 128 || | |||
228 | getSVEStackSize(MF)); | |||
229 | } | |||
230 | ||||
231 | /// hasFP - Return true if the specified function should have a dedicated frame | |||
232 | /// pointer register. | |||
233 | bool AArch64FrameLowering::hasFP(const MachineFunction &MF) const { | |||
234 | const MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
235 | const TargetRegisterInfo *RegInfo = MF.getSubtarget().getRegisterInfo(); | |||
236 | // Win64 EH requires a frame pointer if funclets are present, as the locals | |||
237 | // are accessed off the frame pointer in both the parent function and the | |||
238 | // funclets. | |||
239 | if (MF.hasEHFunclets()) | |||
240 | return true; | |||
241 | // Retain behavior of always omitting the FP for leaf functions when possible. | |||
242 | if (MFI.hasCalls() && MF.getTarget().Options.DisableFramePointerElim(MF)) | |||
243 | return true; | |||
244 | if (MFI.hasVarSizedObjects() || MFI.isFrameAddressTaken() || | |||
245 | MFI.hasStackMap() || MFI.hasPatchPoint() || | |||
246 | RegInfo->needsStackRealignment(MF)) | |||
247 | return true; | |||
248 | // With large callframes around we may need to use FP to access the scavenging | |||
249 | // emergency spillslot. | |||
250 | // | |||
251 | // Unfortunately some calls to hasFP() like machine verifier -> | |||
252 | // getReservedReg() -> hasFP in the middle of global isel are too early | |||
253 | // to know the max call frame size. Hopefully conservatively returning "true" | |||
254 | // in those cases is fine. | |||
255 | // DefaultSafeSPDisplacement is fine as we only emergency spill GP regs. | |||
256 | if (!MFI.isMaxCallFrameSizeComputed() || | |||
257 | MFI.getMaxCallFrameSize() > DefaultSafeSPDisplacement) | |||
258 | return true; | |||
259 | ||||
260 | return false; | |||
261 | } | |||
262 | ||||
263 | /// hasReservedCallFrame - Under normal circumstances, when a frame pointer is | |||
264 | /// not required, we reserve argument space for call sites in the function | |||
265 | /// immediately on entry to the current function. This eliminates the need for | |||
266 | /// add/sub sp brackets around call sites. Returns true if the call frame is | |||
267 | /// included as part of the stack frame. | |||
268 | bool | |||
269 | AArch64FrameLowering::hasReservedCallFrame(const MachineFunction &MF) const { | |||
270 | return !MF.getFrameInfo().hasVarSizedObjects(); | |||
271 | } | |||
272 | ||||
273 | MachineBasicBlock::iterator AArch64FrameLowering::eliminateCallFramePseudoInstr( | |||
274 | MachineFunction &MF, MachineBasicBlock &MBB, | |||
275 | MachineBasicBlock::iterator I) const { | |||
276 | const AArch64InstrInfo *TII = | |||
277 | static_cast<const AArch64InstrInfo *>(MF.getSubtarget().getInstrInfo()); | |||
278 | DebugLoc DL = I->getDebugLoc(); | |||
279 | unsigned Opc = I->getOpcode(); | |||
280 | bool IsDestroy = Opc == TII->getCallFrameDestroyOpcode(); | |||
281 | uint64_t CalleePopAmount = IsDestroy ? I->getOperand(1).getImm() : 0; | |||
282 | ||||
283 | if (!hasReservedCallFrame(MF)) { | |||
284 | unsigned Align = getStackAlignment(); | |||
285 | ||||
286 | int64_t Amount = I->getOperand(0).getImm(); | |||
287 | Amount = alignTo(Amount, Align); | |||
288 | if (!IsDestroy) | |||
289 | Amount = -Amount; | |||
290 | ||||
291 | // N.b. if CalleePopAmount is valid but zero (i.e. callee would pop, but it | |||
292 | // doesn't have to pop anything), then the first operand will be zero too so | |||
293 | // this adjustment is a no-op. | |||
294 | if (CalleePopAmount == 0) { | |||
295 | // FIXME: in-function stack adjustment for calls is limited to 24-bits | |||
296 | // because there's no guaranteed temporary register available. | |||
297 | // | |||
298 | // ADD/SUB (immediate) has only LSL #0 and LSL #12 available. | |||
299 | // 1) For offset <= 12-bit, we use LSL #0 | |||
300 | // 2) For 12-bit <= offset <= 24-bit, we use two instructions. One uses | |||
301 | // LSL #0, and the other uses LSL #12. | |||
302 | // | |||
303 | // Most call frames will be allocated at the start of a function so | |||
304 | // this is OK, but it is a limitation that needs dealing with. | |||
305 | assert(Amount > -0xffffff && Amount < 0xffffff && "call frame too large")((Amount > -0xffffff && Amount < 0xffffff && "call frame too large") ? static_cast<void> (0) : __assert_fail ("Amount > -0xffffff && Amount < 0xffffff && \"call frame too large\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 305, __PRETTY_FUNCTION__)); | |||
306 | emitFrameOffset(MBB, I, DL, AArch64::SP, AArch64::SP, {Amount, MVT::i8}, | |||
307 | TII); | |||
308 | } | |||
309 | } else if (CalleePopAmount != 0) { | |||
310 | // If the calling convention demands that the callee pops arguments from the | |||
311 | // stack, we want to add it back if we have a reserved call frame. | |||
312 | assert(CalleePopAmount < 0xffffff && "call frame too large")((CalleePopAmount < 0xffffff && "call frame too large" ) ? static_cast<void> (0) : __assert_fail ("CalleePopAmount < 0xffffff && \"call frame too large\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 312, __PRETTY_FUNCTION__)); | |||
313 | emitFrameOffset(MBB, I, DL, AArch64::SP, AArch64::SP, | |||
314 | {-(int64_t)CalleePopAmount, MVT::i8}, TII); | |||
315 | } | |||
316 | return MBB.erase(I); | |||
317 | } | |||
318 | ||||
319 | static bool ShouldSignReturnAddress(MachineFunction &MF) { | |||
320 | // The function should be signed in the following situations: | |||
321 | // - sign-return-address=all | |||
322 | // - sign-return-address=non-leaf and the functions spills the LR | |||
323 | ||||
324 | const Function &F = MF.getFunction(); | |||
325 | if (!F.hasFnAttribute("sign-return-address")) | |||
326 | return false; | |||
327 | ||||
328 | StringRef Scope = F.getFnAttribute("sign-return-address").getValueAsString(); | |||
329 | if (Scope.equals("none")) | |||
330 | return false; | |||
331 | ||||
332 | if (Scope.equals("all")) | |||
333 | return true; | |||
334 | ||||
335 | assert(Scope.equals("non-leaf") && "Expected all, none or non-leaf")((Scope.equals("non-leaf") && "Expected all, none or non-leaf" ) ? static_cast<void> (0) : __assert_fail ("Scope.equals(\"non-leaf\") && \"Expected all, none or non-leaf\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 335, __PRETTY_FUNCTION__)); | |||
336 | ||||
337 | for (const auto &Info : MF.getFrameInfo().getCalleeSavedInfo()) | |||
338 | if (Info.getReg() == AArch64::LR) | |||
339 | return true; | |||
340 | ||||
341 | return false; | |||
342 | } | |||
343 | ||||
344 | void AArch64FrameLowering::emitCalleeSavedFrameMoves( | |||
345 | MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI) const { | |||
346 | MachineFunction &MF = *MBB.getParent(); | |||
347 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
348 | const TargetSubtargetInfo &STI = MF.getSubtarget(); | |||
349 | const MCRegisterInfo *MRI = STI.getRegisterInfo(); | |||
350 | const TargetInstrInfo *TII = STI.getInstrInfo(); | |||
351 | DebugLoc DL = MBB.findDebugLoc(MBBI); | |||
352 | ||||
353 | // Add callee saved registers to move list. | |||
354 | const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo(); | |||
355 | if (CSI.empty()) | |||
356 | return; | |||
357 | ||||
358 | for (const auto &Info : CSI) { | |||
359 | unsigned Reg = Info.getReg(); | |||
360 | int64_t Offset = | |||
361 | MFI.getObjectOffset(Info.getFrameIdx()) - getOffsetOfLocalArea(); | |||
362 | unsigned DwarfReg = MRI->getDwarfRegNum(Reg, true); | |||
363 | unsigned CFIIndex = MF.addFrameInst( | |||
364 | MCCFIInstruction::createOffset(nullptr, DwarfReg, Offset)); | |||
365 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION)) | |||
366 | .addCFIIndex(CFIIndex) | |||
367 | .setMIFlags(MachineInstr::FrameSetup); | |||
368 | } | |||
369 | } | |||
370 | ||||
371 | // Find a scratch register that we can use at the start of the prologue to | |||
372 | // re-align the stack pointer. We avoid using callee-save registers since they | |||
373 | // may appear to be free when this is called from canUseAsPrologue (during | |||
374 | // shrink wrapping), but then no longer be free when this is called from | |||
375 | // emitPrologue. | |||
376 | // | |||
377 | // FIXME: This is a bit conservative, since in the above case we could use one | |||
378 | // of the callee-save registers as a scratch temp to re-align the stack pointer, | |||
379 | // but we would then have to make sure that we were in fact saving at least one | |||
380 | // callee-save register in the prologue, which is additional complexity that | |||
381 | // doesn't seem worth the benefit. | |||
382 | static unsigned findScratchNonCalleeSaveRegister(MachineBasicBlock *MBB) { | |||
383 | MachineFunction *MF = MBB->getParent(); | |||
384 | ||||
385 | // If MBB is an entry block, use X9 as the scratch register | |||
386 | if (&MF->front() == MBB) | |||
387 | return AArch64::X9; | |||
388 | ||||
389 | const AArch64Subtarget &Subtarget = MF->getSubtarget<AArch64Subtarget>(); | |||
390 | const AArch64RegisterInfo &TRI = *Subtarget.getRegisterInfo(); | |||
391 | LivePhysRegs LiveRegs(TRI); | |||
392 | LiveRegs.addLiveIns(*MBB); | |||
393 | ||||
394 | // Mark callee saved registers as used so we will not choose them. | |||
395 | const MCPhysReg *CSRegs = MF->getRegInfo().getCalleeSavedRegs(); | |||
396 | for (unsigned i = 0; CSRegs[i]; ++i) | |||
397 | LiveRegs.addReg(CSRegs[i]); | |||
398 | ||||
399 | // Prefer X9 since it was historically used for the prologue scratch reg. | |||
400 | const MachineRegisterInfo &MRI = MF->getRegInfo(); | |||
401 | if (LiveRegs.available(MRI, AArch64::X9)) | |||
402 | return AArch64::X9; | |||
403 | ||||
404 | for (unsigned Reg : AArch64::GPR64RegClass) { | |||
405 | if (LiveRegs.available(MRI, Reg)) | |||
406 | return Reg; | |||
407 | } | |||
408 | return AArch64::NoRegister; | |||
409 | } | |||
410 | ||||
411 | bool AArch64FrameLowering::canUseAsPrologue( | |||
412 | const MachineBasicBlock &MBB) const { | |||
413 | const MachineFunction *MF = MBB.getParent(); | |||
414 | MachineBasicBlock *TmpMBB = const_cast<MachineBasicBlock *>(&MBB); | |||
415 | const AArch64Subtarget &Subtarget = MF->getSubtarget<AArch64Subtarget>(); | |||
416 | const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo(); | |||
417 | ||||
418 | // Don't need a scratch register if we're not going to re-align the stack. | |||
419 | if (!RegInfo->needsStackRealignment(*MF)) | |||
420 | return true; | |||
421 | // Otherwise, we can use any block as long as it has a scratch register | |||
422 | // available. | |||
423 | return findScratchNonCalleeSaveRegister(TmpMBB) != AArch64::NoRegister; | |||
424 | } | |||
425 | ||||
426 | static bool windowsRequiresStackProbe(MachineFunction &MF, | |||
427 | unsigned StackSizeInBytes) { | |||
428 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
429 | if (!Subtarget.isTargetWindows()) | |||
430 | return false; | |||
431 | const Function &F = MF.getFunction(); | |||
432 | // TODO: When implementing stack protectors, take that into account | |||
433 | // for the probe threshold. | |||
434 | unsigned StackProbeSize = 4096; | |||
435 | if (F.hasFnAttribute("stack-probe-size")) | |||
436 | F.getFnAttribute("stack-probe-size") | |||
437 | .getValueAsString() | |||
438 | .getAsInteger(0, StackProbeSize); | |||
439 | return (StackSizeInBytes >= StackProbeSize) && | |||
440 | !F.hasFnAttribute("no-stack-arg-probe"); | |||
441 | } | |||
442 | ||||
443 | bool AArch64FrameLowering::shouldCombineCSRLocalStackBump( | |||
444 | MachineFunction &MF, unsigned StackBumpBytes) const { | |||
445 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
446 | const MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
447 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
448 | const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo(); | |||
449 | ||||
450 | if (MF.getFunction().hasOptSize()) | |||
451 | return false; | |||
452 | ||||
453 | if (AFI->getLocalStackSize() == 0) | |||
454 | return false; | |||
455 | ||||
456 | // 512 is the maximum immediate for stp/ldp that will be used for | |||
457 | // callee-save save/restores | |||
458 | if (StackBumpBytes >= 512 || windowsRequiresStackProbe(MF, StackBumpBytes)) | |||
459 | return false; | |||
460 | ||||
461 | if (MFI.hasVarSizedObjects()) | |||
462 | return false; | |||
463 | ||||
464 | if (RegInfo->needsStackRealignment(MF)) | |||
465 | return false; | |||
466 | ||||
467 | // This isn't strictly necessary, but it simplifies things a bit since the | |||
468 | // current RedZone handling code assumes the SP is adjusted by the | |||
469 | // callee-save save/restore code. | |||
470 | if (canUseRedZone(MF)) | |||
471 | return false; | |||
472 | ||||
473 | // When there is an SVE area on the stack, always allocate the | |||
474 | // callee-saves and spills/locals separately. | |||
475 | if (getSVEStackSize(MF)) | |||
476 | return false; | |||
477 | ||||
478 | return true; | |||
479 | } | |||
480 | ||||
481 | // Given a load or a store instruction, generate an appropriate unwinding SEH | |||
482 | // code on Windows. | |||
483 | static MachineBasicBlock::iterator InsertSEH(MachineBasicBlock::iterator MBBI, | |||
484 | const TargetInstrInfo &TII, | |||
485 | MachineInstr::MIFlag Flag) { | |||
486 | unsigned Opc = MBBI->getOpcode(); | |||
487 | MachineBasicBlock *MBB = MBBI->getParent(); | |||
488 | MachineFunction &MF = *MBB->getParent(); | |||
489 | DebugLoc DL = MBBI->getDebugLoc(); | |||
490 | unsigned ImmIdx = MBBI->getNumOperands() - 1; | |||
491 | int Imm = MBBI->getOperand(ImmIdx).getImm(); | |||
492 | MachineInstrBuilder MIB; | |||
493 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
494 | const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo(); | |||
495 | ||||
496 | switch (Opc) { | |||
497 | default: | |||
498 | llvm_unreachable("No SEH Opcode for this instruction")::llvm::llvm_unreachable_internal("No SEH Opcode for this instruction" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 498); | |||
499 | case AArch64::LDPDpost: | |||
500 | Imm = -Imm; | |||
501 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
502 | case AArch64::STPDpre: { | |||
503 | unsigned Reg0 = RegInfo->getSEHRegNum(MBBI->getOperand(1).getReg()); | |||
504 | unsigned Reg1 = RegInfo->getSEHRegNum(MBBI->getOperand(2).getReg()); | |||
505 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFRegP_X)) | |||
506 | .addImm(Reg0) | |||
507 | .addImm(Reg1) | |||
508 | .addImm(Imm * 8) | |||
509 | .setMIFlag(Flag); | |||
510 | break; | |||
511 | } | |||
512 | case AArch64::LDPXpost: | |||
513 | Imm = -Imm; | |||
514 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
515 | case AArch64::STPXpre: { | |||
516 | Register Reg0 = MBBI->getOperand(1).getReg(); | |||
517 | Register Reg1 = MBBI->getOperand(2).getReg(); | |||
518 | if (Reg0 == AArch64::FP && Reg1 == AArch64::LR) | |||
519 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFPLR_X)) | |||
520 | .addImm(Imm * 8) | |||
521 | .setMIFlag(Flag); | |||
522 | else | |||
523 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveRegP_X)) | |||
524 | .addImm(RegInfo->getSEHRegNum(Reg0)) | |||
525 | .addImm(RegInfo->getSEHRegNum(Reg1)) | |||
526 | .addImm(Imm * 8) | |||
527 | .setMIFlag(Flag); | |||
528 | break; | |||
529 | } | |||
530 | case AArch64::LDRDpost: | |||
531 | Imm = -Imm; | |||
532 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
533 | case AArch64::STRDpre: { | |||
534 | unsigned Reg = RegInfo->getSEHRegNum(MBBI->getOperand(1).getReg()); | |||
535 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFReg_X)) | |||
536 | .addImm(Reg) | |||
537 | .addImm(Imm) | |||
538 | .setMIFlag(Flag); | |||
539 | break; | |||
540 | } | |||
541 | case AArch64::LDRXpost: | |||
542 | Imm = -Imm; | |||
543 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
544 | case AArch64::STRXpre: { | |||
545 | unsigned Reg = RegInfo->getSEHRegNum(MBBI->getOperand(1).getReg()); | |||
546 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveReg_X)) | |||
547 | .addImm(Reg) | |||
548 | .addImm(Imm) | |||
549 | .setMIFlag(Flag); | |||
550 | break; | |||
551 | } | |||
552 | case AArch64::STPDi: | |||
553 | case AArch64::LDPDi: { | |||
554 | unsigned Reg0 = RegInfo->getSEHRegNum(MBBI->getOperand(0).getReg()); | |||
555 | unsigned Reg1 = RegInfo->getSEHRegNum(MBBI->getOperand(1).getReg()); | |||
556 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFRegP)) | |||
557 | .addImm(Reg0) | |||
558 | .addImm(Reg1) | |||
559 | .addImm(Imm * 8) | |||
560 | .setMIFlag(Flag); | |||
561 | break; | |||
562 | } | |||
563 | case AArch64::STPXi: | |||
564 | case AArch64::LDPXi: { | |||
565 | Register Reg0 = MBBI->getOperand(0).getReg(); | |||
566 | Register Reg1 = MBBI->getOperand(1).getReg(); | |||
567 | if (Reg0 == AArch64::FP && Reg1 == AArch64::LR) | |||
568 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFPLR)) | |||
569 | .addImm(Imm * 8) | |||
570 | .setMIFlag(Flag); | |||
571 | else | |||
572 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveRegP)) | |||
573 | .addImm(RegInfo->getSEHRegNum(Reg0)) | |||
574 | .addImm(RegInfo->getSEHRegNum(Reg1)) | |||
575 | .addImm(Imm * 8) | |||
576 | .setMIFlag(Flag); | |||
577 | break; | |||
578 | } | |||
579 | case AArch64::STRXui: | |||
580 | case AArch64::LDRXui: { | |||
581 | int Reg = RegInfo->getSEHRegNum(MBBI->getOperand(0).getReg()); | |||
582 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveReg)) | |||
583 | .addImm(Reg) | |||
584 | .addImm(Imm * 8) | |||
585 | .setMIFlag(Flag); | |||
586 | break; | |||
587 | } | |||
588 | case AArch64::STRDui: | |||
589 | case AArch64::LDRDui: { | |||
590 | unsigned Reg = RegInfo->getSEHRegNum(MBBI->getOperand(0).getReg()); | |||
591 | MIB = BuildMI(MF, DL, TII.get(AArch64::SEH_SaveFReg)) | |||
592 | .addImm(Reg) | |||
593 | .addImm(Imm * 8) | |||
594 | .setMIFlag(Flag); | |||
595 | break; | |||
596 | } | |||
597 | } | |||
598 | auto I = MBB->insertAfter(MBBI, MIB); | |||
599 | return I; | |||
600 | } | |||
601 | ||||
602 | // Fix up the SEH opcode associated with the save/restore instruction. | |||
603 | static void fixupSEHOpcode(MachineBasicBlock::iterator MBBI, | |||
604 | unsigned LocalStackSize) { | |||
605 | MachineOperand *ImmOpnd = nullptr; | |||
606 | unsigned ImmIdx = MBBI->getNumOperands() - 1; | |||
607 | switch (MBBI->getOpcode()) { | |||
608 | default: | |||
609 | llvm_unreachable("Fix the offset in the SEH instruction")::llvm::llvm_unreachable_internal("Fix the offset in the SEH instruction" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 609); | |||
610 | case AArch64::SEH_SaveFPLR: | |||
611 | case AArch64::SEH_SaveRegP: | |||
612 | case AArch64::SEH_SaveReg: | |||
613 | case AArch64::SEH_SaveFRegP: | |||
614 | case AArch64::SEH_SaveFReg: | |||
615 | ImmOpnd = &MBBI->getOperand(ImmIdx); | |||
616 | break; | |||
617 | } | |||
618 | if (ImmOpnd) | |||
619 | ImmOpnd->setImm(ImmOpnd->getImm() + LocalStackSize); | |||
620 | } | |||
621 | ||||
622 | // Convert callee-save register save/restore instruction to do stack pointer | |||
623 | // decrement/increment to allocate/deallocate the callee-save stack area by | |||
624 | // converting store/load to use pre/post increment version. | |||
625 | static MachineBasicBlock::iterator convertCalleeSaveRestoreToSPPrePostIncDec( | |||
626 | MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, | |||
627 | const DebugLoc &DL, const TargetInstrInfo *TII, int CSStackSizeInc, | |||
628 | bool NeedsWinCFI, bool *HasWinCFI, bool InProlog = true) { | |||
629 | // Ignore instructions that do not operate on SP, i.e. shadow call stack | |||
630 | // instructions and associated CFI instruction. | |||
631 | while (MBBI->getOpcode() == AArch64::STRXpost || | |||
632 | MBBI->getOpcode() == AArch64::LDRXpre || | |||
633 | MBBI->getOpcode() == AArch64::CFI_INSTRUCTION) { | |||
634 | if (MBBI->getOpcode() != AArch64::CFI_INSTRUCTION) | |||
635 | assert(MBBI->getOperand(0).getReg() != AArch64::SP)((MBBI->getOperand(0).getReg() != AArch64::SP) ? static_cast <void> (0) : __assert_fail ("MBBI->getOperand(0).getReg() != AArch64::SP" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 635, __PRETTY_FUNCTION__)); | |||
636 | ++MBBI; | |||
637 | } | |||
638 | unsigned NewOpc; | |||
639 | int Scale = 1; | |||
640 | switch (MBBI->getOpcode()) { | |||
641 | default: | |||
642 | llvm_unreachable("Unexpected callee-save save/restore opcode!")::llvm::llvm_unreachable_internal("Unexpected callee-save save/restore opcode!" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 642); | |||
643 | case AArch64::STPXi: | |||
644 | NewOpc = AArch64::STPXpre; | |||
645 | Scale = 8; | |||
646 | break; | |||
647 | case AArch64::STPDi: | |||
648 | NewOpc = AArch64::STPDpre; | |||
649 | Scale = 8; | |||
650 | break; | |||
651 | case AArch64::STPQi: | |||
652 | NewOpc = AArch64::STPQpre; | |||
653 | Scale = 16; | |||
654 | break; | |||
655 | case AArch64::STRXui: | |||
656 | NewOpc = AArch64::STRXpre; | |||
657 | break; | |||
658 | case AArch64::STRDui: | |||
659 | NewOpc = AArch64::STRDpre; | |||
660 | break; | |||
661 | case AArch64::STRQui: | |||
662 | NewOpc = AArch64::STRQpre; | |||
663 | break; | |||
664 | case AArch64::LDPXi: | |||
665 | NewOpc = AArch64::LDPXpost; | |||
666 | Scale = 8; | |||
667 | break; | |||
668 | case AArch64::LDPDi: | |||
669 | NewOpc = AArch64::LDPDpost; | |||
670 | Scale = 8; | |||
671 | break; | |||
672 | case AArch64::LDPQi: | |||
673 | NewOpc = AArch64::LDPQpost; | |||
674 | Scale = 16; | |||
675 | break; | |||
676 | case AArch64::LDRXui: | |||
677 | NewOpc = AArch64::LDRXpost; | |||
678 | break; | |||
679 | case AArch64::LDRDui: | |||
680 | NewOpc = AArch64::LDRDpost; | |||
681 | break; | |||
682 | case AArch64::LDRQui: | |||
683 | NewOpc = AArch64::LDRQpost; | |||
684 | break; | |||
685 | } | |||
686 | // Get rid of the SEH code associated with the old instruction. | |||
687 | if (NeedsWinCFI) { | |||
688 | auto SEH = std::next(MBBI); | |||
689 | if (AArch64InstrInfo::isSEHInstruction(*SEH)) | |||
690 | SEH->eraseFromParent(); | |||
691 | } | |||
692 | ||||
693 | MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(NewOpc)); | |||
694 | MIB.addReg(AArch64::SP, RegState::Define); | |||
695 | ||||
696 | // Copy all operands other than the immediate offset. | |||
697 | unsigned OpndIdx = 0; | |||
698 | for (unsigned OpndEnd = MBBI->getNumOperands() - 1; OpndIdx < OpndEnd; | |||
699 | ++OpndIdx) | |||
700 | MIB.add(MBBI->getOperand(OpndIdx)); | |||
701 | ||||
702 | assert(MBBI->getOperand(OpndIdx).getImm() == 0 &&((MBBI->getOperand(OpndIdx).getImm() == 0 && "Unexpected immediate offset in first/last callee-save save/restore " "instruction!") ? static_cast<void> (0) : __assert_fail ("MBBI->getOperand(OpndIdx).getImm() == 0 && \"Unexpected immediate offset in first/last callee-save save/restore \" \"instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 704, __PRETTY_FUNCTION__)) | |||
703 | "Unexpected immediate offset in first/last callee-save save/restore "((MBBI->getOperand(OpndIdx).getImm() == 0 && "Unexpected immediate offset in first/last callee-save save/restore " "instruction!") ? static_cast<void> (0) : __assert_fail ("MBBI->getOperand(OpndIdx).getImm() == 0 && \"Unexpected immediate offset in first/last callee-save save/restore \" \"instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 704, __PRETTY_FUNCTION__)) | |||
704 | "instruction!")((MBBI->getOperand(OpndIdx).getImm() == 0 && "Unexpected immediate offset in first/last callee-save save/restore " "instruction!") ? static_cast<void> (0) : __assert_fail ("MBBI->getOperand(OpndIdx).getImm() == 0 && \"Unexpected immediate offset in first/last callee-save save/restore \" \"instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 704, __PRETTY_FUNCTION__)); | |||
705 | assert(MBBI->getOperand(OpndIdx - 1).getReg() == AArch64::SP &&((MBBI->getOperand(OpndIdx - 1).getReg() == AArch64::SP && "Unexpected base register in callee-save save/restore instruction!" ) ? static_cast<void> (0) : __assert_fail ("MBBI->getOperand(OpndIdx - 1).getReg() == AArch64::SP && \"Unexpected base register in callee-save save/restore instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 706, __PRETTY_FUNCTION__)) | |||
706 | "Unexpected base register in callee-save save/restore instruction!")((MBBI->getOperand(OpndIdx - 1).getReg() == AArch64::SP && "Unexpected base register in callee-save save/restore instruction!" ) ? static_cast<void> (0) : __assert_fail ("MBBI->getOperand(OpndIdx - 1).getReg() == AArch64::SP && \"Unexpected base register in callee-save save/restore instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 706, __PRETTY_FUNCTION__)); | |||
707 | assert(CSStackSizeInc % Scale == 0)((CSStackSizeInc % Scale == 0) ? static_cast<void> (0) : __assert_fail ("CSStackSizeInc % Scale == 0", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 707, __PRETTY_FUNCTION__)); | |||
708 | MIB.addImm(CSStackSizeInc / Scale); | |||
709 | ||||
710 | MIB.setMIFlags(MBBI->getFlags()); | |||
711 | MIB.setMemRefs(MBBI->memoperands()); | |||
712 | ||||
713 | // Generate a new SEH code that corresponds to the new instruction. | |||
714 | if (NeedsWinCFI) { | |||
715 | *HasWinCFI = true; | |||
716 | InsertSEH(*MIB, *TII, | |||
717 | InProlog ? MachineInstr::FrameSetup : MachineInstr::FrameDestroy); | |||
718 | } | |||
719 | ||||
720 | return std::prev(MBB.erase(MBBI)); | |||
721 | } | |||
722 | ||||
723 | // Fixup callee-save register save/restore instructions to take into account | |||
724 | // combined SP bump by adding the local stack size to the stack offsets. | |||
725 | static void fixupCalleeSaveRestoreStackOffset(MachineInstr &MI, | |||
726 | unsigned LocalStackSize, | |||
727 | bool NeedsWinCFI, | |||
728 | bool *HasWinCFI) { | |||
729 | if (AArch64InstrInfo::isSEHInstruction(MI)) | |||
730 | return; | |||
731 | ||||
732 | unsigned Opc = MI.getOpcode(); | |||
733 | ||||
734 | // Ignore instructions that do not operate on SP, i.e. shadow call stack | |||
735 | // instructions and associated CFI instruction. | |||
736 | if (Opc == AArch64::STRXpost || Opc == AArch64::LDRXpre || | |||
737 | Opc == AArch64::CFI_INSTRUCTION) { | |||
738 | if (Opc != AArch64::CFI_INSTRUCTION) | |||
739 | assert(MI.getOperand(0).getReg() != AArch64::SP)((MI.getOperand(0).getReg() != AArch64::SP) ? static_cast< void> (0) : __assert_fail ("MI.getOperand(0).getReg() != AArch64::SP" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 739, __PRETTY_FUNCTION__)); | |||
740 | return; | |||
741 | } | |||
742 | ||||
743 | unsigned Scale; | |||
744 | switch (Opc) { | |||
745 | case AArch64::STPXi: | |||
746 | case AArch64::STRXui: | |||
747 | case AArch64::STPDi: | |||
748 | case AArch64::STRDui: | |||
749 | case AArch64::LDPXi: | |||
750 | case AArch64::LDRXui: | |||
751 | case AArch64::LDPDi: | |||
752 | case AArch64::LDRDui: | |||
753 | Scale = 8; | |||
754 | break; | |||
755 | case AArch64::STPQi: | |||
756 | case AArch64::STRQui: | |||
757 | case AArch64::LDPQi: | |||
758 | case AArch64::LDRQui: | |||
759 | Scale = 16; | |||
760 | break; | |||
761 | default: | |||
762 | llvm_unreachable("Unexpected callee-save save/restore opcode!")::llvm::llvm_unreachable_internal("Unexpected callee-save save/restore opcode!" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 762); | |||
763 | } | |||
764 | ||||
765 | unsigned OffsetIdx = MI.getNumExplicitOperands() - 1; | |||
766 | assert(MI.getOperand(OffsetIdx - 1).getReg() == AArch64::SP &&((MI.getOperand(OffsetIdx - 1).getReg() == AArch64::SP && "Unexpected base register in callee-save save/restore instruction!" ) ? static_cast<void> (0) : __assert_fail ("MI.getOperand(OffsetIdx - 1).getReg() == AArch64::SP && \"Unexpected base register in callee-save save/restore instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 767, __PRETTY_FUNCTION__)) | |||
767 | "Unexpected base register in callee-save save/restore instruction!")((MI.getOperand(OffsetIdx - 1).getReg() == AArch64::SP && "Unexpected base register in callee-save save/restore instruction!" ) ? static_cast<void> (0) : __assert_fail ("MI.getOperand(OffsetIdx - 1).getReg() == AArch64::SP && \"Unexpected base register in callee-save save/restore instruction!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 767, __PRETTY_FUNCTION__)); | |||
768 | // Last operand is immediate offset that needs fixing. | |||
769 | MachineOperand &OffsetOpnd = MI.getOperand(OffsetIdx); | |||
770 | // All generated opcodes have scaled offsets. | |||
771 | assert(LocalStackSize % Scale == 0)((LocalStackSize % Scale == 0) ? static_cast<void> (0) : __assert_fail ("LocalStackSize % Scale == 0", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 771, __PRETTY_FUNCTION__)); | |||
772 | OffsetOpnd.setImm(OffsetOpnd.getImm() + LocalStackSize / Scale); | |||
773 | ||||
774 | if (NeedsWinCFI) { | |||
775 | *HasWinCFI = true; | |||
776 | auto MBBI = std::next(MachineBasicBlock::iterator(MI)); | |||
777 | assert(MBBI != MI.getParent()->end() && "Expecting a valid instruction")((MBBI != MI.getParent()->end() && "Expecting a valid instruction" ) ? static_cast<void> (0) : __assert_fail ("MBBI != MI.getParent()->end() && \"Expecting a valid instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 777, __PRETTY_FUNCTION__)); | |||
778 | assert(AArch64InstrInfo::isSEHInstruction(*MBBI) &&((AArch64InstrInfo::isSEHInstruction(*MBBI) && "Expecting a SEH instruction" ) ? static_cast<void> (0) : __assert_fail ("AArch64InstrInfo::isSEHInstruction(*MBBI) && \"Expecting a SEH instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 779, __PRETTY_FUNCTION__)) | |||
779 | "Expecting a SEH instruction")((AArch64InstrInfo::isSEHInstruction(*MBBI) && "Expecting a SEH instruction" ) ? static_cast<void> (0) : __assert_fail ("AArch64InstrInfo::isSEHInstruction(*MBBI) && \"Expecting a SEH instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 779, __PRETTY_FUNCTION__)); | |||
780 | fixupSEHOpcode(MBBI, LocalStackSize); | |||
781 | } | |||
782 | } | |||
783 | ||||
784 | static void adaptForLdStOpt(MachineBasicBlock &MBB, | |||
785 | MachineBasicBlock::iterator FirstSPPopI, | |||
786 | MachineBasicBlock::iterator LastPopI) { | |||
787 | // Sometimes (when we restore in the same order as we save), we can end up | |||
788 | // with code like this: | |||
789 | // | |||
790 | // ldp x26, x25, [sp] | |||
791 | // ldp x24, x23, [sp, #16] | |||
792 | // ldp x22, x21, [sp, #32] | |||
793 | // ldp x20, x19, [sp, #48] | |||
794 | // add sp, sp, #64 | |||
795 | // | |||
796 | // In this case, it is always better to put the first ldp at the end, so | |||
797 | // that the load-store optimizer can run and merge the ldp and the add into | |||
798 | // a post-index ldp. | |||
799 | // If we managed to grab the first pop instruction, move it to the end. | |||
800 | if (ReverseCSRRestoreSeq) | |||
801 | MBB.splice(FirstSPPopI, &MBB, LastPopI); | |||
802 | // We should end up with something like this now: | |||
803 | // | |||
804 | // ldp x24, x23, [sp, #16] | |||
805 | // ldp x22, x21, [sp, #32] | |||
806 | // ldp x20, x19, [sp, #48] | |||
807 | // ldp x26, x25, [sp] | |||
808 | // add sp, sp, #64 | |||
809 | // | |||
810 | // and the load-store optimizer can merge the last two instructions into: | |||
811 | // | |||
812 | // ldp x26, x25, [sp], #64 | |||
813 | // | |||
814 | } | |||
815 | ||||
816 | static bool ShouldSignWithAKey(MachineFunction &MF) { | |||
817 | const Function &F = MF.getFunction(); | |||
818 | if (!F.hasFnAttribute("sign-return-address-key")) | |||
819 | return true; | |||
820 | ||||
821 | const StringRef Key = | |||
822 | F.getFnAttribute("sign-return-address-key").getValueAsString(); | |||
823 | assert(Key.equals_lower("a_key") || Key.equals_lower("b_key"))((Key.equals_lower("a_key") || Key.equals_lower("b_key")) ? static_cast <void> (0) : __assert_fail ("Key.equals_lower(\"a_key\") || Key.equals_lower(\"b_key\")" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 823, __PRETTY_FUNCTION__)); | |||
824 | return Key.equals_lower("a_key"); | |||
825 | } | |||
826 | ||||
827 | static bool needsWinCFI(const MachineFunction &MF) { | |||
828 | const Function &F = MF.getFunction(); | |||
829 | return MF.getTarget().getMCAsmInfo()->usesWindowsCFI() && | |||
830 | F.needsUnwindTableEntry(); | |||
831 | } | |||
832 | ||||
833 | static bool isTargetDarwin(const MachineFunction &MF) { | |||
834 | return MF.getSubtarget<AArch64Subtarget>().isTargetDarwin(); | |||
835 | } | |||
836 | ||||
837 | // Convenience function to determine whether I is an SVE callee save. | |||
838 | bool IsSVECalleeSave(MachineBasicBlock::iterator I) { | |||
839 | switch (I->getOpcode()) { | |||
840 | default: | |||
841 | return false; | |||
842 | case AArch64::STR_ZXI: | |||
843 | case AArch64::STR_PXI: | |||
844 | case AArch64::LDR_ZXI: | |||
845 | case AArch64::LDR_PXI: | |||
846 | return I->getFlag(MachineInstr::FrameSetup) || | |||
847 | I->getFlag(MachineInstr::FrameDestroy); | |||
848 | } | |||
849 | } | |||
850 | ||||
851 | void AArch64FrameLowering::emitPrologue(MachineFunction &MF, | |||
852 | MachineBasicBlock &MBB) const { | |||
853 | MachineBasicBlock::iterator MBBI = MBB.begin(); | |||
854 | const MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
855 | const Function &F = MF.getFunction(); | |||
856 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
857 | const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo(); | |||
858 | const TargetInstrInfo *TII = Subtarget.getInstrInfo(); | |||
859 | MachineModuleInfo &MMI = MF.getMMI(); | |||
860 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
861 | bool needsFrameMoves = | |||
862 | MF.needsFrameMoves() && !MF.getTarget().getMCAsmInfo()->usesWindowsCFI(); | |||
863 | bool HasFP = hasFP(MF); | |||
864 | bool NeedsWinCFI = needsWinCFI(MF); | |||
865 | bool HasWinCFI = false; | |||
866 | auto Cleanup = make_scope_exit([&]() { MF.setHasWinCFI(HasWinCFI); }); | |||
867 | ||||
868 | bool IsFunclet = MBB.isEHFuncletEntry(); | |||
869 | ||||
870 | // At this point, we're going to decide whether or not the function uses a | |||
871 | // redzone. In most cases, the function doesn't have a redzone so let's | |||
872 | // assume that's false and set it to true in the case that there's a redzone. | |||
873 | AFI->setHasRedZone(false); | |||
874 | ||||
875 | // Debug location must be unknown since the first debug location is used | |||
876 | // to determine the end of the prologue. | |||
877 | DebugLoc DL; | |||
878 | ||||
879 | if (ShouldSignReturnAddress(MF)) { | |||
880 | if (ShouldSignWithAKey(MF)) | |||
881 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::PACIASP)) | |||
882 | .setMIFlag(MachineInstr::FrameSetup); | |||
883 | else { | |||
884 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::EMITBKEY)) | |||
885 | .setMIFlag(MachineInstr::FrameSetup); | |||
886 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::PACIBSP)) | |||
887 | .setMIFlag(MachineInstr::FrameSetup); | |||
888 | } | |||
889 | ||||
890 | unsigned CFIIndex = | |||
891 | MF.addFrameInst(MCCFIInstruction::createNegateRAState(nullptr)); | |||
892 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION)) | |||
893 | .addCFIIndex(CFIIndex) | |||
894 | .setMIFlags(MachineInstr::FrameSetup); | |||
895 | } | |||
896 | ||||
897 | // All calls are tail calls in GHC calling conv, and functions have no | |||
898 | // prologue/epilogue. | |||
899 | if (MF.getFunction().getCallingConv() == CallingConv::GHC) | |||
900 | return; | |||
901 | ||||
902 | // Set tagged base pointer to the bottom of the stack frame. | |||
903 | // Ideally it should match SP value after prologue. | |||
904 | AFI->setTaggedBasePointerOffset(MFI.getStackSize()); | |||
905 | ||||
906 | const StackOffset &SVEStackSize = getSVEStackSize(MF); | |||
907 | ||||
908 | // getStackSize() includes all the locals in its size calculation. We don't | |||
909 | // include these locals when computing the stack size of a funclet, as they | |||
910 | // are allocated in the parent's stack frame and accessed via the frame | |||
911 | // pointer from the funclet. We only save the callee saved registers in the | |||
912 | // funclet, which are really the callee saved registers of the parent | |||
913 | // function, including the funclet. | |||
914 | int NumBytes = IsFunclet ? (int)getWinEHFuncletFrameSize(MF) | |||
915 | : (int)MFI.getStackSize(); | |||
916 | if (!AFI->hasStackFrame() && !windowsRequiresStackProbe(MF, NumBytes)) { | |||
917 | assert(!HasFP && "unexpected function without stack frame but with FP")((!HasFP && "unexpected function without stack frame but with FP" ) ? static_cast<void> (0) : __assert_fail ("!HasFP && \"unexpected function without stack frame but with FP\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 917, __PRETTY_FUNCTION__)); | |||
918 | assert(!SVEStackSize &&((!SVEStackSize && "unexpected function without stack frame but with SVE objects" ) ? static_cast<void> (0) : __assert_fail ("!SVEStackSize && \"unexpected function without stack frame but with SVE objects\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 919, __PRETTY_FUNCTION__)) | |||
919 | "unexpected function without stack frame but with SVE objects")((!SVEStackSize && "unexpected function without stack frame but with SVE objects" ) ? static_cast<void> (0) : __assert_fail ("!SVEStackSize && \"unexpected function without stack frame but with SVE objects\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 919, __PRETTY_FUNCTION__)); | |||
920 | // All of the stack allocation is for locals. | |||
921 | AFI->setLocalStackSize(NumBytes); | |||
922 | if (!NumBytes) | |||
923 | return; | |||
924 | // REDZONE: If the stack size is less than 128 bytes, we don't need | |||
925 | // to actually allocate. | |||
926 | if (canUseRedZone(MF)) { | |||
927 | AFI->setHasRedZone(true); | |||
928 | ++NumRedZoneFunctions; | |||
929 | } else { | |||
930 | emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP, | |||
931 | {-NumBytes, MVT::i8}, TII, MachineInstr::FrameSetup, | |||
932 | false, NeedsWinCFI, &HasWinCFI); | |||
933 | if (!NeedsWinCFI) { | |||
934 | // Label used to tie together the PROLOG_LABEL and the MachineMoves. | |||
935 | MCSymbol *FrameLabel = MMI.getContext().createTempSymbol(); | |||
936 | // Encode the stack size of the leaf function. | |||
937 | unsigned CFIIndex = MF.addFrameInst( | |||
938 | MCCFIInstruction::createDefCfaOffset(FrameLabel, -NumBytes)); | |||
939 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION)) | |||
940 | .addCFIIndex(CFIIndex) | |||
941 | .setMIFlags(MachineInstr::FrameSetup); | |||
942 | } | |||
943 | } | |||
944 | ||||
945 | if (NeedsWinCFI) { | |||
946 | HasWinCFI = true; | |||
947 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_PrologEnd)) | |||
948 | .setMIFlag(MachineInstr::FrameSetup); | |||
949 | } | |||
950 | ||||
951 | return; | |||
952 | } | |||
953 | ||||
954 | bool IsWin64 = | |||
955 | Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()); | |||
956 | // Var args are accounted for in the containing function, so don't | |||
957 | // include them for funclets. | |||
958 | unsigned FixedObject = (IsWin64 && !IsFunclet) ? | |||
959 | alignTo(AFI->getVarArgsGPRSize(), 16) : 0; | |||
960 | ||||
961 | auto PrologueSaveSize = AFI->getCalleeSavedStackSize() + FixedObject; | |||
962 | // All of the remaining stack allocations are for locals. | |||
963 | AFI->setLocalStackSize(NumBytes - PrologueSaveSize); | |||
964 | bool CombineSPBump = shouldCombineCSRLocalStackBump(MF, NumBytes); | |||
965 | if (CombineSPBump) { | |||
966 | assert(!SVEStackSize && "Cannot combine SP bump with SVE")((!SVEStackSize && "Cannot combine SP bump with SVE") ? static_cast<void> (0) : __assert_fail ("!SVEStackSize && \"Cannot combine SP bump with SVE\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 966, __PRETTY_FUNCTION__)); | |||
967 | emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP, | |||
968 | {-NumBytes, MVT::i8}, TII, MachineInstr::FrameSetup, false, | |||
969 | NeedsWinCFI, &HasWinCFI); | |||
970 | NumBytes = 0; | |||
971 | } else if (PrologueSaveSize != 0) { | |||
972 | MBBI = convertCalleeSaveRestoreToSPPrePostIncDec( | |||
973 | MBB, MBBI, DL, TII, -PrologueSaveSize, NeedsWinCFI, &HasWinCFI); | |||
974 | NumBytes -= PrologueSaveSize; | |||
975 | } | |||
976 | assert(NumBytes >= 0 && "Negative stack allocation size!?")((NumBytes >= 0 && "Negative stack allocation size!?" ) ? static_cast<void> (0) : __assert_fail ("NumBytes >= 0 && \"Negative stack allocation size!?\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 976, __PRETTY_FUNCTION__)); | |||
977 | ||||
978 | // Move past the saves of the callee-saved registers, fixing up the offsets | |||
979 | // and pre-inc if we decided to combine the callee-save and local stack | |||
980 | // pointer bump above. | |||
981 | MachineBasicBlock::iterator End = MBB.end(); | |||
982 | while (MBBI != End && MBBI->getFlag(MachineInstr::FrameSetup) && | |||
983 | !IsSVECalleeSave(MBBI)) { | |||
984 | if (CombineSPBump) | |||
985 | fixupCalleeSaveRestoreStackOffset(*MBBI, AFI->getLocalStackSize(), | |||
986 | NeedsWinCFI, &HasWinCFI); | |||
987 | ++MBBI; | |||
988 | } | |||
989 | ||||
990 | // The code below is not applicable to funclets. We have emitted all the SEH | |||
991 | // opcodes that we needed to emit. The FP and BP belong to the containing | |||
992 | // function. | |||
993 | if (IsFunclet) { | |||
994 | if (NeedsWinCFI) { | |||
995 | HasWinCFI = true; | |||
996 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_PrologEnd)) | |||
997 | .setMIFlag(MachineInstr::FrameSetup); | |||
998 | } | |||
999 | ||||
1000 | // SEH funclets are passed the frame pointer in X1. If the parent | |||
1001 | // function uses the base register, then the base register is used | |||
1002 | // directly, and is not retrieved from X1. | |||
1003 | if (F.hasPersonalityFn()) { | |||
1004 | EHPersonality Per = classifyEHPersonality(F.getPersonalityFn()); | |||
1005 | if (isAsynchronousEHPersonality(Per)) { | |||
1006 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::COPY), AArch64::FP) | |||
1007 | .addReg(AArch64::X1).setMIFlag(MachineInstr::FrameSetup); | |||
1008 | MBB.addLiveIn(AArch64::X1); | |||
1009 | } | |||
1010 | } | |||
1011 | ||||
1012 | return; | |||
1013 | } | |||
1014 | ||||
1015 | if (HasFP) { | |||
1016 | // Only set up FP if we actually need to. | |||
1017 | int FPOffset = isTargetDarwin(MF) ? (AFI->getCalleeSavedStackSize() - 16) : 0; | |||
1018 | ||||
1019 | if (CombineSPBump) | |||
1020 | FPOffset += AFI->getLocalStackSize(); | |||
1021 | ||||
1022 | // Issue sub fp, sp, FPOffset or | |||
1023 | // mov fp,sp when FPOffset is zero. | |||
1024 | // Note: All stores of callee-saved registers are marked as "FrameSetup". | |||
1025 | // This code marks the instruction(s) that set the FP also. | |||
1026 | emitFrameOffset(MBB, MBBI, DL, AArch64::FP, AArch64::SP, | |||
1027 | {FPOffset, MVT::i8}, TII, MachineInstr::FrameSetup, false, | |||
1028 | NeedsWinCFI, &HasWinCFI); | |||
1029 | } | |||
1030 | ||||
1031 | if (windowsRequiresStackProbe(MF, NumBytes)) { | |||
1032 | uint32_t NumWords = NumBytes >> 4; | |||
1033 | if (NeedsWinCFI) { | |||
1034 | HasWinCFI = true; | |||
1035 | // alloc_l can hold at most 256MB, so assume that NumBytes doesn't | |||
1036 | // exceed this amount. We need to move at most 2^24 - 1 into x15. | |||
1037 | // This is at most two instructions, MOVZ follwed by MOVK. | |||
1038 | // TODO: Fix to use multiple stack alloc unwind codes for stacks | |||
1039 | // exceeding 256MB in size. | |||
1040 | if (NumBytes >= (1 << 28)) | |||
1041 | report_fatal_error("Stack size cannot exceed 256MB for stack " | |||
1042 | "unwinding purposes"); | |||
1043 | ||||
1044 | uint32_t LowNumWords = NumWords & 0xFFFF; | |||
1045 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::MOVZXi), AArch64::X15) | |||
1046 | .addImm(LowNumWords) | |||
1047 | .addImm(AArch64_AM::getShifterImm(AArch64_AM::LSL, 0)) | |||
1048 | .setMIFlag(MachineInstr::FrameSetup); | |||
1049 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1050 | .setMIFlag(MachineInstr::FrameSetup); | |||
1051 | if ((NumWords & 0xFFFF0000) != 0) { | |||
1052 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::MOVKXi), AArch64::X15) | |||
1053 | .addReg(AArch64::X15) | |||
1054 | .addImm((NumWords & 0xFFFF0000) >> 16) // High half | |||
1055 | .addImm(AArch64_AM::getShifterImm(AArch64_AM::LSL, 16)) | |||
1056 | .setMIFlag(MachineInstr::FrameSetup); | |||
1057 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1058 | .setMIFlag(MachineInstr::FrameSetup); | |||
1059 | } | |||
1060 | } else { | |||
1061 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::MOVi64imm), AArch64::X15) | |||
1062 | .addImm(NumWords) | |||
1063 | .setMIFlags(MachineInstr::FrameSetup); | |||
1064 | } | |||
1065 | ||||
1066 | switch (MF.getTarget().getCodeModel()) { | |||
1067 | case CodeModel::Tiny: | |||
1068 | case CodeModel::Small: | |||
1069 | case CodeModel::Medium: | |||
1070 | case CodeModel::Kernel: | |||
1071 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::BL)) | |||
1072 | .addExternalSymbol("__chkstk") | |||
1073 | .addReg(AArch64::X15, RegState::Implicit) | |||
1074 | .addReg(AArch64::X16, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1075 | .addReg(AArch64::X17, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1076 | .addReg(AArch64::NZCV, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1077 | .setMIFlags(MachineInstr::FrameSetup); | |||
1078 | if (NeedsWinCFI) { | |||
1079 | HasWinCFI = true; | |||
1080 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1081 | .setMIFlag(MachineInstr::FrameSetup); | |||
1082 | } | |||
1083 | break; | |||
1084 | case CodeModel::Large: | |||
1085 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::MOVaddrEXT)) | |||
1086 | .addReg(AArch64::X16, RegState::Define) | |||
1087 | .addExternalSymbol("__chkstk") | |||
1088 | .addExternalSymbol("__chkstk") | |||
1089 | .setMIFlags(MachineInstr::FrameSetup); | |||
1090 | if (NeedsWinCFI) { | |||
1091 | HasWinCFI = true; | |||
1092 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1093 | .setMIFlag(MachineInstr::FrameSetup); | |||
1094 | } | |||
1095 | ||||
1096 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::BLR)) | |||
1097 | .addReg(AArch64::X16, RegState::Kill) | |||
1098 | .addReg(AArch64::X15, RegState::Implicit | RegState::Define) | |||
1099 | .addReg(AArch64::X16, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1100 | .addReg(AArch64::X17, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1101 | .addReg(AArch64::NZCV, RegState::Implicit | RegState::Define | RegState::Dead) | |||
1102 | .setMIFlags(MachineInstr::FrameSetup); | |||
1103 | if (NeedsWinCFI) { | |||
1104 | HasWinCFI = true; | |||
1105 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1106 | .setMIFlag(MachineInstr::FrameSetup); | |||
1107 | } | |||
1108 | break; | |||
1109 | } | |||
1110 | ||||
1111 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SUBXrx64), AArch64::SP) | |||
1112 | .addReg(AArch64::SP, RegState::Kill) | |||
1113 | .addReg(AArch64::X15, RegState::Kill) | |||
1114 | .addImm(AArch64_AM::getArithExtendImm(AArch64_AM::UXTX, 4)) | |||
1115 | .setMIFlags(MachineInstr::FrameSetup); | |||
1116 | if (NeedsWinCFI) { | |||
1117 | HasWinCFI = true; | |||
1118 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_StackAlloc)) | |||
1119 | .addImm(NumBytes) | |||
1120 | .setMIFlag(MachineInstr::FrameSetup); | |||
1121 | } | |||
1122 | NumBytes = 0; | |||
1123 | } | |||
1124 | ||||
1125 | StackOffset AllocateBefore = SVEStackSize, AllocateAfter = {}; | |||
1126 | MachineBasicBlock::iterator CalleeSavesBegin = MBBI, CalleeSavesEnd = MBBI; | |||
1127 | ||||
1128 | // Process the SVE callee-saves to determine what space needs to be | |||
1129 | // allocated. | |||
1130 | if (AFI->getSVECalleeSavedStackSize()) { | |||
1131 | // Find callee save instructions in frame. | |||
1132 | CalleeSavesBegin = MBBI; | |||
1133 | assert(IsSVECalleeSave(CalleeSavesBegin) && "Unexpected instruction")((IsSVECalleeSave(CalleeSavesBegin) && "Unexpected instruction" ) ? static_cast<void> (0) : __assert_fail ("IsSVECalleeSave(CalleeSavesBegin) && \"Unexpected instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1133, __PRETTY_FUNCTION__)); | |||
1134 | while (IsSVECalleeSave(MBBI) && MBBI != MBB.getFirstTerminator()) | |||
1135 | ++MBBI; | |||
1136 | CalleeSavesEnd = MBBI; | |||
1137 | ||||
1138 | int64_t OffsetToFirstCalleeSaveFromSP = | |||
1139 | MFI.getObjectOffset(AFI->getMaxSVECSFrameIndex()); | |||
1140 | StackOffset OffsetToCalleeSavesFromSP = | |||
1141 | StackOffset(OffsetToFirstCalleeSaveFromSP, MVT::nxv1i8) + SVEStackSize; | |||
1142 | AllocateBefore -= OffsetToCalleeSavesFromSP; | |||
1143 | AllocateAfter = SVEStackSize - AllocateBefore; | |||
1144 | } | |||
1145 | ||||
1146 | // Allocate space for the callee saves (if any). | |||
1147 | emitFrameOffset(MBB, CalleeSavesBegin, DL, AArch64::SP, AArch64::SP, | |||
1148 | -AllocateBefore, TII, | |||
1149 | MachineInstr::FrameSetup); | |||
1150 | ||||
1151 | // Finally allocate remaining SVE stack space. | |||
1152 | emitFrameOffset(MBB, CalleeSavesEnd, DL, AArch64::SP, AArch64::SP, | |||
1153 | -AllocateAfter, TII, | |||
1154 | MachineInstr::FrameSetup); | |||
1155 | ||||
1156 | // Allocate space for the rest of the frame. | |||
1157 | if (NumBytes) { | |||
1158 | const bool NeedsRealignment = RegInfo->needsStackRealignment(MF); | |||
1159 | unsigned scratchSPReg = AArch64::SP; | |||
1160 | ||||
1161 | if (NeedsRealignment) { | |||
1162 | scratchSPReg = findScratchNonCalleeSaveRegister(&MBB); | |||
1163 | assert(scratchSPReg != AArch64::NoRegister)((scratchSPReg != AArch64::NoRegister) ? static_cast<void> (0) : __assert_fail ("scratchSPReg != AArch64::NoRegister", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1163, __PRETTY_FUNCTION__)); | |||
1164 | } | |||
1165 | ||||
1166 | // If we're a leaf function, try using the red zone. | |||
1167 | if (!canUseRedZone(MF)) | |||
1168 | // FIXME: in the case of dynamic re-alignment, NumBytes doesn't have | |||
1169 | // the correct value here, as NumBytes also includes padding bytes, | |||
1170 | // which shouldn't be counted here. | |||
1171 | emitFrameOffset(MBB, MBBI, DL, scratchSPReg, AArch64::SP, | |||
1172 | {-NumBytes, MVT::i8}, TII, MachineInstr::FrameSetup, | |||
1173 | false, NeedsWinCFI, &HasWinCFI); | |||
1174 | ||||
1175 | if (NeedsRealignment) { | |||
1176 | const unsigned Alignment = MFI.getMaxAlignment(); | |||
1177 | const unsigned NrBitsToZero = countTrailingZeros(Alignment); | |||
1178 | assert(NrBitsToZero > 1)((NrBitsToZero > 1) ? static_cast<void> (0) : __assert_fail ("NrBitsToZero > 1", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1178, __PRETTY_FUNCTION__)); | |||
1179 | assert(scratchSPReg != AArch64::SP)((scratchSPReg != AArch64::SP) ? static_cast<void> (0) : __assert_fail ("scratchSPReg != AArch64::SP", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1179, __PRETTY_FUNCTION__)); | |||
1180 | ||||
1181 | // SUB X9, SP, NumBytes | |||
1182 | // -- X9 is temporary register, so shouldn't contain any live data here, | |||
1183 | // -- free to use. This is already produced by emitFrameOffset above. | |||
1184 | // AND SP, X9, 0b11111...0000 | |||
1185 | // The logical immediates have a non-trivial encoding. The following | |||
1186 | // formula computes the encoded immediate with all ones but | |||
1187 | // NrBitsToZero zero bits as least significant bits. | |||
1188 | uint32_t andMaskEncoded = (1 << 12) // = N | |||
1189 | | ((64 - NrBitsToZero) << 6) // immr | |||
1190 | | ((64 - NrBitsToZero - 1) << 0); // imms | |||
1191 | ||||
1192 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::ANDXri), AArch64::SP) | |||
1193 | .addReg(scratchSPReg, RegState::Kill) | |||
1194 | .addImm(andMaskEncoded); | |||
1195 | AFI->setStackRealigned(true); | |||
1196 | if (NeedsWinCFI) { | |||
1197 | HasWinCFI = true; | |||
1198 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_StackAlloc)) | |||
1199 | .addImm(NumBytes & andMaskEncoded) | |||
1200 | .setMIFlag(MachineInstr::FrameSetup); | |||
1201 | } | |||
1202 | } | |||
1203 | } | |||
1204 | ||||
1205 | // If we need a base pointer, set it up here. It's whatever the value of the | |||
1206 | // stack pointer is at this point. Any variable size objects will be allocated | |||
1207 | // after this, so we can still use the base pointer to reference locals. | |||
1208 | // | |||
1209 | // FIXME: Clarify FrameSetup flags here. | |||
1210 | // Note: Use emitFrameOffset() like above for FP if the FrameSetup flag is | |||
1211 | // needed. | |||
1212 | if (RegInfo->hasBasePointer(MF)) { | |||
1213 | TII->copyPhysReg(MBB, MBBI, DL, RegInfo->getBaseRegister(), AArch64::SP, | |||
1214 | false); | |||
1215 | if (NeedsWinCFI) { | |||
1216 | HasWinCFI = true; | |||
1217 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_Nop)) | |||
1218 | .setMIFlag(MachineInstr::FrameSetup); | |||
1219 | } | |||
1220 | } | |||
1221 | ||||
1222 | // The very last FrameSetup instruction indicates the end of prologue. Emit a | |||
1223 | // SEH opcode indicating the prologue end. | |||
1224 | if (NeedsWinCFI && HasWinCFI) { | |||
1225 | BuildMI(MBB, MBBI, DL, TII->get(AArch64::SEH_PrologEnd)) | |||
1226 | .setMIFlag(MachineInstr::FrameSetup); | |||
1227 | } | |||
1228 | ||||
1229 | if (needsFrameMoves) { | |||
1230 | const DataLayout &TD = MF.getDataLayout(); | |||
1231 | const int StackGrowth = isTargetDarwin(MF) | |||
1232 | ? (2 * -TD.getPointerSize(0)) | |||
1233 | : -AFI->getCalleeSavedStackSize(); | |||
1234 | Register FramePtr = RegInfo->getFrameRegister(MF); | |||
1235 | // An example of the prologue: | |||
1236 | // | |||
1237 | // .globl __foo | |||
1238 | // .align 2 | |||
1239 | // __foo: | |||
1240 | // Ltmp0: | |||
1241 | // .cfi_startproc | |||
1242 | // .cfi_personality 155, ___gxx_personality_v0 | |||
1243 | // Leh_func_begin: | |||
1244 | // .cfi_lsda 16, Lexception33 | |||
1245 | // | |||
1246 | // stp xa,bx, [sp, -#offset]! | |||
1247 | // ... | |||
1248 | // stp x28, x27, [sp, #offset-32] | |||
1249 | // stp fp, lr, [sp, #offset-16] | |||
1250 | // add fp, sp, #offset - 16 | |||
1251 | // sub sp, sp, #1360 | |||
1252 | // | |||
1253 | // The Stack: | |||
1254 | // +-------------------------------------------+ | |||
1255 | // 10000 | ........ | ........ | ........ | ........ | | |||
1256 | // 10004 | ........ | ........ | ........ | ........ | | |||
1257 | // +-------------------------------------------+ | |||
1258 | // 10008 | ........ | ........ | ........ | ........ | | |||
1259 | // 1000c | ........ | ........ | ........ | ........ | | |||
1260 | // +===========================================+ | |||
1261 | // 10010 | X28 Register | | |||
1262 | // 10014 | X28 Register | | |||
1263 | // +-------------------------------------------+ | |||
1264 | // 10018 | X27 Register | | |||
1265 | // 1001c | X27 Register | | |||
1266 | // +===========================================+ | |||
1267 | // 10020 | Frame Pointer | | |||
1268 | // 10024 | Frame Pointer | | |||
1269 | // +-------------------------------------------+ | |||
1270 | // 10028 | Link Register | | |||
1271 | // 1002c | Link Register | | |||
1272 | // +===========================================+ | |||
1273 | // 10030 | ........ | ........ | ........ | ........ | | |||
1274 | // 10034 | ........ | ........ | ........ | ........ | | |||
1275 | // +-------------------------------------------+ | |||
1276 | // 10038 | ........ | ........ | ........ | ........ | | |||
1277 | // 1003c | ........ | ........ | ........ | ........ | | |||
1278 | // +-------------------------------------------+ | |||
1279 | // | |||
1280 | // [sp] = 10030 :: >>initial value<< | |||
1281 | // sp = 10020 :: stp fp, lr, [sp, #-16]! | |||
1282 | // fp = sp == 10020 :: mov fp, sp | |||
1283 | // [sp] == 10020 :: stp x28, x27, [sp, #-16]! | |||
1284 | // sp == 10010 :: >>final value<< | |||
1285 | // | |||
1286 | // The frame pointer (w29) points to address 10020. If we use an offset of | |||
1287 | // '16' from 'w29', we get the CFI offsets of -8 for w30, -16 for w29, -24 | |||
1288 | // for w27, and -32 for w28: | |||
1289 | // | |||
1290 | // Ltmp1: | |||
1291 | // .cfi_def_cfa w29, 16 | |||
1292 | // Ltmp2: | |||
1293 | // .cfi_offset w30, -8 | |||
1294 | // Ltmp3: | |||
1295 | // .cfi_offset w29, -16 | |||
1296 | // Ltmp4: | |||
1297 | // .cfi_offset w27, -24 | |||
1298 | // Ltmp5: | |||
1299 | // .cfi_offset w28, -32 | |||
1300 | ||||
1301 | if (HasFP) { | |||
1302 | // Define the current CFA rule to use the provided FP. | |||
1303 | unsigned Reg = RegInfo->getDwarfRegNum(FramePtr, true); | |||
1304 | unsigned CFIIndex = MF.addFrameInst(MCCFIInstruction::createDefCfa( | |||
1305 | nullptr, Reg, StackGrowth - FixedObject)); | |||
1306 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION)) | |||
1307 | .addCFIIndex(CFIIndex) | |||
1308 | .setMIFlags(MachineInstr::FrameSetup); | |||
1309 | } else { | |||
1310 | // Encode the stack size of the leaf function. | |||
1311 | unsigned CFIIndex = MF.addFrameInst( | |||
1312 | MCCFIInstruction::createDefCfaOffset(nullptr, -MFI.getStackSize())); | |||
1313 | BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION)) | |||
1314 | .addCFIIndex(CFIIndex) | |||
1315 | .setMIFlags(MachineInstr::FrameSetup); | |||
1316 | } | |||
1317 | ||||
1318 | // Now emit the moves for whatever callee saved regs we have (including FP, | |||
1319 | // LR if those are saved). | |||
1320 | emitCalleeSavedFrameMoves(MBB, MBBI); | |||
1321 | } | |||
1322 | } | |||
1323 | ||||
1324 | static void InsertReturnAddressAuth(MachineFunction &MF, | |||
1325 | MachineBasicBlock &MBB) { | |||
1326 | if (!ShouldSignReturnAddress(MF)) | |||
1327 | return; | |||
1328 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
1329 | const TargetInstrInfo *TII = Subtarget.getInstrInfo(); | |||
1330 | ||||
1331 | MachineBasicBlock::iterator MBBI = MBB.getFirstTerminator(); | |||
1332 | DebugLoc DL; | |||
1333 | if (MBBI != MBB.end()) | |||
1334 | DL = MBBI->getDebugLoc(); | |||
1335 | ||||
1336 | // The AUTIASP instruction assembles to a hint instruction before v8.3a so | |||
1337 | // this instruction can safely used for any v8a architecture. | |||
1338 | // From v8.3a onwards there are optimised authenticate LR and return | |||
1339 | // instructions, namely RETA{A,B}, that can be used instead. | |||
1340 | if (Subtarget.hasV8_3aOps() && MBBI != MBB.end() && | |||
1341 | MBBI->getOpcode() == AArch64::RET_ReallyLR) { | |||
1342 | BuildMI(MBB, MBBI, DL, | |||
1343 | TII->get(ShouldSignWithAKey(MF) ? AArch64::RETAA : AArch64::RETAB)) | |||
1344 | .copyImplicitOps(*MBBI); | |||
1345 | MBB.erase(MBBI); | |||
1346 | } else { | |||
1347 | BuildMI( | |||
1348 | MBB, MBBI, DL, | |||
1349 | TII->get(ShouldSignWithAKey(MF) ? AArch64::AUTIASP : AArch64::AUTIBSP)) | |||
1350 | .setMIFlag(MachineInstr::FrameDestroy); | |||
1351 | } | |||
1352 | } | |||
1353 | ||||
1354 | static bool isFuncletReturnInstr(const MachineInstr &MI) { | |||
1355 | switch (MI.getOpcode()) { | |||
1356 | default: | |||
1357 | return false; | |||
1358 | case AArch64::CATCHRET: | |||
1359 | case AArch64::CLEANUPRET: | |||
1360 | return true; | |||
1361 | } | |||
1362 | } | |||
1363 | ||||
1364 | void AArch64FrameLowering::emitEpilogue(MachineFunction &MF, | |||
1365 | MachineBasicBlock &MBB) const { | |||
1366 | MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr(); | |||
1367 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
1368 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
1369 | const TargetInstrInfo *TII = Subtarget.getInstrInfo(); | |||
1370 | DebugLoc DL; | |||
1371 | bool IsTailCallReturn = false; | |||
1372 | bool NeedsWinCFI = needsWinCFI(MF); | |||
1373 | bool HasWinCFI = false; | |||
1374 | bool IsFunclet = false; | |||
1375 | auto WinCFI = make_scope_exit([&]() { | |||
1376 | if (!MF.hasWinCFI()) | |||
1377 | MF.setHasWinCFI(HasWinCFI); | |||
1378 | }); | |||
1379 | ||||
1380 | if (MBB.end() != MBBI) { | |||
1381 | DL = MBBI->getDebugLoc(); | |||
1382 | unsigned RetOpcode = MBBI->getOpcode(); | |||
1383 | IsTailCallReturn = RetOpcode == AArch64::TCRETURNdi || | |||
1384 | RetOpcode == AArch64::TCRETURNri || | |||
1385 | RetOpcode == AArch64::TCRETURNriBTI; | |||
1386 | IsFunclet = isFuncletReturnInstr(*MBBI); | |||
1387 | } | |||
1388 | ||||
1389 | int NumBytes = IsFunclet ? (int)getWinEHFuncletFrameSize(MF) | |||
1390 | : MFI.getStackSize(); | |||
1391 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
1392 | ||||
1393 | // All calls are tail calls in GHC calling conv, and functions have no | |||
1394 | // prologue/epilogue. | |||
1395 | if (MF.getFunction().getCallingConv() == CallingConv::GHC) | |||
1396 | return; | |||
1397 | ||||
1398 | // Initial and residual are named for consistency with the prologue. Note that | |||
1399 | // in the epilogue, the residual adjustment is executed first. | |||
1400 | uint64_t ArgumentPopSize = 0; | |||
1401 | if (IsTailCallReturn) { | |||
1402 | MachineOperand &StackAdjust = MBBI->getOperand(1); | |||
1403 | ||||
1404 | // For a tail-call in a callee-pops-arguments environment, some or all of | |||
1405 | // the stack may actually be in use for the call's arguments, this is | |||
1406 | // calculated during LowerCall and consumed here... | |||
1407 | ArgumentPopSize = StackAdjust.getImm(); | |||
1408 | } else { | |||
1409 | // ... otherwise the amount to pop is *all* of the argument space, | |||
1410 | // conveniently stored in the MachineFunctionInfo by | |||
1411 | // LowerFormalArguments. This will, of course, be zero for the C calling | |||
1412 | // convention. | |||
1413 | ArgumentPopSize = AFI->getArgumentStackToRestore(); | |||
1414 | } | |||
1415 | ||||
1416 | // The stack frame should be like below, | |||
1417 | // | |||
1418 | // ---------------------- --- | |||
1419 | // | | | | |||
1420 | // | BytesInStackArgArea| CalleeArgStackSize | |||
1421 | // | (NumReusableBytes) | (of tail call) | |||
1422 | // | | --- | |||
1423 | // | | | | |||
1424 | // ---------------------| --- | | |||
1425 | // | | | | | |||
1426 | // | CalleeSavedReg | | | | |||
1427 | // | (CalleeSavedStackSize)| | | | |||
1428 | // | | | | | |||
1429 | // ---------------------| | NumBytes | |||
1430 | // | | StackSize (StackAdjustUp) | |||
1431 | // | LocalStackSize | | | | |||
1432 | // | (covering callee | | | | |||
1433 | // | args) | | | | |||
1434 | // | | | | | |||
1435 | // ---------------------- --- --- | |||
1436 | // | |||
1437 | // So NumBytes = StackSize + BytesInStackArgArea - CalleeArgStackSize | |||
1438 | // = StackSize + ArgumentPopSize | |||
1439 | // | |||
1440 | // AArch64TargetLowering::LowerCall figures out ArgumentPopSize and keeps | |||
1441 | // it as the 2nd argument of AArch64ISD::TC_RETURN. | |||
1442 | ||||
1443 | auto Cleanup = make_scope_exit([&] { InsertReturnAddressAuth(MF, MBB); }); | |||
1444 | ||||
1445 | bool IsWin64 = | |||
1446 | Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()); | |||
1447 | // Var args are accounted for in the containing function, so don't | |||
1448 | // include them for funclets. | |||
1449 | unsigned FixedObject = | |||
1450 | (IsWin64 && !IsFunclet) ? alignTo(AFI->getVarArgsGPRSize(), 16) : 0; | |||
1451 | ||||
1452 | uint64_t AfterCSRPopSize = ArgumentPopSize; | |||
1453 | auto PrologueSaveSize = AFI->getCalleeSavedStackSize() + FixedObject; | |||
1454 | // We cannot rely on the local stack size set in emitPrologue if the function | |||
1455 | // has funclets, as funclets have different local stack size requirements, and | |||
1456 | // the current value set in emitPrologue may be that of the containing | |||
1457 | // function. | |||
1458 | if (MF.hasEHFunclets()) | |||
1459 | AFI->setLocalStackSize(NumBytes - PrologueSaveSize); | |||
1460 | bool CombineSPBump = shouldCombineCSRLocalStackBump(MF, NumBytes); | |||
1461 | // Assume we can't combine the last pop with the sp restore. | |||
1462 | ||||
1463 | if (!CombineSPBump && PrologueSaveSize != 0) { | |||
1464 | MachineBasicBlock::iterator Pop = std::prev(MBB.getFirstTerminator()); | |||
1465 | while (AArch64InstrInfo::isSEHInstruction(*Pop)) | |||
1466 | Pop = std::prev(Pop); | |||
1467 | // Converting the last ldp to a post-index ldp is valid only if the last | |||
1468 | // ldp's offset is 0. | |||
1469 | const MachineOperand &OffsetOp = Pop->getOperand(Pop->getNumOperands() - 1); | |||
1470 | // If the offset is 0, convert it to a post-index ldp. | |||
1471 | if (OffsetOp.getImm() == 0) | |||
1472 | convertCalleeSaveRestoreToSPPrePostIncDec( | |||
1473 | MBB, Pop, DL, TII, PrologueSaveSize, NeedsWinCFI, &HasWinCFI, false); | |||
1474 | else { | |||
1475 | // If not, make sure to emit an add after the last ldp. | |||
1476 | // We're doing this by transfering the size to be restored from the | |||
1477 | // adjustment *before* the CSR pops to the adjustment *after* the CSR | |||
1478 | // pops. | |||
1479 | AfterCSRPopSize += PrologueSaveSize; | |||
1480 | } | |||
1481 | } | |||
1482 | ||||
1483 | // Move past the restores of the callee-saved registers. | |||
1484 | // If we plan on combining the sp bump of the local stack size and the callee | |||
1485 | // save stack size, we might need to adjust the CSR save and restore offsets. | |||
1486 | MachineBasicBlock::iterator LastPopI = MBB.getFirstTerminator(); | |||
1487 | MachineBasicBlock::iterator Begin = MBB.begin(); | |||
1488 | while (LastPopI != Begin) { | |||
1489 | --LastPopI; | |||
1490 | if (!LastPopI->getFlag(MachineInstr::FrameDestroy) || | |||
1491 | IsSVECalleeSave(LastPopI)) { | |||
1492 | ++LastPopI; | |||
1493 | break; | |||
1494 | } else if (CombineSPBump) | |||
1495 | fixupCalleeSaveRestoreStackOffset(*LastPopI, AFI->getLocalStackSize(), | |||
1496 | NeedsWinCFI, &HasWinCFI); | |||
1497 | } | |||
1498 | ||||
1499 | if (NeedsWinCFI) { | |||
1500 | HasWinCFI = true; | |||
1501 | BuildMI(MBB, LastPopI, DL, TII->get(AArch64::SEH_EpilogStart)) | |||
1502 | .setMIFlag(MachineInstr::FrameDestroy); | |||
1503 | } | |||
1504 | ||||
1505 | const StackOffset &SVEStackSize = getSVEStackSize(MF); | |||
1506 | ||||
1507 | // If there is a single SP update, insert it before the ret and we're done. | |||
1508 | if (CombineSPBump) { | |||
1509 | assert(!SVEStackSize && "Cannot combine SP bump with SVE")((!SVEStackSize && "Cannot combine SP bump with SVE") ? static_cast<void> (0) : __assert_fail ("!SVEStackSize && \"Cannot combine SP bump with SVE\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1509, __PRETTY_FUNCTION__)); | |||
1510 | emitFrameOffset(MBB, MBB.getFirstTerminator(), DL, AArch64::SP, AArch64::SP, | |||
1511 | {NumBytes + (int64_t)AfterCSRPopSize, MVT::i8}, TII, | |||
1512 | MachineInstr::FrameDestroy, false, NeedsWinCFI, &HasWinCFI); | |||
1513 | if (NeedsWinCFI && HasWinCFI) | |||
1514 | BuildMI(MBB, MBB.getFirstTerminator(), DL, | |||
1515 | TII->get(AArch64::SEH_EpilogEnd)) | |||
1516 | .setMIFlag(MachineInstr::FrameDestroy); | |||
1517 | return; | |||
1518 | } | |||
1519 | ||||
1520 | NumBytes -= PrologueSaveSize; | |||
1521 | assert(NumBytes >= 0 && "Negative stack allocation size!?")((NumBytes >= 0 && "Negative stack allocation size!?" ) ? static_cast<void> (0) : __assert_fail ("NumBytes >= 0 && \"Negative stack allocation size!?\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1521, __PRETTY_FUNCTION__)); | |||
1522 | ||||
1523 | // Process the SVE callee-saves to determine what space needs to be | |||
1524 | // deallocated. | |||
1525 | StackOffset DeallocateBefore = {}, DeallocateAfter = SVEStackSize; | |||
1526 | MachineBasicBlock::iterator RestoreBegin = LastPopI, RestoreEnd = LastPopI; | |||
1527 | if (AFI->getSVECalleeSavedStackSize()) { | |||
1528 | RestoreBegin = std::prev(RestoreEnd);; | |||
1529 | while (IsSVECalleeSave(RestoreBegin) && | |||
1530 | RestoreBegin != MBB.begin()) | |||
1531 | --RestoreBegin; | |||
1532 | ++RestoreBegin; | |||
1533 | ||||
1534 | assert(IsSVECalleeSave(RestoreBegin) &&((IsSVECalleeSave(RestoreBegin) && IsSVECalleeSave(std ::prev(RestoreEnd)) && "Unexpected instruction") ? static_cast <void> (0) : __assert_fail ("IsSVECalleeSave(RestoreBegin) && IsSVECalleeSave(std::prev(RestoreEnd)) && \"Unexpected instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1535, __PRETTY_FUNCTION__)) | |||
1535 | IsSVECalleeSave(std::prev(RestoreEnd)) && "Unexpected instruction")((IsSVECalleeSave(RestoreBegin) && IsSVECalleeSave(std ::prev(RestoreEnd)) && "Unexpected instruction") ? static_cast <void> (0) : __assert_fail ("IsSVECalleeSave(RestoreBegin) && IsSVECalleeSave(std::prev(RestoreEnd)) && \"Unexpected instruction\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1535, __PRETTY_FUNCTION__)); | |||
1536 | ||||
1537 | int64_t OffsetToFirstCalleeSaveFromSP = | |||
1538 | MFI.getObjectOffset(AFI->getMaxSVECSFrameIndex()); | |||
1539 | StackOffset OffsetToCalleeSavesFromSP = | |||
1540 | StackOffset(OffsetToFirstCalleeSaveFromSP, MVT::nxv1i8) + SVEStackSize; | |||
1541 | DeallocateBefore = OffsetToCalleeSavesFromSP; | |||
1542 | DeallocateAfter = SVEStackSize - DeallocateBefore; | |||
1543 | } | |||
1544 | ||||
1545 | // Deallocate the SVE area. | |||
1546 | if (SVEStackSize) { | |||
1547 | if (AFI->isStackRealigned()) { | |||
1548 | if (AFI->getSVECalleeSavedStackSize()) | |||
1549 | // Set SP to start of SVE area, from which the callee-save reloads | |||
1550 | // can be done. The code below will deallocate the stack space | |||
1551 | // space by moving FP -> SP. | |||
1552 | emitFrameOffset(MBB, RestoreBegin, DL, AArch64::SP, AArch64::FP, | |||
1553 | -SVEStackSize, TII, MachineInstr::FrameDestroy); | |||
1554 | } else { | |||
1555 | if (AFI->getSVECalleeSavedStackSize()) { | |||
1556 | // Deallocate the non-SVE locals first before we can deallocate (and | |||
1557 | // restore callee saves) from the SVE area. | |||
1558 | emitFrameOffset(MBB, RestoreBegin, DL, AArch64::SP, AArch64::SP, | |||
1559 | {NumBytes, MVT::i8}, TII, MachineInstr::FrameDestroy); | |||
1560 | NumBytes = 0; | |||
1561 | } | |||
1562 | ||||
1563 | emitFrameOffset(MBB, RestoreBegin, DL, AArch64::SP, AArch64::SP, | |||
1564 | DeallocateBefore, TII, MachineInstr::FrameDestroy); | |||
1565 | ||||
1566 | emitFrameOffset(MBB, RestoreEnd, DL, AArch64::SP, AArch64::SP, | |||
1567 | DeallocateAfter, TII, MachineInstr::FrameDestroy); | |||
1568 | } | |||
1569 | } | |||
1570 | ||||
1571 | if (!hasFP(MF)) { | |||
1572 | bool RedZone = canUseRedZone(MF); | |||
1573 | // If this was a redzone leaf function, we don't need to restore the | |||
1574 | // stack pointer (but we may need to pop stack args for fastcc). | |||
1575 | if (RedZone && AfterCSRPopSize == 0) | |||
1576 | return; | |||
1577 | ||||
1578 | bool NoCalleeSaveRestore = PrologueSaveSize == 0; | |||
1579 | int StackRestoreBytes = RedZone ? 0 : NumBytes; | |||
1580 | if (NoCalleeSaveRestore) | |||
1581 | StackRestoreBytes += AfterCSRPopSize; | |||
1582 | ||||
1583 | // If we were able to combine the local stack pop with the argument pop, | |||
1584 | // then we're done. | |||
1585 | bool Done = NoCalleeSaveRestore || AfterCSRPopSize == 0; | |||
1586 | ||||
1587 | // If we're done after this, make sure to help the load store optimizer. | |||
1588 | if (Done) | |||
1589 | adaptForLdStOpt(MBB, MBB.getFirstTerminator(), LastPopI); | |||
1590 | ||||
1591 | emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::SP, | |||
1592 | {StackRestoreBytes, MVT::i8}, TII, | |||
1593 | MachineInstr::FrameDestroy, false, NeedsWinCFI, &HasWinCFI); | |||
1594 | if (Done) { | |||
1595 | if (NeedsWinCFI) { | |||
1596 | HasWinCFI = true; | |||
1597 | BuildMI(MBB, MBB.getFirstTerminator(), DL, | |||
1598 | TII->get(AArch64::SEH_EpilogEnd)) | |||
1599 | .setMIFlag(MachineInstr::FrameDestroy); | |||
1600 | } | |||
1601 | return; | |||
1602 | } | |||
1603 | ||||
1604 | NumBytes = 0; | |||
1605 | } | |||
1606 | ||||
1607 | // Restore the original stack pointer. | |||
1608 | // FIXME: Rather than doing the math here, we should instead just use | |||
1609 | // non-post-indexed loads for the restores if we aren't actually going to | |||
1610 | // be able to save any instructions. | |||
1611 | if (!IsFunclet && (MFI.hasVarSizedObjects() || AFI->isStackRealigned())) { | |||
1612 | int64_t OffsetToFrameRecord = | |||
1613 | isTargetDarwin(MF) ? (-(int64_t)AFI->getCalleeSavedStackSize() + 16) : 0; | |||
1614 | emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::FP, | |||
1615 | {OffsetToFrameRecord, MVT::i8}, | |||
1616 | TII, MachineInstr::FrameDestroy, false, NeedsWinCFI); | |||
1617 | } else if (NumBytes) | |||
1618 | emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::SP, | |||
1619 | {NumBytes, MVT::i8}, TII, MachineInstr::FrameDestroy, false, | |||
1620 | NeedsWinCFI); | |||
1621 | ||||
1622 | // This must be placed after the callee-save restore code because that code | |||
1623 | // assumes the SP is at the same location as it was after the callee-save save | |||
1624 | // code in the prologue. | |||
1625 | if (AfterCSRPopSize) { | |||
1626 | // Find an insertion point for the first ldp so that it goes before the | |||
1627 | // shadow call stack epilog instruction. This ensures that the restore of | |||
1628 | // lr from x18 is placed after the restore from sp. | |||
1629 | auto FirstSPPopI = MBB.getFirstTerminator(); | |||
1630 | while (FirstSPPopI != Begin) { | |||
1631 | auto Prev = std::prev(FirstSPPopI); | |||
1632 | if (Prev->getOpcode() != AArch64::LDRXpre || | |||
1633 | Prev->getOperand(0).getReg() == AArch64::SP) | |||
1634 | break; | |||
1635 | FirstSPPopI = Prev; | |||
1636 | } | |||
1637 | ||||
1638 | adaptForLdStOpt(MBB, FirstSPPopI, LastPopI); | |||
1639 | ||||
1640 | emitFrameOffset(MBB, FirstSPPopI, DL, AArch64::SP, AArch64::SP, | |||
1641 | {(int64_t)AfterCSRPopSize, MVT::i8}, TII, | |||
1642 | MachineInstr::FrameDestroy, false, NeedsWinCFI, &HasWinCFI); | |||
1643 | } | |||
1644 | if (NeedsWinCFI && HasWinCFI) | |||
1645 | BuildMI(MBB, MBB.getFirstTerminator(), DL, TII->get(AArch64::SEH_EpilogEnd)) | |||
1646 | .setMIFlag(MachineInstr::FrameDestroy); | |||
1647 | ||||
1648 | MF.setHasWinCFI(HasWinCFI); | |||
1649 | } | |||
1650 | ||||
1651 | /// getFrameIndexReference - Provide a base+offset reference to an FI slot for | |||
1652 | /// debug info. It's the same as what we use for resolving the code-gen | |||
1653 | /// references for now. FIXME: This can go wrong when references are | |||
1654 | /// SP-relative and simple call frames aren't used. | |||
1655 | int AArch64FrameLowering::getFrameIndexReference(const MachineFunction &MF, | |||
1656 | int FI, | |||
1657 | unsigned &FrameReg) const { | |||
1658 | return resolveFrameIndexReference( | |||
1659 | MF, FI, FrameReg, | |||
1660 | /*PreferFP=*/ | |||
1661 | MF.getFunction().hasFnAttribute(Attribute::SanitizeHWAddress), | |||
1662 | /*ForSimm=*/false) | |||
1663 | .getBytes(); | |||
1664 | } | |||
1665 | ||||
1666 | int AArch64FrameLowering::getNonLocalFrameIndexReference( | |||
1667 | const MachineFunction &MF, int FI) const { | |||
1668 | return getSEHFrameIndexOffset(MF, FI); | |||
1669 | } | |||
1670 | ||||
1671 | static StackOffset getFPOffset(const MachineFunction &MF, int ObjectOffset) { | |||
1672 | const auto *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
1673 | const auto &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
1674 | bool IsWin64 = | |||
1675 | Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()); | |||
1676 | unsigned FixedObject = IsWin64 ? alignTo(AFI->getVarArgsGPRSize(), 16) : 0; | |||
1677 | unsigned FPAdjust = isTargetDarwin(MF) | |||
1678 | ? 16 : AFI->getCalleeSavedStackSize(MF.getFrameInfo()); | |||
1679 | return {ObjectOffset + FixedObject + FPAdjust, MVT::i8}; | |||
1680 | } | |||
1681 | ||||
1682 | static StackOffset getStackOffset(const MachineFunction &MF, int ObjectOffset) { | |||
1683 | const auto &MFI = MF.getFrameInfo(); | |||
1684 | return {ObjectOffset + (int)MFI.getStackSize(), MVT::i8}; | |||
1685 | } | |||
1686 | ||||
1687 | int AArch64FrameLowering::getSEHFrameIndexOffset(const MachineFunction &MF, | |||
1688 | int FI) const { | |||
1689 | const auto *RegInfo = static_cast<const AArch64RegisterInfo *>( | |||
1690 | MF.getSubtarget().getRegisterInfo()); | |||
1691 | int ObjectOffset = MF.getFrameInfo().getObjectOffset(FI); | |||
1692 | return RegInfo->getLocalAddressRegister(MF) == AArch64::FP | |||
1693 | ? getFPOffset(MF, ObjectOffset).getBytes() | |||
1694 | : getStackOffset(MF, ObjectOffset).getBytes(); | |||
1695 | } | |||
1696 | ||||
1697 | StackOffset AArch64FrameLowering::resolveFrameIndexReference( | |||
1698 | const MachineFunction &MF, int FI, unsigned &FrameReg, bool PreferFP, | |||
1699 | bool ForSimm) const { | |||
1700 | const auto &MFI = MF.getFrameInfo(); | |||
1701 | int ObjectOffset = MFI.getObjectOffset(FI); | |||
1702 | bool isFixed = MFI.isFixedObjectIndex(FI); | |||
1703 | bool isSVE = MFI.getStackID(FI) == TargetStackID::SVEVector; | |||
1704 | return resolveFrameOffsetReference(MF, ObjectOffset, isFixed, isSVE, FrameReg, | |||
1705 | PreferFP, ForSimm); | |||
1706 | } | |||
1707 | ||||
1708 | StackOffset AArch64FrameLowering::resolveFrameOffsetReference( | |||
1709 | const MachineFunction &MF, int ObjectOffset, bool isFixed, bool isSVE, | |||
1710 | unsigned &FrameReg, bool PreferFP, bool ForSimm) const { | |||
1711 | const auto &MFI = MF.getFrameInfo(); | |||
1712 | const auto *RegInfo = static_cast<const AArch64RegisterInfo *>( | |||
1713 | MF.getSubtarget().getRegisterInfo()); | |||
1714 | const auto *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
1715 | const auto &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
1716 | ||||
1717 | int FPOffset = getFPOffset(MF, ObjectOffset).getBytes(); | |||
1718 | int Offset = getStackOffset(MF, ObjectOffset).getBytes(); | |||
1719 | bool isCSR = | |||
1720 | !isFixed && ObjectOffset >= -((int)AFI->getCalleeSavedStackSize(MFI)); | |||
1721 | ||||
1722 | const StackOffset &SVEStackSize = getSVEStackSize(MF); | |||
1723 | ||||
1724 | // Use frame pointer to reference fixed objects. Use it for locals if | |||
1725 | // there are VLAs or a dynamically realigned SP (and thus the SP isn't | |||
1726 | // reliable as a base). Make sure useFPForScavengingIndex() does the | |||
1727 | // right thing for the emergency spill slot. | |||
1728 | bool UseFP = false; | |||
1729 | if (AFI->hasStackFrame() && !isSVE) { | |||
1730 | // We shouldn't prefer using the FP when there is an SVE area | |||
1731 | // in between the FP and the non-SVE locals/spills. | |||
1732 | PreferFP &= !SVEStackSize; | |||
1733 | ||||
1734 | // Note: Keeping the following as multiple 'if' statements rather than | |||
1735 | // merging to a single expression for readability. | |||
1736 | // | |||
1737 | // Argument access should always use the FP. | |||
1738 | if (isFixed) { | |||
1739 | UseFP = hasFP(MF); | |||
1740 | } else if (isCSR && RegInfo->needsStackRealignment(MF)) { | |||
1741 | // References to the CSR area must use FP if we're re-aligning the stack | |||
1742 | // since the dynamically-sized alignment padding is between the SP/BP and | |||
1743 | // the CSR area. | |||
1744 | assert(hasFP(MF) && "Re-aligned stack must have frame pointer")((hasFP(MF) && "Re-aligned stack must have frame pointer" ) ? static_cast<void> (0) : __assert_fail ("hasFP(MF) && \"Re-aligned stack must have frame pointer\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1744, __PRETTY_FUNCTION__)); | |||
1745 | UseFP = true; | |||
1746 | } else if (hasFP(MF) && !RegInfo->needsStackRealignment(MF)) { | |||
1747 | // If the FPOffset is negative and we're producing a signed immediate, we | |||
1748 | // have to keep in mind that the available offset range for negative | |||
1749 | // offsets is smaller than for positive ones. If an offset is available | |||
1750 | // via the FP and the SP, use whichever is closest. | |||
1751 | bool FPOffsetFits = !ForSimm || FPOffset >= -256; | |||
1752 | PreferFP |= Offset > -FPOffset; | |||
1753 | ||||
1754 | if (MFI.hasVarSizedObjects()) { | |||
1755 | // If we have variable sized objects, we can use either FP or BP, as the | |||
1756 | // SP offset is unknown. We can use the base pointer if we have one and | |||
1757 | // FP is not preferred. If not, we're stuck with using FP. | |||
1758 | bool CanUseBP = RegInfo->hasBasePointer(MF); | |||
1759 | if (FPOffsetFits && CanUseBP) // Both are ok. Pick the best. | |||
1760 | UseFP = PreferFP; | |||
1761 | else if (!CanUseBP) { // Can't use BP. Forced to use FP. | |||
1762 | assert(!SVEStackSize && "Expected BP to be available")((!SVEStackSize && "Expected BP to be available") ? static_cast <void> (0) : __assert_fail ("!SVEStackSize && \"Expected BP to be available\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1762, __PRETTY_FUNCTION__)); | |||
1763 | UseFP = true; | |||
1764 | } | |||
1765 | // else we can use BP and FP, but the offset from FP won't fit. | |||
1766 | // That will make us scavenge registers which we can probably avoid by | |||
1767 | // using BP. If it won't fit for BP either, we'll scavenge anyway. | |||
1768 | } else if (FPOffset >= 0) { | |||
1769 | // Use SP or FP, whichever gives us the best chance of the offset | |||
1770 | // being in range for direct access. If the FPOffset is positive, | |||
1771 | // that'll always be best, as the SP will be even further away. | |||
1772 | UseFP = true; | |||
1773 | } else if (MF.hasEHFunclets() && !RegInfo->hasBasePointer(MF)) { | |||
1774 | // Funclets access the locals contained in the parent's stack frame | |||
1775 | // via the frame pointer, so we have to use the FP in the parent | |||
1776 | // function. | |||
1777 | (void) Subtarget; | |||
1778 | assert(((Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv ()) && "Funclets should only be present on Win64") ? static_cast <void> (0) : __assert_fail ("Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()) && \"Funclets should only be present on Win64\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1780, __PRETTY_FUNCTION__)) | |||
1779 | Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()) &&((Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv ()) && "Funclets should only be present on Win64") ? static_cast <void> (0) : __assert_fail ("Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()) && \"Funclets should only be present on Win64\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1780, __PRETTY_FUNCTION__)) | |||
1780 | "Funclets should only be present on Win64")((Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv ()) && "Funclets should only be present on Win64") ? static_cast <void> (0) : __assert_fail ("Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv()) && \"Funclets should only be present on Win64\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1780, __PRETTY_FUNCTION__)); | |||
1781 | UseFP = true; | |||
1782 | } else { | |||
1783 | // We have the choice between FP and (SP or BP). | |||
1784 | if (FPOffsetFits && PreferFP) // If FP is the best fit, use it. | |||
1785 | UseFP = true; | |||
1786 | } | |||
1787 | } | |||
1788 | } | |||
1789 | ||||
1790 | assert(((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF) || !UseFP) &&((((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF ) || !UseFP) && "In the presence of dynamic stack pointer realignment, " "non-argument/CSR objects cannot be accessed through the frame pointer" ) ? static_cast<void> (0) : __assert_fail ("((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF) || !UseFP) && \"In the presence of dynamic stack pointer realignment, \" \"non-argument/CSR objects cannot be accessed through the frame pointer\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1792, __PRETTY_FUNCTION__)) | |||
1791 | "In the presence of dynamic stack pointer realignment, "((((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF ) || !UseFP) && "In the presence of dynamic stack pointer realignment, " "non-argument/CSR objects cannot be accessed through the frame pointer" ) ? static_cast<void> (0) : __assert_fail ("((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF) || !UseFP) && \"In the presence of dynamic stack pointer realignment, \" \"non-argument/CSR objects cannot be accessed through the frame pointer\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1792, __PRETTY_FUNCTION__)) | |||
1792 | "non-argument/CSR objects cannot be accessed through the frame pointer")((((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF ) || !UseFP) && "In the presence of dynamic stack pointer realignment, " "non-argument/CSR objects cannot be accessed through the frame pointer" ) ? static_cast<void> (0) : __assert_fail ("((isFixed || isCSR) || !RegInfo->needsStackRealignment(MF) || !UseFP) && \"In the presence of dynamic stack pointer realignment, \" \"non-argument/CSR objects cannot be accessed through the frame pointer\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1792, __PRETTY_FUNCTION__)); | |||
1793 | ||||
1794 | if (isSVE) { | |||
1795 | int64_t OffsetToSVEArea = | |||
1796 | MFI.getStackSize() - AFI->getCalleeSavedStackSize(); | |||
1797 | StackOffset FPOffset = {ObjectOffset, MVT::nxv1i8}; | |||
1798 | StackOffset SPOffset = SVEStackSize + | |||
1799 | StackOffset(ObjectOffset, MVT::nxv1i8) + | |||
1800 | StackOffset(OffsetToSVEArea, MVT::i8); | |||
1801 | // Always use the FP for SVE spills if available and beneficial. | |||
1802 | if (hasFP(MF) && | |||
1803 | (SPOffset.getBytes() || | |||
1804 | FPOffset.getScalableBytes() < SPOffset.getScalableBytes() || | |||
1805 | RegInfo->needsStackRealignment(MF))) { | |||
1806 | FrameReg = RegInfo->getFrameRegister(MF); | |||
1807 | return FPOffset; | |||
1808 | } | |||
1809 | ||||
1810 | FrameReg = RegInfo->hasBasePointer(MF) ? RegInfo->getBaseRegister() | |||
1811 | : (unsigned)AArch64::SP; | |||
1812 | return SPOffset; | |||
1813 | } | |||
1814 | ||||
1815 | StackOffset ScalableOffset = {}; | |||
1816 | if (UseFP && !(isFixed || isCSR)) | |||
1817 | ScalableOffset = -SVEStackSize; | |||
1818 | if (!UseFP && (isFixed || isCSR)) | |||
1819 | ScalableOffset = SVEStackSize; | |||
1820 | ||||
1821 | if (UseFP) { | |||
1822 | FrameReg = RegInfo->getFrameRegister(MF); | |||
1823 | return StackOffset(FPOffset, MVT::i8) + ScalableOffset; | |||
1824 | } | |||
1825 | ||||
1826 | // Use the base pointer if we have one. | |||
1827 | if (RegInfo->hasBasePointer(MF)) | |||
1828 | FrameReg = RegInfo->getBaseRegister(); | |||
1829 | else { | |||
1830 | assert(!MFI.hasVarSizedObjects() &&((!MFI.hasVarSizedObjects() && "Can't use SP when we have var sized objects." ) ? static_cast<void> (0) : __assert_fail ("!MFI.hasVarSizedObjects() && \"Can't use SP when we have var sized objects.\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1831, __PRETTY_FUNCTION__)) | |||
1831 | "Can't use SP when we have var sized objects.")((!MFI.hasVarSizedObjects() && "Can't use SP when we have var sized objects." ) ? static_cast<void> (0) : __assert_fail ("!MFI.hasVarSizedObjects() && \"Can't use SP when we have var sized objects.\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1831, __PRETTY_FUNCTION__)); | |||
1832 | FrameReg = AArch64::SP; | |||
1833 | // If we're using the red zone for this function, the SP won't actually | |||
1834 | // be adjusted, so the offsets will be negative. They're also all | |||
1835 | // within range of the signed 9-bit immediate instructions. | |||
1836 | if (canUseRedZone(MF)) | |||
1837 | Offset -= AFI->getLocalStackSize(); | |||
1838 | } | |||
1839 | ||||
1840 | return StackOffset(Offset, MVT::i8) + ScalableOffset; | |||
1841 | } | |||
1842 | ||||
1843 | static unsigned getPrologueDeath(MachineFunction &MF, unsigned Reg) { | |||
1844 | // Do not set a kill flag on values that are also marked as live-in. This | |||
1845 | // happens with the @llvm-returnaddress intrinsic and with arguments passed in | |||
1846 | // callee saved registers. | |||
1847 | // Omitting the kill flags is conservatively correct even if the live-in | |||
1848 | // is not used after all. | |||
1849 | bool IsLiveIn = MF.getRegInfo().isLiveIn(Reg); | |||
1850 | return getKillRegState(!IsLiveIn); | |||
1851 | } | |||
1852 | ||||
1853 | static bool produceCompactUnwindFrame(MachineFunction &MF) { | |||
1854 | const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>(); | |||
1855 | AttributeList Attrs = MF.getFunction().getAttributes(); | |||
1856 | return Subtarget.isTargetMachO() && | |||
1857 | !(Subtarget.getTargetLowering()->supportSwiftError() && | |||
1858 | Attrs.hasAttrSomewhere(Attribute::SwiftError)); | |||
1859 | } | |||
1860 | ||||
1861 | static bool invalidateWindowsRegisterPairing(unsigned Reg1, unsigned Reg2, | |||
1862 | bool NeedsWinCFI) { | |||
1863 | // If we are generating register pairs for a Windows function that requires | |||
1864 | // EH support, then pair consecutive registers only. There are no unwind | |||
1865 | // opcodes for saves/restores of non-consectuve register pairs. | |||
1866 | // The unwind opcodes are save_regp, save_regp_x, save_fregp, save_frepg_x. | |||
1867 | // https://docs.microsoft.com/en-us/cpp/build/arm64-exception-handling | |||
1868 | ||||
1869 | // TODO: LR can be paired with any register. We don't support this yet in | |||
1870 | // the MCLayer. We need to add support for the save_lrpair unwind code. | |||
1871 | if (!NeedsWinCFI) | |||
1872 | return false; | |||
1873 | if (Reg2 == Reg1 + 1) | |||
1874 | return false; | |||
1875 | return true; | |||
1876 | } | |||
1877 | ||||
1878 | /// Returns true if Reg1 and Reg2 cannot be paired using a ldp/stp instruction. | |||
1879 | /// WindowsCFI requires that only consecutive registers can be paired. | |||
1880 | /// LR and FP need to be allocated together when the frame needs to save | |||
1881 | /// the frame-record. This means any other register pairing with LR is invalid. | |||
1882 | static bool invalidateRegisterPairing(unsigned Reg1, unsigned Reg2, | |||
1883 | bool NeedsWinCFI, bool NeedsFrameRecord) { | |||
1884 | if (NeedsWinCFI) | |||
1885 | return invalidateWindowsRegisterPairing(Reg1, Reg2, true); | |||
1886 | ||||
1887 | // If we need to store the frame record, don't pair any register | |||
1888 | // with LR other than FP. | |||
1889 | if (NeedsFrameRecord) | |||
1890 | return Reg2 == AArch64::LR; | |||
1891 | ||||
1892 | return false; | |||
1893 | } | |||
1894 | ||||
1895 | namespace { | |||
1896 | ||||
1897 | struct RegPairInfo { | |||
1898 | unsigned Reg1 = AArch64::NoRegister; | |||
1899 | unsigned Reg2 = AArch64::NoRegister; | |||
1900 | int FrameIdx; | |||
1901 | int Offset; | |||
1902 | enum RegType { GPR, FPR64, FPR128, PPR, ZPR } Type; | |||
1903 | ||||
1904 | RegPairInfo() = default; | |||
1905 | ||||
1906 | bool isPaired() const { return Reg2 != AArch64::NoRegister; } | |||
1907 | ||||
1908 | unsigned getScale() const { | |||
1909 | switch (Type) { | |||
1910 | case PPR: | |||
1911 | return 2; | |||
1912 | case GPR: | |||
1913 | case FPR64: | |||
1914 | return 8; | |||
1915 | case ZPR: | |||
1916 | case FPR128: | |||
1917 | return 16; | |||
1918 | } | |||
1919 | llvm_unreachable("Unsupported type")::llvm::llvm_unreachable_internal("Unsupported type", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1919); | |||
1920 | } | |||
1921 | ||||
1922 | bool isScalable() const { return Type == PPR || Type == ZPR; } | |||
1923 | }; | |||
1924 | ||||
1925 | } // end anonymous namespace | |||
1926 | ||||
1927 | static void computeCalleeSaveRegisterPairs( | |||
1928 | MachineFunction &MF, const std::vector<CalleeSavedInfo> &CSI, | |||
1929 | const TargetRegisterInfo *TRI, SmallVectorImpl<RegPairInfo> &RegPairs, | |||
1930 | bool &NeedShadowCallStackProlog, bool NeedsFrameRecord) { | |||
1931 | ||||
1932 | if (CSI.empty()) | |||
1933 | return; | |||
1934 | ||||
1935 | bool NeedsWinCFI = needsWinCFI(MF); | |||
1936 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
1937 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
1938 | CallingConv::ID CC = MF.getFunction().getCallingConv(); | |||
1939 | unsigned Count = CSI.size(); | |||
1940 | (void)CC; | |||
1941 | // MachO's compact unwind format relies on all registers being stored in | |||
1942 | // pairs. | |||
1943 | assert((!produceCompactUnwindFrame(MF) ||(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && "Odd number of callee-saved regs to spill!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && \"Odd number of callee-saved regs to spill!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1946, __PRETTY_FUNCTION__)) | |||
1944 | CC == CallingConv::PreserveMost ||(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && "Odd number of callee-saved regs to spill!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && \"Odd number of callee-saved regs to spill!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1946, __PRETTY_FUNCTION__)) | |||
1945 | (Count & 1) == 0) &&(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && "Odd number of callee-saved regs to spill!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && \"Odd number of callee-saved regs to spill!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1946, __PRETTY_FUNCTION__)) | |||
1946 | "Odd number of callee-saved regs to spill!")(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && "Odd number of callee-saved regs to spill!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (Count & 1) == 0) && \"Odd number of callee-saved regs to spill!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1946, __PRETTY_FUNCTION__)); | |||
1947 | int ByteOffset = AFI->getCalleeSavedStackSize(); | |||
1948 | int ScalableByteOffset = AFI->getSVECalleeSavedStackSize(); | |||
1949 | // On Linux, we will have either one or zero non-paired register. On Windows | |||
1950 | // with CFI, we can have multiple unpaired registers in order to utilize the | |||
1951 | // available unwind codes. This flag assures that the alignment fixup is done | |||
1952 | // only once, as intened. | |||
1953 | bool FixupDone = false; | |||
1954 | for (unsigned i = 0; i < Count; ++i) { | |||
1955 | RegPairInfo RPI; | |||
1956 | RPI.Reg1 = CSI[i].getReg(); | |||
1957 | ||||
1958 | if (AArch64::GPR64RegClass.contains(RPI.Reg1)) | |||
1959 | RPI.Type = RegPairInfo::GPR; | |||
1960 | else if (AArch64::FPR64RegClass.contains(RPI.Reg1)) | |||
1961 | RPI.Type = RegPairInfo::FPR64; | |||
1962 | else if (AArch64::FPR128RegClass.contains(RPI.Reg1)) | |||
1963 | RPI.Type = RegPairInfo::FPR128; | |||
1964 | else if (AArch64::ZPRRegClass.contains(RPI.Reg1)) | |||
1965 | RPI.Type = RegPairInfo::ZPR; | |||
1966 | else if (AArch64::PPRRegClass.contains(RPI.Reg1)) | |||
1967 | RPI.Type = RegPairInfo::PPR; | |||
1968 | else | |||
1969 | llvm_unreachable("Unsupported register class.")::llvm::llvm_unreachable_internal("Unsupported register class." , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 1969); | |||
1970 | ||||
1971 | // Add the next reg to the pair if it is in the same register class. | |||
1972 | if (i + 1 < Count) { | |||
1973 | unsigned NextReg = CSI[i + 1].getReg(); | |||
1974 | switch (RPI.Type) { | |||
1975 | case RegPairInfo::GPR: | |||
1976 | if (AArch64::GPR64RegClass.contains(NextReg) && | |||
1977 | !invalidateRegisterPairing(RPI.Reg1, NextReg, NeedsWinCFI, | |||
1978 | NeedsFrameRecord)) | |||
1979 | RPI.Reg2 = NextReg; | |||
1980 | break; | |||
1981 | case RegPairInfo::FPR64: | |||
1982 | if (AArch64::FPR64RegClass.contains(NextReg) && | |||
1983 | !invalidateWindowsRegisterPairing(RPI.Reg1, NextReg, NeedsWinCFI)) | |||
1984 | RPI.Reg2 = NextReg; | |||
1985 | break; | |||
1986 | case RegPairInfo::FPR128: | |||
1987 | if (AArch64::FPR128RegClass.contains(NextReg)) | |||
1988 | RPI.Reg2 = NextReg; | |||
1989 | break; | |||
1990 | case RegPairInfo::PPR: | |||
1991 | case RegPairInfo::ZPR: | |||
1992 | break; | |||
1993 | } | |||
1994 | } | |||
1995 | ||||
1996 | // If either of the registers to be saved is the lr register, it means that | |||
1997 | // we also need to save lr in the shadow call stack. | |||
1998 | if ((RPI.Reg1 == AArch64::LR || RPI.Reg2 == AArch64::LR) && | |||
1999 | MF.getFunction().hasFnAttribute(Attribute::ShadowCallStack)) { | |||
2000 | if (!MF.getSubtarget<AArch64Subtarget>().isXRegisterReserved(18)) | |||
2001 | report_fatal_error("Must reserve x18 to use shadow call stack"); | |||
2002 | NeedShadowCallStackProlog = true; | |||
2003 | } | |||
2004 | ||||
2005 | // GPRs and FPRs are saved in pairs of 64-bit regs. We expect the CSI | |||
2006 | // list to come in sorted by frame index so that we can issue the store | |||
2007 | // pair instructions directly. Assert if we see anything otherwise. | |||
2008 | // | |||
2009 | // The order of the registers in the list is controlled by | |||
2010 | // getCalleeSavedRegs(), so they will always be in-order, as well. | |||
2011 | assert((!RPI.isPaired() ||(((!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1] .getFrameIdx())) && "Out of order callee saved regs!" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1].getFrameIdx())) && \"Out of order callee saved regs!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2013, __PRETTY_FUNCTION__)) | |||
2012 | (CSI[i].getFrameIdx() + 1 == CSI[i + 1].getFrameIdx())) &&(((!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1] .getFrameIdx())) && "Out of order callee saved regs!" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1].getFrameIdx())) && \"Out of order callee saved regs!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2013, __PRETTY_FUNCTION__)) | |||
2013 | "Out of order callee saved regs!")(((!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1] .getFrameIdx())) && "Out of order callee saved regs!" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || (CSI[i].getFrameIdx() + 1 == CSI[i + 1].getFrameIdx())) && \"Out of order callee saved regs!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2013, __PRETTY_FUNCTION__)); | |||
2014 | ||||
2015 | assert((!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64::FP ||(((!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64 ::FP || RPI.Reg1 == AArch64::LR) && "FrameRecord must be allocated together with LR" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64::FP || RPI.Reg1 == AArch64::LR) && \"FrameRecord must be allocated together with LR\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2017, __PRETTY_FUNCTION__)) | |||
2016 | RPI.Reg1 == AArch64::LR) &&(((!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64 ::FP || RPI.Reg1 == AArch64::LR) && "FrameRecord must be allocated together with LR" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64::FP || RPI.Reg1 == AArch64::LR) && \"FrameRecord must be allocated together with LR\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2017, __PRETTY_FUNCTION__)) | |||
2017 | "FrameRecord must be allocated together with LR")(((!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64 ::FP || RPI.Reg1 == AArch64::LR) && "FrameRecord must be allocated together with LR" ) ? static_cast<void> (0) : __assert_fail ("(!RPI.isPaired() || !NeedsFrameRecord || RPI.Reg2 != AArch64::FP || RPI.Reg1 == AArch64::LR) && \"FrameRecord must be allocated together with LR\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2017, __PRETTY_FUNCTION__)); | |||
2018 | ||||
2019 | // MachO's compact unwind format relies on all registers being stored in | |||
2020 | // adjacent register pairs. | |||
2021 | assert((!produceCompactUnwindFrame(MF) ||(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)) | |||
2022 | CC == CallingConv::PreserveMost ||(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)) | |||
2023 | (RPI.isPaired() &&(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)) | |||
2024 | ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) ||(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)) | |||
2025 | RPI.Reg1 + 1 == RPI.Reg2))) &&(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)) | |||
2026 | "Callee-save registers not saved as adjacent register pair!")(((!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && "Callee-save registers not saved as adjacent register pair!" ) ? static_cast<void> (0) : __assert_fail ("(!produceCompactUnwindFrame(MF) || CC == CallingConv::PreserveMost || (RPI.isPaired() && ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) || RPI.Reg1 + 1 == RPI.Reg2))) && \"Callee-save registers not saved as adjacent register pair!\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2026, __PRETTY_FUNCTION__)); | |||
2027 | ||||
2028 | RPI.FrameIdx = CSI[i].getFrameIdx(); | |||
2029 | ||||
2030 | int Scale = RPI.getScale(); | |||
2031 | if (RPI.isScalable()) | |||
2032 | ScalableByteOffset -= Scale; | |||
2033 | else | |||
2034 | ByteOffset -= RPI.isPaired() ? 2 * Scale : Scale; | |||
2035 | ||||
2036 | assert(!(RPI.isScalable() && RPI.isPaired()) &&((!(RPI.isScalable() && RPI.isPaired()) && "Paired spill/fill instructions don't exist for SVE vectors" ) ? static_cast<void> (0) : __assert_fail ("!(RPI.isScalable() && RPI.isPaired()) && \"Paired spill/fill instructions don't exist for SVE vectors\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2037, __PRETTY_FUNCTION__)) | |||
2037 | "Paired spill/fill instructions don't exist for SVE vectors")((!(RPI.isScalable() && RPI.isPaired()) && "Paired spill/fill instructions don't exist for SVE vectors" ) ? static_cast<void> (0) : __assert_fail ("!(RPI.isScalable() && RPI.isPaired()) && \"Paired spill/fill instructions don't exist for SVE vectors\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2037, __PRETTY_FUNCTION__)); | |||
2038 | ||||
2039 | // Round up size of non-pair to pair size if we need to pad the | |||
2040 | // callee-save area to ensure 16-byte alignment. | |||
2041 | if (AFI->hasCalleeSaveStackFreeSpace() && !FixupDone && | |||
2042 | !RPI.isScalable() && RPI.Type != RegPairInfo::FPR128 && | |||
2043 | !RPI.isPaired()) { | |||
2044 | FixupDone = true; | |||
2045 | ByteOffset -= 8; | |||
2046 | assert(ByteOffset % 16 == 0)((ByteOffset % 16 == 0) ? static_cast<void> (0) : __assert_fail ("ByteOffset % 16 == 0", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2046, __PRETTY_FUNCTION__)); | |||
2047 | assert(MFI.getObjectAlignment(RPI.FrameIdx) <= 16)((MFI.getObjectAlignment(RPI.FrameIdx) <= 16) ? static_cast <void> (0) : __assert_fail ("MFI.getObjectAlignment(RPI.FrameIdx) <= 16" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2047, __PRETTY_FUNCTION__)); | |||
2048 | MFI.setObjectAlignment(RPI.FrameIdx, 16); | |||
2049 | } | |||
2050 | ||||
2051 | int Offset = RPI.isScalable() ? ScalableByteOffset : ByteOffset; | |||
2052 | assert(Offset % Scale == 0)((Offset % Scale == 0) ? static_cast<void> (0) : __assert_fail ("Offset % Scale == 0", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2052, __PRETTY_FUNCTION__)); | |||
2053 | RPI.Offset = Offset / Scale; | |||
2054 | ||||
2055 | assert(((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) ||((((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && "Offset out of bounds for LDP/STP immediate" ) ? static_cast<void> (0) : __assert_fail ("((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && \"Offset out of bounds for LDP/STP immediate\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2057, __PRETTY_FUNCTION__)) | |||
2056 | (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) &&((((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && "Offset out of bounds for LDP/STP immediate" ) ? static_cast<void> (0) : __assert_fail ("((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && \"Offset out of bounds for LDP/STP immediate\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2057, __PRETTY_FUNCTION__)) | |||
2057 | "Offset out of bounds for LDP/STP immediate")((((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && "Offset out of bounds for LDP/STP immediate" ) ? static_cast<void> (0) : __assert_fail ("((!RPI.isScalable() && RPI.Offset >= -64 && RPI.Offset <= 63) || (RPI.isScalable() && RPI.Offset >= -256 && RPI.Offset <= 255)) && \"Offset out of bounds for LDP/STP immediate\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2057, __PRETTY_FUNCTION__)); | |||
2058 | ||||
2059 | RegPairs.push_back(RPI); | |||
2060 | if (RPI.isPaired()) | |||
2061 | ++i; | |||
2062 | } | |||
2063 | } | |||
2064 | ||||
2065 | bool AArch64FrameLowering::spillCalleeSavedRegisters( | |||
2066 | MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, | |||
2067 | const std::vector<CalleeSavedInfo> &CSI, | |||
2068 | const TargetRegisterInfo *TRI) const { | |||
2069 | MachineFunction &MF = *MBB.getParent(); | |||
2070 | const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); | |||
2071 | bool NeedsWinCFI = needsWinCFI(MF); | |||
2072 | DebugLoc DL; | |||
2073 | SmallVector<RegPairInfo, 8> RegPairs; | |||
2074 | ||||
2075 | bool NeedShadowCallStackProlog = false; | |||
2076 | computeCalleeSaveRegisterPairs(MF, CSI, TRI, RegPairs, | |||
2077 | NeedShadowCallStackProlog, hasFP(MF)); | |||
2078 | const MachineRegisterInfo &MRI = MF.getRegInfo(); | |||
2079 | ||||
2080 | if (NeedShadowCallStackProlog) { | |||
2081 | // Shadow call stack prolog: str x30, [x18], #8 | |||
2082 | BuildMI(MBB, MI, DL, TII.get(AArch64::STRXpost)) | |||
2083 | .addReg(AArch64::X18, RegState::Define) | |||
2084 | .addReg(AArch64::LR) | |||
2085 | .addReg(AArch64::X18) | |||
2086 | .addImm(8) | |||
2087 | .setMIFlag(MachineInstr::FrameSetup); | |||
2088 | ||||
2089 | if (NeedsWinCFI) | |||
2090 | BuildMI(MBB, MI, DL, TII.get(AArch64::SEH_Nop)) | |||
2091 | .setMIFlag(MachineInstr::FrameSetup); | |||
2092 | ||||
2093 | if (!MF.getFunction().hasFnAttribute(Attribute::NoUnwind)) { | |||
2094 | // Emit a CFI instruction that causes 8 to be subtracted from the value of | |||
2095 | // x18 when unwinding past this frame. | |||
2096 | static const char CFIInst[] = { | |||
2097 | dwarf::DW_CFA_val_expression, | |||
2098 | 18, // register | |||
2099 | 2, // length | |||
2100 | static_cast<char>(unsigned(dwarf::DW_OP_breg18)), | |||
2101 | static_cast<char>(-8) & 0x7f, // addend (sleb128) | |||
2102 | }; | |||
2103 | unsigned CFIIndex = MF.addFrameInst(MCCFIInstruction::createEscape( | |||
2104 | nullptr, StringRef(CFIInst, sizeof(CFIInst)))); | |||
2105 | BuildMI(MBB, MI, DL, TII.get(AArch64::CFI_INSTRUCTION)) | |||
2106 | .addCFIIndex(CFIIndex) | |||
2107 | .setMIFlag(MachineInstr::FrameSetup); | |||
2108 | } | |||
2109 | ||||
2110 | // This instruction also makes x18 live-in to the entry block. | |||
2111 | MBB.addLiveIn(AArch64::X18); | |||
2112 | } | |||
2113 | ||||
2114 | for (auto RPII = RegPairs.rbegin(), RPIE = RegPairs.rend(); RPII != RPIE; | |||
2115 | ++RPII) { | |||
2116 | RegPairInfo RPI = *RPII; | |||
2117 | unsigned Reg1 = RPI.Reg1; | |||
2118 | unsigned Reg2 = RPI.Reg2; | |||
2119 | unsigned StrOpc; | |||
2120 | ||||
2121 | // Issue sequence of spills for cs regs. The first spill may be converted | |||
2122 | // to a pre-decrement store later by emitPrologue if the callee-save stack | |||
2123 | // area allocation can't be combined with the local stack area allocation. | |||
2124 | // For example: | |||
2125 | // stp x22, x21, [sp, #0] // addImm(+0) | |||
2126 | // stp x20, x19, [sp, #16] // addImm(+2) | |||
2127 | // stp fp, lr, [sp, #32] // addImm(+4) | |||
2128 | // Rationale: This sequence saves uop updates compared to a sequence of | |||
2129 | // pre-increment spills like stp xi,xj,[sp,#-16]! | |||
2130 | // Note: Similar rationale and sequence for restores in epilog. | |||
2131 | unsigned Size, Align; | |||
2132 | switch (RPI.Type) { | |||
2133 | case RegPairInfo::GPR: | |||
2134 | StrOpc = RPI.isPaired() ? AArch64::STPXi : AArch64::STRXui; | |||
2135 | Size = 8; | |||
2136 | Align = 8; | |||
2137 | break; | |||
2138 | case RegPairInfo::FPR64: | |||
2139 | StrOpc = RPI.isPaired() ? AArch64::STPDi : AArch64::STRDui; | |||
2140 | Size = 8; | |||
2141 | Align = 8; | |||
2142 | break; | |||
2143 | case RegPairInfo::FPR128: | |||
2144 | StrOpc = RPI.isPaired() ? AArch64::STPQi : AArch64::STRQui; | |||
2145 | Size = 16; | |||
2146 | Align = 16; | |||
2147 | break; | |||
2148 | case RegPairInfo::ZPR: | |||
2149 | StrOpc = AArch64::STR_ZXI; | |||
2150 | Size = 16; | |||
2151 | Align = 16; | |||
2152 | break; | |||
2153 | case RegPairInfo::PPR: | |||
2154 | StrOpc = AArch64::STR_PXI; | |||
2155 | Size = 2; | |||
2156 | Align = 2; | |||
2157 | break; | |||
2158 | } | |||
2159 | LLVM_DEBUG(dbgs() << "CSR spill: (" << printReg(Reg1, TRI);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR spill: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2160 | if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR spill: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2161 | dbgs() << ") -> fi#(" << RPI.FrameIdx;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR spill: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2162 | if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR spill: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2163 | dbgs() << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR spill: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false); | |||
2164 | ||||
2165 | assert((!NeedsWinCFI || !(Reg1 == AArch64::LR && Reg2 == AArch64::FP)) &&(((!NeedsWinCFI || !(Reg1 == AArch64::LR && Reg2 == AArch64 ::FP)) && "Windows unwdinding requires a consecutive (FP,LR) pair" ) ? static_cast<void> (0) : __assert_fail ("(!NeedsWinCFI || !(Reg1 == AArch64::LR && Reg2 == AArch64::FP)) && \"Windows unwdinding requires a consecutive (FP,LR) pair\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2166, __PRETTY_FUNCTION__)) | |||
2166 | "Windows unwdinding requires a consecutive (FP,LR) pair")(((!NeedsWinCFI || !(Reg1 == AArch64::LR && Reg2 == AArch64 ::FP)) && "Windows unwdinding requires a consecutive (FP,LR) pair" ) ? static_cast<void> (0) : __assert_fail ("(!NeedsWinCFI || !(Reg1 == AArch64::LR && Reg2 == AArch64::FP)) && \"Windows unwdinding requires a consecutive (FP,LR) pair\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2166, __PRETTY_FUNCTION__)); | |||
2167 | // Windows unwind codes require consecutive registers if registers are | |||
2168 | // paired. Make the switch here, so that the code below will save (x,x+1) | |||
2169 | // and not (x+1,x). | |||
2170 | unsigned FrameIdxReg1 = RPI.FrameIdx; | |||
2171 | unsigned FrameIdxReg2 = RPI.FrameIdx + 1; | |||
2172 | if (NeedsWinCFI && RPI.isPaired()) { | |||
2173 | std::swap(Reg1, Reg2); | |||
2174 | std::swap(FrameIdxReg1, FrameIdxReg2); | |||
2175 | } | |||
2176 | MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(StrOpc)); | |||
2177 | if (!MRI.isReserved(Reg1)) | |||
2178 | MBB.addLiveIn(Reg1); | |||
2179 | if (RPI.isPaired()) { | |||
2180 | if (!MRI.isReserved(Reg2)) | |||
2181 | MBB.addLiveIn(Reg2); | |||
2182 | MIB.addReg(Reg2, getPrologueDeath(MF, Reg2)); | |||
2183 | MIB.addMemOperand(MF.getMachineMemOperand( | |||
2184 | MachinePointerInfo::getFixedStack(MF, FrameIdxReg2), | |||
2185 | MachineMemOperand::MOStore, Size, Align)); | |||
2186 | } | |||
2187 | MIB.addReg(Reg1, getPrologueDeath(MF, Reg1)) | |||
2188 | .addReg(AArch64::SP) | |||
2189 | .addImm(RPI.Offset) // [sp, #offset*scale], | |||
2190 | // where factor*scale is implicit | |||
2191 | .setMIFlag(MachineInstr::FrameSetup); | |||
2192 | MIB.addMemOperand(MF.getMachineMemOperand( | |||
2193 | MachinePointerInfo::getFixedStack(MF,FrameIdxReg1), | |||
2194 | MachineMemOperand::MOStore, Size, Align)); | |||
2195 | if (NeedsWinCFI) | |||
2196 | InsertSEH(MIB, TII, MachineInstr::FrameSetup); | |||
2197 | ||||
2198 | // Update the StackIDs of the SVE stack slots. | |||
2199 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
2200 | if (RPI.Type == RegPairInfo::ZPR || RPI.Type == RegPairInfo::PPR) | |||
2201 | MFI.setStackID(RPI.FrameIdx, TargetStackID::SVEVector); | |||
2202 | ||||
2203 | } | |||
2204 | return true; | |||
2205 | } | |||
2206 | ||||
2207 | bool AArch64FrameLowering::restoreCalleeSavedRegisters( | |||
2208 | MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, | |||
2209 | std::vector<CalleeSavedInfo> &CSI, | |||
2210 | const TargetRegisterInfo *TRI) const { | |||
2211 | MachineFunction &MF = *MBB.getParent(); | |||
2212 | const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); | |||
2213 | DebugLoc DL; | |||
2214 | SmallVector<RegPairInfo, 8> RegPairs; | |||
2215 | bool NeedsWinCFI = needsWinCFI(MF); | |||
2216 | ||||
2217 | if (MI != MBB.end()) | |||
2218 | DL = MI->getDebugLoc(); | |||
2219 | ||||
2220 | bool NeedShadowCallStackProlog = false; | |||
2221 | computeCalleeSaveRegisterPairs(MF, CSI, TRI, RegPairs, | |||
2222 | NeedShadowCallStackProlog, hasFP(MF)); | |||
2223 | ||||
2224 | auto EmitMI = [&](const RegPairInfo &RPI) { | |||
2225 | unsigned Reg1 = RPI.Reg1; | |||
2226 | unsigned Reg2 = RPI.Reg2; | |||
2227 | ||||
2228 | // Issue sequence of restores for cs regs. The last restore may be converted | |||
2229 | // to a post-increment load later by emitEpilogue if the callee-save stack | |||
2230 | // area allocation can't be combined with the local stack area allocation. | |||
2231 | // For example: | |||
2232 | // ldp fp, lr, [sp, #32] // addImm(+4) | |||
2233 | // ldp x20, x19, [sp, #16] // addImm(+2) | |||
2234 | // ldp x22, x21, [sp, #0] // addImm(+0) | |||
2235 | // Note: see comment in spillCalleeSavedRegisters() | |||
2236 | unsigned LdrOpc; | |||
2237 | unsigned Size, Align; | |||
2238 | switch (RPI.Type) { | |||
2239 | case RegPairInfo::GPR: | |||
2240 | LdrOpc = RPI.isPaired() ? AArch64::LDPXi : AArch64::LDRXui; | |||
2241 | Size = 8; | |||
2242 | Align = 8; | |||
2243 | break; | |||
2244 | case RegPairInfo::FPR64: | |||
2245 | LdrOpc = RPI.isPaired() ? AArch64::LDPDi : AArch64::LDRDui; | |||
2246 | Size = 8; | |||
2247 | Align = 8; | |||
2248 | break; | |||
2249 | case RegPairInfo::FPR128: | |||
2250 | LdrOpc = RPI.isPaired() ? AArch64::LDPQi : AArch64::LDRQui; | |||
2251 | Size = 16; | |||
2252 | Align = 16; | |||
2253 | break; | |||
2254 | case RegPairInfo::ZPR: | |||
2255 | LdrOpc = AArch64::LDR_ZXI; | |||
2256 | Size = 16; | |||
2257 | Align = 16; | |||
2258 | break; | |||
2259 | case RegPairInfo::PPR: | |||
2260 | LdrOpc = AArch64::LDR_PXI; | |||
2261 | Size = 2; | |||
2262 | Align = 2; | |||
2263 | break; | |||
2264 | } | |||
2265 | LLVM_DEBUG(dbgs() << "CSR restore: (" << printReg(Reg1, TRI);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR restore: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2266 | if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR restore: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2267 | dbgs() << ") -> fi#(" << RPI.FrameIdx;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR restore: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2268 | if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR restore: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false) | |||
2269 | dbgs() << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "CSR restore: (" << printReg (Reg1, TRI); if (RPI.isPaired()) dbgs() << ", " << printReg(Reg2, TRI); dbgs() << ") -> fi#(" << RPI.FrameIdx; if (RPI.isPaired()) dbgs() << ", " << RPI.FrameIdx + 1; dbgs() << ")\n"; } } while (false); | |||
2270 | ||||
2271 | // Windows unwind codes require consecutive registers if registers are | |||
2272 | // paired. Make the switch here, so that the code below will save (x,x+1) | |||
2273 | // and not (x+1,x). | |||
2274 | unsigned FrameIdxReg1 = RPI.FrameIdx; | |||
2275 | unsigned FrameIdxReg2 = RPI.FrameIdx + 1; | |||
2276 | if (NeedsWinCFI && RPI.isPaired()) { | |||
2277 | std::swap(Reg1, Reg2); | |||
2278 | std::swap(FrameIdxReg1, FrameIdxReg2); | |||
2279 | } | |||
2280 | MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(LdrOpc)); | |||
2281 | if (RPI.isPaired()) { | |||
2282 | MIB.addReg(Reg2, getDefRegState(true)); | |||
2283 | MIB.addMemOperand(MF.getMachineMemOperand( | |||
2284 | MachinePointerInfo::getFixedStack(MF, FrameIdxReg2), | |||
2285 | MachineMemOperand::MOLoad, Size, Align)); | |||
2286 | } | |||
2287 | MIB.addReg(Reg1, getDefRegState(true)) | |||
2288 | .addReg(AArch64::SP) | |||
2289 | .addImm(RPI.Offset) // [sp, #offset*scale] | |||
2290 | // where factor*scale is implicit | |||
2291 | .setMIFlag(MachineInstr::FrameDestroy); | |||
2292 | MIB.addMemOperand(MF.getMachineMemOperand( | |||
2293 | MachinePointerInfo::getFixedStack(MF, FrameIdxReg1), | |||
2294 | MachineMemOperand::MOLoad, Size, Align)); | |||
2295 | if (NeedsWinCFI) | |||
2296 | InsertSEH(MIB, TII, MachineInstr::FrameDestroy); | |||
2297 | }; | |||
2298 | ||||
2299 | // SVE objects are always restored in reverse order. | |||
2300 | for (const RegPairInfo &RPI : reverse(RegPairs)) | |||
2301 | if (RPI.isScalable()) | |||
2302 | EmitMI(RPI); | |||
2303 | ||||
2304 | if (ReverseCSRRestoreSeq) { | |||
2305 | for (const RegPairInfo &RPI : reverse(RegPairs)) | |||
2306 | if (!RPI.isScalable()) | |||
2307 | EmitMI(RPI); | |||
2308 | } else | |||
2309 | for (const RegPairInfo &RPI : RegPairs) | |||
2310 | if (!RPI.isScalable()) | |||
2311 | EmitMI(RPI); | |||
2312 | ||||
2313 | if (NeedShadowCallStackProlog) { | |||
2314 | // Shadow call stack epilog: ldr x30, [x18, #-8]! | |||
2315 | BuildMI(MBB, MI, DL, TII.get(AArch64::LDRXpre)) | |||
2316 | .addReg(AArch64::X18, RegState::Define) | |||
2317 | .addReg(AArch64::LR, RegState::Define) | |||
2318 | .addReg(AArch64::X18) | |||
2319 | .addImm(-8) | |||
2320 | .setMIFlag(MachineInstr::FrameDestroy); | |||
2321 | } | |||
2322 | ||||
2323 | return true; | |||
2324 | } | |||
2325 | ||||
2326 | void AArch64FrameLowering::determineCalleeSaves(MachineFunction &MF, | |||
2327 | BitVector &SavedRegs, | |||
2328 | RegScavenger *RS) const { | |||
2329 | // All calls are tail calls in GHC calling conv, and functions have no | |||
2330 | // prologue/epilogue. | |||
2331 | if (MF.getFunction().getCallingConv() == CallingConv::GHC) | |||
2332 | return; | |||
2333 | ||||
2334 | TargetFrameLowering::determineCalleeSaves(MF, SavedRegs, RS); | |||
2335 | const AArch64RegisterInfo *RegInfo = static_cast<const AArch64RegisterInfo *>( | |||
2336 | MF.getSubtarget().getRegisterInfo()); | |||
2337 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
2338 | unsigned UnspilledCSGPR = AArch64::NoRegister; | |||
2339 | unsigned UnspilledCSGPRPaired = AArch64::NoRegister; | |||
2340 | ||||
2341 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
2342 | const MCPhysReg *CSRegs = MF.getRegInfo().getCalleeSavedRegs(); | |||
2343 | ||||
2344 | unsigned BasePointerReg = RegInfo->hasBasePointer(MF) | |||
2345 | ? RegInfo->getBaseRegister() | |||
2346 | : (unsigned)AArch64::NoRegister; | |||
2347 | ||||
2348 | unsigned ExtraCSSpill = 0; | |||
2349 | // Figure out which callee-saved registers to save/restore. | |||
2350 | for (unsigned i = 0; CSRegs[i]; ++i) { | |||
2351 | const unsigned Reg = CSRegs[i]; | |||
2352 | ||||
2353 | // Add the base pointer register to SavedRegs if it is callee-save. | |||
2354 | if (Reg == BasePointerReg) | |||
2355 | SavedRegs.set(Reg); | |||
2356 | ||||
2357 | bool RegUsed = SavedRegs.test(Reg); | |||
2358 | unsigned PairedReg = AArch64::NoRegister; | |||
2359 | if (AArch64::GPR64RegClass.contains(Reg) || | |||
2360 | AArch64::FPR64RegClass.contains(Reg) || | |||
2361 | AArch64::FPR128RegClass.contains(Reg)) | |||
2362 | PairedReg = CSRegs[i ^ 1]; | |||
2363 | ||||
2364 | if (!RegUsed) { | |||
2365 | if (AArch64::GPR64RegClass.contains(Reg) && | |||
2366 | !RegInfo->isReservedReg(MF, Reg)) { | |||
2367 | UnspilledCSGPR = Reg; | |||
2368 | UnspilledCSGPRPaired = PairedReg; | |||
2369 | } | |||
2370 | continue; | |||
2371 | } | |||
2372 | ||||
2373 | // MachO's compact unwind format relies on all registers being stored in | |||
2374 | // pairs. | |||
2375 | // FIXME: the usual format is actually better if unwinding isn't needed. | |||
2376 | if (produceCompactUnwindFrame(MF) && PairedReg != AArch64::NoRegister && | |||
2377 | !SavedRegs.test(PairedReg)) { | |||
2378 | SavedRegs.set(PairedReg); | |||
2379 | if (AArch64::GPR64RegClass.contains(PairedReg) && | |||
2380 | !RegInfo->isReservedReg(MF, PairedReg)) | |||
2381 | ExtraCSSpill = PairedReg; | |||
2382 | } | |||
2383 | } | |||
2384 | ||||
2385 | // Calculates the callee saved stack size. | |||
2386 | unsigned CSStackSize = 0; | |||
2387 | unsigned SVECSStackSize = 0; | |||
2388 | const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); | |||
2389 | const MachineRegisterInfo &MRI = MF.getRegInfo(); | |||
2390 | for (unsigned Reg : SavedRegs.set_bits()) { | |||
2391 | auto RegSize = TRI->getRegSizeInBits(Reg, MRI) / 8; | |||
2392 | if (AArch64::PPRRegClass.contains(Reg) || | |||
2393 | AArch64::ZPRRegClass.contains(Reg)) | |||
2394 | SVECSStackSize += RegSize; | |||
2395 | else | |||
2396 | CSStackSize += RegSize; | |||
2397 | } | |||
2398 | ||||
2399 | // Save number of saved regs, so we can easily update CSStackSize later. | |||
2400 | unsigned NumSavedRegs = SavedRegs.count(); | |||
2401 | ||||
2402 | // The frame record needs to be created by saving the appropriate registers | |||
2403 | unsigned EstimatedStackSize = MFI.estimateStackSize(MF); | |||
2404 | if (hasFP(MF) || | |||
2405 | windowsRequiresStackProbe(MF, EstimatedStackSize + CSStackSize + 16)) { | |||
2406 | SavedRegs.set(AArch64::FP); | |||
2407 | SavedRegs.set(AArch64::LR); | |||
2408 | } | |||
2409 | ||||
2410 | LLVM_DEBUG(dbgs() << "*** determineCalleeSaves\nSaved CSRs:";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "*** determineCalleeSaves\nSaved CSRs:" ; for (unsigned Reg : SavedRegs.set_bits()) dbgs() << ' ' << printReg(Reg, RegInfo); dbgs() << "\n";; } } while (false) | |||
2411 | for (unsigned Regdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "*** determineCalleeSaves\nSaved CSRs:" ; for (unsigned Reg : SavedRegs.set_bits()) dbgs() << ' ' << printReg(Reg, RegInfo); dbgs() << "\n";; } } while (false) | |||
2412 | : SavedRegs.set_bits()) dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "*** determineCalleeSaves\nSaved CSRs:" ; for (unsigned Reg : SavedRegs.set_bits()) dbgs() << ' ' << printReg(Reg, RegInfo); dbgs() << "\n";; } } while (false) | |||
2413 | << ' ' << printReg(Reg, RegInfo);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "*** determineCalleeSaves\nSaved CSRs:" ; for (unsigned Reg : SavedRegs.set_bits()) dbgs() << ' ' << printReg(Reg, RegInfo); dbgs() << "\n";; } } while (false) | |||
2414 | dbgs() << "\n";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "*** determineCalleeSaves\nSaved CSRs:" ; for (unsigned Reg : SavedRegs.set_bits()) dbgs() << ' ' << printReg(Reg, RegInfo); dbgs() << "\n";; } } while (false); | |||
2415 | ||||
2416 | // If any callee-saved registers are used, the frame cannot be eliminated. | |||
2417 | int64_t SVEStackSize = | |||
2418 | alignTo(SVECSStackSize + estimateSVEStackObjectOffsets(MFI), 16); | |||
2419 | bool CanEliminateFrame = (SavedRegs.count() == 0) && !SVEStackSize; | |||
2420 | ||||
2421 | // The CSR spill slots have not been allocated yet, so estimateStackSize | |||
2422 | // won't include them. | |||
2423 | unsigned EstimatedStackSizeLimit = estimateRSStackSizeLimit(MF); | |||
2424 | ||||
2425 | // Conservatively always assume BigStack when there are SVE spills. | |||
2426 | bool BigStack = SVEStackSize || | |||
2427 | (EstimatedStackSize + CSStackSize) > EstimatedStackSizeLimit; | |||
2428 | if (BigStack || !CanEliminateFrame || RegInfo->cannotEliminateFrame(MF)) | |||
2429 | AFI->setHasStackFrame(true); | |||
2430 | ||||
2431 | // Estimate if we might need to scavenge a register at some point in order | |||
2432 | // to materialize a stack offset. If so, either spill one additional | |||
2433 | // callee-saved register or reserve a special spill slot to facilitate | |||
2434 | // register scavenging. If we already spilled an extra callee-saved register | |||
2435 | // above to keep the number of spills even, we don't need to do anything else | |||
2436 | // here. | |||
2437 | if (BigStack) { | |||
2438 | if (!ExtraCSSpill && UnspilledCSGPR != AArch64::NoRegister) { | |||
2439 | LLVM_DEBUG(dbgs() << "Spilling " << printReg(UnspilledCSGPR, RegInfo)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Spilling " << printReg (UnspilledCSGPR, RegInfo) << " to get a scratch register.\n" ; } } while (false) | |||
2440 | << " to get a scratch register.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Spilling " << printReg (UnspilledCSGPR, RegInfo) << " to get a scratch register.\n" ; } } while (false); | |||
2441 | SavedRegs.set(UnspilledCSGPR); | |||
2442 | // MachO's compact unwind format relies on all registers being stored in | |||
2443 | // pairs, so if we need to spill one extra for BigStack, then we need to | |||
2444 | // store the pair. | |||
2445 | if (produceCompactUnwindFrame(MF)) | |||
2446 | SavedRegs.set(UnspilledCSGPRPaired); | |||
2447 | ExtraCSSpill = UnspilledCSGPR; | |||
2448 | } | |||
2449 | ||||
2450 | // If we didn't find an extra callee-saved register to spill, create | |||
2451 | // an emergency spill slot. | |||
2452 | if (!ExtraCSSpill || MF.getRegInfo().isPhysRegUsed(ExtraCSSpill)) { | |||
2453 | const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); | |||
2454 | const TargetRegisterClass &RC = AArch64::GPR64RegClass; | |||
2455 | unsigned Size = TRI->getSpillSize(RC); | |||
2456 | unsigned Align = TRI->getSpillAlignment(RC); | |||
2457 | int FI = MFI.CreateStackObject(Size, Align, false); | |||
2458 | RS->addScavengingFrameIndex(FI); | |||
2459 | LLVM_DEBUG(dbgs() << "No available CS registers, allocated fi#" << FIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "No available CS registers, allocated fi#" << FI << " as the emergency spill slot.\n"; } } while (false) | |||
2460 | << " as the emergency spill slot.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "No available CS registers, allocated fi#" << FI << " as the emergency spill slot.\n"; } } while (false); | |||
2461 | } | |||
2462 | } | |||
2463 | ||||
2464 | // Adding the size of additional 64bit GPR saves. | |||
2465 | CSStackSize += 8 * (SavedRegs.count() - NumSavedRegs); | |||
2466 | unsigned AlignedCSStackSize = alignTo(CSStackSize, 16); | |||
2467 | LLVM_DEBUG(dbgs() << "Estimated stack frame size: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Estimated stack frame size: " << EstimatedStackSize + AlignedCSStackSize << " bytes.\n" ; } } while (false) | |||
2468 | << EstimatedStackSize + AlignedCSStackSizedo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Estimated stack frame size: " << EstimatedStackSize + AlignedCSStackSize << " bytes.\n" ; } } while (false) | |||
2469 | << " bytes.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Estimated stack frame size: " << EstimatedStackSize + AlignedCSStackSize << " bytes.\n" ; } } while (false); | |||
2470 | ||||
2471 | assert((!MFI.isCalleeSavedInfoValid() ||(((!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize () == AlignedCSStackSize) && "Should not invalidate callee saved info" ) ? static_cast<void> (0) : __assert_fail ("(!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize() == AlignedCSStackSize) && \"Should not invalidate callee saved info\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2473, __PRETTY_FUNCTION__)) | |||
2472 | AFI->getCalleeSavedStackSize() == AlignedCSStackSize) &&(((!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize () == AlignedCSStackSize) && "Should not invalidate callee saved info" ) ? static_cast<void> (0) : __assert_fail ("(!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize() == AlignedCSStackSize) && \"Should not invalidate callee saved info\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2473, __PRETTY_FUNCTION__)) | |||
2473 | "Should not invalidate callee saved info")(((!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize () == AlignedCSStackSize) && "Should not invalidate callee saved info" ) ? static_cast<void> (0) : __assert_fail ("(!MFI.isCalleeSavedInfoValid() || AFI->getCalleeSavedStackSize() == AlignedCSStackSize) && \"Should not invalidate callee saved info\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2473, __PRETTY_FUNCTION__)); | |||
2474 | ||||
2475 | // Round up to register pair alignment to avoid additional SP adjustment | |||
2476 | // instructions. | |||
2477 | AFI->setCalleeSavedStackSize(AlignedCSStackSize); | |||
2478 | AFI->setCalleeSaveStackHasFreeSpace(AlignedCSStackSize != CSStackSize); | |||
2479 | AFI->setSVECalleeSavedStackSize(alignTo(SVECSStackSize, 16)); | |||
2480 | } | |||
2481 | ||||
2482 | bool AArch64FrameLowering::enableStackSlotScavenging( | |||
2483 | const MachineFunction &MF) const { | |||
2484 | const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
2485 | return AFI->hasCalleeSaveStackFreeSpace(); | |||
2486 | } | |||
2487 | ||||
2488 | /// returns true if there are any SVE callee saves. | |||
2489 | static bool getSVECalleeSaveSlotRange(const MachineFrameInfo &MFI, | |||
2490 | int &Min, int &Max) { | |||
2491 | if (!MFI.isCalleeSavedInfoValid()) | |||
2492 | return false; | |||
2493 | ||||
2494 | Min = std::numeric_limits<int>::max(); | |||
2495 | Max = std::numeric_limits<int>::min(); | |||
2496 | const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo(); | |||
2497 | for (auto &CS : CSI) { | |||
2498 | if (AArch64::ZPRRegClass.contains(CS.getReg()) || | |||
2499 | AArch64::PPRRegClass.contains(CS.getReg())) { | |||
2500 | assert((Max == std::numeric_limits<int>::min() ||(((Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && "SVE CalleeSaves are not consecutive" ) ? static_cast<void> (0) : __assert_fail ("(Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && \"SVE CalleeSaves are not consecutive\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2502, __PRETTY_FUNCTION__)) | |||
2501 | Max + 1 == CS.getFrameIdx()) &&(((Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && "SVE CalleeSaves are not consecutive" ) ? static_cast<void> (0) : __assert_fail ("(Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && \"SVE CalleeSaves are not consecutive\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2502, __PRETTY_FUNCTION__)) | |||
2502 | "SVE CalleeSaves are not consecutive")(((Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && "SVE CalleeSaves are not consecutive" ) ? static_cast<void> (0) : __assert_fail ("(Max == std::numeric_limits<int>::min() || Max + 1 == CS.getFrameIdx()) && \"SVE CalleeSaves are not consecutive\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2502, __PRETTY_FUNCTION__)); | |||
2503 | ||||
2504 | Min = std::min(Min, CS.getFrameIdx()); | |||
2505 | Max = std::max(Max, CS.getFrameIdx()); | |||
2506 | } | |||
2507 | } | |||
2508 | return Min != std::numeric_limits<int>::max(); | |||
2509 | } | |||
2510 | ||||
2511 | // Process all the SVE stack objects and determine offsets for each | |||
2512 | // object. If AssignOffsets is true, the offsets get assigned. | |||
2513 | // Fills in the first and last callee-saved frame indices into | |||
2514 | // Min/MaxCSFrameIndex, respectively. | |||
2515 | // Returns the size of the stack. | |||
2516 | static int64_t determineSVEStackObjectOffsets(MachineFrameInfo &MFI, | |||
2517 | int &MinCSFrameIndex, | |||
2518 | int &MaxCSFrameIndex, | |||
2519 | bool AssignOffsets) { | |||
2520 | // First process all fixed stack objects. | |||
2521 | int64_t Offset = 0; | |||
2522 | for (int I = MFI.getObjectIndexBegin(); I != 0; ++I) | |||
2523 | if (MFI.getStackID(I) == TargetStackID::SVEVector) { | |||
2524 | int64_t FixedOffset = -MFI.getObjectOffset(I); | |||
2525 | if (FixedOffset > Offset) | |||
2526 | Offset = FixedOffset; | |||
2527 | } | |||
2528 | ||||
2529 | // Then process all callee saved slots. | |||
2530 | if (getSVECalleeSaveSlotRange(MFI, MinCSFrameIndex, MaxCSFrameIndex)) { | |||
2531 | // Make sure to align the last callee save slot. | |||
2532 | MFI.setObjectAlignment(MaxCSFrameIndex, 16U); | |||
2533 | ||||
2534 | // Assign offsets to the callee save slots. | |||
2535 | for (int I = MinCSFrameIndex; I <= MaxCSFrameIndex; ++I) { | |||
2536 | Offset += MFI.getObjectSize(I); | |||
2537 | Offset = alignTo(Offset, MFI.getObjectAlignment(I)); | |||
2538 | if (AssignOffsets) { | |||
2539 | LLVM_DEBUG(dbgs() << "alloc FI(" << I << ") at SP[" << Offsetdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "alloc FI(" << I << ") at SP[" << Offset << "]\n"; } } while (false) | |||
2540 | << "]\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "alloc FI(" << I << ") at SP[" << Offset << "]\n"; } } while (false); | |||
2541 | MFI.setObjectOffset(I, -Offset); | |||
2542 | } | |||
2543 | } | |||
2544 | } | |||
2545 | ||||
2546 | // Note: We don't take allocatable stack objects into | |||
2547 | // account yet, because allocation for those is not yet | |||
2548 | // implemented. | |||
2549 | return Offset; | |||
2550 | } | |||
2551 | ||||
2552 | int64_t AArch64FrameLowering::estimateSVEStackObjectOffsets( | |||
2553 | MachineFrameInfo &MFI) const { | |||
2554 | int MinCSFrameIndex, MaxCSFrameIndex; | |||
2555 | return determineSVEStackObjectOffsets(MFI, MinCSFrameIndex, MaxCSFrameIndex, false); | |||
2556 | } | |||
2557 | ||||
2558 | int64_t AArch64FrameLowering::assignSVEStackObjectOffsets( | |||
2559 | MachineFrameInfo &MFI, int &MinCSFrameIndex, int &MaxCSFrameIndex) const { | |||
2560 | return determineSVEStackObjectOffsets(MFI, MinCSFrameIndex, MaxCSFrameIndex, | |||
2561 | true); | |||
2562 | } | |||
2563 | ||||
2564 | void AArch64FrameLowering::processFunctionBeforeFrameFinalized( | |||
2565 | MachineFunction &MF, RegScavenger *RS) const { | |||
2566 | MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
2567 | ||||
2568 | assert(getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown &&((getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown && "Upwards growing stack unsupported") ? static_cast <void> (0) : __assert_fail ("getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown && \"Upwards growing stack unsupported\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2569, __PRETTY_FUNCTION__)) | |||
| ||||
2569 | "Upwards growing stack unsupported")((getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown && "Upwards growing stack unsupported") ? static_cast <void> (0) : __assert_fail ("getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown && \"Upwards growing stack unsupported\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2569, __PRETTY_FUNCTION__)); | |||
2570 | ||||
2571 | int MinCSFrameIndex, MaxCSFrameIndex; | |||
2572 | int64_t SVEStackSize = | |||
2573 | assignSVEStackObjectOffsets(MFI, MinCSFrameIndex, MaxCSFrameIndex); | |||
2574 | ||||
2575 | AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>(); | |||
2576 | AFI->setStackSizeSVE(alignTo(SVEStackSize, 16U)); | |||
2577 | AFI->setMinMaxSVECSFrameIndex(MinCSFrameIndex, MaxCSFrameIndex); | |||
| ||||
2578 | ||||
2579 | // If this function isn't doing Win64-style C++ EH, we don't need to do | |||
2580 | // anything. | |||
2581 | if (!MF.hasEHFunclets()) | |||
2582 | return; | |||
2583 | const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); | |||
2584 | WinEHFuncInfo &EHInfo = *MF.getWinEHFuncInfo(); | |||
2585 | ||||
2586 | MachineBasicBlock &MBB = MF.front(); | |||
2587 | auto MBBI = MBB.begin(); | |||
2588 | while (MBBI != MBB.end() && MBBI->getFlag(MachineInstr::FrameSetup)) | |||
2589 | ++MBBI; | |||
2590 | ||||
2591 | // Create an UnwindHelp object. | |||
2592 | int UnwindHelpFI = | |||
2593 | MFI.CreateStackObject(/*size*/8, /*alignment*/16, false); | |||
2594 | EHInfo.UnwindHelpFrameIdx = UnwindHelpFI; | |||
2595 | // We need to store -2 into the UnwindHelp object at the start of the | |||
2596 | // function. | |||
2597 | DebugLoc DL; | |||
2598 | RS->enterBasicBlockEnd(MBB); | |||
2599 | RS->backward(std::prev(MBBI)); | |||
2600 | unsigned DstReg = RS->FindUnusedReg(&AArch64::GPR64commonRegClass); | |||
2601 | assert(DstReg && "There must be a free register after frame setup")((DstReg && "There must be a free register after frame setup" ) ? static_cast<void> (0) : __assert_fail ("DstReg && \"There must be a free register after frame setup\"" , "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp" , 2601, __PRETTY_FUNCTION__)); | |||
2602 | BuildMI(MBB, MBBI, DL, TII.get(AArch64::MOVi64imm), DstReg).addImm(-2); | |||
2603 | BuildMI(MBB, MBBI, DL, TII.get(AArch64::STURXi)) | |||
2604 | .addReg(DstReg, getKillRegState(true)) | |||
2605 | .addFrameIndex(UnwindHelpFI) | |||
2606 | .addImm(0); | |||
2607 | } | |||
2608 | ||||
2609 | /// For Win64 AArch64 EH, the offset to the Unwind object is from the SP before | |||
2610 | /// the update. This is easily retrieved as it is exactly the offset that is set | |||
2611 | /// in processFunctionBeforeFrameFinalized. | |||
2612 | int AArch64FrameLowering::getFrameIndexReferencePreferSP( | |||
2613 | const MachineFunction &MF, int FI, unsigned &FrameReg, | |||
2614 | bool IgnoreSPUpdates) const { | |||
2615 | const MachineFrameInfo &MFI = MF.getFrameInfo(); | |||
2616 | LLVM_DEBUG(dbgs() << "Offset from the SP for " << FI << " is "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Offset from the SP for " << FI << " is " << MFI.getObjectOffset(FI) << "\n"; } } while (false) | |||
2617 | << MFI.getObjectOffset(FI) << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("frame-info")) { dbgs() << "Offset from the SP for " << FI << " is " << MFI.getObjectOffset(FI) << "\n"; } } while (false); | |||
2618 | FrameReg = AArch64::SP; | |||
2619 | return MFI.getObjectOffset(FI); | |||
2620 | } | |||
2621 | ||||
2622 | /// The parent frame offset (aka dispFrame) is only used on X86_64 to retrieve | |||
2623 | /// the parent's frame pointer | |||
2624 | unsigned AArch64FrameLowering::getWinEHParentFrameOffset( | |||
2625 | const MachineFunction &MF) const { | |||
2626 | return 0; | |||
2627 | } | |||
2628 | ||||
2629 | /// Funclets only need to account for space for the callee saved registers, | |||
2630 | /// as the locals are accounted for in the parent's stack frame. | |||
2631 | unsigned AArch64FrameLowering::getWinEHFuncletFrameSize( | |||
2632 | const MachineFunction &MF) const { | |||
2633 | // This is the size of the pushed CSRs. | |||
2634 | unsigned CSSize = | |||
2635 | MF.getInfo<AArch64FunctionInfo>()->getCalleeSavedStackSize(); | |||
2636 | // This is the amount of stack a funclet needs to allocate. | |||
2637 | return alignTo(CSSize + MF.getFrameInfo().getMaxCallFrameSize(), | |||
2638 | getStackAlignment()); | |||
2639 | } |