Revert " [LoongArch][ISel] Check the number of sign bits in `PatGprGpr_32` (#107432)"
[llvm-project.git] / llvm / lib / Target / LoongArch / LoongArchFrameLowering.cpp
blob4e504729b23e2d8f7f6510b335899fb3e6586835
1 //===-- LoongArchFrameLowering.cpp - LoongArch Frame Information -*- 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 LoongArch implementation of TargetFrameLowering class.
11 //===----------------------------------------------------------------------===//
13 #include "LoongArchFrameLowering.h"
14 #include "LoongArchMachineFunctionInfo.h"
15 #include "LoongArchSubtarget.h"
16 #include "MCTargetDesc/LoongArchBaseInfo.h"
17 #include "MCTargetDesc/LoongArchMCTargetDesc.h"
18 #include "llvm/CodeGen/MachineFrameInfo.h"
19 #include "llvm/CodeGen/MachineFunction.h"
20 #include "llvm/CodeGen/MachineInstrBuilder.h"
21 #include "llvm/CodeGen/MachineRegisterInfo.h"
22 #include "llvm/CodeGen/RegisterScavenging.h"
23 #include "llvm/IR/DiagnosticInfo.h"
24 #include "llvm/MC/MCDwarf.h"
26 using namespace llvm;
28 #define DEBUG_TYPE "loongarch-frame-lowering"
30 // Return true if the specified function should have a dedicated frame
31 // pointer register. This is true if frame pointer elimination is
32 // disabled, if it needs dynamic stack realignment, if the function has
33 // variable sized allocas, or if the frame address is taken.
34 bool LoongArchFrameLowering::hasFP(const MachineFunction &MF) const {
35 const TargetRegisterInfo *RegInfo = MF.getSubtarget().getRegisterInfo();
37 const MachineFrameInfo &MFI = MF.getFrameInfo();
38 return MF.getTarget().Options.DisableFramePointerElim(MF) ||
39 RegInfo->hasStackRealignment(MF) || MFI.hasVarSizedObjects() ||
40 MFI.isFrameAddressTaken();
43 bool LoongArchFrameLowering::hasBP(const MachineFunction &MF) const {
44 const MachineFrameInfo &MFI = MF.getFrameInfo();
45 const TargetRegisterInfo *TRI = STI.getRegisterInfo();
47 return MFI.hasVarSizedObjects() && TRI->hasStackRealignment(MF);
50 void LoongArchFrameLowering::adjustReg(MachineBasicBlock &MBB,
51 MachineBasicBlock::iterator MBBI,
52 const DebugLoc &DL, Register DestReg,
53 Register SrcReg, int64_t Val,
54 MachineInstr::MIFlag Flag) const {
55 const LoongArchInstrInfo *TII = STI.getInstrInfo();
56 bool IsLA64 = STI.is64Bit();
57 unsigned Addi = IsLA64 ? LoongArch::ADDI_D : LoongArch::ADDI_W;
59 if (DestReg == SrcReg && Val == 0)
60 return;
62 if (isInt<12>(Val)) {
63 // addi.w/d $DstReg, $SrcReg, Val
64 BuildMI(MBB, MBBI, DL, TII->get(Addi), DestReg)
65 .addReg(SrcReg)
66 .addImm(Val)
67 .setMIFlag(Flag);
68 return;
71 // Try to split the offset across two ADDIs. We need to keep the stack pointer
72 // aligned after each ADDI. We need to determine the maximum value we can put
73 // in each ADDI. In the negative direction, we can use -2048 which is always
74 // sufficiently aligned. In the positive direction, we need to find the
75 // largest 12-bit immediate that is aligned. Exclude -4096 since it can be
76 // created with LU12I.W.
77 assert(getStackAlign().value() < 2048 && "Stack alignment too large");
78 int64_t MaxPosAdjStep = 2048 - getStackAlign().value();
79 if (Val > -4096 && Val <= (2 * MaxPosAdjStep)) {
80 int64_t FirstAdj = Val < 0 ? -2048 : MaxPosAdjStep;
81 Val -= FirstAdj;
82 BuildMI(MBB, MBBI, DL, TII->get(Addi), DestReg)
83 .addReg(SrcReg)
84 .addImm(FirstAdj)
85 .setMIFlag(Flag);
86 BuildMI(MBB, MBBI, DL, TII->get(Addi), DestReg)
87 .addReg(DestReg, RegState::Kill)
88 .addImm(Val)
89 .setMIFlag(Flag);
90 return;
93 unsigned Opc = IsLA64 ? LoongArch::ADD_D : LoongArch::ADD_W;
94 if (Val < 0) {
95 Val = -Val;
96 Opc = IsLA64 ? LoongArch::SUB_D : LoongArch::SUB_W;
99 MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
100 Register ScratchReg = MRI.createVirtualRegister(&LoongArch::GPRRegClass);
101 TII->movImm(MBB, MBBI, DL, ScratchReg, Val, Flag);
102 BuildMI(MBB, MBBI, DL, TII->get(Opc), DestReg)
103 .addReg(SrcReg)
104 .addReg(ScratchReg, RegState::Kill)
105 .setMIFlag(Flag);
108 // Determine the size of the frame and maximum call frame size.
109 void LoongArchFrameLowering::determineFrameLayout(MachineFunction &MF) const {
110 MachineFrameInfo &MFI = MF.getFrameInfo();
112 // Get the number of bytes to allocate from the FrameInfo.
113 uint64_t FrameSize = MFI.getStackSize();
115 // Make sure the frame is aligned.
116 FrameSize = alignTo(FrameSize, getStackAlign());
118 // Update frame info.
119 MFI.setStackSize(FrameSize);
122 static uint64_t estimateFunctionSizeInBytes(const LoongArchInstrInfo *TII,
123 const MachineFunction &MF) {
124 uint64_t FuncSize = 0;
125 for (auto &MBB : MF)
126 for (auto &MI : MBB)
127 FuncSize += TII->getInstSizeInBytes(MI);
128 return FuncSize;
131 static bool needScavSlotForCFR(MachineFunction &MF) {
132 if (!MF.getSubtarget<LoongArchSubtarget>().hasBasicF())
133 return false;
134 for (auto &MBB : MF)
135 for (auto &MI : MBB)
136 if (MI.getOpcode() == LoongArch::PseudoST_CFR)
137 return true;
138 return false;
141 void LoongArchFrameLowering::processFunctionBeforeFrameFinalized(
142 MachineFunction &MF, RegScavenger *RS) const {
143 const LoongArchRegisterInfo *RI = STI.getRegisterInfo();
144 const TargetRegisterClass &RC = LoongArch::GPRRegClass;
145 const LoongArchInstrInfo *TII = STI.getInstrInfo();
146 LoongArchMachineFunctionInfo *LAFI =
147 MF.getInfo<LoongArchMachineFunctionInfo>();
148 MachineFrameInfo &MFI = MF.getFrameInfo();
150 unsigned ScavSlotsNum = 0;
152 // Far branches beyond 27-bit offset require a spill slot for scratch
153 // register.
154 bool IsLargeFunction = !isInt<27>(estimateFunctionSizeInBytes(TII, MF));
155 if (IsLargeFunction)
156 ScavSlotsNum = 1;
158 // estimateStackSize has been observed to under-estimate the final stack
159 // size, so give ourselves wiggle-room by checking for stack size
160 // representable an 11-bit signed field rather than 12-bits.
161 if (!isInt<11>(MFI.estimateStackSize(MF)))
162 ScavSlotsNum = std::max(ScavSlotsNum, 1u);
164 // For CFR spill.
165 if (needScavSlotForCFR(MF))
166 ++ScavSlotsNum;
168 // Create emergency spill slots.
169 for (unsigned i = 0; i < ScavSlotsNum; ++i) {
170 int FI = MFI.CreateStackObject(RI->getSpillSize(RC), RI->getSpillAlign(RC),
171 false);
172 RS->addScavengingFrameIndex(FI);
173 if (IsLargeFunction && LAFI->getBranchRelaxationSpillFrameIndex() == -1)
174 LAFI->setBranchRelaxationSpillFrameIndex(FI);
175 LLVM_DEBUG(dbgs() << "Allocated FI(" << FI
176 << ") as the emergency spill slot.\n");
180 void LoongArchFrameLowering::emitPrologue(MachineFunction &MF,
181 MachineBasicBlock &MBB) const {
182 MachineFrameInfo &MFI = MF.getFrameInfo();
183 auto *LoongArchFI = MF.getInfo<LoongArchMachineFunctionInfo>();
184 const LoongArchRegisterInfo *RI = STI.getRegisterInfo();
185 const LoongArchInstrInfo *TII = STI.getInstrInfo();
186 MachineBasicBlock::iterator MBBI = MBB.begin();
187 bool IsLA64 = STI.is64Bit();
189 Register SPReg = LoongArch::R3;
190 Register FPReg = LoongArch::R22;
192 // Debug location must be unknown since the first debug location is used
193 // to determine the end of the prologue.
194 DebugLoc DL;
195 // All calls are tail calls in GHC calling conv, and functions have no
196 // prologue/epilogue.
197 if (MF.getFunction().getCallingConv() == CallingConv::GHC)
198 return;
199 // Determine the correct frame layout
200 determineFrameLayout(MF);
202 // First, compute final stack size.
203 uint64_t StackSize = MFI.getStackSize();
204 uint64_t RealStackSize = StackSize;
206 // Early exit if there is no need to allocate space in the stack.
207 if (StackSize == 0 && !MFI.adjustsStack())
208 return;
210 uint64_t FirstSPAdjustAmount = getFirstSPAdjustAmount(MF);
211 // Split the SP adjustment to reduce the offsets of callee saved spill.
212 if (FirstSPAdjustAmount)
213 StackSize = FirstSPAdjustAmount;
215 // Adjust stack.
216 adjustReg(MBB, MBBI, DL, SPReg, SPReg, -StackSize, MachineInstr::FrameSetup);
217 // Emit ".cfi_def_cfa_offset StackSize".
218 unsigned CFIIndex =
219 MF.addFrameInst(MCCFIInstruction::cfiDefCfaOffset(nullptr, StackSize));
220 BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION))
221 .addCFIIndex(CFIIndex)
222 .setMIFlag(MachineInstr::FrameSetup);
224 const auto &CSI = MFI.getCalleeSavedInfo();
226 // The frame pointer is callee-saved, and code has been generated for us to
227 // save it to the stack. We need to skip over the storing of callee-saved
228 // registers as the frame pointer must be modified after it has been saved
229 // to the stack, not before.
230 std::advance(MBBI, CSI.size());
232 // Iterate over list of callee-saved registers and emit .cfi_offset
233 // directives.
234 for (const auto &Entry : CSI) {
235 int64_t Offset = MFI.getObjectOffset(Entry.getFrameIdx());
236 unsigned CFIIndex = MF.addFrameInst(MCCFIInstruction::createOffset(
237 nullptr, RI->getDwarfRegNum(Entry.getReg(), true), Offset));
238 BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION))
239 .addCFIIndex(CFIIndex)
240 .setMIFlag(MachineInstr::FrameSetup);
243 // Generate new FP.
244 if (hasFP(MF)) {
245 adjustReg(MBB, MBBI, DL, FPReg, SPReg,
246 StackSize - LoongArchFI->getVarArgsSaveSize(),
247 MachineInstr::FrameSetup);
249 // Emit ".cfi_def_cfa $fp, LoongArchFI->getVarArgsSaveSize()"
250 unsigned CFIIndex = MF.addFrameInst(
251 MCCFIInstruction::cfiDefCfa(nullptr, RI->getDwarfRegNum(FPReg, true),
252 LoongArchFI->getVarArgsSaveSize()));
253 BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION))
254 .addCFIIndex(CFIIndex)
255 .setMIFlag(MachineInstr::FrameSetup);
258 // Emit the second SP adjustment after saving callee saved registers.
259 if (FirstSPAdjustAmount) {
260 uint64_t SecondSPAdjustAmount = RealStackSize - FirstSPAdjustAmount;
261 assert(SecondSPAdjustAmount > 0 &&
262 "SecondSPAdjustAmount should be greater than zero");
263 adjustReg(MBB, MBBI, DL, SPReg, SPReg, -SecondSPAdjustAmount,
264 MachineInstr::FrameSetup);
266 if (!hasFP(MF)) {
267 // If we are using a frame-pointer, and thus emitted ".cfi_def_cfa fp, 0",
268 // don't emit an sp-based .cfi_def_cfa_offset
269 // Emit ".cfi_def_cfa_offset RealStackSize"
270 unsigned CFIIndex = MF.addFrameInst(
271 MCCFIInstruction::cfiDefCfaOffset(nullptr, RealStackSize));
272 BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION))
273 .addCFIIndex(CFIIndex)
274 .setMIFlag(MachineInstr::FrameSetup);
278 if (hasFP(MF)) {
279 // Realign stack.
280 if (RI->hasStackRealignment(MF)) {
281 unsigned Align = Log2(MFI.getMaxAlign());
282 assert(Align > 0 && "The stack realignment size is invalid!");
283 BuildMI(MBB, MBBI, DL,
284 TII->get(IsLA64 ? LoongArch::BSTRINS_D : LoongArch::BSTRINS_W),
285 SPReg)
286 .addReg(SPReg)
287 .addReg(LoongArch::R0)
288 .addImm(Align - 1)
289 .addImm(0)
290 .setMIFlag(MachineInstr::FrameSetup);
291 // FP will be used to restore the frame in the epilogue, so we need
292 // another base register BP to record SP after re-alignment. SP will
293 // track the current stack after allocating variable sized objects.
294 if (hasBP(MF)) {
295 // move BP, $sp
296 BuildMI(MBB, MBBI, DL, TII->get(LoongArch::OR),
297 LoongArchABI::getBPReg())
298 .addReg(SPReg)
299 .addReg(LoongArch::R0)
300 .setMIFlag(MachineInstr::FrameSetup);
306 void LoongArchFrameLowering::emitEpilogue(MachineFunction &MF,
307 MachineBasicBlock &MBB) const {
308 const LoongArchRegisterInfo *RI = STI.getRegisterInfo();
309 MachineFrameInfo &MFI = MF.getFrameInfo();
310 auto *LoongArchFI = MF.getInfo<LoongArchMachineFunctionInfo>();
311 Register SPReg = LoongArch::R3;
312 // All calls are tail calls in GHC calling conv, and functions have no
313 // prologue/epilogue.
314 if (MF.getFunction().getCallingConv() == CallingConv::GHC)
315 return;
316 MachineBasicBlock::iterator MBBI = MBB.getFirstTerminator();
317 DebugLoc DL = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
319 const auto &CSI = MFI.getCalleeSavedInfo();
320 // Skip to before the restores of callee-saved registers.
321 auto LastFrameDestroy = MBBI;
322 if (!CSI.empty())
323 LastFrameDestroy = std::prev(MBBI, CSI.size());
325 // Get the number of bytes from FrameInfo.
326 uint64_t StackSize = MFI.getStackSize();
328 // Restore the stack pointer.
329 if (RI->hasStackRealignment(MF) || MFI.hasVarSizedObjects()) {
330 assert(hasFP(MF) && "frame pointer should not have been eliminated");
331 adjustReg(MBB, LastFrameDestroy, DL, SPReg, LoongArch::R22,
332 -StackSize + LoongArchFI->getVarArgsSaveSize(),
333 MachineInstr::FrameDestroy);
336 uint64_t FirstSPAdjustAmount = getFirstSPAdjustAmount(MF);
337 if (FirstSPAdjustAmount) {
338 uint64_t SecondSPAdjustAmount = StackSize - FirstSPAdjustAmount;
339 assert(SecondSPAdjustAmount > 0 &&
340 "SecondSPAdjustAmount should be greater than zero");
342 adjustReg(MBB, LastFrameDestroy, DL, SPReg, SPReg, SecondSPAdjustAmount,
343 MachineInstr::FrameDestroy);
344 StackSize = FirstSPAdjustAmount;
347 // Deallocate stack
348 adjustReg(MBB, MBBI, DL, SPReg, SPReg, StackSize, MachineInstr::FrameDestroy);
351 // We would like to split the SP adjustment to reduce prologue/epilogue
352 // as following instructions. In this way, the offset of the callee saved
353 // register could fit in a single store.
354 // e.g.
355 // addi.d $sp, $sp, -2032
356 // st.d $ra, $sp, 2024
357 // st.d $fp, $sp, 2016
358 // addi.d $sp, $sp, -16
359 uint64_t LoongArchFrameLowering::getFirstSPAdjustAmount(
360 const MachineFunction &MF) const {
361 const MachineFrameInfo &MFI = MF.getFrameInfo();
362 const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
364 // Return the FirstSPAdjustAmount if the StackSize can not fit in a signed
365 // 12-bit and there exists a callee-saved register needing to be pushed.
366 if (!isInt<12>(MFI.getStackSize()) && (CSI.size() > 0)) {
367 // FirstSPAdjustAmount is chosen as (2048 - StackAlign) because 2048 will
368 // cause sp = sp + 2048 in the epilogue to be split into multiple
369 // instructions. Offsets smaller than 2048 can fit in a single load/store
370 // instruction, and we have to stick with the stack alignment.
371 // So (2048 - StackAlign) will satisfy the stack alignment.
372 return 2048 - getStackAlign().value();
374 return 0;
377 void LoongArchFrameLowering::determineCalleeSaves(MachineFunction &MF,
378 BitVector &SavedRegs,
379 RegScavenger *RS) const {
380 TargetFrameLowering::determineCalleeSaves(MF, SavedRegs, RS);
381 // Unconditionally spill RA and FP only if the function uses a frame
382 // pointer.
383 if (hasFP(MF)) {
384 SavedRegs.set(LoongArch::R1);
385 SavedRegs.set(LoongArch::R22);
387 // Mark BP as used if function has dedicated base pointer.
388 if (hasBP(MF))
389 SavedRegs.set(LoongArchABI::getBPReg());
392 // Do not preserve stack space within prologue for outgoing variables if the
393 // function contains variable size objects.
394 // Let eliminateCallFramePseudoInstr preserve stack space for it.
395 bool LoongArchFrameLowering::hasReservedCallFrame(
396 const MachineFunction &MF) const {
397 return !MF.getFrameInfo().hasVarSizedObjects();
400 // Eliminate ADJCALLSTACKDOWN, ADJCALLSTACKUP pseudo instructions.
401 MachineBasicBlock::iterator
402 LoongArchFrameLowering::eliminateCallFramePseudoInstr(
403 MachineFunction &MF, MachineBasicBlock &MBB,
404 MachineBasicBlock::iterator MI) const {
405 Register SPReg = LoongArch::R3;
406 DebugLoc DL = MI->getDebugLoc();
408 if (!hasReservedCallFrame(MF)) {
409 // If space has not been reserved for a call frame, ADJCALLSTACKDOWN and
410 // ADJCALLSTACKUP must be converted to instructions manipulating the stack
411 // pointer. This is necessary when there is a variable length stack
412 // allocation (e.g. alloca), which means it's not possible to allocate
413 // space for outgoing arguments from within the function prologue.
414 int64_t Amount = MI->getOperand(0).getImm();
416 if (Amount != 0) {
417 // Ensure the stack remains aligned after adjustment.
418 Amount = alignSPAdjust(Amount);
420 if (MI->getOpcode() == LoongArch::ADJCALLSTACKDOWN)
421 Amount = -Amount;
423 adjustReg(MBB, MI, DL, SPReg, SPReg, Amount, MachineInstr::NoFlags);
427 return MBB.erase(MI);
430 bool LoongArchFrameLowering::spillCalleeSavedRegisters(
431 MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
432 ArrayRef<CalleeSavedInfo> CSI, const TargetRegisterInfo *TRI) const {
433 if (CSI.empty())
434 return true;
436 MachineFunction *MF = MBB.getParent();
437 const TargetInstrInfo &TII = *MF->getSubtarget().getInstrInfo();
439 // Insert the spill to the stack frame.
440 for (auto &CS : CSI) {
441 Register Reg = CS.getReg();
442 // If the register is RA and the return address is taken by method
443 // LoongArchTargetLowering::lowerRETURNADDR, don't set kill flag.
444 bool IsKill =
445 !(Reg == LoongArch::R1 && MF->getFrameInfo().isReturnAddressTaken());
446 const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
447 TII.storeRegToStackSlot(MBB, MI, Reg, IsKill, CS.getFrameIdx(), RC, TRI,
448 Register());
451 return true;
454 StackOffset LoongArchFrameLowering::getFrameIndexReference(
455 const MachineFunction &MF, int FI, Register &FrameReg) const {
456 const MachineFrameInfo &MFI = MF.getFrameInfo();
457 const TargetRegisterInfo *RI = MF.getSubtarget().getRegisterInfo();
458 auto *LoongArchFI = MF.getInfo<LoongArchMachineFunctionInfo>();
459 uint64_t StackSize = MFI.getStackSize();
460 uint64_t FirstSPAdjustAmount = getFirstSPAdjustAmount(MF);
462 // Callee-saved registers should be referenced relative to the stack
463 // pointer (positive offset), otherwise use the frame pointer (negative
464 // offset).
465 const auto &CSI = MFI.getCalleeSavedInfo();
466 int MinCSFI = 0;
467 int MaxCSFI = -1;
468 StackOffset Offset =
469 StackOffset::getFixed(MFI.getObjectOffset(FI) - getOffsetOfLocalArea() +
470 MFI.getOffsetAdjustment());
472 if (CSI.size()) {
473 MinCSFI = CSI[0].getFrameIdx();
474 MaxCSFI = CSI[CSI.size() - 1].getFrameIdx();
477 if (FI >= MinCSFI && FI <= MaxCSFI) {
478 FrameReg = LoongArch::R3;
479 if (FirstSPAdjustAmount)
480 Offset += StackOffset::getFixed(FirstSPAdjustAmount);
481 else
482 Offset += StackOffset::getFixed(StackSize);
483 } else if (RI->hasStackRealignment(MF) && !MFI.isFixedObjectIndex(FI)) {
484 // If the stack was realigned, the frame pointer is set in order to allow
485 // SP to be restored, so we need another base register to record the stack
486 // after realignment.
487 FrameReg = hasBP(MF) ? LoongArchABI::getBPReg() : LoongArch::R3;
488 Offset += StackOffset::getFixed(StackSize);
489 } else {
490 FrameReg = RI->getFrameRegister(MF);
491 if (hasFP(MF))
492 Offset += StackOffset::getFixed(LoongArchFI->getVarArgsSaveSize());
493 else
494 Offset += StackOffset::getFixed(StackSize);
497 return Offset;
500 bool LoongArchFrameLowering::enableShrinkWrapping(
501 const MachineFunction &MF) const {
502 // Keep the conventional code flow when not optimizing.
503 if (MF.getFunction().hasOptNone())
504 return false;
506 return true;