xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AVR/AVRRegisterInfo.cpp (revision 700637cbb5e582861067a11aaca4d053546871d2)
1 //===-- AVRRegisterInfo.cpp - AVR Register Information --------------------===//
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 AVR implementation of the TargetRegisterInfo class.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "AVRRegisterInfo.h"
14 
15 #include "llvm/ADT/BitVector.h"
16 #include "llvm/CodeGen/MachineFrameInfo.h"
17 #include "llvm/CodeGen/MachineFunction.h"
18 #include "llvm/CodeGen/MachineInstrBuilder.h"
19 #include "llvm/CodeGen/TargetFrameLowering.h"
20 
21 #include "AVR.h"
22 #include "AVRInstrInfo.h"
23 #include "AVRMachineFunctionInfo.h"
24 #include "AVRTargetMachine.h"
25 #include "MCTargetDesc/AVRMCTargetDesc.h"
26 
27 #define GET_REGINFO_TARGET_DESC
28 #include "AVRGenRegisterInfo.inc"
29 
30 namespace llvm {
31 
AVRRegisterInfo()32 AVRRegisterInfo::AVRRegisterInfo() : AVRGenRegisterInfo(0) {}
33 
34 const uint16_t *
getCalleeSavedRegs(const MachineFunction * MF) const35 AVRRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
36   const AVRMachineFunctionInfo *AFI = MF->getInfo<AVRMachineFunctionInfo>();
37   const AVRSubtarget &STI = MF->getSubtarget<AVRSubtarget>();
38   if (STI.hasTinyEncoding())
39     return AFI->isInterruptOrSignalHandler() ? CSR_InterruptsTiny_SaveList
40                                              : CSR_NormalTiny_SaveList;
41   else
42     return AFI->isInterruptOrSignalHandler() ? CSR_Interrupts_SaveList
43                                              : CSR_Normal_SaveList;
44 }
45 
46 const uint32_t *
getCallPreservedMask(const MachineFunction & MF,CallingConv::ID CC) const47 AVRRegisterInfo::getCallPreservedMask(const MachineFunction &MF,
48                                       CallingConv::ID CC) const {
49   const AVRSubtarget &STI = MF.getSubtarget<AVRSubtarget>();
50   return STI.hasTinyEncoding() ? CSR_NormalTiny_RegMask : CSR_Normal_RegMask;
51 }
52 
getReservedRegs(const MachineFunction & MF) const53 BitVector AVRRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
54   BitVector Reserved(getNumRegs());
55 
56   // Reserve the intermediate result registers r1 and r2
57   // The result of instructions like 'mul' is always stored here.
58   // R0/R1/R1R0 are always reserved on both avr and avrtiny.
59   Reserved.set(AVR::R0);
60   Reserved.set(AVR::R1);
61   Reserved.set(AVR::R1R0);
62 
63   // Reserve the stack pointer.
64   Reserved.set(AVR::SPL);
65   Reserved.set(AVR::SPH);
66   Reserved.set(AVR::SP);
67 
68   // Reserve R2~R17 only on avrtiny.
69   if (MF.getSubtarget<AVRSubtarget>().hasTinyEncoding()) {
70     // Reserve 8-bit registers R2~R15, Rtmp(R16) and Zero(R17).
71     for (unsigned Reg = AVR::R2; Reg <= AVR::R17; Reg++)
72       Reserved.set(Reg);
73     // Reserve 16-bit registers R3R2~R18R17.
74     for (unsigned Reg = AVR::R3R2; Reg <= AVR::R18R17; Reg++)
75       Reserved.set(Reg);
76   }
77 
78   // We tenatively reserve the frame pointer register r29:r28 because the
79   // function may require one, but we cannot tell until register allocation
80   // is complete, which can be too late.
81   //
82   // Instead we just unconditionally reserve the Y register.
83   //
84   // TODO: Write a pass to enumerate functions which reserved the Y register
85   //       but didn't end up needing a frame pointer. In these, we can
86   //       convert one or two of the spills inside to use the Y register.
87   Reserved.set(AVR::R28);
88   Reserved.set(AVR::R29);
89   Reserved.set(AVR::R29R28);
90 
91   return Reserved;
92 }
93 
94 const TargetRegisterClass *
getLargestLegalSuperClass(const TargetRegisterClass * RC,const MachineFunction & MF) const95 AVRRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC,
96                                            const MachineFunction &MF) const {
97   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
98   if (TRI->isTypeLegalForClass(*RC, MVT::i16)) {
99     return &AVR::DREGSRegClass;
100   }
101 
102   if (TRI->isTypeLegalForClass(*RC, MVT::i8)) {
103     return &AVR::GPR8RegClass;
104   }
105 
106   llvm_unreachable("Invalid register size");
107 }
108 
109 /// Fold a frame offset shared between two add instructions into a single one.
foldFrameOffset(MachineBasicBlock::iterator & II,int & Offset,Register DstReg)110 static void foldFrameOffset(MachineBasicBlock::iterator &II, int &Offset,
111                             Register DstReg) {
112   MachineInstr &MI = *II;
113   int Opcode = MI.getOpcode();
114 
115   // Don't bother trying if the next instruction is not an add or a sub.
116   if ((Opcode != AVR::SUBIWRdK) && (Opcode != AVR::ADIWRdK)) {
117     return;
118   }
119 
120   // Check that DstReg matches with next instruction, otherwise the instruction
121   // is not related to stack address manipulation.
122   if (DstReg != MI.getOperand(0).getReg()) {
123     return;
124   }
125 
126   // Add the offset in the next instruction to our offset.
127   switch (Opcode) {
128   case AVR::SUBIWRdK:
129     Offset += -MI.getOperand(2).getImm();
130     break;
131   case AVR::ADIWRdK:
132     Offset += MI.getOperand(2).getImm();
133     break;
134   }
135 
136   // Finally remove the instruction.
137   II++;
138   MI.eraseFromParent();
139 }
140 
eliminateFrameIndex(MachineBasicBlock::iterator II,int SPAdj,unsigned FIOperandNum,RegScavenger * RS) const141 bool AVRRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
142                                           int SPAdj, unsigned FIOperandNum,
143                                           RegScavenger *RS) const {
144   assert(SPAdj == 0 && "Unexpected SPAdj value");
145 
146   MachineInstr &MI = *II;
147   DebugLoc dl = MI.getDebugLoc();
148   MachineBasicBlock &MBB = *MI.getParent();
149   const MachineFunction &MF = *MBB.getParent();
150   const AVRTargetMachine &TM = (const AVRTargetMachine &)MF.getTarget();
151   const TargetInstrInfo &TII = *TM.getSubtargetImpl()->getInstrInfo();
152   const MachineFrameInfo &MFI = MF.getFrameInfo();
153   const TargetFrameLowering *TFI = TM.getSubtargetImpl()->getFrameLowering();
154   const AVRSubtarget &STI = MF.getSubtarget<AVRSubtarget>();
155   int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
156   int Offset = MFI.getObjectOffset(FrameIndex);
157 
158   // Add one to the offset because SP points to an empty slot.
159   Offset += MFI.getStackSize() - TFI->getOffsetOfLocalArea() + 1;
160   // Fold incoming offset.
161   Offset += MI.getOperand(FIOperandNum + 1).getImm();
162 
163   // This is actually "load effective address" of the stack slot
164   // instruction. We have only two-address instructions, thus we need to
165   // expand it into move + add.
166   if (MI.getOpcode() == AVR::FRMIDX) {
167     Register DstReg = MI.getOperand(0).getReg();
168     assert(DstReg != AVR::R29R28 && "Dest reg cannot be the frame pointer");
169 
170     // Copy the frame pointer.
171     if (STI.hasMOVW()) {
172       BuildMI(MBB, MI, dl, TII.get(AVR::MOVWRdRr), DstReg).addReg(AVR::R29R28);
173     } else {
174       Register DstLoReg, DstHiReg;
175       splitReg(DstReg, DstLoReg, DstHiReg);
176       BuildMI(MBB, MI, dl, TII.get(AVR::MOVRdRr), DstLoReg).addReg(AVR::R28);
177       BuildMI(MBB, MI, dl, TII.get(AVR::MOVRdRr), DstHiReg).addReg(AVR::R29);
178     }
179 
180     assert(Offset > 0 && "Invalid offset");
181 
182     // We need to materialize the offset via an add instruction.
183     unsigned Opcode;
184 
185     II++; // Skip over the FRMIDX instruction.
186 
187     // Generally, to load a frame address two add instructions are emitted that
188     // could get folded into a single one:
189     //  movw    r31:r30, r29:r28
190     //  adiw    r31:r30, 29
191     //  adiw    r31:r30, 16
192     // to:
193     //  movw    r31:r30, r29:r28
194     //  adiw    r31:r30, 45
195     if (II != MBB.end())
196       foldFrameOffset(II, Offset, DstReg);
197 
198     // Select the best opcode based on DstReg and the offset size.
199     switch (DstReg) {
200     case AVR::R25R24:
201     case AVR::R27R26:
202     case AVR::R31R30: {
203       if (isUInt<6>(Offset) && STI.hasADDSUBIW()) {
204         Opcode = AVR::ADIWRdK;
205         break;
206       }
207       [[fallthrough]];
208     }
209     default: {
210       // This opcode will get expanded into a pair of subi/sbci.
211       Opcode = AVR::SUBIWRdK;
212       Offset = -Offset;
213       break;
214     }
215     }
216 
217     MachineInstr *New = BuildMI(MBB, II, dl, TII.get(Opcode), DstReg)
218                             .addReg(DstReg, RegState::Kill)
219                             .addImm(Offset);
220     New->getOperand(3).setIsDead();
221 
222     MI.eraseFromParent(); // remove FRMIDX
223 
224     return false;
225   }
226 
227   // On most AVRs, we can use an offset up to 62 for load/store with
228   // displacement (63 for byte values, 62 for word values). However, the
229   // "reduced tiny" cores don't support load/store with displacement. So for
230   // them, we force an offset of 0 meaning that any positive offset will require
231   // adjusting the frame pointer.
232   int MaxOffset = STI.hasTinyEncoding() ? 0 : 62;
233 
234   // If the offset is too big we have to adjust and restore the frame pointer
235   // to materialize a valid load/store with displacement.
236   //: TODO: consider using only one adiw/sbiw chain for more than one frame
237   //: index
238   if (Offset > MaxOffset) {
239     unsigned AddOpc = AVR::ADIWRdK, SubOpc = AVR::SBIWRdK;
240     int AddOffset = Offset - MaxOffset;
241 
242     // For huge offsets where adiw/sbiw cannot be used use a pair of subi/sbci.
243     if ((Offset - MaxOffset) > 63 || !STI.hasADDSUBIW()) {
244       AddOpc = AVR::SUBIWRdK;
245       SubOpc = AVR::SUBIWRdK;
246       AddOffset = -AddOffset;
247     }
248 
249     // It is possible that the spiller places this frame instruction in between
250     // a compare and branch, invalidating the contents of SREG set by the
251     // compare instruction because of the add/sub pairs. Conservatively save and
252     // restore SREG before and after each add/sub pair.
253     BuildMI(MBB, II, dl, TII.get(AVR::INRdA), STI.getTmpRegister())
254         .addImm(STI.getIORegSREG());
255 
256     MachineInstr *New = BuildMI(MBB, II, dl, TII.get(AddOpc), AVR::R29R28)
257                             .addReg(AVR::R29R28, RegState::Kill)
258                             .addImm(AddOffset);
259     New->getOperand(3).setIsDead();
260 
261     // Restore SREG.
262     BuildMI(MBB, std::next(II), dl, TII.get(AVR::OUTARr))
263         .addImm(STI.getIORegSREG())
264         .addReg(STI.getTmpRegister(), RegState::Kill);
265 
266     // No need to set SREG as dead here otherwise if the next instruction is a
267     // cond branch it will be using a dead register.
268     BuildMI(MBB, std::next(II), dl, TII.get(SubOpc), AVR::R29R28)
269         .addReg(AVR::R29R28, RegState::Kill)
270         .addImm(Offset - MaxOffset);
271 
272     Offset = MaxOffset;
273   }
274 
275   MI.getOperand(FIOperandNum).ChangeToRegister(AVR::R29R28, false);
276   assert(isUInt<6>(Offset) && "Offset is out of range");
277   MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
278   return false;
279 }
280 
getFrameRegister(const MachineFunction & MF) const281 Register AVRRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
282   const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering();
283   if (TFI->hasFP(MF)) {
284     // The Y pointer register
285     return AVR::R28;
286   }
287 
288   return AVR::SP;
289 }
290 
291 const TargetRegisterClass *
getPointerRegClass(const MachineFunction & MF,unsigned Kind) const292 AVRRegisterInfo::getPointerRegClass(const MachineFunction &MF,
293                                     unsigned Kind) const {
294   // FIXME: Currently we're using avr-gcc as reference, so we restrict
295   // ptrs to Y and Z regs. Though avr-gcc has buggy implementation
296   // of memory constraint, so we can fix it and bit avr-gcc here ;-)
297   return &AVR::PTRDISPREGSRegClass;
298 }
299 
splitReg(Register Reg,Register & LoReg,Register & HiReg) const300 void AVRRegisterInfo::splitReg(Register Reg, Register &LoReg,
301                                Register &HiReg) const {
302   assert(AVR::DREGSRegClass.contains(Reg) && "can only split 16-bit registers");
303 
304   LoReg = getSubReg(Reg, AVR::sub_lo);
305   HiReg = getSubReg(Reg, AVR::sub_hi);
306 }
307 
shouldCoalesce(MachineInstr * MI,const TargetRegisterClass * SrcRC,unsigned SubReg,const TargetRegisterClass * DstRC,unsigned DstSubReg,const TargetRegisterClass * NewRC,LiveIntervals & LIS) const308 bool AVRRegisterInfo::shouldCoalesce(
309     MachineInstr *MI, const TargetRegisterClass *SrcRC, unsigned SubReg,
310     const TargetRegisterClass *DstRC, unsigned DstSubReg,
311     const TargetRegisterClass *NewRC, LiveIntervals &LIS) const {
312   if (this->getRegClass(AVR::PTRDISPREGSRegClassID)->hasSubClassEq(NewRC)) {
313     return false;
314   }
315 
316   return TargetRegisterInfo::shouldCoalesce(MI, SrcRC, SubReg, DstRC, DstSubReg,
317                                             NewRC, LIS);
318 }
319 
320 } // end of namespace llvm
321