1 //===-- ARMExpandPseudoInsts.cpp - Expand pseudo instructions -------------===//
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
7 //===----------------------------------------------------------------------===//
9 // This file contains a pass that expands pseudo instructions into target
10 // instructions to allow proper scheduling, if-conversion, and other late
11 // optimizations. This pass should be run after register allocation but before
12 // the post-regalloc scheduling pass.
14 //===----------------------------------------------------------------------===//
17 #include "ARMBaseInstrInfo.h"
18 #include "ARMBaseRegisterInfo.h"
19 #include "ARMConstantPoolValue.h"
20 #include "ARMMachineFunctionInfo.h"
21 #include "ARMSubtarget.h"
22 #include "MCTargetDesc/ARMAddressingModes.h"
23 #include "llvm/CodeGen/LivePhysRegs.h"
24 #include "llvm/CodeGen/MachineFrameInfo.h"
25 #include "llvm/CodeGen/MachineFunctionPass.h"
26 #include "llvm/Support/Debug.h"
30 #define DEBUG_TYPE "arm-pseudo"
33 VerifyARMPseudo("verify-arm-pseudo-expand", cl::Hidden,
34 cl::desc("Verify machine code after expanding ARM pseudos"));
36 #define ARM_EXPAND_PSEUDO_NAME "ARM pseudo instruction expansion pass"
39 class ARMExpandPseudo : public MachineFunctionPass {
42 ARMExpandPseudo() : MachineFunctionPass(ID) {}
44 const ARMBaseInstrInfo *TII;
45 const TargetRegisterInfo *TRI;
46 const ARMSubtarget *STI;
49 bool runOnMachineFunction(MachineFunction &Fn) override;
51 MachineFunctionProperties getRequiredProperties() const override {
52 return MachineFunctionProperties().set(
53 MachineFunctionProperties::Property::NoVRegs);
56 StringRef getPassName() const override {
57 return ARM_EXPAND_PSEUDO_NAME;
61 void TransferImpOps(MachineInstr &OldMI,
62 MachineInstrBuilder &UseMI, MachineInstrBuilder &DefMI);
63 bool ExpandMI(MachineBasicBlock &MBB,
64 MachineBasicBlock::iterator MBBI,
65 MachineBasicBlock::iterator &NextMBBI);
66 bool ExpandMBB(MachineBasicBlock &MBB);
67 void ExpandVLD(MachineBasicBlock::iterator &MBBI);
68 void ExpandVST(MachineBasicBlock::iterator &MBBI);
69 void ExpandLaneOp(MachineBasicBlock::iterator &MBBI);
70 void ExpandVTBL(MachineBasicBlock::iterator &MBBI,
71 unsigned Opc, bool IsExt);
72 void ExpandMOV32BitImm(MachineBasicBlock &MBB,
73 MachineBasicBlock::iterator &MBBI);
74 bool ExpandCMP_SWAP(MachineBasicBlock &MBB,
75 MachineBasicBlock::iterator MBBI, unsigned LdrexOp,
76 unsigned StrexOp, unsigned UxtOp,
77 MachineBasicBlock::iterator &NextMBBI);
79 bool ExpandCMP_SWAP_64(MachineBasicBlock &MBB,
80 MachineBasicBlock::iterator MBBI,
81 MachineBasicBlock::iterator &NextMBBI);
83 char ARMExpandPseudo::ID = 0;
86 INITIALIZE_PASS(ARMExpandPseudo, DEBUG_TYPE, ARM_EXPAND_PSEUDO_NAME, false,
89 /// TransferImpOps - Transfer implicit operands on the pseudo instruction to
90 /// the instructions created from the expansion.
91 void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI,
92 MachineInstrBuilder &UseMI,
93 MachineInstrBuilder &DefMI) {
94 const MCInstrDesc &Desc = OldMI.getDesc();
95 for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands();
97 const MachineOperand &MO = OldMI.getOperand(i);
98 assert(MO.isReg() && MO.getReg());
107 // Constants for register spacing in NEON load/store instructions.
108 // For quad-register load-lane and store-lane pseudo instructors, the
109 // spacing is initially assumed to be EvenDblSpc, and that is changed to
110 // OddDblSpc depending on the lane number operand.
111 enum NEONRegSpacing {
113 SingleLowSpc , // Single spacing, low registers, three and four vectors.
114 SingleHighQSpc, // Single spacing, high registers, four vectors.
115 SingleHighTSpc, // Single spacing, high registers, three vectors.
120 // Entries for NEON load/store information table. The table is sorted by
121 // PseudoOpc for fast binary-search lookups.
122 struct NEONLdStTableEntry {
127 bool hasWritebackOperand;
128 uint8_t RegSpacing; // One of type NEONRegSpacing
129 uint8_t NumRegs; // D registers loaded or stored
130 uint8_t RegElts; // elements per D register; used for lane ops
131 // FIXME: Temporary flag to denote whether the real instruction takes
132 // a single register (like the encoding) or all of the registers in
133 // the list (like the asm syntax and the isel DAG). When all definitions
134 // are converted to take only the single encoded register, this will
136 bool copyAllListRegs;
138 // Comparison methods for binary search of the table.
139 bool operator<(const NEONLdStTableEntry &TE) const {
140 return PseudoOpc < TE.PseudoOpc;
142 friend bool operator<(const NEONLdStTableEntry &TE, unsigned PseudoOpc) {
143 return TE.PseudoOpc < PseudoOpc;
145 friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned PseudoOpc,
146 const NEONLdStTableEntry &TE) {
147 return PseudoOpc < TE.PseudoOpc;
152 static const NEONLdStTableEntry NEONLdStTable[] = {
153 { ARM::VLD1LNq16Pseudo, ARM::VLD1LNd16, true, false, false, EvenDblSpc, 1, 4 ,true},
154 { ARM::VLD1LNq16Pseudo_UPD, ARM::VLD1LNd16_UPD, true, true, true, EvenDblSpc, 1, 4 ,true},
155 { ARM::VLD1LNq32Pseudo, ARM::VLD1LNd32, true, false, false, EvenDblSpc, 1, 2 ,true},
156 { ARM::VLD1LNq32Pseudo_UPD, ARM::VLD1LNd32_UPD, true, true, true, EvenDblSpc, 1, 2 ,true},
157 { ARM::VLD1LNq8Pseudo, ARM::VLD1LNd8, true, false, false, EvenDblSpc, 1, 8 ,true},
158 { ARM::VLD1LNq8Pseudo_UPD, ARM::VLD1LNd8_UPD, true, true, true, EvenDblSpc, 1, 8 ,true},
160 { ARM::VLD1d16QPseudo, ARM::VLD1d16Q, true, false, false, SingleSpc, 4, 4 ,false},
161 { ARM::VLD1d16TPseudo, ARM::VLD1d16T, true, false, false, SingleSpc, 3, 4 ,false},
162 { ARM::VLD1d32QPseudo, ARM::VLD1d32Q, true, false, false, SingleSpc, 4, 2 ,false},
163 { ARM::VLD1d32TPseudo, ARM::VLD1d32T, true, false, false, SingleSpc, 3, 2 ,false},
164 { ARM::VLD1d64QPseudo, ARM::VLD1d64Q, true, false, false, SingleSpc, 4, 1 ,false},
165 { ARM::VLD1d64QPseudoWB_fixed, ARM::VLD1d64Qwb_fixed, true, true, false, SingleSpc, 4, 1 ,false},
166 { ARM::VLD1d64QPseudoWB_register, ARM::VLD1d64Qwb_register, true, true, true, SingleSpc, 4, 1 ,false},
167 { ARM::VLD1d64TPseudo, ARM::VLD1d64T, true, false, false, SingleSpc, 3, 1 ,false},
168 { ARM::VLD1d64TPseudoWB_fixed, ARM::VLD1d64Twb_fixed, true, true, false, SingleSpc, 3, 1 ,false},
169 { ARM::VLD1d64TPseudoWB_register, ARM::VLD1d64Twb_register, true, true, true, SingleSpc, 3, 1 ,false},
170 { ARM::VLD1d8QPseudo, ARM::VLD1d8Q, true, false, false, SingleSpc, 4, 8 ,false},
171 { ARM::VLD1d8TPseudo, ARM::VLD1d8T, true, false, false, SingleSpc, 3, 8 ,false},
172 { ARM::VLD1q16HighQPseudo, ARM::VLD1d16Q, true, false, false, SingleHighQSpc, 4, 4 ,false},
173 { ARM::VLD1q16HighTPseudo, ARM::VLD1d16T, true, false, false, SingleHighTSpc, 3, 4 ,false},
174 { ARM::VLD1q16LowQPseudo_UPD, ARM::VLD1d16Qwb_fixed, true, true, true, SingleLowSpc, 4, 4 ,false},
175 { ARM::VLD1q16LowTPseudo_UPD, ARM::VLD1d16Twb_fixed, true, true, true, SingleLowSpc, 3, 4 ,false},
176 { ARM::VLD1q32HighQPseudo, ARM::VLD1d32Q, true, false, false, SingleHighQSpc, 4, 2 ,false},
177 { ARM::VLD1q32HighTPseudo, ARM::VLD1d32T, true, false, false, SingleHighTSpc, 3, 2 ,false},
178 { ARM::VLD1q32LowQPseudo_UPD, ARM::VLD1d32Qwb_fixed, true, true, true, SingleLowSpc, 4, 2 ,false},
179 { ARM::VLD1q32LowTPseudo_UPD, ARM::VLD1d32Twb_fixed, true, true, true, SingleLowSpc, 3, 2 ,false},
180 { ARM::VLD1q64HighQPseudo, ARM::VLD1d64Q, true, false, false, SingleHighQSpc, 4, 1 ,false},
181 { ARM::VLD1q64HighTPseudo, ARM::VLD1d64T, true, false, false, SingleHighTSpc, 3, 1 ,false},
182 { ARM::VLD1q64LowQPseudo_UPD, ARM::VLD1d64Qwb_fixed, true, true, true, SingleLowSpc, 4, 1 ,false},
183 { ARM::VLD1q64LowTPseudo_UPD, ARM::VLD1d64Twb_fixed, true, true, true, SingleLowSpc, 3, 1 ,false},
184 { ARM::VLD1q8HighQPseudo, ARM::VLD1d8Q, true, false, false, SingleHighQSpc, 4, 8 ,false},
185 { ARM::VLD1q8HighTPseudo, ARM::VLD1d8T, true, false, false, SingleHighTSpc, 3, 8 ,false},
186 { ARM::VLD1q8LowQPseudo_UPD, ARM::VLD1d8Qwb_fixed, true, true, true, SingleLowSpc, 4, 8 ,false},
187 { ARM::VLD1q8LowTPseudo_UPD, ARM::VLD1d8Twb_fixed, true, true, true, SingleLowSpc, 3, 8 ,false},
189 { ARM::VLD2DUPq16EvenPseudo, ARM::VLD2DUPd16x2, true, false, false, EvenDblSpc, 2, 4 ,false},
190 { ARM::VLD2DUPq16OddPseudo, ARM::VLD2DUPd16x2, true, false, false, OddDblSpc, 2, 4 ,false},
191 { ARM::VLD2DUPq32EvenPseudo, ARM::VLD2DUPd32x2, true, false, false, EvenDblSpc, 2, 2 ,false},
192 { ARM::VLD2DUPq32OddPseudo, ARM::VLD2DUPd32x2, true, false, false, OddDblSpc, 2, 2 ,false},
193 { ARM::VLD2DUPq8EvenPseudo, ARM::VLD2DUPd8x2, true, false, false, EvenDblSpc, 2, 8 ,false},
194 { ARM::VLD2DUPq8OddPseudo, ARM::VLD2DUPd8x2, true, false, false, OddDblSpc, 2, 8 ,false},
196 { ARM::VLD2LNd16Pseudo, ARM::VLD2LNd16, true, false, false, SingleSpc, 2, 4 ,true},
197 { ARM::VLD2LNd16Pseudo_UPD, ARM::VLD2LNd16_UPD, true, true, true, SingleSpc, 2, 4 ,true},
198 { ARM::VLD2LNd32Pseudo, ARM::VLD2LNd32, true, false, false, SingleSpc, 2, 2 ,true},
199 { ARM::VLD2LNd32Pseudo_UPD, ARM::VLD2LNd32_UPD, true, true, true, SingleSpc, 2, 2 ,true},
200 { ARM::VLD2LNd8Pseudo, ARM::VLD2LNd8, true, false, false, SingleSpc, 2, 8 ,true},
201 { ARM::VLD2LNd8Pseudo_UPD, ARM::VLD2LNd8_UPD, true, true, true, SingleSpc, 2, 8 ,true},
202 { ARM::VLD2LNq16Pseudo, ARM::VLD2LNq16, true, false, false, EvenDblSpc, 2, 4 ,true},
203 { ARM::VLD2LNq16Pseudo_UPD, ARM::VLD2LNq16_UPD, true, true, true, EvenDblSpc, 2, 4 ,true},
204 { ARM::VLD2LNq32Pseudo, ARM::VLD2LNq32, true, false, false, EvenDblSpc, 2, 2 ,true},
205 { ARM::VLD2LNq32Pseudo_UPD, ARM::VLD2LNq32_UPD, true, true, true, EvenDblSpc, 2, 2 ,true},
207 { ARM::VLD2q16Pseudo, ARM::VLD2q16, true, false, false, SingleSpc, 4, 4 ,false},
208 { ARM::VLD2q16PseudoWB_fixed, ARM::VLD2q16wb_fixed, true, true, false, SingleSpc, 4, 4 ,false},
209 { ARM::VLD2q16PseudoWB_register, ARM::VLD2q16wb_register, true, true, true, SingleSpc, 4, 4 ,false},
210 { ARM::VLD2q32Pseudo, ARM::VLD2q32, true, false, false, SingleSpc, 4, 2 ,false},
211 { ARM::VLD2q32PseudoWB_fixed, ARM::VLD2q32wb_fixed, true, true, false, SingleSpc, 4, 2 ,false},
212 { ARM::VLD2q32PseudoWB_register, ARM::VLD2q32wb_register, true, true, true, SingleSpc, 4, 2 ,false},
213 { ARM::VLD2q8Pseudo, ARM::VLD2q8, true, false, false, SingleSpc, 4, 8 ,false},
214 { ARM::VLD2q8PseudoWB_fixed, ARM::VLD2q8wb_fixed, true, true, false, SingleSpc, 4, 8 ,false},
215 { ARM::VLD2q8PseudoWB_register, ARM::VLD2q8wb_register, true, true, true, SingleSpc, 4, 8 ,false},
217 { ARM::VLD3DUPd16Pseudo, ARM::VLD3DUPd16, true, false, false, SingleSpc, 3, 4,true},
218 { ARM::VLD3DUPd16Pseudo_UPD, ARM::VLD3DUPd16_UPD, true, true, true, SingleSpc, 3, 4,true},
219 { ARM::VLD3DUPd32Pseudo, ARM::VLD3DUPd32, true, false, false, SingleSpc, 3, 2,true},
220 { ARM::VLD3DUPd32Pseudo_UPD, ARM::VLD3DUPd32_UPD, true, true, true, SingleSpc, 3, 2,true},
221 { ARM::VLD3DUPd8Pseudo, ARM::VLD3DUPd8, true, false, false, SingleSpc, 3, 8,true},
222 { ARM::VLD3DUPd8Pseudo_UPD, ARM::VLD3DUPd8_UPD, true, true, true, SingleSpc, 3, 8,true},
223 { ARM::VLD3DUPq16EvenPseudo, ARM::VLD3DUPq16, true, false, false, EvenDblSpc, 3, 4 ,true},
224 { ARM::VLD3DUPq16OddPseudo, ARM::VLD3DUPq16, true, false, false, OddDblSpc, 3, 4 ,true},
225 { ARM::VLD3DUPq32EvenPseudo, ARM::VLD3DUPq32, true, false, false, EvenDblSpc, 3, 2 ,true},
226 { ARM::VLD3DUPq32OddPseudo, ARM::VLD3DUPq32, true, false, false, OddDblSpc, 3, 2 ,true},
227 { ARM::VLD3DUPq8EvenPseudo, ARM::VLD3DUPq8, true, false, false, EvenDblSpc, 3, 8 ,true},
228 { ARM::VLD3DUPq8OddPseudo, ARM::VLD3DUPq8, true, false, false, OddDblSpc, 3, 8 ,true},
230 { ARM::VLD3LNd16Pseudo, ARM::VLD3LNd16, true, false, false, SingleSpc, 3, 4 ,true},
231 { ARM::VLD3LNd16Pseudo_UPD, ARM::VLD3LNd16_UPD, true, true, true, SingleSpc, 3, 4 ,true},
232 { ARM::VLD3LNd32Pseudo, ARM::VLD3LNd32, true, false, false, SingleSpc, 3, 2 ,true},
233 { ARM::VLD3LNd32Pseudo_UPD, ARM::VLD3LNd32_UPD, true, true, true, SingleSpc, 3, 2 ,true},
234 { ARM::VLD3LNd8Pseudo, ARM::VLD3LNd8, true, false, false, SingleSpc, 3, 8 ,true},
235 { ARM::VLD3LNd8Pseudo_UPD, ARM::VLD3LNd8_UPD, true, true, true, SingleSpc, 3, 8 ,true},
236 { ARM::VLD3LNq16Pseudo, ARM::VLD3LNq16, true, false, false, EvenDblSpc, 3, 4 ,true},
237 { ARM::VLD3LNq16Pseudo_UPD, ARM::VLD3LNq16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true},
238 { ARM::VLD3LNq32Pseudo, ARM::VLD3LNq32, true, false, false, EvenDblSpc, 3, 2 ,true},
239 { ARM::VLD3LNq32Pseudo_UPD, ARM::VLD3LNq32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true},
241 { ARM::VLD3d16Pseudo, ARM::VLD3d16, true, false, false, SingleSpc, 3, 4 ,true},
242 { ARM::VLD3d16Pseudo_UPD, ARM::VLD3d16_UPD, true, true, true, SingleSpc, 3, 4 ,true},
243 { ARM::VLD3d32Pseudo, ARM::VLD3d32, true, false, false, SingleSpc, 3, 2 ,true},
244 { ARM::VLD3d32Pseudo_UPD, ARM::VLD3d32_UPD, true, true, true, SingleSpc, 3, 2 ,true},
245 { ARM::VLD3d8Pseudo, ARM::VLD3d8, true, false, false, SingleSpc, 3, 8 ,true},
246 { ARM::VLD3d8Pseudo_UPD, ARM::VLD3d8_UPD, true, true, true, SingleSpc, 3, 8 ,true},
248 { ARM::VLD3q16Pseudo_UPD, ARM::VLD3q16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true},
249 { ARM::VLD3q16oddPseudo, ARM::VLD3q16, true, false, false, OddDblSpc, 3, 4 ,true},
250 { ARM::VLD3q16oddPseudo_UPD, ARM::VLD3q16_UPD, true, true, true, OddDblSpc, 3, 4 ,true},
251 { ARM::VLD3q32Pseudo_UPD, ARM::VLD3q32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true},
252 { ARM::VLD3q32oddPseudo, ARM::VLD3q32, true, false, false, OddDblSpc, 3, 2 ,true},
253 { ARM::VLD3q32oddPseudo_UPD, ARM::VLD3q32_UPD, true, true, true, OddDblSpc, 3, 2 ,true},
254 { ARM::VLD3q8Pseudo_UPD, ARM::VLD3q8_UPD, true, true, true, EvenDblSpc, 3, 8 ,true},
255 { ARM::VLD3q8oddPseudo, ARM::VLD3q8, true, false, false, OddDblSpc, 3, 8 ,true},
256 { ARM::VLD3q8oddPseudo_UPD, ARM::VLD3q8_UPD, true, true, true, OddDblSpc, 3, 8 ,true},
258 { ARM::VLD4DUPd16Pseudo, ARM::VLD4DUPd16, true, false, false, SingleSpc, 4, 4,true},
259 { ARM::VLD4DUPd16Pseudo_UPD, ARM::VLD4DUPd16_UPD, true, true, true, SingleSpc, 4, 4,true},
260 { ARM::VLD4DUPd32Pseudo, ARM::VLD4DUPd32, true, false, false, SingleSpc, 4, 2,true},
261 { ARM::VLD4DUPd32Pseudo_UPD, ARM::VLD4DUPd32_UPD, true, true, true, SingleSpc, 4, 2,true},
262 { ARM::VLD4DUPd8Pseudo, ARM::VLD4DUPd8, true, false, false, SingleSpc, 4, 8,true},
263 { ARM::VLD4DUPd8Pseudo_UPD, ARM::VLD4DUPd8_UPD, true, true, true, SingleSpc, 4, 8,true},
264 { ARM::VLD4DUPq16EvenPseudo, ARM::VLD4DUPq16, true, false, false, EvenDblSpc, 4, 4 ,true},
265 { ARM::VLD4DUPq16OddPseudo, ARM::VLD4DUPq16, true, false, false, OddDblSpc, 4, 4 ,true},
266 { ARM::VLD4DUPq32EvenPseudo, ARM::VLD4DUPq32, true, false, false, EvenDblSpc, 4, 2 ,true},
267 { ARM::VLD4DUPq32OddPseudo, ARM::VLD4DUPq32, true, false, false, OddDblSpc, 4, 2 ,true},
268 { ARM::VLD4DUPq8EvenPseudo, ARM::VLD4DUPq8, true, false, false, EvenDblSpc, 4, 8 ,true},
269 { ARM::VLD4DUPq8OddPseudo, ARM::VLD4DUPq8, true, false, false, OddDblSpc, 4, 8 ,true},
271 { ARM::VLD4LNd16Pseudo, ARM::VLD4LNd16, true, false, false, SingleSpc, 4, 4 ,true},
272 { ARM::VLD4LNd16Pseudo_UPD, ARM::VLD4LNd16_UPD, true, true, true, SingleSpc, 4, 4 ,true},
273 { ARM::VLD4LNd32Pseudo, ARM::VLD4LNd32, true, false, false, SingleSpc, 4, 2 ,true},
274 { ARM::VLD4LNd32Pseudo_UPD, ARM::VLD4LNd32_UPD, true, true, true, SingleSpc, 4, 2 ,true},
275 { ARM::VLD4LNd8Pseudo, ARM::VLD4LNd8, true, false, false, SingleSpc, 4, 8 ,true},
276 { ARM::VLD4LNd8Pseudo_UPD, ARM::VLD4LNd8_UPD, true, true, true, SingleSpc, 4, 8 ,true},
277 { ARM::VLD4LNq16Pseudo, ARM::VLD4LNq16, true, false, false, EvenDblSpc, 4, 4 ,true},
278 { ARM::VLD4LNq16Pseudo_UPD, ARM::VLD4LNq16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true},
279 { ARM::VLD4LNq32Pseudo, ARM::VLD4LNq32, true, false, false, EvenDblSpc, 4, 2 ,true},
280 { ARM::VLD4LNq32Pseudo_UPD, ARM::VLD4LNq32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true},
282 { ARM::VLD4d16Pseudo, ARM::VLD4d16, true, false, false, SingleSpc, 4, 4 ,true},
283 { ARM::VLD4d16Pseudo_UPD, ARM::VLD4d16_UPD, true, true, true, SingleSpc, 4, 4 ,true},
284 { ARM::VLD4d32Pseudo, ARM::VLD4d32, true, false, false, SingleSpc, 4, 2 ,true},
285 { ARM::VLD4d32Pseudo_UPD, ARM::VLD4d32_UPD, true, true, true, SingleSpc, 4, 2 ,true},
286 { ARM::VLD4d8Pseudo, ARM::VLD4d8, true, false, false, SingleSpc, 4, 8 ,true},
287 { ARM::VLD4d8Pseudo_UPD, ARM::VLD4d8_UPD, true, true, true, SingleSpc, 4, 8 ,true},
289 { ARM::VLD4q16Pseudo_UPD, ARM::VLD4q16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true},
290 { ARM::VLD4q16oddPseudo, ARM::VLD4q16, true, false, false, OddDblSpc, 4, 4 ,true},
291 { ARM::VLD4q16oddPseudo_UPD, ARM::VLD4q16_UPD, true, true, true, OddDblSpc, 4, 4 ,true},
292 { ARM::VLD4q32Pseudo_UPD, ARM::VLD4q32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true},
293 { ARM::VLD4q32oddPseudo, ARM::VLD4q32, true, false, false, OddDblSpc, 4, 2 ,true},
294 { ARM::VLD4q32oddPseudo_UPD, ARM::VLD4q32_UPD, true, true, true, OddDblSpc, 4, 2 ,true},
295 { ARM::VLD4q8Pseudo_UPD, ARM::VLD4q8_UPD, true, true, true, EvenDblSpc, 4, 8 ,true},
296 { ARM::VLD4q8oddPseudo, ARM::VLD4q8, true, false, false, OddDblSpc, 4, 8 ,true},
297 { ARM::VLD4q8oddPseudo_UPD, ARM::VLD4q8_UPD, true, true, true, OddDblSpc, 4, 8 ,true},
299 { ARM::VST1LNq16Pseudo, ARM::VST1LNd16, false, false, false, EvenDblSpc, 1, 4 ,true},
300 { ARM::VST1LNq16Pseudo_UPD, ARM::VST1LNd16_UPD, false, true, true, EvenDblSpc, 1, 4 ,true},
301 { ARM::VST1LNq32Pseudo, ARM::VST1LNd32, false, false, false, EvenDblSpc, 1, 2 ,true},
302 { ARM::VST1LNq32Pseudo_UPD, ARM::VST1LNd32_UPD, false, true, true, EvenDblSpc, 1, 2 ,true},
303 { ARM::VST1LNq8Pseudo, ARM::VST1LNd8, false, false, false, EvenDblSpc, 1, 8 ,true},
304 { ARM::VST1LNq8Pseudo_UPD, ARM::VST1LNd8_UPD, false, true, true, EvenDblSpc, 1, 8 ,true},
306 { ARM::VST1d16QPseudo, ARM::VST1d16Q, false, false, false, SingleSpc, 4, 4 ,false},
307 { ARM::VST1d16TPseudo, ARM::VST1d16T, false, false, false, SingleSpc, 3, 4 ,false},
308 { ARM::VST1d32QPseudo, ARM::VST1d32Q, false, false, false, SingleSpc, 4, 2 ,false},
309 { ARM::VST1d32TPseudo, ARM::VST1d32T, false, false, false, SingleSpc, 3, 2 ,false},
310 { ARM::VST1d64QPseudo, ARM::VST1d64Q, false, false, false, SingleSpc, 4, 1 ,false},
311 { ARM::VST1d64QPseudoWB_fixed, ARM::VST1d64Qwb_fixed, false, true, false, SingleSpc, 4, 1 ,false},
312 { ARM::VST1d64QPseudoWB_register, ARM::VST1d64Qwb_register, false, true, true, SingleSpc, 4, 1 ,false},
313 { ARM::VST1d64TPseudo, ARM::VST1d64T, false, false, false, SingleSpc, 3, 1 ,false},
314 { ARM::VST1d64TPseudoWB_fixed, ARM::VST1d64Twb_fixed, false, true, false, SingleSpc, 3, 1 ,false},
315 { ARM::VST1d64TPseudoWB_register, ARM::VST1d64Twb_register, false, true, true, SingleSpc, 3, 1 ,false},
316 { ARM::VST1d8QPseudo, ARM::VST1d8Q, false, false, false, SingleSpc, 4, 8 ,false},
317 { ARM::VST1d8TPseudo, ARM::VST1d8T, false, false, false, SingleSpc, 3, 8 ,false},
318 { ARM::VST1q16HighQPseudo, ARM::VST1d16Q, false, false, false, SingleHighQSpc, 4, 4 ,false},
319 { ARM::VST1q16HighTPseudo, ARM::VST1d16T, false, false, false, SingleHighTSpc, 3, 4 ,false},
320 { ARM::VST1q16LowQPseudo_UPD, ARM::VST1d16Qwb_fixed, false, true, true, SingleLowSpc, 4, 4 ,false},
321 { ARM::VST1q16LowTPseudo_UPD, ARM::VST1d16Twb_fixed, false, true, true, SingleLowSpc, 3, 4 ,false},
322 { ARM::VST1q32HighQPseudo, ARM::VST1d32Q, false, false, false, SingleHighQSpc, 4, 2 ,false},
323 { ARM::VST1q32HighTPseudo, ARM::VST1d32T, false, false, false, SingleHighTSpc, 3, 2 ,false},
324 { ARM::VST1q32LowQPseudo_UPD, ARM::VST1d32Qwb_fixed, false, true, true, SingleLowSpc, 4, 2 ,false},
325 { ARM::VST1q32LowTPseudo_UPD, ARM::VST1d32Twb_fixed, false, true, true, SingleLowSpc, 3, 2 ,false},
326 { ARM::VST1q64HighQPseudo, ARM::VST1d64Q, false, false, false, SingleHighQSpc, 4, 1 ,false},
327 { ARM::VST1q64HighTPseudo, ARM::VST1d64T, false, false, false, SingleHighTSpc, 3, 1 ,false},
328 { ARM::VST1q64LowQPseudo_UPD, ARM::VST1d64Qwb_fixed, false, true, true, SingleLowSpc, 4, 1 ,false},
329 { ARM::VST1q64LowTPseudo_UPD, ARM::VST1d64Twb_fixed, false, true, true, SingleLowSpc, 3, 1 ,false},
330 { ARM::VST1q8HighQPseudo, ARM::VST1d8Q, false, false, false, SingleHighQSpc, 4, 8 ,false},
331 { ARM::VST1q8HighTPseudo, ARM::VST1d8T, false, false, false, SingleHighTSpc, 3, 8 ,false},
332 { ARM::VST1q8LowQPseudo_UPD, ARM::VST1d8Qwb_fixed, false, true, true, SingleLowSpc, 4, 8 ,false},
333 { ARM::VST1q8LowTPseudo_UPD, ARM::VST1d8Twb_fixed, false, true, true, SingleLowSpc, 3, 8 ,false},
335 { ARM::VST2LNd16Pseudo, ARM::VST2LNd16, false, false, false, SingleSpc, 2, 4 ,true},
336 { ARM::VST2LNd16Pseudo_UPD, ARM::VST2LNd16_UPD, false, true, true, SingleSpc, 2, 4 ,true},
337 { ARM::VST2LNd32Pseudo, ARM::VST2LNd32, false, false, false, SingleSpc, 2, 2 ,true},
338 { ARM::VST2LNd32Pseudo_UPD, ARM::VST2LNd32_UPD, false, true, true, SingleSpc, 2, 2 ,true},
339 { ARM::VST2LNd8Pseudo, ARM::VST2LNd8, false, false, false, SingleSpc, 2, 8 ,true},
340 { ARM::VST2LNd8Pseudo_UPD, ARM::VST2LNd8_UPD, false, true, true, SingleSpc, 2, 8 ,true},
341 { ARM::VST2LNq16Pseudo, ARM::VST2LNq16, false, false, false, EvenDblSpc, 2, 4,true},
342 { ARM::VST2LNq16Pseudo_UPD, ARM::VST2LNq16_UPD, false, true, true, EvenDblSpc, 2, 4,true},
343 { ARM::VST2LNq32Pseudo, ARM::VST2LNq32, false, false, false, EvenDblSpc, 2, 2,true},
344 { ARM::VST2LNq32Pseudo_UPD, ARM::VST2LNq32_UPD, false, true, true, EvenDblSpc, 2, 2,true},
346 { ARM::VST2q16Pseudo, ARM::VST2q16, false, false, false, SingleSpc, 4, 4 ,false},
347 { ARM::VST2q16PseudoWB_fixed, ARM::VST2q16wb_fixed, false, true, false, SingleSpc, 4, 4 ,false},
348 { ARM::VST2q16PseudoWB_register, ARM::VST2q16wb_register, false, true, true, SingleSpc, 4, 4 ,false},
349 { ARM::VST2q32Pseudo, ARM::VST2q32, false, false, false, SingleSpc, 4, 2 ,false},
350 { ARM::VST2q32PseudoWB_fixed, ARM::VST2q32wb_fixed, false, true, false, SingleSpc, 4, 2 ,false},
351 { ARM::VST2q32PseudoWB_register, ARM::VST2q32wb_register, false, true, true, SingleSpc, 4, 2 ,false},
352 { ARM::VST2q8Pseudo, ARM::VST2q8, false, false, false, SingleSpc, 4, 8 ,false},
353 { ARM::VST2q8PseudoWB_fixed, ARM::VST2q8wb_fixed, false, true, false, SingleSpc, 4, 8 ,false},
354 { ARM::VST2q8PseudoWB_register, ARM::VST2q8wb_register, false, true, true, SingleSpc, 4, 8 ,false},
356 { ARM::VST3LNd16Pseudo, ARM::VST3LNd16, false, false, false, SingleSpc, 3, 4 ,true},
357 { ARM::VST3LNd16Pseudo_UPD, ARM::VST3LNd16_UPD, false, true, true, SingleSpc, 3, 4 ,true},
358 { ARM::VST3LNd32Pseudo, ARM::VST3LNd32, false, false, false, SingleSpc, 3, 2 ,true},
359 { ARM::VST3LNd32Pseudo_UPD, ARM::VST3LNd32_UPD, false, true, true, SingleSpc, 3, 2 ,true},
360 { ARM::VST3LNd8Pseudo, ARM::VST3LNd8, false, false, false, SingleSpc, 3, 8 ,true},
361 { ARM::VST3LNd8Pseudo_UPD, ARM::VST3LNd8_UPD, false, true, true, SingleSpc, 3, 8 ,true},
362 { ARM::VST3LNq16Pseudo, ARM::VST3LNq16, false, false, false, EvenDblSpc, 3, 4,true},
363 { ARM::VST3LNq16Pseudo_UPD, ARM::VST3LNq16_UPD, false, true, true, EvenDblSpc, 3, 4,true},
364 { ARM::VST3LNq32Pseudo, ARM::VST3LNq32, false, false, false, EvenDblSpc, 3, 2,true},
365 { ARM::VST3LNq32Pseudo_UPD, ARM::VST3LNq32_UPD, false, true, true, EvenDblSpc, 3, 2,true},
367 { ARM::VST3d16Pseudo, ARM::VST3d16, false, false, false, SingleSpc, 3, 4 ,true},
368 { ARM::VST3d16Pseudo_UPD, ARM::VST3d16_UPD, false, true, true, SingleSpc, 3, 4 ,true},
369 { ARM::VST3d32Pseudo, ARM::VST3d32, false, false, false, SingleSpc, 3, 2 ,true},
370 { ARM::VST3d32Pseudo_UPD, ARM::VST3d32_UPD, false, true, true, SingleSpc, 3, 2 ,true},
371 { ARM::VST3d8Pseudo, ARM::VST3d8, false, false, false, SingleSpc, 3, 8 ,true},
372 { ARM::VST3d8Pseudo_UPD, ARM::VST3d8_UPD, false, true, true, SingleSpc, 3, 8 ,true},
374 { ARM::VST3q16Pseudo_UPD, ARM::VST3q16_UPD, false, true, true, EvenDblSpc, 3, 4 ,true},
375 { ARM::VST3q16oddPseudo, ARM::VST3q16, false, false, false, OddDblSpc, 3, 4 ,true},
376 { ARM::VST3q16oddPseudo_UPD, ARM::VST3q16_UPD, false, true, true, OddDblSpc, 3, 4 ,true},
377 { ARM::VST3q32Pseudo_UPD, ARM::VST3q32_UPD, false, true, true, EvenDblSpc, 3, 2 ,true},
378 { ARM::VST3q32oddPseudo, ARM::VST3q32, false, false, false, OddDblSpc, 3, 2 ,true},
379 { ARM::VST3q32oddPseudo_UPD, ARM::VST3q32_UPD, false, true, true, OddDblSpc, 3, 2 ,true},
380 { ARM::VST3q8Pseudo_UPD, ARM::VST3q8_UPD, false, true, true, EvenDblSpc, 3, 8 ,true},
381 { ARM::VST3q8oddPseudo, ARM::VST3q8, false, false, false, OddDblSpc, 3, 8 ,true},
382 { ARM::VST3q8oddPseudo_UPD, ARM::VST3q8_UPD, false, true, true, OddDblSpc, 3, 8 ,true},
384 { ARM::VST4LNd16Pseudo, ARM::VST4LNd16, false, false, false, SingleSpc, 4, 4 ,true},
385 { ARM::VST4LNd16Pseudo_UPD, ARM::VST4LNd16_UPD, false, true, true, SingleSpc, 4, 4 ,true},
386 { ARM::VST4LNd32Pseudo, ARM::VST4LNd32, false, false, false, SingleSpc, 4, 2 ,true},
387 { ARM::VST4LNd32Pseudo_UPD, ARM::VST4LNd32_UPD, false, true, true, SingleSpc, 4, 2 ,true},
388 { ARM::VST4LNd8Pseudo, ARM::VST4LNd8, false, false, false, SingleSpc, 4, 8 ,true},
389 { ARM::VST4LNd8Pseudo_UPD, ARM::VST4LNd8_UPD, false, true, true, SingleSpc, 4, 8 ,true},
390 { ARM::VST4LNq16Pseudo, ARM::VST4LNq16, false, false, false, EvenDblSpc, 4, 4,true},
391 { ARM::VST4LNq16Pseudo_UPD, ARM::VST4LNq16_UPD, false, true, true, EvenDblSpc, 4, 4,true},
392 { ARM::VST4LNq32Pseudo, ARM::VST4LNq32, false, false, false, EvenDblSpc, 4, 2,true},
393 { ARM::VST4LNq32Pseudo_UPD, ARM::VST4LNq32_UPD, false, true, true, EvenDblSpc, 4, 2,true},
395 { ARM::VST4d16Pseudo, ARM::VST4d16, false, false, false, SingleSpc, 4, 4 ,true},
396 { ARM::VST4d16Pseudo_UPD, ARM::VST4d16_UPD, false, true, true, SingleSpc, 4, 4 ,true},
397 { ARM::VST4d32Pseudo, ARM::VST4d32, false, false, false, SingleSpc, 4, 2 ,true},
398 { ARM::VST4d32Pseudo_UPD, ARM::VST4d32_UPD, false, true, true, SingleSpc, 4, 2 ,true},
399 { ARM::VST4d8Pseudo, ARM::VST4d8, false, false, false, SingleSpc, 4, 8 ,true},
400 { ARM::VST4d8Pseudo_UPD, ARM::VST4d8_UPD, false, true, true, SingleSpc, 4, 8 ,true},
402 { ARM::VST4q16Pseudo_UPD, ARM::VST4q16_UPD, false, true, true, EvenDblSpc, 4, 4 ,true},
403 { ARM::VST4q16oddPseudo, ARM::VST4q16, false, false, false, OddDblSpc, 4, 4 ,true},
404 { ARM::VST4q16oddPseudo_UPD, ARM::VST4q16_UPD, false, true, true, OddDblSpc, 4, 4 ,true},
405 { ARM::VST4q32Pseudo_UPD, ARM::VST4q32_UPD, false, true, true, EvenDblSpc, 4, 2 ,true},
406 { ARM::VST4q32oddPseudo, ARM::VST4q32, false, false, false, OddDblSpc, 4, 2 ,true},
407 { ARM::VST4q32oddPseudo_UPD, ARM::VST4q32_UPD, false, true, true, OddDblSpc, 4, 2 ,true},
408 { ARM::VST4q8Pseudo_UPD, ARM::VST4q8_UPD, false, true, true, EvenDblSpc, 4, 8 ,true},
409 { ARM::VST4q8oddPseudo, ARM::VST4q8, false, false, false, OddDblSpc, 4, 8 ,true},
410 { ARM::VST4q8oddPseudo_UPD, ARM::VST4q8_UPD, false, true, true, OddDblSpc, 4, 8 ,true}
413 /// LookupNEONLdSt - Search the NEONLdStTable for information about a NEON
414 /// load or store pseudo instruction.
415 static const NEONLdStTableEntry *LookupNEONLdSt(unsigned Opcode) {
417 // Make sure the table is sorted.
418 static std::atomic<bool> TableChecked(false);
419 if (!TableChecked.load(std::memory_order_relaxed)) {
420 assert(std::is_sorted(std::begin(NEONLdStTable), std::end(NEONLdStTable)) &&
421 "NEONLdStTable is not sorted!");
422 TableChecked.store(true, std::memory_order_relaxed);
426 auto I = llvm::lower_bound(NEONLdStTable, Opcode);
427 if (I != std::end(NEONLdStTable) && I->PseudoOpc == Opcode)
432 /// GetDSubRegs - Get 4 D subregisters of a Q, QQ, or QQQQ register,
433 /// corresponding to the specified register spacing. Not all of the results
434 /// are necessarily valid, e.g., a Q register only has 2 D subregisters.
435 static void GetDSubRegs(unsigned Reg, NEONRegSpacing RegSpc,
436 const TargetRegisterInfo *TRI, unsigned &D0,
437 unsigned &D1, unsigned &D2, unsigned &D3) {
438 if (RegSpc == SingleSpc || RegSpc == SingleLowSpc) {
439 D0 = TRI->getSubReg(Reg, ARM::dsub_0);
440 D1 = TRI->getSubReg(Reg, ARM::dsub_1);
441 D2 = TRI->getSubReg(Reg, ARM::dsub_2);
442 D3 = TRI->getSubReg(Reg, ARM::dsub_3);
443 } else if (RegSpc == SingleHighQSpc) {
444 D0 = TRI->getSubReg(Reg, ARM::dsub_4);
445 D1 = TRI->getSubReg(Reg, ARM::dsub_5);
446 D2 = TRI->getSubReg(Reg, ARM::dsub_6);
447 D3 = TRI->getSubReg(Reg, ARM::dsub_7);
448 } else if (RegSpc == SingleHighTSpc) {
449 D0 = TRI->getSubReg(Reg, ARM::dsub_3);
450 D1 = TRI->getSubReg(Reg, ARM::dsub_4);
451 D2 = TRI->getSubReg(Reg, ARM::dsub_5);
452 D3 = TRI->getSubReg(Reg, ARM::dsub_6);
453 } else if (RegSpc == EvenDblSpc) {
454 D0 = TRI->getSubReg(Reg, ARM::dsub_0);
455 D1 = TRI->getSubReg(Reg, ARM::dsub_2);
456 D2 = TRI->getSubReg(Reg, ARM::dsub_4);
457 D3 = TRI->getSubReg(Reg, ARM::dsub_6);
459 assert(RegSpc == OddDblSpc && "unknown register spacing");
460 D0 = TRI->getSubReg(Reg, ARM::dsub_1);
461 D1 = TRI->getSubReg(Reg, ARM::dsub_3);
462 D2 = TRI->getSubReg(Reg, ARM::dsub_5);
463 D3 = TRI->getSubReg(Reg, ARM::dsub_7);
467 /// ExpandVLD - Translate VLD pseudo instructions with Q, QQ or QQQQ register
468 /// operands to real VLD instructions with D register operands.
469 void ARMExpandPseudo::ExpandVLD(MachineBasicBlock::iterator &MBBI) {
470 MachineInstr &MI = *MBBI;
471 MachineBasicBlock &MBB = *MI.getParent();
472 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump());
474 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
475 assert(TableEntry && TableEntry->IsLoad && "NEONLdStTable lookup failed");
476 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing;
477 unsigned NumRegs = TableEntry->NumRegs;
479 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
480 TII->get(TableEntry->RealOpc));
483 bool DstIsDead = MI.getOperand(OpIdx).isDead();
484 Register DstReg = MI.getOperand(OpIdx++).getReg();
485 if(TableEntry->RealOpc == ARM::VLD2DUPd8x2 ||
486 TableEntry->RealOpc == ARM::VLD2DUPd16x2 ||
487 TableEntry->RealOpc == ARM::VLD2DUPd32x2) {
488 unsigned SubRegIndex;
489 if (RegSpc == EvenDblSpc) {
490 SubRegIndex = ARM::dsub_0;
492 assert(RegSpc == OddDblSpc && "Unexpected spacing!");
493 SubRegIndex = ARM::dsub_1;
495 Register SubReg = TRI->getSubReg(DstReg, SubRegIndex);
496 unsigned DstRegPair = TRI->getMatchingSuperReg(SubReg, ARM::dsub_0,
497 &ARM::DPairSpcRegClass);
498 MIB.addReg(DstRegPair, RegState::Define | getDeadRegState(DstIsDead));
500 unsigned D0, D1, D2, D3;
501 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3);
502 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead));
503 if (NumRegs > 1 && TableEntry->copyAllListRegs)
504 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
505 if (NumRegs > 2 && TableEntry->copyAllListRegs)
506 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead));
507 if (NumRegs > 3 && TableEntry->copyAllListRegs)
508 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead));
511 if (TableEntry->isUpdating)
512 MIB.add(MI.getOperand(OpIdx++));
514 // Copy the addrmode6 operands.
515 MIB.add(MI.getOperand(OpIdx++));
516 MIB.add(MI.getOperand(OpIdx++));
518 // Copy the am6offset operand.
519 if (TableEntry->hasWritebackOperand) {
520 // TODO: The writing-back pseudo instructions we translate here are all
521 // defined to take am6offset nodes that are capable to represent both fixed
522 // and register forms. Some real instructions, however, do not rely on
523 // am6offset and have separate definitions for such forms. When this is the
524 // case, fixed forms do not take any offset nodes, so here we skip them for
525 // such instructions. Once all real and pseudo writing-back instructions are
526 // rewritten without use of am6offset nodes, this code will go away.
527 const MachineOperand &AM6Offset = MI.getOperand(OpIdx++);
528 if (TableEntry->RealOpc == ARM::VLD1d8Qwb_fixed ||
529 TableEntry->RealOpc == ARM::VLD1d16Qwb_fixed ||
530 TableEntry->RealOpc == ARM::VLD1d32Qwb_fixed ||
531 TableEntry->RealOpc == ARM::VLD1d64Qwb_fixed ||
532 TableEntry->RealOpc == ARM::VLD1d8Twb_fixed ||
533 TableEntry->RealOpc == ARM::VLD1d16Twb_fixed ||
534 TableEntry->RealOpc == ARM::VLD1d32Twb_fixed ||
535 TableEntry->RealOpc == ARM::VLD1d64Twb_fixed) {
536 assert(AM6Offset.getReg() == 0 &&
537 "A fixed writing-back pseudo instruction provides an offset "
544 // For an instruction writing double-spaced subregs, the pseudo instruction
545 // has an extra operand that is a use of the super-register. Record the
546 // operand index and skip over it.
547 unsigned SrcOpIdx = 0;
548 if(TableEntry->RealOpc != ARM::VLD2DUPd8x2 &&
549 TableEntry->RealOpc != ARM::VLD2DUPd16x2 &&
550 TableEntry->RealOpc != ARM::VLD2DUPd32x2) {
551 if (RegSpc == EvenDblSpc || RegSpc == OddDblSpc ||
552 RegSpc == SingleLowSpc || RegSpc == SingleHighQSpc ||
553 RegSpc == SingleHighTSpc)
557 // Copy the predicate operands.
558 MIB.add(MI.getOperand(OpIdx++));
559 MIB.add(MI.getOperand(OpIdx++));
561 // Copy the super-register source operand used for double-spaced subregs over
562 // to the new instruction as an implicit operand.
564 MachineOperand MO = MI.getOperand(SrcOpIdx);
565 MO.setImplicit(true);
568 // Add an implicit def for the super-register.
569 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
570 TransferImpOps(MI, MIB, MIB);
572 // Transfer memoperands.
573 MIB.cloneMemRefs(MI);
574 MI.eraseFromParent();
575 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump(););
578 /// ExpandVST - Translate VST pseudo instructions with Q, QQ or QQQQ register
579 /// operands to real VST instructions with D register operands.
580 void ARMExpandPseudo::ExpandVST(MachineBasicBlock::iterator &MBBI) {
581 MachineInstr &MI = *MBBI;
582 MachineBasicBlock &MBB = *MI.getParent();
583 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump());
585 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
586 assert(TableEntry && !TableEntry->IsLoad && "NEONLdStTable lookup failed");
587 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing;
588 unsigned NumRegs = TableEntry->NumRegs;
590 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
591 TII->get(TableEntry->RealOpc));
593 if (TableEntry->isUpdating)
594 MIB.add(MI.getOperand(OpIdx++));
596 // Copy the addrmode6 operands.
597 MIB.add(MI.getOperand(OpIdx++));
598 MIB.add(MI.getOperand(OpIdx++));
600 if (TableEntry->hasWritebackOperand) {
601 // TODO: The writing-back pseudo instructions we translate here are all
602 // defined to take am6offset nodes that are capable to represent both fixed
603 // and register forms. Some real instructions, however, do not rely on
604 // am6offset and have separate definitions for such forms. When this is the
605 // case, fixed forms do not take any offset nodes, so here we skip them for
606 // such instructions. Once all real and pseudo writing-back instructions are
607 // rewritten without use of am6offset nodes, this code will go away.
608 const MachineOperand &AM6Offset = MI.getOperand(OpIdx++);
609 if (TableEntry->RealOpc == ARM::VST1d8Qwb_fixed ||
610 TableEntry->RealOpc == ARM::VST1d16Qwb_fixed ||
611 TableEntry->RealOpc == ARM::VST1d32Qwb_fixed ||
612 TableEntry->RealOpc == ARM::VST1d64Qwb_fixed ||
613 TableEntry->RealOpc == ARM::VST1d8Twb_fixed ||
614 TableEntry->RealOpc == ARM::VST1d16Twb_fixed ||
615 TableEntry->RealOpc == ARM::VST1d32Twb_fixed ||
616 TableEntry->RealOpc == ARM::VST1d64Twb_fixed) {
617 assert(AM6Offset.getReg() == 0 &&
618 "A fixed writing-back pseudo instruction provides an offset "
625 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
626 bool SrcIsUndef = MI.getOperand(OpIdx).isUndef();
627 Register SrcReg = MI.getOperand(OpIdx++).getReg();
628 unsigned D0, D1, D2, D3;
629 GetDSubRegs(SrcReg, RegSpc, TRI, D0, D1, D2, D3);
630 MIB.addReg(D0, getUndefRegState(SrcIsUndef));
631 if (NumRegs > 1 && TableEntry->copyAllListRegs)
632 MIB.addReg(D1, getUndefRegState(SrcIsUndef));
633 if (NumRegs > 2 && TableEntry->copyAllListRegs)
634 MIB.addReg(D2, getUndefRegState(SrcIsUndef));
635 if (NumRegs > 3 && TableEntry->copyAllListRegs)
636 MIB.addReg(D3, getUndefRegState(SrcIsUndef));
638 // Copy the predicate operands.
639 MIB.add(MI.getOperand(OpIdx++));
640 MIB.add(MI.getOperand(OpIdx++));
642 if (SrcIsKill && !SrcIsUndef) // Add an implicit kill for the super-reg.
643 MIB->addRegisterKilled(SrcReg, TRI, true);
644 else if (!SrcIsUndef)
645 MIB.addReg(SrcReg, RegState::Implicit); // Add implicit uses for src reg.
646 TransferImpOps(MI, MIB, MIB);
648 // Transfer memoperands.
649 MIB.cloneMemRefs(MI);
650 MI.eraseFromParent();
651 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump(););
654 /// ExpandLaneOp - Translate VLD*LN and VST*LN instructions with Q, QQ or QQQQ
655 /// register operands to real instructions with D register operands.
656 void ARMExpandPseudo::ExpandLaneOp(MachineBasicBlock::iterator &MBBI) {
657 MachineInstr &MI = *MBBI;
658 MachineBasicBlock &MBB = *MI.getParent();
659 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump());
661 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
662 assert(TableEntry && "NEONLdStTable lookup failed");
663 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing;
664 unsigned NumRegs = TableEntry->NumRegs;
665 unsigned RegElts = TableEntry->RegElts;
667 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
668 TII->get(TableEntry->RealOpc));
670 // The lane operand is always the 3rd from last operand, before the 2
671 // predicate operands.
672 unsigned Lane = MI.getOperand(MI.getDesc().getNumOperands() - 3).getImm();
674 // Adjust the lane and spacing as needed for Q registers.
675 assert(RegSpc != OddDblSpc && "unexpected register spacing for VLD/VST-lane");
676 if (RegSpc == EvenDblSpc && Lane >= RegElts) {
680 assert(Lane < RegElts && "out of range lane for VLD/VST-lane");
682 unsigned D0 = 0, D1 = 0, D2 = 0, D3 = 0;
684 bool DstIsDead = false;
685 if (TableEntry->IsLoad) {
686 DstIsDead = MI.getOperand(OpIdx).isDead();
687 DstReg = MI.getOperand(OpIdx++).getReg();
688 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3);
689 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead));
691 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
693 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead));
695 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead));
698 if (TableEntry->isUpdating)
699 MIB.add(MI.getOperand(OpIdx++));
701 // Copy the addrmode6 operands.
702 MIB.add(MI.getOperand(OpIdx++));
703 MIB.add(MI.getOperand(OpIdx++));
704 // Copy the am6offset operand.
705 if (TableEntry->hasWritebackOperand)
706 MIB.add(MI.getOperand(OpIdx++));
708 // Grab the super-register source.
709 MachineOperand MO = MI.getOperand(OpIdx++);
710 if (!TableEntry->IsLoad)
711 GetDSubRegs(MO.getReg(), RegSpc, TRI, D0, D1, D2, D3);
713 // Add the subregs as sources of the new instruction.
714 unsigned SrcFlags = (getUndefRegState(MO.isUndef()) |
715 getKillRegState(MO.isKill()));
716 MIB.addReg(D0, SrcFlags);
718 MIB.addReg(D1, SrcFlags);
720 MIB.addReg(D2, SrcFlags);
722 MIB.addReg(D3, SrcFlags);
724 // Add the lane number operand.
728 // Copy the predicate operands.
729 MIB.add(MI.getOperand(OpIdx++));
730 MIB.add(MI.getOperand(OpIdx++));
732 // Copy the super-register source to be an implicit source.
733 MO.setImplicit(true);
735 if (TableEntry->IsLoad)
736 // Add an implicit def for the super-register.
737 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
738 TransferImpOps(MI, MIB, MIB);
739 // Transfer memoperands.
740 MIB.cloneMemRefs(MI);
741 MI.eraseFromParent();
744 /// ExpandVTBL - Translate VTBL and VTBX pseudo instructions with Q or QQ
745 /// register operands to real instructions with D register operands.
746 void ARMExpandPseudo::ExpandVTBL(MachineBasicBlock::iterator &MBBI,
747 unsigned Opc, bool IsExt) {
748 MachineInstr &MI = *MBBI;
749 MachineBasicBlock &MBB = *MI.getParent();
750 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump());
752 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc));
755 // Transfer the destination register operand.
756 MIB.add(MI.getOperand(OpIdx++));
758 MachineOperand VdSrc(MI.getOperand(OpIdx++));
762 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
763 Register SrcReg = MI.getOperand(OpIdx++).getReg();
764 unsigned D0, D1, D2, D3;
765 GetDSubRegs(SrcReg, SingleSpc, TRI, D0, D1, D2, D3);
768 // Copy the other source register operand.
769 MachineOperand VmSrc(MI.getOperand(OpIdx++));
772 // Copy the predicate operands.
773 MIB.add(MI.getOperand(OpIdx++));
774 MIB.add(MI.getOperand(OpIdx++));
776 // Add an implicit kill and use for the super-reg.
777 MIB.addReg(SrcReg, RegState::Implicit | getKillRegState(SrcIsKill));
778 TransferImpOps(MI, MIB, MIB);
779 MI.eraseFromParent();
780 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump(););
783 static bool IsAnAddressOperand(const MachineOperand &MO) {
784 // This check is overly conservative. Unless we are certain that the machine
785 // operand is not a symbol reference, we return that it is a symbol reference.
786 // This is important as the load pair may not be split up Windows.
787 switch (MO.getType()) {
788 case MachineOperand::MO_Register:
789 case MachineOperand::MO_Immediate:
790 case MachineOperand::MO_CImmediate:
791 case MachineOperand::MO_FPImmediate:
792 case MachineOperand::MO_ShuffleMask:
794 case MachineOperand::MO_MachineBasicBlock:
796 case MachineOperand::MO_FrameIndex:
798 case MachineOperand::MO_ConstantPoolIndex:
799 case MachineOperand::MO_TargetIndex:
800 case MachineOperand::MO_JumpTableIndex:
801 case MachineOperand::MO_ExternalSymbol:
802 case MachineOperand::MO_GlobalAddress:
803 case MachineOperand::MO_BlockAddress:
805 case MachineOperand::MO_RegisterMask:
806 case MachineOperand::MO_RegisterLiveOut:
808 case MachineOperand::MO_Metadata:
809 case MachineOperand::MO_MCSymbol:
811 case MachineOperand::MO_CFIIndex:
813 case MachineOperand::MO_IntrinsicID:
814 case MachineOperand::MO_Predicate:
815 llvm_unreachable("should not exist post-isel");
817 llvm_unreachable("unhandled machine operand type");
820 static MachineOperand makeImplicit(const MachineOperand &MO) {
821 MachineOperand NewMO = MO;
826 void ARMExpandPseudo::ExpandMOV32BitImm(MachineBasicBlock &MBB,
827 MachineBasicBlock::iterator &MBBI) {
828 MachineInstr &MI = *MBBI;
829 unsigned Opcode = MI.getOpcode();
830 unsigned PredReg = 0;
831 ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg);
832 Register DstReg = MI.getOperand(0).getReg();
833 bool DstIsDead = MI.getOperand(0).isDead();
834 bool isCC = Opcode == ARM::MOVCCi32imm || Opcode == ARM::t2MOVCCi32imm;
835 const MachineOperand &MO = MI.getOperand(isCC ? 2 : 1);
836 bool RequiresBundling = STI->isTargetWindows() && IsAnAddressOperand(MO);
837 MachineInstrBuilder LO16, HI16;
838 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump());
840 if (!STI->hasV6T2Ops() &&
841 (Opcode == ARM::MOVi32imm || Opcode == ARM::MOVCCi32imm)) {
842 // FIXME Windows CE supports older ARM CPUs
843 assert(!STI->isTargetWindows() && "Windows on ARM requires ARMv7+");
845 // Expand into a movi + orr.
846 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVi), DstReg);
847 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::ORRri))
848 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
851 assert (MO.isImm() && "MOVi32imm w/ non-immediate source operand!");
852 unsigned ImmVal = (unsigned)MO.getImm();
853 unsigned SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(ImmVal);
854 unsigned SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(ImmVal);
855 LO16 = LO16.addImm(SOImmValV1);
856 HI16 = HI16.addImm(SOImmValV2);
857 LO16.cloneMemRefs(MI);
858 HI16.cloneMemRefs(MI);
859 LO16.addImm(Pred).addReg(PredReg).add(condCodeOp());
860 HI16.addImm(Pred).addReg(PredReg).add(condCodeOp());
862 LO16.add(makeImplicit(MI.getOperand(1)));
863 TransferImpOps(MI, LO16, HI16);
864 MI.eraseFromParent();
868 unsigned LO16Opc = 0;
869 unsigned HI16Opc = 0;
870 if (Opcode == ARM::t2MOVi32imm || Opcode == ARM::t2MOVCCi32imm) {
871 LO16Opc = ARM::t2MOVi16;
872 HI16Opc = ARM::t2MOVTi16;
874 LO16Opc = ARM::MOVi16;
875 HI16Opc = ARM::MOVTi16;
878 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LO16Opc), DstReg);
879 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc))
880 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
883 switch (MO.getType()) {
884 case MachineOperand::MO_Immediate: {
885 unsigned Imm = MO.getImm();
886 unsigned Lo16 = Imm & 0xffff;
887 unsigned Hi16 = (Imm >> 16) & 0xffff;
888 LO16 = LO16.addImm(Lo16);
889 HI16 = HI16.addImm(Hi16);
892 case MachineOperand::MO_ExternalSymbol: {
893 const char *ES = MO.getSymbolName();
894 unsigned TF = MO.getTargetFlags();
895 LO16 = LO16.addExternalSymbol(ES, TF | ARMII::MO_LO16);
896 HI16 = HI16.addExternalSymbol(ES, TF | ARMII::MO_HI16);
900 const GlobalValue *GV = MO.getGlobal();
901 unsigned TF = MO.getTargetFlags();
902 LO16 = LO16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_LO16);
903 HI16 = HI16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_HI16);
908 LO16.cloneMemRefs(MI);
909 HI16.cloneMemRefs(MI);
910 LO16.addImm(Pred).addReg(PredReg);
911 HI16.addImm(Pred).addReg(PredReg);
913 if (RequiresBundling)
914 finalizeBundle(MBB, LO16->getIterator(), MBBI->getIterator());
917 LO16.add(makeImplicit(MI.getOperand(1)));
918 TransferImpOps(MI, LO16, HI16);
919 MI.eraseFromParent();
920 LLVM_DEBUG(dbgs() << "To: "; LO16.getInstr()->dump(););
921 LLVM_DEBUG(dbgs() << "And: "; HI16.getInstr()->dump(););
924 /// Expand a CMP_SWAP pseudo-inst to an ldrex/strex loop as simply as
925 /// possible. This only gets used at -O0 so we don't care about efficiency of
926 /// the generated code.
927 bool ARMExpandPseudo::ExpandCMP_SWAP(MachineBasicBlock &MBB,
928 MachineBasicBlock::iterator MBBI,
929 unsigned LdrexOp, unsigned StrexOp,
931 MachineBasicBlock::iterator &NextMBBI) {
932 bool IsThumb = STI->isThumb();
933 MachineInstr &MI = *MBBI;
934 DebugLoc DL = MI.getDebugLoc();
935 const MachineOperand &Dest = MI.getOperand(0);
936 Register TempReg = MI.getOperand(1).getReg();
937 // Duplicating undef operands into 2 instructions does not guarantee the same
938 // value on both; However undef should be replaced by xzr anyway.
939 assert(!MI.getOperand(2).isUndef() && "cannot handle undef");
940 Register AddrReg = MI.getOperand(2).getReg();
941 Register DesiredReg = MI.getOperand(3).getReg();
942 Register NewReg = MI.getOperand(4).getReg();
944 MachineFunction *MF = MBB.getParent();
945 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
946 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
947 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
949 MF->insert(++MBB.getIterator(), LoadCmpBB);
950 MF->insert(++LoadCmpBB->getIterator(), StoreBB);
951 MF->insert(++StoreBB->getIterator(), DoneBB);
954 MachineInstrBuilder MIB =
955 BuildMI(MBB, MBBI, DL, TII->get(UxtOp), DesiredReg)
956 .addReg(DesiredReg, RegState::Kill);
959 MIB.add(predOps(ARMCC::AL));
963 // ldrex rDest, [rAddr]
964 // cmp rDest, rDesired
967 MachineInstrBuilder MIB;
968 MIB = BuildMI(LoadCmpBB, DL, TII->get(LdrexOp), Dest.getReg());
970 if (LdrexOp == ARM::t2LDREX)
971 MIB.addImm(0); // a 32-bit Thumb ldrex (only) allows an offset.
972 MIB.add(predOps(ARMCC::AL));
974 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr;
975 BuildMI(LoadCmpBB, DL, TII->get(CMPrr))
976 .addReg(Dest.getReg(), getKillRegState(Dest.isDead()))
978 .add(predOps(ARMCC::AL));
979 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc;
980 BuildMI(LoadCmpBB, DL, TII->get(Bcc))
983 .addReg(ARM::CPSR, RegState::Kill);
984 LoadCmpBB->addSuccessor(DoneBB);
985 LoadCmpBB->addSuccessor(StoreBB);
988 // strex rTempReg, rNew, [rAddr]
991 MIB = BuildMI(StoreBB, DL, TII->get(StrexOp), TempReg)
994 if (StrexOp == ARM::t2STREX)
995 MIB.addImm(0); // a 32-bit Thumb strex (only) allows an offset.
996 MIB.add(predOps(ARMCC::AL));
998 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri;
999 BuildMI(StoreBB, DL, TII->get(CMPri))
1000 .addReg(TempReg, RegState::Kill)
1002 .add(predOps(ARMCC::AL));
1003 BuildMI(StoreBB, DL, TII->get(Bcc))
1006 .addReg(ARM::CPSR, RegState::Kill);
1007 StoreBB->addSuccessor(LoadCmpBB);
1008 StoreBB->addSuccessor(DoneBB);
1010 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end());
1011 DoneBB->transferSuccessors(&MBB);
1013 MBB.addSuccessor(LoadCmpBB);
1015 NextMBBI = MBB.end();
1016 MI.eraseFromParent();
1018 // Recompute livein lists.
1019 LivePhysRegs LiveRegs;
1020 computeAndAddLiveIns(LiveRegs, *DoneBB);
1021 computeAndAddLiveIns(LiveRegs, *StoreBB);
1022 computeAndAddLiveIns(LiveRegs, *LoadCmpBB);
1023 // Do an extra pass around the loop to get loop carried registers right.
1024 StoreBB->clearLiveIns();
1025 computeAndAddLiveIns(LiveRegs, *StoreBB);
1026 LoadCmpBB->clearLiveIns();
1027 computeAndAddLiveIns(LiveRegs, *LoadCmpBB);
1032 /// ARM's ldrexd/strexd take a consecutive register pair (represented as a
1033 /// single GPRPair register), Thumb's take two separate registers so we need to
1034 /// extract the subregs from the pair.
1035 static void addExclusiveRegPair(MachineInstrBuilder &MIB, MachineOperand &Reg,
1036 unsigned Flags, bool IsThumb,
1037 const TargetRegisterInfo *TRI) {
1039 Register RegLo = TRI->getSubReg(Reg.getReg(), ARM::gsub_0);
1040 Register RegHi = TRI->getSubReg(Reg.getReg(), ARM::gsub_1);
1041 MIB.addReg(RegLo, Flags);
1042 MIB.addReg(RegHi, Flags);
1044 MIB.addReg(Reg.getReg(), Flags);
1047 /// Expand a 64-bit CMP_SWAP to an ldrexd/strexd loop.
1048 bool ARMExpandPseudo::ExpandCMP_SWAP_64(MachineBasicBlock &MBB,
1049 MachineBasicBlock::iterator MBBI,
1050 MachineBasicBlock::iterator &NextMBBI) {
1051 bool IsThumb = STI->isThumb();
1052 MachineInstr &MI = *MBBI;
1053 DebugLoc DL = MI.getDebugLoc();
1054 MachineOperand &Dest = MI.getOperand(0);
1055 Register TempReg = MI.getOperand(1).getReg();
1056 // Duplicating undef operands into 2 instructions does not guarantee the same
1057 // value on both; However undef should be replaced by xzr anyway.
1058 assert(!MI.getOperand(2).isUndef() && "cannot handle undef");
1059 Register AddrReg = MI.getOperand(2).getReg();
1060 Register DesiredReg = MI.getOperand(3).getReg();
1061 MachineOperand New = MI.getOperand(4);
1062 New.setIsKill(false);
1064 Register DestLo = TRI->getSubReg(Dest.getReg(), ARM::gsub_0);
1065 Register DestHi = TRI->getSubReg(Dest.getReg(), ARM::gsub_1);
1066 Register DesiredLo = TRI->getSubReg(DesiredReg, ARM::gsub_0);
1067 Register DesiredHi = TRI->getSubReg(DesiredReg, ARM::gsub_1);
1069 MachineFunction *MF = MBB.getParent();
1070 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
1071 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
1072 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock());
1074 MF->insert(++MBB.getIterator(), LoadCmpBB);
1075 MF->insert(++LoadCmpBB->getIterator(), StoreBB);
1076 MF->insert(++StoreBB->getIterator(), DoneBB);
1079 // ldrexd rDestLo, rDestHi, [rAddr]
1080 // cmp rDestLo, rDesiredLo
1081 // sbcs dead rTempReg, rDestHi, rDesiredHi
1083 unsigned LDREXD = IsThumb ? ARM::t2LDREXD : ARM::LDREXD;
1084 MachineInstrBuilder MIB;
1085 MIB = BuildMI(LoadCmpBB, DL, TII->get(LDREXD));
1086 addExclusiveRegPair(MIB, Dest, RegState::Define, IsThumb, TRI);
1087 MIB.addReg(AddrReg).add(predOps(ARMCC::AL));
1089 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr;
1090 BuildMI(LoadCmpBB, DL, TII->get(CMPrr))
1091 .addReg(DestLo, getKillRegState(Dest.isDead()))
1093 .add(predOps(ARMCC::AL));
1095 BuildMI(LoadCmpBB, DL, TII->get(CMPrr))
1096 .addReg(DestHi, getKillRegState(Dest.isDead()))
1098 .addImm(ARMCC::EQ).addReg(ARM::CPSR, RegState::Kill);
1100 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc;
1101 BuildMI(LoadCmpBB, DL, TII->get(Bcc))
1104 .addReg(ARM::CPSR, RegState::Kill);
1105 LoadCmpBB->addSuccessor(DoneBB);
1106 LoadCmpBB->addSuccessor(StoreBB);
1109 // strexd rTempReg, rNewLo, rNewHi, [rAddr]
1112 unsigned STREXD = IsThumb ? ARM::t2STREXD : ARM::STREXD;
1113 MIB = BuildMI(StoreBB, DL, TII->get(STREXD), TempReg);
1114 unsigned Flags = getKillRegState(New.isDead());
1115 addExclusiveRegPair(MIB, New, Flags, IsThumb, TRI);
1116 MIB.addReg(AddrReg).add(predOps(ARMCC::AL));
1118 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri;
1119 BuildMI(StoreBB, DL, TII->get(CMPri))
1120 .addReg(TempReg, RegState::Kill)
1122 .add(predOps(ARMCC::AL));
1123 BuildMI(StoreBB, DL, TII->get(Bcc))
1126 .addReg(ARM::CPSR, RegState::Kill);
1127 StoreBB->addSuccessor(LoadCmpBB);
1128 StoreBB->addSuccessor(DoneBB);
1130 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end());
1131 DoneBB->transferSuccessors(&MBB);
1133 MBB.addSuccessor(LoadCmpBB);
1135 NextMBBI = MBB.end();
1136 MI.eraseFromParent();
1138 // Recompute livein lists.
1139 LivePhysRegs LiveRegs;
1140 computeAndAddLiveIns(LiveRegs, *DoneBB);
1141 computeAndAddLiveIns(LiveRegs, *StoreBB);
1142 computeAndAddLiveIns(LiveRegs, *LoadCmpBB);
1143 // Do an extra pass around the loop to get loop carried registers right.
1144 StoreBB->clearLiveIns();
1145 computeAndAddLiveIns(LiveRegs, *StoreBB);
1146 LoadCmpBB->clearLiveIns();
1147 computeAndAddLiveIns(LiveRegs, *LoadCmpBB);
1153 bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
1154 MachineBasicBlock::iterator MBBI,
1155 MachineBasicBlock::iterator &NextMBBI) {
1156 MachineInstr &MI = *MBBI;
1157 unsigned Opcode = MI.getOpcode();
1162 case ARM::TCRETURNdi:
1163 case ARM::TCRETURNri: {
1164 MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
1165 assert(MBBI->isReturn() &&
1166 "Can only insert epilog into returning blocks");
1167 unsigned RetOpcode = MBBI->getOpcode();
1168 DebugLoc dl = MBBI->getDebugLoc();
1169 const ARMBaseInstrInfo &TII = *static_cast<const ARMBaseInstrInfo *>(
1170 MBB.getParent()->getSubtarget().getInstrInfo());
1172 // Tail call return: adjust the stack pointer and jump to callee.
1173 MBBI = MBB.getLastNonDebugInstr();
1174 MachineOperand &JumpTarget = MBBI->getOperand(0);
1176 // Jump to label or value in register.
1177 if (RetOpcode == ARM::TCRETURNdi) {
1180 ? (STI->isTargetMachO() ? ARM::tTAILJMPd : ARM::tTAILJMPdND)
1182 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(TCOpcode));
1183 if (JumpTarget.isGlobal())
1184 MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
1185 JumpTarget.getTargetFlags());
1187 assert(JumpTarget.isSymbol());
1188 MIB.addExternalSymbol(JumpTarget.getSymbolName(),
1189 JumpTarget.getTargetFlags());
1192 // Add the default predicate in Thumb mode.
1194 MIB.add(predOps(ARMCC::AL));
1195 } else if (RetOpcode == ARM::TCRETURNri) {
1197 STI->isThumb() ? ARM::tTAILJMPr
1198 : (STI->hasV4TOps() ? ARM::TAILJMPr : ARM::TAILJMPr4);
1199 BuildMI(MBB, MBBI, dl,
1201 .addReg(JumpTarget.getReg(), RegState::Kill);
1204 auto NewMI = std::prev(MBBI);
1205 for (unsigned i = 1, e = MBBI->getNumOperands(); i != e; ++i)
1206 NewMI->addOperand(MBBI->getOperand(i));
1209 // Update call site info and delete the pseudo instruction TCRETURN.
1210 MBB.getParent()->moveCallSiteInfo(&MI, &*NewMI);
1218 case ARM::VMOVDcc: {
1219 unsigned newOpc = Opcode != ARM::VMOVDcc ? ARM::VMOVS : ARM::VMOVD;
1220 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(newOpc),
1221 MI.getOperand(1).getReg())
1222 .add(MI.getOperand(2))
1223 .addImm(MI.getOperand(3).getImm()) // 'pred'
1224 .add(MI.getOperand(4))
1225 .add(makeImplicit(MI.getOperand(1)));
1227 MI.eraseFromParent();
1232 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVr : ARM::MOVr;
1233 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc),
1234 MI.getOperand(1).getReg())
1235 .add(MI.getOperand(2))
1236 .addImm(MI.getOperand(3).getImm()) // 'pred'
1237 .add(MI.getOperand(4))
1238 .add(condCodeOp()) // 's' bit
1239 .add(makeImplicit(MI.getOperand(1)));
1241 MI.eraseFromParent();
1244 case ARM::MOVCCsi: {
1245 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi),
1246 (MI.getOperand(1).getReg()))
1247 .add(MI.getOperand(2))
1248 .addImm(MI.getOperand(3).getImm())
1249 .addImm(MI.getOperand(4).getImm()) // 'pred'
1250 .add(MI.getOperand(5))
1251 .add(condCodeOp()) // 's' bit
1252 .add(makeImplicit(MI.getOperand(1)));
1254 MI.eraseFromParent();
1257 case ARM::MOVCCsr: {
1258 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsr),
1259 (MI.getOperand(1).getReg()))
1260 .add(MI.getOperand(2))
1261 .add(MI.getOperand(3))
1262 .addImm(MI.getOperand(4).getImm())
1263 .addImm(MI.getOperand(5).getImm()) // 'pred'
1264 .add(MI.getOperand(6))
1265 .add(condCodeOp()) // 's' bit
1266 .add(makeImplicit(MI.getOperand(1)));
1268 MI.eraseFromParent();
1271 case ARM::t2MOVCCi16:
1272 case ARM::MOVCCi16: {
1273 unsigned NewOpc = AFI->isThumbFunction() ? ARM::t2MOVi16 : ARM::MOVi16;
1274 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc),
1275 MI.getOperand(1).getReg())
1276 .addImm(MI.getOperand(2).getImm())
1277 .addImm(MI.getOperand(3).getImm()) // 'pred'
1278 .add(MI.getOperand(4))
1279 .add(makeImplicit(MI.getOperand(1)));
1280 MI.eraseFromParent();
1285 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVi : ARM::MOVi;
1286 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc),
1287 MI.getOperand(1).getReg())
1288 .addImm(MI.getOperand(2).getImm())
1289 .addImm(MI.getOperand(3).getImm()) // 'pred'
1290 .add(MI.getOperand(4))
1291 .add(condCodeOp()) // 's' bit
1292 .add(makeImplicit(MI.getOperand(1)));
1294 MI.eraseFromParent();
1299 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MVNi : ARM::MVNi;
1300 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc),
1301 MI.getOperand(1).getReg())
1302 .addImm(MI.getOperand(2).getImm())
1303 .addImm(MI.getOperand(3).getImm()) // 'pred'
1304 .add(MI.getOperand(4))
1305 .add(condCodeOp()) // 's' bit
1306 .add(makeImplicit(MI.getOperand(1)));
1308 MI.eraseFromParent();
1311 case ARM::t2MOVCClsl:
1312 case ARM::t2MOVCClsr:
1313 case ARM::t2MOVCCasr:
1314 case ARM::t2MOVCCror: {
1317 case ARM::t2MOVCClsl: NewOpc = ARM::t2LSLri; break;
1318 case ARM::t2MOVCClsr: NewOpc = ARM::t2LSRri; break;
1319 case ARM::t2MOVCCasr: NewOpc = ARM::t2ASRri; break;
1320 case ARM::t2MOVCCror: NewOpc = ARM::t2RORri; break;
1321 default: llvm_unreachable("unexpeced conditional move");
1323 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc),
1324 MI.getOperand(1).getReg())
1325 .add(MI.getOperand(2))
1326 .addImm(MI.getOperand(3).getImm())
1327 .addImm(MI.getOperand(4).getImm()) // 'pred'
1328 .add(MI.getOperand(5))
1329 .add(condCodeOp()) // 's' bit
1330 .add(makeImplicit(MI.getOperand(1)));
1331 MI.eraseFromParent();
1334 case ARM::Int_eh_sjlj_dispatchsetup: {
1335 MachineFunction &MF = *MI.getParent()->getParent();
1336 const ARMBaseInstrInfo *AII =
1337 static_cast<const ARMBaseInstrInfo*>(TII);
1338 const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
1339 // For functions using a base pointer, we rematerialize it (via the frame
1340 // pointer) here since eh.sjlj.setjmp and eh.sjlj.longjmp don't do it
1341 // for us. Otherwise, expand to nothing.
1342 if (RI.hasBasePointer(MF)) {
1343 int32_t NumBytes = AFI->getFramePtrSpillOffset();
1344 Register FramePtr = RI.getFrameRegister(MF);
1345 assert(MF.getSubtarget().getFrameLowering()->hasFP(MF) &&
1346 "base pointer without frame pointer?");
1348 if (AFI->isThumb2Function()) {
1349 emitT2RegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6,
1350 FramePtr, -NumBytes, ARMCC::AL, 0, *TII);
1351 } else if (AFI->isThumbFunction()) {
1352 emitThumbRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6,
1353 FramePtr, -NumBytes, *TII, RI);
1355 emitARMRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6,
1356 FramePtr, -NumBytes, ARMCC::AL, 0,
1359 // If there's dynamic realignment, adjust for it.
1360 if (RI.needsStackRealignment(MF)) {
1361 MachineFrameInfo &MFI = MF.getFrameInfo();
1362 unsigned MaxAlign = MFI.getMaxAlignment();
1363 assert (!AFI->isThumb1OnlyFunction());
1364 // Emit bic r6, r6, MaxAlign
1365 assert(MaxAlign <= 256 && "The BIC instruction cannot encode "
1366 "immediates larger than 256 with all lower "
1368 unsigned bicOpc = AFI->isThumbFunction() ?
1369 ARM::t2BICri : ARM::BICri;
1370 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(bicOpc), ARM::R6)
1371 .addReg(ARM::R6, RegState::Kill)
1372 .addImm(MaxAlign - 1)
1373 .add(predOps(ARMCC::AL))
1378 MI.eraseFromParent();
1382 case ARM::MOVsrl_flag:
1383 case ARM::MOVsra_flag: {
1384 // These are just fancy MOVs instructions.
1385 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi),
1386 MI.getOperand(0).getReg())
1387 .add(MI.getOperand(1))
1388 .addImm(ARM_AM::getSORegOpc(
1389 (Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr : ARM_AM::asr), 1))
1390 .add(predOps(ARMCC::AL))
1391 .addReg(ARM::CPSR, RegState::Define);
1392 MI.eraseFromParent();
1396 // This encodes as "MOVs Rd, Rm, rrx
1397 MachineInstrBuilder MIB =
1398 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi),
1399 MI.getOperand(0).getReg())
1400 .add(MI.getOperand(1))
1401 .addImm(ARM_AM::getSORegOpc(ARM_AM::rrx, 0))
1402 .add(predOps(ARMCC::AL))
1404 TransferImpOps(MI, MIB, MIB);
1405 MI.eraseFromParent();
1410 const bool Thumb = Opcode == ARM::tTPsoft;
1412 MachineInstrBuilder MIB;
1413 if (STI->genLongCalls()) {
1414 MachineFunction *MF = MBB.getParent();
1415 MachineConstantPool *MCP = MF->getConstantPool();
1416 unsigned PCLabelID = AFI->createPICLabelUId();
1417 MachineConstantPoolValue *CPV =
1418 ARMConstantPoolSymbol::Create(MF->getFunction().getContext(),
1419 "__aeabi_read_tp", PCLabelID, 0);
1420 Register Reg = MI.getOperand(0).getReg();
1421 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
1422 TII->get(Thumb ? ARM::tLDRpci : ARM::LDRi12), Reg)
1423 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, 4));
1426 MIB.add(predOps(ARMCC::AL));
1428 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
1429 TII->get(Thumb ? ARM::tBLXr : ARM::BLX));
1431 MIB.add(predOps(ARMCC::AL));
1432 MIB.addReg(Reg, RegState::Kill);
1434 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
1435 TII->get(Thumb ? ARM::tBL : ARM::BL));
1437 MIB.add(predOps(ARMCC::AL));
1438 MIB.addExternalSymbol("__aeabi_read_tp", 0);
1441 MIB.cloneMemRefs(MI);
1442 TransferImpOps(MI, MIB, MIB);
1443 MI.getMF()->moveCallSiteInfo(&MI, &*MIB);
1444 MI.eraseFromParent();
1447 case ARM::tLDRpci_pic:
1448 case ARM::t2LDRpci_pic: {
1449 unsigned NewLdOpc = (Opcode == ARM::tLDRpci_pic)
1450 ? ARM::tLDRpci : ARM::t2LDRpci;
1451 Register DstReg = MI.getOperand(0).getReg();
1452 bool DstIsDead = MI.getOperand(0).isDead();
1453 MachineInstrBuilder MIB1 =
1454 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewLdOpc), DstReg)
1455 .add(MI.getOperand(1))
1456 .add(predOps(ARMCC::AL));
1457 MIB1.cloneMemRefs(MI);
1458 MachineInstrBuilder MIB2 =
1459 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tPICADD))
1460 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
1462 .add(MI.getOperand(2));
1463 TransferImpOps(MI, MIB1, MIB2);
1464 MI.eraseFromParent();
1468 case ARM::LDRLIT_ga_abs:
1469 case ARM::LDRLIT_ga_pcrel:
1470 case ARM::LDRLIT_ga_pcrel_ldr:
1471 case ARM::tLDRLIT_ga_abs:
1472 case ARM::tLDRLIT_ga_pcrel: {
1473 Register DstReg = MI.getOperand(0).getReg();
1474 bool DstIsDead = MI.getOperand(0).isDead();
1475 const MachineOperand &MO1 = MI.getOperand(1);
1476 auto Flags = MO1.getTargetFlags();
1477 const GlobalValue *GV = MO1.getGlobal();
1479 Opcode != ARM::tLDRLIT_ga_pcrel && Opcode != ARM::tLDRLIT_ga_abs;
1481 Opcode != ARM::LDRLIT_ga_abs && Opcode != ARM::tLDRLIT_ga_abs;
1482 unsigned LDRLITOpc = IsARM ? ARM::LDRi12 : ARM::tLDRpci;
1483 unsigned PICAddOpc =
1485 ? (Opcode == ARM::LDRLIT_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD)
1488 // We need a new const-pool entry to load from.
1489 MachineConstantPool *MCP = MBB.getParent()->getConstantPool();
1490 unsigned ARMPCLabelIndex = 0;
1491 MachineConstantPoolValue *CPV;
1494 unsigned PCAdj = IsARM ? 8 : 4;
1495 auto Modifier = (Flags & ARMII::MO_GOT)
1497 : ARMCP::no_modifier;
1498 ARMPCLabelIndex = AFI->createPICLabelUId();
1499 CPV = ARMConstantPoolConstant::Create(
1500 GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj, Modifier,
1501 /*AddCurrentAddr*/ Modifier == ARMCP::GOT_PREL);
1503 CPV = ARMConstantPoolConstant::Create(GV, ARMCP::no_modifier);
1505 MachineInstrBuilder MIB =
1506 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LDRLITOpc), DstReg)
1507 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, 4));
1510 MIB.add(predOps(ARMCC::AL));
1513 MachineInstrBuilder MIB =
1514 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(PICAddOpc))
1515 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
1517 .addImm(ARMPCLabelIndex);
1520 MIB.add(predOps(ARMCC::AL));
1523 MI.eraseFromParent();
1526 case ARM::MOV_ga_pcrel:
1527 case ARM::MOV_ga_pcrel_ldr:
1528 case ARM::t2MOV_ga_pcrel: {
1529 // Expand into movw + movw. Also "add pc" / ldr [pc] in PIC mode.
1530 unsigned LabelId = AFI->createPICLabelUId();
1531 Register DstReg = MI.getOperand(0).getReg();
1532 bool DstIsDead = MI.getOperand(0).isDead();
1533 const MachineOperand &MO1 = MI.getOperand(1);
1534 const GlobalValue *GV = MO1.getGlobal();
1535 unsigned TF = MO1.getTargetFlags();
1536 bool isARM = Opcode != ARM::t2MOV_ga_pcrel;
1537 unsigned LO16Opc = isARM ? ARM::MOVi16_ga_pcrel : ARM::t2MOVi16_ga_pcrel;
1538 unsigned HI16Opc = isARM ? ARM::MOVTi16_ga_pcrel :ARM::t2MOVTi16_ga_pcrel;
1539 unsigned LO16TF = TF | ARMII::MO_LO16;
1540 unsigned HI16TF = TF | ARMII::MO_HI16;
1541 unsigned PICAddOpc = isARM
1542 ? (Opcode == ARM::MOV_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD)
1544 MachineInstrBuilder MIB1 = BuildMI(MBB, MBBI, MI.getDebugLoc(),
1545 TII->get(LO16Opc), DstReg)
1546 .addGlobalAddress(GV, MO1.getOffset(), TF | LO16TF)
1549 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc), DstReg)
1551 .addGlobalAddress(GV, MO1.getOffset(), TF | HI16TF)
1554 MachineInstrBuilder MIB3 = BuildMI(MBB, MBBI, MI.getDebugLoc(),
1555 TII->get(PICAddOpc))
1556 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
1557 .addReg(DstReg).addImm(LabelId);
1559 MIB3.add(predOps(ARMCC::AL));
1560 if (Opcode == ARM::MOV_ga_pcrel_ldr)
1561 MIB3.cloneMemRefs(MI);
1563 TransferImpOps(MI, MIB1, MIB3);
1564 MI.eraseFromParent();
1568 case ARM::MOVi32imm:
1569 case ARM::MOVCCi32imm:
1570 case ARM::t2MOVi32imm:
1571 case ARM::t2MOVCCi32imm:
1572 ExpandMOV32BitImm(MBB, MBBI);
1575 case ARM::SUBS_PC_LR: {
1576 MachineInstrBuilder MIB =
1577 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::SUBri), ARM::PC)
1579 .add(MI.getOperand(0))
1580 .add(MI.getOperand(1))
1581 .add(MI.getOperand(2))
1582 .addReg(ARM::CPSR, RegState::Undef);
1583 TransferImpOps(MI, MIB, MIB);
1584 MI.eraseFromParent();
1587 case ARM::VLDMQIA: {
1588 unsigned NewOpc = ARM::VLDMDIA;
1589 MachineInstrBuilder MIB =
1590 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc));
1593 // Grab the Q register destination.
1594 bool DstIsDead = MI.getOperand(OpIdx).isDead();
1595 Register DstReg = MI.getOperand(OpIdx++).getReg();
1597 // Copy the source register.
1598 MIB.add(MI.getOperand(OpIdx++));
1600 // Copy the predicate operands.
1601 MIB.add(MI.getOperand(OpIdx++));
1602 MIB.add(MI.getOperand(OpIdx++));
1604 // Add the destination operands (D subregs).
1605 Register D0 = TRI->getSubReg(DstReg, ARM::dsub_0);
1606 Register D1 = TRI->getSubReg(DstReg, ARM::dsub_1);
1607 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead))
1608 .addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
1610 // Add an implicit def for the super-register.
1611 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
1612 TransferImpOps(MI, MIB, MIB);
1613 MIB.cloneMemRefs(MI);
1614 MI.eraseFromParent();
1618 case ARM::VSTMQIA: {
1619 unsigned NewOpc = ARM::VSTMDIA;
1620 MachineInstrBuilder MIB =
1621 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc));
1624 // Grab the Q register source.
1625 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
1626 Register SrcReg = MI.getOperand(OpIdx++).getReg();
1628 // Copy the destination register.
1629 MachineOperand Dst(MI.getOperand(OpIdx++));
1632 // Copy the predicate operands.
1633 MIB.add(MI.getOperand(OpIdx++));
1634 MIB.add(MI.getOperand(OpIdx++));
1636 // Add the source operands (D subregs).
1637 Register D0 = TRI->getSubReg(SrcReg, ARM::dsub_0);
1638 Register D1 = TRI->getSubReg(SrcReg, ARM::dsub_1);
1639 MIB.addReg(D0, SrcIsKill ? RegState::Kill : 0)
1640 .addReg(D1, SrcIsKill ? RegState::Kill : 0);
1642 if (SrcIsKill) // Add an implicit kill for the Q register.
1643 MIB->addRegisterKilled(SrcReg, TRI, true);
1645 TransferImpOps(MI, MIB, MIB);
1646 MIB.cloneMemRefs(MI);
1647 MI.eraseFromParent();
1651 case ARM::VLD2q8Pseudo:
1652 case ARM::VLD2q16Pseudo:
1653 case ARM::VLD2q32Pseudo:
1654 case ARM::VLD2q8PseudoWB_fixed:
1655 case ARM::VLD2q16PseudoWB_fixed:
1656 case ARM::VLD2q32PseudoWB_fixed:
1657 case ARM::VLD2q8PseudoWB_register:
1658 case ARM::VLD2q16PseudoWB_register:
1659 case ARM::VLD2q32PseudoWB_register:
1660 case ARM::VLD3d8Pseudo:
1661 case ARM::VLD3d16Pseudo:
1662 case ARM::VLD3d32Pseudo:
1663 case ARM::VLD1d8TPseudo:
1664 case ARM::VLD1d16TPseudo:
1665 case ARM::VLD1d32TPseudo:
1666 case ARM::VLD1d64TPseudo:
1667 case ARM::VLD1d64TPseudoWB_fixed:
1668 case ARM::VLD1d64TPseudoWB_register:
1669 case ARM::VLD3d8Pseudo_UPD:
1670 case ARM::VLD3d16Pseudo_UPD:
1671 case ARM::VLD3d32Pseudo_UPD:
1672 case ARM::VLD3q8Pseudo_UPD:
1673 case ARM::VLD3q16Pseudo_UPD:
1674 case ARM::VLD3q32Pseudo_UPD:
1675 case ARM::VLD3q8oddPseudo:
1676 case ARM::VLD3q16oddPseudo:
1677 case ARM::VLD3q32oddPseudo:
1678 case ARM::VLD3q8oddPseudo_UPD:
1679 case ARM::VLD3q16oddPseudo_UPD:
1680 case ARM::VLD3q32oddPseudo_UPD:
1681 case ARM::VLD4d8Pseudo:
1682 case ARM::VLD4d16Pseudo:
1683 case ARM::VLD4d32Pseudo:
1684 case ARM::VLD1d8QPseudo:
1685 case ARM::VLD1d16QPseudo:
1686 case ARM::VLD1d32QPseudo:
1687 case ARM::VLD1d64QPseudo:
1688 case ARM::VLD1d64QPseudoWB_fixed:
1689 case ARM::VLD1d64QPseudoWB_register:
1690 case ARM::VLD1q8HighQPseudo:
1691 case ARM::VLD1q8LowQPseudo_UPD:
1692 case ARM::VLD1q8HighTPseudo:
1693 case ARM::VLD1q8LowTPseudo_UPD:
1694 case ARM::VLD1q16HighQPseudo:
1695 case ARM::VLD1q16LowQPseudo_UPD:
1696 case ARM::VLD1q16HighTPseudo:
1697 case ARM::VLD1q16LowTPseudo_UPD:
1698 case ARM::VLD1q32HighQPseudo:
1699 case ARM::VLD1q32LowQPseudo_UPD:
1700 case ARM::VLD1q32HighTPseudo:
1701 case ARM::VLD1q32LowTPseudo_UPD:
1702 case ARM::VLD1q64HighQPseudo:
1703 case ARM::VLD1q64LowQPseudo_UPD:
1704 case ARM::VLD1q64HighTPseudo:
1705 case ARM::VLD1q64LowTPseudo_UPD:
1706 case ARM::VLD4d8Pseudo_UPD:
1707 case ARM::VLD4d16Pseudo_UPD:
1708 case ARM::VLD4d32Pseudo_UPD:
1709 case ARM::VLD4q8Pseudo_UPD:
1710 case ARM::VLD4q16Pseudo_UPD:
1711 case ARM::VLD4q32Pseudo_UPD:
1712 case ARM::VLD4q8oddPseudo:
1713 case ARM::VLD4q16oddPseudo:
1714 case ARM::VLD4q32oddPseudo:
1715 case ARM::VLD4q8oddPseudo_UPD:
1716 case ARM::VLD4q16oddPseudo_UPD:
1717 case ARM::VLD4q32oddPseudo_UPD:
1718 case ARM::VLD3DUPd8Pseudo:
1719 case ARM::VLD3DUPd16Pseudo:
1720 case ARM::VLD3DUPd32Pseudo:
1721 case ARM::VLD3DUPd8Pseudo_UPD:
1722 case ARM::VLD3DUPd16Pseudo_UPD:
1723 case ARM::VLD3DUPd32Pseudo_UPD:
1724 case ARM::VLD4DUPd8Pseudo:
1725 case ARM::VLD4DUPd16Pseudo:
1726 case ARM::VLD4DUPd32Pseudo:
1727 case ARM::VLD4DUPd8Pseudo_UPD:
1728 case ARM::VLD4DUPd16Pseudo_UPD:
1729 case ARM::VLD4DUPd32Pseudo_UPD:
1730 case ARM::VLD2DUPq8EvenPseudo:
1731 case ARM::VLD2DUPq8OddPseudo:
1732 case ARM::VLD2DUPq16EvenPseudo:
1733 case ARM::VLD2DUPq16OddPseudo:
1734 case ARM::VLD2DUPq32EvenPseudo:
1735 case ARM::VLD2DUPq32OddPseudo:
1736 case ARM::VLD3DUPq8EvenPseudo:
1737 case ARM::VLD3DUPq8OddPseudo:
1738 case ARM::VLD3DUPq16EvenPseudo:
1739 case ARM::VLD3DUPq16OddPseudo:
1740 case ARM::VLD3DUPq32EvenPseudo:
1741 case ARM::VLD3DUPq32OddPseudo:
1742 case ARM::VLD4DUPq8EvenPseudo:
1743 case ARM::VLD4DUPq8OddPseudo:
1744 case ARM::VLD4DUPq16EvenPseudo:
1745 case ARM::VLD4DUPq16OddPseudo:
1746 case ARM::VLD4DUPq32EvenPseudo:
1747 case ARM::VLD4DUPq32OddPseudo:
1751 case ARM::VST2q8Pseudo:
1752 case ARM::VST2q16Pseudo:
1753 case ARM::VST2q32Pseudo:
1754 case ARM::VST2q8PseudoWB_fixed:
1755 case ARM::VST2q16PseudoWB_fixed:
1756 case ARM::VST2q32PseudoWB_fixed:
1757 case ARM::VST2q8PseudoWB_register:
1758 case ARM::VST2q16PseudoWB_register:
1759 case ARM::VST2q32PseudoWB_register:
1760 case ARM::VST3d8Pseudo:
1761 case ARM::VST3d16Pseudo:
1762 case ARM::VST3d32Pseudo:
1763 case ARM::VST1d8TPseudo:
1764 case ARM::VST1d16TPseudo:
1765 case ARM::VST1d32TPseudo:
1766 case ARM::VST1d64TPseudo:
1767 case ARM::VST3d8Pseudo_UPD:
1768 case ARM::VST3d16Pseudo_UPD:
1769 case ARM::VST3d32Pseudo_UPD:
1770 case ARM::VST1d64TPseudoWB_fixed:
1771 case ARM::VST1d64TPseudoWB_register:
1772 case ARM::VST3q8Pseudo_UPD:
1773 case ARM::VST3q16Pseudo_UPD:
1774 case ARM::VST3q32Pseudo_UPD:
1775 case ARM::VST3q8oddPseudo:
1776 case ARM::VST3q16oddPseudo:
1777 case ARM::VST3q32oddPseudo:
1778 case ARM::VST3q8oddPseudo_UPD:
1779 case ARM::VST3q16oddPseudo_UPD:
1780 case ARM::VST3q32oddPseudo_UPD:
1781 case ARM::VST4d8Pseudo:
1782 case ARM::VST4d16Pseudo:
1783 case ARM::VST4d32Pseudo:
1784 case ARM::VST1d8QPseudo:
1785 case ARM::VST1d16QPseudo:
1786 case ARM::VST1d32QPseudo:
1787 case ARM::VST1d64QPseudo:
1788 case ARM::VST4d8Pseudo_UPD:
1789 case ARM::VST4d16Pseudo_UPD:
1790 case ARM::VST4d32Pseudo_UPD:
1791 case ARM::VST1d64QPseudoWB_fixed:
1792 case ARM::VST1d64QPseudoWB_register:
1793 case ARM::VST1q8HighQPseudo:
1794 case ARM::VST1q8LowQPseudo_UPD:
1795 case ARM::VST1q8HighTPseudo:
1796 case ARM::VST1q8LowTPseudo_UPD:
1797 case ARM::VST1q16HighQPseudo:
1798 case ARM::VST1q16LowQPseudo_UPD:
1799 case ARM::VST1q16HighTPseudo:
1800 case ARM::VST1q16LowTPseudo_UPD:
1801 case ARM::VST1q32HighQPseudo:
1802 case ARM::VST1q32LowQPseudo_UPD:
1803 case ARM::VST1q32HighTPseudo:
1804 case ARM::VST1q32LowTPseudo_UPD:
1805 case ARM::VST1q64HighQPseudo:
1806 case ARM::VST1q64LowQPseudo_UPD:
1807 case ARM::VST1q64HighTPseudo:
1808 case ARM::VST1q64LowTPseudo_UPD:
1809 case ARM::VST4q8Pseudo_UPD:
1810 case ARM::VST4q16Pseudo_UPD:
1811 case ARM::VST4q32Pseudo_UPD:
1812 case ARM::VST4q8oddPseudo:
1813 case ARM::VST4q16oddPseudo:
1814 case ARM::VST4q32oddPseudo:
1815 case ARM::VST4q8oddPseudo_UPD:
1816 case ARM::VST4q16oddPseudo_UPD:
1817 case ARM::VST4q32oddPseudo_UPD:
1821 case ARM::VLD1LNq8Pseudo:
1822 case ARM::VLD1LNq16Pseudo:
1823 case ARM::VLD1LNq32Pseudo:
1824 case ARM::VLD1LNq8Pseudo_UPD:
1825 case ARM::VLD1LNq16Pseudo_UPD:
1826 case ARM::VLD1LNq32Pseudo_UPD:
1827 case ARM::VLD2LNd8Pseudo:
1828 case ARM::VLD2LNd16Pseudo:
1829 case ARM::VLD2LNd32Pseudo:
1830 case ARM::VLD2LNq16Pseudo:
1831 case ARM::VLD2LNq32Pseudo:
1832 case ARM::VLD2LNd8Pseudo_UPD:
1833 case ARM::VLD2LNd16Pseudo_UPD:
1834 case ARM::VLD2LNd32Pseudo_UPD:
1835 case ARM::VLD2LNq16Pseudo_UPD:
1836 case ARM::VLD2LNq32Pseudo_UPD:
1837 case ARM::VLD3LNd8Pseudo:
1838 case ARM::VLD3LNd16Pseudo:
1839 case ARM::VLD3LNd32Pseudo:
1840 case ARM::VLD3LNq16Pseudo:
1841 case ARM::VLD3LNq32Pseudo:
1842 case ARM::VLD3LNd8Pseudo_UPD:
1843 case ARM::VLD3LNd16Pseudo_UPD:
1844 case ARM::VLD3LNd32Pseudo_UPD:
1845 case ARM::VLD3LNq16Pseudo_UPD:
1846 case ARM::VLD3LNq32Pseudo_UPD:
1847 case ARM::VLD4LNd8Pseudo:
1848 case ARM::VLD4LNd16Pseudo:
1849 case ARM::VLD4LNd32Pseudo:
1850 case ARM::VLD4LNq16Pseudo:
1851 case ARM::VLD4LNq32Pseudo:
1852 case ARM::VLD4LNd8Pseudo_UPD:
1853 case ARM::VLD4LNd16Pseudo_UPD:
1854 case ARM::VLD4LNd32Pseudo_UPD:
1855 case ARM::VLD4LNq16Pseudo_UPD:
1856 case ARM::VLD4LNq32Pseudo_UPD:
1857 case ARM::VST1LNq8Pseudo:
1858 case ARM::VST1LNq16Pseudo:
1859 case ARM::VST1LNq32Pseudo:
1860 case ARM::VST1LNq8Pseudo_UPD:
1861 case ARM::VST1LNq16Pseudo_UPD:
1862 case ARM::VST1LNq32Pseudo_UPD:
1863 case ARM::VST2LNd8Pseudo:
1864 case ARM::VST2LNd16Pseudo:
1865 case ARM::VST2LNd32Pseudo:
1866 case ARM::VST2LNq16Pseudo:
1867 case ARM::VST2LNq32Pseudo:
1868 case ARM::VST2LNd8Pseudo_UPD:
1869 case ARM::VST2LNd16Pseudo_UPD:
1870 case ARM::VST2LNd32Pseudo_UPD:
1871 case ARM::VST2LNq16Pseudo_UPD:
1872 case ARM::VST2LNq32Pseudo_UPD:
1873 case ARM::VST3LNd8Pseudo:
1874 case ARM::VST3LNd16Pseudo:
1875 case ARM::VST3LNd32Pseudo:
1876 case ARM::VST3LNq16Pseudo:
1877 case ARM::VST3LNq32Pseudo:
1878 case ARM::VST3LNd8Pseudo_UPD:
1879 case ARM::VST3LNd16Pseudo_UPD:
1880 case ARM::VST3LNd32Pseudo_UPD:
1881 case ARM::VST3LNq16Pseudo_UPD:
1882 case ARM::VST3LNq32Pseudo_UPD:
1883 case ARM::VST4LNd8Pseudo:
1884 case ARM::VST4LNd16Pseudo:
1885 case ARM::VST4LNd32Pseudo:
1886 case ARM::VST4LNq16Pseudo:
1887 case ARM::VST4LNq32Pseudo:
1888 case ARM::VST4LNd8Pseudo_UPD:
1889 case ARM::VST4LNd16Pseudo_UPD:
1890 case ARM::VST4LNd32Pseudo_UPD:
1891 case ARM::VST4LNq16Pseudo_UPD:
1892 case ARM::VST4LNq32Pseudo_UPD:
1896 case ARM::VTBL3Pseudo: ExpandVTBL(MBBI, ARM::VTBL3, false); return true;
1897 case ARM::VTBL4Pseudo: ExpandVTBL(MBBI, ARM::VTBL4, false); return true;
1898 case ARM::VTBX3Pseudo: ExpandVTBL(MBBI, ARM::VTBX3, true); return true;
1899 case ARM::VTBX4Pseudo: ExpandVTBL(MBBI, ARM::VTBX4, true); return true;
1901 case ARM::CMP_SWAP_8:
1903 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXB, ARM::t2STREXB,
1904 ARM::tUXTB, NextMBBI);
1906 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXB, ARM::STREXB,
1907 ARM::UXTB, NextMBBI);
1908 case ARM::CMP_SWAP_16:
1910 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXH, ARM::t2STREXH,
1911 ARM::tUXTH, NextMBBI);
1913 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXH, ARM::STREXH,
1914 ARM::UXTH, NextMBBI);
1915 case ARM::CMP_SWAP_32:
1917 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREX, ARM::t2STREX, 0,
1920 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREX, ARM::STREX, 0, NextMBBI);
1922 case ARM::CMP_SWAP_64:
1923 return ExpandCMP_SWAP_64(MBB, MBBI, NextMBBI);
1925 case ARM::tBL_PUSHLR:
1926 case ARM::BL_PUSHLR: {
1927 const bool Thumb = Opcode == ARM::tBL_PUSHLR;
1928 Register Reg = MI.getOperand(0).getReg();
1929 assert(Reg == ARM::LR && "expect LR register!");
1930 MachineInstrBuilder MIB;
1933 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tPUSH))
1934 .add(predOps(ARMCC::AL))
1937 // bl __gnu_mcount_nc
1938 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tBL));
1941 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::STMDB_UPD))
1942 .addReg(ARM::SP, RegState::Define)
1944 .add(predOps(ARMCC::AL))
1947 // bl __gnu_mcount_nc
1948 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::BL));
1950 MIB.cloneMemRefs(MI);
1951 for (unsigned i = 1; i < MI.getNumOperands(); ++i) MIB.add(MI.getOperand(i));
1952 MI.eraseFromParent();
1958 bool ARMExpandPseudo::ExpandMBB(MachineBasicBlock &MBB) {
1959 bool Modified = false;
1961 MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
1963 MachineBasicBlock::iterator NMBBI = std::next(MBBI);
1964 Modified |= ExpandMI(MBB, MBBI, NMBBI);
1971 bool ARMExpandPseudo::runOnMachineFunction(MachineFunction &MF) {
1972 STI = &static_cast<const ARMSubtarget &>(MF.getSubtarget());
1973 TII = STI->getInstrInfo();
1974 TRI = STI->getRegisterInfo();
1975 AFI = MF.getInfo<ARMFunctionInfo>();
1977 LLVM_DEBUG(dbgs() << "********** ARM EXPAND PSEUDO INSTRUCTIONS **********\n"
1978 << "********** Function: " << MF.getName() << '\n');
1980 bool Modified = false;
1981 for (MachineBasicBlock &MBB : MF)
1982 Modified |= ExpandMBB(MBB);
1983 if (VerifyARMPseudo)
1984 MF.verify(this, "After expanding ARM pseudo instructions.");
1986 LLVM_DEBUG(dbgs() << "***************************************************\n");
1990 /// createARMExpandPseudoPass - returns an instance of the pseudo instruction
1992 FunctionPass *llvm::createARMExpandPseudoPass() {
1993 return new ARMExpandPseudo();