1 //===-- ARMExpandPseudoInsts.cpp - Expand pseudo instructions -----*- C++ -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains a pass that expands pseudo instructions into target
11 // instructions to allow proper scheduling, if-conversion, and other late
12 // optimizations. This pass should be run after register allocation but before
13 // the post-regalloc scheduling pass.
15 //===----------------------------------------------------------------------===//
17 #define DEBUG_TYPE "arm-pseudo"
19 #include "ARMAddressingModes.h"
20 #include "ARMBaseInstrInfo.h"
21 #include "ARMBaseRegisterInfo.h"
22 #include "ARMMachineFunctionInfo.h"
23 #include "ARMRegisterInfo.h"
24 #include "llvm/CodeGen/MachineFrameInfo.h"
25 #include "llvm/CodeGen/MachineFunctionPass.h"
26 #include "llvm/CodeGen/MachineInstrBuilder.h"
27 #include "llvm/Target/TargetFrameInfo.h"
28 #include "llvm/Target/TargetRegisterInfo.h"
29 #include "llvm/Support/raw_ostream.h" // FIXME: for debug only. remove!
33 class ARMExpandPseudo : public MachineFunctionPass {
36 ARMExpandPseudo() : MachineFunctionPass(ID) {}
38 const ARMBaseInstrInfo *TII;
39 const TargetRegisterInfo *TRI;
40 const ARMSubtarget *STI;
42 virtual bool runOnMachineFunction(MachineFunction &Fn);
44 virtual const char *getPassName() const {
45 return "ARM pseudo instruction expansion pass";
49 void TransferImpOps(MachineInstr &OldMI,
50 MachineInstrBuilder &UseMI, MachineInstrBuilder &DefMI);
51 bool ExpandMBB(MachineBasicBlock &MBB);
52 void ExpandVLD(MachineBasicBlock::iterator &MBBI);
53 void ExpandVST(MachineBasicBlock::iterator &MBBI);
54 void ExpandLaneOp(MachineBasicBlock::iterator &MBBI);
55 void ExpandVTBL(MachineBasicBlock::iterator &MBBI,
56 unsigned Opc, bool IsExt, unsigned NumRegs);
58 char ARMExpandPseudo::ID = 0;
61 /// TransferImpOps - Transfer implicit operands on the pseudo instruction to
62 /// the instructions created from the expansion.
63 void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI,
64 MachineInstrBuilder &UseMI,
65 MachineInstrBuilder &DefMI) {
66 const TargetInstrDesc &Desc = OldMI.getDesc();
67 for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands();
69 const MachineOperand &MO = OldMI.getOperand(i);
70 assert(MO.isReg() && MO.getReg());
79 // Constants for register spacing in NEON load/store instructions.
80 // For quad-register load-lane and store-lane pseudo instructors, the
81 // spacing is initially assumed to be EvenDblSpc, and that is changed to
82 // OddDblSpc depending on the lane number operand.
89 // Entries for NEON load/store information table. The table is sorted by
90 // PseudoOpc for fast binary-search lookups.
91 struct NEONLdStTableEntry {
96 NEONRegSpacing RegSpacing;
97 unsigned char NumRegs; // D registers loaded or stored
98 unsigned char RegElts; // elements per D register; used for lane ops
100 // Comparison methods for binary search of the table.
101 bool operator<(const NEONLdStTableEntry &TE) const {
102 return PseudoOpc < TE.PseudoOpc;
104 friend bool operator<(const NEONLdStTableEntry &TE, unsigned PseudoOpc) {
105 return TE.PseudoOpc < PseudoOpc;
107 friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned PseudoOpc,
108 const NEONLdStTableEntry &TE) {
109 return PseudoOpc < TE.PseudoOpc;
114 static const NEONLdStTableEntry NEONLdStTable[] = {
115 { ARM::VLD1DUPq16Pseudo, ARM::VLD1DUPq16, true, false, SingleSpc, 2, 4},
116 { ARM::VLD1DUPq16Pseudo_UPD, ARM::VLD1DUPq16_UPD, true, true, SingleSpc, 2, 4},
117 { ARM::VLD1DUPq32Pseudo, ARM::VLD1DUPq32, true, false, SingleSpc, 2, 2},
118 { ARM::VLD1DUPq32Pseudo_UPD, ARM::VLD1DUPq32_UPD, true, true, SingleSpc, 2, 2},
119 { ARM::VLD1DUPq8Pseudo, ARM::VLD1DUPq8, true, false, SingleSpc, 2, 8},
120 { ARM::VLD1DUPq8Pseudo_UPD, ARM::VLD1DUPq8_UPD, true, true, SingleSpc, 2, 8},
122 { ARM::VLD1LNq16Pseudo, ARM::VLD1LNd16, true, false, EvenDblSpc, 1, 4 },
123 { ARM::VLD1LNq16Pseudo_UPD, ARM::VLD1LNd16_UPD, true, true, EvenDblSpc, 1, 4 },
124 { ARM::VLD1LNq32Pseudo, ARM::VLD1LNd32, true, false, EvenDblSpc, 1, 2 },
125 { ARM::VLD1LNq32Pseudo_UPD, ARM::VLD1LNd32_UPD, true, true, EvenDblSpc, 1, 2 },
126 { ARM::VLD1LNq8Pseudo, ARM::VLD1LNd8, true, false, EvenDblSpc, 1, 8 },
127 { ARM::VLD1LNq8Pseudo_UPD, ARM::VLD1LNd8_UPD, true, true, EvenDblSpc, 1, 8 },
129 { ARM::VLD1d64QPseudo, ARM::VLD1d64Q, true, false, SingleSpc, 4, 1 },
130 { ARM::VLD1d64QPseudo_UPD, ARM::VLD1d64Q_UPD, true, true, SingleSpc, 4, 1 },
131 { ARM::VLD1d64TPseudo, ARM::VLD1d64T, true, false, SingleSpc, 3, 1 },
132 { ARM::VLD1d64TPseudo_UPD, ARM::VLD1d64T_UPD, true, true, SingleSpc, 3, 1 },
134 { ARM::VLD1q16Pseudo, ARM::VLD1q16, true, false, SingleSpc, 2, 4 },
135 { ARM::VLD1q16Pseudo_UPD, ARM::VLD1q16_UPD, true, true, SingleSpc, 2, 4 },
136 { ARM::VLD1q32Pseudo, ARM::VLD1q32, true, false, SingleSpc, 2, 2 },
137 { ARM::VLD1q32Pseudo_UPD, ARM::VLD1q32_UPD, true, true, SingleSpc, 2, 2 },
138 { ARM::VLD1q64Pseudo, ARM::VLD1q64, true, false, SingleSpc, 2, 1 },
139 { ARM::VLD1q64Pseudo_UPD, ARM::VLD1q64_UPD, true, true, SingleSpc, 2, 1 },
140 { ARM::VLD1q8Pseudo, ARM::VLD1q8, true, false, SingleSpc, 2, 8 },
141 { ARM::VLD1q8Pseudo_UPD, ARM::VLD1q8_UPD, true, true, SingleSpc, 2, 8 },
143 { ARM::VLD2DUPd16Pseudo, ARM::VLD2DUPd16, true, false, SingleSpc, 2, 4},
144 { ARM::VLD2DUPd16Pseudo_UPD, ARM::VLD2DUPd16_UPD, true, true, SingleSpc, 2, 4},
145 { ARM::VLD2DUPd32Pseudo, ARM::VLD2DUPd32, true, false, SingleSpc, 2, 2},
146 { ARM::VLD2DUPd32Pseudo_UPD, ARM::VLD2DUPd32_UPD, true, true, SingleSpc, 2, 2},
147 { ARM::VLD2DUPd8Pseudo, ARM::VLD2DUPd8, true, false, SingleSpc, 2, 8},
148 { ARM::VLD2DUPd8Pseudo_UPD, ARM::VLD2DUPd8_UPD, true, true, SingleSpc, 2, 8},
150 { ARM::VLD2LNd16Pseudo, ARM::VLD2LNd16, true, false, SingleSpc, 2, 4 },
151 { ARM::VLD2LNd16Pseudo_UPD, ARM::VLD2LNd16_UPD, true, true, SingleSpc, 2, 4 },
152 { ARM::VLD2LNd32Pseudo, ARM::VLD2LNd32, true, false, SingleSpc, 2, 2 },
153 { ARM::VLD2LNd32Pseudo_UPD, ARM::VLD2LNd32_UPD, true, true, SingleSpc, 2, 2 },
154 { ARM::VLD2LNd8Pseudo, ARM::VLD2LNd8, true, false, SingleSpc, 2, 8 },
155 { ARM::VLD2LNd8Pseudo_UPD, ARM::VLD2LNd8_UPD, true, true, SingleSpc, 2, 8 },
156 { ARM::VLD2LNq16Pseudo, ARM::VLD2LNq16, true, false, EvenDblSpc, 2, 4 },
157 { ARM::VLD2LNq16Pseudo_UPD, ARM::VLD2LNq16_UPD, true, true, EvenDblSpc, 2, 4 },
158 { ARM::VLD2LNq32Pseudo, ARM::VLD2LNq32, true, false, EvenDblSpc, 2, 2 },
159 { ARM::VLD2LNq32Pseudo_UPD, ARM::VLD2LNq32_UPD, true, true, EvenDblSpc, 2, 2 },
161 { ARM::VLD2d16Pseudo, ARM::VLD2d16, true, false, SingleSpc, 2, 4 },
162 { ARM::VLD2d16Pseudo_UPD, ARM::VLD2d16_UPD, true, true, SingleSpc, 2, 4 },
163 { ARM::VLD2d32Pseudo, ARM::VLD2d32, true, false, SingleSpc, 2, 2 },
164 { ARM::VLD2d32Pseudo_UPD, ARM::VLD2d32_UPD, true, true, SingleSpc, 2, 2 },
165 { ARM::VLD2d8Pseudo, ARM::VLD2d8, true, false, SingleSpc, 2, 8 },
166 { ARM::VLD2d8Pseudo_UPD, ARM::VLD2d8_UPD, true, true, SingleSpc, 2, 8 },
168 { ARM::VLD2q16Pseudo, ARM::VLD2q16, true, false, SingleSpc, 4, 4 },
169 { ARM::VLD2q16Pseudo_UPD, ARM::VLD2q16_UPD, true, true, SingleSpc, 4, 4 },
170 { ARM::VLD2q32Pseudo, ARM::VLD2q32, true, false, SingleSpc, 4, 2 },
171 { ARM::VLD2q32Pseudo_UPD, ARM::VLD2q32_UPD, true, true, SingleSpc, 4, 2 },
172 { ARM::VLD2q8Pseudo, ARM::VLD2q8, true, false, SingleSpc, 4, 8 },
173 { ARM::VLD2q8Pseudo_UPD, ARM::VLD2q8_UPD, true, true, SingleSpc, 4, 8 },
175 { ARM::VLD3DUPd16Pseudo, ARM::VLD3DUPd16, true, false, SingleSpc, 3, 4},
176 { ARM::VLD3DUPd16Pseudo_UPD, ARM::VLD3DUPd16_UPD, true, true, SingleSpc, 3, 4},
177 { ARM::VLD3DUPd32Pseudo, ARM::VLD3DUPd32, true, false, SingleSpc, 3, 2},
178 { ARM::VLD3DUPd32Pseudo_UPD, ARM::VLD3DUPd32_UPD, true, true, SingleSpc, 3, 2},
179 { ARM::VLD3DUPd8Pseudo, ARM::VLD3DUPd8, true, false, SingleSpc, 3, 8},
180 { ARM::VLD3DUPd8Pseudo_UPD, ARM::VLD3DUPd8_UPD, true, true, SingleSpc, 3, 8},
182 { ARM::VLD3LNd16Pseudo, ARM::VLD3LNd16, true, false, SingleSpc, 3, 4 },
183 { ARM::VLD3LNd16Pseudo_UPD, ARM::VLD3LNd16_UPD, true, true, SingleSpc, 3, 4 },
184 { ARM::VLD3LNd32Pseudo, ARM::VLD3LNd32, true, false, SingleSpc, 3, 2 },
185 { ARM::VLD3LNd32Pseudo_UPD, ARM::VLD3LNd32_UPD, true, true, SingleSpc, 3, 2 },
186 { ARM::VLD3LNd8Pseudo, ARM::VLD3LNd8, true, false, SingleSpc, 3, 8 },
187 { ARM::VLD3LNd8Pseudo_UPD, ARM::VLD3LNd8_UPD, true, true, SingleSpc, 3, 8 },
188 { ARM::VLD3LNq16Pseudo, ARM::VLD3LNq16, true, false, EvenDblSpc, 3, 4 },
189 { ARM::VLD3LNq16Pseudo_UPD, ARM::VLD3LNq16_UPD, true, true, EvenDblSpc, 3, 4 },
190 { ARM::VLD3LNq32Pseudo, ARM::VLD3LNq32, true, false, EvenDblSpc, 3, 2 },
191 { ARM::VLD3LNq32Pseudo_UPD, ARM::VLD3LNq32_UPD, true, true, EvenDblSpc, 3, 2 },
193 { ARM::VLD3d16Pseudo, ARM::VLD3d16, true, false, SingleSpc, 3, 4 },
194 { ARM::VLD3d16Pseudo_UPD, ARM::VLD3d16_UPD, true, true, SingleSpc, 3, 4 },
195 { ARM::VLD3d32Pseudo, ARM::VLD3d32, true, false, SingleSpc, 3, 2 },
196 { ARM::VLD3d32Pseudo_UPD, ARM::VLD3d32_UPD, true, true, SingleSpc, 3, 2 },
197 { ARM::VLD3d8Pseudo, ARM::VLD3d8, true, false, SingleSpc, 3, 8 },
198 { ARM::VLD3d8Pseudo_UPD, ARM::VLD3d8_UPD, true, true, SingleSpc, 3, 8 },
200 { ARM::VLD3q16Pseudo_UPD, ARM::VLD3q16_UPD, true, true, EvenDblSpc, 3, 4 },
201 { ARM::VLD3q16oddPseudo_UPD, ARM::VLD3q16_UPD, true, true, OddDblSpc, 3, 4 },
202 { ARM::VLD3q32Pseudo_UPD, ARM::VLD3q32_UPD, true, true, EvenDblSpc, 3, 2 },
203 { ARM::VLD3q32oddPseudo_UPD, ARM::VLD3q32_UPD, true, true, OddDblSpc, 3, 2 },
204 { ARM::VLD3q8Pseudo_UPD, ARM::VLD3q8_UPD, true, true, EvenDblSpc, 3, 8 },
205 { ARM::VLD3q8oddPseudo_UPD, ARM::VLD3q8_UPD, true, true, OddDblSpc, 3, 8 },
207 { ARM::VLD4DUPd16Pseudo, ARM::VLD4DUPd16, true, false, SingleSpc, 4, 4},
208 { ARM::VLD4DUPd16Pseudo_UPD, ARM::VLD4DUPd16_UPD, true, true, SingleSpc, 4, 4},
209 { ARM::VLD4DUPd32Pseudo, ARM::VLD4DUPd32, true, false, SingleSpc, 4, 2},
210 { ARM::VLD4DUPd32Pseudo_UPD, ARM::VLD4DUPd32_UPD, true, true, SingleSpc, 4, 2},
211 { ARM::VLD4DUPd8Pseudo, ARM::VLD4DUPd8, true, false, SingleSpc, 4, 8},
212 { ARM::VLD4DUPd8Pseudo_UPD, ARM::VLD4DUPd8_UPD, true, true, SingleSpc, 4, 8},
214 { ARM::VLD4LNd16Pseudo, ARM::VLD4LNd16, true, false, SingleSpc, 4, 4 },
215 { ARM::VLD4LNd16Pseudo_UPD, ARM::VLD4LNd16_UPD, true, true, SingleSpc, 4, 4 },
216 { ARM::VLD4LNd32Pseudo, ARM::VLD4LNd32, true, false, SingleSpc, 4, 2 },
217 { ARM::VLD4LNd32Pseudo_UPD, ARM::VLD4LNd32_UPD, true, true, SingleSpc, 4, 2 },
218 { ARM::VLD4LNd8Pseudo, ARM::VLD4LNd8, true, false, SingleSpc, 4, 8 },
219 { ARM::VLD4LNd8Pseudo_UPD, ARM::VLD4LNd8_UPD, true, true, SingleSpc, 4, 8 },
220 { ARM::VLD4LNq16Pseudo, ARM::VLD4LNq16, true, false, EvenDblSpc, 4, 4 },
221 { ARM::VLD4LNq16Pseudo_UPD, ARM::VLD4LNq16_UPD, true, true, EvenDblSpc, 4, 4 },
222 { ARM::VLD4LNq32Pseudo, ARM::VLD4LNq32, true, false, EvenDblSpc, 4, 2 },
223 { ARM::VLD4LNq32Pseudo_UPD, ARM::VLD4LNq32_UPD, true, true, EvenDblSpc, 4, 2 },
225 { ARM::VLD4d16Pseudo, ARM::VLD4d16, true, false, SingleSpc, 4, 4 },
226 { ARM::VLD4d16Pseudo_UPD, ARM::VLD4d16_UPD, true, true, SingleSpc, 4, 4 },
227 { ARM::VLD4d32Pseudo, ARM::VLD4d32, true, false, SingleSpc, 4, 2 },
228 { ARM::VLD4d32Pseudo_UPD, ARM::VLD4d32_UPD, true, true, SingleSpc, 4, 2 },
229 { ARM::VLD4d8Pseudo, ARM::VLD4d8, true, false, SingleSpc, 4, 8 },
230 { ARM::VLD4d8Pseudo_UPD, ARM::VLD4d8_UPD, true, true, SingleSpc, 4, 8 },
232 { ARM::VLD4q16Pseudo_UPD, ARM::VLD4q16_UPD, true, true, EvenDblSpc, 4, 4 },
233 { ARM::VLD4q16oddPseudo_UPD, ARM::VLD4q16_UPD, true, true, OddDblSpc, 4, 4 },
234 { ARM::VLD4q32Pseudo_UPD, ARM::VLD4q32_UPD, true, true, EvenDblSpc, 4, 2 },
235 { ARM::VLD4q32oddPseudo_UPD, ARM::VLD4q32_UPD, true, true, OddDblSpc, 4, 2 },
236 { ARM::VLD4q8Pseudo_UPD, ARM::VLD4q8_UPD, true, true, EvenDblSpc, 4, 8 },
237 { ARM::VLD4q8oddPseudo_UPD, ARM::VLD4q8_UPD, true, true, OddDblSpc, 4, 8 },
239 { ARM::VST1LNq16Pseudo, ARM::VST1LNd16, false, false, EvenDblSpc, 1, 4 },
240 { ARM::VST1LNq16Pseudo_UPD, ARM::VST1LNd16_UPD,false, true, EvenDblSpc, 1, 4 },
241 { ARM::VST1LNq32Pseudo, ARM::VST1LNd32, false, false, EvenDblSpc, 1, 2 },
242 { ARM::VST1LNq32Pseudo_UPD, ARM::VST1LNd32_UPD,false, true, EvenDblSpc, 1, 2 },
243 { ARM::VST1LNq8Pseudo, ARM::VST1LNd8, false, false, EvenDblSpc, 1, 8 },
244 { ARM::VST1LNq8Pseudo_UPD, ARM::VST1LNd8_UPD, false, true, EvenDblSpc, 1, 8 },
246 { ARM::VST1d64QPseudo, ARM::VST1d64Q, false, false, SingleSpc, 4, 1 },
247 { ARM::VST1d64QPseudo_UPD, ARM::VST1d64Q_UPD, false, true, SingleSpc, 4, 1 },
248 { ARM::VST1d64TPseudo, ARM::VST1d64T, false, false, SingleSpc, 3, 1 },
249 { ARM::VST1d64TPseudo_UPD, ARM::VST1d64T_UPD, false, true, SingleSpc, 3, 1 },
251 { ARM::VST1q16Pseudo, ARM::VST1q16, false, false, SingleSpc, 2, 4 },
252 { ARM::VST1q16Pseudo_UPD, ARM::VST1q16_UPD, false, true, SingleSpc, 2, 4 },
253 { ARM::VST1q32Pseudo, ARM::VST1q32, false, false, SingleSpc, 2, 2 },
254 { ARM::VST1q32Pseudo_UPD, ARM::VST1q32_UPD, false, true, SingleSpc, 2, 2 },
255 { ARM::VST1q64Pseudo, ARM::VST1q64, false, false, SingleSpc, 2, 1 },
256 { ARM::VST1q64Pseudo_UPD, ARM::VST1q64_UPD, false, true, SingleSpc, 2, 1 },
257 { ARM::VST1q8Pseudo, ARM::VST1q8, false, false, SingleSpc, 2, 8 },
258 { ARM::VST1q8Pseudo_UPD, ARM::VST1q8_UPD, false, true, SingleSpc, 2, 8 },
260 { ARM::VST2LNd16Pseudo, ARM::VST2LNd16, false, false, SingleSpc, 2, 4 },
261 { ARM::VST2LNd16Pseudo_UPD, ARM::VST2LNd16_UPD, false, true, SingleSpc, 2, 4 },
262 { ARM::VST2LNd32Pseudo, ARM::VST2LNd32, false, false, SingleSpc, 2, 2 },
263 { ARM::VST2LNd32Pseudo_UPD, ARM::VST2LNd32_UPD, false, true, SingleSpc, 2, 2 },
264 { ARM::VST2LNd8Pseudo, ARM::VST2LNd8, false, false, SingleSpc, 2, 8 },
265 { ARM::VST2LNd8Pseudo_UPD, ARM::VST2LNd8_UPD, false, true, SingleSpc, 2, 8 },
266 { ARM::VST2LNq16Pseudo, ARM::VST2LNq16, false, false, EvenDblSpc, 2, 4},
267 { ARM::VST2LNq16Pseudo_UPD, ARM::VST2LNq16_UPD, false, true, EvenDblSpc, 2, 4},
268 { ARM::VST2LNq32Pseudo, ARM::VST2LNq32, false, false, EvenDblSpc, 2, 2},
269 { ARM::VST2LNq32Pseudo_UPD, ARM::VST2LNq32_UPD, false, true, EvenDblSpc, 2, 2},
271 { ARM::VST2d16Pseudo, ARM::VST2d16, false, false, SingleSpc, 2, 4 },
272 { ARM::VST2d16Pseudo_UPD, ARM::VST2d16_UPD, false, true, SingleSpc, 2, 4 },
273 { ARM::VST2d32Pseudo, ARM::VST2d32, false, false, SingleSpc, 2, 2 },
274 { ARM::VST2d32Pseudo_UPD, ARM::VST2d32_UPD, false, true, SingleSpc, 2, 2 },
275 { ARM::VST2d8Pseudo, ARM::VST2d8, false, false, SingleSpc, 2, 8 },
276 { ARM::VST2d8Pseudo_UPD, ARM::VST2d8_UPD, false, true, SingleSpc, 2, 8 },
278 { ARM::VST2q16Pseudo, ARM::VST2q16, false, false, SingleSpc, 4, 4 },
279 { ARM::VST2q16Pseudo_UPD, ARM::VST2q16_UPD, false, true, SingleSpc, 4, 4 },
280 { ARM::VST2q32Pseudo, ARM::VST2q32, false, false, SingleSpc, 4, 2 },
281 { ARM::VST2q32Pseudo_UPD, ARM::VST2q32_UPD, false, true, SingleSpc, 4, 2 },
282 { ARM::VST2q8Pseudo, ARM::VST2q8, false, false, SingleSpc, 4, 8 },
283 { ARM::VST2q8Pseudo_UPD, ARM::VST2q8_UPD, false, true, SingleSpc, 4, 8 },
285 { ARM::VST3LNd16Pseudo, ARM::VST3LNd16, false, false, SingleSpc, 3, 4 },
286 { ARM::VST3LNd16Pseudo_UPD, ARM::VST3LNd16_UPD, false, true, SingleSpc, 3, 4 },
287 { ARM::VST3LNd32Pseudo, ARM::VST3LNd32, false, false, SingleSpc, 3, 2 },
288 { ARM::VST3LNd32Pseudo_UPD, ARM::VST3LNd32_UPD, false, true, SingleSpc, 3, 2 },
289 { ARM::VST3LNd8Pseudo, ARM::VST3LNd8, false, false, SingleSpc, 3, 8 },
290 { ARM::VST3LNd8Pseudo_UPD, ARM::VST3LNd8_UPD, false, true, SingleSpc, 3, 8 },
291 { ARM::VST3LNq16Pseudo, ARM::VST3LNq16, false, false, EvenDblSpc, 3, 4},
292 { ARM::VST3LNq16Pseudo_UPD, ARM::VST3LNq16_UPD, false, true, EvenDblSpc, 3, 4},
293 { ARM::VST3LNq32Pseudo, ARM::VST3LNq32, false, false, EvenDblSpc, 3, 2},
294 { ARM::VST3LNq32Pseudo_UPD, ARM::VST3LNq32_UPD, false, true, EvenDblSpc, 3, 2},
296 { ARM::VST3d16Pseudo, ARM::VST3d16, false, false, SingleSpc, 3, 4 },
297 { ARM::VST3d16Pseudo_UPD, ARM::VST3d16_UPD, false, true, SingleSpc, 3, 4 },
298 { ARM::VST3d32Pseudo, ARM::VST3d32, false, false, SingleSpc, 3, 2 },
299 { ARM::VST3d32Pseudo_UPD, ARM::VST3d32_UPD, false, true, SingleSpc, 3, 2 },
300 { ARM::VST3d8Pseudo, ARM::VST3d8, false, false, SingleSpc, 3, 8 },
301 { ARM::VST3d8Pseudo_UPD, ARM::VST3d8_UPD, false, true, SingleSpc, 3, 8 },
303 { ARM::VST3q16Pseudo_UPD, ARM::VST3q16_UPD, false, true, EvenDblSpc, 3, 4 },
304 { ARM::VST3q16oddPseudo_UPD, ARM::VST3q16_UPD, false, true, OddDblSpc, 3, 4 },
305 { ARM::VST3q32Pseudo_UPD, ARM::VST3q32_UPD, false, true, EvenDblSpc, 3, 2 },
306 { ARM::VST3q32oddPseudo_UPD, ARM::VST3q32_UPD, false, true, OddDblSpc, 3, 2 },
307 { ARM::VST3q8Pseudo_UPD, ARM::VST3q8_UPD, false, true, EvenDblSpc, 3, 8 },
308 { ARM::VST3q8oddPseudo_UPD, ARM::VST3q8_UPD, false, true, OddDblSpc, 3, 8 },
310 { ARM::VST4LNd16Pseudo, ARM::VST4LNd16, false, false, SingleSpc, 4, 4 },
311 { ARM::VST4LNd16Pseudo_UPD, ARM::VST4LNd16_UPD, false, true, SingleSpc, 4, 4 },
312 { ARM::VST4LNd32Pseudo, ARM::VST4LNd32, false, false, SingleSpc, 4, 2 },
313 { ARM::VST4LNd32Pseudo_UPD, ARM::VST4LNd32_UPD, false, true, SingleSpc, 4, 2 },
314 { ARM::VST4LNd8Pseudo, ARM::VST4LNd8, false, false, SingleSpc, 4, 8 },
315 { ARM::VST4LNd8Pseudo_UPD, ARM::VST4LNd8_UPD, false, true, SingleSpc, 4, 8 },
316 { ARM::VST4LNq16Pseudo, ARM::VST4LNq16, false, false, EvenDblSpc, 4, 4},
317 { ARM::VST4LNq16Pseudo_UPD, ARM::VST4LNq16_UPD, false, true, EvenDblSpc, 4, 4},
318 { ARM::VST4LNq32Pseudo, ARM::VST4LNq32, false, false, EvenDblSpc, 4, 2},
319 { ARM::VST4LNq32Pseudo_UPD, ARM::VST4LNq32_UPD, false, true, EvenDblSpc, 4, 2},
321 { ARM::VST4d16Pseudo, ARM::VST4d16, false, false, SingleSpc, 4, 4 },
322 { ARM::VST4d16Pseudo_UPD, ARM::VST4d16_UPD, false, true, SingleSpc, 4, 4 },
323 { ARM::VST4d32Pseudo, ARM::VST4d32, false, false, SingleSpc, 4, 2 },
324 { ARM::VST4d32Pseudo_UPD, ARM::VST4d32_UPD, false, true, SingleSpc, 4, 2 },
325 { ARM::VST4d8Pseudo, ARM::VST4d8, false, false, SingleSpc, 4, 8 },
326 { ARM::VST4d8Pseudo_UPD, ARM::VST4d8_UPD, false, true, SingleSpc, 4, 8 },
328 { ARM::VST4q16Pseudo_UPD, ARM::VST4q16_UPD, false, true, EvenDblSpc, 4, 4 },
329 { ARM::VST4q16oddPseudo_UPD, ARM::VST4q16_UPD, false, true, OddDblSpc, 4, 4 },
330 { ARM::VST4q32Pseudo_UPD, ARM::VST4q32_UPD, false, true, EvenDblSpc, 4, 2 },
331 { ARM::VST4q32oddPseudo_UPD, ARM::VST4q32_UPD, false, true, OddDblSpc, 4, 2 },
332 { ARM::VST4q8Pseudo_UPD, ARM::VST4q8_UPD, false, true, EvenDblSpc, 4, 8 },
333 { ARM::VST4q8oddPseudo_UPD , ARM::VST4q8_UPD, false, true, OddDblSpc, 4, 8 }
336 /// LookupNEONLdSt - Search the NEONLdStTable for information about a NEON
337 /// load or store pseudo instruction.
338 static const NEONLdStTableEntry *LookupNEONLdSt(unsigned Opcode) {
339 unsigned NumEntries = array_lengthof(NEONLdStTable);
342 // Make sure the table is sorted.
343 static bool TableChecked = false;
345 for (unsigned i = 0; i != NumEntries-1; ++i)
346 assert(NEONLdStTable[i] < NEONLdStTable[i+1] &&
347 "NEONLdStTable is not sorted!");
352 const NEONLdStTableEntry *I =
353 std::lower_bound(NEONLdStTable, NEONLdStTable + NumEntries, Opcode);
354 if (I != NEONLdStTable + NumEntries && I->PseudoOpc == Opcode)
359 /// GetDSubRegs - Get 4 D subregisters of a Q, QQ, or QQQQ register,
360 /// corresponding to the specified register spacing. Not all of the results
361 /// are necessarily valid, e.g., a Q register only has 2 D subregisters.
362 static void GetDSubRegs(unsigned Reg, NEONRegSpacing RegSpc,
363 const TargetRegisterInfo *TRI, unsigned &D0,
364 unsigned &D1, unsigned &D2, unsigned &D3) {
365 if (RegSpc == SingleSpc) {
366 D0 = TRI->getSubReg(Reg, ARM::dsub_0);
367 D1 = TRI->getSubReg(Reg, ARM::dsub_1);
368 D2 = TRI->getSubReg(Reg, ARM::dsub_2);
369 D3 = TRI->getSubReg(Reg, ARM::dsub_3);
370 } else if (RegSpc == EvenDblSpc) {
371 D0 = TRI->getSubReg(Reg, ARM::dsub_0);
372 D1 = TRI->getSubReg(Reg, ARM::dsub_2);
373 D2 = TRI->getSubReg(Reg, ARM::dsub_4);
374 D3 = TRI->getSubReg(Reg, ARM::dsub_6);
376 assert(RegSpc == OddDblSpc && "unknown register spacing");
377 D0 = TRI->getSubReg(Reg, ARM::dsub_1);
378 D1 = TRI->getSubReg(Reg, ARM::dsub_3);
379 D2 = TRI->getSubReg(Reg, ARM::dsub_5);
380 D3 = TRI->getSubReg(Reg, ARM::dsub_7);
384 /// ExpandVLD - Translate VLD pseudo instructions with Q, QQ or QQQQ register
385 /// operands to real VLD instructions with D register operands.
386 void ARMExpandPseudo::ExpandVLD(MachineBasicBlock::iterator &MBBI) {
387 MachineInstr &MI = *MBBI;
388 MachineBasicBlock &MBB = *MI.getParent();
390 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
391 assert(TableEntry && TableEntry->IsLoad && "NEONLdStTable lookup failed");
392 NEONRegSpacing RegSpc = TableEntry->RegSpacing;
393 unsigned NumRegs = TableEntry->NumRegs;
395 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
396 TII->get(TableEntry->RealOpc));
399 bool DstIsDead = MI.getOperand(OpIdx).isDead();
400 unsigned DstReg = MI.getOperand(OpIdx++).getReg();
401 unsigned D0, D1, D2, D3;
402 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3);
403 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead))
404 .addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
406 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead));
408 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead));
410 if (TableEntry->HasWriteBack)
411 MIB.addOperand(MI.getOperand(OpIdx++));
413 // Copy the addrmode6 operands.
414 MIB.addOperand(MI.getOperand(OpIdx++));
415 MIB.addOperand(MI.getOperand(OpIdx++));
416 // Copy the am6offset operand.
417 if (TableEntry->HasWriteBack)
418 MIB.addOperand(MI.getOperand(OpIdx++));
420 // For an instruction writing double-spaced subregs, the pseudo instruction
421 // has an extra operand that is a use of the super-register. Record the
422 // operand index and skip over it.
423 unsigned SrcOpIdx = 0;
424 if (RegSpc == EvenDblSpc || RegSpc == OddDblSpc)
427 // Copy the predicate operands.
428 MIB.addOperand(MI.getOperand(OpIdx++));
429 MIB.addOperand(MI.getOperand(OpIdx++));
431 // Copy the super-register source operand used for double-spaced subregs over
432 // to the new instruction as an implicit operand.
434 MachineOperand MO = MI.getOperand(SrcOpIdx);
435 MO.setImplicit(true);
438 // Add an implicit def for the super-register.
439 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
440 TransferImpOps(MI, MIB, MIB);
441 MI.eraseFromParent();
444 /// ExpandVST - Translate VST pseudo instructions with Q, QQ or QQQQ register
445 /// operands to real VST instructions with D register operands.
446 void ARMExpandPseudo::ExpandVST(MachineBasicBlock::iterator &MBBI) {
447 MachineInstr &MI = *MBBI;
448 MachineBasicBlock &MBB = *MI.getParent();
450 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
451 assert(TableEntry && !TableEntry->IsLoad && "NEONLdStTable lookup failed");
452 NEONRegSpacing RegSpc = TableEntry->RegSpacing;
453 unsigned NumRegs = TableEntry->NumRegs;
455 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
456 TII->get(TableEntry->RealOpc));
458 if (TableEntry->HasWriteBack)
459 MIB.addOperand(MI.getOperand(OpIdx++));
461 // Copy the addrmode6 operands.
462 MIB.addOperand(MI.getOperand(OpIdx++));
463 MIB.addOperand(MI.getOperand(OpIdx++));
464 // Copy the am6offset operand.
465 if (TableEntry->HasWriteBack)
466 MIB.addOperand(MI.getOperand(OpIdx++));
468 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
469 unsigned SrcReg = MI.getOperand(OpIdx++).getReg();
470 unsigned D0, D1, D2, D3;
471 GetDSubRegs(SrcReg, RegSpc, TRI, D0, D1, D2, D3);
472 MIB.addReg(D0).addReg(D1);
478 // Copy the predicate operands.
479 MIB.addOperand(MI.getOperand(OpIdx++));
480 MIB.addOperand(MI.getOperand(OpIdx++));
483 // Add an implicit kill for the super-reg.
484 (*MIB).addRegisterKilled(SrcReg, TRI, true);
485 TransferImpOps(MI, MIB, MIB);
486 MI.eraseFromParent();
489 /// ExpandLaneOp - Translate VLD*LN and VST*LN instructions with Q, QQ or QQQQ
490 /// register operands to real instructions with D register operands.
491 void ARMExpandPseudo::ExpandLaneOp(MachineBasicBlock::iterator &MBBI) {
492 MachineInstr &MI = *MBBI;
493 MachineBasicBlock &MBB = *MI.getParent();
495 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
496 assert(TableEntry && "NEONLdStTable lookup failed");
497 NEONRegSpacing RegSpc = TableEntry->RegSpacing;
498 unsigned NumRegs = TableEntry->NumRegs;
499 unsigned RegElts = TableEntry->RegElts;
501 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
502 TII->get(TableEntry->RealOpc));
504 // The lane operand is always the 3rd from last operand, before the 2
505 // predicate operands.
506 unsigned Lane = MI.getOperand(MI.getDesc().getNumOperands() - 3).getImm();
508 // Adjust the lane and spacing as needed for Q registers.
509 assert(RegSpc != OddDblSpc && "unexpected register spacing for VLD/VST-lane");
510 if (RegSpc == EvenDblSpc && Lane >= RegElts) {
514 assert(Lane < RegElts && "out of range lane for VLD/VST-lane");
516 unsigned D0, D1, D2, D3;
518 bool DstIsDead = false;
519 if (TableEntry->IsLoad) {
520 DstIsDead = MI.getOperand(OpIdx).isDead();
521 DstReg = MI.getOperand(OpIdx++).getReg();
522 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3);
523 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead));
525 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
527 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead));
529 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead));
532 if (TableEntry->HasWriteBack)
533 MIB.addOperand(MI.getOperand(OpIdx++));
535 // Copy the addrmode6 operands.
536 MIB.addOperand(MI.getOperand(OpIdx++));
537 MIB.addOperand(MI.getOperand(OpIdx++));
538 // Copy the am6offset operand.
539 if (TableEntry->HasWriteBack)
540 MIB.addOperand(MI.getOperand(OpIdx++));
542 // Grab the super-register source.
543 MachineOperand MO = MI.getOperand(OpIdx++);
544 if (!TableEntry->IsLoad)
545 GetDSubRegs(MO.getReg(), RegSpc, TRI, D0, D1, D2, D3);
547 // Add the subregs as sources of the new instruction.
548 unsigned SrcFlags = (getUndefRegState(MO.isUndef()) |
549 getKillRegState(MO.isKill()));
550 MIB.addReg(D0, SrcFlags);
552 MIB.addReg(D1, SrcFlags);
554 MIB.addReg(D2, SrcFlags);
556 MIB.addReg(D3, SrcFlags);
558 // Add the lane number operand.
562 // Copy the predicate operands.
563 MIB.addOperand(MI.getOperand(OpIdx++));
564 MIB.addOperand(MI.getOperand(OpIdx++));
566 // Copy the super-register source to be an implicit source.
567 MO.setImplicit(true);
569 if (TableEntry->IsLoad)
570 // Add an implicit def for the super-register.
571 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
572 TransferImpOps(MI, MIB, MIB);
573 MI.eraseFromParent();
576 /// ExpandVTBL - Translate VTBL and VTBX pseudo instructions with Q or QQ
577 /// register operands to real instructions with D register operands.
578 void ARMExpandPseudo::ExpandVTBL(MachineBasicBlock::iterator &MBBI,
579 unsigned Opc, bool IsExt, unsigned NumRegs) {
580 MachineInstr &MI = *MBBI;
581 MachineBasicBlock &MBB = *MI.getParent();
583 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc));
586 // Transfer the destination register operand.
587 MIB.addOperand(MI.getOperand(OpIdx++));
589 MIB.addOperand(MI.getOperand(OpIdx++));
591 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
592 unsigned SrcReg = MI.getOperand(OpIdx++).getReg();
593 unsigned D0, D1, D2, D3;
594 GetDSubRegs(SrcReg, SingleSpc, TRI, D0, D1, D2, D3);
595 MIB.addReg(D0).addReg(D1);
601 // Copy the other source register operand.
602 MIB.addOperand(MI.getOperand(OpIdx++));
604 // Copy the predicate operands.
605 MIB.addOperand(MI.getOperand(OpIdx++));
606 MIB.addOperand(MI.getOperand(OpIdx++));
609 // Add an implicit kill for the super-reg.
610 (*MIB).addRegisterKilled(SrcReg, TRI, true);
611 TransferImpOps(MI, MIB, MIB);
612 MI.eraseFromParent();
615 bool ARMExpandPseudo::ExpandMBB(MachineBasicBlock &MBB) {
616 bool Modified = false;
618 MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
620 MachineInstr &MI = *MBBI;
621 MachineBasicBlock::iterator NMBBI = llvm::next(MBBI);
623 bool ModifiedOp = true;
624 unsigned Opcode = MI.getOpcode();
630 case ARM::Int_eh_sjlj_dispatchsetup: {
631 MachineFunction &MF = *MI.getParent()->getParent();
632 const ARMBaseInstrInfo *AII =
633 static_cast<const ARMBaseInstrInfo*>(TII);
634 const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
635 // For functions using a base pointer, we rematerialize it (via the frame
636 // pointer) here since eh.sjlj.setjmp and eh.sjlj.longjmp don't do it
637 // for us. Otherwise, expand to nothing.
638 if (RI.hasBasePointer(MF)) {
639 ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
640 int32_t NumBytes = AFI->getFramePtrSpillOffset();
641 unsigned FramePtr = RI.getFrameRegister(MF);
642 assert(MF.getTarget().getFrameInfo()->hasFP(MF) &&
643 "base pointer without frame pointer?");
645 if (AFI->isThumb2Function()) {
646 llvm::emitT2RegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6,
647 FramePtr, -NumBytes, ARMCC::AL, 0, *TII);
648 } else if (AFI->isThumbFunction()) {
649 llvm::emitThumbRegPlusImmediate(MBB, MBBI, ARM::R6,
651 *TII, RI, MI.getDebugLoc());
653 llvm::emitARMRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6,
654 FramePtr, -NumBytes, ARMCC::AL, 0,
657 // If there's dynamic realignment, adjust for it.
658 if (RI.needsStackRealignment(MF)) {
659 MachineFrameInfo *MFI = MF.getFrameInfo();
660 unsigned MaxAlign = MFI->getMaxAlignment();
661 assert (!AFI->isThumb1OnlyFunction());
662 // Emit bic r6, r6, MaxAlign
663 unsigned bicOpc = AFI->isThumbFunction() ?
664 ARM::t2BICri : ARM::BICri;
665 AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(),
666 TII->get(bicOpc), ARM::R6)
667 .addReg(ARM::R6, RegState::Kill)
668 .addImm(MaxAlign-1)));
672 MI.eraseFromParent();
676 case ARM::MOVsrl_flag:
677 case ARM::MOVsra_flag: {
678 // These are just fancy MOVs insructions.
679 AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVs),
680 MI.getOperand(0).getReg())
681 .addOperand(MI.getOperand(1))
683 .addImm(ARM_AM::getSORegOpc((Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr
685 .addReg(ARM::CPSR, RegState::Define);
686 MI.eraseFromParent();
690 // This encodes as "MOVs Rd, Rm, rrx
691 MachineInstrBuilder MIB =
692 AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVs),
693 MI.getOperand(0).getReg())
694 .addOperand(MI.getOperand(1))
695 .addOperand(MI.getOperand(1))
696 .addImm(ARM_AM::getSORegOpc(ARM_AM::rrx, 0)))
698 TransferImpOps(MI, MIB, MIB);
699 MI.eraseFromParent();
703 MachineInstrBuilder MIB =
704 BuildMI(MBB, MBBI, MI.getDebugLoc(),
706 .addExternalSymbol("__aeabi_read_tp", 0);
708 (*MIB).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
709 TransferImpOps(MI, MIB, MIB);
710 MI.eraseFromParent();
715 case ARM::t2LDRSHpci:
716 case ARM::t2LDRSBpci:
717 case ARM::t2LDRpci: {
719 if (Opcode == ARM::t2LDRpci)
720 NewLdOpc = ARM::t2LDRi12;
721 else if (Opcode == ARM::t2LDRHpci)
722 NewLdOpc = ARM::t2LDRHi12;
723 else if (Opcode == ARM::t2LDRBpci)
724 NewLdOpc = ARM::t2LDRBi12;
725 else if (Opcode == ARM::t2LDRSHpci)
726 NewLdOpc = ARM::t2LDRSHi12;
727 else if (Opcode == ARM::t2LDRSBpci)
728 NewLdOpc = ARM::t2LDRSBi12;
730 llvm_unreachable("Not a known opcode?");
732 unsigned DstReg = MI.getOperand(0).getReg();
733 MachineInstrBuilder MIB =
734 AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(),
735 TII->get(NewLdOpc), DstReg)
737 .addOperand(MI.getOperand(1)));
738 (*MIB).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
739 TransferImpOps(MI, MIB, MIB);
740 MI.eraseFromParent();
743 case ARM::tLDRpci_pic:
744 case ARM::t2LDRpci_pic: {
745 unsigned NewLdOpc = (Opcode == ARM::tLDRpci_pic)
746 ? ARM::tLDRpci : ARM::t2LDRi12;
747 unsigned DstReg = MI.getOperand(0).getReg();
748 bool DstIsDead = MI.getOperand(0).isDead();
749 MachineInstrBuilder MIB1 =
750 BuildMI(MBB, MBBI, MI.getDebugLoc(),
751 TII->get(NewLdOpc), DstReg);
752 if (Opcode == ARM::t2LDRpci_pic) MIB1.addReg(ARM::PC);
753 MIB1.addOperand(MI.getOperand(1));
754 AddDefaultPred(MIB1);
755 (*MIB1).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
756 MachineInstrBuilder MIB2 = BuildMI(MBB, MBBI, MI.getDebugLoc(),
757 TII->get(ARM::tPICADD))
758 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
760 .addOperand(MI.getOperand(2));
761 TransferImpOps(MI, MIB1, MIB2);
762 MI.eraseFromParent();
767 case ARM::MOVCCi32imm:
768 case ARM::t2MOVi32imm:
769 case ARM::t2MOVCCi32imm: {
770 unsigned PredReg = 0;
771 ARMCC::CondCodes Pred = llvm::getInstrPredicate(&MI, PredReg);
772 unsigned DstReg = MI.getOperand(0).getReg();
773 bool DstIsDead = MI.getOperand(0).isDead();
774 bool isCC = Opcode == ARM::MOVCCi32imm || Opcode == ARM::t2MOVCCi32imm;
775 const MachineOperand &MO = MI.getOperand(isCC ? 2 : 1);
776 MachineInstrBuilder LO16, HI16;
778 if (!STI->hasV6T2Ops() &&
779 (Opcode == ARM::MOVi32imm || Opcode == ARM::MOVCCi32imm)) {
780 // Expand into a movi + orr.
781 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVi), DstReg);
782 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::ORRri))
783 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
786 assert (MO.isImm() && "MOVi32imm w/ non-immediate source operand!");
787 unsigned ImmVal = (unsigned)MO.getImm();
788 unsigned SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(ImmVal);
789 unsigned SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(ImmVal);
790 LO16 = LO16.addImm(SOImmValV1);
791 HI16 = HI16.addImm(SOImmValV2);
792 (*LO16).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
793 (*HI16).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
794 LO16.addImm(Pred).addReg(PredReg).addReg(0);
795 HI16.addImm(Pred).addReg(PredReg).addReg(0);
796 TransferImpOps(MI, LO16, HI16);
797 MI.eraseFromParent();
802 (Opcode == ARM::t2MOVi32imm || Opcode == ARM::t2MOVCCi32imm);
804 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(),
805 TII->get(isThumb ? ARM::t2MOVi16 : ARM::MOVi16),
807 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(),
808 TII->get(isThumb ? ARM::t2MOVTi16 : ARM::MOVTi16))
809 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead))
813 unsigned Imm = MO.getImm();
814 unsigned Lo16 = Imm & 0xffff;
815 unsigned Hi16 = (Imm >> 16) & 0xffff;
816 LO16 = LO16.addImm(Lo16);
817 HI16 = HI16.addImm(Hi16);
819 const GlobalValue *GV = MO.getGlobal();
820 unsigned TF = MO.getTargetFlags();
821 LO16 = LO16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_LO16);
822 HI16 = HI16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_HI16);
824 (*LO16).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
825 (*HI16).setMemRefs(MI.memoperands_begin(), MI.memoperands_end());
826 LO16.addImm(Pred).addReg(PredReg);
827 HI16.addImm(Pred).addReg(PredReg);
828 TransferImpOps(MI, LO16, HI16);
829 MI.eraseFromParent();
834 unsigned DstReg = MI.getOperand(0).getReg();
835 bool DstIsDead = MI.getOperand(0).isDead();
836 unsigned EvenDst = TRI->getSubReg(DstReg, ARM::qsub_0);
837 unsigned OddDst = TRI->getSubReg(DstReg, ARM::qsub_1);
838 unsigned SrcReg = MI.getOperand(1).getReg();
839 bool SrcIsKill = MI.getOperand(1).isKill();
840 unsigned EvenSrc = TRI->getSubReg(SrcReg, ARM::qsub_0);
841 unsigned OddSrc = TRI->getSubReg(SrcReg, ARM::qsub_1);
842 MachineInstrBuilder Even =
843 AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(),
844 TII->get(ARM::VMOVQ))
846 RegState::Define | getDeadRegState(DstIsDead))
847 .addReg(EvenSrc, getKillRegState(SrcIsKill)));
848 MachineInstrBuilder Odd =
849 AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(),
850 TII->get(ARM::VMOVQ))
852 RegState::Define | getDeadRegState(DstIsDead))
853 .addReg(OddSrc, getKillRegState(SrcIsKill)));
854 TransferImpOps(MI, Even, Odd);
855 MI.eraseFromParent();
861 unsigned NewOpc = (Opcode == ARM::VLDMQIA) ? ARM::VLDMDIA : ARM::VLDMDDB;
862 MachineInstrBuilder MIB =
863 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc));
866 // Grab the Q register destination.
867 bool DstIsDead = MI.getOperand(OpIdx).isDead();
868 unsigned DstReg = MI.getOperand(OpIdx++).getReg();
870 // Copy the source register.
871 MIB.addOperand(MI.getOperand(OpIdx++));
873 // Copy the predicate operands.
874 MIB.addOperand(MI.getOperand(OpIdx++));
875 MIB.addOperand(MI.getOperand(OpIdx++));
877 // Add the destination operands (D subregs).
878 unsigned D0 = TRI->getSubReg(DstReg, ARM::dsub_0);
879 unsigned D1 = TRI->getSubReg(DstReg, ARM::dsub_1);
880 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead))
881 .addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
883 // Add an implicit def for the super-register.
884 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
885 TransferImpOps(MI, MIB, MIB);
886 MI.eraseFromParent();
892 unsigned NewOpc = (Opcode == ARM::VSTMQIA) ? ARM::VSTMDIA : ARM::VSTMDDB;
893 MachineInstrBuilder MIB =
894 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc));
897 // Grab the Q register source.
898 bool SrcIsKill = MI.getOperand(OpIdx).isKill();
899 unsigned SrcReg = MI.getOperand(OpIdx++).getReg();
901 // Copy the destination register.
902 MIB.addOperand(MI.getOperand(OpIdx++));
904 // Copy the predicate operands.
905 MIB.addOperand(MI.getOperand(OpIdx++));
906 MIB.addOperand(MI.getOperand(OpIdx++));
908 // Add the source operands (D subregs).
909 unsigned D0 = TRI->getSubReg(SrcReg, ARM::dsub_0);
910 unsigned D1 = TRI->getSubReg(SrcReg, ARM::dsub_1);
911 MIB.addReg(D0).addReg(D1);
914 // Add an implicit kill for the Q register.
915 (*MIB).addRegisterKilled(SrcReg, TRI, true);
917 TransferImpOps(MI, MIB, MIB);
918 MI.eraseFromParent();
923 unsigned NewOpc = Opcode == ARM::VDUPfqf ? ARM::VDUPLNfq : ARM::VDUPLNfd;
924 MachineInstrBuilder MIB =
925 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc));
927 unsigned SrcReg = MI.getOperand(1).getReg();
928 unsigned Lane = getARMRegisterNumbering(SrcReg) & 1;
929 unsigned DReg = TRI->getMatchingSuperReg(SrcReg,
930 Lane & 1 ? ARM::ssub_1 : ARM::ssub_0, &ARM::DPR_VFP2RegClass);
931 // The lane is [0,1] for the containing DReg superregister.
932 // Copy the dst/src register operands.
933 MIB.addOperand(MI.getOperand(OpIdx++));
936 // Add the lane select operand.
938 // Add the predicate operands.
939 MIB.addOperand(MI.getOperand(OpIdx++));
940 MIB.addOperand(MI.getOperand(OpIdx++));
942 TransferImpOps(MI, MIB, MIB);
943 MI.eraseFromParent();
947 case ARM::VLD1q8Pseudo:
948 case ARM::VLD1q16Pseudo:
949 case ARM::VLD1q32Pseudo:
950 case ARM::VLD1q64Pseudo:
951 case ARM::VLD1q8Pseudo_UPD:
952 case ARM::VLD1q16Pseudo_UPD:
953 case ARM::VLD1q32Pseudo_UPD:
954 case ARM::VLD1q64Pseudo_UPD:
955 case ARM::VLD2d8Pseudo:
956 case ARM::VLD2d16Pseudo:
957 case ARM::VLD2d32Pseudo:
958 case ARM::VLD2q8Pseudo:
959 case ARM::VLD2q16Pseudo:
960 case ARM::VLD2q32Pseudo:
961 case ARM::VLD2d8Pseudo_UPD:
962 case ARM::VLD2d16Pseudo_UPD:
963 case ARM::VLD2d32Pseudo_UPD:
964 case ARM::VLD2q8Pseudo_UPD:
965 case ARM::VLD2q16Pseudo_UPD:
966 case ARM::VLD2q32Pseudo_UPD:
967 case ARM::VLD3d8Pseudo:
968 case ARM::VLD3d16Pseudo:
969 case ARM::VLD3d32Pseudo:
970 case ARM::VLD1d64TPseudo:
971 case ARM::VLD3d8Pseudo_UPD:
972 case ARM::VLD3d16Pseudo_UPD:
973 case ARM::VLD3d32Pseudo_UPD:
974 case ARM::VLD1d64TPseudo_UPD:
975 case ARM::VLD3q8Pseudo_UPD:
976 case ARM::VLD3q16Pseudo_UPD:
977 case ARM::VLD3q32Pseudo_UPD:
978 case ARM::VLD3q8oddPseudo_UPD:
979 case ARM::VLD3q16oddPseudo_UPD:
980 case ARM::VLD3q32oddPseudo_UPD:
981 case ARM::VLD4d8Pseudo:
982 case ARM::VLD4d16Pseudo:
983 case ARM::VLD4d32Pseudo:
984 case ARM::VLD1d64QPseudo:
985 case ARM::VLD4d8Pseudo_UPD:
986 case ARM::VLD4d16Pseudo_UPD:
987 case ARM::VLD4d32Pseudo_UPD:
988 case ARM::VLD1d64QPseudo_UPD:
989 case ARM::VLD4q8Pseudo_UPD:
990 case ARM::VLD4q16Pseudo_UPD:
991 case ARM::VLD4q32Pseudo_UPD:
992 case ARM::VLD4q8oddPseudo_UPD:
993 case ARM::VLD4q16oddPseudo_UPD:
994 case ARM::VLD4q32oddPseudo_UPD:
995 case ARM::VLD1DUPq8Pseudo:
996 case ARM::VLD1DUPq16Pseudo:
997 case ARM::VLD1DUPq32Pseudo:
998 case ARM::VLD1DUPq8Pseudo_UPD:
999 case ARM::VLD1DUPq16Pseudo_UPD:
1000 case ARM::VLD1DUPq32Pseudo_UPD:
1001 case ARM::VLD2DUPd8Pseudo:
1002 case ARM::VLD2DUPd16Pseudo:
1003 case ARM::VLD2DUPd32Pseudo:
1004 case ARM::VLD2DUPd8Pseudo_UPD:
1005 case ARM::VLD2DUPd16Pseudo_UPD:
1006 case ARM::VLD2DUPd32Pseudo_UPD:
1007 case ARM::VLD3DUPd8Pseudo:
1008 case ARM::VLD3DUPd16Pseudo:
1009 case ARM::VLD3DUPd32Pseudo:
1010 case ARM::VLD3DUPd8Pseudo_UPD:
1011 case ARM::VLD3DUPd16Pseudo_UPD:
1012 case ARM::VLD3DUPd32Pseudo_UPD:
1013 case ARM::VLD4DUPd8Pseudo:
1014 case ARM::VLD4DUPd16Pseudo:
1015 case ARM::VLD4DUPd32Pseudo:
1016 case ARM::VLD4DUPd8Pseudo_UPD:
1017 case ARM::VLD4DUPd16Pseudo_UPD:
1018 case ARM::VLD4DUPd32Pseudo_UPD:
1022 case ARM::VST1q8Pseudo:
1023 case ARM::VST1q16Pseudo:
1024 case ARM::VST1q32Pseudo:
1025 case ARM::VST1q64Pseudo:
1026 case ARM::VST1q8Pseudo_UPD:
1027 case ARM::VST1q16Pseudo_UPD:
1028 case ARM::VST1q32Pseudo_UPD:
1029 case ARM::VST1q64Pseudo_UPD:
1030 case ARM::VST2d8Pseudo:
1031 case ARM::VST2d16Pseudo:
1032 case ARM::VST2d32Pseudo:
1033 case ARM::VST2q8Pseudo:
1034 case ARM::VST2q16Pseudo:
1035 case ARM::VST2q32Pseudo:
1036 case ARM::VST2d8Pseudo_UPD:
1037 case ARM::VST2d16Pseudo_UPD:
1038 case ARM::VST2d32Pseudo_UPD:
1039 case ARM::VST2q8Pseudo_UPD:
1040 case ARM::VST2q16Pseudo_UPD:
1041 case ARM::VST2q32Pseudo_UPD:
1042 case ARM::VST3d8Pseudo:
1043 case ARM::VST3d16Pseudo:
1044 case ARM::VST3d32Pseudo:
1045 case ARM::VST1d64TPseudo:
1046 case ARM::VST3d8Pseudo_UPD:
1047 case ARM::VST3d16Pseudo_UPD:
1048 case ARM::VST3d32Pseudo_UPD:
1049 case ARM::VST1d64TPseudo_UPD:
1050 case ARM::VST3q8Pseudo_UPD:
1051 case ARM::VST3q16Pseudo_UPD:
1052 case ARM::VST3q32Pseudo_UPD:
1053 case ARM::VST3q8oddPseudo_UPD:
1054 case ARM::VST3q16oddPseudo_UPD:
1055 case ARM::VST3q32oddPseudo_UPD:
1056 case ARM::VST4d8Pseudo:
1057 case ARM::VST4d16Pseudo:
1058 case ARM::VST4d32Pseudo:
1059 case ARM::VST1d64QPseudo:
1060 case ARM::VST4d8Pseudo_UPD:
1061 case ARM::VST4d16Pseudo_UPD:
1062 case ARM::VST4d32Pseudo_UPD:
1063 case ARM::VST1d64QPseudo_UPD:
1064 case ARM::VST4q8Pseudo_UPD:
1065 case ARM::VST4q16Pseudo_UPD:
1066 case ARM::VST4q32Pseudo_UPD:
1067 case ARM::VST4q8oddPseudo_UPD:
1068 case ARM::VST4q16oddPseudo_UPD:
1069 case ARM::VST4q32oddPseudo_UPD:
1073 case ARM::VLD1LNq8Pseudo:
1074 case ARM::VLD1LNq16Pseudo:
1075 case ARM::VLD1LNq32Pseudo:
1076 case ARM::VLD1LNq8Pseudo_UPD:
1077 case ARM::VLD1LNq16Pseudo_UPD:
1078 case ARM::VLD1LNq32Pseudo_UPD:
1079 case ARM::VLD2LNd8Pseudo:
1080 case ARM::VLD2LNd16Pseudo:
1081 case ARM::VLD2LNd32Pseudo:
1082 case ARM::VLD2LNq16Pseudo:
1083 case ARM::VLD2LNq32Pseudo:
1084 case ARM::VLD2LNd8Pseudo_UPD:
1085 case ARM::VLD2LNd16Pseudo_UPD:
1086 case ARM::VLD2LNd32Pseudo_UPD:
1087 case ARM::VLD2LNq16Pseudo_UPD:
1088 case ARM::VLD2LNq32Pseudo_UPD:
1089 case ARM::VLD3LNd8Pseudo:
1090 case ARM::VLD3LNd16Pseudo:
1091 case ARM::VLD3LNd32Pseudo:
1092 case ARM::VLD3LNq16Pseudo:
1093 case ARM::VLD3LNq32Pseudo:
1094 case ARM::VLD3LNd8Pseudo_UPD:
1095 case ARM::VLD3LNd16Pseudo_UPD:
1096 case ARM::VLD3LNd32Pseudo_UPD:
1097 case ARM::VLD3LNq16Pseudo_UPD:
1098 case ARM::VLD3LNq32Pseudo_UPD:
1099 case ARM::VLD4LNd8Pseudo:
1100 case ARM::VLD4LNd16Pseudo:
1101 case ARM::VLD4LNd32Pseudo:
1102 case ARM::VLD4LNq16Pseudo:
1103 case ARM::VLD4LNq32Pseudo:
1104 case ARM::VLD4LNd8Pseudo_UPD:
1105 case ARM::VLD4LNd16Pseudo_UPD:
1106 case ARM::VLD4LNd32Pseudo_UPD:
1107 case ARM::VLD4LNq16Pseudo_UPD:
1108 case ARM::VLD4LNq32Pseudo_UPD:
1109 case ARM::VST1LNq8Pseudo:
1110 case ARM::VST1LNq16Pseudo:
1111 case ARM::VST1LNq32Pseudo:
1112 case ARM::VST1LNq8Pseudo_UPD:
1113 case ARM::VST1LNq16Pseudo_UPD:
1114 case ARM::VST1LNq32Pseudo_UPD:
1115 case ARM::VST2LNd8Pseudo:
1116 case ARM::VST2LNd16Pseudo:
1117 case ARM::VST2LNd32Pseudo:
1118 case ARM::VST2LNq16Pseudo:
1119 case ARM::VST2LNq32Pseudo:
1120 case ARM::VST2LNd8Pseudo_UPD:
1121 case ARM::VST2LNd16Pseudo_UPD:
1122 case ARM::VST2LNd32Pseudo_UPD:
1123 case ARM::VST2LNq16Pseudo_UPD:
1124 case ARM::VST2LNq32Pseudo_UPD:
1125 case ARM::VST3LNd8Pseudo:
1126 case ARM::VST3LNd16Pseudo:
1127 case ARM::VST3LNd32Pseudo:
1128 case ARM::VST3LNq16Pseudo:
1129 case ARM::VST3LNq32Pseudo:
1130 case ARM::VST3LNd8Pseudo_UPD:
1131 case ARM::VST3LNd16Pseudo_UPD:
1132 case ARM::VST3LNd32Pseudo_UPD:
1133 case ARM::VST3LNq16Pseudo_UPD:
1134 case ARM::VST3LNq32Pseudo_UPD:
1135 case ARM::VST4LNd8Pseudo:
1136 case ARM::VST4LNd16Pseudo:
1137 case ARM::VST4LNd32Pseudo:
1138 case ARM::VST4LNq16Pseudo:
1139 case ARM::VST4LNq32Pseudo:
1140 case ARM::VST4LNd8Pseudo_UPD:
1141 case ARM::VST4LNd16Pseudo_UPD:
1142 case ARM::VST4LNd32Pseudo_UPD:
1143 case ARM::VST4LNq16Pseudo_UPD:
1144 case ARM::VST4LNq32Pseudo_UPD:
1148 case ARM::VTBL2Pseudo: ExpandVTBL(MBBI, ARM::VTBL2, false, 2); break;
1149 case ARM::VTBL3Pseudo: ExpandVTBL(MBBI, ARM::VTBL3, false, 3); break;
1150 case ARM::VTBL4Pseudo: ExpandVTBL(MBBI, ARM::VTBL4, false, 4); break;
1151 case ARM::VTBX2Pseudo: ExpandVTBL(MBBI, ARM::VTBX2, true, 2); break;
1152 case ARM::VTBX3Pseudo: ExpandVTBL(MBBI, ARM::VTBX3, true, 3); break;
1153 case ARM::VTBX4Pseudo: ExpandVTBL(MBBI, ARM::VTBX4, true, 4); break;
1164 bool ARMExpandPseudo::runOnMachineFunction(MachineFunction &MF) {
1165 TII = static_cast<const ARMBaseInstrInfo*>(MF.getTarget().getInstrInfo());
1166 TRI = MF.getTarget().getRegisterInfo();
1167 STI = &MF.getTarget().getSubtarget<ARMSubtarget>();
1169 bool Modified = false;
1170 for (MachineFunction::iterator MFI = MF.begin(), E = MF.end(); MFI != E;
1172 Modified |= ExpandMBB(*MFI);
1176 /// createARMExpandPseudoPass - returns an instance of the pseudo instruction
1178 FunctionPass *llvm::createARMExpandPseudoPass() {
1179 return new ARMExpandPseudo();