1 //===- X86RegisterInfo.cpp - X86 Register Information -----------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains the X86 implementation of the MRegisterInfo class. This
11 // file is responsible for the frame pointer elimination optimization on X86.
13 //===----------------------------------------------------------------------===//
16 #include "X86RegisterInfo.h"
17 #include "X86InstrBuilder.h"
18 #include "X86MachineFunctionInfo.h"
19 #include "X86Subtarget.h"
20 #include "X86TargetMachine.h"
21 #include "llvm/Constants.h"
22 #include "llvm/Function.h"
23 #include "llvm/Type.h"
24 #include "llvm/CodeGen/ValueTypes.h"
25 #include "llvm/CodeGen/MachineInstrBuilder.h"
26 #include "llvm/CodeGen/MachineFunction.h"
27 #include "llvm/CodeGen/MachineFrameInfo.h"
28 #include "llvm/CodeGen/MachineLocation.h"
29 #include "llvm/CodeGen/SSARegMap.h"
30 #include "llvm/Target/TargetAsmInfo.h"
31 #include "llvm/Target/TargetFrameInfo.h"
32 #include "llvm/Target/TargetInstrInfo.h"
33 #include "llvm/Target/TargetMachine.h"
34 #include "llvm/Target/TargetOptions.h"
35 #include "llvm/Support/CommandLine.h"
36 #include "llvm/ADT/BitVector.h"
37 #include "llvm/ADT/STLExtras.h"
42 NoFusing("disable-spill-fusing",
43 cl::desc("Disable fusing of spill code into instructions"));
45 PrintFailedFusing("print-failed-fuse-candidates",
46 cl::desc("Print instructions that the allocator wants to"
47 " fuse, but the X86 backend currently can't"),
51 X86RegisterInfo::X86RegisterInfo(X86TargetMachine &tm,
52 const TargetInstrInfo &tii)
53 : X86GenRegisterInfo(X86::ADJCALLSTACKDOWN, X86::ADJCALLSTACKUP),
55 // Cache some information.
56 const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
57 Is64Bit = Subtarget->is64Bit();
68 SmallVector<unsigned,16> AmbEntries;
69 static const unsigned OpTbl2Addr[][2] = {
70 { X86::ADC32ri, X86::ADC32mi },
71 { X86::ADC32ri8, X86::ADC32mi8 },
72 { X86::ADC32rr, X86::ADC32mr },
73 { X86::ADC64ri32, X86::ADC64mi32 },
74 { X86::ADC64ri8, X86::ADC64mi8 },
75 { X86::ADC64rr, X86::ADC64mr },
76 { X86::ADD16ri, X86::ADD16mi },
77 { X86::ADD16ri8, X86::ADD16mi8 },
78 { X86::ADD16rr, X86::ADD16mr },
79 { X86::ADD32ri, X86::ADD32mi },
80 { X86::ADD32ri8, X86::ADD32mi8 },
81 { X86::ADD32rr, X86::ADD32mr },
82 { X86::ADD64ri32, X86::ADD64mi32 },
83 { X86::ADD64ri8, X86::ADD64mi8 },
84 { X86::ADD64rr, X86::ADD64mr },
85 { X86::ADD8ri, X86::ADD8mi },
86 { X86::ADD8rr, X86::ADD8mr },
87 { X86::AND16ri, X86::AND16mi },
88 { X86::AND16ri8, X86::AND16mi8 },
89 { X86::AND16rr, X86::AND16mr },
90 { X86::AND32ri, X86::AND32mi },
91 { X86::AND32ri8, X86::AND32mi8 },
92 { X86::AND32rr, X86::AND32mr },
93 { X86::AND64ri32, X86::AND64mi32 },
94 { X86::AND64ri8, X86::AND64mi8 },
95 { X86::AND64rr, X86::AND64mr },
96 { X86::AND8ri, X86::AND8mi },
97 { X86::AND8rr, X86::AND8mr },
98 { X86::DEC16r, X86::DEC16m },
99 { X86::DEC32r, X86::DEC32m },
100 { X86::DEC64_16r, X86::DEC64_16m },
101 { X86::DEC64_32r, X86::DEC64_32m },
102 { X86::DEC64r, X86::DEC64m },
103 { X86::DEC8r, X86::DEC8m },
104 { X86::INC16r, X86::INC16m },
105 { X86::INC32r, X86::INC32m },
106 { X86::INC64_16r, X86::INC64_16m },
107 { X86::INC64_32r, X86::INC64_32m },
108 { X86::INC64r, X86::INC64m },
109 { X86::INC8r, X86::INC8m },
110 { X86::NEG16r, X86::NEG16m },
111 { X86::NEG32r, X86::NEG32m },
112 { X86::NEG64r, X86::NEG64m },
113 { X86::NEG8r, X86::NEG8m },
114 { X86::NOT16r, X86::NOT16m },
115 { X86::NOT32r, X86::NOT32m },
116 { X86::NOT64r, X86::NOT64m },
117 { X86::NOT8r, X86::NOT8m },
118 { X86::OR16ri, X86::OR16mi },
119 { X86::OR16ri8, X86::OR16mi8 },
120 { X86::OR16rr, X86::OR16mr },
121 { X86::OR32ri, X86::OR32mi },
122 { X86::OR32ri8, X86::OR32mi8 },
123 { X86::OR32rr, X86::OR32mr },
124 { X86::OR64ri32, X86::OR64mi32 },
125 { X86::OR64ri8, X86::OR64mi8 },
126 { X86::OR64rr, X86::OR64mr },
127 { X86::OR8ri, X86::OR8mi },
128 { X86::OR8rr, X86::OR8mr },
129 { X86::ROL16r1, X86::ROL16m1 },
130 { X86::ROL16rCL, X86::ROL16mCL },
131 { X86::ROL16ri, X86::ROL16mi },
132 { X86::ROL32r1, X86::ROL32m1 },
133 { X86::ROL32rCL, X86::ROL32mCL },
134 { X86::ROL32ri, X86::ROL32mi },
135 { X86::ROL64r1, X86::ROL64m1 },
136 { X86::ROL64rCL, X86::ROL64mCL },
137 { X86::ROL64ri, X86::ROL64mi },
138 { X86::ROL8r1, X86::ROL8m1 },
139 { X86::ROL8rCL, X86::ROL8mCL },
140 { X86::ROL8ri, X86::ROL8mi },
141 { X86::ROR16r1, X86::ROR16m1 },
142 { X86::ROR16rCL, X86::ROR16mCL },
143 { X86::ROR16ri, X86::ROR16mi },
144 { X86::ROR32r1, X86::ROR32m1 },
145 { X86::ROR32rCL, X86::ROR32mCL },
146 { X86::ROR32ri, X86::ROR32mi },
147 { X86::ROR64r1, X86::ROR64m1 },
148 { X86::ROR64rCL, X86::ROR64mCL },
149 { X86::ROR64ri, X86::ROR64mi },
150 { X86::ROR8r1, X86::ROR8m1 },
151 { X86::ROR8rCL, X86::ROR8mCL },
152 { X86::ROR8ri, X86::ROR8mi },
153 { X86::SAR16r1, X86::SAR16m1 },
154 { X86::SAR16rCL, X86::SAR16mCL },
155 { X86::SAR16ri, X86::SAR16mi },
156 { X86::SAR32r1, X86::SAR32m1 },
157 { X86::SAR32rCL, X86::SAR32mCL },
158 { X86::SAR32ri, X86::SAR32mi },
159 { X86::SAR64r1, X86::SAR64m1 },
160 { X86::SAR64rCL, X86::SAR64mCL },
161 { X86::SAR64ri, X86::SAR64mi },
162 { X86::SAR8r1, X86::SAR8m1 },
163 { X86::SAR8rCL, X86::SAR8mCL },
164 { X86::SAR8ri, X86::SAR8mi },
165 { X86::SBB32ri, X86::SBB32mi },
166 { X86::SBB32ri8, X86::SBB32mi8 },
167 { X86::SBB32rr, X86::SBB32mr },
168 { X86::SBB64ri32, X86::SBB64mi32 },
169 { X86::SBB64ri8, X86::SBB64mi8 },
170 { X86::SBB64rr, X86::SBB64mr },
171 { X86::SHL16r1, X86::SHL16m1 },
172 { X86::SHL16rCL, X86::SHL16mCL },
173 { X86::SHL16ri, X86::SHL16mi },
174 { X86::SHL32r1, X86::SHL32m1 },
175 { X86::SHL32rCL, X86::SHL32mCL },
176 { X86::SHL32ri, X86::SHL32mi },
177 { X86::SHL64r1, X86::SHL64m1 },
178 { X86::SHL64rCL, X86::SHL64mCL },
179 { X86::SHL64ri, X86::SHL64mi },
180 { X86::SHL8r1, X86::SHL8m1 },
181 { X86::SHL8rCL, X86::SHL8mCL },
182 { X86::SHL8ri, X86::SHL8mi },
183 { X86::SHLD16rrCL, X86::SHLD16mrCL },
184 { X86::SHLD16rri8, X86::SHLD16mri8 },
185 { X86::SHLD32rrCL, X86::SHLD32mrCL },
186 { X86::SHLD32rri8, X86::SHLD32mri8 },
187 { X86::SHLD64rrCL, X86::SHLD64mrCL },
188 { X86::SHLD64rri8, X86::SHLD64mri8 },
189 { X86::SHR16r1, X86::SHR16m1 },
190 { X86::SHR16rCL, X86::SHR16mCL },
191 { X86::SHR16ri, X86::SHR16mi },
192 { X86::SHR32r1, X86::SHR32m1 },
193 { X86::SHR32rCL, X86::SHR32mCL },
194 { X86::SHR32ri, X86::SHR32mi },
195 { X86::SHR64r1, X86::SHR64m1 },
196 { X86::SHR64rCL, X86::SHR64mCL },
197 { X86::SHR64ri, X86::SHR64mi },
198 { X86::SHR8r1, X86::SHR8m1 },
199 { X86::SHR8rCL, X86::SHR8mCL },
200 { X86::SHR8ri, X86::SHR8mi },
201 { X86::SHRD16rrCL, X86::SHRD16mrCL },
202 { X86::SHRD16rri8, X86::SHRD16mri8 },
203 { X86::SHRD32rrCL, X86::SHRD32mrCL },
204 { X86::SHRD32rri8, X86::SHRD32mri8 },
205 { X86::SHRD64rrCL, X86::SHRD64mrCL },
206 { X86::SHRD64rri8, X86::SHRD64mri8 },
207 { X86::SUB16ri, X86::SUB16mi },
208 { X86::SUB16ri8, X86::SUB16mi8 },
209 { X86::SUB16rr, X86::SUB16mr },
210 { X86::SUB32ri, X86::SUB32mi },
211 { X86::SUB32ri8, X86::SUB32mi8 },
212 { X86::SUB32rr, X86::SUB32mr },
213 { X86::SUB64ri32, X86::SUB64mi32 },
214 { X86::SUB64ri8, X86::SUB64mi8 },
215 { X86::SUB64rr, X86::SUB64mr },
216 { X86::SUB8ri, X86::SUB8mi },
217 { X86::SUB8rr, X86::SUB8mr },
218 { X86::XOR16ri, X86::XOR16mi },
219 { X86::XOR16ri8, X86::XOR16mi8 },
220 { X86::XOR16rr, X86::XOR16mr },
221 { X86::XOR32ri, X86::XOR32mi },
222 { X86::XOR32ri8, X86::XOR32mi8 },
223 { X86::XOR32rr, X86::XOR32mr },
224 { X86::XOR64ri32, X86::XOR64mi32 },
225 { X86::XOR64ri8, X86::XOR64mi8 },
226 { X86::XOR64rr, X86::XOR64mr },
227 { X86::XOR8ri, X86::XOR8mi },
228 { X86::XOR8rr, X86::XOR8mr }
231 for (unsigned i = 0, e = array_lengthof(OpTbl2Addr); i != e; ++i) {
232 unsigned RegOp = OpTbl2Addr[i][0];
233 unsigned MemOp = OpTbl2Addr[i][1];
234 if (!RegOp2MemOpTable2Addr.insert(std::make_pair((unsigned*)RegOp, MemOp)))
235 assert(false && "Duplicated entries?");
236 unsigned AuxInfo = 0 | (1 << 4) | (1 << 5); // Index 0,folded load and store
237 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
238 std::make_pair(RegOp, AuxInfo))))
239 AmbEntries.push_back(MemOp);
242 // If the third value is 1, then it's folding either a load or a store.
243 static const unsigned OpTbl0[][3] = {
244 { X86::CALL32r, X86::CALL32m, 1 },
245 { X86::CALL64r, X86::CALL64m, 1 },
246 { X86::CMP16ri, X86::CMP16mi, 1 },
247 { X86::CMP16ri8, X86::CMP16mi8, 1 },
248 { X86::CMP32ri, X86::CMP32mi, 1 },
249 { X86::CMP32ri8, X86::CMP32mi8, 1 },
250 { X86::CMP64ri32, X86::CMP64mi32, 1 },
251 { X86::CMP64ri8, X86::CMP64mi8, 1 },
252 { X86::CMP8ri, X86::CMP8mi, 1 },
253 { X86::DIV16r, X86::DIV16m, 1 },
254 { X86::DIV32r, X86::DIV32m, 1 },
255 { X86::DIV64r, X86::DIV64m, 1 },
256 { X86::DIV8r, X86::DIV8m, 1 },
257 { X86::FsMOVAPDrr, X86::MOVSDmr, 0 },
258 { X86::FsMOVAPSrr, X86::MOVSSmr, 0 },
259 { X86::IDIV16r, X86::IDIV16m, 1 },
260 { X86::IDIV32r, X86::IDIV32m, 1 },
261 { X86::IDIV64r, X86::IDIV64m, 1 },
262 { X86::IDIV8r, X86::IDIV8m, 1 },
263 { X86::IMUL16r, X86::IMUL16m, 1 },
264 { X86::IMUL32r, X86::IMUL32m, 1 },
265 { X86::IMUL64r, X86::IMUL64m, 1 },
266 { X86::IMUL8r, X86::IMUL8m, 1 },
267 { X86::JMP32r, X86::JMP32m, 1 },
268 { X86::JMP64r, X86::JMP64m, 1 },
269 { X86::MOV16ri, X86::MOV16mi, 0 },
270 { X86::MOV16rr, X86::MOV16mr, 0 },
271 { X86::MOV16to16_, X86::MOV16_mr, 0 },
272 { X86::MOV32ri, X86::MOV32mi, 0 },
273 { X86::MOV32rr, X86::MOV32mr, 0 },
274 { X86::MOV32to32_, X86::MOV32_mr, 0 },
275 { X86::MOV64ri32, X86::MOV64mi32, 0 },
276 { X86::MOV64rr, X86::MOV64mr, 0 },
277 { X86::MOV8ri, X86::MOV8mi, 0 },
278 { X86::MOV8rr, X86::MOV8mr, 0 },
279 { X86::MOVAPDrr, X86::MOVAPDmr, 0 },
280 { X86::MOVAPSrr, X86::MOVAPSmr, 0 },
281 { X86::MOVPDI2DIrr, X86::MOVPDI2DImr, 0 },
282 { X86::MOVPQIto64rr,X86::MOVPQIto64mr, 0 },
283 { X86::MOVPS2SSrr, X86::MOVPS2SSmr, 0 },
284 { X86::MOVSDrr, X86::MOVSDmr, 0 },
285 { X86::MOVSDto64rr, X86::MOVSDto64mr, 0 },
286 { X86::MOVSS2DIrr, X86::MOVSS2DImr, 0 },
287 { X86::MOVSSrr, X86::MOVSSmr, 0 },
288 { X86::MOVUPDrr, X86::MOVUPDmr, 0 },
289 { X86::MOVUPSrr, X86::MOVUPSmr, 0 },
290 { X86::MUL16r, X86::MUL16m, 1 },
291 { X86::MUL32r, X86::MUL32m, 1 },
292 { X86::MUL64r, X86::MUL64m, 1 },
293 { X86::MUL8r, X86::MUL8m, 1 },
294 { X86::SETAEr, X86::SETAEm, 0 },
295 { X86::SETAr, X86::SETAm, 0 },
296 { X86::SETBEr, X86::SETBEm, 0 },
297 { X86::SETBr, X86::SETBm, 0 },
298 { X86::SETEr, X86::SETEm, 0 },
299 { X86::SETGEr, X86::SETGEm, 0 },
300 { X86::SETGr, X86::SETGm, 0 },
301 { X86::SETLEr, X86::SETLEm, 0 },
302 { X86::SETLr, X86::SETLm, 0 },
303 { X86::SETNEr, X86::SETNEm, 0 },
304 { X86::SETNPr, X86::SETNPm, 0 },
305 { X86::SETNSr, X86::SETNSm, 0 },
306 { X86::SETPr, X86::SETPm, 0 },
307 { X86::SETSr, X86::SETSm, 0 },
308 { X86::TAILJMPr, X86::TAILJMPm, 1 },
309 { X86::TEST16ri, X86::TEST16mi, 1 },
310 { X86::TEST32ri, X86::TEST32mi, 1 },
311 { X86::TEST64ri32, X86::TEST64mi32, 1 },
312 { X86::TEST8ri, X86::TEST8mi, 1 },
313 { X86::XCHG16rr, X86::XCHG16mr, 0 },
314 { X86::XCHG32rr, X86::XCHG32mr, 0 },
315 { X86::XCHG64rr, X86::XCHG64mr, 0 },
316 { X86::XCHG8rr, X86::XCHG8mr, 0 }
319 for (unsigned i = 0, e = array_lengthof(OpTbl0); i != e; ++i) {
320 unsigned RegOp = OpTbl0[i][0];
321 unsigned MemOp = OpTbl0[i][1];
322 if (!RegOp2MemOpTable0.insert(std::make_pair((unsigned*)RegOp, MemOp)))
323 assert(false && "Duplicated entries?");
324 unsigned FoldedLoad = OpTbl0[i][2];
325 // Index 0, folded load or store.
326 unsigned AuxInfo = 0 | (FoldedLoad << 4) | ((FoldedLoad^1) << 5);
327 if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr)
328 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
329 std::make_pair(RegOp, AuxInfo))))
330 AmbEntries.push_back(MemOp);
333 static const unsigned OpTbl1[][2] = {
334 { X86::CMP16rr, X86::CMP16rm },
335 { X86::CMP32rr, X86::CMP32rm },
336 { X86::CMP64rr, X86::CMP64rm },
337 { X86::CMP8rr, X86::CMP8rm },
338 { X86::CVTSD2SSrr, X86::CVTSD2SSrm },
339 { X86::CVTSI2SD64rr, X86::CVTSI2SD64rm },
340 { X86::CVTSI2SDrr, X86::CVTSI2SDrm },
341 { X86::CVTSI2SS64rr, X86::CVTSI2SS64rm },
342 { X86::CVTSI2SSrr, X86::CVTSI2SSrm },
343 { X86::CVTSS2SDrr, X86::CVTSS2SDrm },
344 { X86::CVTTSD2SI64rr, X86::CVTTSD2SI64rm },
345 { X86::CVTTSD2SIrr, X86::CVTTSD2SIrm },
346 { X86::CVTTSS2SI64rr, X86::CVTTSS2SI64rm },
347 { X86::CVTTSS2SIrr, X86::CVTTSS2SIrm },
348 { X86::FsMOVAPDrr, X86::MOVSDrm },
349 { X86::FsMOVAPSrr, X86::MOVSSrm },
350 { X86::IMUL16rri, X86::IMUL16rmi },
351 { X86::IMUL16rri8, X86::IMUL16rmi8 },
352 { X86::IMUL32rri, X86::IMUL32rmi },
353 { X86::IMUL32rri8, X86::IMUL32rmi8 },
354 { X86::IMUL64rri32, X86::IMUL64rmi32 },
355 { X86::IMUL64rri8, X86::IMUL64rmi8 },
356 { X86::Int_CMPSDrr, X86::Int_CMPSDrm },
357 { X86::Int_CMPSSrr, X86::Int_CMPSSrm },
358 { X86::Int_COMISDrr, X86::Int_COMISDrm },
359 { X86::Int_COMISSrr, X86::Int_COMISSrm },
360 { X86::Int_CVTDQ2PDrr, X86::Int_CVTDQ2PDrm },
361 { X86::Int_CVTDQ2PSrr, X86::Int_CVTDQ2PSrm },
362 { X86::Int_CVTPD2DQrr, X86::Int_CVTPD2DQrm },
363 { X86::Int_CVTPD2PSrr, X86::Int_CVTPD2PSrm },
364 { X86::Int_CVTPS2DQrr, X86::Int_CVTPS2DQrm },
365 { X86::Int_CVTPS2PDrr, X86::Int_CVTPS2PDrm },
366 { X86::Int_CVTSD2SI64rr,X86::Int_CVTSD2SI64rm },
367 { X86::Int_CVTSD2SIrr, X86::Int_CVTSD2SIrm },
368 { X86::Int_CVTSD2SSrr, X86::Int_CVTSD2SSrm },
369 { X86::Int_CVTSI2SD64rr,X86::Int_CVTSI2SD64rm },
370 { X86::Int_CVTSI2SDrr, X86::Int_CVTSI2SDrm },
371 { X86::Int_CVTSI2SS64rr,X86::Int_CVTSI2SS64rm },
372 { X86::Int_CVTSI2SSrr, X86::Int_CVTSI2SSrm },
373 { X86::Int_CVTSS2SDrr, X86::Int_CVTSS2SDrm },
374 { X86::Int_CVTSS2SI64rr,X86::Int_CVTSS2SI64rm },
375 { X86::Int_CVTSS2SIrr, X86::Int_CVTSS2SIrm },
376 { X86::Int_CVTTPD2DQrr, X86::Int_CVTTPD2DQrm },
377 { X86::Int_CVTTPS2DQrr, X86::Int_CVTTPS2DQrm },
378 { X86::Int_CVTTSD2SI64rr,X86::Int_CVTTSD2SI64rm },
379 { X86::Int_CVTTSD2SIrr, X86::Int_CVTTSD2SIrm },
380 { X86::Int_CVTTSS2SI64rr,X86::Int_CVTTSS2SI64rm },
381 { X86::Int_CVTTSS2SIrr, X86::Int_CVTTSS2SIrm },
382 { X86::Int_UCOMISDrr, X86::Int_UCOMISDrm },
383 { X86::Int_UCOMISSrr, X86::Int_UCOMISSrm },
384 { X86::MOV16rr, X86::MOV16rm },
385 { X86::MOV16to16_, X86::MOV16_rm },
386 { X86::MOV32rr, X86::MOV32rm },
387 { X86::MOV32to32_, X86::MOV32_rm },
388 { X86::MOV64rr, X86::MOV64rm },
389 { X86::MOV64toPQIrr, X86::MOV64toPQIrm },
390 { X86::MOV64toSDrr, X86::MOV64toSDrm },
391 { X86::MOV8rr, X86::MOV8rm },
392 { X86::MOVAPDrr, X86::MOVAPDrm },
393 { X86::MOVAPSrr, X86::MOVAPSrm },
394 { X86::MOVDDUPrr, X86::MOVDDUPrm },
395 { X86::MOVDI2PDIrr, X86::MOVDI2PDIrm },
396 { X86::MOVDI2SSrr, X86::MOVDI2SSrm },
397 { X86::MOVSD2PDrr, X86::MOVSD2PDrm },
398 { X86::MOVSDrr, X86::MOVSDrm },
399 { X86::MOVSHDUPrr, X86::MOVSHDUPrm },
400 { X86::MOVSLDUPrr, X86::MOVSLDUPrm },
401 { X86::MOVSS2PSrr, X86::MOVSS2PSrm },
402 { X86::MOVSSrr, X86::MOVSSrm },
403 { X86::MOVSX16rr8, X86::MOVSX16rm8 },
404 { X86::MOVSX32rr16, X86::MOVSX32rm16 },
405 { X86::MOVSX32rr8, X86::MOVSX32rm8 },
406 { X86::MOVSX64rr16, X86::MOVSX64rm16 },
407 { X86::MOVSX64rr32, X86::MOVSX64rm32 },
408 { X86::MOVSX64rr8, X86::MOVSX64rm8 },
409 { X86::MOVUPDrr, X86::MOVUPDrm },
410 { X86::MOVUPSrr, X86::MOVUPSrm },
411 { X86::MOVZX16rr8, X86::MOVZX16rm8 },
412 { X86::MOVZX32rr16, X86::MOVZX32rm16 },
413 { X86::MOVZX32rr8, X86::MOVZX32rm8 },
414 { X86::MOVZX64rr16, X86::MOVZX64rm16 },
415 { X86::MOVZX64rr8, X86::MOVZX64rm8 },
416 { X86::PSHUFDri, X86::PSHUFDmi },
417 { X86::PSHUFHWri, X86::PSHUFHWmi },
418 { X86::PSHUFLWri, X86::PSHUFLWmi },
419 { X86::PsMOVZX64rr32, X86::PsMOVZX64rm32 },
420 { X86::RCPPSr, X86::RCPPSm },
421 { X86::RCPPSr_Int, X86::RCPPSm_Int },
422 { X86::RSQRTPSr, X86::RSQRTPSm },
423 { X86::RSQRTPSr_Int, X86::RSQRTPSm_Int },
424 { X86::RSQRTSSr, X86::RSQRTSSm },
425 { X86::RSQRTSSr_Int, X86::RSQRTSSm_Int },
426 { X86::SQRTPDr, X86::SQRTPDm },
427 { X86::SQRTPDr_Int, X86::SQRTPDm_Int },
428 { X86::SQRTPSr, X86::SQRTPSm },
429 { X86::SQRTPSr_Int, X86::SQRTPSm_Int },
430 { X86::SQRTSDr, X86::SQRTSDm },
431 { X86::SQRTSDr_Int, X86::SQRTSDm_Int },
432 { X86::SQRTSSr, X86::SQRTSSm },
433 { X86::SQRTSSr_Int, X86::SQRTSSm_Int },
434 { X86::TEST16rr, X86::TEST16rm },
435 { X86::TEST32rr, X86::TEST32rm },
436 { X86::TEST64rr, X86::TEST64rm },
437 { X86::TEST8rr, X86::TEST8rm },
438 // FIXME: TEST*rr EAX,EAX ---> CMP [mem], 0
439 { X86::UCOMISDrr, X86::UCOMISDrm },
440 { X86::UCOMISSrr, X86::UCOMISSrm },
441 { X86::XCHG16rr, X86::XCHG16rm },
442 { X86::XCHG32rr, X86::XCHG32rm },
443 { X86::XCHG64rr, X86::XCHG64rm },
444 { X86::XCHG8rr, X86::XCHG8rm }
447 for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {
448 unsigned RegOp = OpTbl1[i][0];
449 unsigned MemOp = OpTbl1[i][1];
450 if (!RegOp2MemOpTable1.insert(std::make_pair((unsigned*)RegOp, MemOp)))
451 assert(false && "Duplicated entries?");
452 unsigned AuxInfo = 1 | (1 << 4); // Index 1, folded load
453 if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr)
454 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
455 std::make_pair(RegOp, AuxInfo))))
456 AmbEntries.push_back(MemOp);
459 static const unsigned OpTbl2[][2] = {
460 { X86::ADC32rr, X86::ADC32rm },
461 { X86::ADC64rr, X86::ADC64rm },
462 { X86::ADD16rr, X86::ADD16rm },
463 { X86::ADD32rr, X86::ADD32rm },
464 { X86::ADD64rr, X86::ADD64rm },
465 { X86::ADD8rr, X86::ADD8rm },
466 { X86::ADDPDrr, X86::ADDPDrm },
467 { X86::ADDPSrr, X86::ADDPSrm },
468 { X86::ADDSDrr, X86::ADDSDrm },
469 { X86::ADDSSrr, X86::ADDSSrm },
470 { X86::ADDSUBPDrr, X86::ADDSUBPDrm },
471 { X86::ADDSUBPSrr, X86::ADDSUBPSrm },
472 { X86::AND16rr, X86::AND16rm },
473 { X86::AND32rr, X86::AND32rm },
474 { X86::AND64rr, X86::AND64rm },
475 { X86::AND8rr, X86::AND8rm },
476 { X86::ANDNPDrr, X86::ANDNPDrm },
477 { X86::ANDNPSrr, X86::ANDNPSrm },
478 { X86::ANDPDrr, X86::ANDPDrm },
479 { X86::ANDPSrr, X86::ANDPSrm },
480 { X86::CMOVA16rr, X86::CMOVA16rm },
481 { X86::CMOVA32rr, X86::CMOVA32rm },
482 { X86::CMOVA64rr, X86::CMOVA64rm },
483 { X86::CMOVAE16rr, X86::CMOVAE16rm },
484 { X86::CMOVAE32rr, X86::CMOVAE32rm },
485 { X86::CMOVAE64rr, X86::CMOVAE64rm },
486 { X86::CMOVB16rr, X86::CMOVB16rm },
487 { X86::CMOVB32rr, X86::CMOVB32rm },
488 { X86::CMOVB64rr, X86::CMOVB64rm },
489 { X86::CMOVBE16rr, X86::CMOVBE16rm },
490 { X86::CMOVBE32rr, X86::CMOVBE32rm },
491 { X86::CMOVBE64rr, X86::CMOVBE64rm },
492 { X86::CMOVE16rr, X86::CMOVE16rm },
493 { X86::CMOVE32rr, X86::CMOVE32rm },
494 { X86::CMOVE64rr, X86::CMOVE64rm },
495 { X86::CMOVG16rr, X86::CMOVG16rm },
496 { X86::CMOVG32rr, X86::CMOVG32rm },
497 { X86::CMOVG64rr, X86::CMOVG64rm },
498 { X86::CMOVGE16rr, X86::CMOVGE16rm },
499 { X86::CMOVGE32rr, X86::CMOVGE32rm },
500 { X86::CMOVGE64rr, X86::CMOVGE64rm },
501 { X86::CMOVL16rr, X86::CMOVL16rm },
502 { X86::CMOVL32rr, X86::CMOVL32rm },
503 { X86::CMOVL64rr, X86::CMOVL64rm },
504 { X86::CMOVLE16rr, X86::CMOVLE16rm },
505 { X86::CMOVLE32rr, X86::CMOVLE32rm },
506 { X86::CMOVLE64rr, X86::CMOVLE64rm },
507 { X86::CMOVNE16rr, X86::CMOVNE16rm },
508 { X86::CMOVNE32rr, X86::CMOVNE32rm },
509 { X86::CMOVNE64rr, X86::CMOVNE64rm },
510 { X86::CMOVNP16rr, X86::CMOVNP16rm },
511 { X86::CMOVNP32rr, X86::CMOVNP32rm },
512 { X86::CMOVNP64rr, X86::CMOVNP64rm },
513 { X86::CMOVNS16rr, X86::CMOVNS16rm },
514 { X86::CMOVNS32rr, X86::CMOVNS32rm },
515 { X86::CMOVNS64rr, X86::CMOVNS64rm },
516 { X86::CMOVP16rr, X86::CMOVP16rm },
517 { X86::CMOVP32rr, X86::CMOVP32rm },
518 { X86::CMOVP64rr, X86::CMOVP64rm },
519 { X86::CMOVS16rr, X86::CMOVS16rm },
520 { X86::CMOVS32rr, X86::CMOVS32rm },
521 { X86::CMOVS64rr, X86::CMOVS64rm },
522 { X86::CMPPDrri, X86::CMPPDrmi },
523 { X86::CMPPSrri, X86::CMPPSrmi },
524 { X86::CMPSDrr, X86::CMPSDrm },
525 { X86::CMPSSrr, X86::CMPSSrm },
526 { X86::DIVPDrr, X86::DIVPDrm },
527 { X86::DIVPSrr, X86::DIVPSrm },
528 { X86::DIVSDrr, X86::DIVSDrm },
529 { X86::DIVSSrr, X86::DIVSSrm },
530 { X86::HADDPDrr, X86::HADDPDrm },
531 { X86::HADDPSrr, X86::HADDPSrm },
532 { X86::HSUBPDrr, X86::HSUBPDrm },
533 { X86::HSUBPSrr, X86::HSUBPSrm },
534 { X86::IMUL16rr, X86::IMUL16rm },
535 { X86::IMUL32rr, X86::IMUL32rm },
536 { X86::IMUL64rr, X86::IMUL64rm },
537 { X86::MAXPDrr, X86::MAXPDrm },
538 { X86::MAXPDrr_Int, X86::MAXPDrm_Int },
539 { X86::MAXPSrr, X86::MAXPSrm },
540 { X86::MAXPSrr_Int, X86::MAXPSrm_Int },
541 { X86::MAXSDrr, X86::MAXSDrm },
542 { X86::MAXSDrr_Int, X86::MAXSDrm_Int },
543 { X86::MAXSSrr, X86::MAXSSrm },
544 { X86::MAXSSrr_Int, X86::MAXSSrm_Int },
545 { X86::MINPDrr, X86::MINPDrm },
546 { X86::MINPDrr_Int, X86::MINPDrm_Int },
547 { X86::MINPSrr, X86::MINPSrm },
548 { X86::MINPSrr_Int, X86::MINPSrm_Int },
549 { X86::MINSDrr, X86::MINSDrm },
550 { X86::MINSDrr_Int, X86::MINSDrm_Int },
551 { X86::MINSSrr, X86::MINSSrm },
552 { X86::MINSSrr_Int, X86::MINSSrm_Int },
553 { X86::MULPDrr, X86::MULPDrm },
554 { X86::MULPSrr, X86::MULPSrm },
555 { X86::MULSDrr, X86::MULSDrm },
556 { X86::MULSSrr, X86::MULSSrm },
557 { X86::OR16rr, X86::OR16rm },
558 { X86::OR32rr, X86::OR32rm },
559 { X86::OR64rr, X86::OR64rm },
560 { X86::OR8rr, X86::OR8rm },
561 { X86::ORPDrr, X86::ORPDrm },
562 { X86::ORPSrr, X86::ORPSrm },
563 { X86::PACKSSDWrr, X86::PACKSSDWrm },
564 { X86::PACKSSWBrr, X86::PACKSSWBrm },
565 { X86::PACKUSWBrr, X86::PACKUSWBrm },
566 { X86::PADDBrr, X86::PADDBrm },
567 { X86::PADDDrr, X86::PADDDrm },
568 { X86::PADDQrr, X86::PADDQrm },
569 { X86::PADDSBrr, X86::PADDSBrm },
570 { X86::PADDSWrr, X86::PADDSWrm },
571 { X86::PADDWrr, X86::PADDWrm },
572 { X86::PANDNrr, X86::PANDNrm },
573 { X86::PANDrr, X86::PANDrm },
574 { X86::PAVGBrr, X86::PAVGBrm },
575 { X86::PAVGWrr, X86::PAVGWrm },
576 { X86::PCMPEQBrr, X86::PCMPEQBrm },
577 { X86::PCMPEQDrr, X86::PCMPEQDrm },
578 { X86::PCMPEQWrr, X86::PCMPEQWrm },
579 { X86::PCMPGTBrr, X86::PCMPGTBrm },
580 { X86::PCMPGTDrr, X86::PCMPGTDrm },
581 { X86::PCMPGTWrr, X86::PCMPGTWrm },
582 { X86::PINSRWrri, X86::PINSRWrmi },
583 { X86::PMADDWDrr, X86::PMADDWDrm },
584 { X86::PMAXSWrr, X86::PMAXSWrm },
585 { X86::PMAXUBrr, X86::PMAXUBrm },
586 { X86::PMINSWrr, X86::PMINSWrm },
587 { X86::PMINUBrr, X86::PMINUBrm },
588 { X86::PMULHUWrr, X86::PMULHUWrm },
589 { X86::PMULHWrr, X86::PMULHWrm },
590 { X86::PMULLWrr, X86::PMULLWrm },
591 { X86::PMULUDQrr, X86::PMULUDQrm },
592 { X86::PORrr, X86::PORrm },
593 { X86::PSADBWrr, X86::PSADBWrm },
594 { X86::PSLLDrr, X86::PSLLDrm },
595 { X86::PSLLQrr, X86::PSLLQrm },
596 { X86::PSLLWrr, X86::PSLLWrm },
597 { X86::PSRADrr, X86::PSRADrm },
598 { X86::PSRAWrr, X86::PSRAWrm },
599 { X86::PSRLDrr, X86::PSRLDrm },
600 { X86::PSRLQrr, X86::PSRLQrm },
601 { X86::PSRLWrr, X86::PSRLWrm },
602 { X86::PSUBBrr, X86::PSUBBrm },
603 { X86::PSUBDrr, X86::PSUBDrm },
604 { X86::PSUBSBrr, X86::PSUBSBrm },
605 { X86::PSUBSWrr, X86::PSUBSWrm },
606 { X86::PSUBWrr, X86::PSUBWrm },
607 { X86::PUNPCKHBWrr, X86::PUNPCKHBWrm },
608 { X86::PUNPCKHDQrr, X86::PUNPCKHDQrm },
609 { X86::PUNPCKHQDQrr, X86::PUNPCKHQDQrm },
610 { X86::PUNPCKHWDrr, X86::PUNPCKHWDrm },
611 { X86::PUNPCKLBWrr, X86::PUNPCKLBWrm },
612 { X86::PUNPCKLDQrr, X86::PUNPCKLDQrm },
613 { X86::PUNPCKLQDQrr, X86::PUNPCKLQDQrm },
614 { X86::PUNPCKLWDrr, X86::PUNPCKLWDrm },
615 { X86::PXORrr, X86::PXORrm },
616 { X86::SBB32rr, X86::SBB32rm },
617 { X86::SBB64rr, X86::SBB64rm },
618 { X86::SHUFPDrri, X86::SHUFPDrmi },
619 { X86::SHUFPSrri, X86::SHUFPSrmi },
620 { X86::SUB16rr, X86::SUB16rm },
621 { X86::SUB32rr, X86::SUB32rm },
622 { X86::SUB64rr, X86::SUB64rm },
623 { X86::SUB8rr, X86::SUB8rm },
624 { X86::SUBPDrr, X86::SUBPDrm },
625 { X86::SUBPSrr, X86::SUBPSrm },
626 { X86::SUBSDrr, X86::SUBSDrm },
627 { X86::SUBSSrr, X86::SUBSSrm },
628 // FIXME: TEST*rr -> swapped operand of TEST*mr.
629 { X86::UNPCKHPDrr, X86::UNPCKHPDrm },
630 { X86::UNPCKHPSrr, X86::UNPCKHPSrm },
631 { X86::UNPCKLPDrr, X86::UNPCKLPDrm },
632 { X86::UNPCKLPSrr, X86::UNPCKLPSrm },
633 { X86::XOR16rr, X86::XOR16rm },
634 { X86::XOR32rr, X86::XOR32rm },
635 { X86::XOR64rr, X86::XOR64rm },
636 { X86::XOR8rr, X86::XOR8rm },
637 { X86::XORPDrr, X86::XORPDrm },
638 { X86::XORPSrr, X86::XORPSrm }
641 for (unsigned i = 0, e = array_lengthof(OpTbl2); i != e; ++i) {
642 unsigned RegOp = OpTbl2[i][0];
643 unsigned MemOp = OpTbl2[i][1];
644 if (!RegOp2MemOpTable2.insert(std::make_pair((unsigned*)RegOp, MemOp)))
645 assert(false && "Duplicated entries?");
646 unsigned AuxInfo = 2 | (1 << 4); // Index 1, folded load
647 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
648 std::make_pair(RegOp, AuxInfo))))
649 AmbEntries.push_back(MemOp);
652 // Remove ambiguous entries.
653 assert(AmbEntries.empty() && "Duplicated entries in unfolding maps?");
656 // getX86RegNum - This function maps LLVM register identifiers to their X86
657 // specific numbering, which is used in various places encoding instructions.
659 unsigned X86RegisterInfo::getX86RegNum(unsigned RegNo) {
661 case X86::RAX: case X86::EAX: case X86::AX: case X86::AL: return N86::EAX;
662 case X86::RCX: case X86::ECX: case X86::CX: case X86::CL: return N86::ECX;
663 case X86::RDX: case X86::EDX: case X86::DX: case X86::DL: return N86::EDX;
664 case X86::RBX: case X86::EBX: case X86::BX: case X86::BL: return N86::EBX;
665 case X86::RSP: case X86::ESP: case X86::SP: case X86::SPL: case X86::AH:
667 case X86::RBP: case X86::EBP: case X86::BP: case X86::BPL: case X86::CH:
669 case X86::RSI: case X86::ESI: case X86::SI: case X86::SIL: case X86::DH:
671 case X86::RDI: case X86::EDI: case X86::DI: case X86::DIL: case X86::BH:
674 case X86::R8: case X86::R8D: case X86::R8W: case X86::R8B:
676 case X86::R9: case X86::R9D: case X86::R9W: case X86::R9B:
678 case X86::R10: case X86::R10D: case X86::R10W: case X86::R10B:
680 case X86::R11: case X86::R11D: case X86::R11W: case X86::R11B:
682 case X86::R12: case X86::R12D: case X86::R12W: case X86::R12B:
684 case X86::R13: case X86::R13D: case X86::R13W: case X86::R13B:
686 case X86::R14: case X86::R14D: case X86::R14W: case X86::R14B:
688 case X86::R15: case X86::R15D: case X86::R15W: case X86::R15B:
691 case X86::ST0: case X86::ST1: case X86::ST2: case X86::ST3:
692 case X86::ST4: case X86::ST5: case X86::ST6: case X86::ST7:
693 return RegNo-X86::ST0;
695 case X86::XMM0: case X86::XMM1: case X86::XMM2: case X86::XMM3:
696 case X86::XMM4: case X86::XMM5: case X86::XMM6: case X86::XMM7:
697 return getDwarfRegNum(RegNo) - getDwarfRegNum(X86::XMM0);
698 case X86::XMM8: case X86::XMM9: case X86::XMM10: case X86::XMM11:
699 case X86::XMM12: case X86::XMM13: case X86::XMM14: case X86::XMM15:
700 return getDwarfRegNum(RegNo) - getDwarfRegNum(X86::XMM8);
703 assert(isVirtualRegister(RegNo) && "Unknown physical register!");
704 assert(0 && "Register allocator hasn't allocated reg correctly yet!");
709 bool X86RegisterInfo::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
710 MachineBasicBlock::iterator MI,
711 const std::vector<CalleeSavedInfo> &CSI) const {
715 MachineFunction &MF = *MBB.getParent();
716 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
717 X86FI->setCalleeSavedFrameSize(CSI.size() * SlotSize);
718 unsigned Opc = Is64Bit ? X86::PUSH64r : X86::PUSH32r;
719 for (unsigned i = CSI.size(); i != 0; --i) {
720 unsigned Reg = CSI[i-1].getReg();
721 // Add the callee-saved register as live-in. It's killed at the spill.
723 BuildMI(MBB, MI, TII.get(Opc)).addReg(Reg);
728 bool X86RegisterInfo::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
729 MachineBasicBlock::iterator MI,
730 const std::vector<CalleeSavedInfo> &CSI) const {
734 unsigned Opc = Is64Bit ? X86::POP64r : X86::POP32r;
735 for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
736 unsigned Reg = CSI[i].getReg();
737 BuildMI(MBB, MI, TII.get(Opc), Reg);
742 static const MachineInstrBuilder &X86InstrAddOperand(MachineInstrBuilder &MIB,
743 MachineOperand &MO) {
745 MIB = MIB.addReg(MO.getReg(), MO.isDef(), MO.isImplicit());
746 else if (MO.isImmediate())
747 MIB = MIB.addImm(MO.getImm());
748 else if (MO.isFrameIndex())
749 MIB = MIB.addFrameIndex(MO.getFrameIndex());
750 else if (MO.isGlobalAddress())
751 MIB = MIB.addGlobalAddress(MO.getGlobal(), MO.getOffset());
752 else if (MO.isConstantPoolIndex())
753 MIB = MIB.addConstantPoolIndex(MO.getConstantPoolIndex(), MO.getOffset());
754 else if (MO.isJumpTableIndex())
755 MIB = MIB.addJumpTableIndex(MO.getJumpTableIndex());
756 else if (MO.isExternalSymbol())
757 MIB = MIB.addExternalSymbol(MO.getSymbolName());
759 assert(0 && "Unknown operand for X86InstrAddOperand!");
764 static unsigned getStoreRegOpcode(const TargetRegisterClass *RC) {
766 if (RC == &X86::GR64RegClass) {
768 } else if (RC == &X86::GR32RegClass) {
770 } else if (RC == &X86::GR16RegClass) {
772 } else if (RC == &X86::GR8RegClass) {
774 } else if (RC == &X86::GR32_RegClass) {
776 } else if (RC == &X86::GR16_RegClass) {
778 } else if (RC == &X86::RFP80RegClass) {
779 Opc = X86::ST_FpP80m; // pops
780 } else if (RC == &X86::RFP64RegClass) {
782 } else if (RC == &X86::RFP32RegClass) {
784 } else if (RC == &X86::FR32RegClass) {
786 } else if (RC == &X86::FR64RegClass) {
788 } else if (RC == &X86::VR128RegClass) {
790 } else if (RC == &X86::VR64RegClass) {
791 Opc = X86::MMX_MOVQ64mr;
793 assert(0 && "Unknown regclass");
800 void X86RegisterInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
801 MachineBasicBlock::iterator MI,
802 unsigned SrcReg, int FrameIdx,
803 const TargetRegisterClass *RC) const {
804 unsigned Opc = getStoreRegOpcode(RC);
805 addFrameReference(BuildMI(MBB, MI, TII.get(Opc)), FrameIdx)
806 .addReg(SrcReg, false, false, true);
809 void X86RegisterInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
810 SmallVectorImpl<MachineOperand> &Addr,
811 const TargetRegisterClass *RC,
812 SmallVectorImpl<MachineInstr*> &NewMIs) const {
813 unsigned Opc = getStoreRegOpcode(RC);
814 MachineInstrBuilder MIB = BuildMI(TII.get(Opc));
815 for (unsigned i = 0, e = Addr.size(); i != e; ++i)
816 MIB = X86InstrAddOperand(MIB, Addr[i]);
817 MIB.addReg(SrcReg, false, false, true);
818 NewMIs.push_back(MIB);
821 static unsigned getLoadRegOpcode(const TargetRegisterClass *RC) {
823 if (RC == &X86::GR64RegClass) {
825 } else if (RC == &X86::GR32RegClass) {
827 } else if (RC == &X86::GR16RegClass) {
829 } else if (RC == &X86::GR8RegClass) {
831 } else if (RC == &X86::GR32_RegClass) {
833 } else if (RC == &X86::GR16_RegClass) {
835 } else if (RC == &X86::RFP80RegClass) {
837 } else if (RC == &X86::RFP64RegClass) {
839 } else if (RC == &X86::RFP32RegClass) {
841 } else if (RC == &X86::FR32RegClass) {
843 } else if (RC == &X86::FR64RegClass) {
845 } else if (RC == &X86::VR128RegClass) {
847 } else if (RC == &X86::VR64RegClass) {
848 Opc = X86::MMX_MOVQ64rm;
850 assert(0 && "Unknown regclass");
857 void X86RegisterInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
858 MachineBasicBlock::iterator MI,
859 unsigned DestReg, int FrameIdx,
860 const TargetRegisterClass *RC) const{
861 unsigned Opc = getLoadRegOpcode(RC);
862 addFrameReference(BuildMI(MBB, MI, TII.get(Opc), DestReg), FrameIdx);
865 void X86RegisterInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
866 SmallVectorImpl<MachineOperand> &Addr,
867 const TargetRegisterClass *RC,
868 SmallVectorImpl<MachineInstr*> &NewMIs) const {
869 unsigned Opc = getLoadRegOpcode(RC);
870 MachineInstrBuilder MIB = BuildMI(TII.get(Opc), DestReg);
871 for (unsigned i = 0, e = Addr.size(); i != e; ++i)
872 MIB = X86InstrAddOperand(MIB, Addr[i]);
873 NewMIs.push_back(MIB);
876 void X86RegisterInfo::copyRegToReg(MachineBasicBlock &MBB,
877 MachineBasicBlock::iterator MI,
878 unsigned DestReg, unsigned SrcReg,
879 const TargetRegisterClass *DestRC,
880 const TargetRegisterClass *SrcRC) const {
881 if (DestRC != SrcRC) {
882 // Moving EFLAGS to / from another register requires a push and a pop.
883 if (SrcRC == &X86::CCRRegClass) {
884 assert(SrcReg == X86::EFLAGS);
885 if (DestRC == &X86::GR64RegClass) {
886 BuildMI(MBB, MI, TII.get(X86::PUSHFQ));
887 BuildMI(MBB, MI, TII.get(X86::POP64r), DestReg);
889 } else if (DestRC == &X86::GR32RegClass) {
890 BuildMI(MBB, MI, TII.get(X86::PUSHFD));
891 BuildMI(MBB, MI, TII.get(X86::POP32r), DestReg);
894 } else if (DestRC == &X86::CCRRegClass) {
895 assert(DestReg == X86::EFLAGS);
896 if (SrcRC == &X86::GR64RegClass) {
897 BuildMI(MBB, MI, TII.get(X86::PUSH64r)).addReg(SrcReg);
898 BuildMI(MBB, MI, TII.get(X86::POPFQ));
900 } else if (SrcRC == &X86::GR32RegClass) {
901 BuildMI(MBB, MI, TII.get(X86::PUSH32r)).addReg(SrcReg);
902 BuildMI(MBB, MI, TII.get(X86::POPFD));
906 cerr << "Not yet supported!";
911 if (DestRC == &X86::GR64RegClass) {
913 } else if (DestRC == &X86::GR32RegClass) {
915 } else if (DestRC == &X86::GR16RegClass) {
917 } else if (DestRC == &X86::GR8RegClass) {
919 } else if (DestRC == &X86::GR32_RegClass) {
921 } else if (DestRC == &X86::GR16_RegClass) {
923 } else if (DestRC == &X86::RFP32RegClass) {
924 Opc = X86::MOV_Fp3232;
925 } else if (DestRC == &X86::RFP64RegClass || DestRC == &X86::RSTRegClass) {
926 Opc = X86::MOV_Fp6464;
927 } else if (DestRC == &X86::RFP80RegClass) {
928 Opc = X86::MOV_Fp8080;
929 } else if (DestRC == &X86::FR32RegClass) {
930 Opc = X86::FsMOVAPSrr;
931 } else if (DestRC == &X86::FR64RegClass) {
932 Opc = X86::FsMOVAPDrr;
933 } else if (DestRC == &X86::VR128RegClass) {
935 } else if (DestRC == &X86::VR64RegClass) {
936 Opc = X86::MMX_MOVQ64rr;
938 assert(0 && "Unknown regclass");
941 BuildMI(MBB, MI, TII.get(Opc), DestReg).addReg(SrcReg);
944 const TargetRegisterClass *
945 X86RegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
946 if (RC == &X86::CCRRegClass)
948 return &X86::GR64RegClass;
950 return &X86::GR32RegClass;
954 void X86RegisterInfo::reMaterialize(MachineBasicBlock &MBB,
955 MachineBasicBlock::iterator I,
957 const MachineInstr *Orig) const {
958 // MOV32r0 etc. are implemented with xor which clobbers condition code.
959 // Re-materialize them as movri instructions to avoid side effects.
960 switch (Orig->getOpcode()) {
962 BuildMI(MBB, I, TII.get(X86::MOV8ri), DestReg).addImm(0);
965 BuildMI(MBB, I, TII.get(X86::MOV16ri), DestReg).addImm(0);
968 BuildMI(MBB, I, TII.get(X86::MOV32ri), DestReg).addImm(0);
971 BuildMI(MBB, I, TII.get(X86::MOV64ri32), DestReg).addImm(0);
974 MachineInstr *MI = Orig->clone();
975 MI->getOperand(0).setReg(DestReg);
982 static MachineInstr *FuseTwoAddrInst(unsigned Opcode,
983 SmallVector<MachineOperand,4> &MOs,
984 MachineInstr *MI, const TargetInstrInfo &TII) {
985 // Create the base instruction with the memory operand as the first part.
986 MachineInstr *NewMI = new MachineInstr(TII.get(Opcode), true);
987 MachineInstrBuilder MIB(NewMI);
988 unsigned NumAddrOps = MOs.size();
989 for (unsigned i = 0; i != NumAddrOps; ++i)
990 MIB = X86InstrAddOperand(MIB, MOs[i]);
991 if (NumAddrOps < 4) // FrameIndex only
992 MIB.addImm(1).addReg(0).addImm(0);
994 // Loop over the rest of the ri operands, converting them over.
995 unsigned NumOps = TII.getNumOperands(MI->getOpcode())-2;
996 for (unsigned i = 0; i != NumOps; ++i) {
997 MachineOperand &MO = MI->getOperand(i+2);
998 MIB = X86InstrAddOperand(MIB, MO);
1000 for (unsigned i = NumOps+2, e = MI->getNumOperands(); i != e; ++i) {
1001 MachineOperand &MO = MI->getOperand(i);
1002 MIB = X86InstrAddOperand(MIB, MO);
1007 static MachineInstr *FuseInst(unsigned Opcode, unsigned OpNo,
1008 SmallVector<MachineOperand,4> &MOs,
1009 MachineInstr *MI, const TargetInstrInfo &TII) {
1010 MachineInstr *NewMI = new MachineInstr(TII.get(Opcode), true);
1011 MachineInstrBuilder MIB(NewMI);
1013 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
1014 MachineOperand &MO = MI->getOperand(i);
1016 assert(MO.isRegister() && "Expected to fold into reg operand!");
1017 unsigned NumAddrOps = MOs.size();
1018 for (unsigned i = 0; i != NumAddrOps; ++i)
1019 MIB = X86InstrAddOperand(MIB, MOs[i]);
1020 if (NumAddrOps < 4) // FrameIndex only
1021 MIB.addImm(1).addReg(0).addImm(0);
1023 MIB = X86InstrAddOperand(MIB, MO);
1029 static MachineInstr *MakeM0Inst(const TargetInstrInfo &TII, unsigned Opcode,
1030 SmallVector<MachineOperand,4> &MOs,
1032 MachineInstrBuilder MIB = BuildMI(TII.get(Opcode));
1034 unsigned NumAddrOps = MOs.size();
1035 for (unsigned i = 0; i != NumAddrOps; ++i)
1036 MIB = X86InstrAddOperand(MIB, MOs[i]);
1037 if (NumAddrOps < 4) // FrameIndex only
1038 MIB.addImm(1).addReg(0).addImm(0);
1039 return MIB.addImm(0);
1043 X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
1044 SmallVector<MachineOperand,4> &MOs) const {
1045 const DenseMap<unsigned*, unsigned> *OpcodeTablePtr = NULL;
1046 bool isTwoAddrFold = false;
1047 unsigned NumOps = TII.getNumOperands(MI->getOpcode());
1048 bool isTwoAddr = NumOps > 1 &&
1049 MI->getInstrDescriptor()->getOperandConstraint(1, TOI::TIED_TO) != -1;
1051 MachineInstr *NewMI = NULL;
1052 // Folding a memory location into the two-address part of a two-address
1053 // instruction is different than folding it other places. It requires
1054 // replacing the *two* registers with the memory location.
1055 if (isTwoAddr && NumOps >= 2 && i < 2 &&
1056 MI->getOperand(0).isRegister() &&
1057 MI->getOperand(1).isRegister() &&
1058 MI->getOperand(0).getReg() == MI->getOperand(1).getReg()) {
1059 OpcodeTablePtr = &RegOp2MemOpTable2Addr;
1060 isTwoAddrFold = true;
1061 } else if (i == 0) { // If operand 0
1062 if (MI->getOpcode() == X86::MOV16r0)
1063 NewMI = MakeM0Inst(TII, X86::MOV16mi, MOs, MI);
1064 else if (MI->getOpcode() == X86::MOV32r0)
1065 NewMI = MakeM0Inst(TII, X86::MOV32mi, MOs, MI);
1066 else if (MI->getOpcode() == X86::MOV64r0)
1067 NewMI = MakeM0Inst(TII, X86::MOV64mi32, MOs, MI);
1068 else if (MI->getOpcode() == X86::MOV8r0)
1069 NewMI = MakeM0Inst(TII, X86::MOV8mi, MOs, MI);
1071 NewMI->copyKillDeadInfo(MI);
1075 OpcodeTablePtr = &RegOp2MemOpTable0;
1076 } else if (i == 1) {
1077 OpcodeTablePtr = &RegOp2MemOpTable1;
1078 } else if (i == 2) {
1079 OpcodeTablePtr = &RegOp2MemOpTable2;
1082 // If table selected...
1083 if (OpcodeTablePtr) {
1084 // Find the Opcode to fuse
1085 DenseMap<unsigned*, unsigned>::iterator I =
1086 OpcodeTablePtr->find((unsigned*)MI->getOpcode());
1087 if (I != OpcodeTablePtr->end()) {
1089 NewMI = FuseTwoAddrInst(I->second, MOs, MI, TII);
1091 NewMI = FuseInst(I->second, i, MOs, MI, TII);
1092 NewMI->copyKillDeadInfo(MI);
1098 if (PrintFailedFusing)
1099 cerr << "We failed to fuse ("
1100 << ((i == 1) ? "r" : "s") << "): " << *MI;
1105 MachineInstr* X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned OpNum,
1106 int FrameIndex) const {
1107 // Check switch flag
1108 if (NoFusing) return NULL;
1109 SmallVector<MachineOperand,4> MOs;
1110 MOs.push_back(MachineOperand::CreateFrameIndex(FrameIndex));
1111 return foldMemoryOperand(MI, OpNum, MOs);
1114 MachineInstr* X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned OpNum,
1115 MachineInstr *LoadMI) const {
1116 // Check switch flag
1117 if (NoFusing) return NULL;
1118 SmallVector<MachineOperand,4> MOs;
1119 unsigned NumOps = TII.getNumOperands(LoadMI->getOpcode());
1120 for (unsigned i = NumOps - 4; i != NumOps; ++i)
1121 MOs.push_back(LoadMI->getOperand(i));
1122 return foldMemoryOperand(MI, OpNum, MOs);
1125 unsigned X86RegisterInfo::getOpcodeAfterMemoryFold(unsigned Opc,
1126 unsigned OpNum) const {
1127 // Check switch flag
1128 if (NoFusing) return 0;
1129 const DenseMap<unsigned*, unsigned> *OpcodeTablePtr = NULL;
1130 unsigned NumOps = TII.getNumOperands(Opc);
1131 bool isTwoAddr = NumOps > 1 &&
1132 TII.getOperandConstraint(Opc, 1, TOI::TIED_TO) != -1;
1134 // Folding a memory location into the two-address part of a two-address
1135 // instruction is different than folding it other places. It requires
1136 // replacing the *two* registers with the memory location.
1137 if (isTwoAddr && NumOps >= 2 && OpNum < 2) {
1138 OpcodeTablePtr = &RegOp2MemOpTable2Addr;
1139 } else if (OpNum == 0) { // If operand 0
1142 return X86::MOV16mi;
1144 return X86::MOV32mi;
1146 return X86::MOV64mi32;
1151 OpcodeTablePtr = &RegOp2MemOpTable0;
1152 } else if (OpNum == 1) {
1153 OpcodeTablePtr = &RegOp2MemOpTable1;
1154 } else if (OpNum == 2) {
1155 OpcodeTablePtr = &RegOp2MemOpTable2;
1158 if (OpcodeTablePtr) {
1159 // Find the Opcode to fuse
1160 DenseMap<unsigned*, unsigned>::iterator I =
1161 OpcodeTablePtr->find((unsigned*)Opc);
1162 if (I != OpcodeTablePtr->end())
1168 bool X86RegisterInfo::unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
1169 unsigned Reg, bool UnfoldLoad, bool UnfoldStore,
1170 SmallVectorImpl<MachineInstr*> &NewMIs) const {
1171 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1172 MemOp2RegOpTable.find((unsigned*)MI->getOpcode());
1173 if (I == MemOp2RegOpTable.end())
1175 unsigned Opc = I->second.first;
1176 unsigned Index = I->second.second & 0xf;
1177 bool FoldedLoad = I->second.second & (1 << 4);
1178 bool FoldedStore = I->second.second & (1 << 5);
1179 if (UnfoldLoad && !FoldedLoad)
1181 UnfoldLoad &= FoldedLoad;
1182 if (UnfoldStore && !FoldedStore)
1184 UnfoldStore &= FoldedStore;
1186 const TargetInstrDescriptor &TID = TII.get(Opc);
1187 const TargetOperandInfo &TOI = TID.OpInfo[Index];
1188 const TargetRegisterClass *RC = (TOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1189 ? TII.getPointerRegClass() : getRegClass(TOI.RegClass);
1190 SmallVector<MachineOperand,4> AddrOps;
1191 SmallVector<MachineOperand,2> BeforeOps;
1192 SmallVector<MachineOperand,2> AfterOps;
1193 SmallVector<MachineOperand,4> ImpOps;
1194 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
1195 MachineOperand &Op = MI->getOperand(i);
1196 if (i >= Index && i < Index+4)
1197 AddrOps.push_back(Op);
1198 else if (Op.isRegister() && Op.isImplicit())
1199 ImpOps.push_back(Op);
1201 BeforeOps.push_back(Op);
1203 AfterOps.push_back(Op);
1206 // Emit the load instruction.
1208 loadRegFromAddr(MF, Reg, AddrOps, RC, NewMIs);
1210 // Address operands cannot be marked isKill.
1211 for (unsigned i = 1; i != 5; ++i) {
1212 MachineOperand &MO = NewMIs[0]->getOperand(i);
1213 if (MO.isRegister())
1219 // Emit the data processing instruction.
1220 MachineInstr *DataMI = new MachineInstr(TID, true);
1221 MachineInstrBuilder MIB(DataMI);
1224 MIB.addReg(Reg, true);
1225 for (unsigned i = 0, e = BeforeOps.size(); i != e; ++i)
1226 MIB = X86InstrAddOperand(MIB, BeforeOps[i]);
1229 for (unsigned i = 0, e = AfterOps.size(); i != e; ++i)
1230 MIB = X86InstrAddOperand(MIB, AfterOps[i]);
1231 for (unsigned i = 0, e = ImpOps.size(); i != e; ++i) {
1232 MachineOperand &MO = ImpOps[i];
1233 MIB.addReg(MO.getReg(), MO.isDef(), true, MO.isKill(), MO.isDead());
1235 NewMIs.push_back(MIB);
1237 // Emit the store instruction.
1239 const TargetOperandInfo &DstTOI = TID.OpInfo[0];
1240 const TargetRegisterClass *DstRC = (DstTOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1241 ? TII.getPointerRegClass() : getRegClass(DstTOI.RegClass);
1242 storeRegToAddr(MF, Reg, AddrOps, DstRC, NewMIs);
1250 X86RegisterInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
1251 SmallVectorImpl<SDNode*> &NewNodes) const {
1252 if (!N->isTargetOpcode())
1255 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1256 MemOp2RegOpTable.find((unsigned*)N->getTargetOpcode());
1257 if (I == MemOp2RegOpTable.end())
1259 unsigned Opc = I->second.first;
1260 unsigned Index = I->second.second & 0xf;
1261 bool FoldedLoad = I->second.second & (1 << 4);
1262 bool FoldedStore = I->second.second & (1 << 5);
1263 const TargetInstrDescriptor &TID = TII.get(Opc);
1264 const TargetOperandInfo &TOI = TID.OpInfo[Index];
1265 const TargetRegisterClass *RC = (TOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1266 ? TII.getPointerRegClass() : getRegClass(TOI.RegClass);
1267 std::vector<SDOperand> AddrOps;
1268 std::vector<SDOperand> BeforeOps;
1269 std::vector<SDOperand> AfterOps;
1270 unsigned NumOps = N->getNumOperands();
1271 for (unsigned i = 0; i != NumOps-1; ++i) {
1272 SDOperand Op = N->getOperand(i);
1273 if (i >= Index && i < Index+4)
1274 AddrOps.push_back(Op);
1276 BeforeOps.push_back(Op);
1278 AfterOps.push_back(Op);
1280 SDOperand Chain = N->getOperand(NumOps-1);
1281 AddrOps.push_back(Chain);
1283 // Emit the load instruction.
1286 MVT::ValueType VT = *RC->vt_begin();
1287 Load = DAG.getTargetNode(getLoadRegOpcode(RC), VT, MVT::Other,
1288 &AddrOps[0], AddrOps.size());
1289 NewNodes.push_back(Load);
1292 // Emit the data processing instruction.
1293 std::vector<MVT::ValueType> VTs;
1294 const TargetRegisterClass *DstRC = 0;
1295 if (TID.numDefs > 0) {
1296 const TargetOperandInfo &DstTOI = TID.OpInfo[0];
1297 DstRC = (DstTOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1298 ? TII.getPointerRegClass() : getRegClass(DstTOI.RegClass);
1299 VTs.push_back(*DstRC->vt_begin());
1301 for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
1302 MVT::ValueType VT = N->getValueType(i);
1303 if (VT != MVT::Other && i >= TID.numDefs)
1307 BeforeOps.push_back(SDOperand(Load, 0));
1308 std::copy(AfterOps.begin(), AfterOps.end(), std::back_inserter(BeforeOps));
1309 SDNode *NewNode= DAG.getTargetNode(Opc, VTs, &BeforeOps[0], BeforeOps.size());
1310 NewNodes.push_back(NewNode);
1312 // Emit the store instruction.
1315 AddrOps.push_back(SDOperand(NewNode, 0));
1316 AddrOps.push_back(Chain);
1317 SDNode *Store = DAG.getTargetNode(getStoreRegOpcode(DstRC),
1318 MVT::Other, &AddrOps[0], AddrOps.size());
1319 NewNodes.push_back(Store);
1325 unsigned X86RegisterInfo::getOpcodeAfterMemoryUnfold(unsigned Opc,
1326 bool UnfoldLoad, bool UnfoldStore) const {
1327 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1328 MemOp2RegOpTable.find((unsigned*)Opc);
1329 if (I == MemOp2RegOpTable.end())
1331 bool FoldedLoad = I->second.second & (1 << 4);
1332 bool FoldedStore = I->second.second & (1 << 5);
1333 if (UnfoldLoad && !FoldedLoad)
1335 if (UnfoldStore && !FoldedStore)
1337 return I->second.first;
1341 X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
1342 static const unsigned CalleeSavedRegs32Bit[] = {
1343 X86::ESI, X86::EDI, X86::EBX, X86::EBP, 0
1346 static const unsigned CalleeSavedRegs32EHRet[] = {
1347 X86::EAX, X86::EDX, X86::ESI, X86::EDI, X86::EBX, X86::EBP, 0
1350 static const unsigned CalleeSavedRegs64Bit[] = {
1351 X86::RBX, X86::R12, X86::R13, X86::R14, X86::R15, X86::RBP, 0
1355 return CalleeSavedRegs64Bit;
1358 MachineFrameInfo *MFI = MF->getFrameInfo();
1359 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1360 if (MMI && MMI->callsEHReturn())
1361 return CalleeSavedRegs32EHRet;
1363 return CalleeSavedRegs32Bit;
1367 const TargetRegisterClass* const*
1368 X86RegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
1369 static const TargetRegisterClass * const CalleeSavedRegClasses32Bit[] = {
1370 &X86::GR32RegClass, &X86::GR32RegClass,
1371 &X86::GR32RegClass, &X86::GR32RegClass, 0
1373 static const TargetRegisterClass * const CalleeSavedRegClasses32EHRet[] = {
1374 &X86::GR32RegClass, &X86::GR32RegClass,
1375 &X86::GR32RegClass, &X86::GR32RegClass,
1376 &X86::GR32RegClass, &X86::GR32RegClass, 0
1378 static const TargetRegisterClass * const CalleeSavedRegClasses64Bit[] = {
1379 &X86::GR64RegClass, &X86::GR64RegClass,
1380 &X86::GR64RegClass, &X86::GR64RegClass,
1381 &X86::GR64RegClass, &X86::GR64RegClass, 0
1385 return CalleeSavedRegClasses64Bit;
1388 MachineFrameInfo *MFI = MF->getFrameInfo();
1389 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1390 if (MMI && MMI->callsEHReturn())
1391 return CalleeSavedRegClasses32EHRet;
1393 return CalleeSavedRegClasses32Bit;
1398 BitVector X86RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
1399 BitVector Reserved(getNumRegs());
1400 Reserved.set(X86::RSP);
1401 Reserved.set(X86::ESP);
1402 Reserved.set(X86::SP);
1403 Reserved.set(X86::SPL);
1405 Reserved.set(X86::RBP);
1406 Reserved.set(X86::EBP);
1407 Reserved.set(X86::BP);
1408 Reserved.set(X86::BPL);
1413 //===----------------------------------------------------------------------===//
1414 // Stack Frame Processing methods
1415 //===----------------------------------------------------------------------===//
1417 // hasFP - Return true if the specified function should have a dedicated frame
1418 // pointer register. This is true if the function has variable sized allocas or
1419 // if frame pointer elimination is disabled.
1421 bool X86RegisterInfo::hasFP(const MachineFunction &MF) const {
1422 MachineFrameInfo *MFI = MF.getFrameInfo();
1423 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1425 return (NoFramePointerElim ||
1426 MFI->hasVarSizedObjects() ||
1427 MF.getInfo<X86MachineFunctionInfo>()->getForceFramePointer() ||
1428 (MMI && MMI->callsUnwindInit()));
1431 bool X86RegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
1432 return !MF.getFrameInfo()->hasVarSizedObjects();
1435 void X86RegisterInfo::
1436 eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
1437 MachineBasicBlock::iterator I) const {
1438 if (!hasReservedCallFrame(MF)) {
1439 // If the stack pointer can be changed after prologue, turn the
1440 // adjcallstackup instruction into a 'sub ESP, <amt>' and the
1441 // adjcallstackdown instruction into 'add ESP, <amt>'
1442 // TODO: consider using push / pop instead of sub + store / add
1443 MachineInstr *Old = I;
1444 uint64_t Amount = Old->getOperand(0).getImm();
1446 // We need to keep the stack aligned properly. To do this, we round the
1447 // amount of space needed for the outgoing arguments up to the next
1448 // alignment boundary.
1449 unsigned Align = MF.getTarget().getFrameInfo()->getStackAlignment();
1450 Amount = (Amount+Align-1)/Align*Align;
1452 MachineInstr *New = 0;
1453 if (Old->getOpcode() == X86::ADJCALLSTACKDOWN) {
1454 New=BuildMI(TII.get(Is64Bit ? X86::SUB64ri32 : X86::SUB32ri), StackPtr)
1455 .addReg(StackPtr).addImm(Amount);
1457 assert(Old->getOpcode() == X86::ADJCALLSTACKUP);
1458 // factor out the amount the callee already popped.
1459 uint64_t CalleeAmt = Old->getOperand(1).getImm();
1460 Amount -= CalleeAmt;
1462 unsigned Opc = (Amount < 128) ?
1463 (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
1464 (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri);
1465 New = BuildMI(TII.get(Opc), StackPtr)
1466 .addReg(StackPtr).addImm(Amount);
1470 // Replace the pseudo instruction with a new instruction...
1471 if (New) MBB.insert(I, New);
1473 } else if (I->getOpcode() == X86::ADJCALLSTACKUP) {
1474 // If we are performing frame pointer elimination and if the callee pops
1475 // something off the stack pointer, add it back. We do this until we have
1476 // more advanced stack pointer tracking ability.
1477 if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
1478 unsigned Opc = (CalleeAmt < 128) ?
1479 (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
1480 (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri);
1482 BuildMI(TII.get(Opc), StackPtr).addReg(StackPtr).addImm(CalleeAmt);
1490 void X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
1491 int SPAdj, RegScavenger *RS) const{
1492 assert(SPAdj == 0 && "Unexpected");
1495 MachineInstr &MI = *II;
1496 MachineFunction &MF = *MI.getParent()->getParent();
1497 while (!MI.getOperand(i).isFrameIndex()) {
1499 assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
1502 int FrameIndex = MI.getOperand(i).getFrameIndex();
1503 // This must be part of a four operand memory reference. Replace the
1504 // FrameIndex with base register with EBP. Add an offset to the offset.
1505 MI.getOperand(i).ChangeToRegister(hasFP(MF) ? FramePtr : StackPtr, false);
1507 // Now add the frame object offset to the offset from EBP.
1508 int64_t Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
1509 MI.getOperand(i+3).getImm()+SlotSize;
1512 Offset += MF.getFrameInfo()->getStackSize();
1514 Offset += SlotSize; // Skip the saved EBP
1515 // Skip the RETADDR move area
1516 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1517 int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1518 if (TailCallReturnAddrDelta < 0) Offset -= TailCallReturnAddrDelta;
1521 MI.getOperand(i+3).ChangeToImmediate(Offset);
1525 X86RegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF) const{
1526 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1527 int32_t TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1528 if (TailCallReturnAddrDelta < 0) {
1529 // create RETURNADDR area
1539 CreateFixedObject(-TailCallReturnAddrDelta,
1540 (-1*SlotSize)+TailCallReturnAddrDelta);
1543 assert((TailCallReturnAddrDelta <= 0) &&
1544 "The Delta should always be zero or negative");
1545 // Create a frame entry for the EBP register that must be saved.
1546 int FrameIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize,
1548 TailCallReturnAddrDelta);
1549 assert(FrameIdx == MF.getFrameInfo()->getObjectIndexBegin() &&
1550 "Slot for EBP register must be last in order to be found!");
1554 /// emitSPUpdate - Emit a series of instructions to increment / decrement the
1555 /// stack pointer by a constant value.
1557 void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
1558 unsigned StackPtr, int64_t NumBytes, bool Is64Bit,
1559 const TargetInstrInfo &TII) {
1560 bool isSub = NumBytes < 0;
1561 uint64_t Offset = isSub ? -NumBytes : NumBytes;
1562 unsigned Opc = isSub
1564 (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
1565 (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri))
1567 (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
1568 (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri));
1569 uint64_t Chunk = (1LL << 31) - 1;
1572 uint64_t ThisVal = (Offset > Chunk) ? Chunk : Offset;
1573 BuildMI(MBB, MBBI, TII.get(Opc), StackPtr).addReg(StackPtr).addImm(ThisVal);
1578 // mergeSPUpdatesUp - Merge two stack-manipulating instructions upper iterator.
1580 void mergeSPUpdatesUp(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
1581 unsigned StackPtr, uint64_t *NumBytes = NULL) {
1582 if (MBBI == MBB.begin()) return;
1584 MachineBasicBlock::iterator PI = prior(MBBI);
1585 unsigned Opc = PI->getOpcode();
1586 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1587 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1588 PI->getOperand(0).getReg() == StackPtr) {
1590 *NumBytes += PI->getOperand(2).getImm();
1592 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1593 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1594 PI->getOperand(0).getReg() == StackPtr) {
1596 *NumBytes -= PI->getOperand(2).getImm();
1601 // mergeSPUpdatesUp - Merge two stack-manipulating instructions lower iterator.
1603 void mergeSPUpdatesDown(MachineBasicBlock &MBB,
1604 MachineBasicBlock::iterator &MBBI,
1605 unsigned StackPtr, uint64_t *NumBytes = NULL) {
1608 if (MBBI == MBB.end()) return;
1610 MachineBasicBlock::iterator NI = next(MBBI);
1611 if (NI == MBB.end()) return;
1613 unsigned Opc = NI->getOpcode();
1614 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1615 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1616 NI->getOperand(0).getReg() == StackPtr) {
1618 *NumBytes -= NI->getOperand(2).getImm();
1621 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1622 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1623 NI->getOperand(0).getReg() == StackPtr) {
1625 *NumBytes += NI->getOperand(2).getImm();
1631 /// mergeSPUpdates - Checks the instruction before/after the passed
1632 /// instruction. If it is an ADD/SUB instruction it is deleted
1633 /// argument and the stack adjustment is returned as a positive value for ADD
1634 /// and a negative for SUB.
1635 static int mergeSPUpdates(MachineBasicBlock &MBB,
1636 MachineBasicBlock::iterator &MBBI,
1638 bool doMergeWithPrevious) {
1640 if ((doMergeWithPrevious && MBBI == MBB.begin()) ||
1641 (!doMergeWithPrevious && MBBI == MBB.end()))
1646 MachineBasicBlock::iterator PI = doMergeWithPrevious ? prior(MBBI) : MBBI;
1647 MachineBasicBlock::iterator NI = doMergeWithPrevious ? 0 : next(MBBI);
1648 unsigned Opc = PI->getOpcode();
1649 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1650 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1651 PI->getOperand(0).getReg() == StackPtr){
1652 Offset += PI->getOperand(2).getImm();
1654 if (!doMergeWithPrevious) MBBI = NI;
1655 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1656 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1657 PI->getOperand(0).getReg() == StackPtr) {
1658 Offset -= PI->getOperand(2).getImm();
1660 if (!doMergeWithPrevious) MBBI = NI;
1666 void X86RegisterInfo::emitPrologue(MachineFunction &MF) const {
1667 MachineBasicBlock &MBB = MF.front(); // Prolog goes in entry BB
1668 MachineFrameInfo *MFI = MF.getFrameInfo();
1669 unsigned Align = MF.getTarget().getFrameInfo()->getStackAlignment();
1670 const Function* Fn = MF.getFunction();
1671 const X86Subtarget* Subtarget = &MF.getTarget().getSubtarget<X86Subtarget>();
1672 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1673 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1674 MachineBasicBlock::iterator MBBI = MBB.begin();
1676 // Prepare for frame info.
1677 unsigned FrameLabelId = 0;
1679 // Get the number of bytes to allocate from the FrameInfo.
1680 uint64_t StackSize = MFI->getStackSize();
1681 // Add RETADDR move area to callee saved frame size.
1682 int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1683 if (TailCallReturnAddrDelta < 0)
1684 X86FI->setCalleeSavedFrameSize(
1685 X86FI->getCalleeSavedFrameSize() +(-TailCallReturnAddrDelta));
1686 uint64_t NumBytes = StackSize - X86FI->getCalleeSavedFrameSize();
1688 // Insert stack pointer adjustment for later moving of return addr. Only
1689 // applies to tail call optimized functions where the callee argument stack
1690 // size is bigger than the callers.
1691 if (TailCallReturnAddrDelta < 0) {
1692 BuildMI(MBB, MBBI, TII.get(Is64Bit? X86::SUB64ri32 : X86::SUB32ri),
1693 StackPtr).addReg(StackPtr).addImm(-TailCallReturnAddrDelta);
1697 // Get the offset of the stack slot for the EBP register... which is
1698 // guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
1699 // Update the frame offset adjustment.
1700 MFI->setOffsetAdjustment(SlotSize-NumBytes);
1702 // Save EBP into the appropriate stack slot...
1703 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::PUSH64r : X86::PUSH32r))
1705 NumBytes -= SlotSize;
1707 if (MMI && MMI->needsFrameInfo()) {
1708 // Mark effective beginning of when frame pointer becomes valid.
1709 FrameLabelId = MMI->NextLabelID();
1710 BuildMI(MBB, MBBI, TII.get(X86::LABEL)).addImm(FrameLabelId);
1713 // Update EBP with the new base value...
1714 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr), FramePtr)
1718 unsigned ReadyLabelId = 0;
1719 if (MMI && MMI->needsFrameInfo()) {
1720 // Mark effective beginning of when frame pointer is ready.
1721 ReadyLabelId = MMI->NextLabelID();
1722 BuildMI(MBB, MBBI, TII.get(X86::LABEL)).addImm(ReadyLabelId);
1725 // Skip the callee-saved push instructions.
1726 while (MBBI != MBB.end() &&
1727 (MBBI->getOpcode() == X86::PUSH32r ||
1728 MBBI->getOpcode() == X86::PUSH64r))
1731 if (NumBytes) { // adjust stack pointer: ESP -= numbytes
1732 if (NumBytes >= 4096 && Subtarget->isTargetCygMing()) {
1733 // Check, whether EAX is livein for this function
1734 bool isEAXAlive = false;
1735 for (MachineFunction::livein_iterator II = MF.livein_begin(),
1736 EE = MF.livein_end(); (II != EE) && !isEAXAlive; ++II) {
1737 unsigned Reg = II->first;
1738 isEAXAlive = (Reg == X86::EAX || Reg == X86::AX ||
1739 Reg == X86::AH || Reg == X86::AL);
1742 // Function prologue calls _alloca to probe the stack when allocating
1743 // more than 4k bytes in one go. Touching the stack at 4K increments is
1744 // necessary to ensure that the guard pages used by the OS virtual memory
1745 // manager are allocated in correct sequence.
1747 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(NumBytes);
1748 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32))
1749 .addExternalSymbol("_alloca");
1752 BuildMI(MBB, MBBI, TII.get(X86::PUSH32r), X86::EAX);
1753 // Allocate NumBytes-4 bytes on stack. We'll also use 4 already
1754 // allocated bytes for EAX.
1755 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(NumBytes-4);
1756 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32))
1757 .addExternalSymbol("_alloca");
1759 MachineInstr *MI = addRegOffset(BuildMI(TII.get(X86::MOV32rm),X86::EAX),
1760 StackPtr, NumBytes-4);
1761 MBB.insert(MBBI, MI);
1764 // If there is an SUB32ri of ESP immediately before this instruction,
1765 // merge the two. This can be the case when tail call elimination is
1766 // enabled and the callee has more arguments then the caller.
1767 NumBytes -= mergeSPUpdates(MBB, MBBI, StackPtr, true);
1768 // If there is an ADD32ri or SUB32ri of ESP immediately after this
1769 // instruction, merge the two instructions.
1770 mergeSPUpdatesDown(MBB, MBBI, StackPtr, &NumBytes);
1773 emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, TII);
1777 if (MMI && MMI->needsFrameInfo()) {
1778 std::vector<MachineMove> &Moves = MMI->getFrameMoves();
1779 const TargetData *TD = MF.getTarget().getTargetData();
1781 // Calculate amount of bytes used for return address storing
1783 (MF.getTarget().getFrameInfo()->getStackGrowthDirection() ==
1784 TargetFrameInfo::StackGrowsUp ?
1785 TD->getPointerSize() : -TD->getPointerSize());
1788 // Show update of SP.
1791 MachineLocation SPDst(MachineLocation::VirtualFP);
1792 MachineLocation SPSrc(MachineLocation::VirtualFP, 2*stackGrowth);
1793 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1795 MachineLocation SPDst(MachineLocation::VirtualFP);
1796 MachineLocation SPSrc(MachineLocation::VirtualFP, -StackSize+stackGrowth);
1797 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1800 //FIXME: Verify & implement for FP
1801 MachineLocation SPDst(StackPtr);
1802 MachineLocation SPSrc(StackPtr, stackGrowth);
1803 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1806 // Add callee saved registers to move list.
1807 const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
1809 // FIXME: This is dirty hack. The code itself is pretty mess right now.
1810 // It should be rewritten from scratch and generalized sometimes.
1812 // Determine maximum offset (minumum due to stack growth)
1813 int64_t MaxOffset = 0;
1814 for (unsigned I = 0, E = CSI.size(); I!=E; ++I)
1815 MaxOffset = std::min(MaxOffset,
1816 MFI->getObjectOffset(CSI[I].getFrameIdx()));
1818 // Calculate offsets
1819 for (unsigned I = 0, E = CSI.size(); I!=E; ++I) {
1820 int64_t Offset = MFI->getObjectOffset(CSI[I].getFrameIdx());
1821 unsigned Reg = CSI[I].getReg();
1822 Offset = (MaxOffset-Offset+3*stackGrowth);
1823 MachineLocation CSDst(MachineLocation::VirtualFP, Offset);
1824 MachineLocation CSSrc(Reg);
1825 Moves.push_back(MachineMove(FrameLabelId, CSDst, CSSrc));
1830 MachineLocation FPDst(MachineLocation::VirtualFP, 2*stackGrowth);
1831 MachineLocation FPSrc(FramePtr);
1832 Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
1835 MachineLocation FPDst(hasFP(MF) ? FramePtr : StackPtr);
1836 MachineLocation FPSrc(MachineLocation::VirtualFP);
1837 Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
1840 // If it's main() on Cygwin\Mingw32 we should align stack as well
1841 if (Fn->hasExternalLinkage() && Fn->getName() == "main" &&
1842 Subtarget->isTargetCygMing()) {
1843 BuildMI(MBB, MBBI, TII.get(X86::AND32ri), X86::ESP)
1844 .addReg(X86::ESP).addImm(-Align);
1847 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(Align);
1848 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32)).addExternalSymbol("_alloca");
1852 void X86RegisterInfo::emitEpilogue(MachineFunction &MF,
1853 MachineBasicBlock &MBB) const {
1854 const MachineFrameInfo *MFI = MF.getFrameInfo();
1855 const Function* Fn = MF.getFunction();
1856 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1857 const X86Subtarget* Subtarget = &MF.getTarget().getSubtarget<X86Subtarget>();
1858 MachineBasicBlock::iterator MBBI = prior(MBB.end());
1859 unsigned RetOpcode = MBBI->getOpcode();
1861 switch (RetOpcode) {
1864 case X86::TCRETURNdi:
1865 case X86::TCRETURNri:
1866 case X86::TCRETURNri64:
1867 case X86::TCRETURNdi64:
1868 case X86::EH_RETURN:
1871 case X86::TAILJMPm: break; // These are ok
1873 assert(0 && "Can only insert epilog into returning blocks");
1876 // Get the number of bytes to allocate from the FrameInfo
1877 uint64_t StackSize = MFI->getStackSize();
1878 unsigned CSSize = X86FI->getCalleeSavedFrameSize();
1879 uint64_t NumBytes = StackSize - CSSize;
1883 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::POP64r : X86::POP32r), FramePtr);
1884 NumBytes -= SlotSize;
1887 // Skip the callee-saved pop instructions.
1888 while (MBBI != MBB.begin()) {
1889 MachineBasicBlock::iterator PI = prior(MBBI);
1890 unsigned Opc = PI->getOpcode();
1891 if (Opc != X86::POP32r && Opc != X86::POP64r && !TII.isTerminatorInstr(Opc))
1896 // If there is an ADD32ri or SUB32ri of ESP immediately before this
1897 // instruction, merge the two instructions.
1898 if (NumBytes || MFI->hasVarSizedObjects())
1899 mergeSPUpdatesUp(MBB, MBBI, StackPtr, &NumBytes);
1901 // If dynamic alloca is used, then reset esp to point to the last callee-saved
1902 // slot before popping them off! Also, if it's main() on Cygwin/Mingw32 we
1903 // aligned stack in the prologue, - revert stack changes back. Note: we're
1904 // assuming, that frame pointer was forced for main()
1905 if (MFI->hasVarSizedObjects() ||
1906 (Fn->hasExternalLinkage() && Fn->getName() == "main" &&
1907 Subtarget->isTargetCygMing())) {
1908 unsigned Opc = Is64Bit ? X86::LEA64r : X86::LEA32r;
1910 MachineInstr *MI = addRegOffset(BuildMI(TII.get(Opc), StackPtr),
1912 MBB.insert(MBBI, MI);
1914 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),StackPtr).
1920 // adjust stack pointer back: ESP += numbytes
1922 emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
1924 // We're returning from function via eh_return.
1925 if (RetOpcode == X86::EH_RETURN) {
1926 MBBI = prior(MBB.end());
1927 MachineOperand &DestAddr = MBBI->getOperand(0);
1928 assert(DestAddr.isRegister() && "Offset should be in register!");
1929 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),StackPtr).
1930 addReg(DestAddr.getReg());
1931 // Tail call return: adjust the stack pointer and jump to callee
1932 } else if (RetOpcode == X86::TCRETURNri || RetOpcode == X86::TCRETURNdi ||
1933 RetOpcode== X86::TCRETURNri64 || RetOpcode == X86::TCRETURNdi64) {
1934 MBBI = prior(MBB.end());
1935 MachineOperand &JumpTarget = MBBI->getOperand(0);
1936 MachineOperand &StackAdjust = MBBI->getOperand(1);
1937 assert( StackAdjust.isImmediate() && "Expecting immediate value.");
1939 // Adjust stack pointer.
1940 int StackAdj = StackAdjust.getImm();
1941 int MaxTCDelta = X86FI->getTCReturnAddrDelta();
1943 assert(MaxTCDelta <= 0 && "MaxTCDelta should never be positive");
1944 // Incoporate the retaddr area.
1945 Offset = StackAdj-MaxTCDelta;
1946 assert(Offset >= 0 && "Offset should never be negative");
1948 // Check for possible merge with preceeding ADD instruction.
1949 Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
1950 emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, TII);
1952 // Jump to label or value in register.
1953 if (RetOpcode == X86::TCRETURNdi|| RetOpcode == X86::TCRETURNdi64)
1954 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPd)).
1955 addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
1956 else if (RetOpcode== X86::TCRETURNri64) {
1957 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPr64), JumpTarget.getReg());
1959 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPr), JumpTarget.getReg());
1960 // Delete the pseudo instruction TCRETURN.
1962 } else if ((RetOpcode == X86::RET || RetOpcode == X86::RETI) &&
1963 (X86FI->getTCReturnAddrDelta() < 0)) {
1964 // Add the return addr area delta back since we are not tail calling.
1965 int delta = -1*X86FI->getTCReturnAddrDelta();
1966 MBBI = prior(MBB.end());
1967 // Check for possible merge with preceeding ADD instruction.
1968 delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
1969 emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, TII);
1973 unsigned X86RegisterInfo::getRARegister() const {
1975 return X86::RIP; // Should have dwarf #16
1977 return X86::EIP; // Should have dwarf #8
1980 unsigned X86RegisterInfo::getFrameRegister(MachineFunction &MF) const {
1981 return hasFP(MF) ? FramePtr : StackPtr;
1984 void X86RegisterInfo::getInitialFrameState(std::vector<MachineMove> &Moves)
1986 // Calculate amount of bytes used for return address storing
1987 int stackGrowth = (Is64Bit ? -8 : -4);
1989 // Initial state of the frame pointer is esp+4.
1990 MachineLocation Dst(MachineLocation::VirtualFP);
1991 MachineLocation Src(StackPtr, stackGrowth);
1992 Moves.push_back(MachineMove(0, Dst, Src));
1994 // Add return address to move list
1995 MachineLocation CSDst(StackPtr, stackGrowth);
1996 MachineLocation CSSrc(getRARegister());
1997 Moves.push_back(MachineMove(0, CSDst, CSSrc));
2000 unsigned X86RegisterInfo::getEHExceptionRegister() const {
2001 assert(0 && "What is the exception register");
2005 unsigned X86RegisterInfo::getEHHandlerRegister() const {
2006 assert(0 && "What is the exception handler register");
2011 unsigned getX86SubSuperRegister(unsigned Reg, MVT::ValueType VT, bool High) {
2013 default: return Reg;
2018 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2020 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2022 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2024 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2030 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2032 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2034 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2036 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2038 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2040 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2042 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2044 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2046 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2048 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2050 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2052 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2054 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2056 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2058 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2060 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2066 default: return Reg;
2067 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2069 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2071 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2073 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2075 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2077 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2079 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2081 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2083 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2085 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2087 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2089 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2091 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2093 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2095 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2097 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2102 default: return Reg;
2103 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2105 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2107 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2109 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2111 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2113 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2115 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2117 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2119 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2121 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2123 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2125 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2127 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2129 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2131 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2133 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2138 default: return Reg;
2139 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2141 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2143 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2145 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2147 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2149 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2151 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2153 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2155 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2157 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2159 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2161 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2163 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2165 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2167 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2169 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2178 #include "X86GenRegisterInfo.inc"