1 //===- X86RegisterInfo.cpp - X86 Register Information -----------*- 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 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/MachineModuleInfo.h"
30 #include "llvm/CodeGen/MachineRegisterInfo.h"
31 #include "llvm/Target/TargetAsmInfo.h"
32 #include "llvm/Target/TargetFrameInfo.h"
33 #include "llvm/Target/TargetInstrInfo.h"
34 #include "llvm/Target/TargetMachine.h"
35 #include "llvm/Target/TargetOptions.h"
36 #include "llvm/Support/CommandLine.h"
37 #include "llvm/ADT/BitVector.h"
38 #include "llvm/ADT/STLExtras.h"
43 NoFusing("disable-spill-fusing",
44 cl::desc("Disable fusing of spill code into instructions"));
46 PrintFailedFusing("print-failed-fuse-candidates",
47 cl::desc("Print instructions that the allocator wants to"
48 " fuse, but the X86 backend currently can't"),
52 X86RegisterInfo::X86RegisterInfo(X86TargetMachine &tm,
53 const TargetInstrInfo &tii)
54 : X86GenRegisterInfo(X86::ADJCALLSTACKDOWN, X86::ADJCALLSTACKUP),
56 // Cache some information.
57 const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
58 Is64Bit = Subtarget->is64Bit();
59 StackAlign = TM.getFrameInfo()->getStackAlignment();
70 SmallVector<unsigned,16> AmbEntries;
71 static const unsigned OpTbl2Addr[][2] = {
72 { X86::ADC32ri, X86::ADC32mi },
73 { X86::ADC32ri8, X86::ADC32mi8 },
74 { X86::ADC32rr, X86::ADC32mr },
75 { X86::ADC64ri32, X86::ADC64mi32 },
76 { X86::ADC64ri8, X86::ADC64mi8 },
77 { X86::ADC64rr, X86::ADC64mr },
78 { X86::ADD16ri, X86::ADD16mi },
79 { X86::ADD16ri8, X86::ADD16mi8 },
80 { X86::ADD16rr, X86::ADD16mr },
81 { X86::ADD32ri, X86::ADD32mi },
82 { X86::ADD32ri8, X86::ADD32mi8 },
83 { X86::ADD32rr, X86::ADD32mr },
84 { X86::ADD64ri32, X86::ADD64mi32 },
85 { X86::ADD64ri8, X86::ADD64mi8 },
86 { X86::ADD64rr, X86::ADD64mr },
87 { X86::ADD8ri, X86::ADD8mi },
88 { X86::ADD8rr, X86::ADD8mr },
89 { X86::AND16ri, X86::AND16mi },
90 { X86::AND16ri8, X86::AND16mi8 },
91 { X86::AND16rr, X86::AND16mr },
92 { X86::AND32ri, X86::AND32mi },
93 { X86::AND32ri8, X86::AND32mi8 },
94 { X86::AND32rr, X86::AND32mr },
95 { X86::AND64ri32, X86::AND64mi32 },
96 { X86::AND64ri8, X86::AND64mi8 },
97 { X86::AND64rr, X86::AND64mr },
98 { X86::AND8ri, X86::AND8mi },
99 { X86::AND8rr, X86::AND8mr },
100 { X86::DEC16r, X86::DEC16m },
101 { X86::DEC32r, X86::DEC32m },
102 { X86::DEC64_16r, X86::DEC64_16m },
103 { X86::DEC64_32r, X86::DEC64_32m },
104 { X86::DEC64r, X86::DEC64m },
105 { X86::DEC8r, X86::DEC8m },
106 { X86::INC16r, X86::INC16m },
107 { X86::INC32r, X86::INC32m },
108 { X86::INC64_16r, X86::INC64_16m },
109 { X86::INC64_32r, X86::INC64_32m },
110 { X86::INC64r, X86::INC64m },
111 { X86::INC8r, X86::INC8m },
112 { X86::NEG16r, X86::NEG16m },
113 { X86::NEG32r, X86::NEG32m },
114 { X86::NEG64r, X86::NEG64m },
115 { X86::NEG8r, X86::NEG8m },
116 { X86::NOT16r, X86::NOT16m },
117 { X86::NOT32r, X86::NOT32m },
118 { X86::NOT64r, X86::NOT64m },
119 { X86::NOT8r, X86::NOT8m },
120 { X86::OR16ri, X86::OR16mi },
121 { X86::OR16ri8, X86::OR16mi8 },
122 { X86::OR16rr, X86::OR16mr },
123 { X86::OR32ri, X86::OR32mi },
124 { X86::OR32ri8, X86::OR32mi8 },
125 { X86::OR32rr, X86::OR32mr },
126 { X86::OR64ri32, X86::OR64mi32 },
127 { X86::OR64ri8, X86::OR64mi8 },
128 { X86::OR64rr, X86::OR64mr },
129 { X86::OR8ri, X86::OR8mi },
130 { X86::OR8rr, X86::OR8mr },
131 { X86::ROL16r1, X86::ROL16m1 },
132 { X86::ROL16rCL, X86::ROL16mCL },
133 { X86::ROL16ri, X86::ROL16mi },
134 { X86::ROL32r1, X86::ROL32m1 },
135 { X86::ROL32rCL, X86::ROL32mCL },
136 { X86::ROL32ri, X86::ROL32mi },
137 { X86::ROL64r1, X86::ROL64m1 },
138 { X86::ROL64rCL, X86::ROL64mCL },
139 { X86::ROL64ri, X86::ROL64mi },
140 { X86::ROL8r1, X86::ROL8m1 },
141 { X86::ROL8rCL, X86::ROL8mCL },
142 { X86::ROL8ri, X86::ROL8mi },
143 { X86::ROR16r1, X86::ROR16m1 },
144 { X86::ROR16rCL, X86::ROR16mCL },
145 { X86::ROR16ri, X86::ROR16mi },
146 { X86::ROR32r1, X86::ROR32m1 },
147 { X86::ROR32rCL, X86::ROR32mCL },
148 { X86::ROR32ri, X86::ROR32mi },
149 { X86::ROR64r1, X86::ROR64m1 },
150 { X86::ROR64rCL, X86::ROR64mCL },
151 { X86::ROR64ri, X86::ROR64mi },
152 { X86::ROR8r1, X86::ROR8m1 },
153 { X86::ROR8rCL, X86::ROR8mCL },
154 { X86::ROR8ri, X86::ROR8mi },
155 { X86::SAR16r1, X86::SAR16m1 },
156 { X86::SAR16rCL, X86::SAR16mCL },
157 { X86::SAR16ri, X86::SAR16mi },
158 { X86::SAR32r1, X86::SAR32m1 },
159 { X86::SAR32rCL, X86::SAR32mCL },
160 { X86::SAR32ri, X86::SAR32mi },
161 { X86::SAR64r1, X86::SAR64m1 },
162 { X86::SAR64rCL, X86::SAR64mCL },
163 { X86::SAR64ri, X86::SAR64mi },
164 { X86::SAR8r1, X86::SAR8m1 },
165 { X86::SAR8rCL, X86::SAR8mCL },
166 { X86::SAR8ri, X86::SAR8mi },
167 { X86::SBB32ri, X86::SBB32mi },
168 { X86::SBB32ri8, X86::SBB32mi8 },
169 { X86::SBB32rr, X86::SBB32mr },
170 { X86::SBB64ri32, X86::SBB64mi32 },
171 { X86::SBB64ri8, X86::SBB64mi8 },
172 { X86::SBB64rr, X86::SBB64mr },
173 { X86::SHL16r1, X86::SHL16m1 },
174 { X86::SHL16rCL, X86::SHL16mCL },
175 { X86::SHL16ri, X86::SHL16mi },
176 { X86::SHL32r1, X86::SHL32m1 },
177 { X86::SHL32rCL, X86::SHL32mCL },
178 { X86::SHL32ri, X86::SHL32mi },
179 { X86::SHL64r1, X86::SHL64m1 },
180 { X86::SHL64rCL, X86::SHL64mCL },
181 { X86::SHL64ri, X86::SHL64mi },
182 { X86::SHL8r1, X86::SHL8m1 },
183 { X86::SHL8rCL, X86::SHL8mCL },
184 { X86::SHL8ri, X86::SHL8mi },
185 { X86::SHLD16rrCL, X86::SHLD16mrCL },
186 { X86::SHLD16rri8, X86::SHLD16mri8 },
187 { X86::SHLD32rrCL, X86::SHLD32mrCL },
188 { X86::SHLD32rri8, X86::SHLD32mri8 },
189 { X86::SHLD64rrCL, X86::SHLD64mrCL },
190 { X86::SHLD64rri8, X86::SHLD64mri8 },
191 { X86::SHR16r1, X86::SHR16m1 },
192 { X86::SHR16rCL, X86::SHR16mCL },
193 { X86::SHR16ri, X86::SHR16mi },
194 { X86::SHR32r1, X86::SHR32m1 },
195 { X86::SHR32rCL, X86::SHR32mCL },
196 { X86::SHR32ri, X86::SHR32mi },
197 { X86::SHR64r1, X86::SHR64m1 },
198 { X86::SHR64rCL, X86::SHR64mCL },
199 { X86::SHR64ri, X86::SHR64mi },
200 { X86::SHR8r1, X86::SHR8m1 },
201 { X86::SHR8rCL, X86::SHR8mCL },
202 { X86::SHR8ri, X86::SHR8mi },
203 { X86::SHRD16rrCL, X86::SHRD16mrCL },
204 { X86::SHRD16rri8, X86::SHRD16mri8 },
205 { X86::SHRD32rrCL, X86::SHRD32mrCL },
206 { X86::SHRD32rri8, X86::SHRD32mri8 },
207 { X86::SHRD64rrCL, X86::SHRD64mrCL },
208 { X86::SHRD64rri8, X86::SHRD64mri8 },
209 { X86::SUB16ri, X86::SUB16mi },
210 { X86::SUB16ri8, X86::SUB16mi8 },
211 { X86::SUB16rr, X86::SUB16mr },
212 { X86::SUB32ri, X86::SUB32mi },
213 { X86::SUB32ri8, X86::SUB32mi8 },
214 { X86::SUB32rr, X86::SUB32mr },
215 { X86::SUB64ri32, X86::SUB64mi32 },
216 { X86::SUB64ri8, X86::SUB64mi8 },
217 { X86::SUB64rr, X86::SUB64mr },
218 { X86::SUB8ri, X86::SUB8mi },
219 { X86::SUB8rr, X86::SUB8mr },
220 { X86::XOR16ri, X86::XOR16mi },
221 { X86::XOR16ri8, X86::XOR16mi8 },
222 { X86::XOR16rr, X86::XOR16mr },
223 { X86::XOR32ri, X86::XOR32mi },
224 { X86::XOR32ri8, X86::XOR32mi8 },
225 { X86::XOR32rr, X86::XOR32mr },
226 { X86::XOR64ri32, X86::XOR64mi32 },
227 { X86::XOR64ri8, X86::XOR64mi8 },
228 { X86::XOR64rr, X86::XOR64mr },
229 { X86::XOR8ri, X86::XOR8mi },
230 { X86::XOR8rr, X86::XOR8mr }
233 for (unsigned i = 0, e = array_lengthof(OpTbl2Addr); i != e; ++i) {
234 unsigned RegOp = OpTbl2Addr[i][0];
235 unsigned MemOp = OpTbl2Addr[i][1];
236 if (!RegOp2MemOpTable2Addr.insert(std::make_pair((unsigned*)RegOp, MemOp)))
237 assert(false && "Duplicated entries?");
238 unsigned AuxInfo = 0 | (1 << 4) | (1 << 5); // Index 0,folded load and store
239 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
240 std::make_pair(RegOp, AuxInfo))))
241 AmbEntries.push_back(MemOp);
244 // If the third value is 1, then it's folding either a load or a store.
245 static const unsigned OpTbl0[][3] = {
246 { X86::CALL32r, X86::CALL32m, 1 },
247 { X86::CALL64r, X86::CALL64m, 1 },
248 { X86::CMP16ri, X86::CMP16mi, 1 },
249 { X86::CMP16ri8, X86::CMP16mi8, 1 },
250 { X86::CMP32ri, X86::CMP32mi, 1 },
251 { X86::CMP32ri8, X86::CMP32mi8, 1 },
252 { X86::CMP64ri32, X86::CMP64mi32, 1 },
253 { X86::CMP64ri8, X86::CMP64mi8, 1 },
254 { X86::CMP8ri, X86::CMP8mi, 1 },
255 { X86::DIV16r, X86::DIV16m, 1 },
256 { X86::DIV32r, X86::DIV32m, 1 },
257 { X86::DIV64r, X86::DIV64m, 1 },
258 { X86::DIV8r, X86::DIV8m, 1 },
259 { X86::FsMOVAPDrr, X86::MOVSDmr, 0 },
260 { X86::FsMOVAPSrr, X86::MOVSSmr, 0 },
261 { X86::IDIV16r, X86::IDIV16m, 1 },
262 { X86::IDIV32r, X86::IDIV32m, 1 },
263 { X86::IDIV64r, X86::IDIV64m, 1 },
264 { X86::IDIV8r, X86::IDIV8m, 1 },
265 { X86::IMUL16r, X86::IMUL16m, 1 },
266 { X86::IMUL32r, X86::IMUL32m, 1 },
267 { X86::IMUL64r, X86::IMUL64m, 1 },
268 { X86::IMUL8r, X86::IMUL8m, 1 },
269 { X86::JMP32r, X86::JMP32m, 1 },
270 { X86::JMP64r, X86::JMP64m, 1 },
271 { X86::MOV16ri, X86::MOV16mi, 0 },
272 { X86::MOV16rr, X86::MOV16mr, 0 },
273 { X86::MOV16to16_, X86::MOV16_mr, 0 },
274 { X86::MOV32ri, X86::MOV32mi, 0 },
275 { X86::MOV32rr, X86::MOV32mr, 0 },
276 { X86::MOV32to32_, X86::MOV32_mr, 0 },
277 { X86::MOV64ri32, X86::MOV64mi32, 0 },
278 { X86::MOV64rr, X86::MOV64mr, 0 },
279 { X86::MOV8ri, X86::MOV8mi, 0 },
280 { X86::MOV8rr, X86::MOV8mr, 0 },
281 { X86::MOVAPDrr, X86::MOVAPDmr, 0 },
282 { X86::MOVAPSrr, X86::MOVAPSmr, 0 },
283 { X86::MOVPDI2DIrr, X86::MOVPDI2DImr, 0 },
284 { X86::MOVPQIto64rr,X86::MOVPQI2QImr, 0 },
285 { X86::MOVPS2SSrr, X86::MOVPS2SSmr, 0 },
286 { X86::MOVSDrr, X86::MOVSDmr, 0 },
287 { X86::MOVSDto64rr, X86::MOVSDto64mr, 0 },
288 { X86::MOVSS2DIrr, X86::MOVSS2DImr, 0 },
289 { X86::MOVSSrr, X86::MOVSSmr, 0 },
290 { X86::MOVUPDrr, X86::MOVUPDmr, 0 },
291 { X86::MOVUPSrr, X86::MOVUPSmr, 0 },
292 { X86::MUL16r, X86::MUL16m, 1 },
293 { X86::MUL32r, X86::MUL32m, 1 },
294 { X86::MUL64r, X86::MUL64m, 1 },
295 { X86::MUL8r, X86::MUL8m, 1 },
296 { X86::SETAEr, X86::SETAEm, 0 },
297 { X86::SETAr, X86::SETAm, 0 },
298 { X86::SETBEr, X86::SETBEm, 0 },
299 { X86::SETBr, X86::SETBm, 0 },
300 { X86::SETEr, X86::SETEm, 0 },
301 { X86::SETGEr, X86::SETGEm, 0 },
302 { X86::SETGr, X86::SETGm, 0 },
303 { X86::SETLEr, X86::SETLEm, 0 },
304 { X86::SETLr, X86::SETLm, 0 },
305 { X86::SETNEr, X86::SETNEm, 0 },
306 { X86::SETNPr, X86::SETNPm, 0 },
307 { X86::SETNSr, X86::SETNSm, 0 },
308 { X86::SETPr, X86::SETPm, 0 },
309 { X86::SETSr, X86::SETSm, 0 },
310 { X86::TAILJMPr, X86::TAILJMPm, 1 },
311 { X86::TEST16ri, X86::TEST16mi, 1 },
312 { X86::TEST32ri, X86::TEST32mi, 1 },
313 { X86::TEST64ri32, X86::TEST64mi32, 1 },
314 { X86::TEST8ri, X86::TEST8mi, 1 },
315 { X86::XCHG16rr, X86::XCHG16mr, 0 },
316 { X86::XCHG32rr, X86::XCHG32mr, 0 },
317 { X86::XCHG64rr, X86::XCHG64mr, 0 },
318 { X86::XCHG8rr, X86::XCHG8mr, 0 }
321 for (unsigned i = 0, e = array_lengthof(OpTbl0); i != e; ++i) {
322 unsigned RegOp = OpTbl0[i][0];
323 unsigned MemOp = OpTbl0[i][1];
324 if (!RegOp2MemOpTable0.insert(std::make_pair((unsigned*)RegOp, MemOp)))
325 assert(false && "Duplicated entries?");
326 unsigned FoldedLoad = OpTbl0[i][2];
327 // Index 0, folded load or store.
328 unsigned AuxInfo = 0 | (FoldedLoad << 4) | ((FoldedLoad^1) << 5);
329 if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr)
330 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
331 std::make_pair(RegOp, AuxInfo))))
332 AmbEntries.push_back(MemOp);
335 static const unsigned OpTbl1[][2] = {
336 { X86::CMP16rr, X86::CMP16rm },
337 { X86::CMP32rr, X86::CMP32rm },
338 { X86::CMP64rr, X86::CMP64rm },
339 { X86::CMP8rr, X86::CMP8rm },
340 { X86::CVTSD2SSrr, X86::CVTSD2SSrm },
341 { X86::CVTSI2SD64rr, X86::CVTSI2SD64rm },
342 { X86::CVTSI2SDrr, X86::CVTSI2SDrm },
343 { X86::CVTSI2SS64rr, X86::CVTSI2SS64rm },
344 { X86::CVTSI2SSrr, X86::CVTSI2SSrm },
345 { X86::CVTSS2SDrr, X86::CVTSS2SDrm },
346 { X86::CVTTSD2SI64rr, X86::CVTTSD2SI64rm },
347 { X86::CVTTSD2SIrr, X86::CVTTSD2SIrm },
348 { X86::CVTTSS2SI64rr, X86::CVTTSS2SI64rm },
349 { X86::CVTTSS2SIrr, X86::CVTTSS2SIrm },
350 { X86::FsMOVAPDrr, X86::MOVSDrm },
351 { X86::FsMOVAPSrr, X86::MOVSSrm },
352 { X86::IMUL16rri, X86::IMUL16rmi },
353 { X86::IMUL16rri8, X86::IMUL16rmi8 },
354 { X86::IMUL32rri, X86::IMUL32rmi },
355 { X86::IMUL32rri8, X86::IMUL32rmi8 },
356 { X86::IMUL64rri32, X86::IMUL64rmi32 },
357 { X86::IMUL64rri8, X86::IMUL64rmi8 },
358 { X86::Int_CMPSDrr, X86::Int_CMPSDrm },
359 { X86::Int_CMPSSrr, X86::Int_CMPSSrm },
360 { X86::Int_COMISDrr, X86::Int_COMISDrm },
361 { X86::Int_COMISSrr, X86::Int_COMISSrm },
362 { X86::Int_CVTDQ2PDrr, X86::Int_CVTDQ2PDrm },
363 { X86::Int_CVTDQ2PSrr, X86::Int_CVTDQ2PSrm },
364 { X86::Int_CVTPD2DQrr, X86::Int_CVTPD2DQrm },
365 { X86::Int_CVTPD2PSrr, X86::Int_CVTPD2PSrm },
366 { X86::Int_CVTPS2DQrr, X86::Int_CVTPS2DQrm },
367 { X86::Int_CVTPS2PDrr, X86::Int_CVTPS2PDrm },
368 { X86::Int_CVTSD2SI64rr,X86::Int_CVTSD2SI64rm },
369 { X86::Int_CVTSD2SIrr, X86::Int_CVTSD2SIrm },
370 { X86::Int_CVTSD2SSrr, X86::Int_CVTSD2SSrm },
371 { X86::Int_CVTSI2SD64rr,X86::Int_CVTSI2SD64rm },
372 { X86::Int_CVTSI2SDrr, X86::Int_CVTSI2SDrm },
373 { X86::Int_CVTSI2SS64rr,X86::Int_CVTSI2SS64rm },
374 { X86::Int_CVTSI2SSrr, X86::Int_CVTSI2SSrm },
375 { X86::Int_CVTSS2SDrr, X86::Int_CVTSS2SDrm },
376 { X86::Int_CVTSS2SI64rr,X86::Int_CVTSS2SI64rm },
377 { X86::Int_CVTSS2SIrr, X86::Int_CVTSS2SIrm },
378 { X86::Int_CVTTPD2DQrr, X86::Int_CVTTPD2DQrm },
379 { X86::Int_CVTTPS2DQrr, X86::Int_CVTTPS2DQrm },
380 { X86::Int_CVTTSD2SI64rr,X86::Int_CVTTSD2SI64rm },
381 { X86::Int_CVTTSD2SIrr, X86::Int_CVTTSD2SIrm },
382 { X86::Int_CVTTSS2SI64rr,X86::Int_CVTTSS2SI64rm },
383 { X86::Int_CVTTSS2SIrr, X86::Int_CVTTSS2SIrm },
384 { X86::Int_UCOMISDrr, X86::Int_UCOMISDrm },
385 { X86::Int_UCOMISSrr, X86::Int_UCOMISSrm },
386 { X86::MOV16rr, X86::MOV16rm },
387 { X86::MOV16to16_, X86::MOV16_rm },
388 { X86::MOV32rr, X86::MOV32rm },
389 { X86::MOV32to32_, X86::MOV32_rm },
390 { X86::MOV64rr, X86::MOV64rm },
391 { X86::MOV64toPQIrr, X86::MOVQI2PQIrm },
392 { X86::MOV64toSDrr, X86::MOV64toSDrm },
393 { X86::MOV8rr, X86::MOV8rm },
394 { X86::MOVAPDrr, X86::MOVAPDrm },
395 { X86::MOVAPSrr, X86::MOVAPSrm },
396 { X86::MOVDDUPrr, X86::MOVDDUPrm },
397 { X86::MOVDI2PDIrr, X86::MOVDI2PDIrm },
398 { X86::MOVDI2SSrr, X86::MOVDI2SSrm },
399 { X86::MOVSD2PDrr, X86::MOVSD2PDrm },
400 { X86::MOVSDrr, X86::MOVSDrm },
401 { X86::MOVSHDUPrr, X86::MOVSHDUPrm },
402 { X86::MOVSLDUPrr, X86::MOVSLDUPrm },
403 { X86::MOVSS2PSrr, X86::MOVSS2PSrm },
404 { X86::MOVSSrr, X86::MOVSSrm },
405 { X86::MOVSX16rr8, X86::MOVSX16rm8 },
406 { X86::MOVSX32rr16, X86::MOVSX32rm16 },
407 { X86::MOVSX32rr8, X86::MOVSX32rm8 },
408 { X86::MOVSX64rr16, X86::MOVSX64rm16 },
409 { X86::MOVSX64rr32, X86::MOVSX64rm32 },
410 { X86::MOVSX64rr8, X86::MOVSX64rm8 },
411 { X86::MOVUPDrr, X86::MOVUPDrm },
412 { X86::MOVUPSrr, X86::MOVUPSrm },
413 { X86::MOVZDI2PDIrr, X86::MOVZDI2PDIrm },
414 { X86::MOVZQI2PQIrr, X86::MOVZQI2PQIrm },
415 { X86::MOVZPQILo2PQIrr, X86::MOVZPQILo2PQIrm },
416 { X86::MOVZX16rr8, X86::MOVZX16rm8 },
417 { X86::MOVZX32rr16, X86::MOVZX32rm16 },
418 { X86::MOVZX32rr8, X86::MOVZX32rm8 },
419 { X86::MOVZX64rr16, X86::MOVZX64rm16 },
420 { X86::MOVZX64rr8, X86::MOVZX64rm8 },
421 { X86::PSHUFDri, X86::PSHUFDmi },
422 { X86::PSHUFHWri, X86::PSHUFHWmi },
423 { X86::PSHUFLWri, X86::PSHUFLWmi },
424 { X86::PsMOVZX64rr32, X86::PsMOVZX64rm32 },
425 { X86::RCPPSr, X86::RCPPSm },
426 { X86::RCPPSr_Int, X86::RCPPSm_Int },
427 { X86::RSQRTPSr, X86::RSQRTPSm },
428 { X86::RSQRTPSr_Int, X86::RSQRTPSm_Int },
429 { X86::RSQRTSSr, X86::RSQRTSSm },
430 { X86::RSQRTSSr_Int, X86::RSQRTSSm_Int },
431 { X86::SQRTPDr, X86::SQRTPDm },
432 { X86::SQRTPDr_Int, X86::SQRTPDm_Int },
433 { X86::SQRTPSr, X86::SQRTPSm },
434 { X86::SQRTPSr_Int, X86::SQRTPSm_Int },
435 { X86::SQRTSDr, X86::SQRTSDm },
436 { X86::SQRTSDr_Int, X86::SQRTSDm_Int },
437 { X86::SQRTSSr, X86::SQRTSSm },
438 { X86::SQRTSSr_Int, X86::SQRTSSm_Int },
439 { X86::TEST16rr, X86::TEST16rm },
440 { X86::TEST32rr, X86::TEST32rm },
441 { X86::TEST64rr, X86::TEST64rm },
442 { X86::TEST8rr, X86::TEST8rm },
443 // FIXME: TEST*rr EAX,EAX ---> CMP [mem], 0
444 { X86::UCOMISDrr, X86::UCOMISDrm },
445 { X86::UCOMISSrr, X86::UCOMISSrm },
446 { X86::XCHG16rr, X86::XCHG16rm },
447 { X86::XCHG32rr, X86::XCHG32rm },
448 { X86::XCHG64rr, X86::XCHG64rm },
449 { X86::XCHG8rr, X86::XCHG8rm }
452 for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {
453 unsigned RegOp = OpTbl1[i][0];
454 unsigned MemOp = OpTbl1[i][1];
455 if (!RegOp2MemOpTable1.insert(std::make_pair((unsigned*)RegOp, MemOp)))
456 assert(false && "Duplicated entries?");
457 unsigned AuxInfo = 1 | (1 << 4); // Index 1, folded load
458 if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr)
459 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
460 std::make_pair(RegOp, AuxInfo))))
461 AmbEntries.push_back(MemOp);
464 static const unsigned OpTbl2[][2] = {
465 { X86::ADC32rr, X86::ADC32rm },
466 { X86::ADC64rr, X86::ADC64rm },
467 { X86::ADD16rr, X86::ADD16rm },
468 { X86::ADD32rr, X86::ADD32rm },
469 { X86::ADD64rr, X86::ADD64rm },
470 { X86::ADD8rr, X86::ADD8rm },
471 { X86::ADDPDrr, X86::ADDPDrm },
472 { X86::ADDPSrr, X86::ADDPSrm },
473 { X86::ADDSDrr, X86::ADDSDrm },
474 { X86::ADDSSrr, X86::ADDSSrm },
475 { X86::ADDSUBPDrr, X86::ADDSUBPDrm },
476 { X86::ADDSUBPSrr, X86::ADDSUBPSrm },
477 { X86::AND16rr, X86::AND16rm },
478 { X86::AND32rr, X86::AND32rm },
479 { X86::AND64rr, X86::AND64rm },
480 { X86::AND8rr, X86::AND8rm },
481 { X86::ANDNPDrr, X86::ANDNPDrm },
482 { X86::ANDNPSrr, X86::ANDNPSrm },
483 { X86::ANDPDrr, X86::ANDPDrm },
484 { X86::ANDPSrr, X86::ANDPSrm },
485 { X86::CMOVA16rr, X86::CMOVA16rm },
486 { X86::CMOVA32rr, X86::CMOVA32rm },
487 { X86::CMOVA64rr, X86::CMOVA64rm },
488 { X86::CMOVAE16rr, X86::CMOVAE16rm },
489 { X86::CMOVAE32rr, X86::CMOVAE32rm },
490 { X86::CMOVAE64rr, X86::CMOVAE64rm },
491 { X86::CMOVB16rr, X86::CMOVB16rm },
492 { X86::CMOVB32rr, X86::CMOVB32rm },
493 { X86::CMOVB64rr, X86::CMOVB64rm },
494 { X86::CMOVBE16rr, X86::CMOVBE16rm },
495 { X86::CMOVBE32rr, X86::CMOVBE32rm },
496 { X86::CMOVBE64rr, X86::CMOVBE64rm },
497 { X86::CMOVE16rr, X86::CMOVE16rm },
498 { X86::CMOVE32rr, X86::CMOVE32rm },
499 { X86::CMOVE64rr, X86::CMOVE64rm },
500 { X86::CMOVG16rr, X86::CMOVG16rm },
501 { X86::CMOVG32rr, X86::CMOVG32rm },
502 { X86::CMOVG64rr, X86::CMOVG64rm },
503 { X86::CMOVGE16rr, X86::CMOVGE16rm },
504 { X86::CMOVGE32rr, X86::CMOVGE32rm },
505 { X86::CMOVGE64rr, X86::CMOVGE64rm },
506 { X86::CMOVL16rr, X86::CMOVL16rm },
507 { X86::CMOVL32rr, X86::CMOVL32rm },
508 { X86::CMOVL64rr, X86::CMOVL64rm },
509 { X86::CMOVLE16rr, X86::CMOVLE16rm },
510 { X86::CMOVLE32rr, X86::CMOVLE32rm },
511 { X86::CMOVLE64rr, X86::CMOVLE64rm },
512 { X86::CMOVNE16rr, X86::CMOVNE16rm },
513 { X86::CMOVNE32rr, X86::CMOVNE32rm },
514 { X86::CMOVNE64rr, X86::CMOVNE64rm },
515 { X86::CMOVNP16rr, X86::CMOVNP16rm },
516 { X86::CMOVNP32rr, X86::CMOVNP32rm },
517 { X86::CMOVNP64rr, X86::CMOVNP64rm },
518 { X86::CMOVNS16rr, X86::CMOVNS16rm },
519 { X86::CMOVNS32rr, X86::CMOVNS32rm },
520 { X86::CMOVNS64rr, X86::CMOVNS64rm },
521 { X86::CMOVP16rr, X86::CMOVP16rm },
522 { X86::CMOVP32rr, X86::CMOVP32rm },
523 { X86::CMOVP64rr, X86::CMOVP64rm },
524 { X86::CMOVS16rr, X86::CMOVS16rm },
525 { X86::CMOVS32rr, X86::CMOVS32rm },
526 { X86::CMOVS64rr, X86::CMOVS64rm },
527 { X86::CMPPDrri, X86::CMPPDrmi },
528 { X86::CMPPSrri, X86::CMPPSrmi },
529 { X86::CMPSDrr, X86::CMPSDrm },
530 { X86::CMPSSrr, X86::CMPSSrm },
531 { X86::DIVPDrr, X86::DIVPDrm },
532 { X86::DIVPSrr, X86::DIVPSrm },
533 { X86::DIVSDrr, X86::DIVSDrm },
534 { X86::DIVSSrr, X86::DIVSSrm },
535 { X86::HADDPDrr, X86::HADDPDrm },
536 { X86::HADDPSrr, X86::HADDPSrm },
537 { X86::HSUBPDrr, X86::HSUBPDrm },
538 { X86::HSUBPSrr, X86::HSUBPSrm },
539 { X86::IMUL16rr, X86::IMUL16rm },
540 { X86::IMUL32rr, X86::IMUL32rm },
541 { X86::IMUL64rr, X86::IMUL64rm },
542 { X86::MAXPDrr, X86::MAXPDrm },
543 { X86::MAXPDrr_Int, X86::MAXPDrm_Int },
544 { X86::MAXPSrr, X86::MAXPSrm },
545 { X86::MAXPSrr_Int, X86::MAXPSrm_Int },
546 { X86::MAXSDrr, X86::MAXSDrm },
547 { X86::MAXSDrr_Int, X86::MAXSDrm_Int },
548 { X86::MAXSSrr, X86::MAXSSrm },
549 { X86::MAXSSrr_Int, X86::MAXSSrm_Int },
550 { X86::MINPDrr, X86::MINPDrm },
551 { X86::MINPDrr_Int, X86::MINPDrm_Int },
552 { X86::MINPSrr, X86::MINPSrm },
553 { X86::MINPSrr_Int, X86::MINPSrm_Int },
554 { X86::MINSDrr, X86::MINSDrm },
555 { X86::MINSDrr_Int, X86::MINSDrm_Int },
556 { X86::MINSSrr, X86::MINSSrm },
557 { X86::MINSSrr_Int, X86::MINSSrm_Int },
558 { X86::MULPDrr, X86::MULPDrm },
559 { X86::MULPSrr, X86::MULPSrm },
560 { X86::MULSDrr, X86::MULSDrm },
561 { X86::MULSSrr, X86::MULSSrm },
562 { X86::OR16rr, X86::OR16rm },
563 { X86::OR32rr, X86::OR32rm },
564 { X86::OR64rr, X86::OR64rm },
565 { X86::OR8rr, X86::OR8rm },
566 { X86::ORPDrr, X86::ORPDrm },
567 { X86::ORPSrr, X86::ORPSrm },
568 { X86::PACKSSDWrr, X86::PACKSSDWrm },
569 { X86::PACKSSWBrr, X86::PACKSSWBrm },
570 { X86::PACKUSWBrr, X86::PACKUSWBrm },
571 { X86::PADDBrr, X86::PADDBrm },
572 { X86::PADDDrr, X86::PADDDrm },
573 { X86::PADDQrr, X86::PADDQrm },
574 { X86::PADDSBrr, X86::PADDSBrm },
575 { X86::PADDSWrr, X86::PADDSWrm },
576 { X86::PADDWrr, X86::PADDWrm },
577 { X86::PANDNrr, X86::PANDNrm },
578 { X86::PANDrr, X86::PANDrm },
579 { X86::PAVGBrr, X86::PAVGBrm },
580 { X86::PAVGWrr, X86::PAVGWrm },
581 { X86::PCMPEQBrr, X86::PCMPEQBrm },
582 { X86::PCMPEQDrr, X86::PCMPEQDrm },
583 { X86::PCMPEQWrr, X86::PCMPEQWrm },
584 { X86::PCMPGTBrr, X86::PCMPGTBrm },
585 { X86::PCMPGTDrr, X86::PCMPGTDrm },
586 { X86::PCMPGTWrr, X86::PCMPGTWrm },
587 { X86::PINSRWrri, X86::PINSRWrmi },
588 { X86::PMADDWDrr, X86::PMADDWDrm },
589 { X86::PMAXSWrr, X86::PMAXSWrm },
590 { X86::PMAXUBrr, X86::PMAXUBrm },
591 { X86::PMINSWrr, X86::PMINSWrm },
592 { X86::PMINUBrr, X86::PMINUBrm },
593 { X86::PMULHUWrr, X86::PMULHUWrm },
594 { X86::PMULHWrr, X86::PMULHWrm },
595 { X86::PMULLWrr, X86::PMULLWrm },
596 { X86::PMULUDQrr, X86::PMULUDQrm },
597 { X86::PORrr, X86::PORrm },
598 { X86::PSADBWrr, X86::PSADBWrm },
599 { X86::PSLLDrr, X86::PSLLDrm },
600 { X86::PSLLQrr, X86::PSLLQrm },
601 { X86::PSLLWrr, X86::PSLLWrm },
602 { X86::PSRADrr, X86::PSRADrm },
603 { X86::PSRAWrr, X86::PSRAWrm },
604 { X86::PSRLDrr, X86::PSRLDrm },
605 { X86::PSRLQrr, X86::PSRLQrm },
606 { X86::PSRLWrr, X86::PSRLWrm },
607 { X86::PSUBBrr, X86::PSUBBrm },
608 { X86::PSUBDrr, X86::PSUBDrm },
609 { X86::PSUBSBrr, X86::PSUBSBrm },
610 { X86::PSUBSWrr, X86::PSUBSWrm },
611 { X86::PSUBWrr, X86::PSUBWrm },
612 { X86::PUNPCKHBWrr, X86::PUNPCKHBWrm },
613 { X86::PUNPCKHDQrr, X86::PUNPCKHDQrm },
614 { X86::PUNPCKHQDQrr, X86::PUNPCKHQDQrm },
615 { X86::PUNPCKHWDrr, X86::PUNPCKHWDrm },
616 { X86::PUNPCKLBWrr, X86::PUNPCKLBWrm },
617 { X86::PUNPCKLDQrr, X86::PUNPCKLDQrm },
618 { X86::PUNPCKLQDQrr, X86::PUNPCKLQDQrm },
619 { X86::PUNPCKLWDrr, X86::PUNPCKLWDrm },
620 { X86::PXORrr, X86::PXORrm },
621 { X86::SBB32rr, X86::SBB32rm },
622 { X86::SBB64rr, X86::SBB64rm },
623 { X86::SHUFPDrri, X86::SHUFPDrmi },
624 { X86::SHUFPSrri, X86::SHUFPSrmi },
625 { X86::SUB16rr, X86::SUB16rm },
626 { X86::SUB32rr, X86::SUB32rm },
627 { X86::SUB64rr, X86::SUB64rm },
628 { X86::SUB8rr, X86::SUB8rm },
629 { X86::SUBPDrr, X86::SUBPDrm },
630 { X86::SUBPSrr, X86::SUBPSrm },
631 { X86::SUBSDrr, X86::SUBSDrm },
632 { X86::SUBSSrr, X86::SUBSSrm },
633 // FIXME: TEST*rr -> swapped operand of TEST*mr.
634 { X86::UNPCKHPDrr, X86::UNPCKHPDrm },
635 { X86::UNPCKHPSrr, X86::UNPCKHPSrm },
636 { X86::UNPCKLPDrr, X86::UNPCKLPDrm },
637 { X86::UNPCKLPSrr, X86::UNPCKLPSrm },
638 { X86::XOR16rr, X86::XOR16rm },
639 { X86::XOR32rr, X86::XOR32rm },
640 { X86::XOR64rr, X86::XOR64rm },
641 { X86::XOR8rr, X86::XOR8rm },
642 { X86::XORPDrr, X86::XORPDrm },
643 { X86::XORPSrr, X86::XORPSrm }
646 for (unsigned i = 0, e = array_lengthof(OpTbl2); i != e; ++i) {
647 unsigned RegOp = OpTbl2[i][0];
648 unsigned MemOp = OpTbl2[i][1];
649 if (!RegOp2MemOpTable2.insert(std::make_pair((unsigned*)RegOp, MemOp)))
650 assert(false && "Duplicated entries?");
651 unsigned AuxInfo = 2 | (1 << 4); // Index 1, folded load
652 if (!MemOp2RegOpTable.insert(std::make_pair((unsigned*)MemOp,
653 std::make_pair(RegOp, AuxInfo))))
654 AmbEntries.push_back(MemOp);
657 // Remove ambiguous entries.
658 assert(AmbEntries.empty() && "Duplicated entries in unfolding maps?");
661 // getDwarfRegNum - This function maps LLVM register identifiers to the
662 // Dwarf specific numbering, used in debug info and exception tables.
664 int X86RegisterInfo::getDwarfRegNum(unsigned RegNo, bool isEH) const {
665 const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
666 unsigned Flavour = DWARFFlavour::X86_64;
667 if (!Subtarget->is64Bit()) {
668 if (Subtarget->isTargetDarwin()) {
669 Flavour = DWARFFlavour::X86_32_Darwin;
670 } else if (Subtarget->isTargetCygMing()) {
671 // Unsupported by now, just quick fallback
672 Flavour = DWARFFlavour::X86_32_ELF;
674 Flavour = DWARFFlavour::X86_32_ELF;
678 return X86GenRegisterInfo::getDwarfRegNumFull(RegNo, Flavour);
681 // getX86RegNum - This function maps LLVM register identifiers to their X86
682 // specific numbering, which is used in various places encoding instructions.
684 unsigned X86RegisterInfo::getX86RegNum(unsigned RegNo) {
686 case X86::RAX: case X86::EAX: case X86::AX: case X86::AL: return N86::EAX;
687 case X86::RCX: case X86::ECX: case X86::CX: case X86::CL: return N86::ECX;
688 case X86::RDX: case X86::EDX: case X86::DX: case X86::DL: return N86::EDX;
689 case X86::RBX: case X86::EBX: case X86::BX: case X86::BL: return N86::EBX;
690 case X86::RSP: case X86::ESP: case X86::SP: case X86::SPL: case X86::AH:
692 case X86::RBP: case X86::EBP: case X86::BP: case X86::BPL: case X86::CH:
694 case X86::RSI: case X86::ESI: case X86::SI: case X86::SIL: case X86::DH:
696 case X86::RDI: case X86::EDI: case X86::DI: case X86::DIL: case X86::BH:
699 case X86::R8: case X86::R8D: case X86::R8W: case X86::R8B:
701 case X86::R9: case X86::R9D: case X86::R9W: case X86::R9B:
703 case X86::R10: case X86::R10D: case X86::R10W: case X86::R10B:
705 case X86::R11: case X86::R11D: case X86::R11W: case X86::R11B:
707 case X86::R12: case X86::R12D: case X86::R12W: case X86::R12B:
709 case X86::R13: case X86::R13D: case X86::R13W: case X86::R13B:
711 case X86::R14: case X86::R14D: case X86::R14W: case X86::R14B:
713 case X86::R15: case X86::R15D: case X86::R15W: case X86::R15B:
716 case X86::ST0: case X86::ST1: case X86::ST2: case X86::ST3:
717 case X86::ST4: case X86::ST5: case X86::ST6: case X86::ST7:
718 return RegNo-X86::ST0;
720 case X86::XMM0: case X86::XMM8: case X86::MM0:
722 case X86::XMM1: case X86::XMM9: case X86::MM1:
724 case X86::XMM2: case X86::XMM10: case X86::MM2:
726 case X86::XMM3: case X86::XMM11: case X86::MM3:
728 case X86::XMM4: case X86::XMM12: case X86::MM4:
730 case X86::XMM5: case X86::XMM13: case X86::MM5:
732 case X86::XMM6: case X86::XMM14: case X86::MM6:
734 case X86::XMM7: case X86::XMM15: case X86::MM7:
738 assert(isVirtualRegister(RegNo) && "Unknown physical register!");
739 assert(0 && "Register allocator hasn't allocated reg correctly yet!");
744 bool X86RegisterInfo::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
745 MachineBasicBlock::iterator MI,
746 const std::vector<CalleeSavedInfo> &CSI) const {
750 MachineFunction &MF = *MBB.getParent();
751 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
752 X86FI->setCalleeSavedFrameSize(CSI.size() * SlotSize);
753 unsigned Opc = Is64Bit ? X86::PUSH64r : X86::PUSH32r;
754 for (unsigned i = CSI.size(); i != 0; --i) {
755 unsigned Reg = CSI[i-1].getReg();
756 // Add the callee-saved register as live-in. It's killed at the spill.
758 BuildMI(MBB, MI, TII.get(Opc)).addReg(Reg);
763 bool X86RegisterInfo::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
764 MachineBasicBlock::iterator MI,
765 const std::vector<CalleeSavedInfo> &CSI) const {
769 unsigned Opc = Is64Bit ? X86::POP64r : X86::POP32r;
770 for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
771 unsigned Reg = CSI[i].getReg();
772 BuildMI(MBB, MI, TII.get(Opc), Reg);
777 static const MachineInstrBuilder &X86InstrAddOperand(MachineInstrBuilder &MIB,
778 MachineOperand &MO) {
780 MIB = MIB.addReg(MO.getReg(), MO.isDef(), MO.isImplicit(),
781 false, false, MO.getSubReg());
782 else if (MO.isImmediate())
783 MIB = MIB.addImm(MO.getImm());
784 else if (MO.isFrameIndex())
785 MIB = MIB.addFrameIndex(MO.getIndex());
786 else if (MO.isGlobalAddress())
787 MIB = MIB.addGlobalAddress(MO.getGlobal(), MO.getOffset());
788 else if (MO.isConstantPoolIndex())
789 MIB = MIB.addConstantPoolIndex(MO.getIndex(), MO.getOffset());
790 else if (MO.isJumpTableIndex())
791 MIB = MIB.addJumpTableIndex(MO.getIndex());
792 else if (MO.isExternalSymbol())
793 MIB = MIB.addExternalSymbol(MO.getSymbolName());
795 assert(0 && "Unknown operand for X86InstrAddOperand!");
800 static unsigned getStoreRegOpcode(const TargetRegisterClass *RC,
801 unsigned StackAlign) {
803 if (RC == &X86::GR64RegClass) {
805 } else if (RC == &X86::GR32RegClass) {
807 } else if (RC == &X86::GR16RegClass) {
809 } else if (RC == &X86::GR8RegClass) {
811 } else if (RC == &X86::GR32_RegClass) {
813 } else if (RC == &X86::GR16_RegClass) {
815 } else if (RC == &X86::RFP80RegClass) {
816 Opc = X86::ST_FpP80m; // pops
817 } else if (RC == &X86::RFP64RegClass) {
819 } else if (RC == &X86::RFP32RegClass) {
821 } else if (RC == &X86::FR32RegClass) {
823 } else if (RC == &X86::FR64RegClass) {
825 } else if (RC == &X86::VR128RegClass) {
826 // FIXME: Use movaps once we are capable of selectively
827 // aligning functions that spill SSE registers on 16-byte boundaries.
828 Opc = StackAlign >= 16 ? X86::MOVAPSmr : X86::MOVUPSmr;
829 } else if (RC == &X86::VR64RegClass) {
830 Opc = X86::MMX_MOVQ64mr;
832 assert(0 && "Unknown regclass");
839 void X86RegisterInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
840 MachineBasicBlock::iterator MI,
841 unsigned SrcReg, bool isKill, int FrameIdx,
842 const TargetRegisterClass *RC) const {
843 unsigned Opc = getStoreRegOpcode(RC, StackAlign);
844 addFrameReference(BuildMI(MBB, MI, TII.get(Opc)), FrameIdx)
845 .addReg(SrcReg, false, false, isKill);
848 void X86RegisterInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
850 SmallVectorImpl<MachineOperand> &Addr,
851 const TargetRegisterClass *RC,
852 SmallVectorImpl<MachineInstr*> &NewMIs) const {
853 unsigned Opc = getStoreRegOpcode(RC, StackAlign);
854 MachineInstrBuilder MIB = BuildMI(TII.get(Opc));
855 for (unsigned i = 0, e = Addr.size(); i != e; ++i)
856 MIB = X86InstrAddOperand(MIB, Addr[i]);
857 MIB.addReg(SrcReg, false, false, isKill);
858 NewMIs.push_back(MIB);
861 static unsigned getLoadRegOpcode(const TargetRegisterClass *RC,
862 unsigned StackAlign) {
864 if (RC == &X86::GR64RegClass) {
866 } else if (RC == &X86::GR32RegClass) {
868 } else if (RC == &X86::GR16RegClass) {
870 } else if (RC == &X86::GR8RegClass) {
872 } else if (RC == &X86::GR32_RegClass) {
874 } else if (RC == &X86::GR16_RegClass) {
876 } else if (RC == &X86::RFP80RegClass) {
878 } else if (RC == &X86::RFP64RegClass) {
880 } else if (RC == &X86::RFP32RegClass) {
882 } else if (RC == &X86::FR32RegClass) {
884 } else if (RC == &X86::FR64RegClass) {
886 } else if (RC == &X86::VR128RegClass) {
887 // FIXME: Use movaps once we are capable of selectively
888 // aligning functions that spill SSE registers on 16-byte boundaries.
889 Opc = StackAlign >= 16 ? X86::MOVAPSrm : X86::MOVUPSrm;
890 } else if (RC == &X86::VR64RegClass) {
891 Opc = X86::MMX_MOVQ64rm;
893 assert(0 && "Unknown regclass");
900 void X86RegisterInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
901 MachineBasicBlock::iterator MI,
902 unsigned DestReg, int FrameIdx,
903 const TargetRegisterClass *RC) const{
904 unsigned Opc = getLoadRegOpcode(RC, StackAlign);
905 addFrameReference(BuildMI(MBB, MI, TII.get(Opc), DestReg), FrameIdx);
908 void X86RegisterInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
909 SmallVectorImpl<MachineOperand> &Addr,
910 const TargetRegisterClass *RC,
911 SmallVectorImpl<MachineInstr*> &NewMIs) const {
912 unsigned Opc = getLoadRegOpcode(RC, StackAlign);
913 MachineInstrBuilder MIB = BuildMI(TII.get(Opc), DestReg);
914 for (unsigned i = 0, e = Addr.size(); i != e; ++i)
915 MIB = X86InstrAddOperand(MIB, Addr[i]);
916 NewMIs.push_back(MIB);
919 const TargetRegisterClass *
920 X86RegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
921 if (RC == &X86::CCRRegClass)
923 return &X86::GR64RegClass;
925 return &X86::GR32RegClass;
929 void X86RegisterInfo::reMaterialize(MachineBasicBlock &MBB,
930 MachineBasicBlock::iterator I,
932 const MachineInstr *Orig) const {
933 // MOV32r0 etc. are implemented with xor which clobbers condition code.
934 // Re-materialize them as movri instructions to avoid side effects.
935 switch (Orig->getOpcode()) {
937 BuildMI(MBB, I, TII.get(X86::MOV8ri), DestReg).addImm(0);
940 BuildMI(MBB, I, TII.get(X86::MOV16ri), DestReg).addImm(0);
943 BuildMI(MBB, I, TII.get(X86::MOV32ri), DestReg).addImm(0);
946 BuildMI(MBB, I, TII.get(X86::MOV64ri32), DestReg).addImm(0);
949 MachineInstr *MI = Orig->clone();
950 MI->getOperand(0).setReg(DestReg);
957 static MachineInstr *FuseTwoAddrInst(unsigned Opcode,
958 SmallVector<MachineOperand,4> &MOs,
959 MachineInstr *MI, const TargetInstrInfo &TII) {
960 // Create the base instruction with the memory operand as the first part.
961 MachineInstr *NewMI = new MachineInstr(TII.get(Opcode), true);
962 MachineInstrBuilder MIB(NewMI);
963 unsigned NumAddrOps = MOs.size();
964 for (unsigned i = 0; i != NumAddrOps; ++i)
965 MIB = X86InstrAddOperand(MIB, MOs[i]);
966 if (NumAddrOps < 4) // FrameIndex only
967 MIB.addImm(1).addReg(0).addImm(0);
969 // Loop over the rest of the ri operands, converting them over.
970 unsigned NumOps = TII.getNumOperands(MI->getOpcode())-2;
971 for (unsigned i = 0; i != NumOps; ++i) {
972 MachineOperand &MO = MI->getOperand(i+2);
973 MIB = X86InstrAddOperand(MIB, MO);
975 for (unsigned i = NumOps+2, e = MI->getNumOperands(); i != e; ++i) {
976 MachineOperand &MO = MI->getOperand(i);
977 MIB = X86InstrAddOperand(MIB, MO);
982 static MachineInstr *FuseInst(unsigned Opcode, unsigned OpNo,
983 SmallVector<MachineOperand,4> &MOs,
984 MachineInstr *MI, const TargetInstrInfo &TII) {
985 MachineInstr *NewMI = new MachineInstr(TII.get(Opcode), true);
986 MachineInstrBuilder MIB(NewMI);
988 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
989 MachineOperand &MO = MI->getOperand(i);
991 assert(MO.isRegister() && "Expected to fold into reg operand!");
992 unsigned NumAddrOps = MOs.size();
993 for (unsigned i = 0; i != NumAddrOps; ++i)
994 MIB = X86InstrAddOperand(MIB, MOs[i]);
995 if (NumAddrOps < 4) // FrameIndex only
996 MIB.addImm(1).addReg(0).addImm(0);
998 MIB = X86InstrAddOperand(MIB, MO);
1004 static MachineInstr *MakeM0Inst(const TargetInstrInfo &TII, unsigned Opcode,
1005 SmallVector<MachineOperand,4> &MOs,
1007 MachineInstrBuilder MIB = BuildMI(TII.get(Opcode));
1009 unsigned NumAddrOps = MOs.size();
1010 for (unsigned i = 0; i != NumAddrOps; ++i)
1011 MIB = X86InstrAddOperand(MIB, MOs[i]);
1012 if (NumAddrOps < 4) // FrameIndex only
1013 MIB.addImm(1).addReg(0).addImm(0);
1014 return MIB.addImm(0);
1018 X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
1019 SmallVector<MachineOperand,4> &MOs) const {
1020 const DenseMap<unsigned*, unsigned> *OpcodeTablePtr = NULL;
1021 bool isTwoAddrFold = false;
1022 unsigned NumOps = TII.getNumOperands(MI->getOpcode());
1023 bool isTwoAddr = NumOps > 1 &&
1024 MI->getInstrDescriptor()->getOperandConstraint(1, TOI::TIED_TO) != -1;
1026 MachineInstr *NewMI = NULL;
1027 // Folding a memory location into the two-address part of a two-address
1028 // instruction is different than folding it other places. It requires
1029 // replacing the *two* registers with the memory location.
1030 if (isTwoAddr && NumOps >= 2 && i < 2 &&
1031 MI->getOperand(0).isRegister() &&
1032 MI->getOperand(1).isRegister() &&
1033 MI->getOperand(0).getReg() == MI->getOperand(1).getReg()) {
1034 OpcodeTablePtr = &RegOp2MemOpTable2Addr;
1035 isTwoAddrFold = true;
1036 } else if (i == 0) { // If operand 0
1037 if (MI->getOpcode() == X86::MOV16r0)
1038 NewMI = MakeM0Inst(TII, X86::MOV16mi, MOs, MI);
1039 else if (MI->getOpcode() == X86::MOV32r0)
1040 NewMI = MakeM0Inst(TII, X86::MOV32mi, MOs, MI);
1041 else if (MI->getOpcode() == X86::MOV64r0)
1042 NewMI = MakeM0Inst(TII, X86::MOV64mi32, MOs, MI);
1043 else if (MI->getOpcode() == X86::MOV8r0)
1044 NewMI = MakeM0Inst(TII, X86::MOV8mi, MOs, MI);
1046 NewMI->copyKillDeadInfo(MI);
1050 OpcodeTablePtr = &RegOp2MemOpTable0;
1051 } else if (i == 1) {
1052 OpcodeTablePtr = &RegOp2MemOpTable1;
1053 } else if (i == 2) {
1054 OpcodeTablePtr = &RegOp2MemOpTable2;
1057 // If table selected...
1058 if (OpcodeTablePtr) {
1059 // Find the Opcode to fuse
1060 DenseMap<unsigned*, unsigned>::iterator I =
1061 OpcodeTablePtr->find((unsigned*)MI->getOpcode());
1062 if (I != OpcodeTablePtr->end()) {
1064 NewMI = FuseTwoAddrInst(I->second, MOs, MI, TII);
1066 NewMI = FuseInst(I->second, i, MOs, MI, TII);
1067 NewMI->copyKillDeadInfo(MI);
1073 if (PrintFailedFusing)
1074 cerr << "We failed to fuse ("
1075 << ((i == 1) ? "r" : "s") << "): " << *MI;
1080 MachineInstr* X86RegisterInfo::foldMemoryOperand(MachineInstr *MI,
1081 SmallVectorImpl<unsigned> &Ops,
1082 int FrameIndex) const {
1083 // Check switch flag
1084 if (NoFusing) return NULL;
1086 if (Ops.size() == 2 && Ops[0] == 0 && Ops[1] == 1) {
1087 unsigned NewOpc = 0;
1088 switch (MI->getOpcode()) {
1089 default: return NULL;
1090 case X86::TEST8rr: NewOpc = X86::CMP8ri; break;
1091 case X86::TEST16rr: NewOpc = X86::CMP16ri; break;
1092 case X86::TEST32rr: NewOpc = X86::CMP32ri; break;
1093 case X86::TEST64rr: NewOpc = X86::CMP64ri32; break;
1095 // Change to CMPXXri r, 0 first.
1096 MI->setInstrDescriptor(TII.get(NewOpc));
1097 MI->getOperand(1).ChangeToImmediate(0);
1098 } else if (Ops.size() != 1)
1101 SmallVector<MachineOperand,4> MOs;
1102 MOs.push_back(MachineOperand::CreateFI(FrameIndex));
1103 return foldMemoryOperand(MI, Ops[0], MOs);
1106 MachineInstr* X86RegisterInfo::foldMemoryOperand(MachineInstr *MI,
1107 SmallVectorImpl<unsigned> &Ops,
1108 MachineInstr *LoadMI) const {
1109 // Check switch flag
1110 if (NoFusing) return NULL;
1112 if (Ops.size() == 2 && Ops[0] == 0 && Ops[1] == 1) {
1113 unsigned NewOpc = 0;
1114 switch (MI->getOpcode()) {
1115 default: return NULL;
1116 case X86::TEST8rr: NewOpc = X86::CMP8ri; break;
1117 case X86::TEST16rr: NewOpc = X86::CMP16ri; break;
1118 case X86::TEST32rr: NewOpc = X86::CMP32ri; break;
1119 case X86::TEST64rr: NewOpc = X86::CMP64ri32; break;
1121 // Change to CMPXXri r, 0 first.
1122 MI->setInstrDescriptor(TII.get(NewOpc));
1123 MI->getOperand(1).ChangeToImmediate(0);
1124 } else if (Ops.size() != 1)
1127 SmallVector<MachineOperand,4> MOs;
1128 unsigned NumOps = TII.getNumOperands(LoadMI->getOpcode());
1129 for (unsigned i = NumOps - 4; i != NumOps; ++i)
1130 MOs.push_back(LoadMI->getOperand(i));
1131 return foldMemoryOperand(MI, Ops[0], MOs);
1135 bool X86RegisterInfo::canFoldMemoryOperand(MachineInstr *MI,
1136 SmallVectorImpl<unsigned> &Ops) const {
1137 // Check switch flag
1138 if (NoFusing) return 0;
1140 if (Ops.size() == 2 && Ops[0] == 0 && Ops[1] == 1) {
1141 switch (MI->getOpcode()) {
1142 default: return false;
1151 if (Ops.size() != 1)
1154 unsigned OpNum = Ops[0];
1155 unsigned Opc = MI->getOpcode();
1156 unsigned NumOps = TII.getNumOperands(Opc);
1157 bool isTwoAddr = NumOps > 1 &&
1158 TII.getOperandConstraint(Opc, 1, TOI::TIED_TO) != -1;
1160 // Folding a memory location into the two-address part of a two-address
1161 // instruction is different than folding it other places. It requires
1162 // replacing the *two* registers with the memory location.
1163 const DenseMap<unsigned*, unsigned> *OpcodeTablePtr = NULL;
1164 if (isTwoAddr && NumOps >= 2 && OpNum < 2) {
1165 OpcodeTablePtr = &RegOp2MemOpTable2Addr;
1166 } else if (OpNum == 0) { // If operand 0
1175 OpcodeTablePtr = &RegOp2MemOpTable0;
1176 } else if (OpNum == 1) {
1177 OpcodeTablePtr = &RegOp2MemOpTable1;
1178 } else if (OpNum == 2) {
1179 OpcodeTablePtr = &RegOp2MemOpTable2;
1182 if (OpcodeTablePtr) {
1183 // Find the Opcode to fuse
1184 DenseMap<unsigned*, unsigned>::iterator I =
1185 OpcodeTablePtr->find((unsigned*)Opc);
1186 if (I != OpcodeTablePtr->end())
1192 bool X86RegisterInfo::unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
1193 unsigned Reg, bool UnfoldLoad, bool UnfoldStore,
1194 SmallVectorImpl<MachineInstr*> &NewMIs) const {
1195 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1196 MemOp2RegOpTable.find((unsigned*)MI->getOpcode());
1197 if (I == MemOp2RegOpTable.end())
1199 unsigned Opc = I->second.first;
1200 unsigned Index = I->second.second & 0xf;
1201 bool FoldedLoad = I->second.second & (1 << 4);
1202 bool FoldedStore = I->second.second & (1 << 5);
1203 if (UnfoldLoad && !FoldedLoad)
1205 UnfoldLoad &= FoldedLoad;
1206 if (UnfoldStore && !FoldedStore)
1208 UnfoldStore &= FoldedStore;
1210 const TargetInstrDescriptor &TID = TII.get(Opc);
1211 const TargetOperandInfo &TOI = TID.OpInfo[Index];
1212 const TargetRegisterClass *RC = (TOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1213 ? TII.getPointerRegClass() : getRegClass(TOI.RegClass);
1214 SmallVector<MachineOperand,4> AddrOps;
1215 SmallVector<MachineOperand,2> BeforeOps;
1216 SmallVector<MachineOperand,2> AfterOps;
1217 SmallVector<MachineOperand,4> ImpOps;
1218 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
1219 MachineOperand &Op = MI->getOperand(i);
1220 if (i >= Index && i < Index+4)
1221 AddrOps.push_back(Op);
1222 else if (Op.isRegister() && Op.isImplicit())
1223 ImpOps.push_back(Op);
1225 BeforeOps.push_back(Op);
1227 AfterOps.push_back(Op);
1230 // Emit the load instruction.
1232 loadRegFromAddr(MF, Reg, AddrOps, RC, NewMIs);
1234 // Address operands cannot be marked isKill.
1235 for (unsigned i = 1; i != 5; ++i) {
1236 MachineOperand &MO = NewMIs[0]->getOperand(i);
1237 if (MO.isRegister())
1238 MO.setIsKill(false);
1243 // Emit the data processing instruction.
1244 MachineInstr *DataMI = new MachineInstr(TID, true);
1245 MachineInstrBuilder MIB(DataMI);
1248 MIB.addReg(Reg, true);
1249 for (unsigned i = 0, e = BeforeOps.size(); i != e; ++i)
1250 MIB = X86InstrAddOperand(MIB, BeforeOps[i]);
1253 for (unsigned i = 0, e = AfterOps.size(); i != e; ++i)
1254 MIB = X86InstrAddOperand(MIB, AfterOps[i]);
1255 for (unsigned i = 0, e = ImpOps.size(); i != e; ++i) {
1256 MachineOperand &MO = ImpOps[i];
1257 MIB.addReg(MO.getReg(), MO.isDef(), true, MO.isKill(), MO.isDead());
1259 // Change CMP32ri r, 0 back to TEST32rr r, r, etc.
1260 unsigned NewOpc = 0;
1261 switch (DataMI->getOpcode()) {
1263 case X86::CMP64ri32:
1267 MachineOperand &MO0 = DataMI->getOperand(0);
1268 MachineOperand &MO1 = DataMI->getOperand(1);
1269 if (MO1.getImm() == 0) {
1270 switch (DataMI->getOpcode()) {
1272 case X86::CMP64ri32: NewOpc = X86::TEST64rr; break;
1273 case X86::CMP32ri: NewOpc = X86::TEST32rr; break;
1274 case X86::CMP16ri: NewOpc = X86::TEST16rr; break;
1275 case X86::CMP8ri: NewOpc = X86::TEST8rr; break;
1277 DataMI->setInstrDescriptor(TII.get(NewOpc));
1278 MO1.ChangeToRegister(MO0.getReg(), false);
1282 NewMIs.push_back(DataMI);
1284 // Emit the store instruction.
1286 const TargetOperandInfo &DstTOI = TID.OpInfo[0];
1287 const TargetRegisterClass *DstRC = (DstTOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1288 ? TII.getPointerRegClass() : getRegClass(DstTOI.RegClass);
1289 storeRegToAddr(MF, Reg, true, AddrOps, DstRC, NewMIs);
1297 X86RegisterInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
1298 SmallVectorImpl<SDNode*> &NewNodes) const {
1299 if (!N->isTargetOpcode())
1302 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1303 MemOp2RegOpTable.find((unsigned*)N->getTargetOpcode());
1304 if (I == MemOp2RegOpTable.end())
1306 unsigned Opc = I->second.first;
1307 unsigned Index = I->second.second & 0xf;
1308 bool FoldedLoad = I->second.second & (1 << 4);
1309 bool FoldedStore = I->second.second & (1 << 5);
1310 const TargetInstrDescriptor &TID = TII.get(Opc);
1311 const TargetOperandInfo &TOI = TID.OpInfo[Index];
1312 const TargetRegisterClass *RC = (TOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1313 ? TII.getPointerRegClass() : getRegClass(TOI.RegClass);
1314 std::vector<SDOperand> AddrOps;
1315 std::vector<SDOperand> BeforeOps;
1316 std::vector<SDOperand> AfterOps;
1317 unsigned NumOps = N->getNumOperands();
1318 for (unsigned i = 0; i != NumOps-1; ++i) {
1319 SDOperand Op = N->getOperand(i);
1320 if (i >= Index && i < Index+4)
1321 AddrOps.push_back(Op);
1323 BeforeOps.push_back(Op);
1325 AfterOps.push_back(Op);
1327 SDOperand Chain = N->getOperand(NumOps-1);
1328 AddrOps.push_back(Chain);
1330 // Emit the load instruction.
1333 MVT::ValueType VT = *RC->vt_begin();
1334 Load = DAG.getTargetNode(getLoadRegOpcode(RC, StackAlign), VT, MVT::Other,
1335 &AddrOps[0], AddrOps.size());
1336 NewNodes.push_back(Load);
1339 // Emit the data processing instruction.
1340 std::vector<MVT::ValueType> VTs;
1341 const TargetRegisterClass *DstRC = 0;
1342 if (TID.numDefs > 0) {
1343 const TargetOperandInfo &DstTOI = TID.OpInfo[0];
1344 DstRC = (DstTOI.Flags & M_LOOK_UP_PTR_REG_CLASS)
1345 ? TII.getPointerRegClass() : getRegClass(DstTOI.RegClass);
1346 VTs.push_back(*DstRC->vt_begin());
1348 for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
1349 MVT::ValueType VT = N->getValueType(i);
1350 if (VT != MVT::Other && i >= TID.numDefs)
1354 BeforeOps.push_back(SDOperand(Load, 0));
1355 std::copy(AfterOps.begin(), AfterOps.end(), std::back_inserter(BeforeOps));
1356 SDNode *NewNode= DAG.getTargetNode(Opc, VTs, &BeforeOps[0], BeforeOps.size());
1357 NewNodes.push_back(NewNode);
1359 // Emit the store instruction.
1362 AddrOps.push_back(SDOperand(NewNode, 0));
1363 AddrOps.push_back(Chain);
1364 SDNode *Store = DAG.getTargetNode(getStoreRegOpcode(DstRC, StackAlign),
1365 MVT::Other, &AddrOps[0], AddrOps.size());
1366 NewNodes.push_back(Store);
1372 unsigned X86RegisterInfo::getOpcodeAfterMemoryUnfold(unsigned Opc,
1373 bool UnfoldLoad, bool UnfoldStore) const {
1374 DenseMap<unsigned*, std::pair<unsigned,unsigned> >::iterator I =
1375 MemOp2RegOpTable.find((unsigned*)Opc);
1376 if (I == MemOp2RegOpTable.end())
1378 bool FoldedLoad = I->second.second & (1 << 4);
1379 bool FoldedStore = I->second.second & (1 << 5);
1380 if (UnfoldLoad && !FoldedLoad)
1382 if (UnfoldStore && !FoldedStore)
1384 return I->second.first;
1388 X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
1389 static const unsigned CalleeSavedRegs32Bit[] = {
1390 X86::ESI, X86::EDI, X86::EBX, X86::EBP, 0
1393 static const unsigned CalleeSavedRegs32EHRet[] = {
1394 X86::EAX, X86::EDX, X86::ESI, X86::EDI, X86::EBX, X86::EBP, 0
1397 static const unsigned CalleeSavedRegs64Bit[] = {
1398 X86::RBX, X86::R12, X86::R13, X86::R14, X86::R15, X86::RBP, 0
1402 return CalleeSavedRegs64Bit;
1405 MachineFrameInfo *MFI = MF->getFrameInfo();
1406 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1407 if (MMI && MMI->callsEHReturn())
1408 return CalleeSavedRegs32EHRet;
1410 return CalleeSavedRegs32Bit;
1414 const TargetRegisterClass* const*
1415 X86RegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
1416 static const TargetRegisterClass * const CalleeSavedRegClasses32Bit[] = {
1417 &X86::GR32RegClass, &X86::GR32RegClass,
1418 &X86::GR32RegClass, &X86::GR32RegClass, 0
1420 static const TargetRegisterClass * const CalleeSavedRegClasses32EHRet[] = {
1421 &X86::GR32RegClass, &X86::GR32RegClass,
1422 &X86::GR32RegClass, &X86::GR32RegClass,
1423 &X86::GR32RegClass, &X86::GR32RegClass, 0
1425 static const TargetRegisterClass * const CalleeSavedRegClasses64Bit[] = {
1426 &X86::GR64RegClass, &X86::GR64RegClass,
1427 &X86::GR64RegClass, &X86::GR64RegClass,
1428 &X86::GR64RegClass, &X86::GR64RegClass, 0
1432 return CalleeSavedRegClasses64Bit;
1435 MachineFrameInfo *MFI = MF->getFrameInfo();
1436 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1437 if (MMI && MMI->callsEHReturn())
1438 return CalleeSavedRegClasses32EHRet;
1440 return CalleeSavedRegClasses32Bit;
1445 BitVector X86RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
1446 BitVector Reserved(getNumRegs());
1447 Reserved.set(X86::RSP);
1448 Reserved.set(X86::ESP);
1449 Reserved.set(X86::SP);
1450 Reserved.set(X86::SPL);
1452 Reserved.set(X86::RBP);
1453 Reserved.set(X86::EBP);
1454 Reserved.set(X86::BP);
1455 Reserved.set(X86::BPL);
1460 //===----------------------------------------------------------------------===//
1461 // Stack Frame Processing methods
1462 //===----------------------------------------------------------------------===//
1464 // hasFP - Return true if the specified function should have a dedicated frame
1465 // pointer register. This is true if the function has variable sized allocas or
1466 // if frame pointer elimination is disabled.
1468 bool X86RegisterInfo::hasFP(const MachineFunction &MF) const {
1469 MachineFrameInfo *MFI = MF.getFrameInfo();
1470 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1472 return (NoFramePointerElim ||
1473 MFI->hasVarSizedObjects() ||
1474 MF.getInfo<X86MachineFunctionInfo>()->getForceFramePointer() ||
1475 (MMI && MMI->callsUnwindInit()));
1478 bool X86RegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
1479 return !MF.getFrameInfo()->hasVarSizedObjects();
1482 void X86RegisterInfo::
1483 eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
1484 MachineBasicBlock::iterator I) const {
1485 if (!hasReservedCallFrame(MF)) {
1486 // If the stack pointer can be changed after prologue, turn the
1487 // adjcallstackup instruction into a 'sub ESP, <amt>' and the
1488 // adjcallstackdown instruction into 'add ESP, <amt>'
1489 // TODO: consider using push / pop instead of sub + store / add
1490 MachineInstr *Old = I;
1491 uint64_t Amount = Old->getOperand(0).getImm();
1493 // We need to keep the stack aligned properly. To do this, we round the
1494 // amount of space needed for the outgoing arguments up to the next
1495 // alignment boundary.
1496 Amount = (Amount+StackAlign-1)/StackAlign*StackAlign;
1498 MachineInstr *New = 0;
1499 if (Old->getOpcode() == X86::ADJCALLSTACKDOWN) {
1500 New=BuildMI(TII.get(Is64Bit ? X86::SUB64ri32 : X86::SUB32ri), StackPtr)
1501 .addReg(StackPtr).addImm(Amount);
1503 assert(Old->getOpcode() == X86::ADJCALLSTACKUP);
1504 // factor out the amount the callee already popped.
1505 uint64_t CalleeAmt = Old->getOperand(1).getImm();
1506 Amount -= CalleeAmt;
1508 unsigned Opc = (Amount < 128) ?
1509 (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
1510 (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri);
1511 New = BuildMI(TII.get(Opc), StackPtr).addReg(StackPtr).addImm(Amount);
1515 // Replace the pseudo instruction with a new instruction...
1516 if (New) MBB.insert(I, New);
1518 } else if (I->getOpcode() == X86::ADJCALLSTACKUP) {
1519 // If we are performing frame pointer elimination and if the callee pops
1520 // something off the stack pointer, add it back. We do this until we have
1521 // more advanced stack pointer tracking ability.
1522 if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
1523 unsigned Opc = (CalleeAmt < 128) ?
1524 (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
1525 (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri);
1527 BuildMI(TII.get(Opc), StackPtr).addReg(StackPtr).addImm(CalleeAmt);
1535 void X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
1536 int SPAdj, RegScavenger *RS) const{
1537 assert(SPAdj == 0 && "Unexpected");
1540 MachineInstr &MI = *II;
1541 MachineFunction &MF = *MI.getParent()->getParent();
1542 while (!MI.getOperand(i).isFrameIndex()) {
1544 assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
1547 int FrameIndex = MI.getOperand(i).getIndex();
1548 // This must be part of a four operand memory reference. Replace the
1549 // FrameIndex with base register with EBP. Add an offset to the offset.
1550 MI.getOperand(i).ChangeToRegister(hasFP(MF) ? FramePtr : StackPtr, false);
1552 // Now add the frame object offset to the offset from EBP.
1553 int64_t Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
1554 MI.getOperand(i+3).getImm()+SlotSize;
1557 Offset += MF.getFrameInfo()->getStackSize();
1559 Offset += SlotSize; // Skip the saved EBP
1560 // Skip the RETADDR move area
1561 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1562 int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1563 if (TailCallReturnAddrDelta < 0) Offset -= TailCallReturnAddrDelta;
1566 MI.getOperand(i+3).ChangeToImmediate(Offset);
1570 X86RegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF) const{
1571 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1572 int32_t TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1573 if (TailCallReturnAddrDelta < 0) {
1574 // create RETURNADDR area
1584 CreateFixedObject(-TailCallReturnAddrDelta,
1585 (-1*SlotSize)+TailCallReturnAddrDelta);
1588 assert((TailCallReturnAddrDelta <= 0) &&
1589 "The Delta should always be zero or negative");
1590 // Create a frame entry for the EBP register that must be saved.
1591 int FrameIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize,
1593 TailCallReturnAddrDelta);
1594 assert(FrameIdx == MF.getFrameInfo()->getObjectIndexBegin() &&
1595 "Slot for EBP register must be last in order to be found!");
1599 /// emitSPUpdate - Emit a series of instructions to increment / decrement the
1600 /// stack pointer by a constant value.
1602 void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
1603 unsigned StackPtr, int64_t NumBytes, bool Is64Bit,
1604 const TargetInstrInfo &TII) {
1605 bool isSub = NumBytes < 0;
1606 uint64_t Offset = isSub ? -NumBytes : NumBytes;
1607 unsigned Opc = isSub
1609 (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
1610 (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri))
1612 (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
1613 (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri));
1614 uint64_t Chunk = (1LL << 31) - 1;
1617 uint64_t ThisVal = (Offset > Chunk) ? Chunk : Offset;
1618 BuildMI(MBB, MBBI, TII.get(Opc), StackPtr).addReg(StackPtr).addImm(ThisVal);
1623 // mergeSPUpdatesUp - Merge two stack-manipulating instructions upper iterator.
1625 void mergeSPUpdatesUp(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
1626 unsigned StackPtr, uint64_t *NumBytes = NULL) {
1627 if (MBBI == MBB.begin()) return;
1629 MachineBasicBlock::iterator PI = prior(MBBI);
1630 unsigned Opc = PI->getOpcode();
1631 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1632 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1633 PI->getOperand(0).getReg() == StackPtr) {
1635 *NumBytes += PI->getOperand(2).getImm();
1637 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1638 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1639 PI->getOperand(0).getReg() == StackPtr) {
1641 *NumBytes -= PI->getOperand(2).getImm();
1646 // mergeSPUpdatesUp - Merge two stack-manipulating instructions lower iterator.
1648 void mergeSPUpdatesDown(MachineBasicBlock &MBB,
1649 MachineBasicBlock::iterator &MBBI,
1650 unsigned StackPtr, uint64_t *NumBytes = NULL) {
1653 if (MBBI == MBB.end()) return;
1655 MachineBasicBlock::iterator NI = next(MBBI);
1656 if (NI == MBB.end()) return;
1658 unsigned Opc = NI->getOpcode();
1659 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1660 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1661 NI->getOperand(0).getReg() == StackPtr) {
1663 *NumBytes -= NI->getOperand(2).getImm();
1666 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1667 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1668 NI->getOperand(0).getReg() == StackPtr) {
1670 *NumBytes += NI->getOperand(2).getImm();
1676 /// mergeSPUpdates - Checks the instruction before/after the passed
1677 /// instruction. If it is an ADD/SUB instruction it is deleted
1678 /// argument and the stack adjustment is returned as a positive value for ADD
1679 /// and a negative for SUB.
1680 static int mergeSPUpdates(MachineBasicBlock &MBB,
1681 MachineBasicBlock::iterator &MBBI,
1683 bool doMergeWithPrevious) {
1685 if ((doMergeWithPrevious && MBBI == MBB.begin()) ||
1686 (!doMergeWithPrevious && MBBI == MBB.end()))
1691 MachineBasicBlock::iterator PI = doMergeWithPrevious ? prior(MBBI) : MBBI;
1692 MachineBasicBlock::iterator NI = doMergeWithPrevious ? 0 : next(MBBI);
1693 unsigned Opc = PI->getOpcode();
1694 if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
1695 Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
1696 PI->getOperand(0).getReg() == StackPtr){
1697 Offset += PI->getOperand(2).getImm();
1699 if (!doMergeWithPrevious) MBBI = NI;
1700 } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
1701 Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
1702 PI->getOperand(0).getReg() == StackPtr) {
1703 Offset -= PI->getOperand(2).getImm();
1705 if (!doMergeWithPrevious) MBBI = NI;
1711 void X86RegisterInfo::emitPrologue(MachineFunction &MF) const {
1712 MachineBasicBlock &MBB = MF.front(); // Prolog goes in entry BB
1713 MachineFrameInfo *MFI = MF.getFrameInfo();
1714 const Function* Fn = MF.getFunction();
1715 const X86Subtarget* Subtarget = &MF.getTarget().getSubtarget<X86Subtarget>();
1716 MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
1717 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1718 MachineBasicBlock::iterator MBBI = MBB.begin();
1720 // Prepare for frame info.
1721 unsigned FrameLabelId = 0;
1723 // Get the number of bytes to allocate from the FrameInfo.
1724 uint64_t StackSize = MFI->getStackSize();
1725 // Add RETADDR move area to callee saved frame size.
1726 int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
1727 if (TailCallReturnAddrDelta < 0)
1728 X86FI->setCalleeSavedFrameSize(
1729 X86FI->getCalleeSavedFrameSize() +(-TailCallReturnAddrDelta));
1730 uint64_t NumBytes = StackSize - X86FI->getCalleeSavedFrameSize();
1732 // Insert stack pointer adjustment for later moving of return addr. Only
1733 // applies to tail call optimized functions where the callee argument stack
1734 // size is bigger than the callers.
1735 if (TailCallReturnAddrDelta < 0) {
1736 BuildMI(MBB, MBBI, TII.get(Is64Bit? X86::SUB64ri32 : X86::SUB32ri),
1737 StackPtr).addReg(StackPtr).addImm(-TailCallReturnAddrDelta);
1741 // Get the offset of the stack slot for the EBP register... which is
1742 // guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
1743 // Update the frame offset adjustment.
1744 MFI->setOffsetAdjustment(SlotSize-NumBytes);
1746 // Save EBP into the appropriate stack slot...
1747 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::PUSH64r : X86::PUSH32r))
1749 NumBytes -= SlotSize;
1751 if (MMI && MMI->needsFrameInfo()) {
1752 // Mark effective beginning of when frame pointer becomes valid.
1753 FrameLabelId = MMI->NextLabelID();
1754 BuildMI(MBB, MBBI, TII.get(X86::LABEL)).addImm(FrameLabelId);
1757 // Update EBP with the new base value...
1758 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr), FramePtr)
1762 unsigned ReadyLabelId = 0;
1763 if (MMI && MMI->needsFrameInfo()) {
1764 // Mark effective beginning of when frame pointer is ready.
1765 ReadyLabelId = MMI->NextLabelID();
1766 BuildMI(MBB, MBBI, TII.get(X86::LABEL)).addImm(ReadyLabelId);
1769 // Skip the callee-saved push instructions.
1770 while (MBBI != MBB.end() &&
1771 (MBBI->getOpcode() == X86::PUSH32r ||
1772 MBBI->getOpcode() == X86::PUSH64r))
1775 if (NumBytes) { // adjust stack pointer: ESP -= numbytes
1776 if (NumBytes >= 4096 && Subtarget->isTargetCygMing()) {
1777 // Check, whether EAX is livein for this function
1778 bool isEAXAlive = false;
1779 for (MachineRegisterInfo::livein_iterator
1780 II = MF.getRegInfo().livein_begin(),
1781 EE = MF.getRegInfo().livein_end(); (II != EE) && !isEAXAlive; ++II) {
1782 unsigned Reg = II->first;
1783 isEAXAlive = (Reg == X86::EAX || Reg == X86::AX ||
1784 Reg == X86::AH || Reg == X86::AL);
1787 // Function prologue calls _alloca to probe the stack when allocating
1788 // more than 4k bytes in one go. Touching the stack at 4K increments is
1789 // necessary to ensure that the guard pages used by the OS virtual memory
1790 // manager are allocated in correct sequence.
1792 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(NumBytes);
1793 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32))
1794 .addExternalSymbol("_alloca");
1797 BuildMI(MBB, MBBI, TII.get(X86::PUSH32r), X86::EAX);
1798 // Allocate NumBytes-4 bytes on stack. We'll also use 4 already
1799 // allocated bytes for EAX.
1800 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(NumBytes-4);
1801 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32))
1802 .addExternalSymbol("_alloca");
1804 MachineInstr *MI = addRegOffset(BuildMI(TII.get(X86::MOV32rm),X86::EAX),
1805 StackPtr, NumBytes-4);
1806 MBB.insert(MBBI, MI);
1809 // If there is an SUB32ri of ESP immediately before this instruction,
1810 // merge the two. This can be the case when tail call elimination is
1811 // enabled and the callee has more arguments then the caller.
1812 NumBytes -= mergeSPUpdates(MBB, MBBI, StackPtr, true);
1813 // If there is an ADD32ri or SUB32ri of ESP immediately after this
1814 // instruction, merge the two instructions.
1815 mergeSPUpdatesDown(MBB, MBBI, StackPtr, &NumBytes);
1818 emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, TII);
1822 if (MMI && MMI->needsFrameInfo()) {
1823 std::vector<MachineMove> &Moves = MMI->getFrameMoves();
1824 const TargetData *TD = MF.getTarget().getTargetData();
1826 // Calculate amount of bytes used for return address storing
1828 (MF.getTarget().getFrameInfo()->getStackGrowthDirection() ==
1829 TargetFrameInfo::StackGrowsUp ?
1830 TD->getPointerSize() : -TD->getPointerSize());
1833 // Show update of SP.
1836 MachineLocation SPDst(MachineLocation::VirtualFP);
1837 MachineLocation SPSrc(MachineLocation::VirtualFP, 2*stackGrowth);
1838 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1840 MachineLocation SPDst(MachineLocation::VirtualFP);
1841 MachineLocation SPSrc(MachineLocation::VirtualFP,
1842 -StackSize+stackGrowth);
1843 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1846 //FIXME: Verify & implement for FP
1847 MachineLocation SPDst(StackPtr);
1848 MachineLocation SPSrc(StackPtr, stackGrowth);
1849 Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
1852 // Add callee saved registers to move list.
1853 const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
1855 // FIXME: This is dirty hack. The code itself is pretty mess right now.
1856 // It should be rewritten from scratch and generalized sometimes.
1858 // Determine maximum offset (minumum due to stack growth)
1859 int64_t MaxOffset = 0;
1860 for (unsigned I = 0, E = CSI.size(); I!=E; ++I)
1861 MaxOffset = std::min(MaxOffset,
1862 MFI->getObjectOffset(CSI[I].getFrameIdx()));
1864 // Calculate offsets
1865 int64_t saveAreaOffset = (hasFP(MF) ? 3 : 2)*stackGrowth;
1866 for (unsigned I = 0, E = CSI.size(); I!=E; ++I) {
1867 int64_t Offset = MFI->getObjectOffset(CSI[I].getFrameIdx());
1868 unsigned Reg = CSI[I].getReg();
1869 Offset = (MaxOffset-Offset+saveAreaOffset);
1870 MachineLocation CSDst(MachineLocation::VirtualFP, Offset);
1871 MachineLocation CSSrc(Reg);
1872 Moves.push_back(MachineMove(FrameLabelId, CSDst, CSSrc));
1877 MachineLocation FPDst(MachineLocation::VirtualFP, 2*stackGrowth);
1878 MachineLocation FPSrc(FramePtr);
1879 Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
1882 MachineLocation FPDst(hasFP(MF) ? FramePtr : StackPtr);
1883 MachineLocation FPSrc(MachineLocation::VirtualFP);
1884 Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
1887 // If it's main() on Cygwin\Mingw32 we should align stack as well
1888 if (Fn->hasExternalLinkage() && Fn->getName() == "main" &&
1889 Subtarget->isTargetCygMing()) {
1890 BuildMI(MBB, MBBI, TII.get(X86::AND32ri), X86::ESP)
1891 .addReg(X86::ESP).addImm(-StackAlign);
1894 BuildMI(MBB, MBBI, TII.get(X86::MOV32ri), X86::EAX).addImm(StackAlign);
1895 BuildMI(MBB, MBBI, TII.get(X86::CALLpcrel32)).addExternalSymbol("_alloca");
1899 void X86RegisterInfo::emitEpilogue(MachineFunction &MF,
1900 MachineBasicBlock &MBB) const {
1901 const MachineFrameInfo *MFI = MF.getFrameInfo();
1902 const Function* Fn = MF.getFunction();
1903 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
1904 const X86Subtarget* Subtarget = &MF.getTarget().getSubtarget<X86Subtarget>();
1905 MachineBasicBlock::iterator MBBI = prior(MBB.end());
1906 unsigned RetOpcode = MBBI->getOpcode();
1908 switch (RetOpcode) {
1911 case X86::TCRETURNdi:
1912 case X86::TCRETURNri:
1913 case X86::TCRETURNri64:
1914 case X86::TCRETURNdi64:
1915 case X86::EH_RETURN:
1918 case X86::TAILJMPm: break; // These are ok
1920 assert(0 && "Can only insert epilog into returning blocks");
1923 // Get the number of bytes to allocate from the FrameInfo
1924 uint64_t StackSize = MFI->getStackSize();
1925 unsigned CSSize = X86FI->getCalleeSavedFrameSize();
1926 uint64_t NumBytes = StackSize - CSSize;
1930 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::POP64r : X86::POP32r), FramePtr);
1931 NumBytes -= SlotSize;
1934 // Skip the callee-saved pop instructions.
1935 while (MBBI != MBB.begin()) {
1936 MachineBasicBlock::iterator PI = prior(MBBI);
1937 unsigned Opc = PI->getOpcode();
1938 if (Opc != X86::POP32r && Opc != X86::POP64r && !TII.isTerminatorInstr(Opc))
1943 // If there is an ADD32ri or SUB32ri of ESP immediately before this
1944 // instruction, merge the two instructions.
1945 if (NumBytes || MFI->hasVarSizedObjects())
1946 mergeSPUpdatesUp(MBB, MBBI, StackPtr, &NumBytes);
1948 // If dynamic alloca is used, then reset esp to point to the last callee-saved
1949 // slot before popping them off! Also, if it's main() on Cygwin/Mingw32 we
1950 // aligned stack in the prologue, - revert stack changes back. Note: we're
1951 // assuming, that frame pointer was forced for main()
1952 if (MFI->hasVarSizedObjects() ||
1953 (Fn->hasExternalLinkage() && Fn->getName() == "main" &&
1954 Subtarget->isTargetCygMing())) {
1955 unsigned Opc = Is64Bit ? X86::LEA64r : X86::LEA32r;
1957 MachineInstr *MI = addRegOffset(BuildMI(TII.get(Opc), StackPtr),
1959 MBB.insert(MBBI, MI);
1961 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),StackPtr).
1967 // adjust stack pointer back: ESP += numbytes
1969 emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
1971 // We're returning from function via eh_return.
1972 if (RetOpcode == X86::EH_RETURN) {
1973 MBBI = prior(MBB.end());
1974 MachineOperand &DestAddr = MBBI->getOperand(0);
1975 assert(DestAddr.isRegister() && "Offset should be in register!");
1976 BuildMI(MBB, MBBI, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),StackPtr).
1977 addReg(DestAddr.getReg());
1978 // Tail call return: adjust the stack pointer and jump to callee
1979 } else if (RetOpcode == X86::TCRETURNri || RetOpcode == X86::TCRETURNdi ||
1980 RetOpcode== X86::TCRETURNri64 || RetOpcode == X86::TCRETURNdi64) {
1981 MBBI = prior(MBB.end());
1982 MachineOperand &JumpTarget = MBBI->getOperand(0);
1983 MachineOperand &StackAdjust = MBBI->getOperand(1);
1984 assert( StackAdjust.isImmediate() && "Expecting immediate value.");
1986 // Adjust stack pointer.
1987 int StackAdj = StackAdjust.getImm();
1988 int MaxTCDelta = X86FI->getTCReturnAddrDelta();
1990 assert(MaxTCDelta <= 0 && "MaxTCDelta should never be positive");
1991 // Incoporate the retaddr area.
1992 Offset = StackAdj-MaxTCDelta;
1993 assert(Offset >= 0 && "Offset should never be negative");
1995 // Check for possible merge with preceeding ADD instruction.
1996 Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
1997 emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, TII);
1999 // Jump to label or value in register.
2000 if (RetOpcode == X86::TCRETURNdi|| RetOpcode == X86::TCRETURNdi64)
2001 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPd)).
2002 addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
2003 else if (RetOpcode== X86::TCRETURNri64) {
2004 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPr64), JumpTarget.getReg());
2006 BuildMI(MBB, MBBI, TII.get(X86::TAILJMPr), JumpTarget.getReg());
2007 // Delete the pseudo instruction TCRETURN.
2009 } else if ((RetOpcode == X86::RET || RetOpcode == X86::RETI) &&
2010 (X86FI->getTCReturnAddrDelta() < 0)) {
2011 // Add the return addr area delta back since we are not tail calling.
2012 int delta = -1*X86FI->getTCReturnAddrDelta();
2013 MBBI = prior(MBB.end());
2014 // Check for possible merge with preceeding ADD instruction.
2015 delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
2016 emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, TII);
2020 unsigned X86RegisterInfo::getRARegister() const {
2022 return X86::RIP; // Should have dwarf #16
2024 return X86::EIP; // Should have dwarf #8
2027 unsigned X86RegisterInfo::getFrameRegister(MachineFunction &MF) const {
2028 return hasFP(MF) ? FramePtr : StackPtr;
2031 void X86RegisterInfo::getInitialFrameState(std::vector<MachineMove> &Moves)
2033 // Calculate amount of bytes used for return address storing
2034 int stackGrowth = (Is64Bit ? -8 : -4);
2036 // Initial state of the frame pointer is esp+4.
2037 MachineLocation Dst(MachineLocation::VirtualFP);
2038 MachineLocation Src(StackPtr, stackGrowth);
2039 Moves.push_back(MachineMove(0, Dst, Src));
2041 // Add return address to move list
2042 MachineLocation CSDst(StackPtr, stackGrowth);
2043 MachineLocation CSSrc(getRARegister());
2044 Moves.push_back(MachineMove(0, CSDst, CSSrc));
2047 unsigned X86RegisterInfo::getEHExceptionRegister() const {
2048 assert(0 && "What is the exception register");
2052 unsigned X86RegisterInfo::getEHHandlerRegister() const {
2053 assert(0 && "What is the exception handler register");
2058 unsigned getX86SubSuperRegister(unsigned Reg, MVT::ValueType VT, bool High) {
2060 default: return Reg;
2065 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2067 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2069 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2071 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2077 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2079 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2081 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2083 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2085 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2087 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2089 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2091 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2093 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2095 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2097 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2099 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2101 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2103 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2105 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2107 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2113 default: return Reg;
2114 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2116 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2118 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2120 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2122 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2124 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2126 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2128 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2130 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2132 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2134 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2136 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2138 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2140 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2142 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2144 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2149 default: return Reg;
2150 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2152 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2154 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2156 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2158 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2160 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2162 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2164 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2166 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2168 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2170 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2172 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2174 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2176 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2178 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2180 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2185 default: return Reg;
2186 case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
2188 case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
2190 case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
2192 case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
2194 case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
2196 case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
2198 case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
2200 case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
2202 case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
2204 case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
2206 case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
2208 case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
2210 case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
2212 case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
2214 case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
2216 case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
2225 #include "X86GenRegisterInfo.inc"