blob: 28e30769abc83a7cf5eecbf5ba171d296728c1b2 [file] [log] [blame]
Daniel Dunbar092a9dd2009-07-17 20:42:00 +00001//===-- X86AsmParser.cpp - Parse X86 assembly to MCInst instructions ------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9
Chris Lattner98986712010-01-14 22:21:20 +000010#include "llvm/Target/TargetAsmParser.h"
Daniel Dunbar4cb1e132009-07-18 23:03:22 +000011#include "X86.h"
Daniel Dunbar54074b52010-07-19 05:44:09 +000012#include "X86Subtarget.h"
Chris Lattner33d60d52010-09-22 04:11:10 +000013#include "llvm/Target/TargetRegistry.h"
14#include "llvm/Target/TargetAsmParser.h"
Kevin Enderby9c656452009-09-10 20:51:44 +000015#include "llvm/MC/MCStreamer.h"
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +000016#include "llvm/MC/MCExpr.h"
Daniel Dunbara027d222009-07-31 02:32:59 +000017#include "llvm/MC/MCInst.h"
Chris Lattnerc6ef2772010-01-22 01:44:57 +000018#include "llvm/MC/MCParser/MCAsmLexer.h"
19#include "llvm/MC/MCParser/MCAsmParser.h"
20#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
Chris Lattner33d60d52010-09-22 04:11:10 +000021#include "llvm/ADT/SmallString.h"
22#include "llvm/ADT/SmallVector.h"
23#include "llvm/ADT/StringExtras.h"
24#include "llvm/ADT/StringSwitch.h"
25#include "llvm/ADT/Twine.h"
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000026#include "llvm/Support/SourceMgr.h"
Daniel Dunbar09062b12010-08-12 00:55:42 +000027#include "llvm/Support/raw_ostream.h"
Daniel Dunbar092a9dd2009-07-17 20:42:00 +000028using namespace llvm;
29
30namespace {
Benjamin Kramerc6b79ac2009-07-31 11:35:26 +000031struct X86Operand;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000032
33class X86ATTAsmParser : public TargetAsmParser {
34 MCAsmParser &Parser;
Daniel Dunbard73ada72010-07-19 00:33:49 +000035 TargetMachine &TM;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000036
Daniel Dunbarf98bc632010-03-18 20:06:02 +000037protected:
38 unsigned Is64Bit : 1;
Michael J. Spencerc0c8df32010-10-09 11:00:50 +000039
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000040private:
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000041 MCAsmParser &getParser() const { return Parser; }
42
43 MCAsmLexer &getLexer() const { return Parser.getLexer(); }
44
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000045 bool Error(SMLoc L, const Twine &Msg) { return Parser.Error(L, Msg); }
46
Chris Lattner29ef9a22010-01-15 18:51:29 +000047 bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000048
Chris Lattner309264d2010-01-15 18:44:13 +000049 X86Operand *ParseOperand();
Chris Lattnereef6d782010-04-17 18:56:34 +000050 X86Operand *ParseMemOperand(unsigned SegReg, SMLoc StartLoc);
Kevin Enderby9c656452009-09-10 20:51:44 +000051
52 bool ParseDirectiveWord(unsigned Size, SMLoc L);
53
Chris Lattner7036f8b2010-09-29 01:42:58 +000054 bool MatchAndEmitInstruction(SMLoc IDLoc,
Chris Lattner7c51a312010-09-29 01:50:45 +000055 SmallVectorImpl<MCParsedAsmOperand*> &Operands,
Chris Lattner7036f8b2010-09-29 01:42:58 +000056 MCStreamer &Out);
Daniel Dunbar20927f22009-08-07 08:26:05 +000057
Daniel Dunbar54074b52010-07-19 05:44:09 +000058 /// @name Auto-generated Matcher Functions
59 /// {
Michael J. Spencerc0c8df32010-10-09 11:00:50 +000060
Chris Lattner0692ee62010-09-06 19:11:01 +000061#define GET_ASSEMBLER_HEADER
62#include "X86GenAsmMatcher.inc"
Michael J. Spencerc0c8df32010-10-09 11:00:50 +000063
Daniel Dunbar0e2771f2009-07-29 00:02:19 +000064 /// }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000065
66public:
Chris Lattner9a82e702010-10-30 04:57:14 +000067 X86ATTAsmParser(const Target &T, MCAsmParser &parser, TargetMachine &TM)
68 : TargetAsmParser(T), Parser(parser), TM(TM) {
Michael J. Spencerc0c8df32010-10-09 11:00:50 +000069
Daniel Dunbar54074b52010-07-19 05:44:09 +000070 // Initialize the set of available features.
71 setAvailableFeatures(ComputeAvailableFeatures(
72 &TM.getSubtarget<X86Subtarget>()));
73 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000074
Benjamin Kramer38e59892010-07-14 22:38:02 +000075 virtual bool ParseInstruction(StringRef Name, SMLoc NameLoc,
Chris Lattner98986712010-01-14 22:21:20 +000076 SmallVectorImpl<MCParsedAsmOperand*> &Operands);
Kevin Enderby9c656452009-09-10 20:51:44 +000077
78 virtual bool ParseDirective(AsmToken DirectiveID);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +000079};
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +000080
Daniel Dunbarf98bc632010-03-18 20:06:02 +000081class X86_32ATTAsmParser : public X86ATTAsmParser {
82public:
Chris Lattner9a82e702010-10-30 04:57:14 +000083 X86_32ATTAsmParser(const Target &T, MCAsmParser &Parser, TargetMachine &TM)
84 : X86ATTAsmParser(T, Parser, TM) {
Daniel Dunbarf98bc632010-03-18 20:06:02 +000085 Is64Bit = false;
86 }
87};
88
89class X86_64ATTAsmParser : public X86ATTAsmParser {
90public:
Chris Lattner9a82e702010-10-30 04:57:14 +000091 X86_64ATTAsmParser(const Target &T, MCAsmParser &Parser, TargetMachine &TM)
92 : X86ATTAsmParser(T, Parser, TM) {
Daniel Dunbarf98bc632010-03-18 20:06:02 +000093 Is64Bit = true;
94 }
95};
96
Chris Lattner37dfdec2009-07-29 06:33:53 +000097} // end anonymous namespace
98
Sean Callanane9b466d2010-01-23 00:40:33 +000099/// @name Auto-generated Match Functions
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000100/// {
Sean Callanane9b466d2010-01-23 00:40:33 +0000101
Chris Lattnerb8d6e982010-02-09 00:34:28 +0000102static unsigned MatchRegisterName(StringRef Name);
Sean Callanane9b466d2010-01-23 00:40:33 +0000103
104/// }
Chris Lattner37dfdec2009-07-29 06:33:53 +0000105
106namespace {
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000107
108/// X86Operand - Instances of this class represent a parsed X86 machine
109/// instruction.
Chris Lattner45220a82010-01-14 21:20:55 +0000110struct X86Operand : public MCParsedAsmOperand {
Chris Lattner1f19f0f2010-01-15 19:06:59 +0000111 enum KindTy {
Daniel Dunbar20927f22009-08-07 08:26:05 +0000112 Token,
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000113 Register,
114 Immediate,
115 Memory
116 } Kind;
117
Chris Lattner29ef9a22010-01-15 18:51:29 +0000118 SMLoc StartLoc, EndLoc;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000119
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000120 union {
121 struct {
Daniel Dunbar20927f22009-08-07 08:26:05 +0000122 const char *Data;
123 unsigned Length;
124 } Tok;
125
126 struct {
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000127 unsigned RegNo;
128 } Reg;
129
130 struct {
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000131 const MCExpr *Val;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000132 } Imm;
133
134 struct {
135 unsigned SegReg;
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000136 const MCExpr *Disp;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000137 unsigned BaseReg;
138 unsigned IndexReg;
139 unsigned Scale;
140 } Mem;
Daniel Dunbardbd692a2009-07-20 20:01:54 +0000141 };
Daniel Dunbar092a9dd2009-07-17 20:42:00 +0000142
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000143 X86Operand(KindTy K, SMLoc Start, SMLoc End)
Chris Lattner1f19f0f2010-01-15 19:06:59 +0000144 : Kind(K), StartLoc(Start), EndLoc(End) {}
Daniel Dunbarc918d602010-05-04 16:12:42 +0000145
Chris Lattner1f19f0f2010-01-15 19:06:59 +0000146 /// getStartLoc - Get the location of the first token of this operand.
147 SMLoc getStartLoc() const { return StartLoc; }
148 /// getEndLoc - Get the location of the last token of this operand.
149 SMLoc getEndLoc() const { return EndLoc; }
150
Daniel Dunbarb3cb6962010-08-11 06:37:04 +0000151 virtual void dump(raw_ostream &OS) const {}
152
Daniel Dunbar20927f22009-08-07 08:26:05 +0000153 StringRef getToken() const {
154 assert(Kind == Token && "Invalid access!");
155 return StringRef(Tok.Data, Tok.Length);
156 }
Daniel Dunbarc918d602010-05-04 16:12:42 +0000157 void setTokenValue(StringRef Value) {
158 assert(Kind == Token && "Invalid access!");
159 Tok.Data = Value.data();
160 Tok.Length = Value.size();
161 }
Daniel Dunbar20927f22009-08-07 08:26:05 +0000162
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000163 unsigned getReg() const {
164 assert(Kind == Register && "Invalid access!");
165 return Reg.RegNo;
166 }
Daniel Dunbara2edbab2009-07-28 20:47:52 +0000167
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000168 const MCExpr *getImm() const {
Daniel Dunbar022e2a82009-07-31 20:53:16 +0000169 assert(Kind == Immediate && "Invalid access!");
170 return Imm.Val;
171 }
172
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000173 const MCExpr *getMemDisp() const {
Daniel Dunbar022e2a82009-07-31 20:53:16 +0000174 assert(Kind == Memory && "Invalid access!");
175 return Mem.Disp;
176 }
177 unsigned getMemSegReg() const {
178 assert(Kind == Memory && "Invalid access!");
179 return Mem.SegReg;
180 }
181 unsigned getMemBaseReg() const {
182 assert(Kind == Memory && "Invalid access!");
183 return Mem.BaseReg;
184 }
185 unsigned getMemIndexReg() const {
186 assert(Kind == Memory && "Invalid access!");
187 return Mem.IndexReg;
188 }
189 unsigned getMemScale() const {
190 assert(Kind == Memory && "Invalid access!");
191 return Mem.Scale;
192 }
193
Daniel Dunbara3741fa2009-08-08 07:50:56 +0000194 bool isToken() const {return Kind == Token; }
Daniel Dunbar20927f22009-08-07 08:26:05 +0000195
196 bool isImm() const { return Kind == Immediate; }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000197
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000198 bool isImmSExti16i8() const {
Daniel Dunbar5fe63382009-08-09 07:20:21 +0000199 if (!isImm())
200 return false;
201
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000202 // If this isn't a constant expr, just assume it fits and let relaxation
203 // handle it.
204 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
205 if (!CE)
206 return true;
Daniel Dunbar5fe63382009-08-09 07:20:21 +0000207
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000208 // Otherwise, check the value is in a range that makes sense for this
209 // extension.
210 uint64_t Value = CE->getValue();
211 return (( Value <= 0x000000000000007FULL)||
212 (0x000000000000FF80ULL <= Value && Value <= 0x000000000000FFFFULL)||
213 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
Daniel Dunbar5fe63382009-08-09 07:20:21 +0000214 }
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000215 bool isImmSExti32i8() const {
Daniel Dunbar1fe591d2010-05-20 20:20:39 +0000216 if (!isImm())
217 return false;
218
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000219 // If this isn't a constant expr, just assume it fits and let relaxation
220 // handle it.
221 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
222 if (!CE)
223 return true;
Daniel Dunbar1fe591d2010-05-20 20:20:39 +0000224
Daniel Dunbar62e4c672010-05-22 21:02:33 +0000225 // Otherwise, check the value is in a range that makes sense for this
226 // extension.
227 uint64_t Value = CE->getValue();
228 return (( Value <= 0x000000000000007FULL)||
229 (0x00000000FFFFFF80ULL <= Value && Value <= 0x00000000FFFFFFFFULL)||
230 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
231 }
232 bool isImmSExti64i8() const {
233 if (!isImm())
234 return false;
235
236 // If this isn't a constant expr, just assume it fits and let relaxation
237 // handle it.
238 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
239 if (!CE)
240 return true;
241
242 // Otherwise, check the value is in a range that makes sense for this
243 // extension.
244 uint64_t Value = CE->getValue();
245 return (( Value <= 0x000000000000007FULL)||
246 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
247 }
248 bool isImmSExti64i32() const {
249 if (!isImm())
250 return false;
251
252 // If this isn't a constant expr, just assume it fits and let relaxation
253 // handle it.
254 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
255 if (!CE)
256 return true;
257
258 // Otherwise, check the value is in a range that makes sense for this
259 // extension.
260 uint64_t Value = CE->getValue();
261 return (( Value <= 0x000000007FFFFFFFULL)||
262 (0xFFFFFFFF80000000ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
Daniel Dunbar1fe591d2010-05-20 20:20:39 +0000263 }
264
Daniel Dunbar20927f22009-08-07 08:26:05 +0000265 bool isMem() const { return Kind == Memory; }
266
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000267 bool isAbsMem() const {
268 return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
Daniel Dunbar7b9147a2010-02-02 21:44:16 +0000269 !getMemIndexReg() && getMemScale() == 1;
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000270 }
271
Daniel Dunbar20927f22009-08-07 08:26:05 +0000272 bool isReg() const { return Kind == Register; }
273
Daniel Dunbar9c60f532010-02-13 00:17:21 +0000274 void addExpr(MCInst &Inst, const MCExpr *Expr) const {
275 // Add as immediates when possible.
276 if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr))
277 Inst.addOperand(MCOperand::CreateImm(CE->getValue()));
278 else
279 Inst.addOperand(MCOperand::CreateExpr(Expr));
280 }
281
Daniel Dunbar5c468e32009-08-10 21:00:45 +0000282 void addRegOperands(MCInst &Inst, unsigned N) const {
Daniel Dunbar20927f22009-08-07 08:26:05 +0000283 assert(N == 1 && "Invalid number of operands!");
284 Inst.addOperand(MCOperand::CreateReg(getReg()));
285 }
286
Daniel Dunbar5c468e32009-08-10 21:00:45 +0000287 void addImmOperands(MCInst &Inst, unsigned N) const {
Daniel Dunbar20927f22009-08-07 08:26:05 +0000288 assert(N == 1 && "Invalid number of operands!");
Daniel Dunbar9c60f532010-02-13 00:17:21 +0000289 addExpr(Inst, getImm());
Daniel Dunbar20927f22009-08-07 08:26:05 +0000290 }
291
Daniel Dunbar5c468e32009-08-10 21:00:45 +0000292 void addMemOperands(MCInst &Inst, unsigned N) const {
Daniel Dunbarec2b1f12010-01-30 00:24:00 +0000293 assert((N == 5) && "Invalid number of operands!");
Daniel Dunbar20927f22009-08-07 08:26:05 +0000294 Inst.addOperand(MCOperand::CreateReg(getMemBaseReg()));
295 Inst.addOperand(MCOperand::CreateImm(getMemScale()));
296 Inst.addOperand(MCOperand::CreateReg(getMemIndexReg()));
Daniel Dunbar9c60f532010-02-13 00:17:21 +0000297 addExpr(Inst, getMemDisp());
Daniel Dunbarec2b1f12010-01-30 00:24:00 +0000298 Inst.addOperand(MCOperand::CreateReg(getMemSegReg()));
299 }
Daniel Dunbar20927f22009-08-07 08:26:05 +0000300
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000301 void addAbsMemOperands(MCInst &Inst, unsigned N) const {
302 assert((N == 1) && "Invalid number of operands!");
303 Inst.addOperand(MCOperand::CreateExpr(getMemDisp()));
304 }
305
Chris Lattnerb4307b32010-01-15 19:28:38 +0000306 static X86Operand *CreateToken(StringRef Str, SMLoc Loc) {
307 X86Operand *Res = new X86Operand(Token, Loc, Loc);
Chris Lattner29ef9a22010-01-15 18:51:29 +0000308 Res->Tok.Data = Str.data();
309 Res->Tok.Length = Str.size();
Daniel Dunbar20927f22009-08-07 08:26:05 +0000310 return Res;
311 }
312
Chris Lattner29ef9a22010-01-15 18:51:29 +0000313 static X86Operand *CreateReg(unsigned RegNo, SMLoc StartLoc, SMLoc EndLoc) {
Chris Lattner1f19f0f2010-01-15 19:06:59 +0000314 X86Operand *Res = new X86Operand(Register, StartLoc, EndLoc);
Chris Lattner29ef9a22010-01-15 18:51:29 +0000315 Res->Reg.RegNo = RegNo;
Chris Lattner29ef9a22010-01-15 18:51:29 +0000316 return Res;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000317 }
Daniel Dunbar20927f22009-08-07 08:26:05 +0000318
Chris Lattnerb4307b32010-01-15 19:28:38 +0000319 static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc){
320 X86Operand *Res = new X86Operand(Immediate, StartLoc, EndLoc);
Chris Lattner29ef9a22010-01-15 18:51:29 +0000321 Res->Imm.Val = Val;
322 return Res;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000323 }
Daniel Dunbar20927f22009-08-07 08:26:05 +0000324
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000325 /// Create an absolute memory operand.
326 static X86Operand *CreateMem(const MCExpr *Disp, SMLoc StartLoc,
327 SMLoc EndLoc) {
328 X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc);
329 Res->Mem.SegReg = 0;
330 Res->Mem.Disp = Disp;
331 Res->Mem.BaseReg = 0;
332 Res->Mem.IndexReg = 0;
Daniel Dunbar7b9147a2010-02-02 21:44:16 +0000333 Res->Mem.Scale = 1;
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000334 return Res;
335 }
336
337 /// Create a generalized memory operand.
Chris Lattner309264d2010-01-15 18:44:13 +0000338 static X86Operand *CreateMem(unsigned SegReg, const MCExpr *Disp,
339 unsigned BaseReg, unsigned IndexReg,
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000340 unsigned Scale, SMLoc StartLoc, SMLoc EndLoc) {
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000341 // We should never just have a displacement, that should be parsed as an
342 // absolute memory operand.
Daniel Dunbarc09e4112009-07-31 22:22:54 +0000343 assert((SegReg || BaseReg || IndexReg) && "Invalid memory operand!");
344
Daniel Dunbar022e2a82009-07-31 20:53:16 +0000345 // The scale should always be one of {1,2,4,8}.
346 assert(((Scale == 1 || Scale == 2 || Scale == 4 || Scale == 8)) &&
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000347 "Invalid scale!");
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000348 X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc);
Chris Lattner29ef9a22010-01-15 18:51:29 +0000349 Res->Mem.SegReg = SegReg;
350 Res->Mem.Disp = Disp;
351 Res->Mem.BaseReg = BaseReg;
352 Res->Mem.IndexReg = IndexReg;
353 Res->Mem.Scale = Scale;
354 return Res;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000355 }
356};
Daniel Dunbara3af3702009-07-20 18:55:04 +0000357
Chris Lattner37dfdec2009-07-29 06:33:53 +0000358} // end anonymous namespace.
Daniel Dunbara2edbab2009-07-28 20:47:52 +0000359
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000360
Chris Lattner29ef9a22010-01-15 18:51:29 +0000361bool X86ATTAsmParser::ParseRegister(unsigned &RegNo,
362 SMLoc &StartLoc, SMLoc &EndLoc) {
Chris Lattner23075742010-01-15 18:27:19 +0000363 RegNo = 0;
Sean Callanan18b83232010-01-19 21:44:56 +0000364 const AsmToken &TokPercent = Parser.getTok();
Kevin Enderby7b4608d2009-09-03 17:15:07 +0000365 assert(TokPercent.is(AsmToken::Percent) && "Invalid token kind!");
Chris Lattner29ef9a22010-01-15 18:51:29 +0000366 StartLoc = TokPercent.getLoc();
Sean Callananb9a25b72010-01-19 20:27:46 +0000367 Parser.Lex(); // Eat percent token.
Kevin Enderby7b4608d2009-09-03 17:15:07 +0000368
Sean Callanan18b83232010-01-19 21:44:56 +0000369 const AsmToken &Tok = Parser.getTok();
Kevin Enderby0d6cd002009-09-16 17:18:29 +0000370 if (Tok.isNot(AsmToken::Identifier))
371 return Error(Tok.getLoc(), "invalid register name");
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000372
Daniel Dunbar0e2771f2009-07-29 00:02:19 +0000373 // FIXME: Validate register for the current architecture; we have to do
374 // validation later, so maybe there is no need for this here.
Kevin Enderby7b4608d2009-09-03 17:15:07 +0000375 RegNo = MatchRegisterName(Tok.getString());
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000376
Chris Lattner33d60d52010-09-22 04:11:10 +0000377 // If the match failed, try the register name as lowercase.
378 if (RegNo == 0)
379 RegNo = MatchRegisterName(LowercaseString(Tok.getString()));
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000380
Bruno Cardoso Lopes3c8e1be2010-07-24 00:06:39 +0000381 // FIXME: This should be done using Requires<In32BitMode> and
382 // Requires<In64BitMode> so "eiz" usage in 64-bit instructions
383 // can be also checked.
384 if (RegNo == X86::RIZ && !Is64Bit)
385 return Error(Tok.getLoc(), "riz register in 64-bit mode only");
386
Chris Lattner33d60d52010-09-22 04:11:10 +0000387 // Parse "%st" as "%st(0)" and "%st(1)", which is multiple tokens.
388 if (RegNo == 0 && (Tok.getString() == "st" || Tok.getString() == "ST")) {
Chris Lattnere16b0fc2010-02-09 00:49:22 +0000389 RegNo = X86::ST0;
390 EndLoc = Tok.getLoc();
391 Parser.Lex(); // Eat 'st'
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000392
Chris Lattnere16b0fc2010-02-09 00:49:22 +0000393 // Check to see if we have '(4)' after %st.
394 if (getLexer().isNot(AsmToken::LParen))
395 return false;
396 // Lex the paren.
397 getParser().Lex();
398
399 const AsmToken &IntTok = Parser.getTok();
400 if (IntTok.isNot(AsmToken::Integer))
401 return Error(IntTok.getLoc(), "expected stack index");
402 switch (IntTok.getIntVal()) {
403 case 0: RegNo = X86::ST0; break;
404 case 1: RegNo = X86::ST1; break;
405 case 2: RegNo = X86::ST2; break;
406 case 3: RegNo = X86::ST3; break;
407 case 4: RegNo = X86::ST4; break;
408 case 5: RegNo = X86::ST5; break;
409 case 6: RegNo = X86::ST6; break;
410 case 7: RegNo = X86::ST7; break;
411 default: return Error(IntTok.getLoc(), "invalid stack index");
412 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000413
Chris Lattnere16b0fc2010-02-09 00:49:22 +0000414 if (getParser().Lex().isNot(AsmToken::RParen))
415 return Error(Parser.getTok().getLoc(), "expected ')'");
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000416
Chris Lattnere16b0fc2010-02-09 00:49:22 +0000417 EndLoc = Tok.getLoc();
418 Parser.Lex(); // Eat ')'
419 return false;
420 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000421
Chris Lattner645b2092010-06-24 07:29:18 +0000422 // If this is "db[0-7]", match it as an alias
423 // for dr[0-7].
424 if (RegNo == 0 && Tok.getString().size() == 3 &&
425 Tok.getString().startswith("db")) {
426 switch (Tok.getString()[2]) {
427 case '0': RegNo = X86::DR0; break;
428 case '1': RegNo = X86::DR1; break;
429 case '2': RegNo = X86::DR2; break;
430 case '3': RegNo = X86::DR3; break;
431 case '4': RegNo = X86::DR4; break;
432 case '5': RegNo = X86::DR5; break;
433 case '6': RegNo = X86::DR6; break;
434 case '7': RegNo = X86::DR7; break;
435 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000436
Chris Lattner645b2092010-06-24 07:29:18 +0000437 if (RegNo != 0) {
438 EndLoc = Tok.getLoc();
439 Parser.Lex(); // Eat it.
440 return false;
441 }
442 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000443
Daniel Dunbar245f0582009-08-08 21:22:41 +0000444 if (RegNo == 0)
Daniel Dunbar0e2771f2009-07-29 00:02:19 +0000445 return Error(Tok.getLoc(), "invalid register name");
446
Chris Lattner29ef9a22010-01-15 18:51:29 +0000447 EndLoc = Tok.getLoc();
Sean Callananb9a25b72010-01-19 20:27:46 +0000448 Parser.Lex(); // Eat identifier token.
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000449 return false;
Daniel Dunbar092a9dd2009-07-17 20:42:00 +0000450}
451
Chris Lattner309264d2010-01-15 18:44:13 +0000452X86Operand *X86ATTAsmParser::ParseOperand() {
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000453 switch (getLexer().getKind()) {
454 default:
Chris Lattnereef6d782010-04-17 18:56:34 +0000455 // Parse a memory operand with no segment register.
456 return ParseMemOperand(0, Parser.getTok().getLoc());
Chris Lattner23075742010-01-15 18:27:19 +0000457 case AsmToken::Percent: {
Chris Lattnereef6d782010-04-17 18:56:34 +0000458 // Read the register.
Chris Lattner23075742010-01-15 18:27:19 +0000459 unsigned RegNo;
Chris Lattner29ef9a22010-01-15 18:51:29 +0000460 SMLoc Start, End;
461 if (ParseRegister(RegNo, Start, End)) return 0;
Bruno Cardoso Lopes3c8e1be2010-07-24 00:06:39 +0000462 if (RegNo == X86::EIZ || RegNo == X86::RIZ) {
463 Error(Start, "eiz and riz can only be used as index registers");
464 return 0;
465 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000466
Chris Lattnereef6d782010-04-17 18:56:34 +0000467 // If this is a segment register followed by a ':', then this is the start
468 // of a memory reference, otherwise this is a normal register reference.
469 if (getLexer().isNot(AsmToken::Colon))
470 return X86Operand::CreateReg(RegNo, Start, End);
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000471
472
Chris Lattnereef6d782010-04-17 18:56:34 +0000473 getParser().Lex(); // Eat the colon.
474 return ParseMemOperand(RegNo, Start);
Chris Lattner23075742010-01-15 18:27:19 +0000475 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000476 case AsmToken::Dollar: {
477 // $42 -> immediate.
Sean Callanan18b83232010-01-19 21:44:56 +0000478 SMLoc Start = Parser.getTok().getLoc(), End;
Sean Callananb9a25b72010-01-19 20:27:46 +0000479 Parser.Lex();
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000480 const MCExpr *Val;
Chris Lattner54482b42010-01-15 19:39:23 +0000481 if (getParser().ParseExpression(Val, End))
Chris Lattner309264d2010-01-15 18:44:13 +0000482 return 0;
Chris Lattnerb4307b32010-01-15 19:28:38 +0000483 return X86Operand::CreateImm(Val, Start, End);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000484 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000485 }
Daniel Dunbardbd692a2009-07-20 20:01:54 +0000486}
487
Chris Lattnereef6d782010-04-17 18:56:34 +0000488/// ParseMemOperand: segment: disp(basereg, indexreg, scale). The '%ds:' prefix
489/// has already been parsed if present.
490X86Operand *X86ATTAsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000491
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000492 // We have to disambiguate a parenthesized expression "(4+5)" from the start
493 // of a memory operand with a missing displacement "(%ebx)" or "(,%eax)". The
Chris Lattner75f265f2010-01-24 01:07:33 +0000494 // only way to do this without lookahead is to eat the '(' and see what is
495 // after it.
Daniel Dunbar8c2eebe2009-08-31 08:08:38 +0000496 const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext());
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000497 if (getLexer().isNot(AsmToken::LParen)) {
Chris Lattner54482b42010-01-15 19:39:23 +0000498 SMLoc ExprEnd;
499 if (getParser().ParseExpression(Disp, ExprEnd)) return 0;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000500
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000501 // After parsing the base expression we could either have a parenthesized
502 // memory address or not. If not, return now. If so, eat the (.
503 if (getLexer().isNot(AsmToken::LParen)) {
Daniel Dunbarc09e4112009-07-31 22:22:54 +0000504 // Unless we have a segment register, treat this as an immediate.
Chris Lattner309264d2010-01-15 18:44:13 +0000505 if (SegReg == 0)
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000506 return X86Operand::CreateMem(Disp, MemStart, ExprEnd);
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000507 return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000508 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000509
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000510 // Eat the '('.
Sean Callananb9a25b72010-01-19 20:27:46 +0000511 Parser.Lex();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000512 } else {
513 // Okay, we have a '('. We don't know if this is an expression or not, but
514 // so we have to eat the ( to see beyond it.
Sean Callanan18b83232010-01-19 21:44:56 +0000515 SMLoc LParenLoc = Parser.getTok().getLoc();
Sean Callananb9a25b72010-01-19 20:27:46 +0000516 Parser.Lex(); // Eat the '('.
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000517
Kevin Enderby7b4608d2009-09-03 17:15:07 +0000518 if (getLexer().is(AsmToken::Percent) || getLexer().is(AsmToken::Comma)) {
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000519 // Nothing to do here, fall into the code below with the '(' part of the
520 // memory operand consumed.
521 } else {
Chris Lattnerb4307b32010-01-15 19:28:38 +0000522 SMLoc ExprEnd;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000523
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000524 // It must be an parenthesized expression, parse it now.
Chris Lattnerb4307b32010-01-15 19:28:38 +0000525 if (getParser().ParseParenExpression(Disp, ExprEnd))
Chris Lattner309264d2010-01-15 18:44:13 +0000526 return 0;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000527
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000528 // After parsing the base expression we could either have a parenthesized
529 // memory address or not. If not, return now. If so, eat the (.
530 if (getLexer().isNot(AsmToken::LParen)) {
Daniel Dunbarc09e4112009-07-31 22:22:54 +0000531 // Unless we have a segment register, treat this as an immediate.
Chris Lattner309264d2010-01-15 18:44:13 +0000532 if (SegReg == 0)
Daniel Dunbarb834f5d2010-01-30 01:02:48 +0000533 return X86Operand::CreateMem(Disp, LParenLoc, ExprEnd);
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000534 return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000535 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000536
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000537 // Eat the '('.
Sean Callananb9a25b72010-01-19 20:27:46 +0000538 Parser.Lex();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000539 }
540 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000541
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000542 // If we reached here, then we just ate the ( of the memory operand. Process
543 // the rest of the memory operand.
Daniel Dunbar022e2a82009-07-31 20:53:16 +0000544 unsigned BaseReg = 0, IndexReg = 0, Scale = 1;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000545
Chris Lattner29ef9a22010-01-15 18:51:29 +0000546 if (getLexer().is(AsmToken::Percent)) {
547 SMLoc L;
548 if (ParseRegister(BaseReg, L, L)) return 0;
Bruno Cardoso Lopes3c8e1be2010-07-24 00:06:39 +0000549 if (BaseReg == X86::EIZ || BaseReg == X86::RIZ) {
550 Error(L, "eiz and riz can only be used as index registers");
551 return 0;
552 }
Chris Lattner29ef9a22010-01-15 18:51:29 +0000553 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000554
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000555 if (getLexer().is(AsmToken::Comma)) {
Sean Callananb9a25b72010-01-19 20:27:46 +0000556 Parser.Lex(); // Eat the comma.
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000557
558 // Following the comma we should have either an index register, or a scale
559 // value. We don't support the later form, but we want to parse it
560 // correctly.
561 //
562 // Not that even though it would be completely consistent to support syntax
Bruno Cardoso Lopes3c8e1be2010-07-24 00:06:39 +0000563 // like "1(%eax,,1)", the assembler doesn't. Use "eiz" or "riz" for this.
Kevin Enderby7b4608d2009-09-03 17:15:07 +0000564 if (getLexer().is(AsmToken::Percent)) {
Chris Lattner29ef9a22010-01-15 18:51:29 +0000565 SMLoc L;
566 if (ParseRegister(IndexReg, L, L)) return 0;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000567
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000568 if (getLexer().isNot(AsmToken::RParen)) {
569 // Parse the scale amount:
570 // ::= ',' [scale-expression]
Chris Lattner309264d2010-01-15 18:44:13 +0000571 if (getLexer().isNot(AsmToken::Comma)) {
Sean Callanan18b83232010-01-19 21:44:56 +0000572 Error(Parser.getTok().getLoc(),
Chris Lattner309264d2010-01-15 18:44:13 +0000573 "expected comma in scale expression");
574 return 0;
575 }
Sean Callananb9a25b72010-01-19 20:27:46 +0000576 Parser.Lex(); // Eat the comma.
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000577
578 if (getLexer().isNot(AsmToken::RParen)) {
Sean Callanan18b83232010-01-19 21:44:56 +0000579 SMLoc Loc = Parser.getTok().getLoc();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000580
581 int64_t ScaleVal;
582 if (getParser().ParseAbsoluteExpression(ScaleVal))
Chris Lattner309264d2010-01-15 18:44:13 +0000583 return 0;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000584
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000585 // Validate the scale amount.
Chris Lattner309264d2010-01-15 18:44:13 +0000586 if (ScaleVal != 1 && ScaleVal != 2 && ScaleVal != 4 && ScaleVal != 8){
587 Error(Loc, "scale factor in address must be 1, 2, 4 or 8");
588 return 0;
589 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000590 Scale = (unsigned)ScaleVal;
591 }
592 }
593 } else if (getLexer().isNot(AsmToken::RParen)) {
Daniel Dunbaree910252010-08-24 19:13:38 +0000594 // A scale amount without an index is ignored.
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000595 // index.
Sean Callanan18b83232010-01-19 21:44:56 +0000596 SMLoc Loc = Parser.getTok().getLoc();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000597
598 int64_t Value;
599 if (getParser().ParseAbsoluteExpression(Value))
Chris Lattner309264d2010-01-15 18:44:13 +0000600 return 0;
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000601
Daniel Dunbaree910252010-08-24 19:13:38 +0000602 if (Value != 1)
603 Warning(Loc, "scale factor without index register is ignored");
604 Scale = 1;
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000605 }
606 }
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000607
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000608 // Ok, we've eaten the memory operand, verify we have a ')' and eat it too.
Chris Lattner309264d2010-01-15 18:44:13 +0000609 if (getLexer().isNot(AsmToken::RParen)) {
Sean Callanan18b83232010-01-19 21:44:56 +0000610 Error(Parser.getTok().getLoc(), "unexpected token in memory operand");
Chris Lattner309264d2010-01-15 18:44:13 +0000611 return 0;
612 }
Sean Callanan18b83232010-01-19 21:44:56 +0000613 SMLoc MemEnd = Parser.getTok().getLoc();
Sean Callananb9a25b72010-01-19 20:27:46 +0000614 Parser.Lex(); // Eat the ')'.
Bruno Cardoso Lopesf64a7d42010-07-23 22:15:26 +0000615
Chris Lattner0a3c5a52010-01-15 19:33:43 +0000616 return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale,
617 MemStart, MemEnd);
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000618}
619
Chris Lattner98986712010-01-14 22:21:20 +0000620bool X86ATTAsmParser::
Benjamin Kramer38e59892010-07-14 22:38:02 +0000621ParseInstruction(StringRef Name, SMLoc NameLoc,
Chris Lattner98986712010-01-14 22:21:20 +0000622 SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
Chris Lattner693173f2010-10-30 19:23:13 +0000623 // FIXME: This is not correct at all.
624 if (Name == "movzx") Name = "movzb";
625
626 StringRef PatchedName = Name;
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000627
628 // FIXME: Hack to recognize cmp<comparison code>{ss,sd,ps,pd}.
629 const MCExpr *ExtraImmOp = 0;
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000630 if ((PatchedName.startswith("cmp") || PatchedName.startswith("vcmp")) &&
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000631 (PatchedName.endswith("ss") || PatchedName.endswith("sd") ||
632 PatchedName.endswith("ps") || PatchedName.endswith("pd"))) {
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000633 bool IsVCMP = PatchedName.startswith("vcmp");
634 unsigned SSECCIdx = IsVCMP ? 4 : 3;
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000635 unsigned SSEComparisonCode = StringSwitch<unsigned>(
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000636 PatchedName.slice(SSECCIdx, PatchedName.size() - 2))
Bruno Cardoso Lopescc69e132010-07-07 22:24:03 +0000637 .Case("eq", 0)
638 .Case("lt", 1)
639 .Case("le", 2)
640 .Case("unord", 3)
641 .Case("neq", 4)
642 .Case("nlt", 5)
643 .Case("nle", 6)
644 .Case("ord", 7)
645 .Case("eq_uq", 8)
646 .Case("nge", 9)
647 .Case("ngt", 0x0A)
648 .Case("false", 0x0B)
649 .Case("neq_oq", 0x0C)
650 .Case("ge", 0x0D)
651 .Case("gt", 0x0E)
652 .Case("true", 0x0F)
653 .Case("eq_os", 0x10)
654 .Case("lt_oq", 0x11)
655 .Case("le_oq", 0x12)
656 .Case("unord_s", 0x13)
657 .Case("neq_us", 0x14)
658 .Case("nlt_uq", 0x15)
659 .Case("nle_uq", 0x16)
660 .Case("ord_s", 0x17)
661 .Case("eq_us", 0x18)
662 .Case("nge_uq", 0x19)
663 .Case("ngt_uq", 0x1A)
664 .Case("false_os", 0x1B)
665 .Case("neq_os", 0x1C)
666 .Case("ge_oq", 0x1D)
667 .Case("gt_oq", 0x1E)
668 .Case("true_us", 0x1F)
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000669 .Default(~0U);
670 if (SSEComparisonCode != ~0U) {
671 ExtraImmOp = MCConstantExpr::Create(SSEComparisonCode,
672 getParser().getContext());
673 if (PatchedName.endswith("ss")) {
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000674 PatchedName = IsVCMP ? "vcmpss" : "cmpss";
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000675 } else if (PatchedName.endswith("sd")) {
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000676 PatchedName = IsVCMP ? "vcmpsd" : "cmpsd";
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000677 } else if (PatchedName.endswith("ps")) {
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000678 PatchedName = IsVCMP ? "vcmpps" : "cmpps";
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000679 } else {
680 assert(PatchedName.endswith("pd") && "Unexpected mnemonic!");
Bruno Cardoso Lopes428256b2010-06-23 21:10:57 +0000681 PatchedName = IsVCMP ? "vcmppd" : "cmppd";
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000682 }
683 }
684 }
Bruno Cardoso Lopesf528d2b2010-07-23 18:41:12 +0000685
686 // FIXME: Hack to recognize vpclmul<src1_quadword, src2_quadword>dq
687 if (PatchedName.startswith("vpclmul")) {
688 unsigned CLMULQuadWordSelect = StringSwitch<unsigned>(
689 PatchedName.slice(7, PatchedName.size() - 2))
690 .Case("lqlq", 0x00) // src1[63:0], src2[63:0]
691 .Case("hqlq", 0x01) // src1[127:64], src2[63:0]
692 .Case("lqhq", 0x10) // src1[63:0], src2[127:64]
693 .Case("hqhq", 0x11) // src1[127:64], src2[127:64]
694 .Default(~0U);
695 if (CLMULQuadWordSelect != ~0U) {
696 ExtraImmOp = MCConstantExpr::Create(CLMULQuadWordSelect,
697 getParser().getContext());
698 assert(PatchedName.endswith("dq") && "Unexpected mnemonic!");
699 PatchedName = "vpclmulqdq";
700 }
701 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000702
Daniel Dunbar1b6c0602010-02-10 21:19:28 +0000703 Operands.push_back(X86Operand::CreateToken(PatchedName, NameLoc));
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000704
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000705 if (ExtraImmOp)
706 Operands.push_back(X86Operand::CreateImm(ExtraImmOp, NameLoc, NameLoc));
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000707
708
Chris Lattner2544f422010-09-08 05:17:37 +0000709 // Determine whether this is an instruction prefix.
710 bool isPrefix =
Chris Lattner693173f2010-10-30 19:23:13 +0000711 Name == "lock" || Name == "rep" ||
712 Name == "repe" || Name == "repz" ||
713 Name == "repne" || Name == "repnz";
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000714
715
Chris Lattner2544f422010-09-08 05:17:37 +0000716 // This does the actual operand parsing. Don't parse any more if we have a
717 // prefix juxtaposed with an operation like "lock incl 4(%rax)", because we
718 // just want to parse the "lock" as the first instruction and the "incl" as
719 // the next one.
720 if (getLexer().isNot(AsmToken::EndOfStatement) && !isPrefix) {
Daniel Dunbar0db68f42009-08-11 05:00:25 +0000721
722 // Parse '*' modifier.
723 if (getLexer().is(AsmToken::Star)) {
Sean Callanan18b83232010-01-19 21:44:56 +0000724 SMLoc Loc = Parser.getTok().getLoc();
Chris Lattnerb4307b32010-01-15 19:28:38 +0000725 Operands.push_back(X86Operand::CreateToken("*", Loc));
Sean Callananb9a25b72010-01-19 20:27:46 +0000726 Parser.Lex(); // Eat the star.
Daniel Dunbar0db68f42009-08-11 05:00:25 +0000727 }
728
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000729 // Read the first operand.
Chris Lattner309264d2010-01-15 18:44:13 +0000730 if (X86Operand *Op = ParseOperand())
731 Operands.push_back(Op);
Chris Lattnercbf8a982010-09-11 16:18:25 +0000732 else {
733 Parser.EatToEndOfStatement();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000734 return true;
Chris Lattnercbf8a982010-09-11 16:18:25 +0000735 }
Daniel Dunbar39e2dd72010-05-25 19:49:32 +0000736
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000737 while (getLexer().is(AsmToken::Comma)) {
Sean Callananb9a25b72010-01-19 20:27:46 +0000738 Parser.Lex(); // Eat the comma.
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000739
740 // Parse and remember the operand.
Chris Lattner309264d2010-01-15 18:44:13 +0000741 if (X86Operand *Op = ParseOperand())
742 Operands.push_back(Op);
Chris Lattnercbf8a982010-09-11 16:18:25 +0000743 else {
744 Parser.EatToEndOfStatement();
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000745 return true;
Chris Lattnercbf8a982010-09-11 16:18:25 +0000746 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000747 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000748
Chris Lattnercbf8a982010-09-11 16:18:25 +0000749 if (getLexer().isNot(AsmToken::EndOfStatement)) {
750 Parser.EatToEndOfStatement();
Chris Lattner2544f422010-09-08 05:17:37 +0000751 return TokError("unexpected token in argument list");
Chris Lattnercbf8a982010-09-11 16:18:25 +0000752 }
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000753 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000754
Chris Lattner2544f422010-09-08 05:17:37 +0000755 if (getLexer().is(AsmToken::EndOfStatement))
756 Parser.Lex(); // Consume the EndOfStatement
Daniel Dunbar16cdcb32009-07-28 22:40:46 +0000757
Chris Lattner508fc472010-10-05 21:09:45 +0000758 // Hack to allow 'movq <largeimm>, <reg>' as an alias for movabsq.
759 if ((Name == "movq" || Name == "mov") && Operands.size() == 3 &&
760 static_cast<X86Operand*>(Operands[2])->isReg() &&
761 static_cast<X86Operand*>(Operands[1])->isImm() &&
762 !static_cast<X86Operand*>(Operands[1])->isImmSExti64i32()) {
763 delete Operands[0];
764 Operands[0] = X86Operand::CreateToken("movabsq", NameLoc);
765 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000766
Chris Lattnere9e16a32010-09-15 04:33:27 +0000767 // FIXME: Hack to handle recognize s{hr,ar,hl} $1, <op>. Canonicalize to
Chris Lattneree211d02010-09-11 16:32:12 +0000768 // "shift <op>".
Daniel Dunbard5e77052010-03-13 00:47:29 +0000769 if ((Name.startswith("shr") || Name.startswith("sar") ||
Chris Lattner1a1ecc92010-10-30 18:13:10 +0000770 Name.startswith("shl") || Name.startswith("sal")) &&
Chris Lattner47ab90b2010-09-06 18:32:06 +0000771 Operands.size() == 3) {
772 X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]);
773 if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) &&
774 cast<MCConstantExpr>(Op1->getImm())->getValue() == 1) {
775 delete Operands[1];
776 Operands.erase(Operands.begin() + 1);
777 }
Daniel Dunbarf2de13f2010-03-20 22:36:38 +0000778 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000779
Chris Lattnere9e16a32010-09-15 04:33:27 +0000780 // FIXME: Hack to handle recognize "rc[lr] <op>" -> "rcl $1, <op>".
781 if ((Name.startswith("rcl") || Name.startswith("rcr")) &&
782 Operands.size() == 2) {
783 const MCExpr *One = MCConstantExpr::Create(1, getParser().getContext());
784 Operands.push_back(X86Operand::CreateImm(One, NameLoc, NameLoc));
785 std::swap(Operands[1], Operands[2]);
786 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000787
Chris Lattnercfad5642010-09-15 04:37:18 +0000788 // FIXME: Hack to handle recognize "sh[lr]d op,op" -> "shld $1, op,op".
789 if ((Name.startswith("shld") || Name.startswith("shrd")) &&
790 Operands.size() == 3) {
791 const MCExpr *One = MCConstantExpr::Create(1, getParser().getContext());
792 Operands.insert(Operands.begin()+1,
793 X86Operand::CreateImm(One, NameLoc, NameLoc));
794 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000795
Daniel Dunbard5e77052010-03-13 00:47:29 +0000796
Chris Lattneree211d02010-09-11 16:32:12 +0000797 // FIXME: Hack to handle recognize "in[bwl] <op>". Canonicalize it to
798 // "inb <op>, %al".
799 if ((Name == "inb" || Name == "inw" || Name == "inl") &&
800 Operands.size() == 2) {
801 unsigned Reg;
802 if (Name[2] == 'b')
803 Reg = MatchRegisterName("al");
804 else if (Name[2] == 'w')
805 Reg = MatchRegisterName("ax");
806 else
807 Reg = MatchRegisterName("eax");
808 SMLoc Loc = Operands.back()->getEndLoc();
809 Operands.push_back(X86Operand::CreateReg(Reg, Loc, Loc));
810 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000811
Chris Lattneree211d02010-09-11 16:32:12 +0000812 // FIXME: Hack to handle recognize "out[bwl] <op>". Canonicalize it to
813 // "outb %al, <op>".
814 if ((Name == "outb" || Name == "outw" || Name == "outl") &&
815 Operands.size() == 2) {
816 unsigned Reg;
817 if (Name[3] == 'b')
818 Reg = MatchRegisterName("al");
819 else if (Name[3] == 'w')
820 Reg = MatchRegisterName("ax");
821 else
822 Reg = MatchRegisterName("eax");
823 SMLoc Loc = Operands.back()->getEndLoc();
824 Operands.push_back(X86Operand::CreateReg(Reg, Loc, Loc));
825 std::swap(Operands[1], Operands[2]);
826 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000827
Chris Lattneref63c9a2010-09-14 23:34:29 +0000828 // FIXME: Hack to handle "out[bwl]? %al, (%dx)" -> "outb %al, %dx".
829 if ((Name == "outb" || Name == "outw" || Name == "outl" || Name == "out") &&
830 Operands.size() == 3) {
831 X86Operand &Op = *(X86Operand*)Operands.back();
832 if (Op.isMem() && Op.Mem.SegReg == 0 &&
833 isa<MCConstantExpr>(Op.Mem.Disp) &&
834 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 &&
835 Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) {
836 SMLoc Loc = Op.getEndLoc();
837 Operands.back() = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc);
838 delete &Op;
839 }
840 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000841
Kevin Enderbycf50a532010-05-25 20:52:34 +0000842 // FIXME: Hack to handle "f{mul*,add*,sub*,div*} $op, st(0)" the same as
843 // "f{mul*,add*,sub*,div*} $op"
844 if ((Name.startswith("fmul") || Name.startswith("fadd") ||
845 Name.startswith("fsub") || Name.startswith("fdiv")) &&
846 Operands.size() == 3 &&
847 static_cast<X86Operand*>(Operands[2])->isReg() &&
848 static_cast<X86Operand*>(Operands[2])->getReg() == X86::ST0) {
849 delete Operands[2];
850 Operands.erase(Operands.begin() + 2);
851 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000852
Chris Lattnercb296ec2010-09-27 07:08:21 +0000853 // FIXME: Hack to handle "f{mulp,addp} st(0), $op" the same as
Chris Lattnera25f9332010-09-29 18:39:16 +0000854 // "f{mulp,addp} $op", since they commute. We also allow fdivrp/fsubrp even
855 // though they don't commute, solely because gas does support this.
856 if ((Name=="fmulp" || Name=="faddp" || Name=="fsubrp" || Name=="fdivrp") &&
857 Operands.size() == 3 &&
Chris Lattner2c5291b2010-09-22 06:26:39 +0000858 static_cast<X86Operand*>(Operands[1])->isReg() &&
859 static_cast<X86Operand*>(Operands[1])->getReg() == X86::ST0) {
860 delete Operands[1];
861 Operands.erase(Operands.begin() + 1);
862 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000863
Daniel Dunbarfba88d42010-08-24 19:37:56 +0000864 // FIXME: Hack to handle "imul <imm>, B" which is an alias for "imul <imm>, B,
865 // B".
Daniel Dunbarae528f62010-08-24 19:24:18 +0000866 if (Name.startswith("imul") && Operands.size() == 3 &&
Daniel Dunbarfba88d42010-08-24 19:37:56 +0000867 static_cast<X86Operand*>(Operands[1])->isImm() &&
Daniel Dunbarae528f62010-08-24 19:24:18 +0000868 static_cast<X86Operand*>(Operands.back())->isReg()) {
869 X86Operand *Op = static_cast<X86Operand*>(Operands.back());
870 Operands.push_back(X86Operand::CreateReg(Op->getReg(), Op->getStartLoc(),
871 Op->getEndLoc()));
872 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000873
Chris Lattnerc5cebeb2010-09-06 23:51:44 +0000874 // 'sldt <mem>' can be encoded with either sldtw or sldtq with the same
875 // effect (both store to a 16-bit mem). Force to sldtw to avoid ambiguity
876 // errors, since its encoding is the most compact.
877 if (Name == "sldt" && Operands.size() == 2 &&
Benjamin Krameraceeb3a2010-09-07 14:40:58 +0000878 static_cast<X86Operand*>(Operands[1])->isMem()) {
879 delete Operands[0];
Chris Lattnerc5cebeb2010-09-06 23:51:44 +0000880 Operands[0] = X86Operand::CreateToken("sldtw", NameLoc);
Benjamin Krameraceeb3a2010-09-07 14:40:58 +0000881 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000882
Chris Lattner9607c402010-09-08 04:53:27 +0000883 // The assembler accepts "xchgX <reg>, <mem>" and "xchgX <mem>, <reg>" as
884 // synonyms. Our tables only have the "<reg>, <mem>" form, so if we see the
885 // other operand order, swap them.
Chris Lattner90b54542010-09-08 22:27:05 +0000886 if (Name == "xchgb" || Name == "xchgw" || Name == "xchgl" || Name == "xchgq"||
887 Name == "xchg")
Chris Lattner9607c402010-09-08 04:53:27 +0000888 if (Operands.size() == 3 &&
889 static_cast<X86Operand*>(Operands[1])->isMem() &&
890 static_cast<X86Operand*>(Operands[2])->isReg()) {
891 std::swap(Operands[1], Operands[2]);
892 }
Daniel Dunbarae528f62010-08-24 19:24:18 +0000893
Chris Lattnerc8ae35a2010-09-08 05:51:12 +0000894 // The assembler accepts "testX <reg>, <mem>" and "testX <mem>, <reg>" as
895 // synonyms. Our tables only have the "<mem>, <reg>" form, so if we see the
896 // other operand order, swap them.
Chris Lattner90b54542010-09-08 22:27:05 +0000897 if (Name == "testb" || Name == "testw" || Name == "testl" || Name == "testq"||
898 Name == "test")
Chris Lattnerc8ae35a2010-09-08 05:51:12 +0000899 if (Operands.size() == 3 &&
900 static_cast<X86Operand*>(Operands[1])->isReg() &&
901 static_cast<X86Operand*>(Operands[2])->isMem()) {
902 std::swap(Operands[1], Operands[2]);
903 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000904
Chris Lattner2d592d12010-09-15 04:04:33 +0000905 // The assembler accepts these instructions with no operand as a synonym for
906 // an instruction acting on st(1). e.g. "fxch" -> "fxch %st(1)".
907 if ((Name == "fxch" || Name == "fucom" || Name == "fucomp" ||
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000908 Name == "faddp" || Name == "fsubp" || Name == "fsubrp" ||
Chris Lattner2d592d12010-09-15 04:04:33 +0000909 Name == "fmulp" || Name == "fdivp" || Name == "fdivrp") &&
910 Operands.size() == 1) {
911 Operands.push_back(X86Operand::CreateReg(MatchRegisterName("st(1)"),
912 NameLoc, NameLoc));
913 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000914
Chris Lattner8f777a22010-09-15 04:08:38 +0000915 // The assembler accepts these instructions with two few operands as a synonym
916 // for taking %st(1),%st(0) or X, %st(0).
Kevin Enderby41e8cc72010-10-27 00:59:28 +0000917 if ((Name == "fcomi" || Name == "fucomi" || Name == "fucompi" ||
918 Name == "fcompi" ) &&
919 Operands.size() < 3) {
Chris Lattner8f777a22010-09-15 04:08:38 +0000920 if (Operands.size() == 1)
921 Operands.push_back(X86Operand::CreateReg(MatchRegisterName("st(1)"),
922 NameLoc, NameLoc));
923 Operands.push_back(X86Operand::CreateReg(MatchRegisterName("st(0)"),
924 NameLoc, NameLoc));
925 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000926
Chris Lattner84f362d2010-09-15 04:15:16 +0000927 // The assembler accepts various amounts of brokenness for fnstsw.
Chris Lattner8b260a72010-10-30 18:07:17 +0000928 if (Name == "fnstsw" || Name == "fnstsww") {
Chris Lattner84f362d2010-09-15 04:15:16 +0000929 if (Operands.size() == 2 &&
930 static_cast<X86Operand*>(Operands[1])->isReg()) {
931 // "fnstsw al" and "fnstsw eax" -> "fnstw"
932 unsigned Reg = static_cast<X86Operand*>(Operands[1])->Reg.RegNo;
933 if (Reg == MatchRegisterName("eax") ||
934 Reg == MatchRegisterName("al")) {
935 delete Operands[1];
936 Operands.pop_back();
937 }
938 }
939
940 // "fnstw" -> "fnstw %ax"
941 if (Operands.size() == 1)
942 Operands.push_back(X86Operand::CreateReg(MatchRegisterName("ax"),
943 NameLoc, NameLoc));
944 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000945
Chris Lattnercbb44262010-09-15 05:25:21 +0000946 // jmp $42,$5 -> ljmp, similarly for call.
947 if ((Name.startswith("call") || Name.startswith("jmp")) &&
948 Operands.size() == 3 &&
949 static_cast<X86Operand*>(Operands[1])->isImm() &&
950 static_cast<X86Operand*>(Operands[2])->isImm()) {
951 const char *NewOpName = StringSwitch<const char *>(Name)
952 .Case("jmp", "ljmp")
953 .Case("jmpw", "ljmpw")
954 .Case("jmpl", "ljmpl")
955 .Case("jmpq", "ljmpq")
956 .Case("call", "lcall")
957 .Case("callw", "lcallw")
958 .Case("calll", "lcalll")
959 .Case("callq", "lcallq")
960 .Default(0);
961 if (NewOpName) {
962 delete Operands[0];
963 Operands[0] = X86Operand::CreateToken(NewOpName, NameLoc);
Chris Lattnerd0bcc9a2010-09-15 05:30:20 +0000964 Name = NewOpName;
Chris Lattnercbb44262010-09-15 05:25:21 +0000965 }
966 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000967
Chris Lattnerd0bcc9a2010-09-15 05:30:20 +0000968 // lcall and ljmp -> lcalll and ljmpl
969 if ((Name == "lcall" || Name == "ljmp") && Operands.size() == 3) {
970 delete Operands[0];
971 Operands[0] = X86Operand::CreateToken(Name == "lcall" ? "lcalll" : "ljmpl",
972 NameLoc);
973 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000974
Chris Lattner1eb1b682010-09-22 05:49:14 +0000975 // call foo is not ambiguous with callw.
976 if (Name == "call" && Operands.size() == 2) {
977 const char *NewName = Is64Bit ? "callq" : "calll";
978 delete Operands[0];
979 Operands[0] = X86Operand::CreateToken(NewName, NameLoc);
980 Name = NewName;
981 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000982
Chris Lattner61129252010-09-22 03:50:32 +0000983 // movsd -> movsl (when no operands are specified).
984 if (Name == "movsd" && Operands.size() == 1) {
985 delete Operands[0];
986 Operands[0] = X86Operand::CreateToken("movsl", NameLoc);
987 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000988
Chris Lattner0c289c12010-09-22 04:04:03 +0000989 // fstp <mem> -> fstps <mem>. Without this, we'll default to fstpl due to
990 // suffix searching.
991 if (Name == "fstp" && Operands.size() == 2 &&
992 static_cast<X86Operand*>(Operands[1])->isMem()) {
993 delete Operands[0];
994 Operands[0] = X86Operand::CreateToken("fstps", NameLoc);
995 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +0000996
997
Chris Lattnerfd8fddd2010-09-27 04:23:03 +0000998 // "clr <reg>" -> "xor <reg>, <reg>".
999 if ((Name == "clrb" || Name == "clrw" || Name == "clrl" || Name == "clrq" ||
1000 Name == "clr") && Operands.size() == 2 &&
1001 static_cast<X86Operand*>(Operands[1])->isReg()) {
1002 unsigned RegNo = static_cast<X86Operand*>(Operands[1])->getReg();
1003 Operands.push_back(X86Operand::CreateReg(RegNo, NameLoc, NameLoc));
1004 delete Operands[0];
1005 Operands[0] = X86Operand::CreateToken("xor", NameLoc);
1006 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001007
Kevin Enderby7aef62f2010-10-18 17:04:36 +00001008 // FIXME: Hack to handle recognize "aa[dm]" -> "aa[dm] $0xA".
1009 if ((Name.startswith("aad") || Name.startswith("aam")) &&
1010 Operands.size() == 1) {
1011 const MCExpr *A = MCConstantExpr::Create(0xA, getParser().getContext());
1012 Operands.push_back(X86Operand::CreateImm(A, NameLoc, NameLoc));
1013 }
1014
Kevin Enderby87f4a1a2010-10-19 00:01:44 +00001015 // "lgdtl" is not ambiguous 32-bit mode and is the same as "lgdt".
1016 // "lgdtq" is not ambiguous 64-bit mode and is the same as "lgdt".
1017 if ((Name == "lgdtl" && Is64Bit == false) ||
1018 (Name == "lgdtq" && Is64Bit == true)) {
1019 const char *NewName = "lgdt";
1020 delete Operands[0];
1021 Operands[0] = X86Operand::CreateToken(NewName, NameLoc);
1022 Name = NewName;
1023 }
1024
1025 // "lidtl" is not ambiguous 32-bit mode and is the same as "lidt".
1026 // "lidtq" is not ambiguous 64-bit mode and is the same as "lidt".
1027 if ((Name == "lidtl" && Is64Bit == false) ||
1028 (Name == "lidtq" && Is64Bit == true)) {
1029 const char *NewName = "lidt";
1030 delete Operands[0];
1031 Operands[0] = X86Operand::CreateToken(NewName, NameLoc);
1032 Name = NewName;
1033 }
1034
1035 // "sgdtl" is not ambiguous 32-bit mode and is the same as "sgdt".
1036 // "sgdtq" is not ambiguous 64-bit mode and is the same as "sgdt".
1037 if ((Name == "sgdtl" && Is64Bit == false) ||
1038 (Name == "sgdtq" && Is64Bit == true)) {
1039 const char *NewName = "sgdt";
1040 delete Operands[0];
1041 Operands[0] = X86Operand::CreateToken(NewName, NameLoc);
1042 Name = NewName;
1043 }
1044
1045 // "sidtl" is not ambiguous 32-bit mode and is the same as "sidt".
1046 // "sidtq" is not ambiguous 64-bit mode and is the same as "sidt".
1047 if ((Name == "sidtl" && Is64Bit == false) ||
1048 (Name == "sidtq" && Is64Bit == true)) {
1049 const char *NewName = "sidt";
1050 delete Operands[0];
1051 Operands[0] = X86Operand::CreateToken(NewName, NameLoc);
1052 Name = NewName;
1053 }
1054
Chris Lattner98986712010-01-14 22:21:20 +00001055 return false;
Daniel Dunbara3af3702009-07-20 18:55:04 +00001056}
1057
Chris Lattner2d592d12010-09-15 04:04:33 +00001058bool X86ATTAsmParser::
Chris Lattner7036f8b2010-09-29 01:42:58 +00001059MatchAndEmitInstruction(SMLoc IDLoc,
Chris Lattner7c51a312010-09-29 01:50:45 +00001060 SmallVectorImpl<MCParsedAsmOperand*> &Operands,
Chris Lattner7036f8b2010-09-29 01:42:58 +00001061 MCStreamer &Out) {
Daniel Dunbarf1e29d42010-08-12 00:55:38 +00001062 assert(!Operands.empty() && "Unexpect empty operand list!");
Chris Lattner7c51a312010-09-29 01:50:45 +00001063 X86Operand *Op = static_cast<X86Operand*>(Operands[0]);
1064 assert(Op->isToken() && "Leading operand should always be a mnemonic!");
Daniel Dunbarf1e29d42010-08-12 00:55:38 +00001065
Chris Lattner7c51a312010-09-29 01:50:45 +00001066 // First, handle aliases that expand to multiple instructions.
1067 // FIXME: This should be replaced with a real .td file alias mechanism.
Andrew Trick0966ec02010-10-22 03:58:29 +00001068 if (Op->getToken() == "fstsw" || Op->getToken() == "fstcw" ||
Chris Lattner8b260a72010-10-30 18:07:17 +00001069 Op->getToken() == "fstsww" || Op->getToken() == "fstcww" ||
Chris Lattner905f2e02010-09-30 17:11:29 +00001070 Op->getToken() == "finit" || Op->getToken() == "fsave" ||
Kevin Enderby5a378072010-10-27 02:53:04 +00001071 Op->getToken() == "fstenv" || Op->getToken() == "fclex") {
Chris Lattner7c51a312010-09-29 01:50:45 +00001072 MCInst Inst;
1073 Inst.setOpcode(X86::WAIT);
1074 Out.EmitInstruction(Inst);
1075
Chris Lattner0bb83a82010-09-30 16:39:29 +00001076 const char *Repl =
1077 StringSwitch<const char*>(Op->getToken())
Chris Lattner8b260a72010-10-30 18:07:17 +00001078 .Case("finit", "fninit")
1079 .Case("fsave", "fnsave")
1080 .Case("fstcw", "fnstcw")
1081 .Case("fstcww", "fnstcw")
Chris Lattner905f2e02010-09-30 17:11:29 +00001082 .Case("fstenv", "fnstenv")
Chris Lattner8b260a72010-10-30 18:07:17 +00001083 .Case("fstsw", "fnstsw")
1084 .Case("fstsww", "fnstsw")
1085 .Case("fclex", "fnclex")
Chris Lattner0bb83a82010-09-30 16:39:29 +00001086 .Default(0);
1087 assert(Repl && "Unknown wait-prefixed instruction");
Benjamin Kramerb0f96fa2010-10-01 12:25:27 +00001088 delete Operands[0];
Chris Lattner0bb83a82010-09-30 16:39:29 +00001089 Operands[0] = X86Operand::CreateToken(Repl, IDLoc);
Chris Lattner7c51a312010-09-29 01:50:45 +00001090 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001091
Chris Lattnera008e8a2010-09-06 21:54:15 +00001092 bool WasOriginallyInvalidOperand = false;
Chris Lattnerce4a3352010-09-06 22:11:18 +00001093 unsigned OrigErrorInfo;
Chris Lattner7036f8b2010-09-29 01:42:58 +00001094 MCInst Inst;
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001095
Daniel Dunbarc918d602010-05-04 16:12:42 +00001096 // First, try a direct match.
Chris Lattnerce4a3352010-09-06 22:11:18 +00001097 switch (MatchInstructionImpl(Operands, Inst, OrigErrorInfo)) {
Chris Lattnerec6789f2010-09-06 20:08:02 +00001098 case Match_Success:
Chris Lattner7036f8b2010-09-29 01:42:58 +00001099 Out.EmitInstruction(Inst);
Daniel Dunbarc918d602010-05-04 16:12:42 +00001100 return false;
Chris Lattnerec6789f2010-09-06 20:08:02 +00001101 case Match_MissingFeature:
1102 Error(IDLoc, "instruction requires a CPU feature not currently enabled");
1103 return true;
Chris Lattnera008e8a2010-09-06 21:54:15 +00001104 case Match_InvalidOperand:
1105 WasOriginallyInvalidOperand = true;
1106 break;
1107 case Match_MnemonicFail:
Chris Lattnerec6789f2010-09-06 20:08:02 +00001108 break;
1109 }
Daniel Dunbarc918d602010-05-04 16:12:42 +00001110
Daniel Dunbarc918d602010-05-04 16:12:42 +00001111 // FIXME: Ideally, we would only attempt suffix matches for things which are
1112 // valid prefixes, and we could just infer the right unambiguous
1113 // type. However, that requires substantially more matcher support than the
1114 // following hack.
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001115
Daniel Dunbarc918d602010-05-04 16:12:42 +00001116 // Change the operand to point to a temporary token.
Daniel Dunbarc918d602010-05-04 16:12:42 +00001117 StringRef Base = Op->getToken();
Daniel Dunbarf1e29d42010-08-12 00:55:38 +00001118 SmallString<16> Tmp;
1119 Tmp += Base;
1120 Tmp += ' ';
1121 Op->setTokenValue(Tmp.str());
Daniel Dunbarc918d602010-05-04 16:12:42 +00001122
1123 // Check for the various suffix matches.
1124 Tmp[Base.size()] = 'b';
Chris Lattnerce4a3352010-09-06 22:11:18 +00001125 unsigned BErrorInfo, WErrorInfo, LErrorInfo, QErrorInfo;
1126 MatchResultTy MatchB = MatchInstructionImpl(Operands, Inst, BErrorInfo);
Daniel Dunbarc918d602010-05-04 16:12:42 +00001127 Tmp[Base.size()] = 'w';
Chris Lattnerce4a3352010-09-06 22:11:18 +00001128 MatchResultTy MatchW = MatchInstructionImpl(Operands, Inst, WErrorInfo);
Daniel Dunbarc918d602010-05-04 16:12:42 +00001129 Tmp[Base.size()] = 'l';
Chris Lattnerce4a3352010-09-06 22:11:18 +00001130 MatchResultTy MatchL = MatchInstructionImpl(Operands, Inst, LErrorInfo);
Daniel Dunbar04814492010-05-12 00:54:20 +00001131 Tmp[Base.size()] = 'q';
Chris Lattnerce4a3352010-09-06 22:11:18 +00001132 MatchResultTy MatchQ = MatchInstructionImpl(Operands, Inst, QErrorInfo);
Daniel Dunbarc918d602010-05-04 16:12:42 +00001133
1134 // Restore the old token.
1135 Op->setTokenValue(Base);
1136
1137 // If exactly one matched, then we treat that as a successful match (and the
1138 // instruction will already have been filled in correctly, since the failing
1139 // matches won't have modified it).
Chris Lattnerec6789f2010-09-06 20:08:02 +00001140 unsigned NumSuccessfulMatches =
1141 (MatchB == Match_Success) + (MatchW == Match_Success) +
1142 (MatchL == Match_Success) + (MatchQ == Match_Success);
Chris Lattner7036f8b2010-09-29 01:42:58 +00001143 if (NumSuccessfulMatches == 1) {
1144 Out.EmitInstruction(Inst);
Daniel Dunbarc918d602010-05-04 16:12:42 +00001145 return false;
Chris Lattner7036f8b2010-09-29 01:42:58 +00001146 }
Daniel Dunbarc918d602010-05-04 16:12:42 +00001147
Chris Lattnerec6789f2010-09-06 20:08:02 +00001148 // Otherwise, the match failed, try to produce a decent error message.
Daniel Dunbarf1e29d42010-08-12 00:55:38 +00001149
Daniel Dunbar09062b12010-08-12 00:55:42 +00001150 // If we had multiple suffix matches, then identify this as an ambiguous
1151 // match.
Chris Lattnerec6789f2010-09-06 20:08:02 +00001152 if (NumSuccessfulMatches > 1) {
Daniel Dunbar09062b12010-08-12 00:55:42 +00001153 char MatchChars[4];
1154 unsigned NumMatches = 0;
Chris Lattnerec6789f2010-09-06 20:08:02 +00001155 if (MatchB == Match_Success)
Daniel Dunbar09062b12010-08-12 00:55:42 +00001156 MatchChars[NumMatches++] = 'b';
Chris Lattnerec6789f2010-09-06 20:08:02 +00001157 if (MatchW == Match_Success)
Daniel Dunbar09062b12010-08-12 00:55:42 +00001158 MatchChars[NumMatches++] = 'w';
Chris Lattnerec6789f2010-09-06 20:08:02 +00001159 if (MatchL == Match_Success)
Daniel Dunbar09062b12010-08-12 00:55:42 +00001160 MatchChars[NumMatches++] = 'l';
Chris Lattnerec6789f2010-09-06 20:08:02 +00001161 if (MatchQ == Match_Success)
Daniel Dunbar09062b12010-08-12 00:55:42 +00001162 MatchChars[NumMatches++] = 'q';
1163
1164 SmallString<126> Msg;
1165 raw_svector_ostream OS(Msg);
1166 OS << "ambiguous instructions require an explicit suffix (could be ";
1167 for (unsigned i = 0; i != NumMatches; ++i) {
1168 if (i != 0)
1169 OS << ", ";
1170 if (i + 1 == NumMatches)
1171 OS << "or ";
1172 OS << "'" << Base << MatchChars[i] << "'";
1173 }
1174 OS << ")";
1175 Error(IDLoc, OS.str());
Chris Lattnerec6789f2010-09-06 20:08:02 +00001176 return true;
Daniel Dunbar09062b12010-08-12 00:55:42 +00001177 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001178
Chris Lattnera008e8a2010-09-06 21:54:15 +00001179 // Okay, we know that none of the variants matched successfully.
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001180
Chris Lattnera008e8a2010-09-06 21:54:15 +00001181 // If all of the instructions reported an invalid mnemonic, then the original
1182 // mnemonic was invalid.
1183 if ((MatchB == Match_MnemonicFail) && (MatchW == Match_MnemonicFail) &&
1184 (MatchL == Match_MnemonicFail) && (MatchQ == Match_MnemonicFail)) {
Chris Lattnerce4a3352010-09-06 22:11:18 +00001185 if (!WasOriginallyInvalidOperand) {
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001186 Error(IDLoc, "invalid instruction mnemonic '" + Base + "'");
Chris Lattnerce4a3352010-09-06 22:11:18 +00001187 return true;
1188 }
1189
1190 // Recover location info for the operand if we know which was the problem.
1191 SMLoc ErrorLoc = IDLoc;
1192 if (OrigErrorInfo != ~0U) {
Chris Lattnerf8840122010-09-15 03:50:11 +00001193 if (OrigErrorInfo >= Operands.size())
1194 return Error(IDLoc, "too few operands for instruction");
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001195
Chris Lattnerce4a3352010-09-06 22:11:18 +00001196 ErrorLoc = ((X86Operand*)Operands[OrigErrorInfo])->getStartLoc();
1197 if (ErrorLoc == SMLoc()) ErrorLoc = IDLoc;
1198 }
1199
Chris Lattnerf8840122010-09-15 03:50:11 +00001200 return Error(ErrorLoc, "invalid operand for instruction");
Chris Lattnera008e8a2010-09-06 21:54:15 +00001201 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001202
Chris Lattnerec6789f2010-09-06 20:08:02 +00001203 // If one instruction matched with a missing feature, report this as a
1204 // missing feature.
1205 if ((MatchB == Match_MissingFeature) + (MatchW == Match_MissingFeature) +
Chris Lattnera008e8a2010-09-06 21:54:15 +00001206 (MatchL == Match_MissingFeature) + (MatchQ == Match_MissingFeature) == 1){
Chris Lattnerec6789f2010-09-06 20:08:02 +00001207 Error(IDLoc, "instruction requires a CPU feature not currently enabled");
1208 return true;
1209 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001210
Chris Lattnera008e8a2010-09-06 21:54:15 +00001211 // If one instruction matched with an invalid operand, report this as an
1212 // operand failure.
1213 if ((MatchB == Match_InvalidOperand) + (MatchW == Match_InvalidOperand) +
1214 (MatchL == Match_InvalidOperand) + (MatchQ == Match_InvalidOperand) == 1){
1215 Error(IDLoc, "invalid operand for instruction");
1216 return true;
1217 }
Michael J. Spencerc0c8df32010-10-09 11:00:50 +00001218
Chris Lattnerec6789f2010-09-06 20:08:02 +00001219 // If all of these were an outright failure, report it in a useless way.
1220 // FIXME: We should give nicer diagnostics about the exact failure.
Chris Lattnera008e8a2010-09-06 21:54:15 +00001221 Error(IDLoc, "unknown use of instruction mnemonic without a size suffix");
Daniel Dunbarc918d602010-05-04 16:12:42 +00001222 return true;
1223}
1224
1225
Chris Lattner537ca842010-10-30 17:38:55 +00001226bool X86ATTAsmParser::ParseDirective(AsmToken DirectiveID) {
1227 StringRef IDVal = DirectiveID.getIdentifier();
1228 if (IDVal == ".word")
1229 return ParseDirectiveWord(2, DirectiveID.getLoc());
1230 return true;
1231}
1232
1233/// ParseDirectiveWord
1234/// ::= .word [ expression (, expression)* ]
1235bool X86ATTAsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
1236 if (getLexer().isNot(AsmToken::EndOfStatement)) {
1237 for (;;) {
1238 const MCExpr *Value;
1239 if (getParser().ParseExpression(Value))
1240 return true;
1241
1242 getParser().getStreamer().EmitValue(Value, Size, 0 /*addrspace*/);
1243
1244 if (getLexer().is(AsmToken::EndOfStatement))
1245 break;
1246
1247 // FIXME: Improve diagnostic.
1248 if (getLexer().isNot(AsmToken::Comma))
1249 return Error(L, "unexpected token in directive");
1250 Parser.Lex();
1251 }
1252 }
1253
1254 Parser.Lex();
1255 return false;
1256}
1257
1258
1259
1260
Sean Callanane88f5522010-01-23 02:43:15 +00001261extern "C" void LLVMInitializeX86AsmLexer();
1262
Daniel Dunbar092a9dd2009-07-17 20:42:00 +00001263// Force static initialization.
1264extern "C" void LLVMInitializeX86AsmParser() {
Daniel Dunbarf98bc632010-03-18 20:06:02 +00001265 RegisterAsmParser<X86_32ATTAsmParser> X(TheX86_32Target);
1266 RegisterAsmParser<X86_64ATTAsmParser> Y(TheX86_64Target);
Sean Callanane88f5522010-01-23 02:43:15 +00001267 LLVMInitializeX86AsmLexer();
Daniel Dunbar092a9dd2009-07-17 20:42:00 +00001268}
Daniel Dunbar0e2771f2009-07-29 00:02:19 +00001269
Chris Lattner0692ee62010-09-06 19:11:01 +00001270#define GET_REGISTER_MATCHER
1271#define GET_MATCHER_IMPLEMENTATION
Daniel Dunbar0e2771f2009-07-29 00:02:19 +00001272#include "X86GenAsmMatcher.inc"