blob: 792a91332ee1a25168efdcb36821d986ee74b648 [file] [log] [blame]
Chad Rosier3d45a772012-08-17 21:27:25 +00001//===--- SemaStmtAsm.cpp - Semantic Analysis for Asm Statements -----------===//
Chad Rosier4b5e48d2012-08-17 21:19:40 +00002//
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//
10// This file implements semantic analysis for inline asm statements.
11//
12//===----------------------------------------------------------------------===//
13
14#include "clang/Sema/SemaInternal.h"
15#include "clang/Sema/Scope.h"
16#include "clang/Sema/ScopeInfo.h"
17#include "clang/Sema/Initialization.h"
18#include "clang/Sema/Lookup.h"
19#include "clang/AST/TypeLoc.h"
20#include "clang/Lex/Preprocessor.h"
21#include "clang/Basic/TargetInfo.h"
22#include "llvm/ADT/ArrayRef.h"
23#include "llvm/ADT/BitVector.h"
24#include "llvm/ADT/SmallString.h"
25#include "llvm/MC/MCAsmInfo.h"
26#include "llvm/MC/MCContext.h"
Chad Rosier6e97be72012-08-22 23:42:09 +000027#include "llvm/MC/MCExpr.h"
Chad Rosier4b5e48d2012-08-17 21:19:40 +000028#include "llvm/MC/MCInst.h"
29#include "llvm/MC/MCInstPrinter.h"
30#include "llvm/MC/MCInstrInfo.h"
31#include "llvm/MC/MCObjectFileInfo.h"
32#include "llvm/MC/MCRegisterInfo.h"
33#include "llvm/MC/MCStreamer.h"
34#include "llvm/MC/MCSubtargetInfo.h"
Chad Rosier6e97be72012-08-22 23:42:09 +000035#include "llvm/MC/MCSymbol.h"
Chad Rosier4b5e48d2012-08-17 21:19:40 +000036#include "llvm/MC/MCTargetAsmParser.h"
37#include "llvm/MC/MCParser/MCAsmLexer.h"
38#include "llvm/MC/MCParser/MCAsmParser.h"
Chad Rosierfb700262012-09-11 23:13:15 +000039#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
Chad Rosier4b5e48d2012-08-17 21:19:40 +000040#include "llvm/Support/SourceMgr.h"
41#include "llvm/Support/TargetRegistry.h"
42#include "llvm/Support/TargetSelect.h"
43using namespace clang;
44using namespace sema;
45
46/// CheckAsmLValue - GNU C has an extremely ugly extension whereby they silently
47/// ignore "noop" casts in places where an lvalue is required by an inline asm.
48/// We emulate this behavior when -fheinous-gnu-extensions is specified, but
49/// provide a strong guidance to not use it.
50///
51/// This method checks to see if the argument is an acceptable l-value and
52/// returns false if it is a case we can handle.
53static bool CheckAsmLValue(const Expr *E, Sema &S) {
54 // Type dependent expressions will be checked during instantiation.
55 if (E->isTypeDependent())
56 return false;
57
58 if (E->isLValue())
59 return false; // Cool, this is an lvalue.
60
61 // Okay, this is not an lvalue, but perhaps it is the result of a cast that we
62 // are supposed to allow.
63 const Expr *E2 = E->IgnoreParenNoopCasts(S.Context);
64 if (E != E2 && E2->isLValue()) {
65 if (!S.getLangOpts().HeinousExtensions)
66 S.Diag(E2->getLocStart(), diag::err_invalid_asm_cast_lvalue)
67 << E->getSourceRange();
68 else
69 S.Diag(E2->getLocStart(), diag::warn_invalid_asm_cast_lvalue)
70 << E->getSourceRange();
71 // Accept, even if we emitted an error diagnostic.
72 return false;
73 }
74
75 // None of the above, just randomly invalid non-lvalue.
76 return true;
77}
78
79/// isOperandMentioned - Return true if the specified operand # is mentioned
80/// anywhere in the decomposed asm string.
81static bool isOperandMentioned(unsigned OpNo,
Chad Rosierdf5faf52012-08-25 00:11:56 +000082 ArrayRef<GCCAsmStmt::AsmStringPiece> AsmStrPieces) {
Chad Rosier4b5e48d2012-08-17 21:19:40 +000083 for (unsigned p = 0, e = AsmStrPieces.size(); p != e; ++p) {
Chad Rosierdf5faf52012-08-25 00:11:56 +000084 const GCCAsmStmt::AsmStringPiece &Piece = AsmStrPieces[p];
Chad Rosier4b5e48d2012-08-17 21:19:40 +000085 if (!Piece.isOperand()) continue;
86
87 // If this is a reference to the input and if the input was the smaller
88 // one, then we have to reject this asm.
89 if (Piece.getOperandNo() == OpNo)
90 return true;
91 }
92 return false;
93}
94
Chad Rosierdf5faf52012-08-25 00:11:56 +000095StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
96 bool IsVolatile, unsigned NumOutputs,
97 unsigned NumInputs, IdentifierInfo **Names,
98 MultiExprArg constraints, MultiExprArg exprs,
99 Expr *asmString, MultiExprArg clobbers,
100 SourceLocation RParenLoc) {
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000101 unsigned NumClobbers = clobbers.size();
102 StringLiteral **Constraints =
Benjamin Kramer5354e772012-08-23 23:38:35 +0000103 reinterpret_cast<StringLiteral**>(constraints.data());
104 Expr **Exprs = exprs.data();
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000105 StringLiteral *AsmString = cast<StringLiteral>(asmString);
Benjamin Kramer5354e772012-08-23 23:38:35 +0000106 StringLiteral **Clobbers = reinterpret_cast<StringLiteral**>(clobbers.data());
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000107
108 SmallVector<TargetInfo::ConstraintInfo, 4> OutputConstraintInfos;
109
110 // The parser verifies that there is a string literal here.
111 if (!AsmString->isAscii())
112 return StmtError(Diag(AsmString->getLocStart(),diag::err_asm_wide_character)
113 << AsmString->getSourceRange());
114
115 for (unsigned i = 0; i != NumOutputs; i++) {
116 StringLiteral *Literal = Constraints[i];
117 if (!Literal->isAscii())
118 return StmtError(Diag(Literal->getLocStart(),diag::err_asm_wide_character)
119 << Literal->getSourceRange());
120
121 StringRef OutputName;
122 if (Names[i])
123 OutputName = Names[i]->getName();
124
125 TargetInfo::ConstraintInfo Info(Literal->getString(), OutputName);
126 if (!Context.getTargetInfo().validateOutputConstraint(Info))
127 return StmtError(Diag(Literal->getLocStart(),
128 diag::err_asm_invalid_output_constraint)
129 << Info.getConstraintStr());
130
131 // Check that the output exprs are valid lvalues.
132 Expr *OutputExpr = Exprs[i];
133 if (CheckAsmLValue(OutputExpr, *this)) {
134 return StmtError(Diag(OutputExpr->getLocStart(),
135 diag::err_asm_invalid_lvalue_in_output)
136 << OutputExpr->getSourceRange());
137 }
138
139 OutputConstraintInfos.push_back(Info);
140 }
141
142 SmallVector<TargetInfo::ConstraintInfo, 4> InputConstraintInfos;
143
144 for (unsigned i = NumOutputs, e = NumOutputs + NumInputs; i != e; i++) {
145 StringLiteral *Literal = Constraints[i];
146 if (!Literal->isAscii())
147 return StmtError(Diag(Literal->getLocStart(),diag::err_asm_wide_character)
148 << Literal->getSourceRange());
149
150 StringRef InputName;
151 if (Names[i])
152 InputName = Names[i]->getName();
153
154 TargetInfo::ConstraintInfo Info(Literal->getString(), InputName);
155 if (!Context.getTargetInfo().validateInputConstraint(OutputConstraintInfos.data(),
156 NumOutputs, Info)) {
157 return StmtError(Diag(Literal->getLocStart(),
158 diag::err_asm_invalid_input_constraint)
159 << Info.getConstraintStr());
160 }
161
162 Expr *InputExpr = Exprs[i];
163
164 // Only allow void types for memory constraints.
165 if (Info.allowsMemory() && !Info.allowsRegister()) {
166 if (CheckAsmLValue(InputExpr, *this))
167 return StmtError(Diag(InputExpr->getLocStart(),
168 diag::err_asm_invalid_lvalue_in_input)
169 << Info.getConstraintStr()
170 << InputExpr->getSourceRange());
171 }
172
173 if (Info.allowsRegister()) {
174 if (InputExpr->getType()->isVoidType()) {
175 return StmtError(Diag(InputExpr->getLocStart(),
176 diag::err_asm_invalid_type_in_input)
177 << InputExpr->getType() << Info.getConstraintStr()
178 << InputExpr->getSourceRange());
179 }
180 }
181
182 ExprResult Result = DefaultFunctionArrayLvalueConversion(Exprs[i]);
183 if (Result.isInvalid())
184 return StmtError();
185
186 Exprs[i] = Result.take();
187 InputConstraintInfos.push_back(Info);
188 }
189
190 // Check that the clobbers are valid.
191 for (unsigned i = 0; i != NumClobbers; i++) {
192 StringLiteral *Literal = Clobbers[i];
193 if (!Literal->isAscii())
194 return StmtError(Diag(Literal->getLocStart(),diag::err_asm_wide_character)
195 << Literal->getSourceRange());
196
197 StringRef Clobber = Literal->getString();
198
199 if (!Context.getTargetInfo().isValidClobber(Clobber))
200 return StmtError(Diag(Literal->getLocStart(),
201 diag::err_asm_unknown_register_name) << Clobber);
202 }
203
Chad Rosierdf5faf52012-08-25 00:11:56 +0000204 GCCAsmStmt *NS =
205 new (Context) GCCAsmStmt(Context, AsmLoc, IsSimple, IsVolatile, NumOutputs,
206 NumInputs, Names, Constraints, Exprs, AsmString,
207 NumClobbers, Clobbers, RParenLoc);
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000208 // Validate the asm string, ensuring it makes sense given the operands we
209 // have.
Chad Rosierdf5faf52012-08-25 00:11:56 +0000210 SmallVector<GCCAsmStmt::AsmStringPiece, 8> Pieces;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000211 unsigned DiagOffs;
212 if (unsigned DiagID = NS->AnalyzeAsmString(Pieces, Context, DiagOffs)) {
213 Diag(getLocationOfStringLiteralByte(AsmString, DiagOffs), DiagID)
214 << AsmString->getSourceRange();
215 return StmtError();
216 }
217
218 // Validate tied input operands for type mismatches.
219 for (unsigned i = 0, e = InputConstraintInfos.size(); i != e; ++i) {
220 TargetInfo::ConstraintInfo &Info = InputConstraintInfos[i];
221
222 // If this is a tied constraint, verify that the output and input have
223 // either exactly the same type, or that they are int/ptr operands with the
224 // same size (int/long, int*/long, are ok etc).
225 if (!Info.hasTiedOperand()) continue;
226
227 unsigned TiedTo = Info.getTiedOperand();
228 unsigned InputOpNo = i+NumOutputs;
229 Expr *OutputExpr = Exprs[TiedTo];
230 Expr *InputExpr = Exprs[InputOpNo];
231
232 if (OutputExpr->isTypeDependent() || InputExpr->isTypeDependent())
233 continue;
234
235 QualType InTy = InputExpr->getType();
236 QualType OutTy = OutputExpr->getType();
237 if (Context.hasSameType(InTy, OutTy))
238 continue; // All types can be tied to themselves.
239
240 // Decide if the input and output are in the same domain (integer/ptr or
241 // floating point.
242 enum AsmDomain {
243 AD_Int, AD_FP, AD_Other
244 } InputDomain, OutputDomain;
245
246 if (InTy->isIntegerType() || InTy->isPointerType())
247 InputDomain = AD_Int;
248 else if (InTy->isRealFloatingType())
249 InputDomain = AD_FP;
250 else
251 InputDomain = AD_Other;
252
253 if (OutTy->isIntegerType() || OutTy->isPointerType())
254 OutputDomain = AD_Int;
255 else if (OutTy->isRealFloatingType())
256 OutputDomain = AD_FP;
257 else
258 OutputDomain = AD_Other;
259
260 // They are ok if they are the same size and in the same domain. This
261 // allows tying things like:
262 // void* to int*
263 // void* to int if they are the same size.
264 // double to long double if they are the same size.
265 //
266 uint64_t OutSize = Context.getTypeSize(OutTy);
267 uint64_t InSize = Context.getTypeSize(InTy);
268 if (OutSize == InSize && InputDomain == OutputDomain &&
269 InputDomain != AD_Other)
270 continue;
271
272 // If the smaller input/output operand is not mentioned in the asm string,
273 // then we can promote the smaller one to a larger input and the asm string
274 // won't notice.
275 bool SmallerValueMentioned = false;
276
277 // If this is a reference to the input and if the input was the smaller
278 // one, then we have to reject this asm.
279 if (isOperandMentioned(InputOpNo, Pieces)) {
280 // This is a use in the asm string of the smaller operand. Since we
281 // codegen this by promoting to a wider value, the asm will get printed
282 // "wrong".
283 SmallerValueMentioned |= InSize < OutSize;
284 }
285 if (isOperandMentioned(TiedTo, Pieces)) {
286 // If this is a reference to the output, and if the output is the larger
287 // value, then it's ok because we'll promote the input to the larger type.
288 SmallerValueMentioned |= OutSize < InSize;
289 }
290
291 // If the smaller value wasn't mentioned in the asm string, and if the
292 // output was a register, just extend the shorter one to the size of the
293 // larger one.
294 if (!SmallerValueMentioned && InputDomain != AD_Other &&
295 OutputConstraintInfos[TiedTo].allowsRegister())
296 continue;
297
298 // Either both of the operands were mentioned or the smaller one was
299 // mentioned. One more special case that we'll allow: if the tied input is
300 // integer, unmentioned, and is a constant, then we'll allow truncating it
301 // down to the size of the destination.
302 if (InputDomain == AD_Int && OutputDomain == AD_Int &&
303 !isOperandMentioned(InputOpNo, Pieces) &&
304 InputExpr->isEvaluatable(Context)) {
305 CastKind castKind =
306 (OutTy->isBooleanType() ? CK_IntegralToBoolean : CK_IntegralCast);
307 InputExpr = ImpCastExprToType(InputExpr, OutTy, castKind).take();
308 Exprs[InputOpNo] = InputExpr;
309 NS->setInputExpr(i, InputExpr);
310 continue;
311 }
312
313 Diag(InputExpr->getLocStart(),
314 diag::err_asm_tying_incompatible_types)
315 << InTy << OutTy << OutputExpr->getSourceRange()
316 << InputExpr->getSourceRange();
317 return StmtError();
318 }
319
320 return Owned(NS);
321}
322
323// isMSAsmKeyword - Return true if this is an MS-style inline asm keyword. These
324// require special handling.
325static bool isMSAsmKeyword(StringRef Name) {
326 bool Ret = llvm::StringSwitch<bool>(Name)
327 .Cases("EVEN", "ALIGN", true) // Alignment directives.
328 .Cases("LENGTH", "SIZE", "TYPE", true) // Type and variable sizes.
329 .Case("_emit", true) // _emit Pseudoinstruction.
330 .Default(false);
331 return Ret;
332}
333
Chad Rosier6e97be72012-08-22 23:42:09 +0000334// getIdentifierInfo - Given a Name and a range of tokens, find the associated
335// IdentifierInfo*.
336static IdentifierInfo *getIdentifierInfo(StringRef Name,
337 ArrayRef<Token> AsmToks,
338 unsigned Begin, unsigned End) {
339 for (unsigned i = Begin; i <= End; ++i) {
340 IdentifierInfo *II = AsmToks[i].getIdentifierInfo();
341 if (II && II->getName() == Name)
342 return II;
343 }
344 return 0;
345}
346
Chad Rosier358ab762012-08-22 21:08:06 +0000347// getSpelling - Get the spelling of the AsmTok token.
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000348static StringRef getSpelling(Sema &SemaRef, Token AsmTok) {
349 StringRef Asm;
350 SmallString<512> TokenBuf;
351 TokenBuf.resize(512);
352 bool StringInvalid = false;
353 Asm = SemaRef.PP.getSpelling(AsmTok, TokenBuf, &StringInvalid);
354 assert (!StringInvalid && "Expected valid string!");
355 return Asm;
356}
357
Chad Rosier358ab762012-08-22 21:08:06 +0000358// Determine if we should bail on this MSAsm instruction.
Chad Rosier2735df22012-08-22 19:18:30 +0000359static bool bailOnMSAsm(std::vector<StringRef> Piece) {
360 for (unsigned i = 0, e = Piece.size(); i != e; ++i)
361 if (isMSAsmKeyword(Piece[i]))
362 return true;
363 return false;
364}
365
Chad Rosier358ab762012-08-22 21:08:06 +0000366// Determine if we should bail on this MSAsm block.
Chad Rosier2735df22012-08-22 19:18:30 +0000367static bool bailOnMSAsm(std::vector<std::vector<StringRef> > Pieces) {
368 for (unsigned i = 0, e = Pieces.size(); i != e; ++i)
369 if (bailOnMSAsm(Pieces[i]))
370 return true;
371 return false;
372}
373
Chad Rosier358ab762012-08-22 21:08:06 +0000374// Determine if this is a simple MSAsm instruction.
Chad Rosier98ac6082012-08-21 23:09:21 +0000375static bool isSimpleMSAsm(std::vector<StringRef> &Pieces,
376 const TargetInfo &TI) {
377 if (isMSAsmKeyword(Pieces[0]))
378 return false;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000379
Chad Rosier98ac6082012-08-21 23:09:21 +0000380 for (unsigned i = 1, e = Pieces.size(); i != e; ++i)
381 if (!TI.isValidGCCRegisterName(Pieces[i]))
382 return false;
Chad Rosier153f8ec2012-08-22 19:50:28 +0000383 return true;
384}
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000385
Chad Rosier358ab762012-08-22 21:08:06 +0000386// Determine if this is a simple MSAsm block.
Chad Rosier153f8ec2012-08-22 19:50:28 +0000387static bool isSimpleMSAsm(std::vector<std::vector<StringRef> > Pieces,
388 const TargetInfo &TI) {
389 for (unsigned i = 0, e = Pieces.size(); i != e; ++i)
390 if (!isSimpleMSAsm(Pieces[i], TI))
391 return false;
Chad Rosier98ac6082012-08-21 23:09:21 +0000392 return true;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000393}
394
Bob Wilson40d39e32012-09-24 19:57:55 +0000395// Break the AsmString into pieces (i.e., mnemonic and operands).
Chad Rosier38c71d32012-08-21 21:56:39 +0000396static void buildMSAsmPieces(StringRef Asm, std::vector<StringRef> &Pieces) {
397 std::pair<StringRef,StringRef> Split = Asm.split(' ');
398
399 // Mnemonic
400 Pieces.push_back(Split.first);
401 Asm = Split.second;
402
403 // Operands
404 while (!Asm.empty()) {
405 Split = Asm.split(", ");
406 Pieces.push_back(Split.first);
407 Asm = Split.second;
408 }
409}
410
Chad Rosierf0fbd772012-08-22 21:04:07 +0000411static void buildMSAsmPieces(std::vector<std::string> &AsmStrings,
412 std::vector<std::vector<StringRef> > &Pieces) {
413 for (unsigned i = 0, e = AsmStrings.size(); i != e; ++i)
414 buildMSAsmPieces(AsmStrings[i], Pieces[i]);
415}
416
Chad Rosierb55e6022012-09-13 00:06:55 +0000417// Build the individual assembly instruction(s) and place them in the AsmStrings
Bob Wilsonb0f6b9c2012-09-24 19:57:59 +0000418// vector. These strings are fed to the AsmParser. Returns true on error.
419static bool buildMSAsmStrings(Sema &SemaRef,
420 SourceLocation AsmLoc,
421 ArrayRef<Token> AsmToks,
Chad Rosierb55e6022012-09-13 00:06:55 +0000422 std::vector<std::string> &AsmStrings,
Chad Rosier9072a022012-08-22 20:30:58 +0000423 std::vector<std::pair<unsigned,unsigned> > &AsmTokRanges) {
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000424 assert (!AsmToks.empty() && "Didn't expect an empty AsmToks!");
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000425
426 SmallString<512> Asm;
Chad Rosier9072a022012-08-22 20:30:58 +0000427 unsigned startTok = 0;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000428 for (unsigned i = 0, e = AsmToks.size(); i < e; ++i) {
Bob Wilson40d39e32012-09-24 19:57:55 +0000429 bool isNewAsm = ((i == 0) ||
430 AsmToks[i].isAtStartOfLine() ||
431 AsmToks[i].is(tok::kw_asm));
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000432
433 if (isNewAsm) {
Chad Rosier38c71d32012-08-21 21:56:39 +0000434 if (i) {
Benjamin Kramer32f3acc2012-08-24 20:43:21 +0000435 AsmStrings.push_back(Asm.str());
Chad Rosier9072a022012-08-22 20:30:58 +0000436 AsmTokRanges.push_back(std::make_pair(startTok, i-1));
437 startTok = i;
Chad Rosier38c71d32012-08-21 21:56:39 +0000438 Asm.clear();
Chad Rosier38c71d32012-08-21 21:56:39 +0000439 }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000440 if (AsmToks[i].is(tok::kw_asm)) {
441 i++; // Skip __asm
Bob Wilsonb0f6b9c2012-09-24 19:57:59 +0000442 if (i == e) {
443 SemaRef.Diag(AsmLoc, diag::err_asm_empty);
444 return true;
445 }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000446 }
447 }
448
Chad Rosierb55e6022012-09-13 00:06:55 +0000449 if (i && AsmToks[i].hasLeadingSpace() && !isNewAsm)
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000450 Asm += ' ';
Chad Rosier4de97162012-09-11 00:51:28 +0000451
452 StringRef Spelling = getSpelling(SemaRef, AsmToks[i]);
453 Asm += Spelling;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000454 }
Benjamin Kramer32f3acc2012-08-24 20:43:21 +0000455 AsmStrings.push_back(Asm.str());
Chad Rosier9072a022012-08-22 20:30:58 +0000456 AsmTokRanges.push_back(std::make_pair(startTok, AsmToks.size()-1));
Bob Wilsonb0f6b9c2012-09-24 19:57:59 +0000457
458 return false;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000459}
460
Chad Rosierc6916492012-09-06 19:56:25 +0000461#define DEF_SIMPLE_MSASM(STR) \
Chad Rosier89fb6d72012-08-28 20:28:20 +0000462 MSAsmStmt *NS = \
463 new (Context) MSAsmStmt(Context, AsmLoc, LBraceLoc, /*IsSimple*/ true, \
464 /*IsVolatile*/ true, AsmToks, Inputs, Outputs, \
Chad Rosierc6916492012-09-06 19:56:25 +0000465 InputExprs, OutputExprs, STR, Constraints, \
Chad Rosier89fb6d72012-08-28 20:28:20 +0000466 Clobbers, EndLoc);
Chad Rosier2735df22012-08-22 19:18:30 +0000467
Chad Rosierb55e6022012-09-13 00:06:55 +0000468StmtResult Sema::ActOnMSAsmStmt(SourceLocation AsmLoc, SourceLocation LBraceLoc,
469 ArrayRef<Token> AsmToks,SourceLocation EndLoc) {
Chad Rosier89fb6d72012-08-28 20:28:20 +0000470 SmallVector<StringRef, 4> Constraints;
471 std::vector<std::string> InputConstraints;
472 std::vector<std::string> OutputConstraints;
Chad Rosier4112a4c2012-08-28 20:35:06 +0000473 SmallVector<StringRef, 4> Clobbers;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000474 std::set<std::string> ClobberRegs;
Chad Rosierb55e6022012-09-13 00:06:55 +0000475
476 // FIXME: Use a struct to hold the various expression information.
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000477 SmallVector<IdentifierInfo*, 4> Inputs;
478 SmallVector<IdentifierInfo*, 4> Outputs;
Chad Rosier633abb02012-08-24 00:07:09 +0000479 SmallVector<Expr*, 4> InputExprs;
480 SmallVector<Expr*, 4> OutputExprs;
Chad Rosieracc22b62012-09-06 19:35:00 +0000481 SmallVector<std::string, 4> InputExprNames;
482 SmallVector<std::string, 4> OutputExprNames;
Chad Rosierb55e6022012-09-13 00:06:55 +0000483 SmallVector<unsigned, 4> InputExprStrIdx;
484 SmallVector<unsigned, 4> OutputExprStrIdx;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000485
486 // Empty asm statements don't need to instantiate the AsmParser, etc.
Chad Rosierc6916492012-09-06 19:56:25 +0000487 StringRef EmptyAsmStr;
488 if (AsmToks.empty()) { DEF_SIMPLE_MSASM(EmptyAsmStr); return Owned(NS); }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000489
Chad Rosier38c71d32012-08-21 21:56:39 +0000490 std::vector<std::string> AsmStrings;
Chad Rosier9072a022012-08-22 20:30:58 +0000491 std::vector<std::pair<unsigned,unsigned> > AsmTokRanges;
Bob Wilsonb0f6b9c2012-09-24 19:57:59 +0000492 if (buildMSAsmStrings(*this, AsmLoc, AsmToks, AsmStrings, AsmTokRanges))
493 return StmtError();
Chad Rosier38c71d32012-08-21 21:56:39 +0000494
Chad Rosiere78460f2012-08-22 20:57:07 +0000495 std::vector<std::vector<StringRef> > Pieces(AsmStrings.size());
Chad Rosierf0fbd772012-08-22 21:04:07 +0000496 buildMSAsmPieces(AsmStrings, Pieces);
Chad Rosier153f8ec2012-08-22 19:50:28 +0000497
498 bool IsSimple = isSimpleMSAsm(Pieces, Context.getTargetInfo());
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000499
Chad Rosier2735df22012-08-22 19:18:30 +0000500 // AsmParser doesn't fully support these asm statements.
Chad Rosierc6916492012-09-06 19:56:25 +0000501 if (bailOnMSAsm(Pieces)) { DEF_SIMPLE_MSASM(EmptyAsmStr); return Owned(NS); }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000502
503 // Initialize targets and assembly printers/parsers.
504 llvm::InitializeAllTargetInfos();
505 llvm::InitializeAllTargetMCs();
506 llvm::InitializeAllAsmParsers();
507
508 // Get the target specific parser.
509 std::string Error;
510 const std::string &TT = Context.getTargetInfo().getTriple().getTriple();
511 const llvm::Target *TheTarget(llvm::TargetRegistry::lookupTarget(TT, Error));
512
513 OwningPtr<llvm::MCAsmInfo> MAI(TheTarget->createMCAsmInfo(TT));
514 OwningPtr<llvm::MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TT));
515 OwningPtr<llvm::MCObjectFileInfo> MOFI(new llvm::MCObjectFileInfo());
516 OwningPtr<llvm::MCSubtargetInfo>
517 STI(TheTarget->createMCSubtargetInfo(TT, "", ""));
518
Chad Rosier25bd2982012-08-23 15:44:35 +0000519 for (unsigned StrIdx = 0, e = AsmStrings.size(); StrIdx != e; ++StrIdx) {
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000520 llvm::SourceMgr SrcMgr;
521 llvm::MCContext Ctx(*MAI, *MRI, MOFI.get(), &SrcMgr);
522 llvm::MemoryBuffer *Buffer =
Chad Rosier25bd2982012-08-23 15:44:35 +0000523 llvm::MemoryBuffer::getMemBuffer(AsmStrings[StrIdx], "<inline asm>");
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000524
525 // Tell SrcMgr about this buffer, which is what the parser will pick up.
526 SrcMgr.AddNewSourceBuffer(Buffer, llvm::SMLoc());
527
528 OwningPtr<llvm::MCStreamer> Str(createNullStreamer(Ctx));
529 OwningPtr<llvm::MCAsmParser>
530 Parser(createMCAsmParser(SrcMgr, Ctx, *Str.get(), *MAI));
531 OwningPtr<llvm::MCTargetAsmParser>
532 TargetParser(TheTarget->createMCAsmParser(*STI, *Parser));
533 // Change to the Intel dialect.
534 Parser->setAssemblerDialect(1);
535 Parser->setTargetParser(*TargetParser.get());
536
537 // Prime the lexer.
538 Parser->Lex();
539
540 // Parse the opcode.
541 StringRef IDVal;
542 Parser->ParseIdentifier(IDVal);
543
544 // Canonicalize the opcode to lower case.
Chad Rosierbe5c3fb2012-09-03 02:30:13 +0000545 SmallString<128> OpcodeStr;
Chad Rosierb706d902012-08-28 22:08:58 +0000546 for (unsigned i = 0, e = IDVal.size(); i != e; ++i)
Chad Rosierbe5c3fb2012-09-03 02:30:13 +0000547 OpcodeStr.push_back(tolower(IDVal[i]));
Chad Rosier3d442192012-09-21 22:22:39 +0000548 // FIXME: Convert to a StmtError.
549 assert(TargetParser->mnemonicIsValid(OpcodeStr) && "Invalid mnemonic!");
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000550
551 // Parse the operands.
552 llvm::SMLoc IDLoc;
553 SmallVector<llvm::MCParsedAsmOperand*, 8> Operands;
Chad Rosierbe5c3fb2012-09-03 02:30:13 +0000554 bool HadError = TargetParser->ParseInstruction(OpcodeStr.str(), IDLoc,
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000555 Operands);
Chad Rosier2735df22012-08-22 19:18:30 +0000556 // If we had an error parsing the operands, fail gracefully.
Chad Rosierc6916492012-09-06 19:56:25 +0000557 if (HadError) { DEF_SIMPLE_MSASM(EmptyAsmStr); return Owned(NS); }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000558
559 // Match the MCInstr.
Chad Rosierbe5c3fb2012-09-03 02:30:13 +0000560 unsigned Kind;
Chad Rosier83591b62012-08-21 18:15:08 +0000561 unsigned ErrorInfo;
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000562 SmallVector<llvm::MCInst, 2> Instrs;
Chad Rosier7065c522012-09-03 03:16:15 +0000563 HadError = TargetParser->MatchInstruction(IDLoc, Kind, Operands, Instrs,
564 ErrorInfo,
Chad Rosier51a6b3f2012-08-21 19:37:55 +0000565 /*matchingInlineAsm*/ true);
Chad Rosier2735df22012-08-22 19:18:30 +0000566 // If we had an error parsing the operands, fail gracefully.
Chad Rosierc6916492012-09-06 19:56:25 +0000567 if (HadError) { DEF_SIMPLE_MSASM(EmptyAsmStr); return Owned(NS); }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000568
569 // Get the instruction descriptor.
Bob Wilson317d8f32012-09-25 16:30:16 +0000570 llvm::MCInst Inst = Instrs.back();
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000571 const llvm::MCInstrInfo *MII = TheTarget->createMCInstrInfo();
572 const llvm::MCInstrDesc &Desc = MII->get(Inst.getOpcode());
573 llvm::MCInstPrinter *IP =
574 TheTarget->createMCInstPrinter(1, *MAI, *MII, *MRI, *STI);
575
Chad Rosier6e97be72012-08-22 23:42:09 +0000576 // Build the list of clobbers, outputs and inputs.
Chad Rosierfd5e56e2012-08-22 22:10:51 +0000577 unsigned NumDefs = Desc.getNumDefs();
Chad Rosier1b497f22012-09-03 20:40:52 +0000578 for (unsigned i = 1, e = Operands.size(); i != e; ++i) {
Chad Rosierfb700262012-09-11 23:13:15 +0000579 if (Operands[i]->isToken() || Operands[i]->isImm())
580 continue;
581
Chad Rosier3ef95a52012-09-13 00:20:37 +0000582 // FIXME: The getMCInstOperandNum() function does not work with tied
583 // operands or custom converters.
Chad Rosier1b497f22012-09-03 20:40:52 +0000584 unsigned NumMCOperands;
Chad Rosier7935dde2012-09-24 22:58:50 +0000585 unsigned MCIdx = TargetParser->getMCInstOperandNum(Kind, Operands, i,
586 NumMCOperands);
Chad Rosier1b497f22012-09-03 20:40:52 +0000587 assert (NumMCOperands && "Expected at least 1 MCOperand!");
Chad Rosier7f9678b2012-09-12 18:14:25 +0000588
589 for (unsigned j = MCIdx, e = MCIdx + NumMCOperands; j != e; ++j) {
590 const llvm::MCOperand &Op = Inst.getOperand(j);
591
592 // Skip immediates.
593 if (Op.isImm() || Op.isFPImm())
594 continue;
595
596 // Skip invalid register operands.
597 if (Op.isReg() && Op.getReg() == 0)
598 continue;
599
600 // Register/Clobber.
601 if (Op.isReg() && NumDefs && (j < NumDefs)) {
602 std::string Reg;
603 llvm::raw_string_ostream OS(Reg);
604 IP->printRegName(OS, Op.getReg());
605
606 StringRef Clobber(OS.str());
607 if (!Context.getTargetInfo().isValidClobber(Clobber))
608 return StmtError(
609 Diag(AsmLoc, diag::err_asm_unknown_register_name) << Clobber);
610 ClobberRegs.insert(Reg);
611 continue;
Chad Rosier1b497f22012-09-03 20:40:52 +0000612 }
Chad Rosier7f9678b2012-09-12 18:14:25 +0000613 // Expr/Input or Output.
614 if (Op.isExpr()) {
615 const llvm::MCExpr *Expr = Op.getExpr();
616 const llvm::MCSymbolRefExpr *SymRef;
617 if ((SymRef = dyn_cast<llvm::MCSymbolRefExpr>(Expr))) {
618 StringRef Name = SymRef->getSymbol().getName();
619 IdentifierInfo *II = getIdentifierInfo(Name, AsmToks,
620 AsmTokRanges[StrIdx].first,
621 AsmTokRanges[StrIdx].second);
622 if (II) {
623 CXXScopeSpec SS;
624 UnqualifiedId Id;
625 SourceLocation Loc;
626 Id.setIdentifier(II, AsmLoc);
627 ExprResult Result = ActOnIdExpression(getCurScope(), SS, Loc, Id,
628 false, false);
629 if (!Result.isInvalid()) {
Chad Rosier3ef95a52012-09-13 00:20:37 +0000630 // FIXME: Determine the proper constraints.
Chad Rosier7f9678b2012-09-12 18:14:25 +0000631 bool isMemDef = (i == 1) && Desc.mayStore();
632 if (isMemDef) {
633 Outputs.push_back(II);
634 OutputExprs.push_back(Result.take());
635 OutputExprNames.push_back(Name.str());
Chad Rosierb55e6022012-09-13 00:06:55 +0000636 OutputExprStrIdx.push_back(StrIdx);
Chad Rosier7f9678b2012-09-12 18:14:25 +0000637 OutputConstraints.push_back("=r");
638 } else {
639 Inputs.push_back(II);
640 InputExprs.push_back(Result.take());
641 InputExprNames.push_back(Name.str());
Chad Rosierb55e6022012-09-13 00:06:55 +0000642 InputExprStrIdx.push_back(StrIdx);
Chad Rosier7f9678b2012-09-12 18:14:25 +0000643 InputConstraints.push_back("r");
644 }
Chad Rosier1016bdf2012-08-24 16:38:58 +0000645 }
Chad Rosier633abb02012-08-24 00:07:09 +0000646 }
647 }
Chad Rosier6e97be72012-08-22 23:42:09 +0000648 }
Chad Rosierfd5e56e2012-08-22 22:10:51 +0000649 }
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000650 }
651 }
652 for (std::set<std::string>::iterator I = ClobberRegs.begin(),
653 E = ClobberRegs.end(); I != E; ++I)
654 Clobbers.push_back(*I);
655
Chad Rosier89fb6d72012-08-28 20:28:20 +0000656 // Merge the output and input constraints. Output constraints are expected
657 // first.
658 for (std::vector<std::string>::iterator I = OutputConstraints.begin(),
659 E = OutputConstraints.end(); I != E; ++I)
660 Constraints.push_back(*I);
661
662 for (std::vector<std::string>::iterator I = InputConstraints.begin(),
663 E = InputConstraints.end(); I != E; ++I)
664 Constraints.push_back(*I);
665
Chad Rosieracc22b62012-09-06 19:35:00 +0000666 // Enumerate the AsmString expressions.
Chad Rosieracc22b62012-09-06 19:35:00 +0000667 unsigned OpNum = 0;
668 for (unsigned i = 0, e = OutputExprNames.size(); i != e; ++i, ++OpNum) {
Chad Rosierb55e6022012-09-13 00:06:55 +0000669 unsigned StrIdx = OutputExprStrIdx[i];
670 // Iterate over the assembly instruction pieces, skipping the mnemonic.
671 for (unsigned j = 1, f = Pieces[StrIdx].size(); j != f; ++j) {
672 // If the operand and the expression name match, then rewrite the operand.
673 if (OutputExprNames[i] == Pieces[StrIdx][j]) {
674 SmallString<32> Res;
675 llvm::raw_svector_ostream OS(Res);
676 OS << '$' << OpNum;
677 OutputExprNames[i] = OS.str();
678 Pieces[StrIdx][j] = OutputExprNames[i];
679 break;
680 }
681 // Check to see if the expression is a substring of the asm piece.
682 std::pair< StringRef, StringRef > Split = Pieces[StrIdx][j].split(' ');
683 bool isKeyword = llvm::StringSwitch<bool>(Split.first)
684 .Cases("BYTE", "byte", "WORD", "word", "DWORD", true)
685 .Cases("dword", "QWORD", "qword", "XWORD", "xword", true)
686 .Cases("XMMWORD", "xmmword", "YMMWORD", "ymmword", true)
687 .Default(false);
688 if (isKeyword &&
689 Split.second.find_first_of(OutputExprNames[i]) != StringRef::npos) {
690 // Is is a substring, do the replacement.
691 SmallString<32> Res;
692 llvm::raw_svector_ostream OS(Res);
693 OS << '$' << OpNum;
694 std::string piece = Pieces[StrIdx][j].str();
695 size_t found = piece.find(InputExprNames[i]);
696 piece.replace(found, InputExprNames[i].size(), OS.str());
697 OutputExprNames[i] = piece;
698 Pieces[StrIdx][j] = OutputExprNames[i];
699 break;
700 }
701 }
Chad Rosieracc22b62012-09-06 19:35:00 +0000702 }
703 for (unsigned i = 0, e = InputExprNames.size(); i != e; ++i, ++OpNum) {
Chad Rosierb55e6022012-09-13 00:06:55 +0000704 unsigned StrIdx = InputExprStrIdx[i];
705 // Iterate over the assembly instruction pieces, skipping the mnemonic.
706 for (unsigned j = 1, f = Pieces[StrIdx].size(); j != f; ++j) {
707 // If the operand and the expression name match, then rewrite the operand.
708 if (InputExprNames[i] == Pieces[StrIdx][j]) {
709 SmallString<32> Res;
710 llvm::raw_svector_ostream OS(Res);
711 OS << '$' << OpNum;
712 InputExprNames[i] = OS.str();
713 Pieces[StrIdx][j] = InputExprNames[i];
714 break;
715 }
716 // Check to see if the expression is a substring of the asm piece.
717 std::pair< StringRef, StringRef > Split = Pieces[StrIdx][j].split(' ');
718 bool isKeyword = llvm::StringSwitch<bool>(Split.first)
719 .Cases("BYTE", "byte", "WORD", "word", "DWORD", true)
720 .Cases("dword", "QWORD", "qword", "XWORD", "xword", true)
721 .Cases("XMMWORD", "xmmword", "YMMWORD", "ymmword", true)
722 .Default(false);
723 if (isKeyword &&
724 Split.second.find_first_of(InputExprNames[i]) != StringRef::npos) {
725 // It is a substring, do the replacement.
726 SmallString<32> Res;
727 llvm::raw_svector_ostream OS(Res);
728 OS << '$' << OpNum;
729 std::string piece = Pieces[StrIdx][j].str();
730 size_t found = piece.find(InputExprNames[i]);
731 piece.replace(found, InputExprNames[i].size(), OS.str());
732 InputExprNames[i] = piece;
733 Pieces[StrIdx][j] = InputExprNames[i];
734 break;
735 }
736 }
737 }
738
739 // Emit the IR assembly string.
740 std::string AsmString;
741 for (unsigned i = 0, e = Pieces.size(); i != e; ++i) {
742 // Skip empty asm stmts.
743 if (Pieces[i].empty()) continue;
744
745 if (i > 0)
746 AsmString += "\n\t";
747
748 // Emit the mnemonic.
749 AsmString += Pieces[i][0];
750 if (Pieces[i].size() > 1)
751 AsmString += ' ';
752
753 // Emit the operands adding $$ to constants.
754 for (unsigned j = 1, f = Pieces[i].size(); j != f; ++j) {
755 if (j > 1) AsmString += ", ";
756 unsigned Val;
757 if (!Pieces[i][j].getAsInteger(0, Val))
758 AsmString += "$$";
759
760 AsmString += Pieces[i][j];
761 }
Chad Rosieracc22b62012-09-06 19:35:00 +0000762 }
763
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000764 MSAsmStmt *NS =
765 new (Context) MSAsmStmt(Context, AsmLoc, LBraceLoc, IsSimple,
766 /*IsVolatile*/ true, AsmToks, Inputs, Outputs,
Chad Rosier89fb6d72012-08-28 20:28:20 +0000767 InputExprs, OutputExprs, AsmString, Constraints,
768 Clobbers, EndLoc);
Chad Rosier4b5e48d2012-08-17 21:19:40 +0000769 return Owned(NS);
770}