Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 1 | //===- Parser.cpp - MLIR Parser Implementation ----------------------------===// |
| 2 | // |
| 3 | // Copyright 2019 The MLIR Authors. |
| 4 | // |
| 5 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | // you may not use this file except in compliance with the License. |
| 7 | // You may obtain a copy of the License at |
| 8 | // |
| 9 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | // |
| 11 | // Unless required by applicable law or agreed to in writing, software |
| 12 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | // See the License for the specific language governing permissions and |
| 15 | // limitations under the License. |
| 16 | // ============================================================================= |
| 17 | // |
| 18 | // This file implements the parser for the MLIR textual form. |
| 19 | // |
| 20 | //===----------------------------------------------------------------------===// |
| 21 | |
| 22 | #include "mlir/Parser.h" |
| 23 | #include "Lexer.h" |
| 24 | #include "mlir/IR/Module.h" |
| 25 | #include "llvm/Support/SourceMgr.h" |
| 26 | using namespace mlir; |
| 27 | using llvm::SourceMgr; |
| 28 | |
| 29 | namespace { |
| 30 | /// Simple enum to make code read better. Failure is "true" in a boolean |
| 31 | /// context. |
| 32 | enum ParseResult { |
| 33 | ParseSuccess, |
| 34 | ParseFailure |
| 35 | }; |
| 36 | |
| 37 | /// Main parser implementation. |
| 38 | class Parser { |
| 39 | public: |
| 40 | Parser(llvm::SourceMgr &sourceMgr) : lex(sourceMgr), curToken(lex.lexToken()){ |
| 41 | module.reset(new Module()); |
| 42 | } |
| 43 | |
| 44 | Module *parseModule(); |
| 45 | private: |
| 46 | // State. |
| 47 | Lexer lex; |
| 48 | |
| 49 | // This is the next token that hasn't been consumed yet. |
| 50 | Token curToken; |
| 51 | |
| 52 | // This is the result module we are parsing into. |
| 53 | std::unique_ptr<Module> module; |
| 54 | |
| 55 | private: |
| 56 | // Helper methods. |
| 57 | |
| 58 | /// Emit an error and return failure. |
| 59 | ParseResult emitError(const Twine &message); |
| 60 | |
| 61 | /// Advance the current lexer onto the next token. |
| 62 | void consumeToken() { |
| 63 | assert(curToken.isNot(Token::eof, Token::error) && |
| 64 | "shouldn't advance past EOF or errors"); |
| 65 | curToken = lex.lexToken(); |
| 66 | } |
| 67 | |
| 68 | /// Advance the current lexer onto the next token, asserting what the expected |
| 69 | /// current token is. This is preferred to the above method because it leads |
| 70 | /// to more self-documenting code with better checking. |
| 71 | void consumeToken(Token::TokenKind kind) { |
| 72 | assert(curToken.is(kind) && "consumed an unexpected token"); |
| 73 | consumeToken(); |
| 74 | } |
| 75 | |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 76 | /// If the current token has the specified kind, consume it and return true. |
| 77 | /// If not, return false. |
| 78 | bool consumeIf(Token::TokenKind kind) { |
| 79 | if (curToken.isNot(kind)) |
| 80 | return false; |
| 81 | consumeToken(kind); |
| 82 | return true; |
| 83 | } |
| 84 | |
| 85 | ParseResult parseCommaSeparatedList(Token::TokenKind rightToken, |
| 86 | const std::function<ParseResult()> &parseElement, |
| 87 | bool allowEmptyList = true); |
| 88 | |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 89 | // Type parsing. |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 90 | ParseResult parsePrimitiveType(); |
| 91 | ParseResult parseElementType(); |
| 92 | ParseResult parseVectorType(); |
| 93 | ParseResult parseDimensionListRanked(SmallVectorImpl<int> &dimensions); |
| 94 | ParseResult parseTensorType(); |
| 95 | ParseResult parseMemRefType(); |
| 96 | ParseResult parseFunctionType(); |
| 97 | ParseResult parseType(); |
| 98 | ParseResult parseTypeList(); |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 99 | |
| 100 | // Top level entity parsing. |
| 101 | ParseResult parseFunctionSignature(StringRef &name); |
| 102 | ParseResult parseExtFunc(); |
| 103 | }; |
| 104 | } // end anonymous namespace |
| 105 | |
| 106 | //===----------------------------------------------------------------------===// |
| 107 | // Helper methods. |
| 108 | //===----------------------------------------------------------------------===// |
| 109 | |
| 110 | ParseResult Parser::emitError(const Twine &message) { |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 111 | // If we hit a parse error in response to a lexer error, then the lexer |
| 112 | // already emitted an error. |
| 113 | if (curToken.is(Token::error)) |
| 114 | return ParseFailure; |
| 115 | |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 116 | // TODO(clattner): If/when we want to implement a -verify mode, this will need |
| 117 | // to package up errors into SMDiagnostic and report them. |
| 118 | lex.getSourceMgr().PrintMessage(curToken.getLoc(), SourceMgr::DK_Error, |
| 119 | message); |
| 120 | return ParseFailure; |
| 121 | } |
| 122 | |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 123 | /// Parse a comma-separated list of elements, terminated with an arbitrary |
| 124 | /// token. This allows empty lists if allowEmptyList is true. |
| 125 | /// |
| 126 | /// abstract-list ::= rightToken // if allowEmptyList == true |
| 127 | /// abstract-list ::= element (',' element)* rightToken |
| 128 | /// |
| 129 | ParseResult Parser:: |
| 130 | parseCommaSeparatedList(Token::TokenKind rightToken, |
| 131 | const std::function<ParseResult()> &parseElement, |
| 132 | bool allowEmptyList) { |
| 133 | // Handle the empty case. |
| 134 | if (curToken.is(rightToken)) { |
| 135 | if (!allowEmptyList) |
| 136 | return emitError("expected list element"); |
| 137 | consumeToken(rightToken); |
| 138 | return ParseSuccess; |
| 139 | } |
| 140 | |
| 141 | // Non-empty case starts with an element. |
| 142 | if (parseElement()) |
| 143 | return ParseFailure; |
| 144 | |
| 145 | // Otherwise we have a list of comma separated elements. |
| 146 | while (consumeIf(Token::comma)) { |
| 147 | if (parseElement()) |
| 148 | return ParseFailure; |
| 149 | } |
| 150 | |
| 151 | // Consume the end character. |
| 152 | if (!consumeIf(rightToken)) |
| 153 | return emitError("expected ',' or ')'"); |
| 154 | |
| 155 | return ParseSuccess; |
| 156 | } |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 157 | |
| 158 | //===----------------------------------------------------------------------===// |
| 159 | // Type Parsing |
| 160 | //===----------------------------------------------------------------------===// |
| 161 | |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 162 | /// Parse the low-level fixed dtypes in the system. |
| 163 | /// |
| 164 | /// primitive-type |
| 165 | /// ::= `f16` | `bf16` | `f32` | `f64` // Floating point |
| 166 | /// | `i1` | `i8` | `i16` | `i32` | `i64` // Sized integers |
| 167 | /// | `int` |
| 168 | /// |
| 169 | ParseResult Parser::parsePrimitiveType() { |
| 170 | // TODO: Build IR objects. |
| 171 | switch (curToken.getKind()) { |
| 172 | default: return emitError("expected type"); |
| 173 | case Token::kw_bf16: |
| 174 | consumeToken(Token::kw_bf16); |
| 175 | return ParseSuccess; |
| 176 | case Token::kw_f16: |
| 177 | consumeToken(Token::kw_f16); |
| 178 | return ParseSuccess; |
| 179 | case Token::kw_f32: |
| 180 | consumeToken(Token::kw_f32); |
| 181 | return ParseSuccess; |
| 182 | case Token::kw_f64: |
| 183 | consumeToken(Token::kw_f64); |
| 184 | return ParseSuccess; |
| 185 | case Token::kw_i1: |
| 186 | consumeToken(Token::kw_i1); |
| 187 | return ParseSuccess; |
| 188 | case Token::kw_i16: |
| 189 | consumeToken(Token::kw_i16); |
| 190 | return ParseSuccess; |
| 191 | case Token::kw_i32: |
| 192 | consumeToken(Token::kw_i32); |
| 193 | return ParseSuccess; |
| 194 | case Token::kw_i64: |
| 195 | consumeToken(Token::kw_i64); |
| 196 | return ParseSuccess; |
| 197 | case Token::kw_i8: |
| 198 | consumeToken(Token::kw_i8); |
| 199 | return ParseSuccess; |
| 200 | case Token::kw_int: |
| 201 | consumeToken(Token::kw_int); |
| 202 | return ParseSuccess; |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | /// Parse the element type of a tensor or memref type. |
| 207 | /// |
| 208 | /// element-type ::= primitive-type | vector-type |
| 209 | /// |
| 210 | ParseResult Parser::parseElementType() { |
| 211 | if (curToken.is(Token::kw_vector)) |
| 212 | return parseVectorType(); |
| 213 | |
| 214 | return parsePrimitiveType(); |
| 215 | } |
| 216 | |
| 217 | /// Parse a vector type. |
| 218 | /// |
| 219 | /// vector-type ::= `vector` `<` const-dimension-list primitive-type `>` |
| 220 | /// const-dimension-list ::= (integer-literal `x`)+ |
| 221 | /// |
| 222 | ParseResult Parser::parseVectorType() { |
| 223 | consumeToken(Token::kw_vector); |
| 224 | |
| 225 | if (!consumeIf(Token::less)) |
| 226 | return emitError("expected '<' in vector type"); |
| 227 | |
| 228 | if (curToken.isNot(Token::integer)) |
| 229 | return emitError("expected dimension size in vector type"); |
| 230 | |
| 231 | SmallVector<unsigned, 4> dimensions; |
| 232 | while (curToken.is(Token::integer)) { |
| 233 | // Make sure this integer value is in bound and valid. |
| 234 | auto dimension = curToken.getUnsignedIntegerValue(); |
| 235 | if (!dimension.hasValue()) |
| 236 | return emitError("invalid dimension in vector type"); |
| 237 | dimensions.push_back(dimension.getValue()); |
| 238 | |
| 239 | consumeToken(Token::integer); |
| 240 | |
| 241 | // Make sure we have an 'x' or something like 'xbf32'. |
| 242 | if (curToken.isNot(Token::bare_identifier) || |
| 243 | curToken.getSpelling()[0] != 'x') |
| 244 | return emitError("expected 'x' in vector dimension list"); |
| 245 | |
| 246 | // If we had a prefix of 'x', lex the next token immediately after the 'x'. |
| 247 | if (curToken.getSpelling().size() != 1) |
| 248 | lex.resetPointer(curToken.getSpelling().data()+1); |
| 249 | |
| 250 | // Consume the 'x'. |
| 251 | consumeToken(Token::bare_identifier); |
| 252 | } |
| 253 | |
| 254 | // Parse the element type. |
| 255 | if (parsePrimitiveType()) |
| 256 | return ParseFailure; |
| 257 | |
| 258 | if (!consumeIf(Token::greater)) |
| 259 | return emitError("expected '>' in vector type"); |
| 260 | |
| 261 | // TODO: Form IR object. |
| 262 | |
| 263 | return ParseSuccess; |
| 264 | } |
| 265 | |
| 266 | /// Parse a dimension list of a tensor or memref type. This populates the |
| 267 | /// dimension list, returning -1 for the '?' dimensions. |
| 268 | /// |
| 269 | /// dimension-list-ranked ::= (dimension `x`)* |
| 270 | /// dimension ::= `?` | integer-literal |
| 271 | /// |
| 272 | ParseResult Parser::parseDimensionListRanked(SmallVectorImpl<int> &dimensions) { |
| 273 | while (curToken.isAny(Token::integer, Token::question)) { |
| 274 | if (consumeIf(Token::question)) { |
| 275 | dimensions.push_back(-1); |
| 276 | } else { |
| 277 | // Make sure this integer value is in bound and valid. |
| 278 | auto dimension = curToken.getUnsignedIntegerValue(); |
| 279 | if (!dimension.hasValue() || (int)dimension.getValue() < 0) |
| 280 | return emitError("invalid dimension"); |
| 281 | dimensions.push_back((int)dimension.getValue()); |
| 282 | consumeToken(Token::integer); |
| 283 | } |
| 284 | |
| 285 | // Make sure we have an 'x' or something like 'xbf32'. |
| 286 | if (curToken.isNot(Token::bare_identifier) || |
| 287 | curToken.getSpelling()[0] != 'x') |
| 288 | return emitError("expected 'x' in dimension list"); |
| 289 | |
| 290 | // If we had a prefix of 'x', lex the next token immediately after the 'x'. |
| 291 | if (curToken.getSpelling().size() != 1) |
| 292 | lex.resetPointer(curToken.getSpelling().data()+1); |
| 293 | |
| 294 | // Consume the 'x'. |
| 295 | consumeToken(Token::bare_identifier); |
| 296 | } |
| 297 | |
| 298 | return ParseSuccess; |
| 299 | } |
| 300 | |
| 301 | /// Parse a tensor type. |
| 302 | /// |
| 303 | /// tensor-type ::= `tensor` `<` dimension-list element-type `>` |
| 304 | /// dimension-list ::= dimension-list-ranked | `??` |
| 305 | /// |
| 306 | ParseResult Parser::parseTensorType() { |
| 307 | consumeToken(Token::kw_tensor); |
| 308 | |
| 309 | if (!consumeIf(Token::less)) |
| 310 | return emitError("expected '<' in tensor type"); |
| 311 | |
| 312 | bool isUnranked; |
| 313 | SmallVector<int, 4> dimensions; |
| 314 | |
| 315 | if (consumeIf(Token::questionquestion)) { |
| 316 | isUnranked = true; |
| 317 | } else { |
| 318 | isUnranked = false; |
| 319 | if (parseDimensionListRanked(dimensions)) |
| 320 | return ParseFailure; |
| 321 | } |
| 322 | |
| 323 | // Parse the element type. |
| 324 | if (parseElementType()) |
| 325 | return ParseFailure; |
| 326 | |
| 327 | if (!consumeIf(Token::greater)) |
| 328 | return emitError("expected '>' in tensor type"); |
| 329 | |
| 330 | // TODO: Form IR object. |
| 331 | |
| 332 | return ParseSuccess; |
| 333 | } |
| 334 | |
| 335 | /// Parse a memref type. |
| 336 | /// |
| 337 | /// memref-type ::= `memref` `<` dimension-list-ranked element-type |
| 338 | /// (`,` semi-affine-map-composition)? (`,` memory-space)? `>` |
| 339 | /// |
| 340 | /// semi-affine-map-composition ::= (semi-affine-map `,` )* semi-affine-map |
| 341 | /// memory-space ::= integer-literal /* | TODO: address-space-id */ |
| 342 | /// |
| 343 | ParseResult Parser::parseMemRefType() { |
| 344 | consumeToken(Token::kw_memref); |
| 345 | |
| 346 | if (!consumeIf(Token::less)) |
| 347 | return emitError("expected '<' in memref type"); |
| 348 | |
| 349 | SmallVector<int, 4> dimensions; |
| 350 | if (parseDimensionListRanked(dimensions)) |
| 351 | return ParseFailure; |
| 352 | |
| 353 | // Parse the element type. |
| 354 | if (parseElementType()) |
| 355 | return ParseFailure; |
| 356 | |
| 357 | // TODO: Parse semi-affine-map-composition. |
| 358 | // TODO: Parse memory-space. |
| 359 | |
| 360 | if (!consumeIf(Token::greater)) |
| 361 | return emitError("expected '>' in memref type"); |
| 362 | |
| 363 | // TODO: Form IR object. |
| 364 | |
| 365 | return ParseSuccess; |
| 366 | } |
| 367 | |
| 368 | |
| 369 | |
| 370 | /// Parse a function type. |
| 371 | /// |
| 372 | /// function-type ::= type-list-parens `->` type-list |
| 373 | /// |
| 374 | ParseResult Parser::parseFunctionType() { |
| 375 | assert(curToken.is(Token::l_paren)); |
| 376 | |
| 377 | if (parseTypeList()) |
| 378 | return ParseFailure; |
| 379 | |
| 380 | if (!consumeIf(Token::arrow)) |
| 381 | return emitError("expected '->' in function type"); |
| 382 | |
| 383 | if (parseTypeList()) |
| 384 | return ParseFailure; |
| 385 | |
| 386 | // TODO: Build IR object. |
| 387 | return ParseSuccess; |
| 388 | } |
| 389 | |
| 390 | |
| 391 | /// Parse an arbitrary type. |
| 392 | /// |
| 393 | /// type ::= primitive-type |
| 394 | /// | vector-type |
| 395 | /// | tensor-type |
| 396 | /// | memref-type |
| 397 | /// | function-type |
| 398 | /// element-type ::= primitive-type | vector-type |
| 399 | /// |
| 400 | ParseResult Parser::parseType() { |
| 401 | switch (curToken.getKind()) { |
| 402 | case Token::kw_memref: return parseMemRefType(); |
| 403 | case Token::kw_tensor: return parseTensorType(); |
| 404 | case Token::kw_vector: return parseVectorType(); |
| 405 | case Token::l_paren: return parseFunctionType(); |
| 406 | default: |
| 407 | return parsePrimitiveType(); |
| 408 | } |
| 409 | } |
| 410 | |
| 411 | /// Parse a "type list", which is a singular type, or a parenthesized list of |
| 412 | /// types. |
| 413 | /// |
| 414 | /// type-list ::= type-list-parens | type |
| 415 | /// type-list-parens ::= `(` `)` |
| 416 | /// | `(` type (`,` type)* `)` |
| 417 | /// |
| 418 | ParseResult Parser::parseTypeList() { |
| 419 | // If there is no parens, then it must be a singular type. |
| 420 | if (!consumeIf(Token::l_paren)) |
| 421 | return parseType(); |
| 422 | |
| 423 | if (parseCommaSeparatedList(Token::r_paren, |
| 424 | [&]() -> ParseResult { |
| 425 | // TODO: Add to list of IR values we're parsing. |
| 426 | return parseType(); |
| 427 | })) { |
| 428 | return ParseFailure; |
| 429 | } |
| 430 | |
| 431 | // TODO: Build IR objects. |
| 432 | return ParseSuccess; |
| 433 | } |
| 434 | |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 435 | |
| 436 | //===----------------------------------------------------------------------===// |
| 437 | // Top-level entity parsing. |
| 438 | //===----------------------------------------------------------------------===// |
| 439 | |
| 440 | /// Parse a function signature, starting with a name and including the parameter |
| 441 | /// list. |
| 442 | /// |
| 443 | /// argument-list ::= type (`,` type)* | /*empty*/ |
| 444 | /// function-signature ::= function-id `(` argument-list `)` (`->` type-list)? |
| 445 | /// |
| 446 | ParseResult Parser::parseFunctionSignature(StringRef &name) { |
| 447 | if (curToken.isNot(Token::at_identifier)) |
| 448 | return emitError("expected a function identifier like '@foo'"); |
| 449 | |
| 450 | name = curToken.getSpelling().drop_front(); |
| 451 | consumeToken(Token::at_identifier); |
| 452 | |
| 453 | if (curToken.isNot(Token::l_paren)) |
| 454 | return emitError("expected '(' in function signature"); |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 455 | |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 456 | if (parseTypeList()) |
| 457 | return ParseFailure; |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 458 | |
Chris Lattner | bb8fafc | 2018-06-22 15:52:02 -0700 | [diff] [blame^] | 459 | // Parse the return type if present. |
| 460 | if (consumeIf(Token::arrow)) { |
| 461 | if (parseTypeList()) |
| 462 | return ParseFailure; |
| 463 | |
| 464 | // TODO: Build IR object. |
| 465 | } |
Chris Lattner | e79379a | 2018-06-22 10:39:19 -0700 | [diff] [blame] | 466 | |
| 467 | return ParseSuccess; |
| 468 | } |
| 469 | |
| 470 | |
| 471 | /// External function declarations. |
| 472 | /// |
| 473 | /// ext-func ::= `extfunc` function-signature |
| 474 | /// |
| 475 | ParseResult Parser::parseExtFunc() { |
| 476 | consumeToken(Token::kw_extfunc); |
| 477 | |
| 478 | StringRef name; |
| 479 | if (parseFunctionSignature(name)) |
| 480 | return ParseFailure; |
| 481 | |
| 482 | |
| 483 | // Okay, the external function definition was parsed correctly. |
| 484 | module->functionList.push_back(new Function(name)); |
| 485 | return ParseSuccess; |
| 486 | } |
| 487 | |
| 488 | |
| 489 | /// This is the top-level module parser. |
| 490 | Module *Parser::parseModule() { |
| 491 | while (1) { |
| 492 | switch (curToken.getKind()) { |
| 493 | default: |
| 494 | emitError("expected a top level entity"); |
| 495 | return nullptr; |
| 496 | |
| 497 | // If we got to the end of the file, then we're done. |
| 498 | case Token::eof: |
| 499 | return module.release(); |
| 500 | |
| 501 | // If we got an error token, then the lexer already emitted an error, just |
| 502 | // stop. Someday we could introduce error recovery if there was demand for |
| 503 | // it. |
| 504 | case Token::error: |
| 505 | return nullptr; |
| 506 | |
| 507 | case Token::kw_extfunc: |
| 508 | if (parseExtFunc()) |
| 509 | return nullptr; |
| 510 | break; |
| 511 | |
| 512 | // TODO: cfgfunc, mlfunc, affine entity declarations, etc. |
| 513 | } |
| 514 | } |
| 515 | } |
| 516 | |
| 517 | //===----------------------------------------------------------------------===// |
| 518 | |
| 519 | /// This parses the file specified by the indicated SourceMgr and returns an |
| 520 | /// MLIR module if it was valid. If not, it emits diagnostics and returns null. |
| 521 | Module *mlir::parseSourceFile(llvm::SourceMgr &sourceMgr) { |
| 522 | return Parser(sourceMgr).parseModule(); |
| 523 | } |