Olivier Deprez | f4ef2d0 | 2021-04-20 13:36:24 +0200 | [diff] [blame] | 1 | //===- llvm/Bitcode/BitcodeConvenience.h - Convenience Wrappers -*- C++ -*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | /// |
| 9 | /// \file Convenience wrappers for the LLVM bitcode format and bitstream APIs. |
| 10 | /// |
| 11 | /// This allows you to use a sort of DSL to declare and use bitcode |
| 12 | /// abbreviations and records. Example: |
| 13 | /// |
| 14 | /// \code |
| 15 | /// using Metadata = BCRecordLayout< |
| 16 | /// METADATA_ID, // ID |
| 17 | /// BCFixed<16>, // Module format major version |
| 18 | /// BCFixed<16>, // Module format minor version |
| 19 | /// BCBlob // misc. version information |
| 20 | /// >; |
| 21 | /// Metadata metadata(Out); |
| 22 | /// metadata.emit(ScratchRecord, VERSION_MAJOR, VERSION_MINOR, Data); |
| 23 | /// \endcode |
| 24 | /// |
| 25 | /// For details on the bitcode format, see |
| 26 | /// http://llvm.org/docs/BitCodeFormat.html |
| 27 | /// |
| 28 | //===----------------------------------------------------------------------===// |
| 29 | |
| 30 | #ifndef LLVM_BITCODE_BITCODECONVENIENCE_H |
| 31 | #define LLVM_BITCODE_BITCODECONVENIENCE_H |
| 32 | |
| 33 | #include "llvm/Bitstream/BitCodes.h" |
| 34 | #include "llvm/Bitstream/BitstreamWriter.h" |
| 35 | #include <cstdint> |
| 36 | |
| 37 | namespace llvm { |
| 38 | namespace detail { |
| 39 | /// Convenience base for all kinds of bitcode abbreviation fields. |
| 40 | /// |
| 41 | /// This just defines common properties queried by the metaprogramming. |
| 42 | template <bool Compound = false> class BCField { |
| 43 | public: |
| 44 | static const bool IsCompound = Compound; |
| 45 | |
| 46 | /// Asserts that the given data is a valid value for this field. |
| 47 | template <typename T> static void assertValid(const T &data) {} |
| 48 | |
| 49 | /// Converts a raw numeric representation of this value to its preferred |
| 50 | /// type. |
| 51 | template <typename T> static T convert(T rawValue) { return rawValue; } |
| 52 | }; |
| 53 | } // namespace detail |
| 54 | |
| 55 | /// Represents a literal operand in a bitcode record. |
| 56 | /// |
| 57 | /// The value of a literal operand is the same for all instances of the record, |
| 58 | /// so it is only emitted in the abbreviation definition. |
| 59 | /// |
| 60 | /// Note that because this uses a compile-time template, you cannot have a |
| 61 | /// literal operand that is fixed at run-time without dropping down to the |
| 62 | /// raw LLVM APIs. |
| 63 | template <uint64_t Value> class BCLiteral : public detail::BCField<> { |
| 64 | public: |
| 65 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 66 | abbrev.Add(llvm::BitCodeAbbrevOp(Value)); |
| 67 | } |
| 68 | |
| 69 | template <typename T> static void assertValid(const T &data) { |
| 70 | assert(data == Value && "data value does not match declared literal value"); |
| 71 | } |
| 72 | }; |
| 73 | |
| 74 | /// Represents a fixed-width value in a bitcode record. |
| 75 | /// |
| 76 | /// Note that the LLVM bitcode format only supports unsigned values. |
| 77 | template <unsigned Width> class BCFixed : public detail::BCField<> { |
| 78 | public: |
| 79 | static_assert(Width <= 64, "fixed-width field is too large"); |
| 80 | |
| 81 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 82 | abbrev.Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::Fixed, Width)); |
| 83 | } |
| 84 | |
| 85 | static void assertValid(const bool &data) { |
| 86 | assert(llvm::isUInt<Width>(data) && |
| 87 | "data value does not fit in the given bit width"); |
| 88 | } |
| 89 | |
| 90 | template <typename T> static void assertValid(const T &data) { |
| 91 | assert(data >= 0 && "cannot encode signed integers"); |
| 92 | assert(llvm::isUInt<Width>(data) && |
| 93 | "data value does not fit in the given bit width"); |
| 94 | } |
| 95 | }; |
| 96 | |
| 97 | /// Represents a variable-width value in a bitcode record. |
| 98 | /// |
| 99 | /// The \p Width parameter should include the continuation bit. |
| 100 | /// |
| 101 | /// Note that the LLVM bitcode format only supports unsigned values. |
| 102 | template <unsigned Width> class BCVBR : public detail::BCField<> { |
| 103 | static_assert(Width >= 2, "width does not have room for continuation bit"); |
| 104 | |
| 105 | public: |
| 106 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 107 | abbrev.Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::VBR, Width)); |
| 108 | } |
| 109 | |
| 110 | template <typename T> static void assertValid(const T &data) { |
| 111 | assert(data >= 0 && "cannot encode signed integers"); |
| 112 | } |
| 113 | }; |
| 114 | |
| 115 | /// Represents a character encoded in LLVM's Char6 encoding. |
| 116 | /// |
| 117 | /// This format is suitable for encoding decimal numbers (without signs or |
| 118 | /// exponents) and C identifiers (without dollar signs), but not much else. |
| 119 | /// |
| 120 | /// \sa http://llvm.org/docs/BitCodeFormat.html#char6-encoded-value |
| 121 | class BCChar6 : public detail::BCField<> { |
| 122 | public: |
| 123 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 124 | abbrev.Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::Char6)); |
| 125 | } |
| 126 | |
| 127 | template <typename T> static void assertValid(const T &data) { |
| 128 | assert(llvm::BitCodeAbbrevOp::isChar6(data) && "invalid Char6 data"); |
| 129 | } |
| 130 | |
| 131 | template <typename T> char convert(T rawValue) { |
| 132 | return static_cast<char>(rawValue); |
| 133 | } |
| 134 | }; |
| 135 | |
| 136 | /// Represents an untyped blob of bytes. |
| 137 | /// |
| 138 | /// If present, this must be the last field in a record. |
| 139 | class BCBlob : public detail::BCField<true> { |
| 140 | public: |
| 141 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 142 | abbrev.Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::Blob)); |
| 143 | } |
| 144 | }; |
| 145 | |
| 146 | /// Represents an array of some other type. |
| 147 | /// |
| 148 | /// If present, this must be the last field in a record. |
| 149 | template <typename ElementTy> class BCArray : public detail::BCField<true> { |
| 150 | static_assert(!ElementTy::IsCompound, "arrays can only contain scalar types"); |
| 151 | |
| 152 | public: |
| 153 | static void emitOp(llvm::BitCodeAbbrev &abbrev) { |
| 154 | abbrev.Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::Array)); |
| 155 | ElementTy::emitOp(abbrev); |
| 156 | } |
| 157 | }; |
| 158 | |
| 159 | namespace detail { |
| 160 | /// Attaches the last field to an abbreviation. |
| 161 | /// |
| 162 | /// This is the base case for \c emitOps. |
| 163 | /// |
| 164 | /// \sa BCRecordLayout::emitAbbrev |
| 165 | template <typename FieldTy> static void emitOps(llvm::BitCodeAbbrev &abbrev) { |
| 166 | FieldTy::emitOp(abbrev); |
| 167 | } |
| 168 | |
| 169 | /// Attaches fields to an abbreviation. |
| 170 | /// |
| 171 | /// This is the recursive case for \c emitOps. |
| 172 | /// |
| 173 | /// \sa BCRecordLayout::emitAbbrev |
| 174 | template <typename FieldTy, typename Next, typename... Rest> |
| 175 | static void emitOps(llvm::BitCodeAbbrev &abbrev) { |
| 176 | static_assert(!FieldTy::IsCompound, |
| 177 | "arrays and blobs may not appear in the middle of a record"); |
| 178 | FieldTy::emitOp(abbrev); |
| 179 | emitOps<Next, Rest...>(abbrev); |
| 180 | } |
| 181 | |
| 182 | /// Helper class for dealing with a scalar element in the middle of a record. |
| 183 | /// |
| 184 | /// \sa BCRecordLayout |
| 185 | template <typename ElementTy, typename... Fields> class BCRecordCoding { |
| 186 | public: |
| 187 | template <typename BufferTy, typename ElementDataTy, typename... DataTy> |
| 188 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 189 | unsigned code, ElementDataTy element, DataTy &&...data) { |
| 190 | static_assert(!ElementTy::IsCompound, |
| 191 | "arrays and blobs may not appear in the middle of a record"); |
| 192 | ElementTy::assertValid(element); |
| 193 | buffer.push_back(element); |
| 194 | BCRecordCoding<Fields...>::emit(Stream, buffer, code, |
| 195 | std::forward<DataTy>(data)...); |
| 196 | } |
| 197 | |
| 198 | template <typename T, typename ElementDataTy, typename... DataTy> |
| 199 | static void read(ArrayRef<T> buffer, ElementDataTy &element, |
| 200 | DataTy &&...data) { |
| 201 | assert(!buffer.empty() && "too few elements in buffer"); |
| 202 | element = ElementTy::convert(buffer.front()); |
| 203 | BCRecordCoding<Fields...>::read(buffer.slice(1), |
| 204 | std::forward<DataTy>(data)...); |
| 205 | } |
| 206 | |
| 207 | template <typename T, typename... DataTy> |
| 208 | static void read(ArrayRef<T> buffer, NoneType, DataTy &&...data) { |
| 209 | assert(!buffer.empty() && "too few elements in buffer"); |
| 210 | BCRecordCoding<Fields...>::read(buffer.slice(1), |
| 211 | std::forward<DataTy>(data)...); |
| 212 | } |
| 213 | }; |
| 214 | |
| 215 | /// Helper class for dealing with a scalar element at the end of a record. |
| 216 | /// |
| 217 | /// This has a separate implementation because up until now we've only been |
| 218 | /// \em building the record (into a data buffer), and now we need to hand it |
| 219 | /// off to the BitstreamWriter to be emitted. |
| 220 | /// |
| 221 | /// \sa BCRecordLayout |
| 222 | template <typename ElementTy> class BCRecordCoding<ElementTy> { |
| 223 | public: |
| 224 | template <typename BufferTy, typename DataTy> |
| 225 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 226 | unsigned code, const DataTy &data) { |
| 227 | static_assert(!ElementTy::IsCompound, |
| 228 | "arrays and blobs need special handling"); |
| 229 | ElementTy::assertValid(data); |
| 230 | buffer.push_back(data); |
| 231 | Stream.EmitRecordWithAbbrev(code, buffer); |
| 232 | } |
| 233 | |
| 234 | template <typename T, typename DataTy> |
| 235 | static void read(ArrayRef<T> buffer, DataTy &data) { |
| 236 | assert(buffer.size() == 1 && "record data does not match layout"); |
| 237 | data = ElementTy::convert(buffer.front()); |
| 238 | } |
| 239 | |
| 240 | template <typename T> static void read(ArrayRef<T> buffer, NoneType) { |
| 241 | assert(buffer.size() == 1 && "record data does not match layout"); |
| 242 | (void)buffer; |
| 243 | } |
| 244 | |
| 245 | template <typename T> static void read(ArrayRef<T> buffer) = delete; |
| 246 | }; |
| 247 | |
| 248 | /// Helper class for dealing with an array at the end of a record. |
| 249 | /// |
| 250 | /// \sa BCRecordLayout::emitRecord |
| 251 | template <typename ElementTy> class BCRecordCoding<BCArray<ElementTy>> { |
| 252 | public: |
| 253 | template <typename BufferTy> |
| 254 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 255 | unsigned code, StringRef data) { |
| 256 | // TODO: validate array data. |
| 257 | Stream.EmitRecordWithArray(code, buffer, data); |
| 258 | } |
| 259 | |
| 260 | template <typename BufferTy, typename ArrayTy> |
| 261 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 262 | unsigned code, const ArrayTy &array) { |
| 263 | #ifndef NDEBUG |
| 264 | for (auto &element : array) |
| 265 | ElementTy::assertValid(element); |
| 266 | #endif |
| 267 | buffer.reserve(buffer.size() + std::distance(array.begin(), array.end())); |
| 268 | std::copy(array.begin(), array.end(), std::back_inserter(buffer)); |
| 269 | Stream.EmitRecordWithAbbrev(code, buffer); |
| 270 | } |
| 271 | |
| 272 | template <typename BufferTy, typename ElementDataTy, typename... DataTy> |
| 273 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 274 | unsigned code, ElementDataTy element, DataTy... data) { |
| 275 | std::array<ElementDataTy, 1 + sizeof...(data)> array{{element, data...}}; |
| 276 | emit(Stream, buffer, code, array); |
| 277 | } |
| 278 | |
| 279 | template <typename BufferTy> |
| 280 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &Buffer, |
| 281 | unsigned code, NoneType) { |
| 282 | Stream.EmitRecordWithAbbrev(code, Buffer); |
| 283 | } |
| 284 | |
| 285 | template <typename T> |
| 286 | static void read(ArrayRef<T> Buffer, ArrayRef<T> &rawData) { |
| 287 | rawData = Buffer; |
| 288 | } |
| 289 | |
| 290 | template <typename T, typename ArrayTy> |
| 291 | static void read(ArrayRef<T> buffer, ArrayTy &array) { |
| 292 | array.append(llvm::map_iterator(buffer.begin(), T::convert), |
| 293 | llvm::map_iterator(buffer.end(), T::convert)); |
| 294 | } |
| 295 | |
| 296 | template <typename T> static void read(ArrayRef<T> buffer, NoneType) { |
| 297 | (void)buffer; |
| 298 | } |
| 299 | |
| 300 | template <typename T> static void read(ArrayRef<T> buffer) = delete; |
| 301 | }; |
| 302 | |
| 303 | /// Helper class for dealing with a blob at the end of a record. |
| 304 | /// |
| 305 | /// \sa BCRecordLayout |
| 306 | template <> class BCRecordCoding<BCBlob> { |
| 307 | public: |
| 308 | template <typename BufferTy> |
| 309 | static void emit(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 310 | unsigned code, StringRef data) { |
| 311 | Stream.EmitRecordWithBlob(code, buffer, data); |
| 312 | } |
| 313 | |
| 314 | template <typename T> static void read(ArrayRef<T> buffer) { (void)buffer; } |
| 315 | |
| 316 | /// Blob data is not stored in the buffer if you are using the correct |
| 317 | /// accessor; this method should not be used. |
| 318 | template <typename T, typename DataTy> |
| 319 | static void read(ArrayRef<T> buffer, DataTy &data) = delete; |
| 320 | }; |
| 321 | |
| 322 | /// A type trait whose \c type field is the last of its template parameters. |
| 323 | template <typename Head, typename... Tail> struct last_type { |
| 324 | using type = typename last_type<Tail...>::type; |
| 325 | }; |
| 326 | |
| 327 | template <typename Head> struct last_type<Head> { using type = Head; }; |
| 328 | |
| 329 | /// A type trait whose \c value field is \c true if the last type is BCBlob. |
| 330 | template <typename... Types> |
| 331 | using has_blob = std::is_same<BCBlob, typename last_type<int, Types...>::type>; |
| 332 | |
| 333 | /// A type trait whose \c value field is \c true if the given type is a |
| 334 | /// BCArray (of any element kind). |
| 335 | template <typename T> struct is_array { |
| 336 | private: |
| 337 | template <typename E> static bool check(BCArray<E> *); |
| 338 | static int check(...); |
| 339 | |
| 340 | public: |
| 341 | typedef bool value_type; |
| 342 | static constexpr bool value = !std::is_same<decltype(check((T *)nullptr)), |
| 343 | decltype(check(false))>::value; |
| 344 | }; |
| 345 | |
| 346 | /// A type trait whose \c value field is \c true if the last type is a |
| 347 | /// BCArray (of any element kind). |
| 348 | template <typename... Types> |
| 349 | using has_array = is_array<typename last_type<int, Types...>::type>; |
| 350 | } // namespace detail |
| 351 | |
| 352 | /// Represents a single bitcode record type. |
| 353 | /// |
| 354 | /// This class template is meant to be instantiated and then given a name, |
| 355 | /// so that from then on that name can be used. |
| 356 | template <typename IDField, typename... Fields> class BCGenericRecordLayout { |
| 357 | llvm::BitstreamWriter &Stream; |
| 358 | |
| 359 | public: |
| 360 | /// The abbreviation code used for this record in the current block. |
| 361 | /// |
| 362 | /// Note that this is not the same as the semantic record code, which is the |
| 363 | /// first field of the record. |
| 364 | const unsigned AbbrevCode; |
| 365 | |
| 366 | /// Create a layout and register it with the given bitstream writer. |
| 367 | explicit BCGenericRecordLayout(llvm::BitstreamWriter &Stream) |
| 368 | : Stream(Stream), AbbrevCode(emitAbbrev(Stream)) {} |
| 369 | |
| 370 | /// Emit a record to the bitstream writer, using the given buffer for scratch |
| 371 | /// space. |
| 372 | /// |
| 373 | /// Note that even fixed arguments must be specified here. |
| 374 | template <typename BufferTy, typename... Data> |
| 375 | void emit(BufferTy &buffer, unsigned id, Data &&...data) const { |
| 376 | emitRecord(Stream, buffer, AbbrevCode, id, std::forward<Data>(data)...); |
| 377 | } |
| 378 | |
| 379 | /// Registers this record's layout with the bitstream reader. |
| 380 | /// |
| 381 | /// eturns The abbreviation code for the newly-registered record type. |
| 382 | static unsigned emitAbbrev(llvm::BitstreamWriter &Stream) { |
| 383 | auto Abbrev = std::make_shared<llvm::BitCodeAbbrev>(); |
| 384 | detail::emitOps<IDField, Fields...>(*Abbrev); |
| 385 | return Stream.EmitAbbrev(std::move(Abbrev)); |
| 386 | } |
| 387 | |
| 388 | /// Emit a record identified by \p abbrCode to bitstream reader \p Stream, |
| 389 | /// using \p buffer for scratch space. |
| 390 | /// |
| 391 | /// Note that even fixed arguments must be specified here. Blobs are passed |
| 392 | /// as StringRefs, while arrays can be passed inline, as aggregates, or as |
| 393 | /// pre-encoded StringRef data. Skipped values and empty arrays should use |
| 394 | /// the special Nothing value. |
| 395 | template <typename BufferTy, typename... Data> |
| 396 | static void emitRecord(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 397 | unsigned abbrCode, unsigned recordID, Data &&...data) { |
| 398 | static_assert(sizeof...(data) <= sizeof...(Fields) || |
| 399 | detail::has_array<Fields...>::value, |
| 400 | "Too many record elements"); |
| 401 | static_assert(sizeof...(data) >= sizeof...(Fields), |
| 402 | "Too few record elements"); |
| 403 | buffer.clear(); |
| 404 | detail::BCRecordCoding<IDField, Fields...>::emit( |
| 405 | Stream, buffer, abbrCode, recordID, std::forward<Data>(data)...); |
| 406 | } |
| 407 | |
| 408 | /// Extract record data from \p buffer into the given data fields. |
| 409 | /// |
| 410 | /// Note that even fixed arguments must be specified here. Pass \c Nothing |
| 411 | /// if you don't care about a particular parameter. Blob data is not included |
| 412 | /// in the buffer and should be handled separately by the caller. |
| 413 | template <typename ElementTy, typename... Data> |
| 414 | static void readRecord(ArrayRef<ElementTy> buffer, Data &&...data) { |
| 415 | static_assert(sizeof...(data) <= sizeof...(Fields), |
| 416 | "Too many record elements"); |
| 417 | static_assert(sizeof...(Fields) <= |
| 418 | sizeof...(data) + detail::has_blob<Fields...>::value, |
| 419 | "Too few record elements"); |
| 420 | return detail::BCRecordCoding<Fields...>::read(buffer, |
| 421 | std::forward<Data>(data)...); |
| 422 | } |
| 423 | |
| 424 | /// Extract record data from \p buffer into the given data fields. |
| 425 | /// |
| 426 | /// Note that even fixed arguments must be specified here. Pass \c Nothing |
| 427 | /// if you don't care about a particular parameter. Blob data is not included |
| 428 | /// in the buffer and should be handled separately by the caller. |
| 429 | template <typename BufferTy, typename... Data> |
| 430 | static void readRecord(BufferTy &buffer, Data &&...data) { |
| 431 | return readRecord(llvm::makeArrayRef(buffer), std::forward<Data>(data)...); |
| 432 | } |
| 433 | }; |
| 434 | |
| 435 | /// A record with a fixed record code. |
| 436 | template <unsigned RecordCode, typename... Fields> |
| 437 | class BCRecordLayout |
| 438 | : public BCGenericRecordLayout<BCLiteral<RecordCode>, Fields...> { |
| 439 | using Base = BCGenericRecordLayout<BCLiteral<RecordCode>, Fields...>; |
| 440 | |
| 441 | public: |
| 442 | enum : unsigned { |
| 443 | /// The record code associated with this layout. |
| 444 | Code = RecordCode |
| 445 | }; |
| 446 | |
| 447 | /// Create a layout and register it with the given bitstream writer. |
| 448 | explicit BCRecordLayout(llvm::BitstreamWriter &Stream) : Base(Stream) {} |
| 449 | |
| 450 | /// Emit a record to the bitstream writer, using the given buffer for scratch |
| 451 | /// space. |
| 452 | /// |
| 453 | /// Note that even fixed arguments must be specified here. |
| 454 | template <typename BufferTy, typename... Data> |
| 455 | void emit(BufferTy &buffer, Data &&...data) const { |
| 456 | Base::emit(buffer, RecordCode, std::forward<Data>(data)...); |
| 457 | } |
| 458 | |
| 459 | /// Emit a record identified by \p abbrCode to bitstream reader \p Stream, |
| 460 | /// using \p buffer for scratch space. |
| 461 | /// |
| 462 | /// Note that even fixed arguments must be specified here. Currently, arrays |
| 463 | /// and blobs can only be passed as StringRefs. |
| 464 | template <typename BufferTy, typename... Data> |
| 465 | static void emitRecord(llvm::BitstreamWriter &Stream, BufferTy &buffer, |
| 466 | unsigned abbrCode, Data &&...data) { |
| 467 | Base::emitRecord(Stream, buffer, abbrCode, RecordCode, |
| 468 | std::forward<Data>(data)...); |
| 469 | } |
| 470 | }; |
| 471 | |
| 472 | /// RAII object to pair entering and exiting a sub-block. |
| 473 | class BCBlockRAII { |
| 474 | llvm::BitstreamWriter &Stream; |
| 475 | |
| 476 | public: |
| 477 | BCBlockRAII(llvm::BitstreamWriter &Stream, unsigned block, unsigned abbrev) |
| 478 | : Stream(Stream) { |
| 479 | Stream.EnterSubblock(block, abbrev); |
| 480 | } |
| 481 | |
| 482 | ~BCBlockRAII() { Stream.ExitBlock(); } |
| 483 | }; |
| 484 | } // namespace llvm |
| 485 | |
| 486 | #endif |