Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | //===- BitstreamWriter.h - Low-level bitstream writer interface -*- 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 | // This header defines the BitstreamWriter class. This class can be used to |
||
| 10 | // write an arbitrary bitstream, regardless of its contents. |
||
| 11 | // |
||
| 12 | //===----------------------------------------------------------------------===// |
||
| 13 | |||
| 14 | #ifndef LLVM_BITSTREAM_BITSTREAMWRITER_H |
||
| 15 | #define LLVM_BITSTREAM_BITSTREAMWRITER_H |
||
| 16 | |||
| 17 | #include "llvm/ADT/ArrayRef.h" |
||
| 18 | #include "llvm/ADT/SmallVector.h" |
||
| 19 | #include "llvm/ADT/StringRef.h" |
||
| 20 | #include "llvm/Bitstream/BitCodes.h" |
||
| 21 | #include "llvm/Support/Endian.h" |
||
| 22 | #include "llvm/Support/MathExtras.h" |
||
| 23 | #include "llvm/Support/raw_ostream.h" |
||
| 24 | #include <algorithm> |
||
| 25 | #include <optional> |
||
| 26 | #include <vector> |
||
| 27 | |||
| 28 | namespace llvm { |
||
| 29 | |||
| 30 | class BitstreamWriter { |
||
| 31 | /// Out - The buffer that keeps unflushed bytes. |
||
| 32 | SmallVectorImpl<char> &Out; |
||
| 33 | |||
| 34 | /// FS - The file stream that Out flushes to. If FS is nullptr, it does not |
||
| 35 | /// support read or seek, Out cannot be flushed until all data are written. |
||
| 36 | raw_fd_stream *FS; |
||
| 37 | |||
| 38 | /// FlushThreshold - If FS is valid, this is the threshold (unit B) to flush |
||
| 39 | /// FS. |
||
| 40 | const uint64_t FlushThreshold; |
||
| 41 | |||
| 42 | /// CurBit - Always between 0 and 31 inclusive, specifies the next bit to use. |
||
| 43 | unsigned CurBit; |
||
| 44 | |||
| 45 | /// CurValue - The current value. Only bits < CurBit are valid. |
||
| 46 | uint32_t CurValue; |
||
| 47 | |||
| 48 | /// CurCodeSize - This is the declared size of code values used for the |
||
| 49 | /// current block, in bits. |
||
| 50 | unsigned CurCodeSize; |
||
| 51 | |||
| 52 | /// BlockInfoCurBID - When emitting a BLOCKINFO_BLOCK, this is the currently |
||
| 53 | /// selected BLOCK ID. |
||
| 54 | unsigned BlockInfoCurBID; |
||
| 55 | |||
| 56 | /// CurAbbrevs - Abbrevs installed at in this block. |
||
| 57 | std::vector<std::shared_ptr<BitCodeAbbrev>> CurAbbrevs; |
||
| 58 | |||
| 59 | struct Block { |
||
| 60 | unsigned PrevCodeSize; |
||
| 61 | size_t StartSizeWord; |
||
| 62 | std::vector<std::shared_ptr<BitCodeAbbrev>> PrevAbbrevs; |
||
| 63 | Block(unsigned PCS, size_t SSW) : PrevCodeSize(PCS), StartSizeWord(SSW) {} |
||
| 64 | }; |
||
| 65 | |||
| 66 | /// BlockScope - This tracks the current blocks that we have entered. |
||
| 67 | std::vector<Block> BlockScope; |
||
| 68 | |||
| 69 | /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks. |
||
| 70 | /// These describe abbreviations that all blocks of the specified ID inherit. |
||
| 71 | struct BlockInfo { |
||
| 72 | unsigned BlockID; |
||
| 73 | std::vector<std::shared_ptr<BitCodeAbbrev>> Abbrevs; |
||
| 74 | }; |
||
| 75 | std::vector<BlockInfo> BlockInfoRecords; |
||
| 76 | |||
| 77 | void WriteWord(unsigned Value) { |
||
| 78 | Value = support::endian::byte_swap<uint32_t, support::little>(Value); |
||
| 79 | Out.append(reinterpret_cast<const char *>(&Value), |
||
| 80 | reinterpret_cast<const char *>(&Value + 1)); |
||
| 81 | } |
||
| 82 | |||
| 83 | uint64_t GetNumOfFlushedBytes() const { return FS ? FS->tell() : 0; } |
||
| 84 | |||
| 85 | size_t GetBufferOffset() const { return Out.size() + GetNumOfFlushedBytes(); } |
||
| 86 | |||
| 87 | size_t GetWordIndex() const { |
||
| 88 | size_t Offset = GetBufferOffset(); |
||
| 89 | assert((Offset & 3) == 0 && "Not 32-bit aligned"); |
||
| 90 | return Offset / 4; |
||
| 91 | } |
||
| 92 | |||
| 93 | /// If the related file stream supports reading, seeking and writing, flush |
||
| 94 | /// the buffer if its size is above a threshold. |
||
| 95 | void FlushToFile() { |
||
| 96 | if (!FS) |
||
| 97 | return; |
||
| 98 | if (Out.size() < FlushThreshold) |
||
| 99 | return; |
||
| 100 | FS->write((char *)&Out.front(), Out.size()); |
||
| 101 | Out.clear(); |
||
| 102 | } |
||
| 103 | |||
| 104 | public: |
||
| 105 | /// Create a BitstreamWriter that writes to Buffer \p O. |
||
| 106 | /// |
||
| 107 | /// \p FS is the file stream that \p O flushes to incrementally. If \p FS is |
||
| 108 | /// null, \p O does not flush incrementially, but writes to disk at the end. |
||
| 109 | /// |
||
| 110 | /// \p FlushThreshold is the threshold (unit M) to flush \p O if \p FS is |
||
| 111 | /// valid. Flushing only occurs at (sub)block boundaries. |
||
| 112 | BitstreamWriter(SmallVectorImpl<char> &O, raw_fd_stream *FS = nullptr, |
||
| 113 | uint32_t FlushThreshold = 512) |
||
| 114 | : Out(O), FS(FS), FlushThreshold(FlushThreshold << 20), CurBit(0), |
||
| 115 | CurValue(0), CurCodeSize(2) {} |
||
| 116 | |||
| 117 | ~BitstreamWriter() { |
||
| 118 | assert(CurBit == 0 && "Unflushed data remaining"); |
||
| 119 | assert(BlockScope.empty() && CurAbbrevs.empty() && "Block imbalance"); |
||
| 120 | } |
||
| 121 | |||
| 122 | /// Retrieve the current position in the stream, in bits. |
||
| 123 | uint64_t GetCurrentBitNo() const { return GetBufferOffset() * 8 + CurBit; } |
||
| 124 | |||
| 125 | /// Retrieve the number of bits currently used to encode an abbrev ID. |
||
| 126 | unsigned GetAbbrevIDWidth() const { return CurCodeSize; } |
||
| 127 | |||
| 128 | //===--------------------------------------------------------------------===// |
||
| 129 | // Basic Primitives for emitting bits to the stream. |
||
| 130 | //===--------------------------------------------------------------------===// |
||
| 131 | |||
| 132 | /// Backpatch a 32-bit word in the output at the given bit offset |
||
| 133 | /// with the specified value. |
||
| 134 | void BackpatchWord(uint64_t BitNo, unsigned NewWord) { |
||
| 135 | using namespace llvm::support; |
||
| 136 | uint64_t ByteNo = BitNo / 8; |
||
| 137 | uint64_t StartBit = BitNo & 7; |
||
| 138 | uint64_t NumOfFlushedBytes = GetNumOfFlushedBytes(); |
||
| 139 | |||
| 140 | if (ByteNo >= NumOfFlushedBytes) { |
||
| 141 | assert((!endian::readAtBitAlignment<uint32_t, little, unaligned>( |
||
| 142 | &Out[ByteNo - NumOfFlushedBytes], StartBit)) && |
||
| 143 | "Expected to be patching over 0-value placeholders"); |
||
| 144 | endian::writeAtBitAlignment<uint32_t, little, unaligned>( |
||
| 145 | &Out[ByteNo - NumOfFlushedBytes], NewWord, StartBit); |
||
| 146 | return; |
||
| 147 | } |
||
| 148 | |||
| 149 | // If the byte offset to backpatch is flushed, use seek to backfill data. |
||
| 150 | // First, save the file position to restore later. |
||
| 151 | uint64_t CurPos = FS->tell(); |
||
| 152 | |||
| 153 | // Copy data to update into Bytes from the file FS and the buffer Out. |
||
| 154 | char Bytes[9]; // Use one more byte to silence a warning from Visual C++. |
||
| 155 | size_t BytesNum = StartBit ? 8 : 4; |
||
| 156 | size_t BytesFromDisk = std::min(static_cast<uint64_t>(BytesNum), NumOfFlushedBytes - ByteNo); |
||
| 157 | size_t BytesFromBuffer = BytesNum - BytesFromDisk; |
||
| 158 | |||
| 159 | // When unaligned, copy existing data into Bytes from the file FS and the |
||
| 160 | // buffer Out so that it can be updated before writing. For debug builds |
||
| 161 | // read bytes unconditionally in order to check that the existing value is 0 |
||
| 162 | // as expected. |
||
| 163 | #ifdef NDEBUG |
||
| 164 | if (StartBit) |
||
| 165 | #endif |
||
| 166 | { |
||
| 167 | FS->seek(ByteNo); |
||
| 168 | ssize_t BytesRead = FS->read(Bytes, BytesFromDisk); |
||
| 169 | (void)BytesRead; // silence warning |
||
| 170 | assert(BytesRead >= 0 && static_cast<size_t>(BytesRead) == BytesFromDisk); |
||
| 171 | for (size_t i = 0; i < BytesFromBuffer; ++i) |
||
| 172 | Bytes[BytesFromDisk + i] = Out[i]; |
||
| 173 | assert((!endian::readAtBitAlignment<uint32_t, little, unaligned>( |
||
| 174 | Bytes, StartBit)) && |
||
| 175 | "Expected to be patching over 0-value placeholders"); |
||
| 176 | } |
||
| 177 | |||
| 178 | // Update Bytes in terms of bit offset and value. |
||
| 179 | endian::writeAtBitAlignment<uint32_t, little, unaligned>(Bytes, NewWord, |
||
| 180 | StartBit); |
||
| 181 | |||
| 182 | // Copy updated data back to the file FS and the buffer Out. |
||
| 183 | FS->seek(ByteNo); |
||
| 184 | FS->write(Bytes, BytesFromDisk); |
||
| 185 | for (size_t i = 0; i < BytesFromBuffer; ++i) |
||
| 186 | Out[i] = Bytes[BytesFromDisk + i]; |
||
| 187 | |||
| 188 | // Restore the file position. |
||
| 189 | FS->seek(CurPos); |
||
| 190 | } |
||
| 191 | |||
| 192 | void BackpatchWord64(uint64_t BitNo, uint64_t Val) { |
||
| 193 | BackpatchWord(BitNo, (uint32_t)Val); |
||
| 194 | BackpatchWord(BitNo + 32, (uint32_t)(Val >> 32)); |
||
| 195 | } |
||
| 196 | |||
| 197 | void Emit(uint32_t Val, unsigned NumBits) { |
||
| 198 | assert(NumBits && NumBits <= 32 && "Invalid value size!"); |
||
| 199 | assert((Val & ~(~0U >> (32-NumBits))) == 0 && "High bits set!"); |
||
| 200 | CurValue |= Val << CurBit; |
||
| 201 | if (CurBit + NumBits < 32) { |
||
| 202 | CurBit += NumBits; |
||
| 203 | return; |
||
| 204 | } |
||
| 205 | |||
| 206 | // Add the current word. |
||
| 207 | WriteWord(CurValue); |
||
| 208 | |||
| 209 | if (CurBit) |
||
| 210 | CurValue = Val >> (32-CurBit); |
||
| 211 | else |
||
| 212 | CurValue = 0; |
||
| 213 | CurBit = (CurBit+NumBits) & 31; |
||
| 214 | } |
||
| 215 | |||
| 216 | void FlushToWord() { |
||
| 217 | if (CurBit) { |
||
| 218 | WriteWord(CurValue); |
||
| 219 | CurBit = 0; |
||
| 220 | CurValue = 0; |
||
| 221 | } |
||
| 222 | } |
||
| 223 | |||
| 224 | void EmitVBR(uint32_t Val, unsigned NumBits) { |
||
| 225 | assert(NumBits <= 32 && "Too many bits to emit!"); |
||
| 226 | uint32_t Threshold = 1U << (NumBits-1); |
||
| 227 | |||
| 228 | // Emit the bits with VBR encoding, NumBits-1 bits at a time. |
||
| 229 | while (Val >= Threshold) { |
||
| 230 | Emit((Val & ((1 << (NumBits-1))-1)) | (1 << (NumBits-1)), NumBits); |
||
| 231 | Val >>= NumBits-1; |
||
| 232 | } |
||
| 233 | |||
| 234 | Emit(Val, NumBits); |
||
| 235 | } |
||
| 236 | |||
| 237 | void EmitVBR64(uint64_t Val, unsigned NumBits) { |
||
| 238 | assert(NumBits <= 32 && "Too many bits to emit!"); |
||
| 239 | if ((uint32_t)Val == Val) |
||
| 240 | return EmitVBR((uint32_t)Val, NumBits); |
||
| 241 | |||
| 242 | uint32_t Threshold = 1U << (NumBits-1); |
||
| 243 | |||
| 244 | // Emit the bits with VBR encoding, NumBits-1 bits at a time. |
||
| 245 | while (Val >= Threshold) { |
||
| 246 | Emit(((uint32_t)Val & ((1 << (NumBits - 1)) - 1)) | (1 << (NumBits - 1)), |
||
| 247 | NumBits); |
||
| 248 | Val >>= NumBits-1; |
||
| 249 | } |
||
| 250 | |||
| 251 | Emit((uint32_t)Val, NumBits); |
||
| 252 | } |
||
| 253 | |||
| 254 | /// EmitCode - Emit the specified code. |
||
| 255 | void EmitCode(unsigned Val) { |
||
| 256 | Emit(Val, CurCodeSize); |
||
| 257 | } |
||
| 258 | |||
| 259 | //===--------------------------------------------------------------------===// |
||
| 260 | // Block Manipulation |
||
| 261 | //===--------------------------------------------------------------------===// |
||
| 262 | |||
| 263 | /// getBlockInfo - If there is block info for the specified ID, return it, |
||
| 264 | /// otherwise return null. |
||
| 265 | BlockInfo *getBlockInfo(unsigned BlockID) { |
||
| 266 | // Common case, the most recent entry matches BlockID. |
||
| 267 | if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID) |
||
| 268 | return &BlockInfoRecords.back(); |
||
| 269 | |||
| 270 | for (BlockInfo &BI : BlockInfoRecords) |
||
| 271 | if (BI.BlockID == BlockID) |
||
| 272 | return &BI; |
||
| 273 | return nullptr; |
||
| 274 | } |
||
| 275 | |||
| 276 | void EnterSubblock(unsigned BlockID, unsigned CodeLen) { |
||
| 277 | // Block header: |
||
| 278 | // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen] |
||
| 279 | EmitCode(bitc::ENTER_SUBBLOCK); |
||
| 280 | EmitVBR(BlockID, bitc::BlockIDWidth); |
||
| 281 | EmitVBR(CodeLen, bitc::CodeLenWidth); |
||
| 282 | FlushToWord(); |
||
| 283 | |||
| 284 | size_t BlockSizeWordIndex = GetWordIndex(); |
||
| 285 | unsigned OldCodeSize = CurCodeSize; |
||
| 286 | |||
| 287 | // Emit a placeholder, which will be replaced when the block is popped. |
||
| 288 | Emit(0, bitc::BlockSizeWidth); |
||
| 289 | |||
| 290 | CurCodeSize = CodeLen; |
||
| 291 | |||
| 292 | // Push the outer block's abbrev set onto the stack, start out with an |
||
| 293 | // empty abbrev set. |
||
| 294 | BlockScope.emplace_back(OldCodeSize, BlockSizeWordIndex); |
||
| 295 | BlockScope.back().PrevAbbrevs.swap(CurAbbrevs); |
||
| 296 | |||
| 297 | // If there is a blockinfo for this BlockID, add all the predefined abbrevs |
||
| 298 | // to the abbrev list. |
||
| 299 | if (BlockInfo *Info = getBlockInfo(BlockID)) |
||
| 300 | append_range(CurAbbrevs, Info->Abbrevs); |
||
| 301 | } |
||
| 302 | |||
| 303 | void ExitBlock() { |
||
| 304 | assert(!BlockScope.empty() && "Block scope imbalance!"); |
||
| 305 | const Block &B = BlockScope.back(); |
||
| 306 | |||
| 307 | // Block tail: |
||
| 308 | // [END_BLOCK, <align4bytes>] |
||
| 309 | EmitCode(bitc::END_BLOCK); |
||
| 310 | FlushToWord(); |
||
| 311 | |||
| 312 | // Compute the size of the block, in words, not counting the size field. |
||
| 313 | size_t SizeInWords = GetWordIndex() - B.StartSizeWord - 1; |
||
| 314 | uint64_t BitNo = uint64_t(B.StartSizeWord) * 32; |
||
| 315 | |||
| 316 | // Update the block size field in the header of this sub-block. |
||
| 317 | BackpatchWord(BitNo, SizeInWords); |
||
| 318 | |||
| 319 | // Restore the inner block's code size and abbrev table. |
||
| 320 | CurCodeSize = B.PrevCodeSize; |
||
| 321 | CurAbbrevs = std::move(B.PrevAbbrevs); |
||
| 322 | BlockScope.pop_back(); |
||
| 323 | FlushToFile(); |
||
| 324 | } |
||
| 325 | |||
| 326 | //===--------------------------------------------------------------------===// |
||
| 327 | // Record Emission |
||
| 328 | //===--------------------------------------------------------------------===// |
||
| 329 | |||
| 330 | private: |
||
| 331 | /// EmitAbbreviatedLiteral - Emit a literal value according to its abbrev |
||
| 332 | /// record. This is a no-op, since the abbrev specifies the literal to use. |
||
| 333 | template<typename uintty> |
||
| 334 | void EmitAbbreviatedLiteral(const BitCodeAbbrevOp &Op, uintty V) { |
||
| 335 | assert(Op.isLiteral() && "Not a literal"); |
||
| 336 | // If the abbrev specifies the literal value to use, don't emit |
||
| 337 | // anything. |
||
| 338 | assert(V == Op.getLiteralValue() && |
||
| 339 | "Invalid abbrev for record!"); |
||
| 340 | } |
||
| 341 | |||
| 342 | /// EmitAbbreviatedField - Emit a single scalar field value with the specified |
||
| 343 | /// encoding. |
||
| 344 | template<typename uintty> |
||
| 345 | void EmitAbbreviatedField(const BitCodeAbbrevOp &Op, uintty V) { |
||
| 346 | assert(!Op.isLiteral() && "Literals should use EmitAbbreviatedLiteral!"); |
||
| 347 | |||
| 348 | // Encode the value as we are commanded. |
||
| 349 | switch (Op.getEncoding()) { |
||
| 350 | default: llvm_unreachable("Unknown encoding!"); |
||
| 351 | case BitCodeAbbrevOp::Fixed: |
||
| 352 | if (Op.getEncodingData()) |
||
| 353 | Emit((unsigned)V, (unsigned)Op.getEncodingData()); |
||
| 354 | break; |
||
| 355 | case BitCodeAbbrevOp::VBR: |
||
| 356 | if (Op.getEncodingData()) |
||
| 357 | EmitVBR64(V, (unsigned)Op.getEncodingData()); |
||
| 358 | break; |
||
| 359 | case BitCodeAbbrevOp::Char6: |
||
| 360 | Emit(BitCodeAbbrevOp::EncodeChar6((char)V), 6); |
||
| 361 | break; |
||
| 362 | } |
||
| 363 | } |
||
| 364 | |||
| 365 | /// EmitRecordWithAbbrevImpl - This is the core implementation of the record |
||
| 366 | /// emission code. If BlobData is non-null, then it specifies an array of |
||
| 367 | /// data that should be emitted as part of the Blob or Array operand that is |
||
| 368 | /// known to exist at the end of the record. If Code is specified, then |
||
| 369 | /// it is the record code to emit before the Vals, which must not contain |
||
| 370 | /// the code. |
||
| 371 | template <typename uintty> |
||
| 372 | void EmitRecordWithAbbrevImpl(unsigned Abbrev, ArrayRef<uintty> Vals, |
||
| 373 | StringRef Blob, std::optional<unsigned> Code) { |
||
| 374 | const char *BlobData = Blob.data(); |
||
| 375 | unsigned BlobLen = (unsigned) Blob.size(); |
||
| 376 | unsigned AbbrevNo = Abbrev-bitc::FIRST_APPLICATION_ABBREV; |
||
| 377 | assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!"); |
||
| 378 | const BitCodeAbbrev *Abbv = CurAbbrevs[AbbrevNo].get(); |
||
| 379 | |||
| 380 | EmitCode(Abbrev); |
||
| 381 | |||
| 382 | unsigned i = 0, e = static_cast<unsigned>(Abbv->getNumOperandInfos()); |
||
| 383 | if (Code) { |
||
| 384 | assert(e && "Expected non-empty abbreviation"); |
||
| 385 | const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i++); |
||
| 386 | |||
| 387 | if (Op.isLiteral()) |
||
| 388 | EmitAbbreviatedLiteral(Op, *Code); |
||
| 389 | else { |
||
| 390 | assert(Op.getEncoding() != BitCodeAbbrevOp::Array && |
||
| 391 | Op.getEncoding() != BitCodeAbbrevOp::Blob && |
||
| 392 | "Expected literal or scalar"); |
||
| 393 | EmitAbbreviatedField(Op, *Code); |
||
| 394 | } |
||
| 395 | } |
||
| 396 | |||
| 397 | unsigned RecordIdx = 0; |
||
| 398 | for (; i != e; ++i) { |
||
| 399 | const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i); |
||
| 400 | if (Op.isLiteral()) { |
||
| 401 | assert(RecordIdx < Vals.size() && "Invalid abbrev/record"); |
||
| 402 | EmitAbbreviatedLiteral(Op, Vals[RecordIdx]); |
||
| 403 | ++RecordIdx; |
||
| 404 | } else if (Op.getEncoding() == BitCodeAbbrevOp::Array) { |
||
| 405 | // Array case. |
||
| 406 | assert(i + 2 == e && "array op not second to last?"); |
||
| 407 | const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i); |
||
| 408 | |||
| 409 | // If this record has blob data, emit it, otherwise we must have record |
||
| 410 | // entries to encode this way. |
||
| 411 | if (BlobData) { |
||
| 412 | assert(RecordIdx == Vals.size() && |
||
| 413 | "Blob data and record entries specified for array!"); |
||
| 414 | // Emit a vbr6 to indicate the number of elements present. |
||
| 415 | EmitVBR(static_cast<uint32_t>(BlobLen), 6); |
||
| 416 | |||
| 417 | // Emit each field. |
||
| 418 | for (unsigned i = 0; i != BlobLen; ++i) |
||
| 419 | EmitAbbreviatedField(EltEnc, (unsigned char)BlobData[i]); |
||
| 420 | |||
| 421 | // Know that blob data is consumed for assertion below. |
||
| 422 | BlobData = nullptr; |
||
| 423 | } else { |
||
| 424 | // Emit a vbr6 to indicate the number of elements present. |
||
| 425 | EmitVBR(static_cast<uint32_t>(Vals.size()-RecordIdx), 6); |
||
| 426 | |||
| 427 | // Emit each field. |
||
| 428 | for (unsigned e = Vals.size(); RecordIdx != e; ++RecordIdx) |
||
| 429 | EmitAbbreviatedField(EltEnc, Vals[RecordIdx]); |
||
| 430 | } |
||
| 431 | } else if (Op.getEncoding() == BitCodeAbbrevOp::Blob) { |
||
| 432 | // If this record has blob data, emit it, otherwise we must have record |
||
| 433 | // entries to encode this way. |
||
| 434 | |||
| 435 | if (BlobData) { |
||
| 436 | assert(RecordIdx == Vals.size() && |
||
| 437 | "Blob data and record entries specified for blob operand!"); |
||
| 438 | |||
| 439 | assert(Blob.data() == BlobData && "BlobData got moved"); |
||
| 440 | assert(Blob.size() == BlobLen && "BlobLen got changed"); |
||
| 441 | emitBlob(Blob); |
||
| 442 | BlobData = nullptr; |
||
| 443 | } else { |
||
| 444 | emitBlob(Vals.slice(RecordIdx)); |
||
| 445 | } |
||
| 446 | } else { // Single scalar field. |
||
| 447 | assert(RecordIdx < Vals.size() && "Invalid abbrev/record"); |
||
| 448 | EmitAbbreviatedField(Op, Vals[RecordIdx]); |
||
| 449 | ++RecordIdx; |
||
| 450 | } |
||
| 451 | } |
||
| 452 | assert(RecordIdx == Vals.size() && "Not all record operands emitted!"); |
||
| 453 | assert(BlobData == nullptr && |
||
| 454 | "Blob data specified for record that doesn't use it!"); |
||
| 455 | } |
||
| 456 | |||
| 457 | public: |
||
| 458 | /// Emit a blob, including flushing before and tail-padding. |
||
| 459 | template <class UIntTy> |
||
| 460 | void emitBlob(ArrayRef<UIntTy> Bytes, bool ShouldEmitSize = true) { |
||
| 461 | // Emit a vbr6 to indicate the number of elements present. |
||
| 462 | if (ShouldEmitSize) |
||
| 463 | EmitVBR(static_cast<uint32_t>(Bytes.size()), 6); |
||
| 464 | |||
| 465 | // Flush to a 32-bit alignment boundary. |
||
| 466 | FlushToWord(); |
||
| 467 | |||
| 468 | // Emit literal bytes. |
||
| 469 | assert(llvm::all_of(Bytes, [](UIntTy B) { return isUInt<8>(B); })); |
||
| 470 | Out.append(Bytes.begin(), Bytes.end()); |
||
| 471 | |||
| 472 | // Align end to 32-bits. |
||
| 473 | while (GetBufferOffset() & 3) |
||
| 474 | Out.push_back(0); |
||
| 475 | } |
||
| 476 | void emitBlob(StringRef Bytes, bool ShouldEmitSize = true) { |
||
| 477 | emitBlob(ArrayRef((const uint8_t *)Bytes.data(), Bytes.size()), |
||
| 478 | ShouldEmitSize); |
||
| 479 | } |
||
| 480 | |||
| 481 | /// EmitRecord - Emit the specified record to the stream, using an abbrev if |
||
| 482 | /// we have one to compress the output. |
||
| 483 | template <typename Container> |
||
| 484 | void EmitRecord(unsigned Code, const Container &Vals, unsigned Abbrev = 0) { |
||
| 485 | if (!Abbrev) { |
||
| 486 | // If we don't have an abbrev to use, emit this in its fully unabbreviated |
||
| 487 | // form. |
||
| 488 | auto Count = static_cast<uint32_t>(std::size(Vals)); |
||
| 489 | EmitCode(bitc::UNABBREV_RECORD); |
||
| 490 | EmitVBR(Code, 6); |
||
| 491 | EmitVBR(Count, 6); |
||
| 492 | for (unsigned i = 0, e = Count; i != e; ++i) |
||
| 493 | EmitVBR64(Vals[i], 6); |
||
| 494 | return; |
||
| 495 | } |
||
| 496 | |||
| 497 | EmitRecordWithAbbrevImpl(Abbrev, ArrayRef(Vals), StringRef(), Code); |
||
| 498 | } |
||
| 499 | |||
| 500 | /// EmitRecordWithAbbrev - Emit a record with the specified abbreviation. |
||
| 501 | /// Unlike EmitRecord, the code for the record should be included in Vals as |
||
| 502 | /// the first entry. |
||
| 503 | template <typename Container> |
||
| 504 | void EmitRecordWithAbbrev(unsigned Abbrev, const Container &Vals) { |
||
| 505 | EmitRecordWithAbbrevImpl(Abbrev, ArrayRef(Vals), StringRef(), std::nullopt); |
||
| 506 | } |
||
| 507 | |||
| 508 | /// EmitRecordWithBlob - Emit the specified record to the stream, using an |
||
| 509 | /// abbrev that includes a blob at the end. The blob data to emit is |
||
| 510 | /// specified by the pointer and length specified at the end. In contrast to |
||
| 511 | /// EmitRecord, this routine expects that the first entry in Vals is the code |
||
| 512 | /// of the record. |
||
| 513 | template <typename Container> |
||
| 514 | void EmitRecordWithBlob(unsigned Abbrev, const Container &Vals, |
||
| 515 | StringRef Blob) { |
||
| 516 | EmitRecordWithAbbrevImpl(Abbrev, ArrayRef(Vals), Blob, std::nullopt); |
||
| 517 | } |
||
| 518 | template <typename Container> |
||
| 519 | void EmitRecordWithBlob(unsigned Abbrev, const Container &Vals, |
||
| 520 | const char *BlobData, unsigned BlobLen) { |
||
| 521 | return EmitRecordWithAbbrevImpl(Abbrev, ArrayRef(Vals), |
||
| 522 | StringRef(BlobData, BlobLen), std::nullopt); |
||
| 523 | } |
||
| 524 | |||
| 525 | /// EmitRecordWithArray - Just like EmitRecordWithBlob, works with records |
||
| 526 | /// that end with an array. |
||
| 527 | template <typename Container> |
||
| 528 | void EmitRecordWithArray(unsigned Abbrev, const Container &Vals, |
||
| 529 | StringRef Array) { |
||
| 530 | EmitRecordWithAbbrevImpl(Abbrev, ArrayRef(Vals), Array, std::nullopt); |
||
| 531 | } |
||
| 532 | template <typename Container> |
||
| 533 | void EmitRecordWithArray(unsigned Abbrev, const Container &Vals, |
||
| 534 | const char *ArrayData, unsigned ArrayLen) { |
||
| 535 | return EmitRecordWithAbbrevImpl( |
||
| 536 | Abbrev, ArrayRef(Vals), StringRef(ArrayData, ArrayLen), std::nullopt); |
||
| 537 | } |
||
| 538 | |||
| 539 | //===--------------------------------------------------------------------===// |
||
| 540 | // Abbrev Emission |
||
| 541 | //===--------------------------------------------------------------------===// |
||
| 542 | |||
| 543 | private: |
||
| 544 | // Emit the abbreviation as a DEFINE_ABBREV record. |
||
| 545 | void EncodeAbbrev(const BitCodeAbbrev &Abbv) { |
||
| 546 | EmitCode(bitc::DEFINE_ABBREV); |
||
| 547 | EmitVBR(Abbv.getNumOperandInfos(), 5); |
||
| 548 | for (unsigned i = 0, e = static_cast<unsigned>(Abbv.getNumOperandInfos()); |
||
| 549 | i != e; ++i) { |
||
| 550 | const BitCodeAbbrevOp &Op = Abbv.getOperandInfo(i); |
||
| 551 | Emit(Op.isLiteral(), 1); |
||
| 552 | if (Op.isLiteral()) { |
||
| 553 | EmitVBR64(Op.getLiteralValue(), 8); |
||
| 554 | } else { |
||
| 555 | Emit(Op.getEncoding(), 3); |
||
| 556 | if (Op.hasEncodingData()) |
||
| 557 | EmitVBR64(Op.getEncodingData(), 5); |
||
| 558 | } |
||
| 559 | } |
||
| 560 | } |
||
| 561 | public: |
||
| 562 | |||
| 563 | /// Emits the abbreviation \p Abbv to the stream. |
||
| 564 | unsigned EmitAbbrev(std::shared_ptr<BitCodeAbbrev> Abbv) { |
||
| 565 | EncodeAbbrev(*Abbv); |
||
| 566 | CurAbbrevs.push_back(std::move(Abbv)); |
||
| 567 | return static_cast<unsigned>(CurAbbrevs.size())-1 + |
||
| 568 | bitc::FIRST_APPLICATION_ABBREV; |
||
| 569 | } |
||
| 570 | |||
| 571 | //===--------------------------------------------------------------------===// |
||
| 572 | // BlockInfo Block Emission |
||
| 573 | //===--------------------------------------------------------------------===// |
||
| 574 | |||
| 575 | /// EnterBlockInfoBlock - Start emitting the BLOCKINFO_BLOCK. |
||
| 576 | void EnterBlockInfoBlock() { |
||
| 577 | EnterSubblock(bitc::BLOCKINFO_BLOCK_ID, 2); |
||
| 578 | BlockInfoCurBID = ~0U; |
||
| 579 | BlockInfoRecords.clear(); |
||
| 580 | } |
||
| 581 | private: |
||
| 582 | /// SwitchToBlockID - If we aren't already talking about the specified block |
||
| 583 | /// ID, emit a BLOCKINFO_CODE_SETBID record. |
||
| 584 | void SwitchToBlockID(unsigned BlockID) { |
||
| 585 | if (BlockInfoCurBID == BlockID) return; |
||
| 586 | SmallVector<unsigned, 2> V; |
||
| 587 | V.push_back(BlockID); |
||
| 588 | EmitRecord(bitc::BLOCKINFO_CODE_SETBID, V); |
||
| 589 | BlockInfoCurBID = BlockID; |
||
| 590 | } |
||
| 591 | |||
| 592 | BlockInfo &getOrCreateBlockInfo(unsigned BlockID) { |
||
| 593 | if (BlockInfo *BI = getBlockInfo(BlockID)) |
||
| 594 | return *BI; |
||
| 595 | |||
| 596 | // Otherwise, add a new record. |
||
| 597 | BlockInfoRecords.emplace_back(); |
||
| 598 | BlockInfoRecords.back().BlockID = BlockID; |
||
| 599 | return BlockInfoRecords.back(); |
||
| 600 | } |
||
| 601 | |||
| 602 | public: |
||
| 603 | |||
| 604 | /// EmitBlockInfoAbbrev - Emit a DEFINE_ABBREV record for the specified |
||
| 605 | /// BlockID. |
||
| 606 | unsigned EmitBlockInfoAbbrev(unsigned BlockID, std::shared_ptr<BitCodeAbbrev> Abbv) { |
||
| 607 | SwitchToBlockID(BlockID); |
||
| 608 | EncodeAbbrev(*Abbv); |
||
| 609 | |||
| 610 | // Add the abbrev to the specified block record. |
||
| 611 | BlockInfo &Info = getOrCreateBlockInfo(BlockID); |
||
| 612 | Info.Abbrevs.push_back(std::move(Abbv)); |
||
| 613 | |||
| 614 | return Info.Abbrevs.size()-1+bitc::FIRST_APPLICATION_ABBREV; |
||
| 615 | } |
||
| 616 | }; |
||
| 617 | |||
| 618 | |||
| 619 | } // End llvm namespace |
||
| 620 | |||
| 621 | #endif |