1 //===- BitstreamReader.cpp - BitstreamReader implementation ---------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9
10 #include "llvm/Bitcode/BitstreamReader.h"
11
12 using namespace llvm;
13
14 //===----------------------------------------------------------------------===//
15 // BitstreamCursor implementation
16 //===----------------------------------------------------------------------===//
17
freeState()18 void BitstreamCursor::freeState() {
19 // Free all the Abbrevs.
20 CurAbbrevs.clear();
21
22 // Free all the Abbrevs in the block scope.
23 BlockScope.clear();
24 }
25
26 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
27 /// the block, and return true if the block has an error.
EnterSubBlock(unsigned BlockID,unsigned * NumWordsP)28 bool BitstreamCursor::EnterSubBlock(unsigned BlockID, unsigned *NumWordsP) {
29 // Save the current block's state on BlockScope.
30 BlockScope.push_back(Block(CurCodeSize));
31 BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
32
33 // Add the abbrevs specific to this block to the CurAbbrevs list.
34 if (const BitstreamReader::BlockInfo *Info =
35 BitStream->getBlockInfo(BlockID)) {
36 CurAbbrevs.insert(CurAbbrevs.end(), Info->Abbrevs.begin(),
37 Info->Abbrevs.end());
38 }
39
40 // Get the codesize of this block.
41 CurCodeSize = ReadVBR(bitc::CodeLenWidth);
42 SkipToFourByteBoundary();
43 unsigned NumWords = Read(bitc::BlockSizeWidth);
44 if (NumWordsP) *NumWordsP = NumWords;
45
46 // Validate that this block is sane.
47 if (CurCodeSize == 0 || AtEndOfStream())
48 return true;
49
50 return false;
51 }
52
readAbbreviatedField(BitstreamCursor & Cursor,const BitCodeAbbrevOp & Op)53 static uint64_t readAbbreviatedField(BitstreamCursor &Cursor,
54 const BitCodeAbbrevOp &Op) {
55 assert(!Op.isLiteral() && "Not to be used with literals!");
56
57 // Decode the value as we are commanded.
58 switch (Op.getEncoding()) {
59 case BitCodeAbbrevOp::Array:
60 case BitCodeAbbrevOp::Blob:
61 llvm_unreachable("Should not reach here");
62 case BitCodeAbbrevOp::Fixed:
63 return Cursor.Read((unsigned)Op.getEncodingData());
64 case BitCodeAbbrevOp::VBR:
65 return Cursor.ReadVBR64((unsigned)Op.getEncodingData());
66 case BitCodeAbbrevOp::Char6:
67 return BitCodeAbbrevOp::DecodeChar6(Cursor.Read(6));
68 }
69 llvm_unreachable("invalid abbreviation encoding");
70 }
71
skipAbbreviatedField(BitstreamCursor & Cursor,const BitCodeAbbrevOp & Op)72 static void skipAbbreviatedField(BitstreamCursor &Cursor,
73 const BitCodeAbbrevOp &Op) {
74 assert(!Op.isLiteral() && "Not to be used with literals!");
75
76 // Decode the value as we are commanded.
77 switch (Op.getEncoding()) {
78 case BitCodeAbbrevOp::Array:
79 case BitCodeAbbrevOp::Blob:
80 llvm_unreachable("Should not reach here");
81 case BitCodeAbbrevOp::Fixed:
82 Cursor.Read((unsigned)Op.getEncodingData());
83 break;
84 case BitCodeAbbrevOp::VBR:
85 Cursor.ReadVBR64((unsigned)Op.getEncodingData());
86 break;
87 case BitCodeAbbrevOp::Char6:
88 Cursor.Read(6);
89 break;
90 }
91 }
92
93
94
95 /// skipRecord - Read the current record and discard it.
skipRecord(unsigned AbbrevID)96 void BitstreamCursor::skipRecord(unsigned AbbrevID) {
97 // Skip unabbreviated records by reading past their entries.
98 if (AbbrevID == bitc::UNABBREV_RECORD) {
99 unsigned Code = ReadVBR(6);
100 (void)Code;
101 unsigned NumElts = ReadVBR(6);
102 for (unsigned i = 0; i != NumElts; ++i)
103 (void)ReadVBR64(6);
104 return;
105 }
106
107 const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
108
109 for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
110 const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
111 if (Op.isLiteral())
112 continue;
113
114 if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
115 Op.getEncoding() != BitCodeAbbrevOp::Blob) {
116 skipAbbreviatedField(*this, Op);
117 continue;
118 }
119
120 if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
121 // Array case. Read the number of elements as a vbr6.
122 unsigned NumElts = ReadVBR(6);
123
124 // Get the element encoding.
125 assert(i+2 == e && "array op not second to last?");
126 const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
127
128 // Read all the elements.
129 for (; NumElts; --NumElts)
130 skipAbbreviatedField(*this, EltEnc);
131 continue;
132 }
133
134 assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
135 // Blob case. Read the number of bytes as a vbr6.
136 unsigned NumElts = ReadVBR(6);
137 SkipToFourByteBoundary(); // 32-bit alignment
138
139 // Figure out where the end of this blob will be including tail padding.
140 size_t NewEnd = GetCurrentBitNo()+((NumElts+3)&~3)*8;
141
142 // If this would read off the end of the bitcode file, just set the
143 // record to empty and return.
144 if (!canSkipToPos(NewEnd/8)) {
145 NextChar = BitStream->getBitcodeBytes().getExtent();
146 break;
147 }
148
149 // Skip over the blob.
150 JumpToBit(NewEnd);
151 }
152 }
153
readRecord(unsigned AbbrevID,SmallVectorImpl<uint64_t> & Vals,StringRef * Blob)154 unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
155 SmallVectorImpl<uint64_t> &Vals,
156 StringRef *Blob) {
157 if (AbbrevID == bitc::UNABBREV_RECORD) {
158 unsigned Code = ReadVBR(6);
159 unsigned NumElts = ReadVBR(6);
160 for (unsigned i = 0; i != NumElts; ++i)
161 Vals.push_back(ReadVBR64(6));
162 return Code;
163 }
164
165 const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
166
167 // Read the record code first.
168 assert(Abbv->getNumOperandInfos() != 0 && "no record code in abbreviation?");
169 const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
170 unsigned Code;
171 if (CodeOp.isLiteral())
172 Code = CodeOp.getLiteralValue();
173 else
174 Code = readAbbreviatedField(*this, CodeOp);
175
176 for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i != e; ++i) {
177 const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
178 if (Op.isLiteral()) {
179 Vals.push_back(Op.getLiteralValue());
180 continue;
181 }
182
183 if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
184 Op.getEncoding() != BitCodeAbbrevOp::Blob) {
185 Vals.push_back(readAbbreviatedField(*this, Op));
186 continue;
187 }
188
189 if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
190 // Array case. Read the number of elements as a vbr6.
191 unsigned NumElts = ReadVBR(6);
192
193 // Get the element encoding.
194 assert(i+2 == e && "array op not second to last?");
195 const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
196
197 // Read all the elements.
198 for (; NumElts; --NumElts)
199 Vals.push_back(readAbbreviatedField(*this, EltEnc));
200 continue;
201 }
202
203 assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
204 // Blob case. Read the number of bytes as a vbr6.
205 unsigned NumElts = ReadVBR(6);
206 SkipToFourByteBoundary(); // 32-bit alignment
207
208 // Figure out where the end of this blob will be including tail padding.
209 size_t CurBitPos = GetCurrentBitNo();
210 size_t NewEnd = CurBitPos+((NumElts+3)&~3)*8;
211
212 // If this would read off the end of the bitcode file, just set the
213 // record to empty and return.
214 if (!canSkipToPos(NewEnd/8)) {
215 Vals.append(NumElts, 0);
216 NextChar = BitStream->getBitcodeBytes().getExtent();
217 break;
218 }
219
220 // Otherwise, inform the streamer that we need these bytes in memory.
221 const char *Ptr = (const char*)
222 BitStream->getBitcodeBytes().getPointer(CurBitPos/8, NumElts);
223
224 // If we can return a reference to the data, do so to avoid copying it.
225 if (Blob) {
226 *Blob = StringRef(Ptr, NumElts);
227 } else {
228 // Otherwise, unpack into Vals with zero extension.
229 for (; NumElts; --NumElts)
230 Vals.push_back((unsigned char)*Ptr++);
231 }
232 // Skip over tail padding.
233 JumpToBit(NewEnd);
234 }
235
236 return Code;
237 }
238
239
ReadAbbrevRecord()240 void BitstreamCursor::ReadAbbrevRecord() {
241 BitCodeAbbrev *Abbv = new BitCodeAbbrev();
242 unsigned NumOpInfo = ReadVBR(5);
243 for (unsigned i = 0; i != NumOpInfo; ++i) {
244 bool IsLiteral = Read(1) ? true : false;
245 if (IsLiteral) {
246 Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
247 continue;
248 }
249
250 BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
251 if (BitCodeAbbrevOp::hasEncodingData(E)) {
252 unsigned Data = ReadVBR64(5);
253
254 // As a special case, handle fixed(0) (i.e., a fixed field with zero bits)
255 // and vbr(0) as a literal zero. This is decoded the same way, and avoids
256 // a slow path in Read() to have to handle reading zero bits.
257 if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
258 Data == 0) {
259 Abbv->Add(BitCodeAbbrevOp(0));
260 continue;
261 }
262
263 Abbv->Add(BitCodeAbbrevOp(E, Data));
264 } else
265 Abbv->Add(BitCodeAbbrevOp(E));
266 }
267 CurAbbrevs.push_back(Abbv);
268 }
269
ReadBlockInfoBlock()270 bool BitstreamCursor::ReadBlockInfoBlock() {
271 // If this is the second stream to get to the block info block, skip it.
272 if (BitStream->hasBlockInfoRecords())
273 return SkipBlock();
274
275 if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
276
277 SmallVector<uint64_t, 64> Record;
278 BitstreamReader::BlockInfo *CurBlockInfo = nullptr;
279
280 // Read all the records for this module.
281 while (1) {
282 BitstreamEntry Entry = advanceSkippingSubblocks(AF_DontAutoprocessAbbrevs);
283
284 switch (Entry.Kind) {
285 case llvm::BitstreamEntry::SubBlock: // Handled for us already.
286 case llvm::BitstreamEntry::Error:
287 return true;
288 case llvm::BitstreamEntry::EndBlock:
289 return false;
290 case llvm::BitstreamEntry::Record:
291 // The interesting case.
292 break;
293 }
294
295 // Read abbrev records, associate them with CurBID.
296 if (Entry.ID == bitc::DEFINE_ABBREV) {
297 if (!CurBlockInfo) return true;
298 ReadAbbrevRecord();
299
300 // ReadAbbrevRecord installs the abbrev in CurAbbrevs. Move it to the
301 // appropriate BlockInfo.
302 CurBlockInfo->Abbrevs.push_back(std::move(CurAbbrevs.back()));
303 CurAbbrevs.pop_back();
304 continue;
305 }
306
307 // Read a record.
308 Record.clear();
309 switch (readRecord(Entry.ID, Record)) {
310 default: break; // Default behavior, ignore unknown content.
311 case bitc::BLOCKINFO_CODE_SETBID:
312 if (Record.size() < 1) return true;
313 CurBlockInfo = &BitStream->getOrCreateBlockInfo((unsigned)Record[0]);
314 break;
315 case bitc::BLOCKINFO_CODE_BLOCKNAME: {
316 if (!CurBlockInfo) return true;
317 if (BitStream->isIgnoringBlockInfoNames()) break; // Ignore name.
318 std::string Name;
319 for (unsigned i = 0, e = Record.size(); i != e; ++i)
320 Name += (char)Record[i];
321 CurBlockInfo->Name = Name;
322 break;
323 }
324 case bitc::BLOCKINFO_CODE_SETRECORDNAME: {
325 if (!CurBlockInfo) return true;
326 if (BitStream->isIgnoringBlockInfoNames()) break; // Ignore name.
327 std::string Name;
328 for (unsigned i = 1, e = Record.size(); i != e; ++i)
329 Name += (char)Record[i];
330 CurBlockInfo->RecordNames.push_back(std::make_pair((unsigned)Record[0],
331 Name));
332 break;
333 }
334 }
335 }
336 }
337
338