1 //===-- llvm/Bitcode/ReaderWriter.h - Bitcode reader/writers ----*- C++ -*-===//
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 // This header defines interfaces to read and write LLVM bitcode files/streams.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_BITCODE_READERWRITER_H
15 #define LLVM_BITCODE_READERWRITER_H
16 
17 #include "llvm/IR/DiagnosticInfo.h"
18 #include "llvm/IR/FunctionInfo.h"
19 #include "llvm/Support/Endian.h"
20 #include "llvm/Support/ErrorOr.h"
21 #include "llvm/Support/MemoryBuffer.h"
22 #include <memory>
23 #include <string>
24 
25 namespace llvm {
26   class BitstreamWriter;
27   class DataStreamer;
28   class LLVMContext;
29   class Module;
30   class ModulePass;
31   class raw_ostream;
32 
33   /// Read the header of the specified bitcode buffer and prepare for lazy
34   /// deserialization of function bodies. If ShouldLazyLoadMetadata is true,
35   /// lazily load metadata as well. If successful, this moves Buffer. On
36   /// error, this *does not* move Buffer.
37   ErrorOr<std::unique_ptr<Module>>
38   getLazyBitcodeModule(std::unique_ptr<MemoryBuffer> &&Buffer,
39                        LLVMContext &Context,
40                        bool ShouldLazyLoadMetadata = false);
41 
42   /// Read the header of the specified stream and prepare for lazy
43   /// deserialization and streaming of function bodies.
44   ErrorOr<std::unique_ptr<Module>>
45   getStreamedBitcodeModule(StringRef Name,
46                            std::unique_ptr<DataStreamer> Streamer,
47                            LLVMContext &Context);
48 
49   /// Read the header of the specified bitcode buffer and extract just the
50   /// triple information. If successful, this returns a string. On error, this
51   /// returns "".
52   std::string getBitcodeTargetTriple(MemoryBufferRef Buffer,
53                                      LLVMContext &Context);
54 
55   /// Read the header of the specified bitcode buffer and extract just the
56   /// producer string information. If successful, this returns a string. On
57   /// error, this returns "".
58   std::string getBitcodeProducerString(MemoryBufferRef Buffer,
59                                        LLVMContext &Context);
60 
61   /// Read the specified bitcode file, returning the module.
62   ErrorOr<std::unique_ptr<Module>> parseBitcodeFile(MemoryBufferRef Buffer,
63                                                     LLVMContext &Context);
64 
65   /// Check if the given bitcode buffer contains a function summary block.
66   bool hasFunctionSummary(MemoryBufferRef Buffer,
67                           DiagnosticHandlerFunction DiagnosticHandler);
68 
69   /// Parse the specified bitcode buffer, returning the function info index.
70   /// If IsLazy is true, parse the entire function summary into
71   /// the index. Otherwise skip the function summary section, and only create
72   /// an index object with a map from function name to function summary offset.
73   /// The index is used to perform lazy function summary reading later.
74   ErrorOr<std::unique_ptr<FunctionInfoIndex>>
75   getFunctionInfoIndex(MemoryBufferRef Buffer,
76                        DiagnosticHandlerFunction DiagnosticHandler,
77                        bool IsLazy = false);
78 
79   /// This method supports lazy reading of function summary data from the
80   /// combined index during function importing. When reading the combined index
81   /// file, getFunctionInfoIndex is first invoked with IsLazy=true.
82   /// Then this method is called for each function considered for importing,
83   /// to parse the summary information for the given function name into
84   /// the index.
85   std::error_code readFunctionSummary(
86       MemoryBufferRef Buffer, DiagnosticHandlerFunction DiagnosticHandler,
87       StringRef FunctionName, std::unique_ptr<FunctionInfoIndex> Index);
88 
89   /// \brief Write the specified module to the specified raw output stream.
90   ///
91   /// For streams where it matters, the given stream should be in "binary"
92   /// mode.
93   ///
94   /// If \c ShouldPreserveUseListOrder, encode the use-list order for each \a
95   /// Value in \c M.  These will be reconstructed exactly when \a M is
96   /// deserialized.
97   ///
98   /// If \c EmitFunctionSummary, emit the function summary index (currently
99   /// for use in ThinLTO optimization).
100   void WriteBitcodeToFile(const Module *M, raw_ostream &Out,
101                           bool ShouldPreserveUseListOrder = false,
102                           bool EmitFunctionSummary = false);
103 
104   /// Write the specified function summary index to the given raw output stream,
105   /// where it will be written in a new bitcode block. This is used when
106   /// writing the combined index file for ThinLTO.
107   void WriteFunctionSummaryToFile(const FunctionInfoIndex &Index,
108                                   raw_ostream &Out);
109 
110   /// isBitcodeWrapper - Return true if the given bytes are the magic bytes
111   /// for an LLVM IR bitcode wrapper.
112   ///
isBitcodeWrapper(const unsigned char * BufPtr,const unsigned char * BufEnd)113   inline bool isBitcodeWrapper(const unsigned char *BufPtr,
114                                const unsigned char *BufEnd) {
115     // See if you can find the hidden message in the magic bytes :-).
116     // (Hint: it's a little-endian encoding.)
117     return BufPtr != BufEnd &&
118            BufPtr[0] == 0xDE &&
119            BufPtr[1] == 0xC0 &&
120            BufPtr[2] == 0x17 &&
121            BufPtr[3] == 0x0B;
122   }
123 
124   /// isRawBitcode - Return true if the given bytes are the magic bytes for
125   /// raw LLVM IR bitcode (without a wrapper).
126   ///
isRawBitcode(const unsigned char * BufPtr,const unsigned char * BufEnd)127   inline bool isRawBitcode(const unsigned char *BufPtr,
128                            const unsigned char *BufEnd) {
129     // These bytes sort of have a hidden message, but it's not in
130     // little-endian this time, and it's a little redundant.
131     return BufPtr != BufEnd &&
132            BufPtr[0] == 'B' &&
133            BufPtr[1] == 'C' &&
134            BufPtr[2] == 0xc0 &&
135            BufPtr[3] == 0xde;
136   }
137 
138   /// isBitcode - Return true if the given bytes are the magic bytes for
139   /// LLVM IR bitcode, either with or without a wrapper.
140   ///
isBitcode(const unsigned char * BufPtr,const unsigned char * BufEnd)141   inline bool isBitcode(const unsigned char *BufPtr,
142                         const unsigned char *BufEnd) {
143     return isBitcodeWrapper(BufPtr, BufEnd) ||
144            isRawBitcode(BufPtr, BufEnd);
145   }
146 
147   /// SkipBitcodeWrapperHeader - Some systems wrap bc files with a special
148   /// header for padding or other reasons.  The format of this header is:
149   ///
150   /// struct bc_header {
151   ///   uint32_t Magic;         // 0x0B17C0DE
152   ///   uint32_t Version;       // Version, currently always 0.
153   ///   uint32_t BitcodeOffset; // Offset to traditional bitcode file.
154   ///   uint32_t BitcodeSize;   // Size of traditional bitcode file.
155   ///   ... potentially other gunk ...
156   /// };
157   ///
158   /// This function is called when we find a file with a matching magic number.
159   /// In this case, skip down to the subsection of the file that is actually a
160   /// BC file.
161   /// If 'VerifyBufferSize' is true, check that the buffer is large enough to
162   /// contain the whole bitcode file.
SkipBitcodeWrapperHeader(const unsigned char * & BufPtr,const unsigned char * & BufEnd,bool VerifyBufferSize)163   inline bool SkipBitcodeWrapperHeader(const unsigned char *&BufPtr,
164                                        const unsigned char *&BufEnd,
165                                        bool VerifyBufferSize) {
166     enum {
167       KnownHeaderSize = 4*4,  // Size of header we read.
168       OffsetField = 2*4,      // Offset in bytes to Offset field.
169       SizeField = 3*4         // Offset in bytes to Size field.
170     };
171 
172     // Must contain the header!
173     if (BufEnd-BufPtr < KnownHeaderSize) return true;
174 
175     unsigned Offset = support::endian::read32le(&BufPtr[OffsetField]);
176     unsigned Size = support::endian::read32le(&BufPtr[SizeField]);
177 
178     // Verify that Offset+Size fits in the file.
179     if (VerifyBufferSize && Offset+Size > unsigned(BufEnd-BufPtr))
180       return true;
181     BufPtr += Offset;
182     BufEnd = BufPtr+Size;
183     return false;
184   }
185 
186   const std::error_category &BitcodeErrorCategory();
187   enum class BitcodeError { InvalidBitcodeSignature = 1, CorruptedBitcode };
make_error_code(BitcodeError E)188   inline std::error_code make_error_code(BitcodeError E) {
189     return std::error_code(static_cast<int>(E), BitcodeErrorCategory());
190   }
191 
192   class BitcodeDiagnosticInfo : public DiagnosticInfo {
193     const Twine &Msg;
194     std::error_code EC;
195 
196   public:
197     BitcodeDiagnosticInfo(std::error_code EC, DiagnosticSeverity Severity,
198                           const Twine &Msg);
199     void print(DiagnosticPrinter &DP) const override;
getError()200     std::error_code getError() const { return EC; }
201 
classof(const DiagnosticInfo * DI)202     static bool classof(const DiagnosticInfo *DI) {
203       return DI->getKind() == DK_Bitcode;
204     }
205   };
206 
207 } // End llvm namespace
208 
209 namespace std {
210 template <> struct is_error_code_enum<llvm::BitcodeError> : std::true_type {};
211 }
212 
213 #endif
214