1 //==-- llvm/CodeGen/DwarfAccelTable.h - Dwarf Accelerator Tables -*- C++ -*-==//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains support for writing dwarf accelerator tables.
12 //===----------------------------------------------------------------------===//
14 #ifndef CODEGEN_ASMPRINTER_DWARFACCELTABLE_H__
15 #define CODEGEN_ASMPRINTER_DWARFACCELTABLE_H__
18 #include "llvm/ADT/ArrayRef.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/MC/MCSymbol.h"
21 #include "llvm/Support/DataTypes.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/Dwarf.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/Format.h"
26 #include "llvm/Support/FormattedStream.h"
30 // The dwarf accelerator tables are an indirect hash table optimized
31 // for null lookup rather than access to known data. They are output into
32 // an on-disk format that looks like this:
46 // where the header contains a magic number, version, type of hash function,
47 // the number of buckets, total number of hashes, and room for a special
48 // struct of data and the length of that struct.
50 // The buckets contain an index (e.g. 6) into the hashes array. The hashes
51 // section contains all of the 32-bit hash values in contiguous memory, and
52 // the offsets contain the offset into the data area for the particular
55 // For a lookup example, we could hash a function name and take it modulo the
56 // number of buckets giving us our bucket. From there we take the bucket value
57 // as an index into the hashes table and look at each successive hash as long
58 // as the hash value is still the same modulo result (bucket value) as earlier.
59 // If we have a match we look at that same entry in the offsets table and
60 // grab the offset in the data for our final match.
68 class DwarfAccelTable {
70 enum HashFunctionType {
74 static uint32_t HashDJB (StringRef Str) {
76 for (unsigned i = 0, e = Str.size(); i != e; ++i)
77 h = ((h << 5) + h) + Str[i];
81 // Helper function to compute the number of buckets needed based on
82 // the number of unique hashes.
83 void ComputeBucketCount (void);
86 uint32_t magic; // 'HASH' magic value to allow endian detection
87 uint16_t version; // Version number.
88 uint16_t hash_function; // The hash function enumeration that was used.
89 uint32_t bucket_count; // The number of buckets in this hash table.
90 uint32_t hashes_count; // The total number of unique hash values
91 // and hash data offsets in this table.
92 uint32_t header_data_len; // The bytes to skip to get to the hash
93 // indexes (buckets) for correct alignment.
94 // Also written to disk is the implementation specific header data.
96 static const uint32_t MagicHash = 0x48415348;
98 TableHeader (uint32_t data_len) :
99 magic (MagicHash), version (1), hash_function (eHashFunctionDJB),
100 bucket_count (0), hashes_count (0), header_data_len (data_len)
104 void print(raw_ostream &O) {
105 O << "Magic: " << format("0x%x", magic) << "\n"
106 << "Version: " << version << "\n"
107 << "Hash Function: " << hash_function << "\n"
108 << "Bucket Count: " << bucket_count << "\n"
109 << "Header Data Length: " << header_data_len << "\n";
111 void dump() { print(dbgs()); }
116 // The HeaderData describes the form of each set of data. In general this
117 // is as a list of atoms (atom_count) where each atom contains a type
118 // (AtomType type) of data, and an encoding form (form). In the case of
119 // data that is referenced via DW_FORM_ref_* the die_offset_base is
120 // used to describe the offset for all forms in the list of atoms.
121 // This also serves as a public interface of sorts.
122 // When written to disk this will have the form:
124 // uint32_t die_offset_base
125 // uint32_t atom_count
129 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
130 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that
131 // contains the item in question
132 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as
133 // DW_FORM_data1 (if no tags exceed 255) or
135 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
136 eAtomTypeTypeFlags = 5u // Flags from enum TypeFlags
140 eTypeFlagClassMask = 0x0000000fu,
142 // Always set for C++, only set for ObjC if this is the
143 // @implementation for a class.
144 eTypeFlagClassIsImplementation = ( 1u << 1 )
147 // Make these public so that they can be used as a general interface to
150 AtomType type; // enum AtomType
151 uint16_t form; // DWARF DW_FORM_ defines
153 Atom(AtomType type, uint16_t form) : type(type), form(form) {}
154 static const char * AtomTypeString(enum AtomType);
156 void print(raw_ostream &O) {
157 O << "Type: " << AtomTypeString(type) << "\n"
158 << "Form: " << dwarf::FormEncodingString(form) << "\n";
167 struct TableHeaderData {
168 uint32_t die_offset_base;
169 SmallVector<Atom, 1> Atoms;
171 TableHeaderData(ArrayRef<Atom> AtomList, uint32_t offset = 0)
172 : die_offset_base(offset), Atoms(AtomList.begin(), AtomList.end()) { }
175 void print (raw_ostream &O) {
176 O << "die_offset_base: " << die_offset_base << "\n";
177 for (size_t i = 0; i < Atoms.size(); i++)
186 // The data itself consists of a str_offset, a count of the DIEs in the
187 // hash and the offsets to the DIEs themselves.
188 // On disk each data section is ended with a 0 KeyType as the end of the
190 // On output this looks like:
191 // uint32_t str_offset
192 // uint32_t hash_data_count
193 // HashData[hash_data_count]
195 struct HashDataContents {
197 char Flags; // Specific flags to output
199 HashDataContents(DIE *D, char Flags) :
203 void print(raw_ostream &O) const {
204 O << " Offset: " << Die->getOffset() << "\n";
205 O << " Tag: " << dwarf::TagString(Die->getTag()) << "\n";
206 O << " Flags: " << Flags << "\n";
215 ArrayRef<HashDataContents*> Data; // offsets
216 HashData(StringRef S, ArrayRef<HashDataContents*> Data)
217 : Str(S), Data(Data) {
218 HashValue = DwarfAccelTable::HashDJB(S);
221 void print(raw_ostream &O) {
222 O << "Name: " << Str << "\n";
223 O << " Hash Value: " << format("0x%x", HashValue) << "\n";
225 if (Sym) Sym->print(O);
228 for (size_t i = 0; i < Data.size(); i++) {
229 O << " Offset: " << Data[i]->Die->getOffset() << "\n";
230 O << " Tag: " << dwarf::TagString(Data[i]->Die->getTag()) << "\n";
231 O << " Flags: " << Data[i]->Flags << "\n";
240 DwarfAccelTable(const DwarfAccelTable&) LLVM_DELETED_FUNCTION;
241 void operator=(const DwarfAccelTable&) LLVM_DELETED_FUNCTION;
243 // Internal Functions
244 void EmitHeader(AsmPrinter *);
245 void EmitBuckets(AsmPrinter *);
246 void EmitHashes(AsmPrinter *);
247 void EmitOffsets(AsmPrinter *, MCSymbol *);
248 void EmitData(AsmPrinter *, DwarfUnits *D);
250 // Allocator for HashData and HashDataContents.
251 BumpPtrAllocator Allocator;
255 TableHeaderData HeaderData;
256 std::vector<HashData*> Data;
259 typedef std::vector<HashDataContents*> DataArray;
260 typedef StringMap<DataArray, BumpPtrAllocator&> StringEntries;
261 StringEntries Entries;
263 // Buckets/Hashes/Offsets
264 typedef std::vector<HashData*> HashList;
265 typedef std::vector<HashList> BucketList;
269 // Public Implementation
271 DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom>);
273 void AddName(StringRef, DIE*, char = 0);
274 void FinalizeTable(AsmPrinter *, StringRef);
275 void Emit(AsmPrinter *, MCSymbol *, DwarfUnits *);
277 void print(raw_ostream &O);
278 void dump() { print(dbgs()); }