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) {}
103 void print(raw_ostream &O) {
104 O << "Magic: " << format("0x%x", magic) << "\n"
105 << "Version: " << version << "\n"
106 << "Hash Function: " << hash_function << "\n"
107 << "Bucket Count: " << bucket_count << "\n"
108 << "Header Data Length: " << header_data_len << "\n";
110 void dump() { print(dbgs()); }
115 // The HeaderData describes the form of each set of data. In general this
116 // is as a list of atoms (atom_count) where each atom contains a type
117 // (AtomType type) of data, and an encoding form (form). In the case of
118 // data that is referenced via DW_FORM_ref_* the die_offset_base is
119 // used to describe the offset for all forms in the list of atoms.
120 // This also serves as a public interface of sorts.
121 // When written to disk this will have the form:
123 // uint32_t die_offset_base
124 // uint32_t atom_count
128 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
129 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that
130 // contains the item in question
131 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as
132 // DW_FORM_data1 (if no tags exceed 255) or
134 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
135 eAtomTypeTypeFlags = 5u // Flags from enum TypeFlags
139 eTypeFlagClassMask = 0x0000000fu,
141 // Always set for C++, only set for ObjC if this is the
142 // @implementation for a class.
143 eTypeFlagClassIsImplementation = (1u << 1)
146 // Make these public so that they can be used as a general interface to
149 AtomType type; // enum AtomType
150 uint16_t form; // DWARF DW_FORM_ defines
152 Atom(AtomType type, uint16_t form) : type(type), form(form) {}
153 static const char *AtomTypeString(enum AtomType);
155 void print(raw_ostream &O) {
156 O << "Type: " << AtomTypeString(type) << "\n"
157 << "Form: " << dwarf::FormEncodingString(form) << "\n";
159 void dump() { print(dbgs()); }
164 struct TableHeaderData {
165 uint32_t die_offset_base;
166 SmallVector<Atom, 1> Atoms;
168 TableHeaderData(ArrayRef<Atom> AtomList, uint32_t offset = 0)
169 : die_offset_base(offset), Atoms(AtomList.begin(), AtomList.end()) {}
172 void print(raw_ostream &O) {
173 O << "die_offset_base: " << die_offset_base << "\n";
174 for (size_t i = 0; i < Atoms.size(); i++)
177 void dump() { print(dbgs()); }
181 // The data itself consists of a str_offset, a count of the DIEs in the
182 // hash and the offsets to the DIEs themselves.
183 // On disk each data section is ended with a 0 KeyType as the end of the
185 // On output this looks like:
186 // uint32_t str_offset
187 // uint32_t hash_data_count
188 // HashData[hash_data_count]
190 struct HashDataContents {
192 char Flags; // Specific flags to output
194 HashDataContents(DIE *D, char Flags) : Die(D), Flags(Flags) {}
196 void print(raw_ostream &O) const {
197 O << " Offset: " << Die->getOffset() << "\n";
198 O << " Tag: " << dwarf::TagString(Die->getTag()) << "\n";
199 O << " Flags: " << Flags << "\n";
209 ArrayRef<HashDataContents *> Data; // offsets
210 HashData(StringRef S, ArrayRef<HashDataContents *> Data)
211 : Str(S), Data(Data) {
212 HashValue = DwarfAccelTable::HashDJB(S);
215 void print(raw_ostream &O) {
216 O << "Name: " << Str << "\n";
217 O << " Hash Value: " << format("0x%x", HashValue) << "\n";
224 for (size_t i = 0; i < Data.size(); i++) {
225 O << " Offset: " << Data[i]->Die->getOffset() << "\n";
226 O << " Tag: " << dwarf::TagString(Data[i]->Die->getTag()) << "\n";
227 O << " Flags: " << Data[i]->Flags << "\n";
230 void dump() { print(dbgs()); }
234 DwarfAccelTable(const DwarfAccelTable &) LLVM_DELETED_FUNCTION;
235 void operator=(const DwarfAccelTable &) LLVM_DELETED_FUNCTION;
237 // Internal Functions
238 void EmitHeader(AsmPrinter *);
239 void EmitBuckets(AsmPrinter *);
240 void EmitHashes(AsmPrinter *);
241 void EmitOffsets(AsmPrinter *, MCSymbol *);
242 void EmitData(AsmPrinter *, DwarfUnits *D);
244 // Allocator for HashData and HashDataContents.
245 BumpPtrAllocator Allocator;
249 TableHeaderData HeaderData;
250 std::vector<HashData *> Data;
253 typedef std::vector<HashDataContents *> DataArray;
254 typedef StringMap<DataArray, BumpPtrAllocator &> StringEntries;
255 StringEntries Entries;
257 // Buckets/Hashes/Offsets
258 typedef std::vector<HashData *> HashList;
259 typedef std::vector<HashList> BucketList;
263 // Public Implementation
265 DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom>);
267 void AddName(StringRef, DIE *, char = 0);
268 void FinalizeTable(AsmPrinter *, StringRef);
269 void Emit(AsmPrinter *, MCSymbol *, DwarfUnits *);
271 void print(raw_ostream &O);
272 void dump() { print(dbgs()); }