DwarfAccelTable.h revision 234982
1234285Sdim//==-- llvm/CodeGen/DwarfAccelTable.h - Dwarf Accelerator Tables -*- C++ -*-==// 2234285Sdim// 3234285Sdim// The LLVM Compiler Infrastructure 4234285Sdim// 5234285Sdim// This file is distributed under the University of Illinois Open Source 6234285Sdim// License. See LICENSE.TXT for details. 7234285Sdim// 8234285Sdim//===----------------------------------------------------------------------===// 9234285Sdim// 10234285Sdim// This file contains support for writing dwarf accelerator tables. 11234285Sdim// 12234285Sdim//===----------------------------------------------------------------------===// 13234285Sdim 14234285Sdim#ifndef CODEGEN_ASMPRINTER_DWARFACCELTABLE_H__ 15234285Sdim#define CODEGEN_ASMPRINTER_DWARFACCELTABLE_H__ 16234285Sdim 17234285Sdim#include "llvm/ADT/StringMap.h" 18234982Sdim#include "llvm/ADT/ArrayRef.h" 19234285Sdim#include "llvm/MC/MCSymbol.h" 20234285Sdim#include "llvm/Support/Dwarf.h" 21234285Sdim#include "llvm/Support/DataTypes.h" 22234285Sdim#include "llvm/Support/Debug.h" 23234285Sdim#include "llvm/Support/ErrorHandling.h" 24234285Sdim#include "llvm/Support/Format.h" 25234285Sdim#include "llvm/Support/FormattedStream.h" 26234285Sdim#include "DIE.h" 27234285Sdim#include <vector> 28234285Sdim#include <map> 29234285Sdim 30234285Sdim// The dwarf accelerator tables are an indirect hash table optimized 31234285Sdim// for null lookup rather than access to known data. They are output into 32234285Sdim// an on-disk format that looks like this: 33234285Sdim// 34234285Sdim// .-------------. 35234285Sdim// | HEADER | 36234285Sdim// |-------------| 37234285Sdim// | BUCKETS | 38234285Sdim// |-------------| 39234285Sdim// | HASHES | 40234285Sdim// |-------------| 41234285Sdim// | OFFSETS | 42234285Sdim// |-------------| 43234285Sdim// | DATA | 44234285Sdim// `-------------' 45234285Sdim// 46234285Sdim// where the header contains a magic number, version, type of hash function, 47234285Sdim// the number of buckets, total number of hashes, and room for a special 48234285Sdim// struct of data and the length of that struct. 49234285Sdim// 50234285Sdim// The buckets contain an index (e.g. 6) into the hashes array. The hashes 51234285Sdim// section contains all of the 32-bit hash values in contiguous memory, and 52234285Sdim// the offsets contain the offset into the data area for the particular 53234285Sdim// hash. 54234285Sdim// 55234285Sdim// For a lookup example, we could hash a function name and take it modulo the 56234285Sdim// number of buckets giving us our bucket. From there we take the bucket value 57234285Sdim// as an index into the hashes table and look at each successive hash as long 58234285Sdim// as the hash value is still the same modulo result (bucket value) as earlier. 59234285Sdim// If we have a match we look at that same entry in the offsets table and 60234285Sdim// grab the offset in the data for our final match. 61234285Sdim 62234285Sdimnamespace llvm { 63234285Sdim 64234285Sdimclass AsmPrinter; 65234285Sdimclass DIE; 66234285Sdimclass DwarfDebug; 67234285Sdim 68234285Sdimclass DwarfAccelTable { 69234285Sdim 70234285Sdim enum HashFunctionType { 71234285Sdim eHashFunctionDJB = 0u 72234285Sdim }; 73234285Sdim 74234285Sdim static uint32_t HashDJB (StringRef Str) { 75234285Sdim uint32_t h = 5381; 76234285Sdim for (unsigned i = 0, e = Str.size(); i != e; ++i) 77234285Sdim h = ((h << 5) + h) + Str[i]; 78234285Sdim return h; 79234285Sdim } 80234285Sdim 81234285Sdim // Helper function to compute the number of buckets needed based on 82234285Sdim // the number of unique hashes. 83234285Sdim void ComputeBucketCount (void); 84234285Sdim 85234285Sdim struct TableHeader { 86234285Sdim uint32_t magic; // 'HASH' magic value to allow endian detection 87234285Sdim uint16_t version; // Version number. 88234285Sdim uint16_t hash_function; // The hash function enumeration that was used. 89234285Sdim uint32_t bucket_count; // The number of buckets in this hash table. 90234285Sdim uint32_t hashes_count; // The total number of unique hash values 91234285Sdim // and hash data offsets in this table. 92234285Sdim uint32_t header_data_len; // The bytes to skip to get to the hash 93234285Sdim // indexes (buckets) for correct alignment. 94234285Sdim // Also written to disk is the implementation specific header data. 95234285Sdim 96234285Sdim static const uint32_t MagicHash = 0x48415348; 97234285Sdim 98234285Sdim TableHeader (uint32_t data_len) : 99234285Sdim magic (MagicHash), version (1), hash_function (eHashFunctionDJB), 100234285Sdim bucket_count (0), hashes_count (0), header_data_len (data_len) 101234285Sdim {} 102234285Sdim 103234285Sdim#ifndef NDEBUG 104234285Sdim void print(raw_ostream &O) { 105234285Sdim O << "Magic: " << format("0x%x", magic) << "\n" 106234285Sdim << "Version: " << version << "\n" 107234285Sdim << "Hash Function: " << hash_function << "\n" 108234285Sdim << "Bucket Count: " << bucket_count << "\n" 109234285Sdim << "Header Data Length: " << header_data_len << "\n"; 110234285Sdim } 111234285Sdim void dump() { print(dbgs()); } 112234285Sdim#endif 113234285Sdim }; 114234285Sdim 115234285Sdimpublic: 116234285Sdim // The HeaderData describes the form of each set of data. In general this 117234285Sdim // is as a list of atoms (atom_count) where each atom contains a type 118234285Sdim // (AtomType type) of data, and an encoding form (form). In the case of 119234285Sdim // data that is referenced via DW_FORM_ref_* the die_offset_base is 120234285Sdim // used to describe the offset for all forms in the list of atoms. 121234285Sdim // This also serves as a public interface of sorts. 122234285Sdim // When written to disk this will have the form: 123234285Sdim // 124234285Sdim // uint32_t die_offset_base 125234285Sdim // uint32_t atom_count 126234285Sdim // atom_count Atoms 127234285Sdim enum AtomType { 128234285Sdim eAtomTypeNULL = 0u, 129234285Sdim eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding 130234285Sdim eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that 131234285Sdim // contains the item in question 132234285Sdim eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as 133234285Sdim // DW_FORM_data1 (if no tags exceed 255) or 134234285Sdim // DW_FORM_data2. 135234285Sdim eAtomTypeNameFlags = 4u, // Flags from enum NameFlags 136234285Sdim eAtomTypeTypeFlags = 5u // Flags from enum TypeFlags 137234285Sdim }; 138234285Sdim 139234285Sdim enum TypeFlags { 140234285Sdim eTypeFlagClassMask = 0x0000000fu, 141234285Sdim 142234285Sdim // Always set for C++, only set for ObjC if this is the 143234285Sdim // @implementation for a class. 144234285Sdim eTypeFlagClassIsImplementation = ( 1u << 1 ) 145234285Sdim }; 146234285Sdim 147234285Sdim // Make these public so that they can be used as a general interface to 148234285Sdim // the class. 149234285Sdim struct Atom { 150234285Sdim AtomType type; // enum AtomType 151234285Sdim uint16_t form; // DWARF DW_FORM_ defines 152234285Sdim 153234285Sdim Atom(AtomType type, uint16_t form) : type(type), form(form) {} 154234285Sdim static const char * AtomTypeString(enum AtomType); 155234285Sdim#ifndef NDEBUG 156234285Sdim void print(raw_ostream &O) { 157234285Sdim O << "Type: " << AtomTypeString(type) << "\n" 158234285Sdim << "Form: " << dwarf::FormEncodingString(form) << "\n"; 159234285Sdim } 160234285Sdim void dump() { 161234285Sdim print(dbgs()); 162234285Sdim } 163234285Sdim#endif 164234285Sdim }; 165234285Sdim 166234285Sdim private: 167234285Sdim struct TableHeaderData { 168234285Sdim uint32_t die_offset_base; 169234982Sdim SmallVector<Atom, 1> Atoms; 170234285Sdim 171234982Sdim TableHeaderData(ArrayRef<Atom> AtomList, uint32_t offset = 0) 172234982Sdim : die_offset_base(offset), Atoms(AtomList.begin(), AtomList.end()) { } 173234982Sdim 174234285Sdim#ifndef NDEBUG 175234285Sdim void print (raw_ostream &O) { 176234285Sdim O << "die_offset_base: " << die_offset_base << "\n"; 177234285Sdim for (size_t i = 0; i < Atoms.size(); i++) 178234285Sdim Atoms[i].print(O); 179234285Sdim } 180234285Sdim void dump() { 181234285Sdim print(dbgs()); 182234285Sdim } 183234285Sdim#endif 184234285Sdim }; 185234285Sdim 186234285Sdim // The data itself consists of a str_offset, a count of the DIEs in the 187234285Sdim // hash and the offsets to the DIEs themselves. 188234285Sdim // On disk each data section is ended with a 0 KeyType as the end of the 189234285Sdim // hash chain. 190234285Sdim // On output this looks like: 191234285Sdim // uint32_t str_offset 192234285Sdim // uint32_t hash_data_count 193234285Sdim // HashData[hash_data_count] 194234285Sdimpublic: 195234285Sdim struct HashDataContents { 196234285Sdim DIE *Die; // Offsets 197234285Sdim char Flags; // Specific flags to output 198234285Sdim 199234285Sdim HashDataContents(DIE *D, char Flags) : 200234285Sdim Die(D), 201234285Sdim Flags(Flags) { } 202234285Sdim #ifndef NDEBUG 203234285Sdim void print(raw_ostream &O) const { 204234285Sdim O << " Offset: " << Die->getOffset() << "\n"; 205234285Sdim O << " Tag: " << dwarf::TagString(Die->getTag()) << "\n"; 206234285Sdim O << " Flags: " << Flags << "\n"; 207234285Sdim } 208234285Sdim #endif 209234285Sdim }; 210234285Sdimprivate: 211234285Sdim struct HashData { 212234285Sdim StringRef Str; 213234285Sdim uint32_t HashValue; 214234285Sdim MCSymbol *Sym; 215234982Sdim ArrayRef<HashDataContents*> Data; // offsets 216234982Sdim HashData(StringRef S, ArrayRef<HashDataContents*> Data) 217234982Sdim : Str(S), Data(Data) { 218234285Sdim HashValue = DwarfAccelTable::HashDJB(S); 219234285Sdim } 220234285Sdim #ifndef NDEBUG 221234285Sdim void print(raw_ostream &O) { 222234285Sdim O << "Name: " << Str << "\n"; 223234285Sdim O << " Hash Value: " << format("0x%x", HashValue) << "\n"; 224234285Sdim O << " Symbol: " ; 225234285Sdim if (Sym) Sym->print(O); 226234285Sdim else O << "<none>"; 227234285Sdim O << "\n"; 228234285Sdim for (size_t i = 0; i < Data.size(); i++) { 229234285Sdim O << " Offset: " << Data[i]->Die->getOffset() << "\n"; 230234285Sdim O << " Tag: " << dwarf::TagString(Data[i]->Die->getTag()) << "\n"; 231234285Sdim O << " Flags: " << Data[i]->Flags << "\n"; 232234285Sdim } 233234285Sdim } 234234285Sdim void dump() { 235234285Sdim print(dbgs()); 236234285Sdim } 237234285Sdim #endif 238234285Sdim }; 239234285Sdim 240234285Sdim DwarfAccelTable(const DwarfAccelTable&); // DO NOT IMPLEMENT 241234285Sdim void operator=(const DwarfAccelTable&); // DO NOT IMPLEMENT 242234285Sdim 243234285Sdim // Internal Functions 244234285Sdim void EmitHeader(AsmPrinter *); 245234285Sdim void EmitBuckets(AsmPrinter *); 246234285Sdim void EmitHashes(AsmPrinter *); 247234285Sdim void EmitOffsets(AsmPrinter *, MCSymbol *); 248234285Sdim void EmitData(AsmPrinter *, DwarfDebug *D); 249234982Sdim 250234982Sdim // Allocator for HashData and HashDataContents. 251234982Sdim BumpPtrAllocator Allocator; 252234982Sdim 253234285Sdim // Output Variables 254234285Sdim TableHeader Header; 255234285Sdim TableHeaderData HeaderData; 256234285Sdim std::vector<HashData*> Data; 257234285Sdim 258234285Sdim // String Data 259234982Sdim typedef std::vector<HashDataContents*> DataArray; 260234982Sdim typedef StringMap<DataArray, BumpPtrAllocator&> StringEntries; 261234285Sdim StringEntries Entries; 262234285Sdim 263234285Sdim // Buckets/Hashes/Offsets 264234285Sdim typedef std::vector<HashData*> HashList; 265234285Sdim typedef std::vector<HashList> BucketList; 266234285Sdim BucketList Buckets; 267234285Sdim HashList Hashes; 268234285Sdim 269234285Sdim // Public Implementation 270234285Sdim public: 271234982Sdim DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom>); 272234285Sdim ~DwarfAccelTable(); 273234285Sdim void AddName(StringRef, DIE*, char = 0); 274234285Sdim void FinalizeTable(AsmPrinter *, const char *); 275234285Sdim void Emit(AsmPrinter *, MCSymbol *, DwarfDebug *); 276234285Sdim#ifndef NDEBUG 277234285Sdim void print(raw_ostream &O); 278234285Sdim void dump() { print(dbgs()); } 279234285Sdim#endif 280234285Sdim}; 281234285Sdim 282234285Sdim} 283234285Sdim#endif 284