DwarfAccelTable.cpp revision be3f051c49699a86d526833d7dbe95645680a340 
1//=-- llvm/CodeGen/DwarfAccelTable.cpp - Dwarf Accelerator Tables -*- 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 file contains support for writing dwarf accelerator tables.
11//
12//===----------------------------------------------------------------------===//
13
14#include "DwarfAccelTable.h"
15#include "DwarfDebug.h"
16#include "DIE.h"
17#include "llvm/ADT/Twine.h"
18#include "llvm/ADT/STLExtras.h"
19#include "llvm/CodeGen/AsmPrinter.h"
20#include "llvm/MC/MCExpr.h"
21#include "llvm/MC/MCStreamer.h"
22#include "llvm/MC/MCSymbol.h"
23#include "llvm/Support/Debug.h"
24
25using namespace llvm;
26
27const char *DwarfAccelTable::Atom::AtomTypeString(enum AtomType AT) {
28  switch (AT) {
29  case eAtomTypeNULL: return "eAtomTypeNULL";
30  case eAtomTypeDIEOffset: return "eAtomTypeDIEOffset";
31  case eAtomTypeCUOffset: return "eAtomTypeCUOffset";
32  case eAtomTypeTag: return "eAtomTypeTag";
33  case eAtomTypeNameFlags: return "eAtomTypeNameFlags";
34  case eAtomTypeTypeFlags: return "eAtomTypeTypeFlags";
35  }
36  llvm_unreachable("invalid AtomType!");
37}
38
39// The general case would need to have a less hard coded size for the
40// length of the HeaderData, however, if we're constructing based on a
41// single Atom then we know it will always be: 4 + 4 + 2 + 2.
42DwarfAccelTable::DwarfAccelTable(DwarfAccelTable::Atom atom) :
43  Header(12),
44  HeaderData(atom) {
45}
46
47// The length of the header data is always going to be 4 + 4 + 4*NumAtoms.
48DwarfAccelTable::DwarfAccelTable(std::vector<DwarfAccelTable::Atom> &atomList) :
49  Header(8 + (atomList.size() * 4)),
50  HeaderData(atomList) {
51}
52
53DwarfAccelTable::~DwarfAccelTable() {
54  for (size_t i = 0, e = Data.size(); i < e; ++i)
55    delete Data[i];
56  for (StringMap<DataArray>::iterator
57         EI = Entries.begin(), EE = Entries.end(); EI != EE; ++EI)
58    for (DataArray::iterator DI = EI->second.begin(),
59           DE = EI->second.end(); DI != DE; ++DI)
60      delete (*DI);
61}
62
63void DwarfAccelTable::AddName(StringRef Name, DIE* die, char Flags) {
64  // If the string is in the list already then add this die to the list
65  // otherwise add a new one.
66  DataArray &DIEs = Entries[Name];
67  DIEs.push_back(new HashDataContents(die, Flags));
68}
69
70void DwarfAccelTable::ComputeBucketCount(void) {
71  // First get the number of unique hashes.
72  std::vector<uint32_t> uniques(Data.size());
73  for (size_t i = 0, e = Data.size(); i < e; ++i)
74    uniques[i] = Data[i]->HashValue;
75  array_pod_sort(uniques.begin(), uniques.end());
76  std::vector<uint32_t>::iterator p =
77    std::unique(uniques.begin(), uniques.end());
78  uint32_t num = std::distance(uniques.begin(), p);
79
80  // Then compute the bucket size, minimum of 1 bucket.
81  if (num > 1024) Header.bucket_count = num/4;
82  if (num > 16) Header.bucket_count = num/2;
83  else Header.bucket_count = num > 0 ? num : 1;
84
85  Header.hashes_count = num;
86}
87
88namespace {
89  // DIESorter - comparison predicate that sorts DIEs by their offset.
90  struct DIESorter {
91    bool operator()(const struct DwarfAccelTable::HashDataContents *A,
92                    const struct DwarfAccelTable::HashDataContents *B) const {
93      return A->Die->getOffset() < B->Die->getOffset();
94    }
95  };
96}
97
98void DwarfAccelTable::FinalizeTable(AsmPrinter *Asm, const char *Prefix) {
99  // Create the individual hash data outputs.
100  for (StringMap<DataArray>::iterator
101         EI = Entries.begin(), EE = Entries.end(); EI != EE; ++EI) {
102    struct HashData *Entry = new HashData((*EI).getKeyData());
103
104    // Unique the entries.
105    std::stable_sort(EI->second.begin(), EI->second.end(), DIESorter());
106    EI->second.erase(std::unique(EI->second.begin(), EI->second.end()),
107                       EI->second.end());
108
109    for (DataArray::const_iterator DI = EI->second.begin(),
110           DE = EI->second.end();
111         DI != DE; ++DI)
112      Entry->addData((*DI));
113    Data.push_back(Entry);
114  }
115
116  // Figure out how many buckets we need, then compute the bucket
117  // contents and the final ordering. We'll emit the hashes and offsets
118  // by doing a walk during the emission phase. We add temporary
119  // symbols to the data so that we can reference them during the offset
120  // later, we'll emit them when we emit the data.
121  ComputeBucketCount();
122
123  // Compute bucket contents and final ordering.
124  Buckets.resize(Header.bucket_count);
125  for (size_t i = 0, e = Data.size(); i < e; ++i) {
126    uint32_t bucket = Data[i]->HashValue % Header.bucket_count;
127    Buckets[bucket].push_back(Data[i]);
128    Data[i]->Sym = Asm->GetTempSymbol(Prefix, i);
129  }
130}
131
132// Emits the header for the table via the AsmPrinter.
133void DwarfAccelTable::EmitHeader(AsmPrinter *Asm) {
134  Asm->OutStreamer.AddComment("Header Magic");
135  Asm->EmitInt32(Header.magic);
136  Asm->OutStreamer.AddComment("Header Version");
137  Asm->EmitInt16(Header.version);
138  Asm->OutStreamer.AddComment("Header Hash Function");
139  Asm->EmitInt16(Header.hash_function);
140  Asm->OutStreamer.AddComment("Header Bucket Count");
141  Asm->EmitInt32(Header.bucket_count);
142  Asm->OutStreamer.AddComment("Header Hash Count");
143  Asm->EmitInt32(Header.hashes_count);
144  Asm->OutStreamer.AddComment("Header Data Length");
145  Asm->EmitInt32(Header.header_data_len);
146  Asm->OutStreamer.AddComment("HeaderData Die Offset Base");
147  Asm->EmitInt32(HeaderData.die_offset_base);
148  Asm->OutStreamer.AddComment("HeaderData Atom Count");
149  Asm->EmitInt32(HeaderData.Atoms.size());
150  for (size_t i = 0; i < HeaderData.Atoms.size(); i++) {
151    Atom A = HeaderData.Atoms[i];
152    Asm->OutStreamer.AddComment(Atom::AtomTypeString(A.type));
153    Asm->EmitInt16(A.type);
154    Asm->OutStreamer.AddComment(dwarf::FormEncodingString(A.form));
155    Asm->EmitInt16(A.form);
156  }
157}
158
159// Walk through and emit the buckets for the table. This will look
160// like a list of numbers of how many elements are in each bucket.
161void DwarfAccelTable::EmitBuckets(AsmPrinter *Asm) {
162  unsigned index = 0;
163  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
164    Asm->OutStreamer.AddComment("Bucket " + Twine(i));
165    if (Buckets[i].size() != 0)
166      Asm->EmitInt32(index);
167    else
168      Asm->EmitInt32(UINT32_MAX);
169    index += Buckets[i].size();
170  }
171}
172
173// Walk through the buckets and emit the individual hashes for each
174// bucket.
175void DwarfAccelTable::EmitHashes(AsmPrinter *Asm) {
176  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
177    for (HashList::const_iterator HI = Buckets[i].begin(),
178           HE = Buckets[i].end(); HI != HE; ++HI) {
179      Asm->OutStreamer.AddComment("Hash in Bucket " + Twine(i));
180      Asm->EmitInt32((*HI)->HashValue);
181    }
182  }
183}
184
185// Walk through the buckets and emit the individual offsets for each
186// element in each bucket. This is done via a symbol subtraction from the
187// beginning of the section. The non-section symbol will be output later
188// when we emit the actual data.
189void DwarfAccelTable::EmitOffsets(AsmPrinter *Asm, MCSymbol *SecBegin) {
190  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
191    for (HashList::const_iterator HI = Buckets[i].begin(),
192           HE = Buckets[i].end(); HI != HE; ++HI) {
193      Asm->OutStreamer.AddComment("Offset in Bucket " + Twine(i));
194      MCContext &Context = Asm->OutStreamer.getContext();
195      const MCExpr *Sub =
196        MCBinaryExpr::CreateSub(MCSymbolRefExpr::Create((*HI)->Sym, Context),
197                                MCSymbolRefExpr::Create(SecBegin, Context),
198                                Context);
199      Asm->OutStreamer.EmitValue(Sub, sizeof(uint32_t), 0);
200    }
201  }
202}
203
204// Walk through the buckets and emit the full data for each element in
205// the bucket. For the string case emit the dies and the various offsets.
206// Terminate each HashData bucket with 0.
207void DwarfAccelTable::EmitData(AsmPrinter *Asm, DwarfDebug *D) {
208  uint64_t PrevHash = UINT64_MAX;
209  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
210    for (HashList::const_iterator HI = Buckets[i].begin(),
211           HE = Buckets[i].end(); HI != HE; ++HI) {
212      // Remember to emit the label for our offset.
213      Asm->OutStreamer.EmitLabel((*HI)->Sym);
214      Asm->OutStreamer.AddComment((*HI)->Str);
215      Asm->EmitSectionOffset(D->getStringPoolEntry((*HI)->Str),
216                             D->getStringPool());
217      Asm->OutStreamer.AddComment("Num DIEs");
218      Asm->EmitInt32((*HI)->Data.size());
219      for (std::vector<struct HashDataContents*>::const_iterator
220             DI = (*HI)->Data.begin(), DE = (*HI)->Data.end();
221           DI != DE; ++DI) {
222        // Emit the DIE offset
223        Asm->EmitInt32((*DI)->Die->getOffset());
224        // If we have multiple Atoms emit that info too.
225        // FIXME: A bit of a hack, we either emit only one atom or all info.
226        if (HeaderData.Atoms.size() > 1) {
227          Asm->EmitInt16((*DI)->Die->getTag());
228          Asm->EmitInt8((*DI)->Flags);
229        }
230      }
231      // Emit a 0 to terminate the data unless we have a hash collision.
232      if (PrevHash != (*HI)->HashValue)
233        Asm->EmitInt32(0);
234      PrevHash = (*HI)->HashValue;
235    }
236  }
237}
238
239// Emit the entire data structure to the output file.
240void DwarfAccelTable::Emit(AsmPrinter *Asm, MCSymbol *SecBegin,
241                           DwarfDebug *D) {
242  // Emit the header.
243  EmitHeader(Asm);
244
245  // Emit the buckets.
246  EmitBuckets(Asm);
247
248  // Emit the hashes.
249  EmitHashes(Asm);
250
251  // Emit the offsets.
252  EmitOffsets(Asm, SecBegin);
253
254  // Emit the hash data.
255  EmitData(Asm, D);
256}
257
258#ifndef NDEBUG
259void DwarfAccelTable::print(raw_ostream &O) {
260
261  Header.print(O);
262  HeaderData.print(O);
263
264  O << "Entries: \n";
265  for (StringMap<DataArray>::const_iterator
266         EI = Entries.begin(), EE = Entries.end(); EI != EE; ++EI) {
267    O << "Name: " << EI->getKeyData() << "\n";
268    for (DataArray::const_iterator DI = EI->second.begin(),
269           DE = EI->second.end();
270         DI != DE; ++DI)
271      (*DI)->print(O);
272  }
273
274  O << "Buckets and Hashes: \n";
275  for (size_t i = 0, e = Buckets.size(); i < e; ++i)
276    for (HashList::const_iterator HI = Buckets[i].begin(),
277           HE = Buckets[i].end(); HI != HE; ++HI)
278      (*HI)->print(O);
279
280  O << "Data: \n";
281    for (std::vector<HashData*>::const_iterator
282           DI = Data.begin(), DE = Data.end(); DI != DE; ++DI)
283      (*DI)->print(O);
284
285
286}
287#endif
288