DwarfAccelTable.cpp revision 36b56886974eae4f9c5ebc96befd3e7bfe5de338 
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 "DIE.h"
16#include "DwarfDebug.h"
17#include "llvm/ADT/STLExtras.h"
18#include "llvm/ADT/Twine.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
27// The length of the header data is always going to be 4 + 4 + 4*NumAtoms.
28DwarfAccelTable::DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom> atomList)
29    : Header(8 + (atomList.size() * 4)), HeaderData(atomList),
30      Entries(Allocator) {}
31
32DwarfAccelTable::~DwarfAccelTable() {}
33
34void DwarfAccelTable::AddName(StringRef Name, const DIE *die, char Flags) {
35  assert(Data.empty() && "Already finalized!");
36  // If the string is in the list already then add this die to the list
37  // otherwise add a new one.
38  DataArray &DIEs = Entries[Name];
39  DIEs.push_back(new (Allocator) HashDataContents(die, Flags));
40}
41
42void DwarfAccelTable::ComputeBucketCount(void) {
43  // First get the number of unique hashes.
44  std::vector<uint32_t> uniques(Data.size());
45  for (size_t i = 0, e = Data.size(); i < e; ++i)
46    uniques[i] = Data[i]->HashValue;
47  array_pod_sort(uniques.begin(), uniques.end());
48  std::vector<uint32_t>::iterator p =
49      std::unique(uniques.begin(), uniques.end());
50  uint32_t num = std::distance(uniques.begin(), p);
51
52  // Then compute the bucket size, minimum of 1 bucket.
53  if (num > 1024)
54    Header.bucket_count = num / 4;
55  if (num > 16)
56    Header.bucket_count = num / 2;
57  else
58    Header.bucket_count = num > 0 ? num : 1;
59
60  Header.hashes_count = num;
61}
62
63// compareDIEs - comparison predicate that sorts DIEs by their offset.
64static bool compareDIEs(const DwarfAccelTable::HashDataContents *A,
65                        const DwarfAccelTable::HashDataContents *B) {
66  return A->Die->getOffset() < B->Die->getOffset();
67}
68
69void DwarfAccelTable::FinalizeTable(AsmPrinter *Asm, StringRef Prefix) {
70  // Create the individual hash data outputs.
71  for (StringMap<DataArray>::iterator EI = Entries.begin(), EE = Entries.end();
72       EI != EE; ++EI) {
73
74    // Unique the entries.
75    std::stable_sort(EI->second.begin(), EI->second.end(), compareDIEs);
76    EI->second.erase(std::unique(EI->second.begin(), EI->second.end()),
77                     EI->second.end());
78
79    HashData *Entry = new (Allocator) HashData(EI->getKey(), EI->second);
80    Data.push_back(Entry);
81  }
82
83  // Figure out how many buckets we need, then compute the bucket
84  // contents and the final ordering. We'll emit the hashes and offsets
85  // by doing a walk during the emission phase. We add temporary
86  // symbols to the data so that we can reference them during the offset
87  // later, we'll emit them when we emit the data.
88  ComputeBucketCount();
89
90  // Compute bucket contents and final ordering.
91  Buckets.resize(Header.bucket_count);
92  for (size_t i = 0, e = Data.size(); i < e; ++i) {
93    uint32_t bucket = Data[i]->HashValue % Header.bucket_count;
94    Buckets[bucket].push_back(Data[i]);
95    Data[i]->Sym = Asm->GetTempSymbol(Prefix, i);
96  }
97}
98
99// Emits the header for the table via the AsmPrinter.
100void DwarfAccelTable::EmitHeader(AsmPrinter *Asm) {
101  Asm->OutStreamer.AddComment("Header Magic");
102  Asm->EmitInt32(Header.magic);
103  Asm->OutStreamer.AddComment("Header Version");
104  Asm->EmitInt16(Header.version);
105  Asm->OutStreamer.AddComment("Header Hash Function");
106  Asm->EmitInt16(Header.hash_function);
107  Asm->OutStreamer.AddComment("Header Bucket Count");
108  Asm->EmitInt32(Header.bucket_count);
109  Asm->OutStreamer.AddComment("Header Hash Count");
110  Asm->EmitInt32(Header.hashes_count);
111  Asm->OutStreamer.AddComment("Header Data Length");
112  Asm->EmitInt32(Header.header_data_len);
113  Asm->OutStreamer.AddComment("HeaderData Die Offset Base");
114  Asm->EmitInt32(HeaderData.die_offset_base);
115  Asm->OutStreamer.AddComment("HeaderData Atom Count");
116  Asm->EmitInt32(HeaderData.Atoms.size());
117  for (size_t i = 0; i < HeaderData.Atoms.size(); i++) {
118    Atom A = HeaderData.Atoms[i];
119    Asm->OutStreamer.AddComment(dwarf::AtomTypeString(A.type));
120    Asm->EmitInt16(A.type);
121    Asm->OutStreamer.AddComment(dwarf::FormEncodingString(A.form));
122    Asm->EmitInt16(A.form);
123  }
124}
125
126// Walk through and emit the buckets for the table. Each index is
127// an offset into the list of hashes.
128void DwarfAccelTable::EmitBuckets(AsmPrinter *Asm) {
129  unsigned index = 0;
130  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
131    Asm->OutStreamer.AddComment("Bucket " + Twine(i));
132    if (Buckets[i].size() != 0)
133      Asm->EmitInt32(index);
134    else
135      Asm->EmitInt32(UINT32_MAX);
136    index += Buckets[i].size();
137  }
138}
139
140// Walk through the buckets and emit the individual hashes for each
141// bucket.
142void DwarfAccelTable::EmitHashes(AsmPrinter *Asm) {
143  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
144    for (HashList::const_iterator HI = Buckets[i].begin(),
145                                  HE = Buckets[i].end();
146         HI != HE; ++HI) {
147      Asm->OutStreamer.AddComment("Hash in Bucket " + Twine(i));
148      Asm->EmitInt32((*HI)->HashValue);
149    }
150  }
151}
152
153// Walk through the buckets and emit the individual offsets for each
154// element in each bucket. This is done via a symbol subtraction from the
155// beginning of the section. The non-section symbol will be output later
156// when we emit the actual data.
157void DwarfAccelTable::EmitOffsets(AsmPrinter *Asm, MCSymbol *SecBegin) {
158  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
159    for (HashList::const_iterator HI = Buckets[i].begin(),
160                                  HE = Buckets[i].end();
161         HI != HE; ++HI) {
162      Asm->OutStreamer.AddComment("Offset in Bucket " + Twine(i));
163      MCContext &Context = Asm->OutStreamer.getContext();
164      const MCExpr *Sub = MCBinaryExpr::CreateSub(
165          MCSymbolRefExpr::Create((*HI)->Sym, Context),
166          MCSymbolRefExpr::Create(SecBegin, Context), Context);
167      Asm->OutStreamer.EmitValue(Sub, sizeof(uint32_t));
168    }
169  }
170}
171
172// Walk through the buckets and emit the full data for each element in
173// the bucket. For the string case emit the dies and the various offsets.
174// Terminate each HashData bucket with 0.
175void DwarfAccelTable::EmitData(AsmPrinter *Asm, DwarfFile *D) {
176  uint64_t PrevHash = UINT64_MAX;
177  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
178    for (HashList::const_iterator HI = Buckets[i].begin(),
179                                  HE = Buckets[i].end();
180         HI != HE; ++HI) {
181      // Remember to emit the label for our offset.
182      Asm->OutStreamer.EmitLabel((*HI)->Sym);
183      Asm->OutStreamer.AddComment((*HI)->Str);
184      Asm->EmitSectionOffset(D->getStringPoolEntry((*HI)->Str),
185                             D->getStringPoolSym());
186      Asm->OutStreamer.AddComment("Num DIEs");
187      Asm->EmitInt32((*HI)->Data.size());
188      for (ArrayRef<HashDataContents *>::const_iterator
189               DI = (*HI)->Data.begin(),
190               DE = (*HI)->Data.end();
191           DI != DE; ++DI) {
192        // Emit the DIE offset
193        Asm->EmitInt32((*DI)->Die->getOffset());
194        // If we have multiple Atoms emit that info too.
195        // FIXME: A bit of a hack, we either emit only one atom or all info.
196        if (HeaderData.Atoms.size() > 1) {
197          Asm->EmitInt16((*DI)->Die->getTag());
198          Asm->EmitInt8((*DI)->Flags);
199        }
200      }
201      // Emit a 0 to terminate the data unless we have a hash collision.
202      if (PrevHash != (*HI)->HashValue)
203        Asm->EmitInt32(0);
204      PrevHash = (*HI)->HashValue;
205    }
206  }
207}
208
209// Emit the entire data structure to the output file.
210void DwarfAccelTable::Emit(AsmPrinter *Asm, MCSymbol *SecBegin, DwarfFile *D) {
211  // Emit the header.
212  EmitHeader(Asm);
213
214  // Emit the buckets.
215  EmitBuckets(Asm);
216
217  // Emit the hashes.
218  EmitHashes(Asm);
219
220  // Emit the offsets.
221  EmitOffsets(Asm, SecBegin);
222
223  // Emit the hash data.
224  EmitData(Asm, D);
225}
226
227#ifndef NDEBUG
228void DwarfAccelTable::print(raw_ostream &O) {
229
230  Header.print(O);
231  HeaderData.print(O);
232
233  O << "Entries: \n";
234  for (StringMap<DataArray>::const_iterator EI = Entries.begin(),
235                                            EE = Entries.end();
236       EI != EE; ++EI) {
237    O << "Name: " << EI->getKeyData() << "\n";
238    for (DataArray::const_iterator DI = EI->second.begin(),
239                                   DE = EI->second.end();
240         DI != DE; ++DI)
241      (*DI)->print(O);
242  }
243
244  O << "Buckets and Hashes: \n";
245  for (size_t i = 0, e = Buckets.size(); i < e; ++i)
246    for (HashList::const_iterator HI = Buckets[i].begin(),
247                                  HE = Buckets[i].end();
248         HI != HE; ++HI)
249      (*HI)->print(O);
250
251  O << "Data: \n";
252  for (std::vector<HashData *>::const_iterator DI = Data.begin(),
253                                               DE = Data.end();
254       DI != DE; ++DI)
255    (*DI)->print(O);
256}
257#endif
258