DwarfAccelTable.cpp revision c36145f19c1e164f7d630b813e9970600d8f2976
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 "llvm/CodeGen/AsmPrinter.h"
15#include "llvm/MC/MCExpr.h"
16#include "llvm/MC/MCStreamer.h"
17#include "llvm/MC/MCSymbol.h"
18#include "llvm/Support/Debug.h"
19#include "DwarfAccelTable.h"
20#include "DwarfDebug.h"
21#include "DIE.h"
22
23using namespace llvm;
24
25const char *DwarfAccelTable::Atom::AtomTypeString(enum AtomType AT) {
26  switch (AT) {
27  default: llvm_unreachable("invalid AtomType!");
28  case eAtomTypeNULL: return "eAtomTypeNULL";
29  case eAtomTypeDIEOffset: return "eAtomTypeDIEOffset";
30  case eAtomTypeCUOffset: return "eAtomTypeCUOffset";
31  case eAtomTypeTag: return "eAtomTypeTag";
32  case eAtomTypeNameFlags: return "eAtomTypeNameFlags";
33  case eAtomTypeTypeFlags: return "eAtomTypeTypeFlags";
34  }
35}
36
37// The general case would need to have a less hard coded size for the
38// length of the HeaderData, however, if we're constructing based on a
39// single Atom then we know it will always be: 4 + 4 + 2 + 2.
40DwarfAccelTable::DwarfAccelTable(DwarfAccelTable::Atom atom) :
41  Header(12),
42  HeaderData(atom) {
43}
44
45// The length of the header data is always going to be 4 + 4 + 4*NumAtoms.
46DwarfAccelTable::DwarfAccelTable(std::vector<DwarfAccelTable::Atom> &atomList) :
47  Header(8 + (atomList.size() * 4)),
48  HeaderData(atomList) {
49}
50
51DwarfAccelTable::~DwarfAccelTable() {
52  for (size_t i = 0, e = Data.size(); i < e; ++i)
53    delete Data[i];
54}
55
56void DwarfAccelTable::AddName(StringRef Name, DIE* die, char Flags) {
57  // If the string is in the list already then add this die to the list
58  // otherwise add a new one.
59  DataArray &DIEs = Entries[Name];
60  DIEs.push_back(new HashDataContents(die, Flags));
61}
62
63void DwarfAccelTable::ComputeBucketCount(void) {
64  // First get the number of unique hashes.
65  std::vector<uint32_t> uniques;
66  uniques.resize(Data.size());
67  for (size_t i = 0, e = Data.size(); i < e; ++i)
68    uniques[i] = Data[i]->HashValue;
69  std::stable_sort(uniques.begin(), uniques.end());
70  std::vector<uint32_t>::iterator p =
71    std::unique(uniques.begin(), uniques.end());
72  uint32_t num = std::distance(uniques.begin(), p);
73
74  // Then compute the bucket size, minimum of 1 bucket.
75  if (num > 1024) Header.bucket_count = num/4;
76  if (num > 16) Header.bucket_count = num/2;
77  else Header.bucket_count = num > 0 ? num : 1;
78
79  Header.hashes_count = num;
80}
81
82namespace {
83  // DIESorter - comparison predicate that sorts DIEs by their offset.
84  struct DIESorter {
85    bool operator()(const struct DwarfAccelTable::HashDataContents *A,
86                    const struct DwarfAccelTable::HashDataContents *B) const {
87      return A->Die->getOffset() < B->Die->getOffset();
88    }
89  };
90}
91
92void DwarfAccelTable::FinalizeTable(AsmPrinter *Asm, const char *Prefix) {
93  // Create the individual hash data outputs.
94  for (StringMap<DataArray>::iterator
95         EI = Entries.begin(), EE = Entries.end(); EI != EE; ++EI) {
96    struct HashData *Entry = new HashData((*EI).getKeyData());
97
98    // Unique the entries.
99    std::stable_sort((*EI).second.begin(), (*EI).second.end(), DIESorter());
100    (*EI).second.erase(std::unique((*EI).second.begin(), (*EI).second.end()),
101                       (*EI).second.end());
102
103    for (DataArray::const_iterator DI = (*EI).second.begin(),
104           DE = (*EI).second.end();
105         DI != DE; ++DI)
106      Entry->addData((*DI));
107    Data.push_back(Entry);
108  }
109
110  // Figure out how many buckets we need, then compute the bucket
111  // contents and the final ordering. We'll emit the hashes and offsets
112  // by doing a walk during the emission phase. We add temporary
113  // symbols to the data so that we can reference them during the offset
114  // later, we'll emit them when we emit the data.
115  ComputeBucketCount();
116
117  // Compute bucket contents and final ordering.
118  Buckets.resize(Header.bucket_count);
119  for (size_t i = 0, e = Data.size(); i < e; ++i) {
120    uint32_t bucket = Data[i]->HashValue % Header.bucket_count;
121    Buckets[bucket].push_back(Data[i]);
122    Data[i]->Sym = Asm->GetTempSymbol(Prefix, i);
123  }
124}
125
126// Emits the header for the table via the AsmPrinter.
127void DwarfAccelTable::EmitHeader(AsmPrinter *Asm) {
128  Asm->OutStreamer.AddComment("Header Magic");
129  Asm->EmitInt32(Header.magic);
130  Asm->OutStreamer.AddComment("Header Version");
131  Asm->EmitInt16(Header.version);
132  Asm->OutStreamer.AddComment("Header Hash Function");
133  Asm->EmitInt16(Header.hash_function);
134  Asm->OutStreamer.AddComment("Header Bucket Count");
135  Asm->EmitInt32(Header.bucket_count);
136  Asm->OutStreamer.AddComment("Header Hash Count");
137  Asm->EmitInt32(Header.hashes_count);
138  Asm->OutStreamer.AddComment("Header Data Length");
139  Asm->EmitInt32(Header.header_data_len);
140  Asm->OutStreamer.AddComment("HeaderData Die Offset Base");
141  Asm->EmitInt32(HeaderData.die_offset_base);
142  Asm->OutStreamer.AddComment("HeaderData Atom Count");
143  Asm->EmitInt32(HeaderData.Atoms.size());
144  for (size_t i = 0; i < HeaderData.Atoms.size(); i++) {
145    Atom A = HeaderData.Atoms[i];
146    Asm->OutStreamer.AddComment(Atom::AtomTypeString(A.type));
147    Asm->EmitInt16(A.type);
148    Asm->OutStreamer.AddComment(dwarf::FormEncodingString(A.form));
149    Asm->EmitInt16(A.form);
150  }
151}
152
153// Walk through and emit the buckets for the table. This will look
154// like a list of numbers of how many elements are in each bucket.
155void DwarfAccelTable::EmitBuckets(AsmPrinter *Asm) {
156  unsigned index = 0;
157  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
158    Asm->OutStreamer.AddComment("Bucket " + Twine(i));
159    if (Buckets[i].size() != 0)
160      Asm->EmitInt32(index);
161    else
162      Asm->EmitInt32(UINT32_MAX);
163    index += Buckets[i].size();
164  }
165}
166
167// Walk through the buckets and emit the individual hashes for each
168// bucket.
169void DwarfAccelTable::EmitHashes(AsmPrinter *Asm) {
170  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
171    for (HashList::const_iterator HI = Buckets[i].begin(),
172           HE = Buckets[i].end(); HI != HE; ++HI) {
173      Asm->OutStreamer.AddComment("Hash in Bucket " + Twine(i));
174      Asm->EmitInt32((*HI)->HashValue);
175    }
176  }
177}
178
179// Walk through the buckets and emit the individual offsets for each
180// element in each bucket. This is done via a symbol subtraction from the
181// beginning of the section. The non-section symbol will be output later
182// when we emit the actual data.
183void DwarfAccelTable::EmitOffsets(AsmPrinter *Asm, MCSymbol *SecBegin) {
184  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
185    for (HashList::const_iterator HI = Buckets[i].begin(),
186           HE = Buckets[i].end(); HI != HE; ++HI) {
187      Asm->OutStreamer.AddComment("Offset in Bucket " + Twine(i));
188      MCContext &Context = Asm->OutStreamer.getContext();
189      const MCExpr *Sub =
190        MCBinaryExpr::CreateSub(MCSymbolRefExpr::Create((*HI)->Sym, Context),
191                                MCSymbolRefExpr::Create(SecBegin, Context),
192                                Context);
193      Asm->OutStreamer.EmitValue(Sub, sizeof(uint32_t), 0);
194    }
195  }
196}
197
198// Walk through the buckets and emit the full data for each element in
199// the bucket. For the string case emit the dies and the various offsets.
200// Terminate each HashData bucket with 0.
201void DwarfAccelTable::EmitData(AsmPrinter *Asm, DwarfDebug *D) {
202  uint64_t PrevHash = UINT64_MAX;
203  for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
204    for (HashList::const_iterator HI = Buckets[i].begin(),
205           HE = Buckets[i].end(); HI != HE; ++HI) {
206      // Remember to emit the label for our offset.
207      Asm->OutStreamer.EmitLabel((*HI)->Sym);
208      Asm->OutStreamer.AddComment((*HI)->Str);
209      Asm->EmitSectionOffset(D->getStringPoolEntry((*HI)->Str),
210                             D->getStringPool());
211      Asm->OutStreamer.AddComment("Num DIEs");
212      Asm->EmitInt32((*HI)->Data.size());
213      for (std::vector<struct HashDataContents*>::const_iterator
214             DI = (*HI)->Data.begin(), DE = (*HI)->Data.end();
215           DI != DE; ++DI) {
216        // Emit the DIE offset
217        Asm->EmitInt32((*DI)->Die->getOffset());
218        // If we have multiple Atoms emit that info too.
219        // FIXME: A bit of a hack, we either emit only one atom or all info.
220        if (HeaderData.Atoms.size() > 1) {
221          Asm->EmitInt16((*DI)->Die->getTag());
222          Asm->EmitInt8((*DI)->Flags);
223        }
224      }
225      // Emit a 0 to terminate the data unless we have a hash collision.
226      if (PrevHash != (*HI)->HashValue)
227        Asm->EmitInt32(0);
228      PrevHash = (*HI)->HashValue;
229    }
230  }
231}
232
233// Emit the entire data structure to the output file.
234void DwarfAccelTable::Emit(AsmPrinter *Asm, MCSymbol *SecBegin,
235                           DwarfDebug *D) {
236  // Emit the header.
237  EmitHeader(Asm);
238
239  // Emit the buckets.
240  EmitBuckets(Asm);
241
242  // Emit the hashes.
243  EmitHashes(Asm);
244
245  // Emit the offsets.
246  EmitOffsets(Asm, SecBegin);
247
248  // Emit the hash data.
249  EmitData(Asm, D);
250}
251
252#ifndef NDEBUG
253void DwarfAccelTable::print(raw_ostream &O) {
254
255  Header.print(O);
256  HeaderData.print(O);
257
258  O << "Entries: \n";
259  for (StringMap<DataArray>::const_iterator
260         EI = Entries.begin(), EE = Entries.end(); EI != EE; ++EI) {
261    O << "Name: " << (*EI).getKeyData() << "\n";
262    for (DataArray::const_iterator DI = (*EI).second.begin(),
263           DE = (*EI).second.end();
264         DI != DE; ++DI)
265      (*DI)->print(O);
266  }
267
268  O << "Buckets and Hashes: \n";
269  for (size_t i = 0, e = Buckets.size(); i < e; ++i)
270    for (HashList::const_iterator HI = Buckets[i].begin(),
271           HE = Buckets[i].end(); HI != HE; ++HI)
272      (*HI)->print(O);
273
274  O << "Data: \n";
275    for (std::vector<HashData*>::const_iterator
276           DI = Data.begin(), DE = Data.end(); DI != DE; ++DI)
277      (*DI)->print(O);
278
279
280}
281#endif
282