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 "DwarfCompileUnit.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/CodeGen/DIE.h"
21 #include "llvm/MC/MCExpr.h"
22 #include "llvm/MC/MCStreamer.h"
23 #include "llvm/MC/MCSymbol.h"
24 #include "llvm/Support/Debug.h"
25
26 using namespace llvm;
27
28 // The length of the header data is always going to be 4 + 4 + 4*NumAtoms.
DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom> atomList)29 DwarfAccelTable::DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom> atomList)
30 : Header(8 + (atomList.size() * 4)), HeaderData(atomList),
31 Entries(Allocator) {}
32
AddName(DwarfStringPoolEntryRef Name,const DIE * die,char Flags)33 void DwarfAccelTable::AddName(DwarfStringPoolEntryRef Name, const DIE *die,
34 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.getString()];
39 assert(!DIEs.Name || DIEs.Name == Name);
40 DIEs.Name = Name;
41 DIEs.Values.push_back(new (Allocator) HashDataContents(die, Flags));
42 }
43
ComputeBucketCount()44 void DwarfAccelTable::ComputeBucketCount() {
45 // First get the number of unique hashes.
46 std::vector<uint32_t> uniques(Data.size());
47 for (size_t i = 0, e = Data.size(); i < e; ++i)
48 uniques[i] = Data[i]->HashValue;
49 array_pod_sort(uniques.begin(), uniques.end());
50 std::vector<uint32_t>::iterator p =
51 std::unique(uniques.begin(), uniques.end());
52 uint32_t num = std::distance(uniques.begin(), p);
53
54 // Then compute the bucket size, minimum of 1 bucket.
55 if (num > 1024)
56 Header.bucket_count = num / 4;
57 else if (num > 16)
58 Header.bucket_count = num / 2;
59 else
60 Header.bucket_count = num > 0 ? num : 1;
61
62 Header.hashes_count = num;
63 }
64
65 // compareDIEs - comparison predicate that sorts DIEs by their offset.
compareDIEs(const DwarfAccelTable::HashDataContents * A,const DwarfAccelTable::HashDataContents * B)66 static bool compareDIEs(const DwarfAccelTable::HashDataContents *A,
67 const DwarfAccelTable::HashDataContents *B) {
68 return A->Die->getOffset() < B->Die->getOffset();
69 }
70
FinalizeTable(AsmPrinter * Asm,StringRef Prefix)71 void DwarfAccelTable::FinalizeTable(AsmPrinter *Asm, StringRef Prefix) {
72 // Create the individual hash data outputs.
73 Data.reserve(Entries.size());
74 for (StringMap<DataArray>::iterator EI = Entries.begin(), EE = Entries.end();
75 EI != EE; ++EI) {
76
77 // Unique the entries.
78 std::stable_sort(EI->second.Values.begin(), EI->second.Values.end(), compareDIEs);
79 EI->second.Values.erase(
80 std::unique(EI->second.Values.begin(), EI->second.Values.end()),
81 EI->second.Values.end());
82
83 HashData *Entry = new (Allocator) HashData(EI->getKey(), EI->second);
84 Data.push_back(Entry);
85 }
86
87 // Figure out how many buckets we need, then compute the bucket
88 // contents and the final ordering. We'll emit the hashes and offsets
89 // by doing a walk during the emission phase. We add temporary
90 // symbols to the data so that we can reference them during the offset
91 // later, we'll emit them when we emit the data.
92 ComputeBucketCount();
93
94 // Compute bucket contents and final ordering.
95 Buckets.resize(Header.bucket_count);
96 for (size_t i = 0, e = Data.size(); i < e; ++i) {
97 uint32_t bucket = Data[i]->HashValue % Header.bucket_count;
98 Buckets[bucket].push_back(Data[i]);
99 Data[i]->Sym = Asm->createTempSymbol(Prefix);
100 }
101
102 // Sort the contents of the buckets by hash value so that hash
103 // collisions end up together. Stable sort makes testing easier and
104 // doesn't cost much more.
105 for (size_t i = 0; i < Buckets.size(); ++i)
106 std::stable_sort(Buckets[i].begin(), Buckets[i].end(),
107 [] (HashData *LHS, HashData *RHS) {
108 return LHS->HashValue < RHS->HashValue;
109 });
110 }
111
112 // Emits the header for the table via the AsmPrinter.
EmitHeader(AsmPrinter * Asm)113 void DwarfAccelTable::EmitHeader(AsmPrinter *Asm) {
114 Asm->OutStreamer->AddComment("Header Magic");
115 Asm->EmitInt32(Header.magic);
116 Asm->OutStreamer->AddComment("Header Version");
117 Asm->EmitInt16(Header.version);
118 Asm->OutStreamer->AddComment("Header Hash Function");
119 Asm->EmitInt16(Header.hash_function);
120 Asm->OutStreamer->AddComment("Header Bucket Count");
121 Asm->EmitInt32(Header.bucket_count);
122 Asm->OutStreamer->AddComment("Header Hash Count");
123 Asm->EmitInt32(Header.hashes_count);
124 Asm->OutStreamer->AddComment("Header Data Length");
125 Asm->EmitInt32(Header.header_data_len);
126 Asm->OutStreamer->AddComment("HeaderData Die Offset Base");
127 Asm->EmitInt32(HeaderData.die_offset_base);
128 Asm->OutStreamer->AddComment("HeaderData Atom Count");
129 Asm->EmitInt32(HeaderData.Atoms.size());
130 for (size_t i = 0; i < HeaderData.Atoms.size(); i++) {
131 Atom A = HeaderData.Atoms[i];
132 Asm->OutStreamer->AddComment(dwarf::AtomTypeString(A.type));
133 Asm->EmitInt16(A.type);
134 Asm->OutStreamer->AddComment(dwarf::FormEncodingString(A.form));
135 Asm->EmitInt16(A.form);
136 }
137 }
138
139 // Walk through and emit the buckets for the table. Each index is
140 // an offset into the list of hashes.
EmitBuckets(AsmPrinter * Asm)141 void DwarfAccelTable::EmitBuckets(AsmPrinter *Asm) {
142 unsigned index = 0;
143 for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
144 Asm->OutStreamer->AddComment("Bucket " + Twine(i));
145 if (Buckets[i].size() != 0)
146 Asm->EmitInt32(index);
147 else
148 Asm->EmitInt32(UINT32_MAX);
149 // Buckets point in the list of hashes, not to the data. Do not
150 // increment the index multiple times in case of hash collisions.
151 uint64_t PrevHash = UINT64_MAX;
152 for (auto *HD : Buckets[i]) {
153 uint32_t HashValue = HD->HashValue;
154 if (PrevHash != HashValue)
155 ++index;
156 PrevHash = HashValue;
157 }
158 }
159 }
160
161 // Walk through the buckets and emit the individual hashes for each
162 // bucket.
EmitHashes(AsmPrinter * Asm)163 void DwarfAccelTable::EmitHashes(AsmPrinter *Asm) {
164 uint64_t PrevHash = UINT64_MAX;
165 for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
166 for (HashList::const_iterator HI = Buckets[i].begin(),
167 HE = Buckets[i].end();
168 HI != HE; ++HI) {
169 uint32_t HashValue = (*HI)->HashValue;
170 if (PrevHash == HashValue)
171 continue;
172 Asm->OutStreamer->AddComment("Hash in Bucket " + Twine(i));
173 Asm->EmitInt32(HashValue);
174 PrevHash = 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.
emitOffsets(AsmPrinter * Asm,const MCSymbol * SecBegin)183 void DwarfAccelTable::emitOffsets(AsmPrinter *Asm, const MCSymbol *SecBegin) {
184 uint64_t PrevHash = UINT64_MAX;
185 for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
186 for (HashList::const_iterator HI = Buckets[i].begin(),
187 HE = Buckets[i].end();
188 HI != HE; ++HI) {
189 uint32_t HashValue = (*HI)->HashValue;
190 if (PrevHash == HashValue)
191 continue;
192 PrevHash = HashValue;
193 Asm->OutStreamer->AddComment("Offset in Bucket " + Twine(i));
194 MCContext &Context = Asm->OutStreamer->getContext();
195 const MCExpr *Sub = MCBinaryExpr::createSub(
196 MCSymbolRefExpr::create((*HI)->Sym, Context),
197 MCSymbolRefExpr::create(SecBegin, Context), Context);
198 Asm->OutStreamer->EmitValue(Sub, sizeof(uint32_t));
199 }
200 }
201 }
202
203 // Walk through the buckets and emit the full data for each element in
204 // the bucket. For the string case emit the dies and the various offsets.
205 // Terminate each HashData bucket with 0.
EmitData(AsmPrinter * Asm,DwarfDebug * D)206 void DwarfAccelTable::EmitData(AsmPrinter *Asm, DwarfDebug *D) {
207 for (size_t i = 0, e = Buckets.size(); i < e; ++i) {
208 uint64_t PrevHash = UINT64_MAX;
209 for (HashList::const_iterator HI = Buckets[i].begin(),
210 HE = Buckets[i].end();
211 HI != HE; ++HI) {
212 // Terminate the previous entry if there is no hash collision
213 // with the current one.
214 if (PrevHash != UINT64_MAX && PrevHash != (*HI)->HashValue)
215 Asm->EmitInt32(0);
216 // Remember to emit the label for our offset.
217 Asm->OutStreamer->EmitLabel((*HI)->Sym);
218 Asm->OutStreamer->AddComment((*HI)->Str);
219 Asm->emitDwarfStringOffset((*HI)->Data.Name);
220 Asm->OutStreamer->AddComment("Num DIEs");
221 Asm->EmitInt32((*HI)->Data.Values.size());
222 for (HashDataContents *HD : (*HI)->Data.Values) {
223 // Emit the DIE offset
224 DwarfCompileUnit *CU = D->lookupUnit(HD->Die->getUnit());
225 assert(CU && "Accelerated DIE should belong to a CU.");
226 Asm->EmitInt32(HD->Die->getOffset() + CU->getDebugInfoOffset());
227 // If we have multiple Atoms emit that info too.
228 // FIXME: A bit of a hack, we either emit only one atom or all info.
229 if (HeaderData.Atoms.size() > 1) {
230 Asm->EmitInt16(HD->Die->getTag());
231 Asm->EmitInt8(HD->Flags);
232 }
233 }
234 PrevHash = (*HI)->HashValue;
235 }
236 // Emit the final end marker for the bucket.
237 if (!Buckets[i].empty())
238 Asm->EmitInt32(0);
239 }
240 }
241
242 // Emit the entire data structure to the output file.
emit(AsmPrinter * Asm,const MCSymbol * SecBegin,DwarfDebug * D)243 void DwarfAccelTable::emit(AsmPrinter *Asm, const MCSymbol *SecBegin,
244 DwarfDebug *D) {
245 // Emit the header.
246 EmitHeader(Asm);
247
248 // Emit the buckets.
249 EmitBuckets(Asm);
250
251 // Emit the hashes.
252 EmitHashes(Asm);
253
254 // Emit the offsets.
255 emitOffsets(Asm, SecBegin);
256
257 // Emit the hash data.
258 EmitData(Asm, D);
259 }
260
261 #ifndef NDEBUG
print(raw_ostream & O)262 void DwarfAccelTable::print(raw_ostream &O) {
263
264 Header.print(O);
265 HeaderData.print(O);
266
267 O << "Entries: \n";
268 for (StringMap<DataArray>::const_iterator EI = Entries.begin(),
269 EE = Entries.end();
270 EI != EE; ++EI) {
271 O << "Name: " << EI->getKeyData() << "\n";
272 for (HashDataContents *HD : EI->second.Values)
273 HD->print(O);
274 }
275
276 O << "Buckets and Hashes: \n";
277 for (size_t i = 0, e = Buckets.size(); i < e; ++i)
278 for (HashList::const_iterator HI = Buckets[i].begin(),
279 HE = Buckets[i].end();
280 HI != HE; ++HI)
281 (*HI)->print(O);
282
283 O << "Data: \n";
284 for (std::vector<HashData *>::const_iterator DI = Data.begin(),
285 DE = Data.end();
286 DI != DE; ++DI)
287 (*DI)->print(O);
288 }
289 #endif
290