1 //===- ICF.cpp ------------------------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // ICF is short for Identical Code Folding. That is a size optimization to 10 // identify and merge two or more read-only sections (typically functions) 11 // that happened to have the same contents. It usually reduces output size 12 // by a few percent. 13 // 14 // On Windows, ICF is enabled by default. 15 // 16 // See ELF/ICF.cpp for the details about the algorithm. 17 // 18 //===----------------------------------------------------------------------===// 19 20 #include "ICF.h" 21 #include "COFFLinkerContext.h" 22 #include "Chunks.h" 23 #include "Symbols.h" 24 #include "lld/Common/ErrorHandler.h" 25 #include "lld/Common/Timer.h" 26 #include "llvm/ADT/Hashing.h" 27 #include "llvm/Support/Debug.h" 28 #include "llvm/Support/Parallel.h" 29 #include "llvm/Support/raw_ostream.h" 30 #include "llvm/Support/xxhash.h" 31 #include <algorithm> 32 #include <atomic> 33 #include <vector> 34 35 using namespace llvm; 36 37 namespace lld { 38 namespace coff { 39 40 class ICF { 41 public: 42 ICF(COFFLinkerContext &c, ICFLevel icfLevel) : icfLevel(icfLevel), ctx(c){}; 43 void run(); 44 45 private: 46 void segregate(size_t begin, size_t end, bool constant); 47 48 bool assocEquals(const SectionChunk *a, const SectionChunk *b); 49 50 bool equalsConstant(const SectionChunk *a, const SectionChunk *b); 51 bool equalsVariable(const SectionChunk *a, const SectionChunk *b); 52 53 bool isEligible(SectionChunk *c); 54 55 size_t findBoundary(size_t begin, size_t end); 56 57 void forEachClassRange(size_t begin, size_t end, 58 std::function<void(size_t, size_t)> fn); 59 60 void forEachClass(std::function<void(size_t, size_t)> fn); 61 62 std::vector<SectionChunk *> chunks; 63 int cnt = 0; 64 std::atomic<bool> repeat = {false}; 65 ICFLevel icfLevel = ICFLevel::All; 66 67 COFFLinkerContext &ctx; 68 }; 69 70 // Returns true if section S is subject of ICF. 71 // 72 // Microsoft's documentation 73 // (https://msdn.microsoft.com/en-us/library/bxwfs976.aspx; visited April 74 // 2017) says that /opt:icf folds both functions and read-only data. 75 // Despite that, the MSVC linker folds only functions. We found 76 // a few instances of programs that are not safe for data merging. 77 // Therefore, we merge only functions just like the MSVC tool. However, we also 78 // merge read-only sections in a couple of cases where the address of the 79 // section is insignificant to the user program and the behaviour matches that 80 // of the Visual C++ linker. 81 bool ICF::isEligible(SectionChunk *c) { 82 // Non-comdat chunks, dead chunks, and writable chunks are not eligible. 83 bool writable = c->getOutputCharacteristics() & llvm::COFF::IMAGE_SCN_MEM_WRITE; 84 if (!c->isCOMDAT() || !c->live || writable) 85 return false; 86 87 // Under regular (not safe) ICF, all code sections are eligible. 88 if ((icfLevel == ICFLevel::All) && 89 c->getOutputCharacteristics() & llvm::COFF::IMAGE_SCN_MEM_EXECUTE) 90 return true; 91 92 // .pdata and .xdata unwind info sections are eligible. 93 StringRef outSecName = c->getSectionName().split('$').first; 94 if (outSecName == ".pdata" || outSecName == ".xdata") 95 return true; 96 97 // So are vtables. 98 if (c->sym && c->sym->getName().startswith("??_7")) 99 return true; 100 101 // Anything else not in an address-significance table is eligible. 102 return !c->keepUnique; 103 } 104 105 // Split an equivalence class into smaller classes. 106 void ICF::segregate(size_t begin, size_t end, bool constant) { 107 while (begin < end) { 108 // Divide [Begin, End) into two. Let Mid be the start index of the 109 // second group. 110 auto bound = std::stable_partition( 111 chunks.begin() + begin + 1, chunks.begin() + end, [&](SectionChunk *s) { 112 if (constant) 113 return equalsConstant(chunks[begin], s); 114 return equalsVariable(chunks[begin], s); 115 }); 116 size_t mid = bound - chunks.begin(); 117 118 // Split [Begin, End) into [Begin, Mid) and [Mid, End). We use Mid as an 119 // equivalence class ID because every group ends with a unique index. 120 for (size_t i = begin; i < mid; ++i) 121 chunks[i]->eqClass[(cnt + 1) % 2] = mid; 122 123 // If we created a group, we need to iterate the main loop again. 124 if (mid != end) 125 repeat = true; 126 127 begin = mid; 128 } 129 } 130 131 // Returns true if two sections' associative children are equal. 132 bool ICF::assocEquals(const SectionChunk *a, const SectionChunk *b) { 133 // Ignore associated metadata sections that don't participate in ICF, such as 134 // debug info and CFGuard metadata. 135 auto considerForICF = [](const SectionChunk &assoc) { 136 StringRef Name = assoc.getSectionName(); 137 return !(Name.startswith(".debug") || Name == ".gfids$y" || 138 Name == ".giats$y" || Name == ".gljmp$y"); 139 }; 140 auto ra = make_filter_range(a->children(), considerForICF); 141 auto rb = make_filter_range(b->children(), considerForICF); 142 return std::equal(ra.begin(), ra.end(), rb.begin(), rb.end(), 143 [&](const SectionChunk &ia, const SectionChunk &ib) { 144 return ia.eqClass[cnt % 2] == ib.eqClass[cnt % 2]; 145 }); 146 } 147 148 // Compare "non-moving" part of two sections, namely everything 149 // except relocation targets. 150 bool ICF::equalsConstant(const SectionChunk *a, const SectionChunk *b) { 151 if (a->relocsSize != b->relocsSize) 152 return false; 153 154 // Compare relocations. 155 auto eq = [&](const coff_relocation &r1, const coff_relocation &r2) { 156 if (r1.Type != r2.Type || 157 r1.VirtualAddress != r2.VirtualAddress) { 158 return false; 159 } 160 Symbol *b1 = a->file->getSymbol(r1.SymbolTableIndex); 161 Symbol *b2 = b->file->getSymbol(r2.SymbolTableIndex); 162 if (b1 == b2) 163 return true; 164 if (auto *d1 = dyn_cast<DefinedRegular>(b1)) 165 if (auto *d2 = dyn_cast<DefinedRegular>(b2)) 166 return d1->getValue() == d2->getValue() && 167 d1->getChunk()->eqClass[cnt % 2] == d2->getChunk()->eqClass[cnt % 2]; 168 return false; 169 }; 170 if (!std::equal(a->getRelocs().begin(), a->getRelocs().end(), 171 b->getRelocs().begin(), eq)) 172 return false; 173 174 // Compare section attributes and contents. 175 return a->getOutputCharacteristics() == b->getOutputCharacteristics() && 176 a->getSectionName() == b->getSectionName() && 177 a->header->SizeOfRawData == b->header->SizeOfRawData && 178 a->checksum == b->checksum && a->getContents() == b->getContents() && 179 assocEquals(a, b); 180 } 181 182 // Compare "moving" part of two sections, namely relocation targets. 183 bool ICF::equalsVariable(const SectionChunk *a, const SectionChunk *b) { 184 // Compare relocations. 185 auto eq = [&](const coff_relocation &r1, const coff_relocation &r2) { 186 Symbol *b1 = a->file->getSymbol(r1.SymbolTableIndex); 187 Symbol *b2 = b->file->getSymbol(r2.SymbolTableIndex); 188 if (b1 == b2) 189 return true; 190 if (auto *d1 = dyn_cast<DefinedRegular>(b1)) 191 if (auto *d2 = dyn_cast<DefinedRegular>(b2)) 192 return d1->getChunk()->eqClass[cnt % 2] == d2->getChunk()->eqClass[cnt % 2]; 193 return false; 194 }; 195 return std::equal(a->getRelocs().begin(), a->getRelocs().end(), 196 b->getRelocs().begin(), eq) && 197 assocEquals(a, b); 198 } 199 200 // Find the first Chunk after Begin that has a different class from Begin. 201 size_t ICF::findBoundary(size_t begin, size_t end) { 202 for (size_t i = begin + 1; i < end; ++i) 203 if (chunks[begin]->eqClass[cnt % 2] != chunks[i]->eqClass[cnt % 2]) 204 return i; 205 return end; 206 } 207 208 void ICF::forEachClassRange(size_t begin, size_t end, 209 std::function<void(size_t, size_t)> fn) { 210 while (begin < end) { 211 size_t mid = findBoundary(begin, end); 212 fn(begin, mid); 213 begin = mid; 214 } 215 } 216 217 // Call Fn on each class group. 218 void ICF::forEachClass(std::function<void(size_t, size_t)> fn) { 219 // If the number of sections are too small to use threading, 220 // call Fn sequentially. 221 if (chunks.size() < 1024) { 222 forEachClassRange(0, chunks.size(), fn); 223 ++cnt; 224 return; 225 } 226 227 // Shard into non-overlapping intervals, and call Fn in parallel. 228 // The sharding must be completed before any calls to Fn are made 229 // so that Fn can modify the Chunks in its shard without causing data 230 // races. 231 const size_t numShards = 256; 232 size_t step = chunks.size() / numShards; 233 size_t boundaries[numShards + 1]; 234 boundaries[0] = 0; 235 boundaries[numShards] = chunks.size(); 236 parallelForEachN(1, numShards, [&](size_t i) { 237 boundaries[i] = findBoundary((i - 1) * step, chunks.size()); 238 }); 239 parallelForEachN(1, numShards + 1, [&](size_t i) { 240 if (boundaries[i - 1] < boundaries[i]) { 241 forEachClassRange(boundaries[i - 1], boundaries[i], fn); 242 } 243 }); 244 ++cnt; 245 } 246 247 // Merge identical COMDAT sections. 248 // Two sections are considered the same if their section headers, 249 // contents and relocations are all the same. 250 void ICF::run() { 251 ScopedTimer t(ctx.icfTimer); 252 253 // Collect only mergeable sections and group by hash value. 254 uint32_t nextId = 1; 255 for (Chunk *c : ctx.symtab.getChunks()) { 256 if (auto *sc = dyn_cast<SectionChunk>(c)) { 257 if (isEligible(sc)) 258 chunks.push_back(sc); 259 else 260 sc->eqClass[0] = nextId++; 261 } 262 } 263 264 // Make sure that ICF doesn't merge sections that are being handled by string 265 // tail merging. 266 for (MergeChunk *mc : ctx.mergeChunkInstances) 267 if (mc) 268 for (SectionChunk *sc : mc->sections) 269 sc->eqClass[0] = nextId++; 270 271 // Initially, we use hash values to partition sections. 272 parallelForEach(chunks, [&](SectionChunk *sc) { 273 sc->eqClass[0] = xxHash64(sc->getContents()); 274 }); 275 276 // Combine the hashes of the sections referenced by each section into its 277 // hash. 278 for (unsigned cnt = 0; cnt != 2; ++cnt) { 279 parallelForEach(chunks, [&](SectionChunk *sc) { 280 uint32_t hash = sc->eqClass[cnt % 2]; 281 for (Symbol *b : sc->symbols()) 282 if (auto *sym = dyn_cast_or_null<DefinedRegular>(b)) 283 hash += sym->getChunk()->eqClass[cnt % 2]; 284 // Set MSB to 1 to avoid collisions with non-hash classes. 285 sc->eqClass[(cnt + 1) % 2] = hash | (1U << 31); 286 }); 287 } 288 289 // From now on, sections in Chunks are ordered so that sections in 290 // the same group are consecutive in the vector. 291 llvm::stable_sort(chunks, [](const SectionChunk *a, const SectionChunk *b) { 292 return a->eqClass[0] < b->eqClass[0]; 293 }); 294 295 // Compare static contents and assign unique IDs for each static content. 296 forEachClass([&](size_t begin, size_t end) { segregate(begin, end, true); }); 297 298 // Split groups by comparing relocations until convergence is obtained. 299 do { 300 repeat = false; 301 forEachClass( 302 [&](size_t begin, size_t end) { segregate(begin, end, false); }); 303 } while (repeat); 304 305 log("ICF needed " + Twine(cnt) + " iterations"); 306 307 // Merge sections in the same classes. 308 forEachClass([&](size_t begin, size_t end) { 309 if (end - begin == 1) 310 return; 311 312 log("Selected " + chunks[begin]->getDebugName()); 313 for (size_t i = begin + 1; i < end; ++i) { 314 log(" Removed " + chunks[i]->getDebugName()); 315 chunks[begin]->replace(chunks[i]); 316 } 317 }); 318 } 319 320 // Entry point to ICF. 321 void doICF(COFFLinkerContext &ctx, ICFLevel icfLevel) { 322 ICF(ctx, icfLevel).run(); 323 } 324 325 } // namespace coff 326 } // namespace lld 327