| 1 | #include <iostream> |
| 2 | #include <iomanip> |
| 3 | #include <vector> |
| 4 | |
| 5 | #include <unordered_map> |
| 6 | |
| 7 | #include <sparsehash/dense_hash_map> |
| 8 | #include <sparsehash/sparse_hash_map> |
| 9 | |
| 10 | #include <Common/Stopwatch.h> |
| 11 | |
| 12 | //#define DBMS_HASH_MAP_COUNT_COLLISIONS |
| 13 | #define DBMS_HASH_MAP_DEBUG_RESIZES |
| 14 | |
| 15 | #include <Core/Types.h> |
| 16 | #include <IO/ReadBufferFromFile.h> |
| 17 | #include <IO/ReadHelpers.h> |
| 18 | #include <Compression/CompressedReadBuffer.h> |
| 19 | #include <common/StringRef.h> |
| 20 | #include <Common/HashTable/HashMap.h> |
| 21 | #include <Interpreters/AggregationCommon.h> |
| 22 | |
| 23 | |
| 24 | struct CompactStringRef |
| 25 | { |
| 26 | union |
| 27 | { |
| 28 | const char * data_mixed = nullptr; |
| 29 | struct |
| 30 | { |
| 31 | char dummy[6]; |
| 32 | UInt16 size; |
| 33 | }; |
| 34 | }; |
| 35 | |
| 36 | CompactStringRef(const char * data_, size_t size_) |
| 37 | { |
| 38 | data_mixed = data_; |
| 39 | size = size_; |
| 40 | } |
| 41 | |
| 42 | CompactStringRef(const unsigned char * data_, size_t size_) : CompactStringRef(reinterpret_cast<const char *>(data_), size_) {} |
| 43 | explicit CompactStringRef(const std::string & s) : CompactStringRef(s.data(), s.size()) {} |
| 44 | CompactStringRef() {} |
| 45 | |
| 46 | const char * data() const { return reinterpret_cast<const char *>(reinterpret_cast<intptr_t>(data_mixed) & 0x0000FFFFFFFFFFFFULL); } |
| 47 | |
| 48 | std::string toString() const { return std::string(data(), size); } |
| 49 | }; |
| 50 | |
| 51 | inline bool operator==(CompactStringRef lhs, CompactStringRef rhs) |
| 52 | { |
| 53 | if (lhs.size != rhs.size) |
| 54 | return false; |
| 55 | |
| 56 | const char * lhs_data = lhs.data(); |
| 57 | const char * rhs_data = rhs.data(); |
| 58 | for (size_t pos = lhs.size - 1; pos < lhs.size; --pos) |
| 59 | if (lhs_data[pos] != rhs_data[pos]) |
| 60 | return false; |
| 61 | |
| 62 | return true; |
| 63 | } |
| 64 | |
| 65 | namespace ZeroTraits |
| 66 | { |
| 67 | template <> |
| 68 | inline bool check<CompactStringRef>(CompactStringRef x) { return nullptr == x.data_mixed; } |
| 69 | |
| 70 | template <> |
| 71 | inline void set<CompactStringRef>(CompactStringRef & x) { x.data_mixed = nullptr; } |
| 72 | } |
| 73 | |
| 74 | template <> |
| 75 | struct DefaultHash<CompactStringRef> |
| 76 | { |
| 77 | size_t operator() (CompactStringRef x) const |
| 78 | { |
| 79 | return CityHash_v1_0_2::CityHash64(x.data(), x.size); |
| 80 | } |
| 81 | }; |
| 82 | |
| 83 | |
| 84 | static inline UInt64 mix(UInt64 h) |
| 85 | { |
| 86 | h ^= h >> 23; |
| 87 | h *= 0x2127599bf4325c37ULL; |
| 88 | h ^= h >> 47; |
| 89 | return h; |
| 90 | } |
| 91 | |
| 92 | struct FastHash64 |
| 93 | { |
| 94 | size_t operator() (CompactStringRef x) const |
| 95 | { |
| 96 | const char * buf = x.data(); |
| 97 | size_t len = x.size; |
| 98 | |
| 99 | const UInt64 m = 0x880355f21e6d1965ULL; |
| 100 | const UInt64 *pos = reinterpret_cast<const UInt64 *>(buf); |
| 101 | const UInt64 *end = pos + (len / 8); |
| 102 | const unsigned char *pos2; |
| 103 | UInt64 h = len * m; |
| 104 | UInt64 v; |
| 105 | |
| 106 | while (pos != end) |
| 107 | { |
| 108 | v = *pos++; |
| 109 | h ^= mix(v); |
| 110 | h *= m; |
| 111 | } |
| 112 | |
| 113 | pos2 = reinterpret_cast<const unsigned char*>(pos); |
| 114 | v = 0; |
| 115 | |
| 116 | switch (len & 7) |
| 117 | { |
| 118 | case 7: v ^= static_cast<UInt64>(pos2[6]) << 48; [[fallthrough]]; |
| 119 | case 6: v ^= static_cast<UInt64>(pos2[5]) << 40; [[fallthrough]]; |
| 120 | case 5: v ^= static_cast<UInt64>(pos2[4]) << 32; [[fallthrough]]; |
| 121 | case 4: v ^= static_cast<UInt64>(pos2[3]) << 24; [[fallthrough]]; |
| 122 | case 3: v ^= static_cast<UInt64>(pos2[2]) << 16; [[fallthrough]]; |
| 123 | case 2: v ^= static_cast<UInt64>(pos2[1]) << 8; [[fallthrough]]; |
| 124 | case 1: v ^= static_cast<UInt64>(pos2[0]); |
| 125 | h ^= mix(v); |
| 126 | h *= m; |
| 127 | } |
| 128 | |
| 129 | return mix(h); |
| 130 | } |
| 131 | }; |
| 132 | |
| 133 | |
| 134 | #if defined(__x86_64__) |
| 135 | struct CrapWow |
| 136 | { |
| 137 | size_t operator() (CompactStringRef x) const |
| 138 | { |
| 139 | const char * key = x.data(); |
| 140 | size_t len = x.size; |
| 141 | size_t seed = 0; |
| 142 | |
| 143 | const UInt64 m = 0x95b47aa3355ba1a1, n = 0x8a970be7488fda55; |
| 144 | |
| 145 | UInt64 hash; |
| 146 | // 3 = m, 4 = n |
| 147 | // r12 = h, r13 = k, ecx = seed, r12 = key |
| 148 | asm( |
| 149 | "leaq (%%rcx,%4), %%r13\n" |
| 150 | "movq %%rdx, %%r14\n" |
| 151 | "movq %%rcx, %%r15\n" |
| 152 | "movq %%rcx, %%r12\n" |
| 153 | "addq %%rax, %%r13\n" |
| 154 | "andq $0xfffffffffffffff0, %%rcx\n" |
| 155 | "jz QW%=\n" |
| 156 | "addq %%rcx, %%r14\n\n" |
| 157 | "negq %%rcx\n" |
| 158 | "XW%=:\n" |
| 159 | "movq %4, %%rax\n" |
| 160 | "mulq (%%r14,%%rcx)\n" |
| 161 | "xorq %%rax, %%r12\n" |
| 162 | "xorq %%rdx, %%r13\n" |
| 163 | "movq %3, %%rax\n" |
| 164 | "mulq 8(%%r14,%%rcx)\n" |
| 165 | "xorq %%rdx, %%r12\n" |
| 166 | "xorq %%rax, %%r13\n" |
| 167 | "addq $16, %%rcx\n" |
| 168 | "jnz XW%=\n" |
| 169 | "QW%=:\n" |
| 170 | "movq %%r15, %%rcx\n" |
| 171 | "andq $8, %%r15\n" |
| 172 | "jz B%=\n" |
| 173 | "movq %4, %%rax\n" |
| 174 | "mulq (%%r14)\n" |
| 175 | "addq $8, %%r14\n" |
| 176 | "xorq %%rax, %%r12\n" |
| 177 | "xorq %%rdx, %%r13\n" |
| 178 | "B%=:\n" |
| 179 | "andq $7, %%rcx\n" |
| 180 | "jz F%=\n" |
| 181 | "movq $1, %%rdx\n" |
| 182 | "shlq $3, %%rcx\n" |
| 183 | "movq %3, %%rax\n" |
| 184 | "shlq %%cl, %%rdx\n" |
| 185 | "addq $-1, %%rdx\n" |
| 186 | "andq (%%r14), %%rdx\n" |
| 187 | "mulq %%rdx\n" |
| 188 | "xorq %%rdx, %%r12\n" |
| 189 | "xorq %%rax, %%r13\n" |
| 190 | "F%=:\n" |
| 191 | "leaq (%%r13,%4), %%rax\n" |
| 192 | "xorq %%r12, %%rax\n" |
| 193 | "mulq %4\n" |
| 194 | "xorq %%rdx, %%rax\n" |
| 195 | "xorq %%r12, %%rax\n" |
| 196 | "xorq %%r13, %%rax\n" |
| 197 | : "=a" (hash), "=c" (key), "=d" (key) |
| 198 | : "r" (m), "r" (n), "a" (seed), "c" (len), "d" (key) |
| 199 | : "%r12" , "%r13" , "%r14" , "%r15" , "cc" |
| 200 | ); |
| 201 | return hash; |
| 202 | } |
| 203 | }; |
| 204 | #endif |
| 205 | |
| 206 | |
| 207 | struct SimpleHash |
| 208 | { |
| 209 | size_t operator() (CompactStringRef x) const |
| 210 | { |
| 211 | const char * pos = x.data(); |
| 212 | size_t size = x.size; |
| 213 | |
| 214 | const char * end = pos + size; |
| 215 | |
| 216 | size_t res = 0; |
| 217 | |
| 218 | if (size == 0) |
| 219 | return 0; |
| 220 | |
| 221 | if (size < 8) |
| 222 | { |
| 223 | memcpy(reinterpret_cast<char *>(&res), pos, size); |
| 224 | return intHash64(res); |
| 225 | } |
| 226 | |
| 227 | while (pos + 8 < end) |
| 228 | { |
| 229 | UInt64 word = *reinterpret_cast<const UInt64 *>(pos); |
| 230 | res = intHash64(word ^ res); |
| 231 | |
| 232 | pos += 8; |
| 233 | } |
| 234 | |
| 235 | UInt64 word = *reinterpret_cast<const UInt64 *>(end - 8); |
| 236 | res = intHash64(word ^ res); |
| 237 | |
| 238 | return res; |
| 239 | } |
| 240 | }; |
| 241 | |
| 242 | |
| 243 | using Key = CompactStringRef; |
| 244 | using Value = UInt64; |
| 245 | |
| 246 | |
| 247 | struct Grower : public HashTableGrower<> |
| 248 | { |
| 249 | /// The state of this structure is enough to get the buffer size of the hash table. |
| 250 | |
| 251 | /// Defines the initial size of the hash table. |
| 252 | static const size_t initial_size_degree = 16; |
| 253 | Grower() { size_degree = initial_size_degree; } |
| 254 | |
| 255 | size_t max_fill = (1ULL << initial_size_degree) * 0.9; |
| 256 | |
| 257 | /// The size of the hash table in the cells. |
| 258 | size_t bufSize() const { return 1ULL << size_degree; } |
| 259 | |
| 260 | size_t maxFill() const { return max_fill /*1 << (size_degree - 1)*/; } |
| 261 | size_t mask() const { return bufSize() - 1; } |
| 262 | |
| 263 | /// From the hash value, get the cell number in the hash table. |
| 264 | size_t place(size_t x) const { return x & mask(); } |
| 265 | |
| 266 | /// The next cell in the collision resolution chain. |
| 267 | size_t next(size_t pos) const { ++pos; return pos & mask(); } |
| 268 | |
| 269 | /// Whether the hash table is sufficiently full. You need to increase the size of the hash table, or remove something unnecessary from it. |
| 270 | bool overflow(size_t elems) const { return elems > maxFill(); } |
| 271 | |
| 272 | /// Increase the size of the hash table. |
| 273 | void increaseSize() |
| 274 | { |
| 275 | size_degree += size_degree >= 23 ? 1 : 2; |
| 276 | max_fill = (1ULL << size_degree) * 0.9; |
| 277 | } |
| 278 | |
| 279 | /// Set the buffer size by the number of elements in the hash table. Used when deserializing a hash table. |
| 280 | [[noreturn]] void set(size_t /*num_elems*/) |
| 281 | { |
| 282 | throw Poco::Exception(__PRETTY_FUNCTION__); |
| 283 | } |
| 284 | }; |
| 285 | |
| 286 | |
| 287 | int main(int argc, char ** argv) |
| 288 | { |
| 289 | if (argc < 3) |
| 290 | { |
| 291 | std::cerr << "Usage: program n m\n" ; |
| 292 | return 1; |
| 293 | } |
| 294 | |
| 295 | size_t n = atoi(argv[1]); |
| 296 | size_t m = atoi(argv[2]); |
| 297 | |
| 298 | DB::Arena pool; |
| 299 | std::vector<Key> data(n); |
| 300 | |
| 301 | std::cerr << "sizeof(Key) = " << sizeof(Key) << ", sizeof(Value) = " << sizeof(Value) << std::endl; |
| 302 | |
| 303 | { |
| 304 | Stopwatch watch; |
| 305 | DB::ReadBufferFromFileDescriptor in1(STDIN_FILENO); |
| 306 | DB::CompressedReadBuffer in2(in1); |
| 307 | |
| 308 | std::string tmp; |
| 309 | for (size_t i = 0; i < n && !in2.eof(); ++i) |
| 310 | { |
| 311 | DB::readStringBinary(tmp, in2); |
| 312 | data[i] = Key(pool.insert(tmp.data(), tmp.size()), tmp.size()); |
| 313 | } |
| 314 | |
| 315 | watch.stop(); |
| 316 | std::cerr << std::fixed << std::setprecision(2) |
| 317 | << "Vector. Size: " << n |
| 318 | << ", elapsed: " << watch.elapsedSeconds() |
| 319 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 320 | << std::endl; |
| 321 | } |
| 322 | |
| 323 | if (!m || m == 1) |
| 324 | { |
| 325 | Stopwatch watch; |
| 326 | |
| 327 | //using Map = HashMap<Key, Value>; |
| 328 | |
| 329 | /// Saving the hash accelerates the resize by about 2 times, and the overall performance by 6-8%. |
| 330 | using Map = HashMapWithSavedHash<Key, Value, DefaultHash<Key>, Grower>; |
| 331 | |
| 332 | Map map; |
| 333 | Map::LookupResult it; |
| 334 | bool inserted; |
| 335 | |
| 336 | for (size_t i = 0; i < n; ++i) |
| 337 | { |
| 338 | map.emplace(data[i], it, inserted); |
| 339 | if (inserted) |
| 340 | it->getMapped() = 0; |
| 341 | ++it->getMapped(); |
| 342 | } |
| 343 | |
| 344 | watch.stop(); |
| 345 | std::cerr << std::fixed << std::setprecision(2) |
| 346 | << "HashMap (CityHash64). Size: " << map.size() |
| 347 | << ", elapsed: " << watch.elapsedSeconds() |
| 348 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 349 | #ifdef DBMS_HASH_MAP_COUNT_COLLISIONS |
| 350 | << ", collisions: " << map.getCollisions() |
| 351 | #endif |
| 352 | << std::endl; |
| 353 | } |
| 354 | |
| 355 | if (!m || m == 2) |
| 356 | { |
| 357 | Stopwatch watch; |
| 358 | |
| 359 | using Map = HashMapWithSavedHash<Key, Value, FastHash64, Grower>; |
| 360 | |
| 361 | Map map; |
| 362 | Map::LookupResult it; |
| 363 | bool inserted; |
| 364 | |
| 365 | for (size_t i = 0; i < n; ++i) |
| 366 | { |
| 367 | map.emplace(data[i], it, inserted); |
| 368 | if (inserted) |
| 369 | it->getMapped() = 0; |
| 370 | ++it->getMapped(); |
| 371 | } |
| 372 | |
| 373 | watch.stop(); |
| 374 | std::cerr << std::fixed << std::setprecision(2) |
| 375 | << "HashMap (FastHash64). Size: " << map.size() |
| 376 | << ", elapsed: " << watch.elapsedSeconds() |
| 377 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 378 | #ifdef DBMS_HASH_MAP_COUNT_COLLISIONS |
| 379 | << ", collisions: " << map.getCollisions() |
| 380 | #endif |
| 381 | << std::endl; |
| 382 | } |
| 383 | |
| 384 | #if defined(__x86_64__) |
| 385 | if (!m || m == 3) |
| 386 | { |
| 387 | Stopwatch watch; |
| 388 | |
| 389 | using Map = HashMapWithSavedHash<Key, Value, CrapWow, Grower>; |
| 390 | |
| 391 | Map map; |
| 392 | Map::LookupResult it; |
| 393 | bool inserted; |
| 394 | |
| 395 | for (size_t i = 0; i < n; ++i) |
| 396 | { |
| 397 | map.emplace(data[i], it, inserted); |
| 398 | if (inserted) |
| 399 | it->getMapped() = 0; |
| 400 | ++it->getMapped(); |
| 401 | } |
| 402 | |
| 403 | watch.stop(); |
| 404 | std::cerr << std::fixed << std::setprecision(2) |
| 405 | << "HashMap (CrapWow). Size: " << map.size() |
| 406 | << ", elapsed: " << watch.elapsedSeconds() |
| 407 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 408 | #ifdef DBMS_HASH_MAP_COUNT_COLLISIONS |
| 409 | << ", collisions: " << map.getCollisions() |
| 410 | #endif |
| 411 | << std::endl; |
| 412 | } |
| 413 | #endif |
| 414 | |
| 415 | if (!m || m == 4) |
| 416 | { |
| 417 | Stopwatch watch; |
| 418 | |
| 419 | using Map = HashMapWithSavedHash<Key, Value, SimpleHash, Grower>; |
| 420 | |
| 421 | Map map; |
| 422 | Map::LookupResult it; |
| 423 | bool inserted; |
| 424 | |
| 425 | for (size_t i = 0; i < n; ++i) |
| 426 | { |
| 427 | map.emplace(data[i], it, inserted); |
| 428 | if (inserted) |
| 429 | it->getMapped() = 0; |
| 430 | ++it->getMapped(); |
| 431 | } |
| 432 | |
| 433 | watch.stop(); |
| 434 | std::cerr << std::fixed << std::setprecision(2) |
| 435 | << "HashMap (SimpleHash). Size: " << map.size() |
| 436 | << ", elapsed: " << watch.elapsedSeconds() |
| 437 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 438 | #ifdef DBMS_HASH_MAP_COUNT_COLLISIONS |
| 439 | << ", collisions: " << map.getCollisions() |
| 440 | #endif |
| 441 | << std::endl; |
| 442 | } |
| 443 | |
| 444 | if (!m || m == 5) |
| 445 | { |
| 446 | Stopwatch watch; |
| 447 | |
| 448 | std::unordered_map<Key, Value, DefaultHash<Key>> map; |
| 449 | for (size_t i = 0; i < n; ++i) |
| 450 | ++map[data[i]]; |
| 451 | |
| 452 | watch.stop(); |
| 453 | std::cerr << std::fixed << std::setprecision(2) |
| 454 | << "std::unordered_map. Size: " << map.size() |
| 455 | << ", elapsed: " << watch.elapsedSeconds() |
| 456 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 457 | << std::endl; |
| 458 | } |
| 459 | |
| 460 | if (!m || m == 6) |
| 461 | { |
| 462 | Stopwatch watch; |
| 463 | |
| 464 | google::dense_hash_map<Key, Value, DefaultHash<Key>> map; |
| 465 | map.set_empty_key(Key("\0" , 1)); |
| 466 | for (size_t i = 0; i < n; ++i) |
| 467 | ++map[data[i]]; |
| 468 | |
| 469 | watch.stop(); |
| 470 | std::cerr << std::fixed << std::setprecision(2) |
| 471 | << "google::dense_hash_map. Size: " << map.size() |
| 472 | << ", elapsed: " << watch.elapsedSeconds() |
| 473 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 474 | << std::endl; |
| 475 | } |
| 476 | |
| 477 | if (!m || m == 7) |
| 478 | { |
| 479 | Stopwatch watch; |
| 480 | |
| 481 | google::sparse_hash_map<Key, Value, DefaultHash<Key>> map; |
| 482 | for (size_t i = 0; i < n; ++i) |
| 483 | ++map[data[i]]; |
| 484 | |
| 485 | watch.stop(); |
| 486 | std::cerr << std::fixed << std::setprecision(2) |
| 487 | << "google::sparse_hash_map. Size: " << map.size() |
| 488 | << ", elapsed: " << watch.elapsedSeconds() |
| 489 | << " (" << n / watch.elapsedSeconds() << " elem/sec.)" |
| 490 | << std::endl; |
| 491 | } |
| 492 | |
| 493 | return 0; |
| 494 | } |
| 495 | |