| 1 | /* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */ |
| 2 | // vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4: |
| 3 | #ident "$Id$" |
| 4 | /*====== |
| 5 | This file is part of PerconaFT. |
| 6 | |
| 7 | |
| 8 | Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved. |
| 9 | |
| 10 | PerconaFT is free software: you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License, version 2, |
| 12 | as published by the Free Software Foundation. |
| 13 | |
| 14 | PerconaFT is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 21 | |
| 22 | ---------------------------------------- |
| 23 | |
| 24 | PerconaFT is free software: you can redistribute it and/or modify |
| 25 | it under the terms of the GNU Affero General Public License, version 3, |
| 26 | as published by the Free Software Foundation. |
| 27 | |
| 28 | PerconaFT is distributed in the hope that it will be useful, |
| 29 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 30 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 31 | GNU Affero General Public License for more details. |
| 32 | |
| 33 | You should have received a copy of the GNU Affero General Public License |
| 34 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 35 | ======= */ |
| 36 | |
| 37 | #ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved." |
| 38 | |
| 39 | #include <string.h> |
| 40 | #include <db.h> |
| 41 | |
| 42 | #include <portability/memory.h> |
| 43 | #include <limits.h> |
| 44 | |
| 45 | namespace toku { |
| 46 | |
| 47 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 48 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::create(void) { |
| 49 | toku_mempool_zero(&this->mp); |
| 50 | this->values_same_size = true; |
| 51 | this->value_length = 0; |
| 52 | this->is_array = true; |
| 53 | this->d.a.num_values = 0; |
| 54 | //TODO: maybe allocate enough space for something by default? |
| 55 | // We may be relying on not needing to allocate space the first time (due to limited time spent while a lock is held) |
| 56 | } |
| 57 | |
| 58 | /** |
| 59 | * Note: create_from_sorted_memory_of_fixed_size_elements does not take ownership of 'mem'. |
| 60 | * Owner is still responsible for freeing it. |
| 61 | * While in the OMT a similar function would steal ownership, this doesn't make sense for the DMT because |
| 62 | * we (usually) have to add padding for alignment (mem has all of the elements PACKED). |
| 63 | * Also all current uses (as of Jan 12, 2014) of this function would require mallocing a new array |
| 64 | * in order to allow stealing. |
| 65 | */ |
| 66 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 67 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::create_from_sorted_memory_of_fixed_size_elements( |
| 68 | const void *mem, |
| 69 | const uint32_t numvalues, |
| 70 | const uint32_t mem_length, |
| 71 | const uint32_t fixed_value_length) { |
| 72 | this->values_same_size = true; |
| 73 | this->value_length = fixed_value_length; |
| 74 | this->is_array = true; |
| 75 | this->d.a.num_values = numvalues; |
| 76 | const uint8_t pad_bytes = get_fixed_length_alignment_overhead(); |
| 77 | uint32_t aligned_memsize = mem_length + numvalues * pad_bytes; |
| 78 | toku_mempool_construct(&this->mp, aligned_memsize); |
| 79 | if (aligned_memsize > 0) { |
| 80 | paranoid_invariant(numvalues > 0); |
| 81 | void *ptr = toku_mempool_malloc(&this->mp, aligned_memsize); |
| 82 | paranoid_invariant_notnull(ptr); |
| 83 | uint8_t * const CAST_FROM_VOIDP(dest, ptr); |
| 84 | const uint8_t * const CAST_FROM_VOIDP(src, mem); |
| 85 | if (pad_bytes == 0) { |
| 86 | paranoid_invariant(aligned_memsize == mem_length); |
| 87 | memcpy(dest, src, aligned_memsize); |
| 88 | } else { |
| 89 | // TODO(leif): check what vectorizes best: multiplying like this or adding to offsets |
| 90 | const uint32_t fixed_len = this->value_length; |
| 91 | const uint32_t fixed_aligned_len = align(this->value_length); |
| 92 | paranoid_invariant(this->d.a.num_values*fixed_len == mem_length); |
| 93 | for (uint32_t i = 0; i < this->d.a.num_values; i++) { |
| 94 | memcpy(&dest[i*fixed_aligned_len], &src[i*fixed_len], fixed_len); |
| 95 | } |
| 96 | } |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 101 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::clone(const dmt &src) { |
| 102 | *this = src; |
| 103 | toku_mempool_clone(&src.mp, &this->mp); |
| 104 | } |
| 105 | |
| 106 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 107 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::clear(void) { |
| 108 | this->is_array = true; |
| 109 | this->d.a.num_values = 0; |
| 110 | this->values_same_size = true; // Reset state |
| 111 | this->value_length = 0; |
| 112 | //TODO(leif): Note that this can mess with our memory_footprint calculation (we may touch past what is marked as 'used' in the mempool) |
| 113 | // One 'fix' is for mempool to also track what was touched, and reset() shouldn't reset that, though realloc() might. |
| 114 | toku_mempool_reset(&this->mp); |
| 115 | } |
| 116 | |
| 117 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 118 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::destroy(void) { |
| 119 | this->clear(); |
| 120 | toku_mempool_destroy(&this->mp); |
| 121 | } |
| 122 | |
| 123 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 124 | uint32_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::size(void) const { |
| 125 | if (this->is_array) { |
| 126 | return this->d.a.num_values; |
| 127 | } else { |
| 128 | return this->nweight(this->d.t.root); |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 133 | uint32_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::nweight(const subtree &subtree) const { |
| 134 | if (subtree.is_null()) { |
| 135 | return 0; |
| 136 | } else { |
| 137 | const dmt_node & node = get_node(subtree); |
| 138 | return node.weight; |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 143 | template<typename dmtcmp_t, int (*h)(const uint32_t size, const dmtdata_t &, const dmtcmp_t &)> |
| 144 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::insert(const dmtwriter_t &value, const dmtcmp_t &v, uint32_t *const idx) { |
| 145 | int r; |
| 146 | uint32_t insert_idx; |
| 147 | |
| 148 | r = this->find_zero<dmtcmp_t, h>(v, nullptr, nullptr, &insert_idx); |
| 149 | if (r==0) { |
| 150 | if (idx) *idx = insert_idx; |
| 151 | return DB_KEYEXIST; |
| 152 | } |
| 153 | if (r != DB_NOTFOUND) return r; |
| 154 | |
| 155 | if ((r = this->insert_at(value, insert_idx))) return r; |
| 156 | if (idx) *idx = insert_idx; |
| 157 | |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 162 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::insert_at(const dmtwriter_t &value, const uint32_t idx) { |
| 163 | if (idx > this->size()) { return EINVAL; } |
| 164 | |
| 165 | bool same_size = this->values_same_size && (this->size() == 0 || value.get_size() == this->value_length); |
| 166 | if (this->is_array) { |
| 167 | if (same_size && idx == this->d.a.num_values) { |
| 168 | return this->insert_at_array_end<true>(value); |
| 169 | } |
| 170 | this->convert_from_array_to_tree(); |
| 171 | } |
| 172 | // Is a tree. |
| 173 | paranoid_invariant(!is_array); |
| 174 | if (!same_size) { |
| 175 | this->values_same_size = false; |
| 176 | this->value_length = 0; |
| 177 | } |
| 178 | |
| 179 | this->maybe_resize_tree(&value); |
| 180 | subtree *rebalance_subtree = nullptr; |
| 181 | this->insert_internal(&this->d.t.root, value, idx, &rebalance_subtree); |
| 182 | if (rebalance_subtree != nullptr) { |
| 183 | this->rebalance(rebalance_subtree); |
| 184 | } |
| 185 | return 0; |
| 186 | } |
| 187 | |
| 188 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 189 | template<bool with_resize> |
| 190 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::insert_at_array_end(const dmtwriter_t& value_in) { |
| 191 | paranoid_invariant(this->is_array); |
| 192 | paranoid_invariant(this->values_same_size); |
| 193 | if (this->d.a.num_values == 0) { |
| 194 | this->value_length = value_in.get_size(); |
| 195 | } |
| 196 | paranoid_invariant(this->value_length == value_in.get_size()); |
| 197 | |
| 198 | if (with_resize) { |
| 199 | this->maybe_resize_array_for_insert(); |
| 200 | } |
| 201 | dmtdata_t *dest = this->alloc_array_value_end(); |
| 202 | value_in.write_to(dest); |
| 203 | return 0; |
| 204 | } |
| 205 | |
| 206 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 207 | dmtdata_t * dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::alloc_array_value_end(void) { |
| 208 | paranoid_invariant(this->is_array); |
| 209 | paranoid_invariant(this->values_same_size); |
| 210 | this->d.a.num_values++; |
| 211 | |
| 212 | void *ptr = toku_mempool_malloc(&this->mp, align(this->value_length)); |
| 213 | paranoid_invariant_notnull(ptr); |
| 214 | paranoid_invariant(reinterpret_cast<size_t>(ptr) % ALIGNMENT == 0); |
| 215 | dmtdata_t *CAST_FROM_VOIDP(n, ptr); |
| 216 | paranoid_invariant(n == get_array_value(this->d.a.num_values - 1)); |
| 217 | return n; |
| 218 | } |
| 219 | |
| 220 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 221 | dmtdata_t * dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_array_value(const uint32_t idx) const { |
| 222 | paranoid_invariant(this->is_array); |
| 223 | paranoid_invariant(this->values_same_size); |
| 224 | |
| 225 | paranoid_invariant(idx < this->d.a.num_values); |
| 226 | return get_array_value_internal(&this->mp, idx); |
| 227 | } |
| 228 | |
| 229 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 230 | dmtdata_t * dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_array_value_internal(const struct mempool *mempool, const uint32_t idx) const { |
| 231 | void* ptr = toku_mempool_get_pointer_from_base_and_offset(mempool, idx * align(this->value_length)); |
| 232 | dmtdata_t *CAST_FROM_VOIDP(value, ptr); |
| 233 | return value; |
| 234 | } |
| 235 | |
| 236 | //TODO(leif) write microbenchmarks to compare growth factor. Note: growth factor here is actually 2.5 because of mempool_construct |
| 237 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 238 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::maybe_resize_array_for_insert(void) { |
| 239 | bool space_available = toku_mempool_get_free_size(&this->mp) >= align(this->value_length); |
| 240 | |
| 241 | if (!space_available) { |
| 242 | const uint32_t n = this->d.a.num_values + 1; |
| 243 | const uint32_t new_n = n <=2 ? 4 : 2*n; |
| 244 | const uint32_t new_space = align(this->value_length) * new_n; |
| 245 | |
| 246 | struct mempool new_kvspace; |
| 247 | toku_mempool_construct(&new_kvspace, new_space); |
| 248 | size_t copy_bytes = this->d.a.num_values * align(this->value_length); |
| 249 | invariant(copy_bytes + align(this->value_length) <= new_space); |
| 250 | paranoid_invariant(copy_bytes <= toku_mempool_get_used_size(&this->mp)); |
| 251 | // Copy over to new mempool |
| 252 | if (this->d.a.num_values > 0) { |
| 253 | void* dest = toku_mempool_malloc(&new_kvspace, copy_bytes); |
| 254 | invariant(dest!=nullptr); |
| 255 | memcpy(dest, get_array_value(0), copy_bytes); |
| 256 | } |
| 257 | toku_mempool_destroy(&this->mp); |
| 258 | this->mp = new_kvspace; |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 263 | uint32_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::align(const uint32_t x) const { |
| 264 | return roundup_to_multiple(ALIGNMENT, x); |
| 265 | } |
| 266 | |
| 267 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 268 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::prepare_for_serialize(void) { |
| 269 | if (!this->is_array) { |
| 270 | this->convert_from_tree_to_array(); |
| 271 | } |
| 272 | } |
| 273 | |
| 274 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 275 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::convert_from_tree_to_array(void) { |
| 276 | paranoid_invariant(!this->is_array); |
| 277 | paranoid_invariant(this->values_same_size); |
| 278 | |
| 279 | const uint32_t num_values = this->size(); |
| 280 | |
| 281 | node_offset *tmp_array; |
| 282 | bool malloced = false; |
| 283 | tmp_array = alloc_temp_node_offsets(num_values); |
| 284 | if (!tmp_array) { |
| 285 | malloced = true; |
| 286 | XMALLOC_N(num_values, tmp_array); |
| 287 | } |
| 288 | this->fill_array_with_subtree_offsets(tmp_array, this->d.t.root); |
| 289 | |
| 290 | struct mempool new_mp; |
| 291 | const uint32_t fixed_len = this->value_length; |
| 292 | const uint32_t fixed_aligned_len = align(this->value_length); |
| 293 | size_t mem_needed = num_values * fixed_aligned_len; |
| 294 | toku_mempool_construct(&new_mp, mem_needed); |
| 295 | uint8_t* CAST_FROM_VOIDP(dest, toku_mempool_malloc(&new_mp, mem_needed)); |
| 296 | paranoid_invariant_notnull(dest); |
| 297 | for (uint32_t i = 0; i < num_values; i++) { |
| 298 | const dmt_node &n = get_node(tmp_array[i]); |
| 299 | memcpy(&dest[i*fixed_aligned_len], &n.value, fixed_len); |
| 300 | } |
| 301 | toku_mempool_destroy(&this->mp); |
| 302 | this->mp = new_mp; |
| 303 | this->is_array = true; |
| 304 | this->d.a.num_values = num_values; |
| 305 | |
| 306 | if (malloced) toku_free(tmp_array); |
| 307 | } |
| 308 | |
| 309 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 310 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::convert_from_array_to_tree(void) { |
| 311 | paranoid_invariant(this->is_array); |
| 312 | paranoid_invariant(this->values_same_size); |
| 313 | |
| 314 | //save array-format information to locals |
| 315 | const uint32_t num_values = this->d.a.num_values; |
| 316 | |
| 317 | node_offset *tmp_array; |
| 318 | bool malloced = false; |
| 319 | tmp_array = alloc_temp_node_offsets(num_values); |
| 320 | if (!tmp_array) { |
| 321 | malloced = true; |
| 322 | XMALLOC_N(num_values, tmp_array); |
| 323 | } |
| 324 | |
| 325 | struct mempool old_mp = this->mp; |
| 326 | size_t mem_needed = num_values * align(this->value_length + __builtin_offsetof(dmt_node, value)); |
| 327 | toku_mempool_construct(&this->mp, mem_needed); |
| 328 | |
| 329 | for (uint32_t i = 0; i < num_values; i++) { |
| 330 | dmtwriter_t writer(this->value_length, get_array_value_internal(&old_mp, i)); |
| 331 | tmp_array[i] = node_malloc_and_set_value(writer); |
| 332 | } |
| 333 | this->is_array = false; |
| 334 | this->rebuild_subtree_from_offsets(&this->d.t.root, tmp_array, num_values); |
| 335 | |
| 336 | if (malloced) toku_free(tmp_array); |
| 337 | toku_mempool_destroy(&old_mp); |
| 338 | } |
| 339 | |
| 340 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 341 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::delete_at(const uint32_t idx) { |
| 342 | uint32_t n = this->size(); |
| 343 | if (idx >= n) { return EINVAL; } |
| 344 | |
| 345 | if (n == 1) { |
| 346 | this->clear(); //Emptying out the entire dmt. |
| 347 | return 0; |
| 348 | } |
| 349 | if (this->is_array) { |
| 350 | this->convert_from_array_to_tree(); |
| 351 | } |
| 352 | paranoid_invariant(!is_array); |
| 353 | |
| 354 | subtree *rebalance_subtree = nullptr; |
| 355 | this->delete_internal(&this->d.t.root, idx, nullptr, &rebalance_subtree); |
| 356 | if (rebalance_subtree != nullptr) { |
| 357 | this->rebalance(rebalance_subtree); |
| 358 | } |
| 359 | this->maybe_resize_tree(nullptr); |
| 360 | return 0; |
| 361 | } |
| 362 | |
| 363 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 364 | template<typename iterate_extra_t, |
| 365 | int (*f)(const uint32_t, const dmtdata_t &, const uint32_t, iterate_extra_t *const)> |
| 366 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate(iterate_extra_t *const ) const { |
| 367 | return this->iterate_on_range<iterate_extra_t, f>(0, this->size(), iterate_extra); |
| 368 | } |
| 369 | |
| 370 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 371 | template<typename iterate_extra_t, |
| 372 | int (*f)(const uint32_t, const dmtdata_t &, const uint32_t, iterate_extra_t *const)> |
| 373 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_on_range(const uint32_t left, const uint32_t right, iterate_extra_t *const ) const { |
| 374 | if (right > this->size()) { return EINVAL; } |
| 375 | if (left == right) { return 0; } |
| 376 | if (this->is_array) { |
| 377 | return this->iterate_internal_array<iterate_extra_t, f>(left, right, iterate_extra); |
| 378 | } |
| 379 | return this->iterate_internal<iterate_extra_t, f>(left, right, this->d.t.root, 0, iterate_extra); |
| 380 | } |
| 381 | |
| 382 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 383 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::verify(void) const { |
| 384 | uint32_t num_values = this->size(); |
| 385 | invariant(num_values < UINT32_MAX); |
| 386 | size_t pool_used = toku_mempool_get_used_size(&this->mp); |
| 387 | size_t pool_size = toku_mempool_get_size(&this->mp); |
| 388 | size_t pool_frag = toku_mempool_get_frag_size(&this->mp); |
| 389 | invariant(pool_used <= pool_size); |
| 390 | if (this->is_array) { |
| 391 | invariant(this->values_same_size); |
| 392 | invariant(num_values == this->d.a.num_values); |
| 393 | |
| 394 | // We know exactly how much memory should be used. |
| 395 | invariant(pool_used == num_values * align(this->value_length)); |
| 396 | |
| 397 | // Array form must have 0 fragmentation in mempool. |
| 398 | invariant(pool_frag == 0); |
| 399 | } else { |
| 400 | if (this->values_same_size) { |
| 401 | // We know exactly how much memory should be used. |
| 402 | invariant(pool_used == num_values * align(this->value_length + __builtin_offsetof(dmt_node, value))); |
| 403 | } else { |
| 404 | // We can only do a lower bound on memory usage. |
| 405 | invariant(pool_used >= num_values * __builtin_offsetof(dmt_node, value)); |
| 406 | } |
| 407 | std::vector<bool> touched(pool_size, false); |
| 408 | verify_internal(this->d.t.root, &touched); |
| 409 | size_t bytes_used = 0; |
| 410 | for (size_t i = 0; i < pool_size; i++) { |
| 411 | if (touched.at(i)) { |
| 412 | ++bytes_used; |
| 413 | } |
| 414 | } |
| 415 | invariant(bytes_used == pool_used); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | // Verifies all weights are internally consistent. |
| 420 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 421 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::verify_internal(const subtree &subtree, std::vector<bool> *touched) const { |
| 422 | if (subtree.is_null()) { |
| 423 | return; |
| 424 | } |
| 425 | const dmt_node &node = get_node(subtree); |
| 426 | |
| 427 | if (this->values_same_size) { |
| 428 | invariant(node.value_length == this->value_length); |
| 429 | } |
| 430 | |
| 431 | size_t offset = toku_mempool_get_offset_from_pointer_and_base(&this->mp, &node); |
| 432 | size_t node_size = align(__builtin_offsetof(dmt_node, value) + node.value_length); |
| 433 | invariant(offset <= touched->size()); |
| 434 | invariant(offset+node_size <= touched->size()); |
| 435 | invariant(offset % ALIGNMENT == 0); |
| 436 | // Mark memory as touched and never allocated to multiple nodes. |
| 437 | for (size_t i = offset; i < offset+node_size; ++i) { |
| 438 | invariant(!touched->at(i)); |
| 439 | touched->at(i) = true; |
| 440 | } |
| 441 | |
| 442 | const uint32_t leftweight = this->nweight(node.left); |
| 443 | const uint32_t rightweight = this->nweight(node.right); |
| 444 | |
| 445 | invariant(leftweight + rightweight + 1 == this->nweight(subtree)); |
| 446 | verify_internal(node.left, touched); |
| 447 | verify_internal(node.right, touched); |
| 448 | } |
| 449 | |
| 450 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 451 | template<typename iterate_extra_t, |
| 452 | int (*f)(const uint32_t, dmtdata_t *, const uint32_t, iterate_extra_t *const)> |
| 453 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_ptr(iterate_extra_t *const ) { |
| 454 | if (this->is_array) { |
| 455 | this->iterate_ptr_internal_array<iterate_extra_t, f>(0, this->size(), iterate_extra); |
| 456 | } else { |
| 457 | this->iterate_ptr_internal<iterate_extra_t, f>(0, this->size(), this->d.t.root, 0, iterate_extra); |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 462 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::fetch(const uint32_t idx, uint32_t *const value_len, dmtdataout_t *const value) const { |
| 463 | if (idx >= this->size()) { return EINVAL; } |
| 464 | if (this->is_array) { |
| 465 | this->fetch_internal_array(idx, value_len, value); |
| 466 | } else { |
| 467 | this->fetch_internal(this->d.t.root, idx, value_len, value); |
| 468 | } |
| 469 | return 0; |
| 470 | } |
| 471 | |
| 472 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 473 | template<typename dmtcmp_t, |
| 474 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 475 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_zero(const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 476 | uint32_t tmp_index; |
| 477 | uint32_t *const child_idxp = (idxp != nullptr) ? idxp : &tmp_index; |
| 478 | int r; |
| 479 | if (this->is_array) { |
| 480 | r = this->find_internal_zero_array<dmtcmp_t, h>(extra, value_len, value, child_idxp); |
| 481 | } |
| 482 | else { |
| 483 | r = this->find_internal_zero<dmtcmp_t, h>(this->d.t.root, extra, value_len, value, child_idxp); |
| 484 | } |
| 485 | return r; |
| 486 | } |
| 487 | |
| 488 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 489 | template<typename dmtcmp_t, |
| 490 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 491 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find(const dmtcmp_t &, int direction, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 492 | uint32_t tmp_index; |
| 493 | uint32_t *const child_idxp = (idxp != nullptr) ? idxp : &tmp_index; |
| 494 | paranoid_invariant(direction != 0); |
| 495 | if (direction < 0) { |
| 496 | if (this->is_array) { |
| 497 | return this->find_internal_minus_array<dmtcmp_t, h>(extra, value_len, value, child_idxp); |
| 498 | } else { |
| 499 | return this->find_internal_minus<dmtcmp_t, h>(this->d.t.root, extra, value_len, value, child_idxp); |
| 500 | } |
| 501 | } else { |
| 502 | if (this->is_array) { |
| 503 | return this->find_internal_plus_array<dmtcmp_t, h>(extra, value_len, value, child_idxp); |
| 504 | } else { |
| 505 | return this->find_internal_plus<dmtcmp_t, h>(this->d.t.root, extra, value_len, value, child_idxp); |
| 506 | } |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 511 | size_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::memory_size(void) { |
| 512 | return (sizeof *this) + toku_mempool_get_size(&this->mp); |
| 513 | } |
| 514 | |
| 515 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 516 | dmt_node_templated<dmtdata_t> & dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_node(const subtree &subtree) const { |
| 517 | paranoid_invariant(!subtree.is_null()); |
| 518 | return get_node(subtree.get_offset()); |
| 519 | } |
| 520 | |
| 521 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 522 | dmt_node_templated<dmtdata_t> & dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_node(const node_offset offset) const { |
| 523 | void* ptr = toku_mempool_get_pointer_from_base_and_offset(&this->mp, offset); |
| 524 | dmt_node *CAST_FROM_VOIDP(node, ptr); |
| 525 | return *node; |
| 526 | } |
| 527 | |
| 528 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 529 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::node_set_value(dmt_node * n, const dmtwriter_t &value) { |
| 530 | n->value_length = value.get_size(); |
| 531 | value.write_to(&n->value); |
| 532 | } |
| 533 | |
| 534 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 535 | node_offset dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::node_malloc_and_set_value(const dmtwriter_t &value) { |
| 536 | size_t val_size = value.get_size(); |
| 537 | size_t size_to_alloc = __builtin_offsetof(dmt_node, value) + val_size; |
| 538 | size_to_alloc = align(size_to_alloc); |
| 539 | void* np = toku_mempool_malloc(&this->mp, size_to_alloc); |
| 540 | paranoid_invariant_notnull(np); |
| 541 | dmt_node *CAST_FROM_VOIDP(n, np); |
| 542 | node_set_value(n, value); |
| 543 | |
| 544 | return toku_mempool_get_offset_from_pointer_and_base(&this->mp, np); |
| 545 | } |
| 546 | |
| 547 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 548 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::node_free(const subtree &st) { |
| 549 | dmt_node &n = get_node(st); |
| 550 | size_t size_to_free = __builtin_offsetof(dmt_node, value) + n.value_length; |
| 551 | size_to_free = align(size_to_free); |
| 552 | toku_mempool_mfree(&this->mp, &n, size_to_free); |
| 553 | } |
| 554 | |
| 555 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 556 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::maybe_resize_tree(const dmtwriter_t * value) { |
| 557 | const ssize_t curr_capacity = toku_mempool_get_size(&this->mp); |
| 558 | const ssize_t curr_free = toku_mempool_get_free_size(&this->mp); |
| 559 | const ssize_t curr_used = toku_mempool_get_used_size(&this->mp); |
| 560 | ssize_t add_size = 0; |
| 561 | if (value) { |
| 562 | add_size = __builtin_offsetof(dmt_node, value) + value->get_size(); |
| 563 | add_size = align(add_size); |
| 564 | } |
| 565 | |
| 566 | const ssize_t need_size = curr_used + add_size; |
| 567 | paranoid_invariant(need_size <= UINT32_MAX); |
| 568 | //TODO(leif) consider different growth rates |
| 569 | const ssize_t new_size = 2*need_size; |
| 570 | paranoid_invariant(new_size <= UINT32_MAX); |
| 571 | |
| 572 | if ((curr_capacity / 2 >= new_size) || // Way too much allocated |
| 573 | (curr_free < add_size)) { // No room in mempool |
| 574 | // Copy all memory and reconstruct dmt in new mempool. |
| 575 | if (curr_free < add_size && toku_mempool_get_frag_size(&this->mp) == 0) { |
| 576 | // TODO(yoni) or TODO(leif) consider doing this not just when frag size is zero, but also when it is a small percentage of the total mempool size |
| 577 | // Offsets remain the same in the new mempool so we can just realloc. |
| 578 | toku_mempool_realloc_larger(&this->mp, new_size); |
| 579 | } else if (!this->d.t.root.is_null()) { |
| 580 | struct mempool new_kvspace; |
| 581 | toku_mempool_construct(&new_kvspace, new_size); |
| 582 | |
| 583 | const dmt_node &n = get_node(this->d.t.root); |
| 584 | node_offset *tmp_array; |
| 585 | bool malloced = false; |
| 586 | tmp_array = alloc_temp_node_offsets(n.weight); |
| 587 | if (!tmp_array) { |
| 588 | malloced = true; |
| 589 | XMALLOC_N(n.weight, tmp_array); |
| 590 | } |
| 591 | this->fill_array_with_subtree_offsets(tmp_array, this->d.t.root); |
| 592 | for (node_offset i = 0; i < n.weight; i++) { |
| 593 | dmt_node &node = get_node(tmp_array[i]); |
| 594 | const size_t bytes_to_copy = __builtin_offsetof(dmt_node, value) + node.value_length; |
| 595 | const size_t bytes_to_alloc = align(bytes_to_copy); |
| 596 | void* newdata = toku_mempool_malloc(&new_kvspace, bytes_to_alloc); |
| 597 | memcpy(newdata, &node, bytes_to_copy); |
| 598 | tmp_array[i] = toku_mempool_get_offset_from_pointer_and_base(&new_kvspace, newdata); |
| 599 | } |
| 600 | |
| 601 | struct mempool old_kvspace = this->mp; |
| 602 | this->mp = new_kvspace; |
| 603 | this->rebuild_subtree_from_offsets(&this->d.t.root, tmp_array, n.weight); |
| 604 | if (malloced) toku_free(tmp_array); |
| 605 | toku_mempool_destroy(&old_kvspace); |
| 606 | } else { |
| 607 | toku_mempool_destroy(&this->mp); |
| 608 | toku_mempool_construct(&this->mp, new_size); |
| 609 | } |
| 610 | } |
| 611 | } |
| 612 | |
| 613 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 614 | bool dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::will_need_rebalance(const subtree &subtree, const int leftmod, const int rightmod) const { |
| 615 | if (subtree.is_null()) { return false; } |
| 616 | const dmt_node &n = get_node(subtree); |
| 617 | // one of the 1's is for the root. |
| 618 | // the other is to take ceil(n/2) |
| 619 | const uint32_t weight_left = this->nweight(n.left) + leftmod; |
| 620 | const uint32_t weight_right = this->nweight(n.right) + rightmod; |
| 621 | return ((1+weight_left < (1+1+weight_right)/2) |
| 622 | || |
| 623 | (1+weight_right < (1+1+weight_left)/2)); |
| 624 | } |
| 625 | |
| 626 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 627 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::insert_internal(subtree *const subtreep, const dmtwriter_t &value, const uint32_t idx, subtree **const rebalance_subtree) { |
| 628 | if (subtreep->is_null()) { |
| 629 | paranoid_invariant_zero(idx); |
| 630 | const node_offset newoffset = this->node_malloc_and_set_value(value); |
| 631 | dmt_node &newnode = get_node(newoffset); |
| 632 | newnode.weight = 1; |
| 633 | newnode.left.set_to_null(); |
| 634 | newnode.right.set_to_null(); |
| 635 | subtreep->set_offset(newoffset); |
| 636 | } else { |
| 637 | dmt_node &n = get_node(*subtreep); |
| 638 | n.weight++; |
| 639 | if (idx <= this->nweight(n.left)) { |
| 640 | if (*rebalance_subtree == nullptr && this->will_need_rebalance(*subtreep, 1, 0)) { |
| 641 | *rebalance_subtree = subtreep; |
| 642 | } |
| 643 | this->insert_internal(&n.left, value, idx, rebalance_subtree); |
| 644 | } else { |
| 645 | if (*rebalance_subtree == nullptr && this->will_need_rebalance(*subtreep, 0, 1)) { |
| 646 | *rebalance_subtree = subtreep; |
| 647 | } |
| 648 | const uint32_t sub_index = idx - this->nweight(n.left) - 1; |
| 649 | this->insert_internal(&n.right, value, sub_index, rebalance_subtree); |
| 650 | } |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 655 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::delete_internal(subtree *const subtreep, const uint32_t idx, subtree *const subtree_replace, subtree **const rebalance_subtree) { |
| 656 | paranoid_invariant_notnull(subtreep); |
| 657 | paranoid_invariant_notnull(rebalance_subtree); |
| 658 | paranoid_invariant(!subtreep->is_null()); |
| 659 | dmt_node &n = get_node(*subtreep); |
| 660 | const uint32_t leftweight = this->nweight(n.left); |
| 661 | if (idx < leftweight) { |
| 662 | n.weight--; |
| 663 | if (*rebalance_subtree == nullptr && this->will_need_rebalance(*subtreep, -1, 0)) { |
| 664 | *rebalance_subtree = subtreep; |
| 665 | } |
| 666 | this->delete_internal(&n.left, idx, subtree_replace, rebalance_subtree); |
| 667 | } else if (idx == leftweight) { |
| 668 | // Found the correct index. |
| 669 | if (n.left.is_null()) { |
| 670 | paranoid_invariant_zero(idx); |
| 671 | // Delete n and let parent point to n.right |
| 672 | subtree ptr_this = *subtreep; |
| 673 | *subtreep = n.right; |
| 674 | subtree to_free; |
| 675 | if (subtree_replace != nullptr) { |
| 676 | // Swap self with the other node. Taking over all responsibility. |
| 677 | to_free = *subtree_replace; |
| 678 | dmt_node &ancestor = get_node(*subtree_replace); |
| 679 | if (*rebalance_subtree == &ancestor.right) { |
| 680 | // Take over rebalance responsibility. |
| 681 | *rebalance_subtree = &n.right; |
| 682 | } |
| 683 | n.weight = ancestor.weight; |
| 684 | n.left = ancestor.left; |
| 685 | n.right = ancestor.right; |
| 686 | *subtree_replace = ptr_this; |
| 687 | } else { |
| 688 | to_free = ptr_this; |
| 689 | } |
| 690 | this->node_free(to_free); |
| 691 | } else if (n.right.is_null()) { |
| 692 | // Delete n and let parent point to n.left |
| 693 | subtree to_free = *subtreep; |
| 694 | *subtreep = n.left; |
| 695 | paranoid_invariant(idx>0); |
| 696 | paranoid_invariant_null(subtree_replace); // To be recursive, we're looking for index 0. n is index > 0 here. |
| 697 | this->node_free(to_free); |
| 698 | } else { |
| 699 | if (*rebalance_subtree == nullptr && this->will_need_rebalance(*subtreep, 0, -1)) { |
| 700 | *rebalance_subtree = subtreep; |
| 701 | } |
| 702 | // don't need to copy up value, it's only used by this |
| 703 | // next call, and when that gets to the bottom there |
| 704 | // won't be any more recursion |
| 705 | n.weight--; |
| 706 | this->delete_internal(&n.right, 0, subtreep, rebalance_subtree); |
| 707 | } |
| 708 | } else { |
| 709 | n.weight--; |
| 710 | if (*rebalance_subtree == nullptr && this->will_need_rebalance(*subtreep, 0, -1)) { |
| 711 | *rebalance_subtree = subtreep; |
| 712 | } |
| 713 | this->delete_internal(&n.right, idx - leftweight - 1, subtree_replace, rebalance_subtree); |
| 714 | } |
| 715 | } |
| 716 | |
| 717 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 718 | template<typename iterate_extra_t, |
| 719 | int (*f)(const uint32_t, const dmtdata_t &, const uint32_t, iterate_extra_t *const)> |
| 720 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_internal_array(const uint32_t left, const uint32_t right, |
| 721 | iterate_extra_t *const ) const { |
| 722 | int r; |
| 723 | for (uint32_t i = left; i < right; ++i) { |
| 724 | r = f(this->value_length, *get_array_value(i), i, iterate_extra); |
| 725 | if (r != 0) { |
| 726 | return r; |
| 727 | } |
| 728 | } |
| 729 | return 0; |
| 730 | } |
| 731 | |
| 732 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 733 | template<typename iterate_extra_t, |
| 734 | int (*f)(const uint32_t, dmtdata_t *, const uint32_t, iterate_extra_t *const)> |
| 735 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_ptr_internal(const uint32_t left, const uint32_t right, |
| 736 | const subtree &subtree, const uint32_t idx, |
| 737 | iterate_extra_t *const ) { |
| 738 | if (!subtree.is_null()) { |
| 739 | dmt_node &n = get_node(subtree); |
| 740 | const uint32_t idx_root = idx + this->nweight(n.left); |
| 741 | if (left < idx_root) { |
| 742 | this->iterate_ptr_internal<iterate_extra_t, f>(left, right, n.left, idx, iterate_extra); |
| 743 | } |
| 744 | if (left <= idx_root && idx_root < right) { |
| 745 | int r = f(n.value_length, &n.value, idx_root, iterate_extra); |
| 746 | lazy_assert_zero(r); |
| 747 | } |
| 748 | if (idx_root + 1 < right) { |
| 749 | this->iterate_ptr_internal<iterate_extra_t, f>(left, right, n.right, idx_root + 1, iterate_extra); |
| 750 | } |
| 751 | } |
| 752 | } |
| 753 | |
| 754 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 755 | template<typename iterate_extra_t, |
| 756 | int (*f)(const uint32_t, dmtdata_t *, const uint32_t, iterate_extra_t *const)> |
| 757 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_ptr_internal_array(const uint32_t left, const uint32_t right, |
| 758 | iterate_extra_t *const ) { |
| 759 | for (uint32_t i = left; i < right; ++i) { |
| 760 | int r = f(this->value_length, get_array_value(i), i, iterate_extra); |
| 761 | lazy_assert_zero(r); |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 766 | template<typename iterate_extra_t, |
| 767 | int (*f)(const uint32_t, const dmtdata_t &, const uint32_t, iterate_extra_t *const)> |
| 768 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::iterate_internal(const uint32_t left, const uint32_t right, |
| 769 | const subtree &subtree, const uint32_t idx, |
| 770 | iterate_extra_t *const ) const { |
| 771 | if (subtree.is_null()) { return 0; } |
| 772 | int r; |
| 773 | const dmt_node &n = get_node(subtree); |
| 774 | const uint32_t idx_root = idx + this->nweight(n.left); |
| 775 | if (left < idx_root) { |
| 776 | r = this->iterate_internal<iterate_extra_t, f>(left, right, n.left, idx, iterate_extra); |
| 777 | if (r != 0) { return r; } |
| 778 | } |
| 779 | if (left <= idx_root && idx_root < right) { |
| 780 | r = f(n.value_length, n.value, idx_root, iterate_extra); |
| 781 | if (r != 0) { return r; } |
| 782 | } |
| 783 | if (idx_root + 1 < right) { |
| 784 | return this->iterate_internal<iterate_extra_t, f>(left, right, n.right, idx_root + 1, iterate_extra); |
| 785 | } |
| 786 | return 0; |
| 787 | } |
| 788 | |
| 789 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 790 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::fetch_internal_array(const uint32_t i, uint32_t *const value_len, dmtdataout_t *const value) const { |
| 791 | copyout(value_len, value, this->value_length, get_array_value(i)); |
| 792 | } |
| 793 | |
| 794 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 795 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::fetch_internal(const subtree &subtree, const uint32_t i, uint32_t *const value_len, dmtdataout_t *const value) const { |
| 796 | dmt_node &n = get_node(subtree); |
| 797 | const uint32_t leftweight = this->nweight(n.left); |
| 798 | if (i < leftweight) { |
| 799 | this->fetch_internal(n.left, i, value_len, value); |
| 800 | } else if (i == leftweight) { |
| 801 | copyout(value_len, value, &n); |
| 802 | } else { |
| 803 | this->fetch_internal(n.right, i - leftweight - 1, value_len, value); |
| 804 | } |
| 805 | } |
| 806 | |
| 807 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 808 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::fill_array_with_subtree_offsets(node_offset *const array, const subtree &subtree) const { |
| 809 | if (!subtree.is_null()) { |
| 810 | const dmt_node &tree = get_node(subtree); |
| 811 | this->fill_array_with_subtree_offsets(&array[0], tree.left); |
| 812 | array[this->nweight(tree.left)] = subtree.get_offset(); |
| 813 | this->fill_array_with_subtree_offsets(&array[this->nweight(tree.left) + 1], tree.right); |
| 814 | } |
| 815 | } |
| 816 | |
| 817 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 818 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::rebuild_subtree_from_offsets(subtree *const subtree, const node_offset *const offsets, const uint32_t numvalues) { |
| 819 | if (numvalues==0) { |
| 820 | subtree->set_to_null(); |
| 821 | } else { |
| 822 | uint32_t halfway = numvalues/2; |
| 823 | subtree->set_offset(offsets[halfway]); |
| 824 | dmt_node &newnode = get_node(offsets[halfway]); |
| 825 | newnode.weight = numvalues; |
| 826 | // value is already in there. |
| 827 | this->rebuild_subtree_from_offsets(&newnode.left, &offsets[0], halfway); |
| 828 | this->rebuild_subtree_from_offsets(&newnode.right, &offsets[halfway+1], numvalues-(halfway+1)); |
| 829 | } |
| 830 | } |
| 831 | |
| 832 | //TODO(leif): Note that this can mess with our memory_footprint calculation (we may touch past what is marked as 'used' in the mempool) |
| 833 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 834 | node_offset* dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::alloc_temp_node_offsets(uint32_t num_offsets) { |
| 835 | size_t mem_needed = num_offsets * sizeof(node_offset); |
| 836 | size_t mem_free; |
| 837 | mem_free = toku_mempool_get_free_size(&this->mp); |
| 838 | node_offset* CAST_FROM_VOIDP(tmp, toku_mempool_get_next_free_ptr(&this->mp)); |
| 839 | if (mem_free >= mem_needed) { |
| 840 | return tmp; |
| 841 | } |
| 842 | return nullptr; |
| 843 | } |
| 844 | |
| 845 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 846 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::rebalance(subtree *const subtree) { |
| 847 | paranoid_invariant(!subtree->is_null()); |
| 848 | |
| 849 | // There is a possible "optimization" here: |
| 850 | // if (this->values_same_size && subtree == &this->d.t.root) { |
| 851 | // this->convert_from_tree_to_array(); |
| 852 | // return; |
| 853 | // } |
| 854 | // but we don't want to do it because it involves actually copying values around |
| 855 | // as opposed to stopping in the middle of rebalancing (like in the OMT) |
| 856 | |
| 857 | node_offset offset = subtree->get_offset(); |
| 858 | const dmt_node &n = get_node(offset); |
| 859 | node_offset *tmp_array; |
| 860 | bool malloced = false; |
| 861 | tmp_array = alloc_temp_node_offsets(n.weight); |
| 862 | if (!tmp_array) { |
| 863 | malloced = true; |
| 864 | XMALLOC_N(n.weight, tmp_array); |
| 865 | } |
| 866 | this->fill_array_with_subtree_offsets(tmp_array, *subtree); |
| 867 | this->rebuild_subtree_from_offsets(subtree, tmp_array, n.weight); |
| 868 | if (malloced) toku_free(tmp_array); |
| 869 | } |
| 870 | |
| 871 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 872 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::copyout(uint32_t *const outlen, dmtdata_t *const out, const dmt_node *const n) { |
| 873 | if (outlen) { |
| 874 | *outlen = n->value_length; |
| 875 | } |
| 876 | if (out) { |
| 877 | *out = n->value; |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 882 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::copyout(uint32_t *const outlen, dmtdata_t **const out, dmt_node *const n) { |
| 883 | if (outlen) { |
| 884 | *outlen = n->value_length; |
| 885 | } |
| 886 | if (out) { |
| 887 | *out = &n->value; |
| 888 | } |
| 889 | } |
| 890 | |
| 891 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 892 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::copyout(uint32_t *const outlen, dmtdata_t *const out, const uint32_t len, const dmtdata_t *const stored_value_ptr) { |
| 893 | if (outlen) { |
| 894 | *outlen = len; |
| 895 | } |
| 896 | if (out) { |
| 897 | *out = *stored_value_ptr; |
| 898 | } |
| 899 | } |
| 900 | |
| 901 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 902 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::copyout(uint32_t *const outlen, dmtdata_t **const out, const uint32_t len, dmtdata_t *const stored_value_ptr) { |
| 903 | if (outlen) { |
| 904 | *outlen = len; |
| 905 | } |
| 906 | if (out) { |
| 907 | *out = stored_value_ptr; |
| 908 | } |
| 909 | } |
| 910 | |
| 911 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 912 | template<typename dmtcmp_t, |
| 913 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 914 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_zero_array(const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 915 | paranoid_invariant_notnull(idxp); |
| 916 | uint32_t min = 0; |
| 917 | uint32_t limit = this->d.a.num_values; |
| 918 | uint32_t best_pos = subtree::NODE_NULL; |
| 919 | uint32_t best_zero = subtree::NODE_NULL; |
| 920 | |
| 921 | while (min!=limit) { |
| 922 | uint32_t mid = (min + limit) / 2; |
| 923 | int hv = h(this->value_length, *get_array_value(mid), extra); |
| 924 | if (hv<0) { |
| 925 | min = mid+1; |
| 926 | } |
| 927 | else if (hv>0) { |
| 928 | best_pos = mid; |
| 929 | limit = mid; |
| 930 | } |
| 931 | else { |
| 932 | best_zero = mid; |
| 933 | limit = mid; |
| 934 | } |
| 935 | } |
| 936 | if (best_zero!=subtree::NODE_NULL) { |
| 937 | //Found a zero |
| 938 | copyout(value_len, value, this->value_length, get_array_value(best_zero)); |
| 939 | *idxp = best_zero; |
| 940 | return 0; |
| 941 | } |
| 942 | if (best_pos!=subtree::NODE_NULL) *idxp = best_pos; |
| 943 | else *idxp = this->d.a.num_values; |
| 944 | return DB_NOTFOUND; |
| 945 | } |
| 946 | |
| 947 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 948 | template<typename dmtcmp_t, |
| 949 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 950 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_zero(const subtree &subtree, const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 951 | paranoid_invariant_notnull(idxp); |
| 952 | if (subtree.is_null()) { |
| 953 | *idxp = 0; |
| 954 | return DB_NOTFOUND; |
| 955 | } |
| 956 | dmt_node &n = get_node(subtree); |
| 957 | int hv = h(n.value_length, n.value, extra); |
| 958 | if (hv<0) { |
| 959 | int r = this->find_internal_zero<dmtcmp_t, h>(n.right, extra, value_len, value, idxp); |
| 960 | *idxp += this->nweight(n.left)+1; |
| 961 | return r; |
| 962 | } else if (hv>0) { |
| 963 | return this->find_internal_zero<dmtcmp_t, h>(n.left, extra, value_len, value, idxp); |
| 964 | } else { |
| 965 | int r = this->find_internal_zero<dmtcmp_t, h>(n.left, extra, value_len, value, idxp); |
| 966 | if (r==DB_NOTFOUND) { |
| 967 | *idxp = this->nweight(n.left); |
| 968 | copyout(value_len, value, &n); |
| 969 | r = 0; |
| 970 | } |
| 971 | return r; |
| 972 | } |
| 973 | } |
| 974 | |
| 975 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 976 | template<typename dmtcmp_t, |
| 977 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 978 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_plus_array(const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 979 | paranoid_invariant_notnull(idxp); |
| 980 | uint32_t min = 0; |
| 981 | uint32_t limit = this->d.a.num_values; |
| 982 | uint32_t best = subtree::NODE_NULL; |
| 983 | |
| 984 | while (min != limit) { |
| 985 | const uint32_t mid = (min + limit) / 2; |
| 986 | const int hv = h(this->value_length, *get_array_value(mid), extra); |
| 987 | if (hv > 0) { |
| 988 | best = mid; |
| 989 | limit = mid; |
| 990 | } else { |
| 991 | min = mid + 1; |
| 992 | } |
| 993 | } |
| 994 | if (best == subtree::NODE_NULL) { return DB_NOTFOUND; } |
| 995 | copyout(value_len, value, this->value_length, get_array_value(best)); |
| 996 | *idxp = best; |
| 997 | return 0; |
| 998 | } |
| 999 | |
| 1000 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1001 | template<typename dmtcmp_t, |
| 1002 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 1003 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_plus(const subtree &subtree, const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 1004 | paranoid_invariant_notnull(idxp); |
| 1005 | if (subtree.is_null()) { |
| 1006 | return DB_NOTFOUND; |
| 1007 | } |
| 1008 | dmt_node & n = get_node(subtree); |
| 1009 | int hv = h(n.value_length, n.value, extra); |
| 1010 | int r; |
| 1011 | if (hv > 0) { |
| 1012 | r = this->find_internal_plus<dmtcmp_t, h>(n.left, extra, value_len, value, idxp); |
| 1013 | if (r == DB_NOTFOUND) { |
| 1014 | *idxp = this->nweight(n.left); |
| 1015 | copyout(value_len, value, &n); |
| 1016 | r = 0; |
| 1017 | } |
| 1018 | } else { |
| 1019 | r = this->find_internal_plus<dmtcmp_t, h>(n.right, extra, value_len, value, idxp); |
| 1020 | if (r == 0) { |
| 1021 | *idxp += this->nweight(n.left) + 1; |
| 1022 | } |
| 1023 | } |
| 1024 | return r; |
| 1025 | } |
| 1026 | |
| 1027 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1028 | template<typename dmtcmp_t, |
| 1029 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 1030 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_minus_array(const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 1031 | paranoid_invariant_notnull(idxp); |
| 1032 | uint32_t min = 0; |
| 1033 | uint32_t limit = this->d.a.num_values; |
| 1034 | uint32_t best = subtree::NODE_NULL; |
| 1035 | |
| 1036 | while (min != limit) { |
| 1037 | const uint32_t mid = (min + limit) / 2; |
| 1038 | const int hv = h(this->value_length, *get_array_value(mid), extra); |
| 1039 | if (hv < 0) { |
| 1040 | best = mid; |
| 1041 | min = mid + 1; |
| 1042 | } else { |
| 1043 | limit = mid; |
| 1044 | } |
| 1045 | } |
| 1046 | if (best == subtree::NODE_NULL) { return DB_NOTFOUND; } |
| 1047 | copyout(value_len, value, this->value_length, get_array_value(best)); |
| 1048 | *idxp = best; |
| 1049 | return 0; |
| 1050 | } |
| 1051 | |
| 1052 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1053 | template<typename dmtcmp_t, |
| 1054 | int (*h)(const uint32_t, const dmtdata_t &, const dmtcmp_t &)> |
| 1055 | int dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::find_internal_minus(const subtree &subtree, const dmtcmp_t &, uint32_t *const value_len, dmtdataout_t *const value, uint32_t *const idxp) const { |
| 1056 | paranoid_invariant_notnull(idxp); |
| 1057 | if (subtree.is_null()) { |
| 1058 | return DB_NOTFOUND; |
| 1059 | } |
| 1060 | dmt_node & n = get_node(subtree); |
| 1061 | int hv = h(n.value_length, n.value, extra); |
| 1062 | if (hv < 0) { |
| 1063 | int r = this->find_internal_minus<dmtcmp_t, h>(n.right, extra, value_len, value, idxp); |
| 1064 | if (r == 0) { |
| 1065 | *idxp += this->nweight(n.left) + 1; |
| 1066 | } else if (r == DB_NOTFOUND) { |
| 1067 | *idxp = this->nweight(n.left); |
| 1068 | copyout(value_len, value, &n); |
| 1069 | r = 0; |
| 1070 | } |
| 1071 | return r; |
| 1072 | } else { |
| 1073 | return this->find_internal_minus<dmtcmp_t, h>(n.left, extra, value_len, value, idxp); |
| 1074 | } |
| 1075 | } |
| 1076 | |
| 1077 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1078 | uint32_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_fixed_length(void) const { |
| 1079 | return this->values_same_size ? this->value_length : 0; |
| 1080 | } |
| 1081 | |
| 1082 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1083 | uint32_t dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::get_fixed_length_alignment_overhead(void) const { |
| 1084 | return this->values_same_size ? align(this->value_length) - this->value_length : 0; |
| 1085 | } |
| 1086 | |
| 1087 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1088 | bool dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::value_length_is_fixed(void) const { |
| 1089 | return this->values_same_size; |
| 1090 | } |
| 1091 | |
| 1092 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1093 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::serialize_values(uint32_t expected_unpadded_memory, struct wbuf *wb) const { |
| 1094 | invariant(this->is_array); |
| 1095 | invariant(this->values_same_size); |
| 1096 | const uint8_t pad_bytes = get_fixed_length_alignment_overhead(); |
| 1097 | const uint32_t fixed_len = this->value_length; |
| 1098 | const uint32_t fixed_aligned_len = align(this->value_length); |
| 1099 | paranoid_invariant(expected_unpadded_memory == this->d.a.num_values * this->value_length); |
| 1100 | paranoid_invariant(toku_mempool_get_used_size(&this->mp) >= |
| 1101 | expected_unpadded_memory + pad_bytes * this->d.a.num_values); |
| 1102 | if (this->d.a.num_values == 0) { |
| 1103 | // Nothing to serialize |
| 1104 | } else if (pad_bytes == 0) { |
| 1105 | // Basically a memcpy |
| 1106 | wbuf_nocrc_literal_bytes(wb, get_array_value(0), expected_unpadded_memory); |
| 1107 | } else { |
| 1108 | uint8_t* const dest = wbuf_nocrc_reserve_literal_bytes(wb, expected_unpadded_memory); |
| 1109 | const uint8_t* const src = reinterpret_cast<uint8_t*>(get_array_value(0)); |
| 1110 | //TODO(leif) maybe look at vectorization here |
| 1111 | for (uint32_t i = 0; i < this->d.a.num_values; i++) { |
| 1112 | memcpy(&dest[i*fixed_len], &src[i*fixed_aligned_len], fixed_len); |
| 1113 | } |
| 1114 | } |
| 1115 | } |
| 1116 | |
| 1117 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1118 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::builder::create(uint32_t _max_values, uint32_t _max_value_bytes) { |
| 1119 | this->max_values = _max_values; |
| 1120 | this->max_value_bytes = _max_value_bytes; |
| 1121 | this->temp.create(); |
| 1122 | paranoid_invariant_null(toku_mempool_get_base(&this->temp.mp)); |
| 1123 | this->temp_valid = true; |
| 1124 | this->sorted_node_offsets = nullptr; |
| 1125 | // Include enough space for alignment padding |
| 1126 | size_t initial_space = (ALIGNMENT - 1) * _max_values + _max_value_bytes; |
| 1127 | |
| 1128 | toku_mempool_construct(&this->temp.mp, initial_space); // Adds 25% |
| 1129 | } |
| 1130 | |
| 1131 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1132 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::builder::append(const dmtwriter_t &value) { |
| 1133 | paranoid_invariant(this->temp_valid); |
| 1134 | //NOTE: Always use d.a.num_values for size because we have not yet created root. |
| 1135 | if (this->temp.values_same_size && (this->temp.d.a.num_values == 0 || value.get_size() == this->temp.value_length)) { |
| 1136 | temp.insert_at_array_end<false>(value); |
| 1137 | return; |
| 1138 | } |
| 1139 | if (this->temp.is_array) { |
| 1140 | // Convert to tree format (without weights and linkage) |
| 1141 | XMALLOC_N(this->max_values, this->sorted_node_offsets); |
| 1142 | |
| 1143 | // Include enough space for alignment padding |
| 1144 | size_t mem_needed = (ALIGNMENT - 1 + __builtin_offsetof(dmt_node, value)) * max_values + max_value_bytes; |
| 1145 | struct mempool old_mp = this->temp.mp; |
| 1146 | |
| 1147 | const uint32_t num_values = this->temp.d.a.num_values; |
| 1148 | toku_mempool_construct(&this->temp.mp, mem_needed); |
| 1149 | |
| 1150 | // Copy over and get node_offsets |
| 1151 | for (uint32_t i = 0; i < num_values; i++) { |
| 1152 | dmtwriter_t writer(this->temp.value_length, this->temp.get_array_value_internal(&old_mp, i)); |
| 1153 | this->sorted_node_offsets[i] = this->temp.node_malloc_and_set_value(writer); |
| 1154 | } |
| 1155 | this->temp.is_array = false; |
| 1156 | this->temp.values_same_size = false; |
| 1157 | this->temp.value_length = 0; |
| 1158 | toku_mempool_destroy(&old_mp); |
| 1159 | } |
| 1160 | paranoid_invariant(!this->temp.is_array); |
| 1161 | this->sorted_node_offsets[this->temp.d.a.num_values++] = this->temp.node_malloc_and_set_value(value); |
| 1162 | } |
| 1163 | |
| 1164 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1165 | bool dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::builder::value_length_is_fixed(void) { |
| 1166 | paranoid_invariant(this->temp_valid); |
| 1167 | return this->temp.values_same_size; |
| 1168 | } |
| 1169 | |
| 1170 | template<typename dmtdata_t, typename dmtdataout_t, typename dmtwriter_t> |
| 1171 | void dmt<dmtdata_t, dmtdataout_t, dmtwriter_t>::builder::build(dmt<dmtdata_t, dmtdataout_t, dmtwriter_t> *dest) { |
| 1172 | invariant(this->temp_valid); |
| 1173 | //NOTE: Always use d.a.num_values for size because we have not yet created root. |
| 1174 | invariant(this->temp.d.a.num_values <= this->max_values); |
| 1175 | // Memory invariant is taken care of incrementally (during append()) |
| 1176 | |
| 1177 | if (!this->temp.is_array) { |
| 1178 | invariant_notnull(this->sorted_node_offsets); |
| 1179 | this->temp.rebuild_subtree_from_offsets(&this->temp.d.t.root, this->sorted_node_offsets, this->temp.d.a.num_values); |
| 1180 | toku_free(this->sorted_node_offsets); |
| 1181 | this->sorted_node_offsets = nullptr; |
| 1182 | } |
| 1183 | paranoid_invariant_null(this->sorted_node_offsets); |
| 1184 | |
| 1185 | const size_t used = toku_mempool_get_used_size(&this->temp.mp); |
| 1186 | const size_t allocated = toku_mempool_get_size(&this->temp.mp); |
| 1187 | // We want to use no more than (about) the actual used space + 25% overhead for mempool growth. |
| 1188 | // When we know the elements are fixed-length, we use the better dmt constructor. |
| 1189 | // In practice, as of Jan 2014, we use the builder in two cases: |
| 1190 | // - When we know the elements are not fixed-length. |
| 1191 | // - During upgrade of a pre version 26 basement node. |
| 1192 | // During upgrade, we will probably wildly overallocate because we don't account for the values that aren't stored in the dmt, so here we want to shrink the mempool. |
| 1193 | // When we know the elements are not fixed-length, we still know how much memory they occupy in total, modulo alignment, so we want to allow for mempool overhead and worst-case alignment overhead, and not shrink the mempool. |
| 1194 | const size_t max_allowed = used + (ALIGNMENT-1) * this->temp.size(); |
| 1195 | const size_t max_allowed_with_mempool_overhead = max_allowed + max_allowed / 4; |
| 1196 | //TODO(leif): get footprint calculation correct (under jemalloc) and add some form of footprint constraint |
| 1197 | if (allocated > max_allowed_with_mempool_overhead) { |
| 1198 | // Reallocate smaller mempool to save memory |
| 1199 | invariant_zero(toku_mempool_get_frag_size(&this->temp.mp)); |
| 1200 | struct mempool new_mp; |
| 1201 | toku_mempool_construct(&new_mp, used); |
| 1202 | void * newbase = toku_mempool_malloc(&new_mp, used); |
| 1203 | invariant_notnull(newbase); |
| 1204 | memcpy(newbase, toku_mempool_get_base(&this->temp.mp), used); |
| 1205 | toku_mempool_destroy(&this->temp.mp); |
| 1206 | this->temp.mp = new_mp; |
| 1207 | } |
| 1208 | |
| 1209 | *dest = this->temp; |
| 1210 | this->temp_valid = false; |
| 1211 | |
| 1212 | } |
| 1213 | } // namespace toku |
| 1214 | |