| 1 | /**************************************************************************/ |
| 2 | /* paged_array.h */ |
| 3 | /**************************************************************************/ |
| 4 | /* This file is part of: */ |
| 5 | /* GODOT ENGINE */ |
| 6 | /* https://godotengine.org */ |
| 7 | /**************************************************************************/ |
| 8 | /* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */ |
| 9 | /* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */ |
| 10 | /* */ |
| 11 | /* Permission is hereby granted, free of charge, to any person obtaining */ |
| 12 | /* a copy of this software and associated documentation files (the */ |
| 13 | /* "Software"), to deal in the Software without restriction, including */ |
| 14 | /* without limitation the rights to use, copy, modify, merge, publish, */ |
| 15 | /* distribute, sublicense, and/or sell copies of the Software, and to */ |
| 16 | /* permit persons to whom the Software is furnished to do so, subject to */ |
| 17 | /* the following conditions: */ |
| 18 | /* */ |
| 19 | /* The above copyright notice and this permission notice shall be */ |
| 20 | /* included in all copies or substantial portions of the Software. */ |
| 21 | /* */ |
| 22 | /* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */ |
| 23 | /* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */ |
| 24 | /* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */ |
| 25 | /* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */ |
| 26 | /* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */ |
| 27 | /* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */ |
| 28 | /* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ |
| 29 | /**************************************************************************/ |
| 30 | |
| 31 | #ifndef PAGED_ARRAY_H |
| 32 | #define PAGED_ARRAY_H |
| 33 | |
| 34 | #include "core/os/memory.h" |
| 35 | #include "core/os/spin_lock.h" |
| 36 | #include "core/typedefs.h" |
| 37 | |
| 38 | #include <type_traits> |
| 39 | |
| 40 | // PagedArray is used mainly for filling a very large array from multiple threads efficiently and without causing major fragmentation |
| 41 | |
| 42 | // PageArrayPool manages central page allocation in a thread safe matter |
| 43 | |
| 44 | template <class T> |
| 45 | class PagedArrayPool { |
| 46 | T **page_pool = nullptr; |
| 47 | uint32_t pages_allocated = 0; |
| 48 | |
| 49 | uint32_t *available_page_pool = nullptr; |
| 50 | uint32_t pages_available = 0; |
| 51 | |
| 52 | uint32_t page_size = 0; |
| 53 | SpinLock spin_lock; |
| 54 | |
| 55 | public: |
| 56 | uint32_t alloc_page() { |
| 57 | spin_lock.lock(); |
| 58 | if (unlikely(pages_available == 0)) { |
| 59 | uint32_t pages_used = pages_allocated; |
| 60 | |
| 61 | pages_allocated++; |
| 62 | page_pool = (T **)memrealloc(page_pool, sizeof(T *) * pages_allocated); |
| 63 | available_page_pool = (uint32_t *)memrealloc(available_page_pool, sizeof(uint32_t) * pages_allocated); |
| 64 | |
| 65 | page_pool[pages_used] = (T *)memalloc(sizeof(T) * page_size); |
| 66 | available_page_pool[0] = pages_used; |
| 67 | |
| 68 | pages_available++; |
| 69 | } |
| 70 | |
| 71 | pages_available--; |
| 72 | uint32_t page = available_page_pool[pages_available]; |
| 73 | spin_lock.unlock(); |
| 74 | |
| 75 | return page; |
| 76 | } |
| 77 | T *get_page(uint32_t p_page_id) { |
| 78 | return page_pool[p_page_id]; |
| 79 | } |
| 80 | |
| 81 | void free_page(uint32_t p_page_id) { |
| 82 | spin_lock.lock(); |
| 83 | available_page_pool[pages_available] = p_page_id; |
| 84 | pages_available++; |
| 85 | spin_lock.unlock(); |
| 86 | } |
| 87 | |
| 88 | uint32_t get_page_size_shift() const { |
| 89 | return get_shift_from_power_of_2(page_size); |
| 90 | } |
| 91 | |
| 92 | uint32_t get_page_size_mask() const { |
| 93 | return page_size - 1; |
| 94 | } |
| 95 | |
| 96 | void reset() { |
| 97 | ERR_FAIL_COND(pages_available < pages_allocated); |
| 98 | if (pages_allocated) { |
| 99 | for (uint32_t i = 0; i < pages_allocated; i++) { |
| 100 | memfree(page_pool[i]); |
| 101 | } |
| 102 | memfree(page_pool); |
| 103 | memfree(available_page_pool); |
| 104 | page_pool = nullptr; |
| 105 | available_page_pool = nullptr; |
| 106 | pages_allocated = 0; |
| 107 | pages_available = 0; |
| 108 | } |
| 109 | } |
| 110 | bool is_configured() const { |
| 111 | return page_size > 0; |
| 112 | } |
| 113 | |
| 114 | void configure(uint32_t p_page_size) { |
| 115 | ERR_FAIL_COND(page_pool != nullptr); // Safety check. |
| 116 | ERR_FAIL_COND(p_page_size == 0); |
| 117 | page_size = nearest_power_of_2_templated(p_page_size); |
| 118 | } |
| 119 | |
| 120 | PagedArrayPool(uint32_t p_page_size = 4096) { // power of 2 recommended because of alignment with OS page sizes. Even if element is bigger, its still a multiple and get rounded amount of pages |
| 121 | configure(p_page_size); |
| 122 | } |
| 123 | |
| 124 | ~PagedArrayPool() { |
| 125 | ERR_FAIL_COND_MSG(pages_available < pages_allocated, "Pages in use exist at exit in PagedArrayPool" ); |
| 126 | reset(); |
| 127 | } |
| 128 | }; |
| 129 | |
| 130 | // PageArray is a local array that is optimized to grow in place, then be cleared often. |
| 131 | // It does so by allocating pages from a PagedArrayPool. |
| 132 | // It is safe to use multiple PagedArrays from different threads, sharing a single PagedArrayPool |
| 133 | |
| 134 | template <class T> |
| 135 | class PagedArray { |
| 136 | PagedArrayPool<T> *page_pool = nullptr; |
| 137 | |
| 138 | T **page_data = nullptr; |
| 139 | uint32_t *page_ids = nullptr; |
| 140 | uint32_t max_pages_used = 0; |
| 141 | uint32_t page_size_shift = 0; |
| 142 | uint32_t page_size_mask = 0; |
| 143 | uint64_t count = 0; |
| 144 | |
| 145 | _FORCE_INLINE_ uint32_t _get_pages_in_use() const { |
| 146 | if (count == 0) { |
| 147 | return 0; |
| 148 | } else { |
| 149 | return ((count - 1) >> page_size_shift) + 1; |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | void _grow_page_array() { |
| 154 | //no more room in the page array to put the new page, make room |
| 155 | if (max_pages_used == 0) { |
| 156 | max_pages_used = 1; |
| 157 | } else { |
| 158 | max_pages_used *= 2; // increase in powers of 2 to keep allocations to minimum |
| 159 | } |
| 160 | page_data = (T **)memrealloc(page_data, sizeof(T *) * max_pages_used); |
| 161 | page_ids = (uint32_t *)memrealloc(page_ids, sizeof(uint32_t) * max_pages_used); |
| 162 | } |
| 163 | |
| 164 | public: |
| 165 | _FORCE_INLINE_ const T &operator[](uint64_t p_index) const { |
| 166 | CRASH_BAD_UNSIGNED_INDEX(p_index, count); |
| 167 | uint32_t page = p_index >> page_size_shift; |
| 168 | uint32_t offset = p_index & page_size_mask; |
| 169 | |
| 170 | return page_data[page][offset]; |
| 171 | } |
| 172 | _FORCE_INLINE_ T &operator[](uint64_t p_index) { |
| 173 | CRASH_BAD_UNSIGNED_INDEX(p_index, count); |
| 174 | uint32_t page = p_index >> page_size_shift; |
| 175 | uint32_t offset = p_index & page_size_mask; |
| 176 | |
| 177 | return page_data[page][offset]; |
| 178 | } |
| 179 | |
| 180 | _FORCE_INLINE_ void push_back(const T &p_value) { |
| 181 | uint32_t remainder = count & page_size_mask; |
| 182 | if (unlikely(remainder == 0)) { |
| 183 | // at 0, so time to request a new page |
| 184 | uint32_t page_count = _get_pages_in_use(); |
| 185 | uint32_t new_page_count = page_count + 1; |
| 186 | |
| 187 | if (unlikely(new_page_count > max_pages_used)) { |
| 188 | ERR_FAIL_NULL(page_pool); // Safety check. |
| 189 | |
| 190 | _grow_page_array(); //keep out of inline |
| 191 | } |
| 192 | |
| 193 | uint32_t page_id = page_pool->alloc_page(); |
| 194 | page_data[page_count] = page_pool->get_page(page_id); |
| 195 | page_ids[page_count] = page_id; |
| 196 | } |
| 197 | |
| 198 | // place the new value |
| 199 | uint32_t page = count >> page_size_shift; |
| 200 | uint32_t offset = count & page_size_mask; |
| 201 | |
| 202 | if (!std::is_trivially_constructible<T>::value) { |
| 203 | memnew_placement(&page_data[page][offset], T(p_value)); |
| 204 | } else { |
| 205 | page_data[page][offset] = p_value; |
| 206 | } |
| 207 | |
| 208 | count++; |
| 209 | } |
| 210 | |
| 211 | _FORCE_INLINE_ void pop_back() { |
| 212 | ERR_FAIL_COND(count == 0); |
| 213 | |
| 214 | if (!std::is_trivially_destructible<T>::value) { |
| 215 | uint32_t page = (count - 1) >> page_size_shift; |
| 216 | uint32_t offset = (count - 1) & page_size_mask; |
| 217 | page_data[page][offset].~T(); |
| 218 | } |
| 219 | |
| 220 | uint32_t remainder = count & page_size_mask; |
| 221 | if (unlikely(remainder == 1)) { |
| 222 | // one element remained, so page must be freed. |
| 223 | uint32_t last_page = _get_pages_in_use() - 1; |
| 224 | page_pool->free_page(page_ids[last_page]); |
| 225 | } |
| 226 | count--; |
| 227 | } |
| 228 | |
| 229 | void clear() { |
| 230 | //destruct if needed |
| 231 | if (!std::is_trivially_destructible<T>::value) { |
| 232 | for (uint64_t i = 0; i < count; i++) { |
| 233 | uint32_t page = i >> page_size_shift; |
| 234 | uint32_t offset = i & page_size_mask; |
| 235 | page_data[page][offset].~T(); |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | //return the pages to the pagepool, so they can be used by another array eventually |
| 240 | uint32_t pages_used = _get_pages_in_use(); |
| 241 | for (uint32_t i = 0; i < pages_used; i++) { |
| 242 | page_pool->free_page(page_ids[i]); |
| 243 | } |
| 244 | |
| 245 | count = 0; |
| 246 | |
| 247 | //note we leave page_data and page_indices intact for next use. If you really want to clear them call reset() |
| 248 | } |
| 249 | |
| 250 | void reset() { |
| 251 | clear(); |
| 252 | if (page_data) { |
| 253 | memfree(page_data); |
| 254 | memfree(page_ids); |
| 255 | page_data = nullptr; |
| 256 | page_ids = nullptr; |
| 257 | max_pages_used = 0; |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | // This takes the pages from a source array and merges them to this one |
| 262 | // resulting order is undefined, but content is merged very efficiently, |
| 263 | // making it ideal to fill content on several threads to later join it. |
| 264 | |
| 265 | void merge_unordered(PagedArray<T> &p_array) { |
| 266 | ERR_FAIL_COND(page_pool != p_array.page_pool); |
| 267 | |
| 268 | uint32_t remainder = count & page_size_mask; |
| 269 | |
| 270 | T *remainder_page = nullptr; |
| 271 | uint32_t remainder_page_id = 0; |
| 272 | |
| 273 | if (remainder > 0) { |
| 274 | uint32_t last_page = _get_pages_in_use() - 1; |
| 275 | remainder_page = page_data[last_page]; |
| 276 | remainder_page_id = page_ids[last_page]; |
| 277 | } |
| 278 | |
| 279 | count -= remainder; |
| 280 | |
| 281 | uint32_t src_page_index = 0; |
| 282 | uint32_t page_size = page_size_mask + 1; |
| 283 | |
| 284 | while (p_array.count > 0) { |
| 285 | uint32_t page_count = _get_pages_in_use(); |
| 286 | uint32_t new_page_count = page_count + 1; |
| 287 | |
| 288 | if (unlikely(new_page_count > max_pages_used)) { |
| 289 | _grow_page_array(); //keep out of inline |
| 290 | } |
| 291 | |
| 292 | page_data[page_count] = p_array.page_data[src_page_index]; |
| 293 | page_ids[page_count] = p_array.page_ids[src_page_index]; |
| 294 | |
| 295 | uint32_t take = MIN(p_array.count, page_size); //pages to take away |
| 296 | p_array.count -= take; |
| 297 | count += take; |
| 298 | src_page_index++; |
| 299 | } |
| 300 | |
| 301 | //handle the remainder page if exists |
| 302 | if (remainder_page) { |
| 303 | uint32_t new_remainder = count & page_size_mask; |
| 304 | |
| 305 | if (new_remainder > 0) { |
| 306 | //must merge old remainder with new remainder |
| 307 | |
| 308 | T *dst_page = page_data[_get_pages_in_use() - 1]; |
| 309 | uint32_t to_copy = MIN(page_size - new_remainder, remainder); |
| 310 | |
| 311 | for (uint32_t i = 0; i < to_copy; i++) { |
| 312 | if (!std::is_trivially_constructible<T>::value) { |
| 313 | memnew_placement(&dst_page[i + new_remainder], T(remainder_page[i + remainder - to_copy])); |
| 314 | } else { |
| 315 | dst_page[i + new_remainder] = remainder_page[i + remainder - to_copy]; |
| 316 | } |
| 317 | |
| 318 | if (!std::is_trivially_destructible<T>::value) { |
| 319 | remainder_page[i + remainder - to_copy].~T(); |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | remainder -= to_copy; //subtract what was copied from remainder |
| 324 | count += to_copy; //add what was copied to the count |
| 325 | |
| 326 | if (remainder == 0) { |
| 327 | //entire remainder copied, let go of remainder page |
| 328 | page_pool->free_page(remainder_page_id); |
| 329 | remainder_page = nullptr; |
| 330 | } |
| 331 | } |
| 332 | |
| 333 | if (remainder > 0) { |
| 334 | //there is still remainder, append it |
| 335 | uint32_t page_count = _get_pages_in_use(); |
| 336 | uint32_t new_page_count = page_count + 1; |
| 337 | |
| 338 | if (unlikely(new_page_count > max_pages_used)) { |
| 339 | _grow_page_array(); //keep out of inline |
| 340 | } |
| 341 | |
| 342 | page_data[page_count] = remainder_page; |
| 343 | page_ids[page_count] = remainder_page_id; |
| 344 | |
| 345 | count += remainder; |
| 346 | } |
| 347 | } |
| 348 | } |
| 349 | |
| 350 | _FORCE_INLINE_ uint64_t size() const { |
| 351 | return count; |
| 352 | } |
| 353 | |
| 354 | void set_page_pool(PagedArrayPool<T> *p_page_pool) { |
| 355 | ERR_FAIL_COND(max_pages_used > 0); // Safety check. |
| 356 | |
| 357 | page_pool = p_page_pool; |
| 358 | page_size_mask = page_pool->get_page_size_mask(); |
| 359 | page_size_shift = page_pool->get_page_size_shift(); |
| 360 | } |
| 361 | |
| 362 | ~PagedArray() { |
| 363 | reset(); |
| 364 | } |
| 365 | }; |
| 366 | |
| 367 | #endif // PAGED_ARRAY_H |
| 368 | |