1 | // © 2017 and later: Unicode, Inc. and others. |
2 | // License & terms of use: http://www.unicode.org/copyright.html |
3 | |
4 | // umutablecptrie.cpp (inspired by utrie2_builder.cpp) |
5 | // created: 2017dec29 Markus W. Scherer |
6 | |
7 | // #define UCPTRIE_DEBUG |
8 | #ifdef UCPTRIE_DEBUG |
9 | # include <stdio.h> |
10 | #endif |
11 | |
12 | #include "unicode/utypes.h" |
13 | #include "unicode/ucptrie.h" |
14 | #include "unicode/umutablecptrie.h" |
15 | #include "unicode/uobject.h" |
16 | #include "unicode/utf16.h" |
17 | #include "cmemory.h" |
18 | #include "uassert.h" |
19 | #include "ucptrie_impl.h" |
20 | |
21 | // ICU-20235 In case Microsoft math.h has defined this, undefine it. |
22 | #ifdef OVERFLOW |
23 | #undef OVERFLOW |
24 | #endif |
25 | |
26 | U_NAMESPACE_BEGIN |
27 | |
28 | namespace { |
29 | |
30 | constexpr int32_t MAX_UNICODE = 0x10ffff; |
31 | |
32 | constexpr int32_t UNICODE_LIMIT = 0x110000; |
33 | constexpr int32_t BMP_LIMIT = 0x10000; |
34 | constexpr int32_t ASCII_LIMIT = 0x80; |
35 | |
36 | constexpr int32_t I_LIMIT = UNICODE_LIMIT >> UCPTRIE_SHIFT_3; |
37 | constexpr int32_t BMP_I_LIMIT = BMP_LIMIT >> UCPTRIE_SHIFT_3; |
38 | constexpr int32_t ASCII_I_LIMIT = ASCII_LIMIT >> UCPTRIE_SHIFT_3; |
39 | |
40 | constexpr int32_t SMALL_DATA_BLOCKS_PER_BMP_BLOCK = (1 << (UCPTRIE_FAST_SHIFT - UCPTRIE_SHIFT_3)); |
41 | |
42 | // Flag values for data blocks. |
43 | constexpr uint8_t ALL_SAME = 0; |
44 | constexpr uint8_t MIXED = 1; |
45 | constexpr uint8_t SAME_AS = 2; |
46 | |
47 | /** Start with allocation of 16k data entries. */ |
48 | constexpr int32_t INITIAL_DATA_LENGTH = ((int32_t)1 << 14); |
49 | |
50 | /** Grow about 8x each time. */ |
51 | constexpr int32_t MEDIUM_DATA_LENGTH = ((int32_t)1 << 17); |
52 | |
53 | /** |
54 | * Maximum length of the build-time data array. |
55 | * One entry per 0x110000 code points. |
56 | */ |
57 | constexpr int32_t MAX_DATA_LENGTH = UNICODE_LIMIT; |
58 | |
59 | // Flag values for index-3 blocks while compacting/building. |
60 | constexpr uint8_t I3_NULL = 0; |
61 | constexpr uint8_t I3_BMP = 1; |
62 | constexpr uint8_t I3_16 = 2; |
63 | constexpr uint8_t I3_18 = 3; |
64 | |
65 | constexpr int32_t INDEX_3_18BIT_BLOCK_LENGTH = UCPTRIE_INDEX_3_BLOCK_LENGTH + UCPTRIE_INDEX_3_BLOCK_LENGTH / 8; |
66 | |
67 | class AllSameBlocks; |
68 | class MixedBlocks; |
69 | |
70 | class MutableCodePointTrie : public UMemory { |
71 | public: |
72 | MutableCodePointTrie(uint32_t initialValue, uint32_t errorValue, UErrorCode &errorCode); |
73 | MutableCodePointTrie(const MutableCodePointTrie &other, UErrorCode &errorCode); |
74 | MutableCodePointTrie(const MutableCodePointTrie &other) = delete; |
75 | ~MutableCodePointTrie(); |
76 | |
77 | MutableCodePointTrie &operator=(const MutableCodePointTrie &other) = delete; |
78 | |
79 | static MutableCodePointTrie *fromUCPMap(const UCPMap *map, UErrorCode &errorCode); |
80 | static MutableCodePointTrie *fromUCPTrie(const UCPTrie *trie, UErrorCode &errorCode); |
81 | |
82 | uint32_t get(UChar32 c) const; |
83 | int32_t getRange(UChar32 start, UCPMapValueFilter *filter, const void *context, |
84 | uint32_t *pValue) const; |
85 | |
86 | void set(UChar32 c, uint32_t value, UErrorCode &errorCode); |
87 | void setRange(UChar32 start, UChar32 end, uint32_t value, UErrorCode &errorCode); |
88 | |
89 | UCPTrie *build(UCPTrieType type, UCPTrieValueWidth valueWidth, UErrorCode &errorCode); |
90 | |
91 | private: |
92 | void clear(); |
93 | |
94 | bool ensureHighStart(UChar32 c); |
95 | int32_t allocDataBlock(int32_t blockLength); |
96 | int32_t getDataBlock(int32_t i); |
97 | |
98 | void maskValues(uint32_t mask); |
99 | UChar32 findHighStart() const; |
100 | int32_t compactWholeDataBlocks(int32_t fastILimit, AllSameBlocks &allSameBlocks); |
101 | int32_t compactData( |
102 | int32_t fastILimit, uint32_t *newData, int32_t newDataCapacity, |
103 | int32_t dataNullIndex, MixedBlocks &mixedBlocks, UErrorCode &errorCode); |
104 | int32_t compactIndex(int32_t fastILimit, MixedBlocks &mixedBlocks, UErrorCode &errorCode); |
105 | int32_t compactTrie(int32_t fastILimit, UErrorCode &errorCode); |
106 | |
107 | uint32_t *index = nullptr; |
108 | int32_t indexCapacity = 0; |
109 | int32_t index3NullOffset = -1; |
110 | uint32_t *data = nullptr; |
111 | int32_t dataCapacity = 0; |
112 | int32_t dataLength = 0; |
113 | int32_t dataNullOffset = -1; |
114 | |
115 | uint32_t origInitialValue; |
116 | uint32_t initialValue; |
117 | uint32_t errorValue; |
118 | UChar32 highStart; |
119 | uint32_t highValue; |
120 | #ifdef UCPTRIE_DEBUG |
121 | public: |
122 | const char *name; |
123 | #endif |
124 | private: |
125 | /** Temporary array while building the final data. */ |
126 | uint16_t *index16 = nullptr; |
127 | uint8_t flags[UNICODE_LIMIT >> UCPTRIE_SHIFT_3]; |
128 | }; |
129 | |
130 | MutableCodePointTrie::MutableCodePointTrie(uint32_t iniValue, uint32_t errValue, UErrorCode &errorCode) : |
131 | origInitialValue(iniValue), initialValue(iniValue), errorValue(errValue), |
132 | highStart(0), highValue(initialValue) |
133 | #ifdef UCPTRIE_DEBUG |
134 | , name("open" ) |
135 | #endif |
136 | { |
137 | if (U_FAILURE(errorCode)) { return; } |
138 | index = (uint32_t *)uprv_malloc(BMP_I_LIMIT * 4); |
139 | data = (uint32_t *)uprv_malloc(INITIAL_DATA_LENGTH * 4); |
140 | if (index == nullptr || data == nullptr) { |
141 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
142 | return; |
143 | } |
144 | indexCapacity = BMP_I_LIMIT; |
145 | dataCapacity = INITIAL_DATA_LENGTH; |
146 | } |
147 | |
148 | MutableCodePointTrie::MutableCodePointTrie(const MutableCodePointTrie &other, UErrorCode &errorCode) : |
149 | index3NullOffset(other.index3NullOffset), |
150 | dataNullOffset(other.dataNullOffset), |
151 | origInitialValue(other.origInitialValue), initialValue(other.initialValue), |
152 | errorValue(other.errorValue), |
153 | highStart(other.highStart), highValue(other.highValue) |
154 | #ifdef UCPTRIE_DEBUG |
155 | , name("mutable clone" ) |
156 | #endif |
157 | { |
158 | if (U_FAILURE(errorCode)) { return; } |
159 | int32_t iCapacity = highStart <= BMP_LIMIT ? BMP_I_LIMIT : I_LIMIT; |
160 | index = (uint32_t *)uprv_malloc(iCapacity * 4); |
161 | data = (uint32_t *)uprv_malloc(other.dataCapacity * 4); |
162 | if (index == nullptr || data == nullptr) { |
163 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
164 | return; |
165 | } |
166 | indexCapacity = iCapacity; |
167 | dataCapacity = other.dataCapacity; |
168 | |
169 | int32_t iLimit = highStart >> UCPTRIE_SHIFT_3; |
170 | uprv_memcpy(flags, other.flags, iLimit); |
171 | uprv_memcpy(index, other.index, iLimit * 4); |
172 | uprv_memcpy(data, other.data, (size_t)other.dataLength * 4); |
173 | dataLength = other.dataLength; |
174 | U_ASSERT(other.index16 == nullptr); |
175 | } |
176 | |
177 | MutableCodePointTrie::~MutableCodePointTrie() { |
178 | uprv_free(index); |
179 | uprv_free(data); |
180 | uprv_free(index16); |
181 | } |
182 | |
183 | MutableCodePointTrie *MutableCodePointTrie::fromUCPMap(const UCPMap *map, UErrorCode &errorCode) { |
184 | // Use the highValue as the initialValue to reduce the highStart. |
185 | uint32_t errorValue = ucpmap_get(map, -1); |
186 | uint32_t initialValue = ucpmap_get(map, 0x10ffff); |
187 | LocalPointer<MutableCodePointTrie> mutableTrie( |
188 | new MutableCodePointTrie(initialValue, errorValue, errorCode), |
189 | errorCode); |
190 | if (U_FAILURE(errorCode)) { |
191 | return nullptr; |
192 | } |
193 | UChar32 start = 0, end; |
194 | uint32_t value; |
195 | while ((end = ucpmap_getRange(map, start, UCPMAP_RANGE_NORMAL, 0, |
196 | nullptr, nullptr, &value)) >= 0) { |
197 | if (value != initialValue) { |
198 | if (start == end) { |
199 | mutableTrie->set(start, value, errorCode); |
200 | } else { |
201 | mutableTrie->setRange(start, end, value, errorCode); |
202 | } |
203 | } |
204 | start = end + 1; |
205 | } |
206 | if (U_SUCCESS(errorCode)) { |
207 | return mutableTrie.orphan(); |
208 | } else { |
209 | return nullptr; |
210 | } |
211 | } |
212 | |
213 | MutableCodePointTrie *MutableCodePointTrie::fromUCPTrie(const UCPTrie *trie, UErrorCode &errorCode) { |
214 | // Use the highValue as the initialValue to reduce the highStart. |
215 | uint32_t errorValue; |
216 | uint32_t initialValue; |
217 | switch (trie->valueWidth) { |
218 | case UCPTRIE_VALUE_BITS_16: |
219 | errorValue = trie->data.ptr16[trie->dataLength - UCPTRIE_ERROR_VALUE_NEG_DATA_OFFSET]; |
220 | initialValue = trie->data.ptr16[trie->dataLength - UCPTRIE_HIGH_VALUE_NEG_DATA_OFFSET]; |
221 | break; |
222 | case UCPTRIE_VALUE_BITS_32: |
223 | errorValue = trie->data.ptr32[trie->dataLength - UCPTRIE_ERROR_VALUE_NEG_DATA_OFFSET]; |
224 | initialValue = trie->data.ptr32[trie->dataLength - UCPTRIE_HIGH_VALUE_NEG_DATA_OFFSET]; |
225 | break; |
226 | case UCPTRIE_VALUE_BITS_8: |
227 | errorValue = trie->data.ptr8[trie->dataLength - UCPTRIE_ERROR_VALUE_NEG_DATA_OFFSET]; |
228 | initialValue = trie->data.ptr8[trie->dataLength - UCPTRIE_HIGH_VALUE_NEG_DATA_OFFSET]; |
229 | break; |
230 | default: |
231 | // Unreachable if the trie is properly initialized. |
232 | errorCode = U_ILLEGAL_ARGUMENT_ERROR; |
233 | return nullptr; |
234 | } |
235 | LocalPointer<MutableCodePointTrie> mutableTrie( |
236 | new MutableCodePointTrie(initialValue, errorValue, errorCode), |
237 | errorCode); |
238 | if (U_FAILURE(errorCode)) { |
239 | return nullptr; |
240 | } |
241 | UChar32 start = 0, end; |
242 | uint32_t value; |
243 | while ((end = ucptrie_getRange(trie, start, UCPMAP_RANGE_NORMAL, 0, |
244 | nullptr, nullptr, &value)) >= 0) { |
245 | if (value != initialValue) { |
246 | if (start == end) { |
247 | mutableTrie->set(start, value, errorCode); |
248 | } else { |
249 | mutableTrie->setRange(start, end, value, errorCode); |
250 | } |
251 | } |
252 | start = end + 1; |
253 | } |
254 | if (U_SUCCESS(errorCode)) { |
255 | return mutableTrie.orphan(); |
256 | } else { |
257 | return nullptr; |
258 | } |
259 | } |
260 | |
261 | void MutableCodePointTrie::clear() { |
262 | index3NullOffset = dataNullOffset = -1; |
263 | dataLength = 0; |
264 | highValue = initialValue = origInitialValue; |
265 | highStart = 0; |
266 | uprv_free(index16); |
267 | index16 = nullptr; |
268 | } |
269 | |
270 | uint32_t MutableCodePointTrie::get(UChar32 c) const { |
271 | if ((uint32_t)c > MAX_UNICODE) { |
272 | return errorValue; |
273 | } |
274 | if (c >= highStart) { |
275 | return highValue; |
276 | } |
277 | int32_t i = c >> UCPTRIE_SHIFT_3; |
278 | if (flags[i] == ALL_SAME) { |
279 | return index[i]; |
280 | } else { |
281 | return data[index[i] + (c & UCPTRIE_SMALL_DATA_MASK)]; |
282 | } |
283 | } |
284 | |
285 | inline uint32_t maybeFilterValue(uint32_t value, uint32_t initialValue, uint32_t nullValue, |
286 | UCPMapValueFilter *filter, const void *context) { |
287 | if (value == initialValue) { |
288 | value = nullValue; |
289 | } else if (filter != nullptr) { |
290 | value = filter(context, value); |
291 | } |
292 | return value; |
293 | } |
294 | |
295 | UChar32 MutableCodePointTrie::getRange( |
296 | UChar32 start, UCPMapValueFilter *filter, const void *context, |
297 | uint32_t *pValue) const { |
298 | if ((uint32_t)start > MAX_UNICODE) { |
299 | return U_SENTINEL; |
300 | } |
301 | if (start >= highStart) { |
302 | if (pValue != nullptr) { |
303 | uint32_t value = highValue; |
304 | if (filter != nullptr) { value = filter(context, value); } |
305 | *pValue = value; |
306 | } |
307 | return MAX_UNICODE; |
308 | } |
309 | uint32_t nullValue = initialValue; |
310 | if (filter != nullptr) { nullValue = filter(context, nullValue); } |
311 | UChar32 c = start; |
312 | uint32_t trieValue, value; |
313 | bool haveValue = false; |
314 | int32_t i = c >> UCPTRIE_SHIFT_3; |
315 | do { |
316 | if (flags[i] == ALL_SAME) { |
317 | uint32_t trieValue2 = index[i]; |
318 | if (haveValue) { |
319 | if (trieValue2 != trieValue) { |
320 | if (filter == nullptr || |
321 | maybeFilterValue(trieValue2, initialValue, nullValue, |
322 | filter, context) != value) { |
323 | return c - 1; |
324 | } |
325 | trieValue = trieValue2; // may or may not help |
326 | } |
327 | } else { |
328 | trieValue = trieValue2; |
329 | value = maybeFilterValue(trieValue2, initialValue, nullValue, filter, context); |
330 | if (pValue != nullptr) { *pValue = value; } |
331 | haveValue = true; |
332 | } |
333 | c = (c + UCPTRIE_SMALL_DATA_BLOCK_LENGTH) & ~UCPTRIE_SMALL_DATA_MASK; |
334 | } else /* MIXED */ { |
335 | int32_t di = index[i] + (c & UCPTRIE_SMALL_DATA_MASK); |
336 | uint32_t trieValue2 = data[di]; |
337 | if (haveValue) { |
338 | if (trieValue2 != trieValue) { |
339 | if (filter == nullptr || |
340 | maybeFilterValue(trieValue2, initialValue, nullValue, |
341 | filter, context) != value) { |
342 | return c - 1; |
343 | } |
344 | trieValue = trieValue2; // may or may not help |
345 | } |
346 | } else { |
347 | trieValue = trieValue2; |
348 | value = maybeFilterValue(trieValue2, initialValue, nullValue, filter, context); |
349 | if (pValue != nullptr) { *pValue = value; } |
350 | haveValue = true; |
351 | } |
352 | while ((++c & UCPTRIE_SMALL_DATA_MASK) != 0) { |
353 | trieValue2 = data[++di]; |
354 | if (trieValue2 != trieValue) { |
355 | if (filter == nullptr || |
356 | maybeFilterValue(trieValue2, initialValue, nullValue, |
357 | filter, context) != value) { |
358 | return c - 1; |
359 | } |
360 | } |
361 | trieValue = trieValue2; // may or may not help |
362 | } |
363 | } |
364 | ++i; |
365 | } while (c < highStart); |
366 | U_ASSERT(haveValue); |
367 | if (maybeFilterValue(highValue, initialValue, nullValue, |
368 | filter, context) != value) { |
369 | return c - 1; |
370 | } else { |
371 | return MAX_UNICODE; |
372 | } |
373 | } |
374 | |
375 | void |
376 | writeBlock(uint32_t *block, uint32_t value) { |
377 | uint32_t *limit = block + UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
378 | while (block < limit) { |
379 | *block++ = value; |
380 | } |
381 | } |
382 | |
383 | bool MutableCodePointTrie::ensureHighStart(UChar32 c) { |
384 | if (c >= highStart) { |
385 | // Round up to a UCPTRIE_CP_PER_INDEX_2_ENTRY boundary to simplify compaction. |
386 | c = (c + UCPTRIE_CP_PER_INDEX_2_ENTRY) & ~(UCPTRIE_CP_PER_INDEX_2_ENTRY - 1); |
387 | int32_t i = highStart >> UCPTRIE_SHIFT_3; |
388 | int32_t iLimit = c >> UCPTRIE_SHIFT_3; |
389 | if (iLimit > indexCapacity) { |
390 | uint32_t *newIndex = (uint32_t *)uprv_malloc(I_LIMIT * 4); |
391 | if (newIndex == nullptr) { return false; } |
392 | uprv_memcpy(newIndex, index, i * 4); |
393 | uprv_free(index); |
394 | index = newIndex; |
395 | indexCapacity = I_LIMIT; |
396 | } |
397 | do { |
398 | flags[i] = ALL_SAME; |
399 | index[i] = initialValue; |
400 | } while(++i < iLimit); |
401 | highStart = c; |
402 | } |
403 | return true; |
404 | } |
405 | |
406 | int32_t MutableCodePointTrie::allocDataBlock(int32_t blockLength) { |
407 | int32_t newBlock = dataLength; |
408 | int32_t newTop = newBlock + blockLength; |
409 | if (newTop > dataCapacity) { |
410 | int32_t capacity; |
411 | if (dataCapacity < MEDIUM_DATA_LENGTH) { |
412 | capacity = MEDIUM_DATA_LENGTH; |
413 | } else if (dataCapacity < MAX_DATA_LENGTH) { |
414 | capacity = MAX_DATA_LENGTH; |
415 | } else { |
416 | // Should never occur. |
417 | // Either MAX_DATA_LENGTH is incorrect, |
418 | // or the code writes more values than should be possible. |
419 | return -1; |
420 | } |
421 | uint32_t *newData = (uint32_t *)uprv_malloc(capacity * 4); |
422 | if (newData == nullptr) { |
423 | return -1; |
424 | } |
425 | uprv_memcpy(newData, data, (size_t)dataLength * 4); |
426 | uprv_free(data); |
427 | data = newData; |
428 | dataCapacity = capacity; |
429 | } |
430 | dataLength = newTop; |
431 | return newBlock; |
432 | } |
433 | |
434 | /** |
435 | * No error checking for illegal arguments. |
436 | * |
437 | * @return -1 if no new data block available (out of memory in data array) |
438 | * @internal |
439 | */ |
440 | int32_t MutableCodePointTrie::getDataBlock(int32_t i) { |
441 | if (flags[i] == MIXED) { |
442 | return index[i]; |
443 | } |
444 | if (i < BMP_I_LIMIT) { |
445 | int32_t newBlock = allocDataBlock(UCPTRIE_FAST_DATA_BLOCK_LENGTH); |
446 | if (newBlock < 0) { return newBlock; } |
447 | int32_t iStart = i & ~(SMALL_DATA_BLOCKS_PER_BMP_BLOCK -1); |
448 | int32_t iLimit = iStart + SMALL_DATA_BLOCKS_PER_BMP_BLOCK; |
449 | do { |
450 | U_ASSERT(flags[iStart] == ALL_SAME); |
451 | writeBlock(data + newBlock, index[iStart]); |
452 | flags[iStart] = MIXED; |
453 | index[iStart++] = newBlock; |
454 | newBlock += UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
455 | } while (iStart < iLimit); |
456 | return index[i]; |
457 | } else { |
458 | int32_t newBlock = allocDataBlock(UCPTRIE_SMALL_DATA_BLOCK_LENGTH); |
459 | if (newBlock < 0) { return newBlock; } |
460 | writeBlock(data + newBlock, index[i]); |
461 | flags[i] = MIXED; |
462 | index[i] = newBlock; |
463 | return newBlock; |
464 | } |
465 | } |
466 | |
467 | void MutableCodePointTrie::set(UChar32 c, uint32_t value, UErrorCode &errorCode) { |
468 | if (U_FAILURE(errorCode)) { |
469 | return; |
470 | } |
471 | if ((uint32_t)c > MAX_UNICODE) { |
472 | errorCode = U_ILLEGAL_ARGUMENT_ERROR; |
473 | return; |
474 | } |
475 | |
476 | int32_t block; |
477 | if (!ensureHighStart(c) || (block = getDataBlock(c >> UCPTRIE_SHIFT_3)) < 0) { |
478 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
479 | return; |
480 | } |
481 | |
482 | data[block + (c & UCPTRIE_SMALL_DATA_MASK)] = value; |
483 | } |
484 | |
485 | void |
486 | fillBlock(uint32_t *block, UChar32 start, UChar32 limit, uint32_t value) { |
487 | uint32_t *pLimit = block + limit; |
488 | block += start; |
489 | while (block < pLimit) { |
490 | *block++ = value; |
491 | } |
492 | } |
493 | |
494 | void MutableCodePointTrie::setRange(UChar32 start, UChar32 end, uint32_t value, UErrorCode &errorCode) { |
495 | if (U_FAILURE(errorCode)) { |
496 | return; |
497 | } |
498 | if ((uint32_t)start > MAX_UNICODE || (uint32_t)end > MAX_UNICODE || start > end) { |
499 | errorCode = U_ILLEGAL_ARGUMENT_ERROR; |
500 | return; |
501 | } |
502 | if (!ensureHighStart(end)) { |
503 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
504 | return; |
505 | } |
506 | |
507 | UChar32 limit = end + 1; |
508 | if (start & UCPTRIE_SMALL_DATA_MASK) { |
509 | // Set partial block at [start..following block boundary[. |
510 | int32_t block = getDataBlock(start >> UCPTRIE_SHIFT_3); |
511 | if (block < 0) { |
512 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
513 | return; |
514 | } |
515 | |
516 | UChar32 nextStart = (start + UCPTRIE_SMALL_DATA_MASK) & ~UCPTRIE_SMALL_DATA_MASK; |
517 | if (nextStart <= limit) { |
518 | fillBlock(data + block, start & UCPTRIE_SMALL_DATA_MASK, UCPTRIE_SMALL_DATA_BLOCK_LENGTH, |
519 | value); |
520 | start = nextStart; |
521 | } else { |
522 | fillBlock(data + block, start & UCPTRIE_SMALL_DATA_MASK, limit & UCPTRIE_SMALL_DATA_MASK, |
523 | value); |
524 | return; |
525 | } |
526 | } |
527 | |
528 | // Number of positions in the last, partial block. |
529 | int32_t rest = limit & UCPTRIE_SMALL_DATA_MASK; |
530 | |
531 | // Round down limit to a block boundary. |
532 | limit &= ~UCPTRIE_SMALL_DATA_MASK; |
533 | |
534 | // Iterate over all-value blocks. |
535 | while (start < limit) { |
536 | int32_t i = start >> UCPTRIE_SHIFT_3; |
537 | if (flags[i] == ALL_SAME) { |
538 | index[i] = value; |
539 | } else /* MIXED */ { |
540 | fillBlock(data + index[i], 0, UCPTRIE_SMALL_DATA_BLOCK_LENGTH, value); |
541 | } |
542 | start += UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
543 | } |
544 | |
545 | if (rest > 0) { |
546 | // Set partial block at [last block boundary..limit[. |
547 | int32_t block = getDataBlock(start >> UCPTRIE_SHIFT_3); |
548 | if (block < 0) { |
549 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
550 | return; |
551 | } |
552 | |
553 | fillBlock(data + block, 0, rest, value); |
554 | } |
555 | } |
556 | |
557 | /* compaction --------------------------------------------------------------- */ |
558 | |
559 | void MutableCodePointTrie::maskValues(uint32_t mask) { |
560 | initialValue &= mask; |
561 | errorValue &= mask; |
562 | highValue &= mask; |
563 | int32_t iLimit = highStart >> UCPTRIE_SHIFT_3; |
564 | for (int32_t i = 0; i < iLimit; ++i) { |
565 | if (flags[i] == ALL_SAME) { |
566 | index[i] &= mask; |
567 | } |
568 | } |
569 | for (int32_t i = 0; i < dataLength; ++i) { |
570 | data[i] &= mask; |
571 | } |
572 | } |
573 | |
574 | template<typename UIntA, typename UIntB> |
575 | bool equalBlocks(const UIntA *s, const UIntB *t, int32_t length) { |
576 | while (length > 0 && *s == *t) { |
577 | ++s; |
578 | ++t; |
579 | --length; |
580 | } |
581 | return length == 0; |
582 | } |
583 | |
584 | bool allValuesSameAs(const uint32_t *p, int32_t length, uint32_t value) { |
585 | const uint32_t *pLimit = p + length; |
586 | while (p < pLimit && *p == value) { ++p; } |
587 | return p == pLimit; |
588 | } |
589 | |
590 | /** Search for an identical block. */ |
591 | int32_t findSameBlock(const uint16_t *p, int32_t pStart, int32_t length, |
592 | const uint16_t *q, int32_t qStart, int32_t blockLength) { |
593 | // Ensure that we do not even partially get past length. |
594 | length -= blockLength; |
595 | |
596 | q += qStart; |
597 | while (pStart <= length) { |
598 | if (equalBlocks(p + pStart, q, blockLength)) { |
599 | return pStart; |
600 | } |
601 | ++pStart; |
602 | } |
603 | return -1; |
604 | } |
605 | |
606 | int32_t findAllSameBlock(const uint32_t *p, int32_t start, int32_t limit, |
607 | uint32_t value, int32_t blockLength) { |
608 | // Ensure that we do not even partially get past limit. |
609 | limit -= blockLength; |
610 | |
611 | for (int32_t block = start; block <= limit; ++block) { |
612 | if (p[block] == value) { |
613 | for (int32_t i = 1;; ++i) { |
614 | if (i == blockLength) { |
615 | return block; |
616 | } |
617 | if (p[block + i] != value) { |
618 | block += i; |
619 | break; |
620 | } |
621 | } |
622 | } |
623 | } |
624 | return -1; |
625 | } |
626 | |
627 | /** |
628 | * Look for maximum overlap of the beginning of the other block |
629 | * with the previous, adjacent block. |
630 | */ |
631 | template<typename UIntA, typename UIntB> |
632 | int32_t getOverlap(const UIntA *p, int32_t length, |
633 | const UIntB *q, int32_t qStart, int32_t blockLength) { |
634 | int32_t overlap = blockLength - 1; |
635 | U_ASSERT(overlap <= length); |
636 | q += qStart; |
637 | while (overlap > 0 && !equalBlocks(p + (length - overlap), q, overlap)) { |
638 | --overlap; |
639 | } |
640 | return overlap; |
641 | } |
642 | |
643 | int32_t getAllSameOverlap(const uint32_t *p, int32_t length, uint32_t value, |
644 | int32_t blockLength) { |
645 | int32_t min = length - (blockLength - 1); |
646 | int32_t i = length; |
647 | while (min < i && p[i - 1] == value) { --i; } |
648 | return length - i; |
649 | } |
650 | |
651 | bool isStartOfSomeFastBlock(uint32_t dataOffset, const uint32_t index[], int32_t fastILimit) { |
652 | for (int32_t i = 0; i < fastILimit; i += SMALL_DATA_BLOCKS_PER_BMP_BLOCK) { |
653 | if (index[i] == dataOffset) { |
654 | return true; |
655 | } |
656 | } |
657 | return false; |
658 | } |
659 | |
660 | /** |
661 | * Finds the start of the last range in the trie by enumerating backward. |
662 | * Indexes for code points higher than this will be omitted. |
663 | */ |
664 | UChar32 MutableCodePointTrie::findHighStart() const { |
665 | int32_t i = highStart >> UCPTRIE_SHIFT_3; |
666 | while (i > 0) { |
667 | bool match; |
668 | if (flags[--i] == ALL_SAME) { |
669 | match = index[i] == highValue; |
670 | } else /* MIXED */ { |
671 | const uint32_t *p = data + index[i]; |
672 | for (int32_t j = 0;; ++j) { |
673 | if (j == UCPTRIE_SMALL_DATA_BLOCK_LENGTH) { |
674 | match = true; |
675 | break; |
676 | } |
677 | if (p[j] != highValue) { |
678 | match = false; |
679 | break; |
680 | } |
681 | } |
682 | } |
683 | if (!match) { |
684 | return (i + 1) << UCPTRIE_SHIFT_3; |
685 | } |
686 | } |
687 | return 0; |
688 | } |
689 | |
690 | class AllSameBlocks { |
691 | public: |
692 | static constexpr int32_t NEW_UNIQUE = -1; |
693 | static constexpr int32_t OVERFLOW = -2; |
694 | |
695 | AllSameBlocks() : length(0), mostRecent(-1) {} |
696 | |
697 | int32_t findOrAdd(int32_t index, int32_t count, uint32_t value) { |
698 | if (mostRecent >= 0 && values[mostRecent] == value) { |
699 | refCounts[mostRecent] += count; |
700 | return indexes[mostRecent]; |
701 | } |
702 | for (int32_t i = 0; i < length; ++i) { |
703 | if (values[i] == value) { |
704 | mostRecent = i; |
705 | refCounts[i] += count; |
706 | return indexes[i]; |
707 | } |
708 | } |
709 | if (length == CAPACITY) { |
710 | return OVERFLOW; |
711 | } |
712 | mostRecent = length; |
713 | indexes[length] = index; |
714 | values[length] = value; |
715 | refCounts[length++] = count; |
716 | return NEW_UNIQUE; |
717 | } |
718 | |
719 | /** Replaces the block which has the lowest reference count. */ |
720 | void add(int32_t index, int32_t count, uint32_t value) { |
721 | U_ASSERT(length == CAPACITY); |
722 | int32_t least = -1; |
723 | int32_t leastCount = I_LIMIT; |
724 | for (int32_t i = 0; i < length; ++i) { |
725 | U_ASSERT(values[i] != value); |
726 | if (refCounts[i] < leastCount) { |
727 | least = i; |
728 | leastCount = refCounts[i]; |
729 | } |
730 | } |
731 | U_ASSERT(least >= 0); |
732 | mostRecent = least; |
733 | indexes[least] = index; |
734 | values[least] = value; |
735 | refCounts[least] = count; |
736 | } |
737 | |
738 | int32_t findMostUsed() const { |
739 | if (length == 0) { return -1; } |
740 | int32_t max = -1; |
741 | int32_t maxCount = 0; |
742 | for (int32_t i = 0; i < length; ++i) { |
743 | if (refCounts[i] > maxCount) { |
744 | max = i; |
745 | maxCount = refCounts[i]; |
746 | } |
747 | } |
748 | return indexes[max]; |
749 | } |
750 | |
751 | private: |
752 | static constexpr int32_t CAPACITY = 32; |
753 | |
754 | int32_t length; |
755 | int32_t mostRecent; |
756 | |
757 | int32_t indexes[CAPACITY]; |
758 | uint32_t values[CAPACITY]; |
759 | int32_t refCounts[CAPACITY]; |
760 | }; |
761 | |
762 | // Custom hash table for mixed-value blocks to be found anywhere in the |
763 | // compacted data or index so far. |
764 | class MixedBlocks { |
765 | public: |
766 | MixedBlocks() {} |
767 | ~MixedBlocks() { |
768 | uprv_free(table); |
769 | } |
770 | |
771 | bool init(int32_t maxLength, int32_t newBlockLength) { |
772 | // We store actual data indexes + 1 to reserve 0 for empty entries. |
773 | int32_t maxDataIndex = maxLength - newBlockLength + 1; |
774 | int32_t newLength; |
775 | if (maxDataIndex <= 0xfff) { // 4k |
776 | newLength = 6007; |
777 | shift = 12; |
778 | mask = 0xfff; |
779 | } else if (maxDataIndex <= 0x7fff) { // 32k |
780 | newLength = 50021; |
781 | shift = 15; |
782 | mask = 0x7fff; |
783 | } else if (maxDataIndex <= 0x1ffff) { // 128k |
784 | newLength = 200003; |
785 | shift = 17; |
786 | mask = 0x1ffff; |
787 | } else { |
788 | // maxDataIndex up to around MAX_DATA_LENGTH, ca. 1.1M |
789 | newLength = 1500007; |
790 | shift = 21; |
791 | mask = 0x1fffff; |
792 | } |
793 | if (newLength > capacity) { |
794 | uprv_free(table); |
795 | table = (uint32_t *)uprv_malloc(newLength * 4); |
796 | if (table == nullptr) { |
797 | return false; |
798 | } |
799 | capacity = newLength; |
800 | } |
801 | length = newLength; |
802 | uprv_memset(table, 0, length * 4); |
803 | |
804 | blockLength = newBlockLength; |
805 | return true; |
806 | } |
807 | |
808 | template<typename UInt> |
809 | void extend(const UInt *data, int32_t minStart, int32_t prevDataLength, int32_t newDataLength) { |
810 | int32_t start = prevDataLength - blockLength; |
811 | if (start >= minStart) { |
812 | ++start; // Skip the last block that we added last time. |
813 | } else { |
814 | start = minStart; // Begin with the first full block. |
815 | } |
816 | for (int32_t end = newDataLength - blockLength; start <= end; ++start) { |
817 | uint32_t hashCode = makeHashCode(data, start); |
818 | addEntry(data, start, hashCode, start); |
819 | } |
820 | } |
821 | |
822 | template<typename UIntA, typename UIntB> |
823 | int32_t findBlock(const UIntA *data, const UIntB *blockData, int32_t blockStart) const { |
824 | uint32_t hashCode = makeHashCode(blockData, blockStart); |
825 | int32_t entryIndex = findEntry(data, blockData, blockStart, hashCode); |
826 | if (entryIndex >= 0) { |
827 | return (table[entryIndex] & mask) - 1; |
828 | } else { |
829 | return -1; |
830 | } |
831 | } |
832 | |
833 | int32_t findAllSameBlock(const uint32_t *data, uint32_t blockValue) const { |
834 | uint32_t hashCode = makeHashCode(blockValue); |
835 | int32_t entryIndex = findEntry(data, blockValue, hashCode); |
836 | if (entryIndex >= 0) { |
837 | return (table[entryIndex] & mask) - 1; |
838 | } else { |
839 | return -1; |
840 | } |
841 | } |
842 | |
843 | private: |
844 | template<typename UInt> |
845 | uint32_t makeHashCode(const UInt *blockData, int32_t blockStart) const { |
846 | int32_t blockLimit = blockStart + blockLength; |
847 | uint32_t hashCode = blockData[blockStart++]; |
848 | do { |
849 | hashCode = 37 * hashCode + blockData[blockStart++]; |
850 | } while (blockStart < blockLimit); |
851 | return hashCode; |
852 | } |
853 | |
854 | uint32_t makeHashCode(uint32_t blockValue) const { |
855 | uint32_t hashCode = blockValue; |
856 | for (int32_t i = 1; i < blockLength; ++i) { |
857 | hashCode = 37 * hashCode + blockValue; |
858 | } |
859 | return hashCode; |
860 | } |
861 | |
862 | template<typename UInt> |
863 | void addEntry(const UInt *data, int32_t blockStart, uint32_t hashCode, int32_t dataIndex) { |
864 | U_ASSERT(0 <= dataIndex && dataIndex < (int32_t)mask); |
865 | int32_t entryIndex = findEntry(data, data, blockStart, hashCode); |
866 | if (entryIndex < 0) { |
867 | table[~entryIndex] = (hashCode << shift) | (dataIndex + 1); |
868 | } |
869 | } |
870 | |
871 | template<typename UIntA, typename UIntB> |
872 | int32_t findEntry(const UIntA *data, const UIntB *blockData, int32_t blockStart, |
873 | uint32_t hashCode) const { |
874 | uint32_t shiftedHashCode = hashCode << shift; |
875 | int32_t initialEntryIndex = (hashCode % (length - 1)) + 1; // 1..length-1 |
876 | for (int32_t entryIndex = initialEntryIndex;;) { |
877 | uint32_t entry = table[entryIndex]; |
878 | if (entry == 0) { |
879 | return ~entryIndex; |
880 | } |
881 | if ((entry & ~mask) == shiftedHashCode) { |
882 | int32_t dataIndex = (entry & mask) - 1; |
883 | if (equalBlocks(data + dataIndex, blockData + blockStart, blockLength)) { |
884 | return entryIndex; |
885 | } |
886 | } |
887 | entryIndex = nextIndex(initialEntryIndex, entryIndex); |
888 | } |
889 | } |
890 | |
891 | int32_t findEntry(const uint32_t *data, uint32_t blockValue, uint32_t hashCode) const { |
892 | uint32_t shiftedHashCode = hashCode << shift; |
893 | int32_t initialEntryIndex = (hashCode % (length - 1)) + 1; // 1..length-1 |
894 | for (int32_t entryIndex = initialEntryIndex;;) { |
895 | uint32_t entry = table[entryIndex]; |
896 | if (entry == 0) { |
897 | return ~entryIndex; |
898 | } |
899 | if ((entry & ~mask) == shiftedHashCode) { |
900 | int32_t dataIndex = (entry & mask) - 1; |
901 | if (allValuesSameAs(data + dataIndex, blockLength, blockValue)) { |
902 | return entryIndex; |
903 | } |
904 | } |
905 | entryIndex = nextIndex(initialEntryIndex, entryIndex); |
906 | } |
907 | } |
908 | |
909 | inline int32_t nextIndex(int32_t initialEntryIndex, int32_t entryIndex) const { |
910 | // U_ASSERT(0 < initialEntryIndex && initialEntryIndex < length); |
911 | return (entryIndex + initialEntryIndex) % length; |
912 | } |
913 | |
914 | // Hash table. |
915 | // The length is a prime number, larger than the maximum data length. |
916 | // The "shift" lower bits store a data index + 1. |
917 | // The remaining upper bits store a partial hashCode of the block data values. |
918 | uint32_t *table = nullptr; |
919 | int32_t capacity = 0; |
920 | int32_t length = 0; |
921 | int32_t shift = 0; |
922 | uint32_t mask = 0; |
923 | |
924 | int32_t blockLength = 0; |
925 | }; |
926 | |
927 | int32_t MutableCodePointTrie::compactWholeDataBlocks(int32_t fastILimit, AllSameBlocks &allSameBlocks) { |
928 | #ifdef UCPTRIE_DEBUG |
929 | bool overflow = false; |
930 | #endif |
931 | |
932 | // ASCII data will be stored as a linear table, even if the following code |
933 | // does not yet count it that way. |
934 | int32_t newDataCapacity = ASCII_LIMIT; |
935 | // Add room for a small data null block in case it would match the start of |
936 | // a fast data block where dataNullOffset must not be set in that case. |
937 | newDataCapacity += UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
938 | // Add room for special values (errorValue, highValue) and padding. |
939 | newDataCapacity += 4; |
940 | int32_t iLimit = highStart >> UCPTRIE_SHIFT_3; |
941 | int32_t blockLength = UCPTRIE_FAST_DATA_BLOCK_LENGTH; |
942 | int32_t inc = SMALL_DATA_BLOCKS_PER_BMP_BLOCK; |
943 | for (int32_t i = 0; i < iLimit; i += inc) { |
944 | if (i == fastILimit) { |
945 | blockLength = UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
946 | inc = 1; |
947 | } |
948 | uint32_t value = index[i]; |
949 | if (flags[i] == MIXED) { |
950 | // Really mixed? |
951 | const uint32_t *p = data + value; |
952 | value = *p; |
953 | if (allValuesSameAs(p + 1, blockLength - 1, value)) { |
954 | flags[i] = ALL_SAME; |
955 | index[i] = value; |
956 | // Fall through to ALL_SAME handling. |
957 | } else { |
958 | newDataCapacity += blockLength; |
959 | continue; |
960 | } |
961 | } else { |
962 | U_ASSERT(flags[i] == ALL_SAME); |
963 | if (inc > 1) { |
964 | // Do all of the fast-range data block's ALL_SAME parts have the same value? |
965 | bool allSame = true; |
966 | int32_t next_i = i + inc; |
967 | for (int32_t j = i + 1; j < next_i; ++j) { |
968 | U_ASSERT(flags[j] == ALL_SAME); |
969 | if (index[j] != value) { |
970 | allSame = false; |
971 | break; |
972 | } |
973 | } |
974 | if (!allSame) { |
975 | // Turn it into a MIXED block. |
976 | if (getDataBlock(i) < 0) { |
977 | return -1; |
978 | } |
979 | newDataCapacity += blockLength; |
980 | continue; |
981 | } |
982 | } |
983 | } |
984 | // Is there another ALL_SAME block with the same value? |
985 | int32_t other = allSameBlocks.findOrAdd(i, inc, value); |
986 | if (other == AllSameBlocks::OVERFLOW) { |
987 | // The fixed-size array overflowed. Slow check for a duplicate block. |
988 | #ifdef UCPTRIE_DEBUG |
989 | if (!overflow) { |
990 | puts("UCPTrie AllSameBlocks overflow" ); |
991 | overflow = true; |
992 | } |
993 | #endif |
994 | int32_t jInc = SMALL_DATA_BLOCKS_PER_BMP_BLOCK; |
995 | for (int32_t j = 0;; j += jInc) { |
996 | if (j == i) { |
997 | allSameBlocks.add(i, inc, value); |
998 | break; |
999 | } |
1000 | if (j == fastILimit) { |
1001 | jInc = 1; |
1002 | } |
1003 | if (flags[j] == ALL_SAME && index[j] == value) { |
1004 | allSameBlocks.add(j, jInc + inc, value); |
1005 | other = j; |
1006 | break; |
1007 | // We could keep counting blocks with the same value |
1008 | // before we add the first one, which may improve compaction in rare cases, |
1009 | // but it would make it slower. |
1010 | } |
1011 | } |
1012 | } |
1013 | if (other >= 0) { |
1014 | flags[i] = SAME_AS; |
1015 | index[i] = other; |
1016 | } else { |
1017 | // New unique same-value block. |
1018 | newDataCapacity += blockLength; |
1019 | } |
1020 | } |
1021 | return newDataCapacity; |
1022 | } |
1023 | |
1024 | #ifdef UCPTRIE_DEBUG |
1025 | # define DEBUG_DO(expr) expr |
1026 | #else |
1027 | # define DEBUG_DO(expr) |
1028 | #endif |
1029 | |
1030 | #ifdef UCPTRIE_DEBUG |
1031 | // Braille symbols: U+28xx = UTF-8 E2 A0 80..E2 A3 BF |
1032 | int32_t appendValue(char s[], int32_t length, uint32_t value) { |
1033 | value ^= value >> 16; |
1034 | value ^= value >> 8; |
1035 | s[length] = 0xE2; |
1036 | s[length + 1] = (char)(0xA0 + ((value >> 6) & 3)); |
1037 | s[length + 2] = (char)(0x80 + (value & 0x3F)); |
1038 | return length + 3; |
1039 | } |
1040 | |
1041 | void printBlock(const uint32_t *block, int32_t blockLength, uint32_t value, |
1042 | UChar32 start, int32_t overlap, uint32_t initialValue) { |
1043 | char s[UCPTRIE_FAST_DATA_BLOCK_LENGTH * 3 + 3]; |
1044 | int32_t length = 0; |
1045 | int32_t i; |
1046 | for (i = 0; i < overlap; ++i) { |
1047 | length = appendValue(s, length, 0); // Braille blank |
1048 | } |
1049 | s[length++] = '|'; |
1050 | for (; i < blockLength; ++i) { |
1051 | if (block != nullptr) { |
1052 | value = block[i]; |
1053 | } |
1054 | if (value == initialValue) { |
1055 | value = 0x40; // Braille lower left dot |
1056 | } |
1057 | length = appendValue(s, length, value); |
1058 | } |
1059 | s[length] = 0; |
1060 | start += overlap; |
1061 | if (start <= 0xffff) { |
1062 | printf(" %04lX %s|\n" , (long)start, s); |
1063 | } else if (start <= 0xfffff) { |
1064 | printf(" %5lX %s|\n" , (long)start, s); |
1065 | } else { |
1066 | printf(" %6lX %s|\n" , (long)start, s); |
1067 | } |
1068 | } |
1069 | #endif |
1070 | |
1071 | /** |
1072 | * Compacts a build-time trie. |
1073 | * |
1074 | * The compaction |
1075 | * - removes blocks that are identical with earlier ones |
1076 | * - overlaps each new non-duplicate block as much as possible with the previously-written one |
1077 | * - works with fast-range data blocks whose length is a multiple of that of |
1078 | * higher-code-point data blocks |
1079 | * |
1080 | * It does not try to find an optimal order of writing, deduplicating, and overlapping blocks. |
1081 | */ |
1082 | int32_t MutableCodePointTrie::compactData( |
1083 | int32_t fastILimit, uint32_t *newData, int32_t newDataCapacity, |
1084 | int32_t dataNullIndex, MixedBlocks &mixedBlocks, UErrorCode &errorCode) { |
1085 | #ifdef UCPTRIE_DEBUG |
1086 | int32_t countSame=0, sumOverlaps=0; |
1087 | bool printData = dataLength == 29088 /* line.brk */ || |
1088 | // dataLength == 30048 /* CanonIterData */ || |
1089 | dataLength == 50400 /* zh.txt~stroke */; |
1090 | #endif |
1091 | |
1092 | // The linear ASCII data has been copied into newData already. |
1093 | int32_t newDataLength = 0; |
1094 | for (int32_t i = 0; newDataLength < ASCII_LIMIT; |
1095 | newDataLength += UCPTRIE_FAST_DATA_BLOCK_LENGTH, i += SMALL_DATA_BLOCKS_PER_BMP_BLOCK) { |
1096 | index[i] = newDataLength; |
1097 | #ifdef UCPTRIE_DEBUG |
1098 | if (printData) { |
1099 | printBlock(newData + newDataLength, UCPTRIE_FAST_DATA_BLOCK_LENGTH, 0, newDataLength, 0, initialValue); |
1100 | } |
1101 | #endif |
1102 | } |
1103 | |
1104 | int32_t blockLength = UCPTRIE_FAST_DATA_BLOCK_LENGTH; |
1105 | if (!mixedBlocks.init(newDataCapacity, blockLength)) { |
1106 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1107 | return 0; |
1108 | } |
1109 | mixedBlocks.extend(newData, 0, 0, newDataLength); |
1110 | |
1111 | int32_t iLimit = highStart >> UCPTRIE_SHIFT_3; |
1112 | int32_t inc = SMALL_DATA_BLOCKS_PER_BMP_BLOCK; |
1113 | int32_t fastLength = 0; |
1114 | for (int32_t i = ASCII_I_LIMIT; i < iLimit; i += inc) { |
1115 | if (i == fastILimit) { |
1116 | blockLength = UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
1117 | inc = 1; |
1118 | fastLength = newDataLength; |
1119 | if (!mixedBlocks.init(newDataCapacity, blockLength)) { |
1120 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1121 | return 0; |
1122 | } |
1123 | mixedBlocks.extend(newData, 0, 0, newDataLength); |
1124 | } |
1125 | if (flags[i] == ALL_SAME) { |
1126 | uint32_t value = index[i]; |
1127 | // Find an earlier part of the data array of length blockLength |
1128 | // that is filled with this value. |
1129 | int32_t n = mixedBlocks.findAllSameBlock(newData, value); |
1130 | // If we find a match, and the current block is the data null block, |
1131 | // and it is not a fast block but matches the start of a fast block, |
1132 | // then we need to continue looking. |
1133 | // This is because this small block is shorter than the fast block, |
1134 | // and not all of the rest of the fast block is filled with this value. |
1135 | // Otherwise trie.getRange() would detect that the fast block starts at |
1136 | // dataNullOffset and assume incorrectly that it is filled with the null value. |
1137 | while (n >= 0 && i == dataNullIndex && i >= fastILimit && n < fastLength && |
1138 | isStartOfSomeFastBlock(n, index, fastILimit)) { |
1139 | n = findAllSameBlock(newData, n + 1, newDataLength, value, blockLength); |
1140 | } |
1141 | if (n >= 0) { |
1142 | DEBUG_DO(++countSame); |
1143 | index[i] = n; |
1144 | } else { |
1145 | n = getAllSameOverlap(newData, newDataLength, value, blockLength); |
1146 | DEBUG_DO(sumOverlaps += n); |
1147 | #ifdef UCPTRIE_DEBUG |
1148 | if (printData) { |
1149 | printBlock(nullptr, blockLength, value, i << UCPTRIE_SHIFT_3, n, initialValue); |
1150 | } |
1151 | #endif |
1152 | index[i] = newDataLength - n; |
1153 | int32_t prevDataLength = newDataLength; |
1154 | while (n < blockLength) { |
1155 | newData[newDataLength++] = value; |
1156 | ++n; |
1157 | } |
1158 | mixedBlocks.extend(newData, 0, prevDataLength, newDataLength); |
1159 | } |
1160 | } else if (flags[i] == MIXED) { |
1161 | const uint32_t *block = data + index[i]; |
1162 | int32_t n = mixedBlocks.findBlock(newData, block, 0); |
1163 | if (n >= 0) { |
1164 | DEBUG_DO(++countSame); |
1165 | index[i] = n; |
1166 | } else { |
1167 | n = getOverlap(newData, newDataLength, block, 0, blockLength); |
1168 | DEBUG_DO(sumOverlaps += n); |
1169 | #ifdef UCPTRIE_DEBUG |
1170 | if (printData) { |
1171 | printBlock(block, blockLength, 0, i << UCPTRIE_SHIFT_3, n, initialValue); |
1172 | } |
1173 | #endif |
1174 | index[i] = newDataLength - n; |
1175 | int32_t prevDataLength = newDataLength; |
1176 | while (n < blockLength) { |
1177 | newData[newDataLength++] = block[n++]; |
1178 | } |
1179 | mixedBlocks.extend(newData, 0, prevDataLength, newDataLength); |
1180 | } |
1181 | } else /* SAME_AS */ { |
1182 | uint32_t j = index[i]; |
1183 | index[i] = index[j]; |
1184 | } |
1185 | } |
1186 | |
1187 | #ifdef UCPTRIE_DEBUG |
1188 | /* we saved some space */ |
1189 | printf("compacting UCPTrie: count of 32-bit data words %lu->%lu countSame=%ld sumOverlaps=%ld\n" , |
1190 | (long)dataLength, (long)newDataLength, (long)countSame, (long)sumOverlaps); |
1191 | #endif |
1192 | return newDataLength; |
1193 | } |
1194 | |
1195 | int32_t MutableCodePointTrie::compactIndex(int32_t fastILimit, MixedBlocks &mixedBlocks, |
1196 | UErrorCode &errorCode) { |
1197 | int32_t fastIndexLength = fastILimit >> (UCPTRIE_FAST_SHIFT - UCPTRIE_SHIFT_3); |
1198 | if ((highStart >> UCPTRIE_FAST_SHIFT) <= fastIndexLength) { |
1199 | // Only the linear fast index, no multi-stage index tables. |
1200 | index3NullOffset = UCPTRIE_NO_INDEX3_NULL_OFFSET; |
1201 | return fastIndexLength; |
1202 | } |
1203 | |
1204 | // Condense the fast index table. |
1205 | // Also, does it contain an index-3 block with all dataNullOffset? |
1206 | uint16_t fastIndex[UCPTRIE_BMP_INDEX_LENGTH]; // fastIndexLength |
1207 | int32_t i3FirstNull = -1; |
1208 | for (int32_t i = 0, j = 0; i < fastILimit; ++j) { |
1209 | uint32_t i3 = index[i]; |
1210 | fastIndex[j] = (uint16_t)i3; |
1211 | if (i3 == (uint32_t)dataNullOffset) { |
1212 | if (i3FirstNull < 0) { |
1213 | i3FirstNull = j; |
1214 | } else if (index3NullOffset < 0 && |
1215 | (j - i3FirstNull + 1) == UCPTRIE_INDEX_3_BLOCK_LENGTH) { |
1216 | index3NullOffset = i3FirstNull; |
1217 | } |
1218 | } else { |
1219 | i3FirstNull = -1; |
1220 | } |
1221 | // Set the index entries that compactData() skipped. |
1222 | // Needed when the multi-stage index covers the fast index range as well. |
1223 | int32_t iNext = i + SMALL_DATA_BLOCKS_PER_BMP_BLOCK; |
1224 | while (++i < iNext) { |
1225 | i3 += UCPTRIE_SMALL_DATA_BLOCK_LENGTH; |
1226 | index[i] = i3; |
1227 | } |
1228 | } |
1229 | |
1230 | if (!mixedBlocks.init(fastIndexLength, UCPTRIE_INDEX_3_BLOCK_LENGTH)) { |
1231 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1232 | return 0; |
1233 | } |
1234 | mixedBlocks.extend(fastIndex, 0, 0, fastIndexLength); |
1235 | |
1236 | // Examine index-3 blocks. For each determine one of: |
1237 | // - same as the index-3 null block |
1238 | // - same as a fast-index block |
1239 | // - 16-bit indexes |
1240 | // - 18-bit indexes |
1241 | // We store this in the first flags entry for the index-3 block. |
1242 | // |
1243 | // Also determine an upper limit for the index-3 table length. |
1244 | int32_t index3Capacity = 0; |
1245 | i3FirstNull = index3NullOffset; |
1246 | bool hasLongI3Blocks = false; |
1247 | // If the fast index covers the whole BMP, then |
1248 | // the multi-stage index is only for supplementary code points. |
1249 | // Otherwise, the multi-stage index covers all of Unicode. |
1250 | int32_t iStart = fastILimit < BMP_I_LIMIT ? 0 : BMP_I_LIMIT; |
1251 | int32_t iLimit = highStart >> UCPTRIE_SHIFT_3; |
1252 | for (int32_t i = iStart; i < iLimit;) { |
1253 | int32_t j = i; |
1254 | int32_t jLimit = i + UCPTRIE_INDEX_3_BLOCK_LENGTH; |
1255 | uint32_t oredI3 = 0; |
1256 | bool isNull = true; |
1257 | do { |
1258 | uint32_t i3 = index[j]; |
1259 | oredI3 |= i3; |
1260 | if (i3 != (uint32_t)dataNullOffset) { |
1261 | isNull = false; |
1262 | } |
1263 | } while (++j < jLimit); |
1264 | if (isNull) { |
1265 | flags[i] = I3_NULL; |
1266 | if (i3FirstNull < 0) { |
1267 | if (oredI3 <= 0xffff) { |
1268 | index3Capacity += UCPTRIE_INDEX_3_BLOCK_LENGTH; |
1269 | } else { |
1270 | index3Capacity += INDEX_3_18BIT_BLOCK_LENGTH; |
1271 | hasLongI3Blocks = true; |
1272 | } |
1273 | i3FirstNull = 0; |
1274 | } |
1275 | } else { |
1276 | if (oredI3 <= 0xffff) { |
1277 | int32_t n = mixedBlocks.findBlock(fastIndex, index, i); |
1278 | if (n >= 0) { |
1279 | flags[i] = I3_BMP; |
1280 | index[i] = n; |
1281 | } else { |
1282 | flags[i] = I3_16; |
1283 | index3Capacity += UCPTRIE_INDEX_3_BLOCK_LENGTH; |
1284 | } |
1285 | } else { |
1286 | flags[i] = I3_18; |
1287 | index3Capacity += INDEX_3_18BIT_BLOCK_LENGTH; |
1288 | hasLongI3Blocks = true; |
1289 | } |
1290 | } |
1291 | i = j; |
1292 | } |
1293 | |
1294 | int32_t index2Capacity = (iLimit - iStart) >> UCPTRIE_SHIFT_2_3; |
1295 | |
1296 | // Length of the index-1 table, rounded up. |
1297 | int32_t index1Length = (index2Capacity + UCPTRIE_INDEX_2_MASK) >> UCPTRIE_SHIFT_1_2; |
1298 | |
1299 | // Index table: Fast index, index-1, index-3, index-2. |
1300 | // +1 for possible index table padding. |
1301 | int32_t index16Capacity = fastIndexLength + index1Length + index3Capacity + index2Capacity + 1; |
1302 | index16 = (uint16_t *)uprv_malloc(index16Capacity * 2); |
1303 | if (index16 == nullptr) { |
1304 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1305 | return 0; |
1306 | } |
1307 | uprv_memcpy(index16, fastIndex, fastIndexLength * 2); |
1308 | |
1309 | if (!mixedBlocks.init(index16Capacity, UCPTRIE_INDEX_3_BLOCK_LENGTH)) { |
1310 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1311 | return 0; |
1312 | } |
1313 | MixedBlocks longI3Blocks; |
1314 | if (hasLongI3Blocks) { |
1315 | if (!longI3Blocks.init(index16Capacity, INDEX_3_18BIT_BLOCK_LENGTH)) { |
1316 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1317 | return 0; |
1318 | } |
1319 | } |
1320 | |
1321 | // Compact the index-3 table and write an uncompacted version of the index-2 table. |
1322 | uint16_t index2[UNICODE_LIMIT >> UCPTRIE_SHIFT_2]; // index2Capacity |
1323 | int32_t i2Length = 0; |
1324 | i3FirstNull = index3NullOffset; |
1325 | int32_t index3Start = fastIndexLength + index1Length; |
1326 | int32_t indexLength = index3Start; |
1327 | for (int32_t i = iStart; i < iLimit; i += UCPTRIE_INDEX_3_BLOCK_LENGTH) { |
1328 | int32_t i3; |
1329 | uint8_t f = flags[i]; |
1330 | if (f == I3_NULL && i3FirstNull < 0) { |
1331 | // First index-3 null block. Write & overlap it like a normal block, then remember it. |
1332 | f = dataNullOffset <= 0xffff ? I3_16 : I3_18; |
1333 | i3FirstNull = 0; |
1334 | } |
1335 | if (f == I3_NULL) { |
1336 | i3 = index3NullOffset; |
1337 | } else if (f == I3_BMP) { |
1338 | i3 = index[i]; |
1339 | } else if (f == I3_16) { |
1340 | int32_t n = mixedBlocks.findBlock(index16, index, i); |
1341 | if (n >= 0) { |
1342 | i3 = n; |
1343 | } else { |
1344 | if (indexLength == index3Start) { |
1345 | // No overlap at the boundary between the index-1 and index-3 tables. |
1346 | n = 0; |
1347 | } else { |
1348 | n = getOverlap(index16, indexLength, |
1349 | index, i, UCPTRIE_INDEX_3_BLOCK_LENGTH); |
1350 | } |
1351 | i3 = indexLength - n; |
1352 | int32_t prevIndexLength = indexLength; |
1353 | while (n < UCPTRIE_INDEX_3_BLOCK_LENGTH) { |
1354 | index16[indexLength++] = index[i + n++]; |
1355 | } |
1356 | mixedBlocks.extend(index16, index3Start, prevIndexLength, indexLength); |
1357 | if (hasLongI3Blocks) { |
1358 | longI3Blocks.extend(index16, index3Start, prevIndexLength, indexLength); |
1359 | } |
1360 | } |
1361 | } else { |
1362 | U_ASSERT(f == I3_18); |
1363 | U_ASSERT(hasLongI3Blocks); |
1364 | // Encode an index-3 block that contains one or more data indexes exceeding 16 bits. |
1365 | int32_t j = i; |
1366 | int32_t jLimit = i + UCPTRIE_INDEX_3_BLOCK_LENGTH; |
1367 | int32_t k = indexLength; |
1368 | do { |
1369 | ++k; |
1370 | uint32_t v = index[j++]; |
1371 | uint32_t upperBits = (v & 0x30000) >> 2; |
1372 | index16[k++] = v; |
1373 | v = index[j++]; |
1374 | upperBits |= (v & 0x30000) >> 4; |
1375 | index16[k++] = v; |
1376 | v = index[j++]; |
1377 | upperBits |= (v & 0x30000) >> 6; |
1378 | index16[k++] = v; |
1379 | v = index[j++]; |
1380 | upperBits |= (v & 0x30000) >> 8; |
1381 | index16[k++] = v; |
1382 | v = index[j++]; |
1383 | upperBits |= (v & 0x30000) >> 10; |
1384 | index16[k++] = v; |
1385 | v = index[j++]; |
1386 | upperBits |= (v & 0x30000) >> 12; |
1387 | index16[k++] = v; |
1388 | v = index[j++]; |
1389 | upperBits |= (v & 0x30000) >> 14; |
1390 | index16[k++] = v; |
1391 | v = index[j++]; |
1392 | upperBits |= (v & 0x30000) >> 16; |
1393 | index16[k++] = v; |
1394 | index16[k - 9] = upperBits; |
1395 | } while (j < jLimit); |
1396 | int32_t n = longI3Blocks.findBlock(index16, index16, indexLength); |
1397 | if (n >= 0) { |
1398 | i3 = n | 0x8000; |
1399 | } else { |
1400 | if (indexLength == index3Start) { |
1401 | // No overlap at the boundary between the index-1 and index-3 tables. |
1402 | n = 0; |
1403 | } else { |
1404 | n = getOverlap(index16, indexLength, |
1405 | index16, indexLength, INDEX_3_18BIT_BLOCK_LENGTH); |
1406 | } |
1407 | i3 = (indexLength - n) | 0x8000; |
1408 | int32_t prevIndexLength = indexLength; |
1409 | if (n > 0) { |
1410 | int32_t start = indexLength; |
1411 | while (n < INDEX_3_18BIT_BLOCK_LENGTH) { |
1412 | index16[indexLength++] = index16[start + n++]; |
1413 | } |
1414 | } else { |
1415 | indexLength += INDEX_3_18BIT_BLOCK_LENGTH; |
1416 | } |
1417 | mixedBlocks.extend(index16, index3Start, prevIndexLength, indexLength); |
1418 | if (hasLongI3Blocks) { |
1419 | longI3Blocks.extend(index16, index3Start, prevIndexLength, indexLength); |
1420 | } |
1421 | } |
1422 | } |
1423 | if (index3NullOffset < 0 && i3FirstNull >= 0) { |
1424 | index3NullOffset = i3; |
1425 | } |
1426 | // Set the index-2 table entry. |
1427 | index2[i2Length++] = i3; |
1428 | } |
1429 | U_ASSERT(i2Length == index2Capacity); |
1430 | U_ASSERT(indexLength <= index3Start + index3Capacity); |
1431 | |
1432 | if (index3NullOffset < 0) { |
1433 | index3NullOffset = UCPTRIE_NO_INDEX3_NULL_OFFSET; |
1434 | } |
1435 | if (indexLength >= (UCPTRIE_NO_INDEX3_NULL_OFFSET + UCPTRIE_INDEX_3_BLOCK_LENGTH)) { |
1436 | // The index-3 offsets exceed 15 bits, or |
1437 | // the last one cannot be distinguished from the no-null-block value. |
1438 | errorCode = U_INDEX_OUTOFBOUNDS_ERROR; |
1439 | return 0; |
1440 | } |
1441 | |
1442 | // Compact the index-2 table and write the index-1 table. |
1443 | static_assert(UCPTRIE_INDEX_2_BLOCK_LENGTH == UCPTRIE_INDEX_3_BLOCK_LENGTH, |
1444 | "must re-init mixedBlocks" ); |
1445 | int32_t blockLength = UCPTRIE_INDEX_2_BLOCK_LENGTH; |
1446 | int32_t i1 = fastIndexLength; |
1447 | for (int32_t i = 0; i < i2Length; i += blockLength) { |
1448 | int32_t n; |
1449 | if ((i2Length - i) >= blockLength) { |
1450 | // normal block |
1451 | U_ASSERT(blockLength == UCPTRIE_INDEX_2_BLOCK_LENGTH); |
1452 | n = mixedBlocks.findBlock(index16, index2, i); |
1453 | } else { |
1454 | // highStart is inside the last index-2 block. Shorten it. |
1455 | blockLength = i2Length - i; |
1456 | n = findSameBlock(index16, index3Start, indexLength, |
1457 | index2, i, blockLength); |
1458 | } |
1459 | int32_t i2; |
1460 | if (n >= 0) { |
1461 | i2 = n; |
1462 | } else { |
1463 | if (indexLength == index3Start) { |
1464 | // No overlap at the boundary between the index-1 and index-3/2 tables. |
1465 | n = 0; |
1466 | } else { |
1467 | n = getOverlap(index16, indexLength, index2, i, blockLength); |
1468 | } |
1469 | i2 = indexLength - n; |
1470 | int32_t prevIndexLength = indexLength; |
1471 | while (n < blockLength) { |
1472 | index16[indexLength++] = index2[i + n++]; |
1473 | } |
1474 | mixedBlocks.extend(index16, index3Start, prevIndexLength, indexLength); |
1475 | } |
1476 | // Set the index-1 table entry. |
1477 | index16[i1++] = i2; |
1478 | } |
1479 | U_ASSERT(i1 == index3Start); |
1480 | U_ASSERT(indexLength <= index16Capacity); |
1481 | |
1482 | #ifdef UCPTRIE_DEBUG |
1483 | /* we saved some space */ |
1484 | printf("compacting UCPTrie: count of 16-bit index words %lu->%lu\n" , |
1485 | (long)iLimit, (long)indexLength); |
1486 | #endif |
1487 | |
1488 | return indexLength; |
1489 | } |
1490 | |
1491 | int32_t MutableCodePointTrie::compactTrie(int32_t fastILimit, UErrorCode &errorCode) { |
1492 | // Find the real highStart and round it up. |
1493 | U_ASSERT((highStart & (UCPTRIE_CP_PER_INDEX_2_ENTRY - 1)) == 0); |
1494 | highValue = get(MAX_UNICODE); |
1495 | int32_t realHighStart = findHighStart(); |
1496 | realHighStart = (realHighStart + (UCPTRIE_CP_PER_INDEX_2_ENTRY - 1)) & |
1497 | ~(UCPTRIE_CP_PER_INDEX_2_ENTRY - 1); |
1498 | if (realHighStart == UNICODE_LIMIT) { |
1499 | highValue = initialValue; |
1500 | } |
1501 | |
1502 | #ifdef UCPTRIE_DEBUG |
1503 | printf("UCPTrie: highStart U+%06lx highValue 0x%lx initialValue 0x%lx\n" , |
1504 | (long)realHighStart, (long)highValue, (long)initialValue); |
1505 | #endif |
1506 | |
1507 | // We always store indexes and data values for the fast range. |
1508 | // Pin highStart to the top of that range while building. |
1509 | UChar32 fastLimit = fastILimit << UCPTRIE_SHIFT_3; |
1510 | if (realHighStart < fastLimit) { |
1511 | for (int32_t i = (realHighStart >> UCPTRIE_SHIFT_3); i < fastILimit; ++i) { |
1512 | flags[i] = ALL_SAME; |
1513 | index[i] = highValue; |
1514 | } |
1515 | highStart = fastLimit; |
1516 | } else { |
1517 | highStart = realHighStart; |
1518 | } |
1519 | |
1520 | uint32_t asciiData[ASCII_LIMIT]; |
1521 | for (int32_t i = 0; i < ASCII_LIMIT; ++i) { |
1522 | asciiData[i] = get(i); |
1523 | } |
1524 | |
1525 | // First we look for which data blocks have the same value repeated over the whole block, |
1526 | // deduplicate such blocks, find a good null data block (for faster enumeration), |
1527 | // and get an upper bound for the necessary data array length. |
1528 | AllSameBlocks allSameBlocks; |
1529 | int32_t newDataCapacity = compactWholeDataBlocks(fastILimit, allSameBlocks); |
1530 | if (newDataCapacity < 0) { |
1531 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1532 | return 0; |
1533 | } |
1534 | uint32_t *newData = (uint32_t *)uprv_malloc(newDataCapacity * 4); |
1535 | if (newData == nullptr) { |
1536 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1537 | return 0; |
1538 | } |
1539 | uprv_memcpy(newData, asciiData, sizeof(asciiData)); |
1540 | |
1541 | int32_t dataNullIndex = allSameBlocks.findMostUsed(); |
1542 | |
1543 | MixedBlocks mixedBlocks; |
1544 | int32_t newDataLength = compactData(fastILimit, newData, newDataCapacity, |
1545 | dataNullIndex, mixedBlocks, errorCode); |
1546 | if (U_FAILURE(errorCode)) { return 0; } |
1547 | U_ASSERT(newDataLength <= newDataCapacity); |
1548 | uprv_free(data); |
1549 | data = newData; |
1550 | dataCapacity = newDataCapacity; |
1551 | dataLength = newDataLength; |
1552 | if (dataLength > (0x3ffff + UCPTRIE_SMALL_DATA_BLOCK_LENGTH)) { |
1553 | // The offset of the last data block is too high to be stored in the index table. |
1554 | errorCode = U_INDEX_OUTOFBOUNDS_ERROR; |
1555 | return 0; |
1556 | } |
1557 | |
1558 | if (dataNullIndex >= 0) { |
1559 | dataNullOffset = index[dataNullIndex]; |
1560 | #ifdef UCPTRIE_DEBUG |
1561 | if (data[dataNullOffset] != initialValue) { |
1562 | printf("UCPTrie initialValue %lx -> more common nullValue %lx\n" , |
1563 | (long)initialValue, (long)data[dataNullOffset]); |
1564 | } |
1565 | #endif |
1566 | initialValue = data[dataNullOffset]; |
1567 | } else { |
1568 | dataNullOffset = UCPTRIE_NO_DATA_NULL_OFFSET; |
1569 | } |
1570 | |
1571 | int32_t indexLength = compactIndex(fastILimit, mixedBlocks, errorCode); |
1572 | highStart = realHighStart; |
1573 | return indexLength; |
1574 | } |
1575 | |
1576 | UCPTrie *MutableCodePointTrie::build(UCPTrieType type, UCPTrieValueWidth valueWidth, UErrorCode &errorCode) { |
1577 | if (U_FAILURE(errorCode)) { |
1578 | return nullptr; |
1579 | } |
1580 | if (type < UCPTRIE_TYPE_FAST || UCPTRIE_TYPE_SMALL < type || |
1581 | valueWidth < UCPTRIE_VALUE_BITS_16 || UCPTRIE_VALUE_BITS_8 < valueWidth) { |
1582 | errorCode = U_ILLEGAL_ARGUMENT_ERROR; |
1583 | return nullptr; |
1584 | } |
1585 | |
1586 | // The mutable trie always stores 32-bit values. |
1587 | // When we build a UCPTrie for a smaller value width, we first mask off unused bits |
1588 | // before compacting the data. |
1589 | switch (valueWidth) { |
1590 | case UCPTRIE_VALUE_BITS_32: |
1591 | break; |
1592 | case UCPTRIE_VALUE_BITS_16: |
1593 | maskValues(0xffff); |
1594 | break; |
1595 | case UCPTRIE_VALUE_BITS_8: |
1596 | maskValues(0xff); |
1597 | break; |
1598 | default: |
1599 | break; |
1600 | } |
1601 | |
1602 | UChar32 fastLimit = type == UCPTRIE_TYPE_FAST ? BMP_LIMIT : UCPTRIE_SMALL_LIMIT; |
1603 | int32_t indexLength = compactTrie(fastLimit >> UCPTRIE_SHIFT_3, errorCode); |
1604 | if (U_FAILURE(errorCode)) { |
1605 | clear(); |
1606 | return nullptr; |
1607 | } |
1608 | |
1609 | // Ensure data table alignment: The index length must be even for uint32_t data. |
1610 | if (valueWidth == UCPTRIE_VALUE_BITS_32 && (indexLength & 1) != 0) { |
1611 | index16[indexLength++] = 0xffee; // arbitrary value |
1612 | } |
1613 | |
1614 | // Make the total trie structure length a multiple of 4 bytes by padding the data table, |
1615 | // and store special values as the last two data values. |
1616 | int32_t length = indexLength * 2; |
1617 | if (valueWidth == UCPTRIE_VALUE_BITS_16) { |
1618 | if (((indexLength ^ dataLength) & 1) != 0) { |
1619 | // padding |
1620 | data[dataLength++] = errorValue; |
1621 | } |
1622 | if (data[dataLength - 1] != errorValue || data[dataLength - 2] != highValue) { |
1623 | data[dataLength++] = highValue; |
1624 | data[dataLength++] = errorValue; |
1625 | } |
1626 | length += dataLength * 2; |
1627 | } else if (valueWidth == UCPTRIE_VALUE_BITS_32) { |
1628 | // 32-bit data words never need padding to a multiple of 4 bytes. |
1629 | if (data[dataLength - 1] != errorValue || data[dataLength - 2] != highValue) { |
1630 | if (data[dataLength - 1] != highValue) { |
1631 | data[dataLength++] = highValue; |
1632 | } |
1633 | data[dataLength++] = errorValue; |
1634 | } |
1635 | length += dataLength * 4; |
1636 | } else { |
1637 | int32_t and3 = (length + dataLength) & 3; |
1638 | if (and3 == 0 && data[dataLength - 1] == errorValue && data[dataLength - 2] == highValue) { |
1639 | // all set |
1640 | } else if(and3 == 3 && data[dataLength - 1] == highValue) { |
1641 | data[dataLength++] = errorValue; |
1642 | } else { |
1643 | while (and3 != 2) { |
1644 | data[dataLength++] = highValue; |
1645 | and3 = (and3 + 1) & 3; |
1646 | } |
1647 | data[dataLength++] = highValue; |
1648 | data[dataLength++] = errorValue; |
1649 | } |
1650 | length += dataLength; |
1651 | } |
1652 | |
1653 | // Calculate the total length of the UCPTrie as a single memory block. |
1654 | length += sizeof(UCPTrie); |
1655 | U_ASSERT((length & 3) == 0); |
1656 | |
1657 | uint8_t *bytes = (uint8_t *)uprv_malloc(length); |
1658 | if (bytes == nullptr) { |
1659 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
1660 | clear(); |
1661 | return nullptr; |
1662 | } |
1663 | UCPTrie *trie = reinterpret_cast<UCPTrie *>(bytes); |
1664 | uprv_memset(trie, 0, sizeof(UCPTrie)); |
1665 | trie->indexLength = indexLength; |
1666 | trie->dataLength = dataLength; |
1667 | |
1668 | trie->highStart = highStart; |
1669 | // Round up shifted12HighStart to a multiple of 0x1000 for easy testing from UTF-8 lead bytes. |
1670 | // Runtime code needs to then test for the real highStart as well. |
1671 | trie->shifted12HighStart = (highStart + 0xfff) >> 12; |
1672 | trie->type = type; |
1673 | trie->valueWidth = valueWidth; |
1674 | |
1675 | trie->index3NullOffset = index3NullOffset; |
1676 | trie->dataNullOffset = dataNullOffset; |
1677 | trie->nullValue = initialValue; |
1678 | |
1679 | bytes += sizeof(UCPTrie); |
1680 | |
1681 | // Fill the index and data arrays. |
1682 | uint16_t *dest16 = (uint16_t *)bytes; |
1683 | trie->index = dest16; |
1684 | |
1685 | if (highStart <= fastLimit) { |
1686 | // Condense only the fast index from the mutable-trie index. |
1687 | for (int32_t i = 0, j = 0; j < indexLength; i += SMALL_DATA_BLOCKS_PER_BMP_BLOCK, ++j) { |
1688 | *dest16++ = (uint16_t)index[i]; // dest16[j] |
1689 | } |
1690 | } else { |
1691 | uprv_memcpy(dest16, index16, indexLength * 2); |
1692 | dest16 += indexLength; |
1693 | } |
1694 | bytes += indexLength * 2; |
1695 | |
1696 | // Write the data array. |
1697 | const uint32_t *p = data; |
1698 | switch (valueWidth) { |
1699 | case UCPTRIE_VALUE_BITS_16: |
1700 | // Write 16-bit data values. |
1701 | trie->data.ptr16 = dest16; |
1702 | for (int32_t i = dataLength; i > 0; --i) { |
1703 | *dest16++ = (uint16_t)*p++; |
1704 | } |
1705 | break; |
1706 | case UCPTRIE_VALUE_BITS_32: |
1707 | // Write 32-bit data values. |
1708 | trie->data.ptr32 = (uint32_t *)bytes; |
1709 | uprv_memcpy(bytes, p, (size_t)dataLength * 4); |
1710 | break; |
1711 | case UCPTRIE_VALUE_BITS_8: |
1712 | // Write 8-bit data values. |
1713 | trie->data.ptr8 = bytes; |
1714 | for (int32_t i = dataLength; i > 0; --i) { |
1715 | *bytes++ = (uint8_t)*p++; |
1716 | } |
1717 | break; |
1718 | default: |
1719 | // Will not occur, valueWidth checked at the beginning. |
1720 | break; |
1721 | } |
1722 | |
1723 | #ifdef UCPTRIE_DEBUG |
1724 | trie->name = name; |
1725 | |
1726 | ucptrie_printLengths(trie, "" ); |
1727 | #endif |
1728 | |
1729 | clear(); |
1730 | return trie; |
1731 | } |
1732 | |
1733 | } // namespace |
1734 | |
1735 | U_NAMESPACE_END |
1736 | |
1737 | U_NAMESPACE_USE |
1738 | |
1739 | U_CAPI UMutableCPTrie * U_EXPORT2 |
1740 | umutablecptrie_open(uint32_t initialValue, uint32_t errorValue, UErrorCode *pErrorCode) { |
1741 | if (U_FAILURE(*pErrorCode)) { |
1742 | return nullptr; |
1743 | } |
1744 | LocalPointer<MutableCodePointTrie> trie( |
1745 | new MutableCodePointTrie(initialValue, errorValue, *pErrorCode), *pErrorCode); |
1746 | if (U_FAILURE(*pErrorCode)) { |
1747 | return nullptr; |
1748 | } |
1749 | return reinterpret_cast<UMutableCPTrie *>(trie.orphan()); |
1750 | } |
1751 | |
1752 | U_CAPI UMutableCPTrie * U_EXPORT2 |
1753 | umutablecptrie_clone(const UMutableCPTrie *other, UErrorCode *pErrorCode) { |
1754 | if (U_FAILURE(*pErrorCode)) { |
1755 | return nullptr; |
1756 | } |
1757 | if (other == nullptr) { |
1758 | return nullptr; |
1759 | } |
1760 | LocalPointer<MutableCodePointTrie> clone( |
1761 | new MutableCodePointTrie(*reinterpret_cast<const MutableCodePointTrie *>(other), *pErrorCode), *pErrorCode); |
1762 | if (U_FAILURE(*pErrorCode)) { |
1763 | return nullptr; |
1764 | } |
1765 | return reinterpret_cast<UMutableCPTrie *>(clone.orphan()); |
1766 | } |
1767 | |
1768 | U_CAPI void U_EXPORT2 |
1769 | umutablecptrie_close(UMutableCPTrie *trie) { |
1770 | delete reinterpret_cast<MutableCodePointTrie *>(trie); |
1771 | } |
1772 | |
1773 | U_CAPI UMutableCPTrie * U_EXPORT2 |
1774 | umutablecptrie_fromUCPMap(const UCPMap *map, UErrorCode *pErrorCode) { |
1775 | if (U_FAILURE(*pErrorCode)) { |
1776 | return nullptr; |
1777 | } |
1778 | if (map == nullptr) { |
1779 | *pErrorCode = U_ILLEGAL_ARGUMENT_ERROR; |
1780 | return nullptr; |
1781 | } |
1782 | return reinterpret_cast<UMutableCPTrie *>(MutableCodePointTrie::fromUCPMap(map, *pErrorCode)); |
1783 | } |
1784 | |
1785 | U_CAPI UMutableCPTrie * U_EXPORT2 |
1786 | umutablecptrie_fromUCPTrie(const UCPTrie *trie, UErrorCode *pErrorCode) { |
1787 | if (U_FAILURE(*pErrorCode)) { |
1788 | return nullptr; |
1789 | } |
1790 | if (trie == nullptr) { |
1791 | *pErrorCode = U_ILLEGAL_ARGUMENT_ERROR; |
1792 | return nullptr; |
1793 | } |
1794 | return reinterpret_cast<UMutableCPTrie *>(MutableCodePointTrie::fromUCPTrie(trie, *pErrorCode)); |
1795 | } |
1796 | |
1797 | U_CAPI uint32_t U_EXPORT2 |
1798 | umutablecptrie_get(const UMutableCPTrie *trie, UChar32 c) { |
1799 | return reinterpret_cast<const MutableCodePointTrie *>(trie)->get(c); |
1800 | } |
1801 | |
1802 | namespace { |
1803 | |
1804 | UChar32 getRange(const void *trie, UChar32 start, |
1805 | UCPMapValueFilter *filter, const void *context, uint32_t *pValue) { |
1806 | return reinterpret_cast<const MutableCodePointTrie *>(trie)-> |
1807 | getRange(start, filter, context, pValue); |
1808 | } |
1809 | |
1810 | } // namespace |
1811 | |
1812 | U_CAPI UChar32 U_EXPORT2 |
1813 | umutablecptrie_getRange(const UMutableCPTrie *trie, UChar32 start, |
1814 | UCPMapRangeOption option, uint32_t surrogateValue, |
1815 | UCPMapValueFilter *filter, const void *context, uint32_t *pValue) { |
1816 | return ucptrie_internalGetRange(getRange, trie, start, |
1817 | option, surrogateValue, |
1818 | filter, context, pValue); |
1819 | } |
1820 | |
1821 | U_CAPI void U_EXPORT2 |
1822 | umutablecptrie_set(UMutableCPTrie *trie, UChar32 c, uint32_t value, UErrorCode *pErrorCode) { |
1823 | if (U_FAILURE(*pErrorCode)) { |
1824 | return; |
1825 | } |
1826 | reinterpret_cast<MutableCodePointTrie *>(trie)->set(c, value, *pErrorCode); |
1827 | } |
1828 | |
1829 | U_CAPI void U_EXPORT2 |
1830 | umutablecptrie_setRange(UMutableCPTrie *trie, UChar32 start, UChar32 end, |
1831 | uint32_t value, UErrorCode *pErrorCode) { |
1832 | if (U_FAILURE(*pErrorCode)) { |
1833 | return; |
1834 | } |
1835 | reinterpret_cast<MutableCodePointTrie *>(trie)->setRange(start, end, value, *pErrorCode); |
1836 | } |
1837 | |
1838 | /* Compact and internally serialize the trie. */ |
1839 | U_CAPI UCPTrie * U_EXPORT2 |
1840 | umutablecptrie_buildImmutable(UMutableCPTrie *trie, UCPTrieType type, UCPTrieValueWidth valueWidth, |
1841 | UErrorCode *pErrorCode) { |
1842 | if (U_FAILURE(*pErrorCode)) { |
1843 | return nullptr; |
1844 | } |
1845 | return reinterpret_cast<MutableCodePointTrie *>(trie)->build(type, valueWidth, *pErrorCode); |
1846 | } |
1847 | |
1848 | #ifdef UCPTRIE_DEBUG |
1849 | U_CFUNC void umutablecptrie_setName(UMutableCPTrie *trie, const char *name) { |
1850 | reinterpret_cast<MutableCodePointTrie *>(trie)->name = name; |
1851 | } |
1852 | #endif |
1853 | |