1 /*
2 **
3 ** Copyright 2010, The Android Open Source Project
4 **
5 ** Licensed under the Apache License, Version 2.0 (the "License");
6 ** you may not use this file except in compliance with the License.
7 ** You may obtain a copy of the License at
8 **
9 ** http://www.apache.org/licenses/LICENSE-2.0
10 **
11 ** Unless required by applicable law or agreed to in writing, software
12 ** distributed under the License is distributed on an "AS IS" BASIS,
13 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 ** See the License for the specific language governing permissions and
15 ** limitations under the License.
16 */
17
18 #ifndef LATINIME_DEFINES_H
19 #define LATINIME_DEFINES_H
20
21 #if defined(FLAG_DO_PROFILE) || defined(FLAG_DBG)
22 #include <cutils/log.h>
23 #define AKLOGE ALOGE
24 #define AKLOGI ALOGI
25
26 #define DUMP_WORD(word, length) do { dumpWord(word, length); } while(0)
27 #define DUMP_WORD_INT(word, length) do { dumpWordInt(word, length); } while(0)
28
dumpWord(const unsigned short * word,const int length)29 static inline void dumpWord(const unsigned short* word, const int length) {
30 static char charBuf[50];
31
32 for (int i = 0; i < length; ++i) {
33 charBuf[i] = word[i];
34 }
35 charBuf[length] = 0;
36 AKLOGI("[ %s ]", charBuf);
37 }
38
dumpWordInt(const int * word,const int length)39 static inline void dumpWordInt(const int* word, const int length) {
40 static char charBuf[50];
41
42 for (int i = 0; i < length; ++i) {
43 charBuf[i] = word[i];
44 }
45 charBuf[length] = 0;
46 AKLOGI("i[ %s ]", charBuf);
47 }
48
49 #else
50 #define AKLOGE(fmt, ...)
51 #define AKLOGI(fmt, ...)
52 #define DUMP_WORD(word, length)
53 #define DUMP_WORD_INT(word, length)
54 #endif
55
56 #ifdef FLAG_DO_PROFILE
57 // Profiler
58 #include <time.h>
59
60 #define PROF_BUF_SIZE 100
61 static float profile_buf[PROF_BUF_SIZE];
62 static float profile_old[PROF_BUF_SIZE];
63 static unsigned int profile_counter[PROF_BUF_SIZE];
64
65 #define PROF_RESET prof_reset()
66 #define PROF_COUNT(prof_buf_id) ++profile_counter[prof_buf_id]
67 #define PROF_OPEN do { PROF_RESET; PROF_START(PROF_BUF_SIZE - 1); } while(0)
68 #define PROF_START(prof_buf_id) do { \
69 PROF_COUNT(prof_buf_id); profile_old[prof_buf_id] = (clock()); } while(0)
70 #define PROF_CLOSE do { PROF_END(PROF_BUF_SIZE - 1); PROF_OUTALL; } while(0)
71 #define PROF_END(prof_buf_id) profile_buf[prof_buf_id] += ((clock()) - profile_old[prof_buf_id])
72 #define PROF_CLOCKOUT(prof_buf_id) \
73 AKLOGI("%s : clock is %f", __FUNCTION__, (clock() - profile_old[prof_buf_id]))
74 #define PROF_OUTALL do { AKLOGI("--- %s ---", __FUNCTION__); prof_out(); } while(0)
75
prof_reset(void)76 static inline void prof_reset(void) {
77 for (int i = 0; i < PROF_BUF_SIZE; ++i) {
78 profile_buf[i] = 0;
79 profile_old[i] = 0;
80 profile_counter[i] = 0;
81 }
82 }
83
prof_out(void)84 static inline void prof_out(void) {
85 if (profile_counter[PROF_BUF_SIZE - 1] != 1) {
86 AKLOGI("Error: You must call PROF_OPEN before PROF_CLOSE.");
87 }
88 AKLOGI("Total time is %6.3f ms.",
89 profile_buf[PROF_BUF_SIZE - 1] * 1000 / (float)CLOCKS_PER_SEC);
90 float all = 0;
91 for (int i = 0; i < PROF_BUF_SIZE - 1; ++i) {
92 all += profile_buf[i];
93 }
94 if (all == 0) all = 1;
95 for (int i = 0; i < PROF_BUF_SIZE - 1; ++i) {
96 if (profile_buf[i] != 0) {
97 AKLOGI("(%d): Used %4.2f%%, %8.4f ms. Called %d times.",
98 i, (profile_buf[i] * 100 / all),
99 profile_buf[i] * 1000 / (float)CLOCKS_PER_SEC, profile_counter[i]);
100 }
101 }
102 }
103
104 #else // FLAG_DO_PROFILE
105 #define PROF_BUF_SIZE 0
106 #define PROF_RESET
107 #define PROF_COUNT(prof_buf_id)
108 #define PROF_OPEN
109 #define PROF_START(prof_buf_id)
110 #define PROF_CLOSE
111 #define PROF_END(prof_buf_id)
112 #define PROF_CLOCK_OUT(prof_buf_id)
113 #define PROF_CLOCKOUT(prof_buf_id)
114 #define PROF_OUTALL
115
116 #endif // FLAG_DO_PROFILE
117
118 #ifdef FLAG_DBG
119 #include <cutils/log.h>
120 #ifndef LOG_TAG
121 #define LOG_TAG "LatinIME: "
122 #endif
123 #define DEBUG_DICT true
124 #define DEBUG_DICT_FULL false
125 #define DEBUG_EDIT_DISTANCE false
126 #define DEBUG_SHOW_FOUND_WORD false
127 #define DEBUG_NODE DEBUG_DICT_FULL
128 #define DEBUG_TRACE DEBUG_DICT_FULL
129 #define DEBUG_PROXIMITY_INFO false
130 #define DEBUG_PROXIMITY_CHARS false
131 #define DEBUG_CORRECTION false
132 #define DEBUG_CORRECTION_FREQ false
133 #define DEBUG_WORDS_PRIORITY_QUEUE false
134
135 #else // FLAG_DBG
136
137 #define DEBUG_DICT false
138 #define DEBUG_DICT_FULL false
139 #define DEBUG_EDIT_DISTANCE false
140 #define DEBUG_SHOW_FOUND_WORD false
141 #define DEBUG_NODE false
142 #define DEBUG_TRACE false
143 #define DEBUG_PROXIMITY_INFO false
144 #define DEBUG_PROXIMITY_CHARS false
145 #define DEBUG_CORRECTION false
146 #define DEBUG_CORRECTION_FREQ false
147 #define DEBUG_WORDS_PRIORITY_QUEUE false
148
149
150 #endif // FLAG_DBG
151
152 #ifndef U_SHORT_MAX
153 #define U_SHORT_MAX 65535 // ((1 << 16) - 1)
154 #endif
155 #ifndef S_INT_MAX
156 #define S_INT_MAX 2147483647 // ((1 << 31) - 1)
157 #endif
158
159 // Define this to use mmap() for dictionary loading. Undefine to use malloc() instead of mmap().
160 // We measured and compared performance of both, and found mmap() is fairly good in terms of
161 // loading time, and acceptable even for several initial lookups which involve page faults.
162 #define USE_MMAP_FOR_DICTIONARY
163
164 // 22-bit address = ~4MB dictionary size limit, which on average would be about 200k-300k words
165 #define ADDRESS_MASK 0x3FFFFF
166
167 // The bit that decides if an address follows in the next 22 bits
168 #define FLAG_ADDRESS_MASK 0x40
169 // The bit that decides if this is a terminal node for a word. The node could still have children,
170 // if the word has other endings.
171 #define FLAG_TERMINAL_MASK 0x80
172
173 #define FLAG_BIGRAM_READ 0x80
174 #define FLAG_BIGRAM_CHILDEXIST 0x40
175 #define FLAG_BIGRAM_CONTINUED 0x80
176 #define FLAG_BIGRAM_FREQ 0x7F
177
178 #define DICTIONARY_VERSION_MIN 200
179 #define NOT_VALID_WORD -99
180 #define NOT_A_CHARACTER -1
181 #define NOT_A_DISTANCE -1
182 #define NOT_A_COORDINATE -1
183 #define EQUIVALENT_CHAR_WITHOUT_DISTANCE_INFO -2
184 #define PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO -3
185 #define ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO -4
186 #define NOT_AN_INDEX -1
187 #define NOT_A_PROBABILITY -1
188
189 #define KEYCODE_SPACE ' '
190
191 #define CALIBRATE_SCORE_BY_TOUCH_COORDINATES true
192
193 #define SUGGEST_WORDS_WITH_MISSING_CHARACTER true
194 #define SUGGEST_WORDS_WITH_EXCESSIVE_CHARACTER true
195 #define SUGGEST_WORDS_WITH_TRANSPOSED_CHARACTERS true
196 #define SUGGEST_MULTIPLE_WORDS true
197
198 // The following "rate"s are used as a multiplier before dividing by 100, so they are in percent.
199 #define WORDS_WITH_MISSING_CHARACTER_DEMOTION_RATE 80
200 #define WORDS_WITH_MISSING_CHARACTER_DEMOTION_START_POS_10X 12
201 #define WORDS_WITH_MISSING_SPACE_CHARACTER_DEMOTION_RATE 58
202 #define WORDS_WITH_MISTYPED_SPACE_DEMOTION_RATE 50
203 #define WORDS_WITH_EXCESSIVE_CHARACTER_DEMOTION_RATE 75
204 #define WORDS_WITH_EXCESSIVE_CHARACTER_OUT_OF_PROXIMITY_DEMOTION_RATE 75
205 #define WORDS_WITH_TRANSPOSED_CHARACTERS_DEMOTION_RATE 70
206 #define FULL_MATCHED_WORDS_PROMOTION_RATE 120
207 #define WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE 90
208 #define WORDS_WITH_ADDITIONAL_PROXIMITY_CHARACTER_DEMOTION_RATE 70
209 #define WORDS_WITH_MATCH_SKIP_PROMOTION_RATE 105
210 #define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE 148
211 #define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_MULTIPLIER 3
212 #define CORRECTION_COUNT_RATE_DEMOTION_RATE_BASE 45
213 #define INPUT_EXCEEDS_OUTPUT_DEMOTION_RATE 70
214 #define FIRST_CHAR_DIFFERENT_DEMOTION_RATE 96
215 #define TWO_WORDS_CAPITALIZED_DEMOTION_RATE 50
216 #define TWO_WORDS_CORRECTION_DEMOTION_BASE 80
217 #define TWO_WORDS_PLUS_OTHER_ERROR_CORRECTION_DEMOTION_DIVIDER 1
218 #define ZERO_DISTANCE_PROMOTION_RATE 110
219 #define NEUTRAL_SCORE_SQUARED_RADIUS 8.0f
220 #define HALF_SCORE_SQUARED_RADIUS 32.0f
221 #define MAX_FREQ 255
222 #define MAX_BIGRAM_FREQ 15
223
224 // This must be greater than or equal to MAX_WORD_LENGTH defined in BinaryDictionary.java
225 // This is only used for the size of array. Not to be used in c functions.
226 #define MAX_WORD_LENGTH_INTERNAL 48
227
228 // This must be equal to ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE in KeyDetector.java
229 #define ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE 2
230
231 // Word limit for sub queues used in WordsPriorityQueuePool. Sub queues are temporary queues used
232 // for better performance.
233 // Holds up to 1 candidate for each word
234 #define SUB_QUEUE_MAX_WORDS 1
235 #define SUB_QUEUE_MAX_COUNT 10
236 #define SUB_QUEUE_MIN_WORD_LENGTH 4
237 // TODO: Extend this limitation
238 #define MULTIPLE_WORDS_SUGGESTION_MAX_WORDS 5
239 // TODO: Remove this limitation
240 #define MULTIPLE_WORDS_SUGGESTION_MAX_WORD_LENGTH 12
241 // TODO: Remove this limitation
242 #define MULTIPLE_WORDS_SUGGESTION_MAX_TOTAL_TRAVERSE_COUNT 45
243 #define MULTIPLE_WORDS_DEMOTION_RATE 80
244 #define MIN_INPUT_LENGTH_FOR_THREE_OR_MORE_WORDS_CORRECTION 6
245
246 #define TWO_WORDS_CORRECTION_WITH_OTHER_ERROR_THRESHOLD 0.35
247 #define START_TWO_WORDS_CORRECTION_THRESHOLD 0.185
248 /* heuristic... This should be changed if we change the unit of the frequency. */
249 #define SUPPRESS_SHORT_MULTIPLE_WORDS_THRESHOLD_FREQ (MAX_FREQ * 58 / 100)
250
251 #define MAX_DEPTH_MULTIPLIER 3
252
253 #define FIRST_WORD_INDEX 0
254
255 // TODO: Reduce this constant if possible; check the maximum number of digraphs in the same
256 // word in the dictionary for languages with digraphs, like German and French
257 #define DEFAULT_MAX_DIGRAPH_SEARCH_DEPTH 5
258
259 // Minimum suggest depth for one word for all cases except for missing space suggestions.
260 #define MIN_SUGGEST_DEPTH 1
261 #define MIN_USER_TYPED_LENGTH_FOR_MULTIPLE_WORD_SUGGESTION 3
262 #define MIN_USER_TYPED_LENGTH_FOR_EXCESSIVE_CHARACTER_SUGGESTION 3
263
264 // Size, in bytes, of the bloom filter index for bigrams
265 // 128 gives us 1024 buckets. The probability of false positive is (1 - e ** (-kn/m))**k,
266 // where k is the number of hash functions, n the number of bigrams, and m the number of
267 // bits we can test.
268 // At the moment 100 is the maximum number of bigrams for a word with the current
269 // dictionaries, so n = 100. 1024 buckets give us m = 1024.
270 // With 1 hash function, our false positive rate is about 9.3%, which should be enough for
271 // our uses since we are only using this to increase average performance. For the record,
272 // k = 2 gives 3.1% and k = 3 gives 1.6%. With k = 1, making m = 2048 gives 4.8%,
273 // and m = 4096 gives 2.4%.
274 #define BIGRAM_FILTER_BYTE_SIZE 128
275 // Must be smaller than BIGRAM_FILTER_BYTE_SIZE * 8, and preferably prime. 1021 is the largest
276 // prime under 128 * 8.
277 #define BIGRAM_FILTER_MODULO 1021
278 #if BIGRAM_FILTER_BYTE_SIZE * 8 < BIGRAM_FILTER_MODULO
279 #error "BIGRAM_FILTER_MODULO is larger than BIGRAM_FILTER_BYTE_SIZE"
280 #endif
281
min(T a,T b)282 template<typename T> inline T min(T a, T b) { return a < b ? a : b; }
max(T a,T b)283 template<typename T> inline T max(T a, T b) { return a > b ? a : b; }
284
285 // The ratio of neutral area radius to sweet spot radius.
286 #define NEUTRAL_AREA_RADIUS_RATIO 1.3f
287
288 // DEBUG
289 #define INPUTLENGTH_FOR_DEBUG -1
290 #define MIN_OUTPUT_INDEX_FOR_DEBUG -1
291
292 #endif // LATINIME_DEFINES_H
293