1 | /* |
2 | * Copyright © 2020 Google, Inc. |
3 | * |
4 | * This is part of HarfBuzz, a text shaping library. |
5 | * |
6 | * Permission is hereby granted, without written agreement and without |
7 | * license or royalty fees, to use, copy, modify, and distribute this |
8 | * software and its documentation for any purpose, provided that the |
9 | * above copyright notice and the following two paragraphs appear in |
10 | * all copies of this software. |
11 | * |
12 | * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR |
13 | * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES |
14 | * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN |
15 | * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH |
16 | * DAMAGE. |
17 | * |
18 | * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, |
19 | * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND |
20 | * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS |
21 | * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO |
22 | * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS. |
23 | * |
24 | * Google Author(s): Garret Rieger |
25 | */ |
26 | |
27 | #ifndef HB_REPACKER_HH |
28 | #define HB_REPACKER_HH |
29 | |
30 | #include "hb-open-type.hh" |
31 | #include "hb-map.hh" |
32 | #include "hb-vector.hh" |
33 | #include "graph/graph.hh" |
34 | #include "graph/gsubgpos-graph.hh" |
35 | #include "graph/serialize.hh" |
36 | |
37 | using graph::graph_t; |
38 | |
39 | /* |
40 | * For a detailed writeup on the overflow resolution algorithm see: |
41 | * docs/repacker.md |
42 | */ |
43 | |
44 | struct lookup_size_t |
45 | { |
46 | unsigned lookup_index; |
47 | size_t size; |
48 | unsigned num_subtables; |
49 | |
50 | static int cmp (const void* a, const void* b) |
51 | { |
52 | return cmp ((const lookup_size_t*) a, |
53 | (const lookup_size_t*) b); |
54 | } |
55 | |
56 | static int cmp (const lookup_size_t* a, const lookup_size_t* b) |
57 | { |
58 | double subtables_per_byte_a = (double) a->num_subtables / (double) a->size; |
59 | double subtables_per_byte_b = (double) b->num_subtables / (double) b->size; |
60 | if (subtables_per_byte_a == subtables_per_byte_b) { |
61 | return b->lookup_index - a->lookup_index; |
62 | } |
63 | |
64 | double cmp = subtables_per_byte_b - subtables_per_byte_a; |
65 | if (cmp < 0) return -1; |
66 | if (cmp > 0) return 1; |
67 | return 0; |
68 | } |
69 | }; |
70 | |
71 | static inline |
72 | bool _presplit_subtables_if_needed (graph::gsubgpos_graph_context_t& ext_context) |
73 | { |
74 | // For each lookup this will check the size of subtables and split them as needed |
75 | // so that no subtable is at risk of overflowing. (where we support splitting for |
76 | // that subtable type). |
77 | // |
78 | // TODO(grieger): de-dup newly added nodes as necessary. Probably just want a full de-dup |
79 | // pass after this processing is done. Not super necessary as splits are |
80 | // only done where overflow is likely, so de-dup probably will get undone |
81 | // later anyways. |
82 | for (unsigned lookup_index : ext_context.lookups.keys ()) |
83 | { |
84 | graph::Lookup* lookup = ext_context.lookups.get(lookup_index); |
85 | if (!lookup->split_subtables_if_needed (ext_context, lookup_index)) |
86 | return false; |
87 | } |
88 | |
89 | return true; |
90 | } |
91 | |
92 | /* |
93 | * Analyze the lookups in a GSUB/GPOS table and decide if any should be promoted |
94 | * to extension lookups. |
95 | */ |
96 | static inline |
97 | bool _promote_extensions_if_needed (graph::gsubgpos_graph_context_t& ext_context) |
98 | { |
99 | // Simple Algorithm (v1, current): |
100 | // 1. Calculate how many bytes each non-extension lookup consumes. |
101 | // 2. Select up to 64k of those to remain as non-extension (greedy, highest subtables per byte first) |
102 | // 3. Promote the rest. |
103 | // |
104 | // Advanced Algorithm (v2, not implemented): |
105 | // 1. Perform connected component analysis using lookups as roots. |
106 | // 2. Compute size of each connected component. |
107 | // 3. Select up to 64k worth of connected components to remain as non-extensions. |
108 | // (greedy, highest subtables per byte first) |
109 | // 4. Promote the rest. |
110 | |
111 | // TODO(garretrieger): support extension demotion, then consider all lookups. Requires advanced algo. |
112 | // TODO(garretrieger): also support extension promotion during iterative resolution phase, then |
113 | // we can use a less conservative threshold here. |
114 | // TODO(grieger): skip this for the 24 bit case. |
115 | if (!ext_context.lookups) return true; |
116 | |
117 | hb_vector_t<lookup_size_t> lookup_sizes; |
118 | lookup_sizes.alloc (ext_context.lookups.get_population (), true); |
119 | |
120 | for (unsigned lookup_index : ext_context.lookups.keys ()) |
121 | { |
122 | const graph::Lookup* lookup = ext_context.lookups.get(lookup_index); |
123 | hb_set_t visited; |
124 | lookup_sizes.push (lookup_size_t { |
125 | lookup_index, |
126 | ext_context.graph.find_subgraph_size (lookup_index, visited), |
127 | lookup->number_of_subtables (), |
128 | }); |
129 | } |
130 | |
131 | lookup_sizes.qsort (); |
132 | |
133 | size_t lookup_list_size = ext_context.graph.vertices_[ext_context.lookup_list_index].table_size (); |
134 | size_t l2_l3_size = lookup_list_size; // Lookup List + Lookups |
135 | size_t l3_l4_size = 0; // Lookups + SubTables |
136 | size_t l4_plus_size = 0; // SubTables + their descendants |
137 | |
138 | // Start by assuming all lookups are using extension subtables, this size will be removed later |
139 | // if it's decided to not make a lookup extension. |
140 | for (auto p : lookup_sizes) |
141 | { |
142 | unsigned subtables_size = p.num_subtables * 8; |
143 | l3_l4_size += subtables_size; |
144 | l4_plus_size += subtables_size; |
145 | } |
146 | |
147 | bool layers_full = false; |
148 | for (auto p : lookup_sizes) |
149 | { |
150 | const graph::Lookup* lookup = ext_context.lookups.get(p.lookup_index); |
151 | if (lookup->is_extension (ext_context.table_tag)) |
152 | // already an extension so size is counted by the loop above. |
153 | continue; |
154 | |
155 | if (!layers_full) |
156 | { |
157 | size_t lookup_size = ext_context.graph.vertices_[p.lookup_index].table_size (); |
158 | hb_set_t visited; |
159 | size_t subtables_size = ext_context.graph.find_subgraph_size (p.lookup_index, visited, 1) - lookup_size; |
160 | size_t remaining_size = p.size - subtables_size - lookup_size; |
161 | |
162 | l2_l3_size += lookup_size; |
163 | l3_l4_size += lookup_size + subtables_size; |
164 | l3_l4_size -= p.num_subtables * 8; |
165 | l4_plus_size += subtables_size + remaining_size; |
166 | |
167 | if (l2_l3_size < (1 << 16) |
168 | && l3_l4_size < (1 << 16) |
169 | && l4_plus_size < (1 << 16)) continue; // this lookup fits within all layers groups |
170 | |
171 | layers_full = true; |
172 | } |
173 | |
174 | if (!ext_context.lookups.get(p.lookup_index)->make_extension (ext_context, p.lookup_index)) |
175 | return false; |
176 | } |
177 | |
178 | return true; |
179 | } |
180 | |
181 | static inline |
182 | bool _try_isolating_subgraphs (const hb_vector_t<graph::overflow_record_t>& overflows, |
183 | graph_t& sorted_graph) |
184 | { |
185 | unsigned space = 0; |
186 | hb_set_t roots_to_isolate; |
187 | |
188 | for (int i = overflows.length - 1; i >= 0; i--) |
189 | { |
190 | const graph::overflow_record_t& r = overflows[i]; |
191 | |
192 | unsigned root; |
193 | unsigned overflow_space = sorted_graph.space_for (r.parent, &root); |
194 | if (!overflow_space) continue; |
195 | if (sorted_graph.num_roots_for_space (overflow_space) <= 1) continue; |
196 | |
197 | if (!space) { |
198 | space = overflow_space; |
199 | } |
200 | |
201 | if (space == overflow_space) |
202 | roots_to_isolate.add(root); |
203 | } |
204 | |
205 | if (!roots_to_isolate) return false; |
206 | |
207 | unsigned maximum_to_move = hb_max ((sorted_graph.num_roots_for_space (space) / 2u), 1u); |
208 | if (roots_to_isolate.get_population () > maximum_to_move) { |
209 | // Only move at most half of the roots in a space at a time. |
210 | unsigned = roots_to_isolate.get_population () - maximum_to_move; |
211 | while (extra--) { |
212 | uint32_t root = HB_SET_VALUE_INVALID; |
213 | roots_to_isolate.previous (&root); |
214 | roots_to_isolate.del (root); |
215 | } |
216 | } |
217 | |
218 | DEBUG_MSG (SUBSET_REPACK, nullptr, |
219 | "Overflow in space %u (%u roots). Moving %u roots to space %u." , |
220 | space, |
221 | sorted_graph.num_roots_for_space (space), |
222 | roots_to_isolate.get_population (), |
223 | sorted_graph.next_space ()); |
224 | |
225 | sorted_graph.isolate_subgraph (roots_to_isolate); |
226 | sorted_graph.move_to_new_space (roots_to_isolate); |
227 | |
228 | return true; |
229 | } |
230 | |
231 | static inline |
232 | bool _process_overflows (const hb_vector_t<graph::overflow_record_t>& overflows, |
233 | hb_set_t& priority_bumped_parents, |
234 | graph_t& sorted_graph) |
235 | { |
236 | bool resolution_attempted = false; |
237 | |
238 | // Try resolving the furthest overflows first. |
239 | for (int i = overflows.length - 1; i >= 0; i--) |
240 | { |
241 | const graph::overflow_record_t& r = overflows[i]; |
242 | const auto& child = sorted_graph.vertices_[r.child]; |
243 | if (child.is_shared ()) |
244 | { |
245 | // The child object is shared, we may be able to eliminate the overflow |
246 | // by duplicating it. |
247 | if (sorted_graph.duplicate (r.parent, r.child) == (unsigned) -1) continue; |
248 | return true; |
249 | } |
250 | |
251 | if (child.is_leaf () && !priority_bumped_parents.has (r.parent)) |
252 | { |
253 | // This object is too far from it's parent, attempt to move it closer. |
254 | // |
255 | // TODO(garretrieger): initially limiting this to leaf's since they can be |
256 | // moved closer with fewer consequences. However, this can |
257 | // likely can be used for non-leafs as well. |
258 | // TODO(garretrieger): also try lowering priority of the parent. Make it |
259 | // get placed further up in the ordering, closer to it's children. |
260 | // this is probably preferable if the total size of the parent object |
261 | // is < then the total size of the children (and the parent can be moved). |
262 | // Since in that case moving the parent will cause a smaller increase in |
263 | // the length of other offsets. |
264 | if (sorted_graph.raise_childrens_priority (r.parent)) { |
265 | priority_bumped_parents.add (r.parent); |
266 | resolution_attempted = true; |
267 | } |
268 | continue; |
269 | } |
270 | |
271 | // TODO(garretrieger): add additional offset resolution strategies |
272 | // - Promotion to extension lookups. |
273 | // - Table splitting. |
274 | } |
275 | |
276 | return resolution_attempted; |
277 | } |
278 | |
279 | inline bool |
280 | hb_resolve_graph_overflows (hb_tag_t table_tag, |
281 | unsigned max_rounds , |
282 | bool recalculate_extensions, |
283 | graph_t& sorted_graph /* IN/OUT */) |
284 | { |
285 | sorted_graph.sort_shortest_distance (); |
286 | if (sorted_graph.in_error ()) |
287 | { |
288 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Sorted graph in error state after initial sort." ); |
289 | return false; |
290 | } |
291 | |
292 | bool will_overflow = graph::will_overflow (sorted_graph); |
293 | if (!will_overflow) |
294 | return true; |
295 | |
296 | graph::gsubgpos_graph_context_t ext_context (table_tag, sorted_graph); |
297 | if ((table_tag == HB_OT_TAG_GPOS |
298 | || table_tag == HB_OT_TAG_GSUB) |
299 | && will_overflow) |
300 | { |
301 | if (recalculate_extensions) |
302 | { |
303 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Splitting subtables if needed." ); |
304 | if (!_presplit_subtables_if_needed (ext_context)) { |
305 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Subtable splitting failed." ); |
306 | return false; |
307 | } |
308 | |
309 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Promoting lookups to extensions if needed." ); |
310 | if (!_promote_extensions_if_needed (ext_context)) { |
311 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Extensions promotion failed." ); |
312 | return false; |
313 | } |
314 | } |
315 | |
316 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Assigning spaces to 32 bit subgraphs." ); |
317 | if (sorted_graph.assign_spaces ()) |
318 | sorted_graph.sort_shortest_distance (); |
319 | else |
320 | sorted_graph.sort_shortest_distance_if_needed (); |
321 | } |
322 | |
323 | unsigned round = 0; |
324 | hb_vector_t<graph::overflow_record_t> overflows; |
325 | // TODO(garretrieger): select a good limit for max rounds. |
326 | while (!sorted_graph.in_error () |
327 | && graph::will_overflow (sorted_graph, &overflows) |
328 | && round < max_rounds) { |
329 | DEBUG_MSG (SUBSET_REPACK, nullptr, "=== Overflow resolution round %u ===" , round); |
330 | print_overflows (sorted_graph, overflows); |
331 | |
332 | hb_set_t priority_bumped_parents; |
333 | |
334 | if (!_try_isolating_subgraphs (overflows, sorted_graph)) |
335 | { |
336 | // Don't count space isolation towards round limit. Only increment |
337 | // round counter if space isolation made no changes. |
338 | round++; |
339 | if (!_process_overflows (overflows, priority_bumped_parents, sorted_graph)) |
340 | { |
341 | DEBUG_MSG (SUBSET_REPACK, nullptr, "No resolution available :(" ); |
342 | break; |
343 | } |
344 | } |
345 | |
346 | sorted_graph.sort_shortest_distance (); |
347 | } |
348 | |
349 | if (sorted_graph.in_error ()) |
350 | { |
351 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Sorted graph in error state." ); |
352 | return false; |
353 | } |
354 | |
355 | if (graph::will_overflow (sorted_graph)) |
356 | { |
357 | DEBUG_MSG (SUBSET_REPACK, nullptr, "Offset overflow resolution failed." ); |
358 | return false; |
359 | } |
360 | |
361 | return true; |
362 | } |
363 | |
364 | /* |
365 | * Attempts to modify the topological sorting of the provided object graph to |
366 | * eliminate offset overflows in the links between objects of the graph. If a |
367 | * non-overflowing ordering is found the updated graph is serialized it into the |
368 | * provided serialization context. |
369 | * |
370 | * If necessary the structure of the graph may be modified in ways that do not |
371 | * affect the functionality of the graph. For example shared objects may be |
372 | * duplicated. |
373 | * |
374 | * For a detailed writeup describing how the algorithm operates see: |
375 | * docs/repacker.md |
376 | */ |
377 | template<typename T> |
378 | inline hb_blob_t* |
379 | hb_resolve_overflows (const T& packed, |
380 | hb_tag_t table_tag, |
381 | unsigned max_rounds = 20, |
382 | bool recalculate_extensions = false) { |
383 | graph_t sorted_graph (packed); |
384 | if (sorted_graph.in_error ()) |
385 | { |
386 | // Invalid graph definition. |
387 | return nullptr; |
388 | } |
389 | |
390 | if (!sorted_graph.is_fully_connected ()) |
391 | { |
392 | sorted_graph.print_orphaned_nodes (); |
393 | return nullptr; |
394 | } |
395 | |
396 | if (sorted_graph.in_error ()) |
397 | { |
398 | // Allocations failed somewhere |
399 | DEBUG_MSG (SUBSET_REPACK, nullptr, |
400 | "Graph is in error, likely due to a memory allocation error." ); |
401 | return nullptr; |
402 | } |
403 | |
404 | if (!hb_resolve_graph_overflows (table_tag, max_rounds, recalculate_extensions, sorted_graph)) |
405 | return nullptr; |
406 | |
407 | return graph::serialize (sorted_graph); |
408 | } |
409 | |
410 | #endif /* HB_REPACKER_HH */ |
411 | |