1 | // Copyright 2012 Google Inc. All Rights Reserved. |
2 | // |
3 | // Use of this source code is governed by a BSD-style license |
4 | // that can be found in the COPYING file in the root of the source |
5 | // tree. An additional intellectual property rights grant can be found |
6 | // in the file PATENTS. All contributing project authors may |
7 | // be found in the AUTHORS file in the root of the source tree. |
8 | // ----------------------------------------------------------------------------- |
9 | // |
10 | // main entry for the decoder |
11 | // |
12 | // Authors: Vikas Arora (vikaas.arora@gmail.com) |
13 | // Jyrki Alakuijala (jyrki@google.com) |
14 | |
15 | #include <stdlib.h> |
16 | |
17 | #include "src/dec/alphai_dec.h" |
18 | #include "src/dec/vp8li_dec.h" |
19 | #include "src/dsp/dsp.h" |
20 | #include "src/dsp/lossless.h" |
21 | #include "src/dsp/lossless_common.h" |
22 | #include "src/dsp/yuv.h" |
23 | #include "src/utils/endian_inl_utils.h" |
24 | #include "src/utils/huffman_utils.h" |
25 | #include "src/utils/utils.h" |
26 | |
27 | #define NUM_ARGB_CACHE_ROWS 16 |
28 | |
29 | static const int kCodeLengthLiterals = 16; |
30 | static const int kCodeLengthRepeatCode = 16; |
31 | static const uint8_t [3] = { 2, 3, 7 }; |
32 | static const uint8_t kCodeLengthRepeatOffsets[3] = { 3, 3, 11 }; |
33 | |
34 | // ----------------------------------------------------------------------------- |
35 | // Five Huffman codes are used at each meta code: |
36 | // 1. green + length prefix codes + color cache codes, |
37 | // 2. alpha, |
38 | // 3. red, |
39 | // 4. blue, and, |
40 | // 5. distance prefix codes. |
41 | typedef enum { |
42 | GREEN = 0, |
43 | RED = 1, |
44 | BLUE = 2, |
45 | ALPHA = 3, |
46 | DIST = 4 |
47 | } HuffIndex; |
48 | |
49 | static const uint16_t kAlphabetSize[HUFFMAN_CODES_PER_META_CODE] = { |
50 | NUM_LITERAL_CODES + NUM_LENGTH_CODES, |
51 | NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES, |
52 | NUM_DISTANCE_CODES |
53 | }; |
54 | |
55 | static const uint8_t kLiteralMap[HUFFMAN_CODES_PER_META_CODE] = { |
56 | 0, 1, 1, 1, 0 |
57 | }; |
58 | |
59 | #define NUM_CODE_LENGTH_CODES 19 |
60 | static const uint8_t kCodeLengthCodeOrder[NUM_CODE_LENGTH_CODES] = { |
61 | 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 |
62 | }; |
63 | |
64 | #define CODE_TO_PLANE_CODES 120 |
65 | static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = { |
66 | 0x18, 0x07, 0x17, 0x19, 0x28, 0x06, 0x27, 0x29, 0x16, 0x1a, |
67 | 0x26, 0x2a, 0x38, 0x05, 0x37, 0x39, 0x15, 0x1b, 0x36, 0x3a, |
68 | 0x25, 0x2b, 0x48, 0x04, 0x47, 0x49, 0x14, 0x1c, 0x35, 0x3b, |
69 | 0x46, 0x4a, 0x24, 0x2c, 0x58, 0x45, 0x4b, 0x34, 0x3c, 0x03, |
70 | 0x57, 0x59, 0x13, 0x1d, 0x56, 0x5a, 0x23, 0x2d, 0x44, 0x4c, |
71 | 0x55, 0x5b, 0x33, 0x3d, 0x68, 0x02, 0x67, 0x69, 0x12, 0x1e, |
72 | 0x66, 0x6a, 0x22, 0x2e, 0x54, 0x5c, 0x43, 0x4d, 0x65, 0x6b, |
73 | 0x32, 0x3e, 0x78, 0x01, 0x77, 0x79, 0x53, 0x5d, 0x11, 0x1f, |
74 | 0x64, 0x6c, 0x42, 0x4e, 0x76, 0x7a, 0x21, 0x2f, 0x75, 0x7b, |
75 | 0x31, 0x3f, 0x63, 0x6d, 0x52, 0x5e, 0x00, 0x74, 0x7c, 0x41, |
76 | 0x4f, 0x10, 0x20, 0x62, 0x6e, 0x30, 0x73, 0x7d, 0x51, 0x5f, |
77 | 0x40, 0x72, 0x7e, 0x61, 0x6f, 0x50, 0x71, 0x7f, 0x60, 0x70 |
78 | }; |
79 | |
80 | // Memory needed for lookup tables of one Huffman tree group. Red, blue, alpha |
81 | // and distance alphabets are constant (256 for red, blue and alpha, 40 for |
82 | // distance) and lookup table sizes for them in worst case are 630 and 410 |
83 | // respectively. Size of green alphabet depends on color cache size and is equal |
84 | // to 256 (green component values) + 24 (length prefix values) |
85 | // + color_cache_size (between 0 and 2048). |
86 | // All values computed for 8-bit first level lookup with Mark Adler's tool: |
87 | // https://github.com/madler/zlib/blob/v1.2.5/examples/enough.c |
88 | #define FIXED_TABLE_SIZE (630 * 3 + 410) |
89 | static const uint16_t kTableSize[12] = { |
90 | FIXED_TABLE_SIZE + 654, |
91 | FIXED_TABLE_SIZE + 656, |
92 | FIXED_TABLE_SIZE + 658, |
93 | FIXED_TABLE_SIZE + 662, |
94 | FIXED_TABLE_SIZE + 670, |
95 | FIXED_TABLE_SIZE + 686, |
96 | FIXED_TABLE_SIZE + 718, |
97 | FIXED_TABLE_SIZE + 782, |
98 | FIXED_TABLE_SIZE + 912, |
99 | FIXED_TABLE_SIZE + 1168, |
100 | FIXED_TABLE_SIZE + 1680, |
101 | FIXED_TABLE_SIZE + 2704 |
102 | }; |
103 | |
104 | static int DecodeImageStream(int xsize, int ysize, |
105 | int is_level0, |
106 | VP8LDecoder* const dec, |
107 | uint32_t** const decoded_data); |
108 | |
109 | //------------------------------------------------------------------------------ |
110 | |
111 | int VP8LCheckSignature(const uint8_t* const data, size_t size) { |
112 | return (size >= VP8L_FRAME_HEADER_SIZE && |
113 | data[0] == VP8L_MAGIC_BYTE && |
114 | (data[4] >> 5) == 0); // version |
115 | } |
116 | |
117 | static int ReadImageInfo(VP8LBitReader* const br, |
118 | int* const width, int* const height, |
119 | int* const has_alpha) { |
120 | if (VP8LReadBits(br, 8) != VP8L_MAGIC_BYTE) return 0; |
121 | *width = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; |
122 | *height = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; |
123 | *has_alpha = VP8LReadBits(br, 1); |
124 | if (VP8LReadBits(br, VP8L_VERSION_BITS) != 0) return 0; |
125 | return !br->eos_; |
126 | } |
127 | |
128 | int VP8LGetInfo(const uint8_t* data, size_t data_size, |
129 | int* const width, int* const height, int* const has_alpha) { |
130 | if (data == NULL || data_size < VP8L_FRAME_HEADER_SIZE) { |
131 | return 0; // not enough data |
132 | } else if (!VP8LCheckSignature(data, data_size)) { |
133 | return 0; // bad signature |
134 | } else { |
135 | int w, h, a; |
136 | VP8LBitReader br; |
137 | VP8LInitBitReader(&br, data, data_size); |
138 | if (!ReadImageInfo(&br, &w, &h, &a)) { |
139 | return 0; |
140 | } |
141 | if (width != NULL) *width = w; |
142 | if (height != NULL) *height = h; |
143 | if (has_alpha != NULL) *has_alpha = a; |
144 | return 1; |
145 | } |
146 | } |
147 | |
148 | //------------------------------------------------------------------------------ |
149 | |
150 | static WEBP_INLINE int GetCopyDistance(int distance_symbol, |
151 | VP8LBitReader* const br) { |
152 | int , offset; |
153 | if (distance_symbol < 4) { |
154 | return distance_symbol + 1; |
155 | } |
156 | extra_bits = (distance_symbol - 2) >> 1; |
157 | offset = (2 + (distance_symbol & 1)) << extra_bits; |
158 | return offset + VP8LReadBits(br, extra_bits) + 1; |
159 | } |
160 | |
161 | static WEBP_INLINE int GetCopyLength(int length_symbol, |
162 | VP8LBitReader* const br) { |
163 | // Length and distance prefixes are encoded the same way. |
164 | return GetCopyDistance(length_symbol, br); |
165 | } |
166 | |
167 | static WEBP_INLINE int PlaneCodeToDistance(int xsize, int plane_code) { |
168 | if (plane_code > CODE_TO_PLANE_CODES) { |
169 | return plane_code - CODE_TO_PLANE_CODES; |
170 | } else { |
171 | const int dist_code = kCodeToPlane[plane_code - 1]; |
172 | const int yoffset = dist_code >> 4; |
173 | const int xoffset = 8 - (dist_code & 0xf); |
174 | const int dist = yoffset * xsize + xoffset; |
175 | return (dist >= 1) ? dist : 1; // dist<1 can happen if xsize is very small |
176 | } |
177 | } |
178 | |
179 | //------------------------------------------------------------------------------ |
180 | // Decodes the next Huffman code from bit-stream. |
181 | // VP8LFillBitWindow(br) needs to be called at minimum every second call |
182 | // to ReadSymbol, in order to pre-fetch enough bits. |
183 | static WEBP_INLINE int ReadSymbol(const HuffmanCode* table, |
184 | VP8LBitReader* const br) { |
185 | int nbits; |
186 | uint32_t val = VP8LPrefetchBits(br); |
187 | table += val & HUFFMAN_TABLE_MASK; |
188 | nbits = table->bits - HUFFMAN_TABLE_BITS; |
189 | if (nbits > 0) { |
190 | VP8LSetBitPos(br, br->bit_pos_ + HUFFMAN_TABLE_BITS); |
191 | val = VP8LPrefetchBits(br); |
192 | table += table->value; |
193 | table += val & ((1 << nbits) - 1); |
194 | } |
195 | VP8LSetBitPos(br, br->bit_pos_ + table->bits); |
196 | return table->value; |
197 | } |
198 | |
199 | // Reads packed symbol depending on GREEN channel |
200 | #define BITS_SPECIAL_MARKER 0x100 // something large enough (and a bit-mask) |
201 | #define PACKED_NON_LITERAL_CODE 0 // must be < NUM_LITERAL_CODES |
202 | static WEBP_INLINE int ReadPackedSymbols(const HTreeGroup* group, |
203 | VP8LBitReader* const br, |
204 | uint32_t* const dst) { |
205 | const uint32_t val = VP8LPrefetchBits(br) & (HUFFMAN_PACKED_TABLE_SIZE - 1); |
206 | const HuffmanCode32 code = group->packed_table[val]; |
207 | assert(group->use_packed_table); |
208 | if (code.bits < BITS_SPECIAL_MARKER) { |
209 | VP8LSetBitPos(br, br->bit_pos_ + code.bits); |
210 | *dst = code.value; |
211 | return PACKED_NON_LITERAL_CODE; |
212 | } else { |
213 | VP8LSetBitPos(br, br->bit_pos_ + code.bits - BITS_SPECIAL_MARKER); |
214 | assert(code.value >= NUM_LITERAL_CODES); |
215 | return code.value; |
216 | } |
217 | } |
218 | |
219 | static int AccumulateHCode(HuffmanCode hcode, int shift, |
220 | HuffmanCode32* const huff) { |
221 | huff->bits += hcode.bits; |
222 | huff->value |= (uint32_t)hcode.value << shift; |
223 | assert(huff->bits <= HUFFMAN_TABLE_BITS); |
224 | return hcode.bits; |
225 | } |
226 | |
227 | static void BuildPackedTable(HTreeGroup* const htree_group) { |
228 | uint32_t code; |
229 | for (code = 0; code < HUFFMAN_PACKED_TABLE_SIZE; ++code) { |
230 | uint32_t bits = code; |
231 | HuffmanCode32* const huff = &htree_group->packed_table[bits]; |
232 | HuffmanCode hcode = htree_group->htrees[GREEN][bits]; |
233 | if (hcode.value >= NUM_LITERAL_CODES) { |
234 | huff->bits = hcode.bits + BITS_SPECIAL_MARKER; |
235 | huff->value = hcode.value; |
236 | } else { |
237 | huff->bits = 0; |
238 | huff->value = 0; |
239 | bits >>= AccumulateHCode(hcode, 8, huff); |
240 | bits >>= AccumulateHCode(htree_group->htrees[RED][bits], 16, huff); |
241 | bits >>= AccumulateHCode(htree_group->htrees[BLUE][bits], 0, huff); |
242 | bits >>= AccumulateHCode(htree_group->htrees[ALPHA][bits], 24, huff); |
243 | (void)bits; |
244 | } |
245 | } |
246 | } |
247 | |
248 | static int ReadHuffmanCodeLengths( |
249 | VP8LDecoder* const dec, const int* const code_length_code_lengths, |
250 | int num_symbols, int* const code_lengths) { |
251 | int ok = 0; |
252 | VP8LBitReader* const br = &dec->br_; |
253 | int symbol; |
254 | int max_symbol; |
255 | int prev_code_len = DEFAULT_CODE_LENGTH; |
256 | HuffmanTables tables; |
257 | |
258 | if (!VP8LHuffmanTablesAllocate(1 << LENGTHS_TABLE_BITS, &tables) || |
259 | !VP8LBuildHuffmanTable(&tables, LENGTHS_TABLE_BITS, |
260 | code_length_code_lengths, NUM_CODE_LENGTH_CODES)) { |
261 | goto End; |
262 | } |
263 | |
264 | if (VP8LReadBits(br, 1)) { // use length |
265 | const int length_nbits = 2 + 2 * VP8LReadBits(br, 3); |
266 | max_symbol = 2 + VP8LReadBits(br, length_nbits); |
267 | if (max_symbol > num_symbols) { |
268 | goto End; |
269 | } |
270 | } else { |
271 | max_symbol = num_symbols; |
272 | } |
273 | |
274 | symbol = 0; |
275 | while (symbol < num_symbols) { |
276 | const HuffmanCode* p; |
277 | int code_len; |
278 | if (max_symbol-- == 0) break; |
279 | VP8LFillBitWindow(br); |
280 | p = &tables.curr_segment->start[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK]; |
281 | VP8LSetBitPos(br, br->bit_pos_ + p->bits); |
282 | code_len = p->value; |
283 | if (code_len < kCodeLengthLiterals) { |
284 | code_lengths[symbol++] = code_len; |
285 | if (code_len != 0) prev_code_len = code_len; |
286 | } else { |
287 | const int use_prev = (code_len == kCodeLengthRepeatCode); |
288 | const int slot = code_len - kCodeLengthLiterals; |
289 | const int = kCodeLengthExtraBits[slot]; |
290 | const int repeat_offset = kCodeLengthRepeatOffsets[slot]; |
291 | int repeat = VP8LReadBits(br, extra_bits) + repeat_offset; |
292 | if (symbol + repeat > num_symbols) { |
293 | goto End; |
294 | } else { |
295 | const int length = use_prev ? prev_code_len : 0; |
296 | while (repeat-- > 0) code_lengths[symbol++] = length; |
297 | } |
298 | } |
299 | } |
300 | ok = 1; |
301 | |
302 | End: |
303 | VP8LHuffmanTablesDeallocate(&tables); |
304 | if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
305 | return ok; |
306 | } |
307 | |
308 | // 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman |
309 | // tree. |
310 | static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec, |
311 | int* const code_lengths, |
312 | HuffmanTables* const table) { |
313 | int ok = 0; |
314 | int size = 0; |
315 | VP8LBitReader* const br = &dec->br_; |
316 | const int simple_code = VP8LReadBits(br, 1); |
317 | |
318 | memset(code_lengths, 0, alphabet_size * sizeof(*code_lengths)); |
319 | |
320 | if (simple_code) { // Read symbols, codes & code lengths directly. |
321 | const int num_symbols = VP8LReadBits(br, 1) + 1; |
322 | const int first_symbol_len_code = VP8LReadBits(br, 1); |
323 | // The first code is either 1 bit or 8 bit code. |
324 | int symbol = VP8LReadBits(br, (first_symbol_len_code == 0) ? 1 : 8); |
325 | code_lengths[symbol] = 1; |
326 | // The second code (if present), is always 8 bits long. |
327 | if (num_symbols == 2) { |
328 | symbol = VP8LReadBits(br, 8); |
329 | code_lengths[symbol] = 1; |
330 | } |
331 | ok = 1; |
332 | } else { // Decode Huffman-coded code lengths. |
333 | int i; |
334 | int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 }; |
335 | const int num_codes = VP8LReadBits(br, 4) + 4; |
336 | if (num_codes > NUM_CODE_LENGTH_CODES) { |
337 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
338 | return 0; |
339 | } |
340 | |
341 | for (i = 0; i < num_codes; ++i) { |
342 | code_length_code_lengths[kCodeLengthCodeOrder[i]] = VP8LReadBits(br, 3); |
343 | } |
344 | ok = ReadHuffmanCodeLengths(dec, code_length_code_lengths, alphabet_size, |
345 | code_lengths); |
346 | } |
347 | |
348 | ok = ok && !br->eos_; |
349 | if (ok) { |
350 | size = VP8LBuildHuffmanTable(table, HUFFMAN_TABLE_BITS, |
351 | code_lengths, alphabet_size); |
352 | } |
353 | if (!ok || size == 0) { |
354 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
355 | return 0; |
356 | } |
357 | return size; |
358 | } |
359 | |
360 | static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize, |
361 | int color_cache_bits, int allow_recursion) { |
362 | int i, j; |
363 | VP8LBitReader* const br = &dec->br_; |
364 | VP8LMetadata* const hdr = &dec->hdr_; |
365 | uint32_t* huffman_image = NULL; |
366 | HTreeGroup* htree_groups = NULL; |
367 | HuffmanTables* huffman_tables = &hdr->huffman_tables_; |
368 | int num_htree_groups = 1; |
369 | int num_htree_groups_max = 1; |
370 | int max_alphabet_size = 0; |
371 | int* code_lengths = NULL; |
372 | const int table_size = kTableSize[color_cache_bits]; |
373 | int* mapping = NULL; |
374 | int ok = 0; |
375 | |
376 | // Check the table has been 0 initialized (through InitMetadata). |
377 | assert(huffman_tables->root.start == NULL); |
378 | assert(huffman_tables->curr_segment == NULL); |
379 | |
380 | if (allow_recursion && VP8LReadBits(br, 1)) { |
381 | // use meta Huffman codes. |
382 | const int huffman_precision = VP8LReadBits(br, 3) + 2; |
383 | const int huffman_xsize = VP8LSubSampleSize(xsize, huffman_precision); |
384 | const int huffman_ysize = VP8LSubSampleSize(ysize, huffman_precision); |
385 | const int huffman_pixs = huffman_xsize * huffman_ysize; |
386 | if (!DecodeImageStream(huffman_xsize, huffman_ysize, 0, dec, |
387 | &huffman_image)) { |
388 | goto Error; |
389 | } |
390 | hdr->huffman_subsample_bits_ = huffman_precision; |
391 | for (i = 0; i < huffman_pixs; ++i) { |
392 | // The huffman data is stored in red and green bytes. |
393 | const int group = (huffman_image[i] >> 8) & 0xffff; |
394 | huffman_image[i] = group; |
395 | if (group >= num_htree_groups_max) { |
396 | num_htree_groups_max = group + 1; |
397 | } |
398 | } |
399 | // Check the validity of num_htree_groups_max. If it seems too big, use a |
400 | // smaller value for later. This will prevent big memory allocations to end |
401 | // up with a bad bitstream anyway. |
402 | // The value of 1000 is totally arbitrary. We know that num_htree_groups_max |
403 | // is smaller than (1 << 16) and should be smaller than the number of pixels |
404 | // (though the format allows it to be bigger). |
405 | if (num_htree_groups_max > 1000 || num_htree_groups_max > xsize * ysize) { |
406 | // Create a mapping from the used indices to the minimal set of used |
407 | // values [0, num_htree_groups) |
408 | mapping = (int*)WebPSafeMalloc(num_htree_groups_max, sizeof(*mapping)); |
409 | if (mapping == NULL) { |
410 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
411 | goto Error; |
412 | } |
413 | // -1 means a value is unmapped, and therefore unused in the Huffman |
414 | // image. |
415 | memset(mapping, 0xff, num_htree_groups_max * sizeof(*mapping)); |
416 | for (num_htree_groups = 0, i = 0; i < huffman_pixs; ++i) { |
417 | // Get the current mapping for the group and remap the Huffman image. |
418 | int* const mapped_group = &mapping[huffman_image[i]]; |
419 | if (*mapped_group == -1) *mapped_group = num_htree_groups++; |
420 | huffman_image[i] = *mapped_group; |
421 | } |
422 | } else { |
423 | num_htree_groups = num_htree_groups_max; |
424 | } |
425 | } |
426 | |
427 | if (br->eos_) goto Error; |
428 | |
429 | // Find maximum alphabet size for the htree group. |
430 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { |
431 | int alphabet_size = kAlphabetSize[j]; |
432 | if (j == 0 && color_cache_bits > 0) { |
433 | alphabet_size += 1 << color_cache_bits; |
434 | } |
435 | if (max_alphabet_size < alphabet_size) { |
436 | max_alphabet_size = alphabet_size; |
437 | } |
438 | } |
439 | |
440 | code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size, |
441 | sizeof(*code_lengths)); |
442 | htree_groups = VP8LHtreeGroupsNew(num_htree_groups); |
443 | |
444 | if (htree_groups == NULL || code_lengths == NULL || |
445 | !VP8LHuffmanTablesAllocate(num_htree_groups * table_size, |
446 | huffman_tables)) { |
447 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
448 | goto Error; |
449 | } |
450 | |
451 | for (i = 0; i < num_htree_groups_max; ++i) { |
452 | // If the index "i" is unused in the Huffman image, just make sure the |
453 | // coefficients are valid but do not store them. |
454 | if (mapping != NULL && mapping[i] == -1) { |
455 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { |
456 | int alphabet_size = kAlphabetSize[j]; |
457 | if (j == 0 && color_cache_bits > 0) { |
458 | alphabet_size += (1 << color_cache_bits); |
459 | } |
460 | // Passing in NULL so that nothing gets filled. |
461 | if (!ReadHuffmanCode(alphabet_size, dec, code_lengths, NULL)) { |
462 | goto Error; |
463 | } |
464 | } |
465 | } else { |
466 | HTreeGroup* const htree_group = |
467 | &htree_groups[(mapping == NULL) ? i : mapping[i]]; |
468 | HuffmanCode** const htrees = htree_group->htrees; |
469 | int size; |
470 | int total_size = 0; |
471 | int is_trivial_literal = 1; |
472 | int max_bits = 0; |
473 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { |
474 | int alphabet_size = kAlphabetSize[j]; |
475 | if (j == 0 && color_cache_bits > 0) { |
476 | alphabet_size += (1 << color_cache_bits); |
477 | } |
478 | size = |
479 | ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables); |
480 | htrees[j] = huffman_tables->curr_segment->curr_table; |
481 | if (size == 0) { |
482 | goto Error; |
483 | } |
484 | if (is_trivial_literal && kLiteralMap[j] == 1) { |
485 | is_trivial_literal = (htrees[j]->bits == 0); |
486 | } |
487 | total_size += htrees[j]->bits; |
488 | huffman_tables->curr_segment->curr_table += size; |
489 | if (j <= ALPHA) { |
490 | int local_max_bits = code_lengths[0]; |
491 | int k; |
492 | for (k = 1; k < alphabet_size; ++k) { |
493 | if (code_lengths[k] > local_max_bits) { |
494 | local_max_bits = code_lengths[k]; |
495 | } |
496 | } |
497 | max_bits += local_max_bits; |
498 | } |
499 | } |
500 | htree_group->is_trivial_literal = is_trivial_literal; |
501 | htree_group->is_trivial_code = 0; |
502 | if (is_trivial_literal) { |
503 | const int red = htrees[RED][0].value; |
504 | const int blue = htrees[BLUE][0].value; |
505 | const int alpha = htrees[ALPHA][0].value; |
506 | htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue; |
507 | if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) { |
508 | htree_group->is_trivial_code = 1; |
509 | htree_group->literal_arb |= htrees[GREEN][0].value << 8; |
510 | } |
511 | } |
512 | htree_group->use_packed_table = |
513 | !htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS); |
514 | if (htree_group->use_packed_table) BuildPackedTable(htree_group); |
515 | } |
516 | } |
517 | ok = 1; |
518 | |
519 | // All OK. Finalize pointers. |
520 | hdr->huffman_image_ = huffman_image; |
521 | hdr->num_htree_groups_ = num_htree_groups; |
522 | hdr->htree_groups_ = htree_groups; |
523 | |
524 | Error: |
525 | WebPSafeFree(code_lengths); |
526 | WebPSafeFree(mapping); |
527 | if (!ok) { |
528 | WebPSafeFree(huffman_image); |
529 | VP8LHuffmanTablesDeallocate(huffman_tables); |
530 | VP8LHtreeGroupsFree(htree_groups); |
531 | } |
532 | return ok; |
533 | } |
534 | |
535 | //------------------------------------------------------------------------------ |
536 | // Scaling. |
537 | |
538 | #if !defined(WEBP_REDUCE_SIZE) |
539 | static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) { |
540 | const int num_channels = 4; |
541 | const int in_width = io->mb_w; |
542 | const int out_width = io->scaled_width; |
543 | const int in_height = io->mb_h; |
544 | const int out_height = io->scaled_height; |
545 | const uint64_t work_size = 2 * num_channels * (uint64_t)out_width; |
546 | rescaler_t* work; // Rescaler work area. |
547 | const uint64_t scaled_data_size = (uint64_t)out_width; |
548 | uint32_t* scaled_data; // Temporary storage for scaled BGRA data. |
549 | const uint64_t memory_size = sizeof(*dec->rescaler) + |
550 | work_size * sizeof(*work) + |
551 | scaled_data_size * sizeof(*scaled_data); |
552 | uint8_t* memory = (uint8_t*)WebPSafeMalloc(memory_size, sizeof(*memory)); |
553 | if (memory == NULL) { |
554 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
555 | return 0; |
556 | } |
557 | assert(dec->rescaler_memory == NULL); |
558 | dec->rescaler_memory = memory; |
559 | |
560 | dec->rescaler = (WebPRescaler*)memory; |
561 | memory += sizeof(*dec->rescaler); |
562 | work = (rescaler_t*)memory; |
563 | memory += work_size * sizeof(*work); |
564 | scaled_data = (uint32_t*)memory; |
565 | |
566 | if (!WebPRescalerInit(dec->rescaler, in_width, in_height, |
567 | (uint8_t*)scaled_data, out_width, out_height, |
568 | 0, num_channels, work)) { |
569 | return 0; |
570 | } |
571 | return 1; |
572 | } |
573 | #endif // WEBP_REDUCE_SIZE |
574 | |
575 | //------------------------------------------------------------------------------ |
576 | // Export to ARGB |
577 | |
578 | #if !defined(WEBP_REDUCE_SIZE) |
579 | |
580 | // We have special "export" function since we need to convert from BGRA |
581 | static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace, |
582 | int rgba_stride, uint8_t* const rgba) { |
583 | uint32_t* const src = (uint32_t*)rescaler->dst; |
584 | uint8_t* dst = rgba; |
585 | const int dst_width = rescaler->dst_width; |
586 | int num_lines_out = 0; |
587 | while (WebPRescalerHasPendingOutput(rescaler)) { |
588 | WebPRescalerExportRow(rescaler); |
589 | WebPMultARGBRow(src, dst_width, 1); |
590 | VP8LConvertFromBGRA(src, dst_width, colorspace, dst); |
591 | dst += rgba_stride; |
592 | ++num_lines_out; |
593 | } |
594 | return num_lines_out; |
595 | } |
596 | |
597 | // Emit scaled rows. |
598 | static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec, |
599 | uint8_t* in, int in_stride, int mb_h, |
600 | uint8_t* const out, int out_stride) { |
601 | const WEBP_CSP_MODE colorspace = dec->output_->colorspace; |
602 | int num_lines_in = 0; |
603 | int num_lines_out = 0; |
604 | while (num_lines_in < mb_h) { |
605 | uint8_t* const row_in = in + (uint64_t)num_lines_in * in_stride; |
606 | uint8_t* const row_out = out + (uint64_t)num_lines_out * out_stride; |
607 | const int lines_left = mb_h - num_lines_in; |
608 | const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); |
609 | int lines_imported; |
610 | assert(needed_lines > 0 && needed_lines <= lines_left); |
611 | WebPMultARGBRows(row_in, in_stride, |
612 | dec->rescaler->src_width, needed_lines, 0); |
613 | lines_imported = |
614 | WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride); |
615 | assert(lines_imported == needed_lines); |
616 | num_lines_in += lines_imported; |
617 | num_lines_out += Export(dec->rescaler, colorspace, out_stride, row_out); |
618 | } |
619 | return num_lines_out; |
620 | } |
621 | |
622 | #endif // WEBP_REDUCE_SIZE |
623 | |
624 | // Emit rows without any scaling. |
625 | static int EmitRows(WEBP_CSP_MODE colorspace, |
626 | const uint8_t* row_in, int in_stride, |
627 | int mb_w, int mb_h, |
628 | uint8_t* const out, int out_stride) { |
629 | int lines = mb_h; |
630 | uint8_t* row_out = out; |
631 | while (lines-- > 0) { |
632 | VP8LConvertFromBGRA((const uint32_t*)row_in, mb_w, colorspace, row_out); |
633 | row_in += in_stride; |
634 | row_out += out_stride; |
635 | } |
636 | return mb_h; // Num rows out == num rows in. |
637 | } |
638 | |
639 | //------------------------------------------------------------------------------ |
640 | // Export to YUVA |
641 | |
642 | static void ConvertToYUVA(const uint32_t* const src, int width, int y_pos, |
643 | const WebPDecBuffer* const output) { |
644 | const WebPYUVABuffer* const buf = &output->u.YUVA; |
645 | |
646 | // first, the luma plane |
647 | WebPConvertARGBToY(src, buf->y + y_pos * buf->y_stride, width); |
648 | |
649 | // then U/V planes |
650 | { |
651 | uint8_t* const u = buf->u + (y_pos >> 1) * buf->u_stride; |
652 | uint8_t* const v = buf->v + (y_pos >> 1) * buf->v_stride; |
653 | // even lines: store values |
654 | // odd lines: average with previous values |
655 | WebPConvertARGBToUV(src, u, v, width, !(y_pos & 1)); |
656 | } |
657 | // Lastly, store alpha if needed. |
658 | if (buf->a != NULL) { |
659 | uint8_t* const a = buf->a + y_pos * buf->a_stride; |
660 | #if defined(WORDS_BIGENDIAN) |
661 | WebPExtractAlpha((uint8_t*)src + 0, 0, width, 1, a, 0); |
662 | #else |
663 | WebPExtractAlpha((uint8_t*)src + 3, 0, width, 1, a, 0); |
664 | #endif |
665 | } |
666 | } |
667 | |
668 | static int ExportYUVA(const VP8LDecoder* const dec, int y_pos) { |
669 | WebPRescaler* const rescaler = dec->rescaler; |
670 | uint32_t* const src = (uint32_t*)rescaler->dst; |
671 | const int dst_width = rescaler->dst_width; |
672 | int num_lines_out = 0; |
673 | while (WebPRescalerHasPendingOutput(rescaler)) { |
674 | WebPRescalerExportRow(rescaler); |
675 | WebPMultARGBRow(src, dst_width, 1); |
676 | ConvertToYUVA(src, dst_width, y_pos, dec->output_); |
677 | ++y_pos; |
678 | ++num_lines_out; |
679 | } |
680 | return num_lines_out; |
681 | } |
682 | |
683 | static int EmitRescaledRowsYUVA(const VP8LDecoder* const dec, |
684 | uint8_t* in, int in_stride, int mb_h) { |
685 | int num_lines_in = 0; |
686 | int y_pos = dec->last_out_row_; |
687 | while (num_lines_in < mb_h) { |
688 | const int lines_left = mb_h - num_lines_in; |
689 | const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); |
690 | int lines_imported; |
691 | WebPMultARGBRows(in, in_stride, dec->rescaler->src_width, needed_lines, 0); |
692 | lines_imported = |
693 | WebPRescalerImport(dec->rescaler, lines_left, in, in_stride); |
694 | assert(lines_imported == needed_lines); |
695 | num_lines_in += lines_imported; |
696 | in += needed_lines * in_stride; |
697 | y_pos += ExportYUVA(dec, y_pos); |
698 | } |
699 | return y_pos; |
700 | } |
701 | |
702 | static int EmitRowsYUVA(const VP8LDecoder* const dec, |
703 | const uint8_t* in, int in_stride, |
704 | int mb_w, int num_rows) { |
705 | int y_pos = dec->last_out_row_; |
706 | while (num_rows-- > 0) { |
707 | ConvertToYUVA((const uint32_t*)in, mb_w, y_pos, dec->output_); |
708 | in += in_stride; |
709 | ++y_pos; |
710 | } |
711 | return y_pos; |
712 | } |
713 | |
714 | //------------------------------------------------------------------------------ |
715 | // Cropping. |
716 | |
717 | // Sets io->mb_y, io->mb_h & io->mb_w according to start row, end row and |
718 | // crop options. Also updates the input data pointer, so that it points to the |
719 | // start of the cropped window. Note that pixels are in ARGB format even if |
720 | // 'in_data' is uint8_t*. |
721 | // Returns true if the crop window is not empty. |
722 | static int SetCropWindow(VP8Io* const io, int y_start, int y_end, |
723 | uint8_t** const in_data, int pixel_stride) { |
724 | assert(y_start < y_end); |
725 | assert(io->crop_left < io->crop_right); |
726 | if (y_end > io->crop_bottom) { |
727 | y_end = io->crop_bottom; // make sure we don't overflow on last row. |
728 | } |
729 | if (y_start < io->crop_top) { |
730 | const int delta = io->crop_top - y_start; |
731 | y_start = io->crop_top; |
732 | *in_data += delta * pixel_stride; |
733 | } |
734 | if (y_start >= y_end) return 0; // Crop window is empty. |
735 | |
736 | *in_data += io->crop_left * sizeof(uint32_t); |
737 | |
738 | io->mb_y = y_start - io->crop_top; |
739 | io->mb_w = io->crop_right - io->crop_left; |
740 | io->mb_h = y_end - y_start; |
741 | return 1; // Non-empty crop window. |
742 | } |
743 | |
744 | //------------------------------------------------------------------------------ |
745 | |
746 | static WEBP_INLINE int GetMetaIndex( |
747 | const uint32_t* const image, int xsize, int bits, int x, int y) { |
748 | if (bits == 0) return 0; |
749 | return image[xsize * (y >> bits) + (x >> bits)]; |
750 | } |
751 | |
752 | static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr, |
753 | int x, int y) { |
754 | const int meta_index = GetMetaIndex(hdr->huffman_image_, hdr->huffman_xsize_, |
755 | hdr->huffman_subsample_bits_, x, y); |
756 | assert(meta_index < hdr->num_htree_groups_); |
757 | return hdr->htree_groups_ + meta_index; |
758 | } |
759 | |
760 | //------------------------------------------------------------------------------ |
761 | // Main loop, with custom row-processing function |
762 | |
763 | typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row); |
764 | |
765 | static void ApplyInverseTransforms(VP8LDecoder* const dec, |
766 | int start_row, int num_rows, |
767 | const uint32_t* const rows) { |
768 | int n = dec->next_transform_; |
769 | const int cache_pixs = dec->width_ * num_rows; |
770 | const int end_row = start_row + num_rows; |
771 | const uint32_t* rows_in = rows; |
772 | uint32_t* const rows_out = dec->argb_cache_; |
773 | |
774 | // Inverse transforms. |
775 | while (n-- > 0) { |
776 | VP8LTransform* const transform = &dec->transforms_[n]; |
777 | VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out); |
778 | rows_in = rows_out; |
779 | } |
780 | if (rows_in != rows_out) { |
781 | // No transform called, hence just copy. |
782 | memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out)); |
783 | } |
784 | } |
785 | |
786 | // Processes (transforms, scales & color-converts) the rows decoded after the |
787 | // last call. |
788 | static void ProcessRows(VP8LDecoder* const dec, int row) { |
789 | const uint32_t* const rows = dec->pixels_ + dec->width_ * dec->last_row_; |
790 | const int num_rows = row - dec->last_row_; |
791 | |
792 | assert(row <= dec->io_->crop_bottom); |
793 | // We can't process more than NUM_ARGB_CACHE_ROWS at a time (that's the size |
794 | // of argb_cache_), but we currently don't need more than that. |
795 | assert(num_rows <= NUM_ARGB_CACHE_ROWS); |
796 | if (num_rows > 0) { // Emit output. |
797 | VP8Io* const io = dec->io_; |
798 | uint8_t* rows_data = (uint8_t*)dec->argb_cache_; |
799 | const int in_stride = io->width * sizeof(uint32_t); // in unit of RGBA |
800 | ApplyInverseTransforms(dec, dec->last_row_, num_rows, rows); |
801 | if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) { |
802 | // Nothing to output (this time). |
803 | } else { |
804 | const WebPDecBuffer* const output = dec->output_; |
805 | if (WebPIsRGBMode(output->colorspace)) { // convert to RGBA |
806 | const WebPRGBABuffer* const buf = &output->u.RGBA; |
807 | uint8_t* const rgba = |
808 | buf->rgba + (int64_t)dec->last_out_row_ * buf->stride; |
809 | const int num_rows_out = |
810 | #if !defined(WEBP_REDUCE_SIZE) |
811 | io->use_scaling ? |
812 | EmitRescaledRowsRGBA(dec, rows_data, in_stride, io->mb_h, |
813 | rgba, buf->stride) : |
814 | #endif // WEBP_REDUCE_SIZE |
815 | EmitRows(output->colorspace, rows_data, in_stride, |
816 | io->mb_w, io->mb_h, rgba, buf->stride); |
817 | // Update 'last_out_row_'. |
818 | dec->last_out_row_ += num_rows_out; |
819 | } else { // convert to YUVA |
820 | dec->last_out_row_ = io->use_scaling ? |
821 | EmitRescaledRowsYUVA(dec, rows_data, in_stride, io->mb_h) : |
822 | EmitRowsYUVA(dec, rows_data, in_stride, io->mb_w, io->mb_h); |
823 | } |
824 | assert(dec->last_out_row_ <= output->height); |
825 | } |
826 | } |
827 | |
828 | // Update 'last_row_'. |
829 | dec->last_row_ = row; |
830 | assert(dec->last_row_ <= dec->height_); |
831 | } |
832 | |
833 | // Row-processing for the special case when alpha data contains only one |
834 | // transform (color indexing), and trivial non-green literals. |
835 | static int Is8bOptimizable(const VP8LMetadata* const hdr) { |
836 | int i; |
837 | if (hdr->color_cache_size_ > 0) return 0; |
838 | // When the Huffman tree contains only one symbol, we can skip the |
839 | // call to ReadSymbol() for red/blue/alpha channels. |
840 | for (i = 0; i < hdr->num_htree_groups_; ++i) { |
841 | HuffmanCode** const htrees = hdr->htree_groups_[i].htrees; |
842 | if (htrees[RED][0].bits > 0) return 0; |
843 | if (htrees[BLUE][0].bits > 0) return 0; |
844 | if (htrees[ALPHA][0].bits > 0) return 0; |
845 | } |
846 | return 1; |
847 | } |
848 | |
849 | static void AlphaApplyFilter(ALPHDecoder* const alph_dec, |
850 | int first_row, int last_row, |
851 | uint8_t* out, int stride) { |
852 | if (alph_dec->filter_ != WEBP_FILTER_NONE) { |
853 | int y; |
854 | const uint8_t* prev_line = alph_dec->prev_line_; |
855 | assert(WebPUnfilters[alph_dec->filter_] != NULL); |
856 | for (y = first_row; y < last_row; ++y) { |
857 | WebPUnfilters[alph_dec->filter_](prev_line, out, out, stride); |
858 | prev_line = out; |
859 | out += stride; |
860 | } |
861 | alph_dec->prev_line_ = prev_line; |
862 | } |
863 | } |
864 | |
865 | static void (VP8LDecoder* const dec, int last_row) { |
866 | // For vertical and gradient filtering, we need to decode the part above the |
867 | // crop_top row, in order to have the correct spatial predictors. |
868 | ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; |
869 | const int top_row = |
870 | (alph_dec->filter_ == WEBP_FILTER_NONE || |
871 | alph_dec->filter_ == WEBP_FILTER_HORIZONTAL) ? dec->io_->crop_top |
872 | : dec->last_row_; |
873 | const int first_row = (dec->last_row_ < top_row) ? top_row : dec->last_row_; |
874 | assert(last_row <= dec->io_->crop_bottom); |
875 | if (last_row > first_row) { |
876 | // Special method for paletted alpha data. We only process the cropped area. |
877 | const int width = dec->io_->width; |
878 | uint8_t* out = alph_dec->output_ + width * first_row; |
879 | const uint8_t* const in = |
880 | (uint8_t*)dec->pixels_ + dec->width_ * first_row; |
881 | VP8LTransform* const transform = &dec->transforms_[0]; |
882 | assert(dec->next_transform_ == 1); |
883 | assert(transform->type_ == COLOR_INDEXING_TRANSFORM); |
884 | VP8LColorIndexInverseTransformAlpha(transform, first_row, last_row, |
885 | in, out); |
886 | AlphaApplyFilter(alph_dec, first_row, last_row, out, width); |
887 | } |
888 | dec->last_row_ = dec->last_out_row_ = last_row; |
889 | } |
890 | |
891 | //------------------------------------------------------------------------------ |
892 | // Helper functions for fast pattern copy (8b and 32b) |
893 | |
894 | // cyclic rotation of pattern word |
895 | static WEBP_INLINE uint32_t Rotate8b(uint32_t V) { |
896 | #if defined(WORDS_BIGENDIAN) |
897 | return ((V & 0xff000000u) >> 24) | (V << 8); |
898 | #else |
899 | return ((V & 0xffu) << 24) | (V >> 8); |
900 | #endif |
901 | } |
902 | |
903 | // copy 1, 2 or 4-bytes pattern |
904 | static WEBP_INLINE void CopySmallPattern8b(const uint8_t* src, uint8_t* dst, |
905 | int length, uint32_t pattern) { |
906 | int i; |
907 | // align 'dst' to 4-bytes boundary. Adjust the pattern along the way. |
908 | while ((uintptr_t)dst & 3) { |
909 | *dst++ = *src++; |
910 | pattern = Rotate8b(pattern); |
911 | --length; |
912 | } |
913 | // Copy the pattern 4 bytes at a time. |
914 | for (i = 0; i < (length >> 2); ++i) { |
915 | ((uint32_t*)dst)[i] = pattern; |
916 | } |
917 | // Finish with left-overs. 'pattern' is still correctly positioned, |
918 | // so no Rotate8b() call is needed. |
919 | for (i <<= 2; i < length; ++i) { |
920 | dst[i] = src[i]; |
921 | } |
922 | } |
923 | |
924 | static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) { |
925 | const uint8_t* src = dst - dist; |
926 | if (length >= 8) { |
927 | uint32_t pattern = 0; |
928 | switch (dist) { |
929 | case 1: |
930 | pattern = src[0]; |
931 | #if defined(__arm__) || defined(_M_ARM) // arm doesn't like multiply that much |
932 | pattern |= pattern << 8; |
933 | pattern |= pattern << 16; |
934 | #elif defined(WEBP_USE_MIPS_DSP_R2) |
935 | __asm__ volatile ("replv.qb %0, %0" : "+r" (pattern)); |
936 | #else |
937 | pattern = 0x01010101u * pattern; |
938 | #endif |
939 | break; |
940 | case 2: |
941 | #if !defined(WORDS_BIGENDIAN) |
942 | memcpy(&pattern, src, sizeof(uint16_t)); |
943 | #else |
944 | pattern = ((uint32_t)src[0] << 8) | src[1]; |
945 | #endif |
946 | #if defined(__arm__) || defined(_M_ARM) |
947 | pattern |= pattern << 16; |
948 | #elif defined(WEBP_USE_MIPS_DSP_R2) |
949 | __asm__ volatile ("replv.ph %0, %0" : "+r" (pattern)); |
950 | #else |
951 | pattern = 0x00010001u * pattern; |
952 | #endif |
953 | break; |
954 | case 4: |
955 | memcpy(&pattern, src, sizeof(uint32_t)); |
956 | break; |
957 | default: |
958 | goto Copy; |
959 | } |
960 | CopySmallPattern8b(src, dst, length, pattern); |
961 | return; |
962 | } |
963 | Copy: |
964 | if (dist >= length) { // no overlap -> use memcpy() |
965 | memcpy(dst, src, length * sizeof(*dst)); |
966 | } else { |
967 | int i; |
968 | for (i = 0; i < length; ++i) dst[i] = src[i]; |
969 | } |
970 | } |
971 | |
972 | // copy pattern of 1 or 2 uint32_t's |
973 | static WEBP_INLINE void CopySmallPattern32b(const uint32_t* src, |
974 | uint32_t* dst, |
975 | int length, uint64_t pattern) { |
976 | int i; |
977 | if ((uintptr_t)dst & 4) { // Align 'dst' to 8-bytes boundary. |
978 | *dst++ = *src++; |
979 | pattern = (pattern >> 32) | (pattern << 32); |
980 | --length; |
981 | } |
982 | assert(0 == ((uintptr_t)dst & 7)); |
983 | for (i = 0; i < (length >> 1); ++i) { |
984 | ((uint64_t*)dst)[i] = pattern; // Copy the pattern 8 bytes at a time. |
985 | } |
986 | if (length & 1) { // Finish with left-over. |
987 | dst[i << 1] = src[i << 1]; |
988 | } |
989 | } |
990 | |
991 | static WEBP_INLINE void CopyBlock32b(uint32_t* const dst, |
992 | int dist, int length) { |
993 | const uint32_t* const src = dst - dist; |
994 | if (dist <= 2 && length >= 4 && ((uintptr_t)dst & 3) == 0) { |
995 | uint64_t pattern; |
996 | if (dist == 1) { |
997 | pattern = (uint64_t)src[0]; |
998 | pattern |= pattern << 32; |
999 | } else { |
1000 | memcpy(&pattern, src, sizeof(pattern)); |
1001 | } |
1002 | CopySmallPattern32b(src, dst, length, pattern); |
1003 | } else if (dist >= length) { // no overlap |
1004 | memcpy(dst, src, length * sizeof(*dst)); |
1005 | } else { |
1006 | int i; |
1007 | for (i = 0; i < length; ++i) dst[i] = src[i]; |
1008 | } |
1009 | } |
1010 | |
1011 | //------------------------------------------------------------------------------ |
1012 | |
1013 | static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data, |
1014 | int width, int height, int last_row) { |
1015 | int ok = 1; |
1016 | int row = dec->last_pixel_ / width; |
1017 | int col = dec->last_pixel_ % width; |
1018 | VP8LBitReader* const br = &dec->br_; |
1019 | VP8LMetadata* const hdr = &dec->hdr_; |
1020 | int pos = dec->last_pixel_; // current position |
1021 | const int end = width * height; // End of data |
1022 | const int last = width * last_row; // Last pixel to decode |
1023 | const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; |
1024 | const int mask = hdr->huffman_mask_; |
1025 | const HTreeGroup* htree_group = |
1026 | (pos < last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; |
1027 | assert(pos <= end); |
1028 | assert(last_row <= height); |
1029 | assert(Is8bOptimizable(hdr)); |
1030 | |
1031 | while (!br->eos_ && pos < last) { |
1032 | int code; |
1033 | // Only update when changing tile. |
1034 | if ((col & mask) == 0) { |
1035 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
1036 | } |
1037 | assert(htree_group != NULL); |
1038 | VP8LFillBitWindow(br); |
1039 | code = ReadSymbol(htree_group->htrees[GREEN], br); |
1040 | if (code < NUM_LITERAL_CODES) { // Literal |
1041 | data[pos] = code; |
1042 | ++pos; |
1043 | ++col; |
1044 | if (col >= width) { |
1045 | col = 0; |
1046 | ++row; |
1047 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
1048 | ExtractPalettedAlphaRows(dec, row); |
1049 | } |
1050 | } |
1051 | } else if (code < len_code_limit) { // Backward reference |
1052 | int dist_code, dist; |
1053 | const int length_sym = code - NUM_LITERAL_CODES; |
1054 | const int length = GetCopyLength(length_sym, br); |
1055 | const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); |
1056 | VP8LFillBitWindow(br); |
1057 | dist_code = GetCopyDistance(dist_symbol, br); |
1058 | dist = PlaneCodeToDistance(width, dist_code); |
1059 | if (pos >= dist && end - pos >= length) { |
1060 | CopyBlock8b(data + pos, dist, length); |
1061 | } else { |
1062 | ok = 0; |
1063 | goto End; |
1064 | } |
1065 | pos += length; |
1066 | col += length; |
1067 | while (col >= width) { |
1068 | col -= width; |
1069 | ++row; |
1070 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
1071 | ExtractPalettedAlphaRows(dec, row); |
1072 | } |
1073 | } |
1074 | if (pos < last && (col & mask)) { |
1075 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
1076 | } |
1077 | } else { // Not reached |
1078 | ok = 0; |
1079 | goto End; |
1080 | } |
1081 | br->eos_ = VP8LIsEndOfStream(br); |
1082 | } |
1083 | // Process the remaining rows corresponding to last row-block. |
1084 | ExtractPalettedAlphaRows(dec, row > last_row ? last_row : row); |
1085 | |
1086 | End: |
1087 | br->eos_ = VP8LIsEndOfStream(br); |
1088 | if (!ok || (br->eos_ && pos < end)) { |
1089 | ok = 0; |
1090 | dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED |
1091 | : VP8_STATUS_BITSTREAM_ERROR; |
1092 | } else { |
1093 | dec->last_pixel_ = pos; |
1094 | } |
1095 | return ok; |
1096 | } |
1097 | |
1098 | static void SaveState(VP8LDecoder* const dec, int last_pixel) { |
1099 | assert(dec->incremental_); |
1100 | dec->saved_br_ = dec->br_; |
1101 | dec->saved_last_pixel_ = last_pixel; |
1102 | if (dec->hdr_.color_cache_size_ > 0) { |
1103 | VP8LColorCacheCopy(&dec->hdr_.color_cache_, &dec->hdr_.saved_color_cache_); |
1104 | } |
1105 | } |
1106 | |
1107 | static void RestoreState(VP8LDecoder* const dec) { |
1108 | assert(dec->br_.eos_); |
1109 | dec->status_ = VP8_STATUS_SUSPENDED; |
1110 | dec->br_ = dec->saved_br_; |
1111 | dec->last_pixel_ = dec->saved_last_pixel_; |
1112 | if (dec->hdr_.color_cache_size_ > 0) { |
1113 | VP8LColorCacheCopy(&dec->hdr_.saved_color_cache_, &dec->hdr_.color_cache_); |
1114 | } |
1115 | } |
1116 | |
1117 | #define SYNC_EVERY_N_ROWS 8 // minimum number of rows between check-points |
1118 | static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data, |
1119 | int width, int height, int last_row, |
1120 | ProcessRowsFunc process_func) { |
1121 | int row = dec->last_pixel_ / width; |
1122 | int col = dec->last_pixel_ % width; |
1123 | VP8LBitReader* const br = &dec->br_; |
1124 | VP8LMetadata* const hdr = &dec->hdr_; |
1125 | uint32_t* src = data + dec->last_pixel_; |
1126 | uint32_t* last_cached = src; |
1127 | uint32_t* const src_end = data + width * height; // End of data |
1128 | uint32_t* const src_last = data + width * last_row; // Last pixel to decode |
1129 | const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; |
1130 | const int color_cache_limit = len_code_limit + hdr->color_cache_size_; |
1131 | int next_sync_row = dec->incremental_ ? row : 1 << 24; |
1132 | VP8LColorCache* const color_cache = |
1133 | (hdr->color_cache_size_ > 0) ? &hdr->color_cache_ : NULL; |
1134 | const int mask = hdr->huffman_mask_; |
1135 | const HTreeGroup* htree_group = |
1136 | (src < src_last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; |
1137 | assert(dec->last_row_ < last_row); |
1138 | assert(src_last <= src_end); |
1139 | |
1140 | while (src < src_last) { |
1141 | int code; |
1142 | if (row >= next_sync_row) { |
1143 | SaveState(dec, (int)(src - data)); |
1144 | next_sync_row = row + SYNC_EVERY_N_ROWS; |
1145 | } |
1146 | // Only update when changing tile. Note we could use this test: |
1147 | // if "((((prev_col ^ col) | prev_row ^ row)) > mask)" -> tile changed |
1148 | // but that's actually slower and needs storing the previous col/row. |
1149 | if ((col & mask) == 0) { |
1150 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
1151 | } |
1152 | assert(htree_group != NULL); |
1153 | if (htree_group->is_trivial_code) { |
1154 | *src = htree_group->literal_arb; |
1155 | goto AdvanceByOne; |
1156 | } |
1157 | VP8LFillBitWindow(br); |
1158 | if (htree_group->use_packed_table) { |
1159 | code = ReadPackedSymbols(htree_group, br, src); |
1160 | if (VP8LIsEndOfStream(br)) break; |
1161 | if (code == PACKED_NON_LITERAL_CODE) goto AdvanceByOne; |
1162 | } else { |
1163 | code = ReadSymbol(htree_group->htrees[GREEN], br); |
1164 | } |
1165 | if (VP8LIsEndOfStream(br)) break; |
1166 | if (code < NUM_LITERAL_CODES) { // Literal |
1167 | if (htree_group->is_trivial_literal) { |
1168 | *src = htree_group->literal_arb | (code << 8); |
1169 | } else { |
1170 | int red, blue, alpha; |
1171 | red = ReadSymbol(htree_group->htrees[RED], br); |
1172 | VP8LFillBitWindow(br); |
1173 | blue = ReadSymbol(htree_group->htrees[BLUE], br); |
1174 | alpha = ReadSymbol(htree_group->htrees[ALPHA], br); |
1175 | if (VP8LIsEndOfStream(br)) break; |
1176 | *src = ((uint32_t)alpha << 24) | (red << 16) | (code << 8) | blue; |
1177 | } |
1178 | AdvanceByOne: |
1179 | ++src; |
1180 | ++col; |
1181 | if (col >= width) { |
1182 | col = 0; |
1183 | ++row; |
1184 | if (process_func != NULL) { |
1185 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
1186 | process_func(dec, row); |
1187 | } |
1188 | } |
1189 | if (color_cache != NULL) { |
1190 | while (last_cached < src) { |
1191 | VP8LColorCacheInsert(color_cache, *last_cached++); |
1192 | } |
1193 | } |
1194 | } |
1195 | } else if (code < len_code_limit) { // Backward reference |
1196 | int dist_code, dist; |
1197 | const int length_sym = code - NUM_LITERAL_CODES; |
1198 | const int length = GetCopyLength(length_sym, br); |
1199 | const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); |
1200 | VP8LFillBitWindow(br); |
1201 | dist_code = GetCopyDistance(dist_symbol, br); |
1202 | dist = PlaneCodeToDistance(width, dist_code); |
1203 | |
1204 | if (VP8LIsEndOfStream(br)) break; |
1205 | if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) { |
1206 | goto Error; |
1207 | } else { |
1208 | CopyBlock32b(src, dist, length); |
1209 | } |
1210 | src += length; |
1211 | col += length; |
1212 | while (col >= width) { |
1213 | col -= width; |
1214 | ++row; |
1215 | if (process_func != NULL) { |
1216 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
1217 | process_func(dec, row); |
1218 | } |
1219 | } |
1220 | } |
1221 | // Because of the check done above (before 'src' was incremented by |
1222 | // 'length'), the following holds true. |
1223 | assert(src <= src_end); |
1224 | if (col & mask) htree_group = GetHtreeGroupForPos(hdr, col, row); |
1225 | if (color_cache != NULL) { |
1226 | while (last_cached < src) { |
1227 | VP8LColorCacheInsert(color_cache, *last_cached++); |
1228 | } |
1229 | } |
1230 | } else if (code < color_cache_limit) { // Color cache |
1231 | const int key = code - len_code_limit; |
1232 | assert(color_cache != NULL); |
1233 | while (last_cached < src) { |
1234 | VP8LColorCacheInsert(color_cache, *last_cached++); |
1235 | } |
1236 | *src = VP8LColorCacheLookup(color_cache, key); |
1237 | goto AdvanceByOne; |
1238 | } else { // Not reached |
1239 | goto Error; |
1240 | } |
1241 | } |
1242 | |
1243 | br->eos_ = VP8LIsEndOfStream(br); |
1244 | if (dec->incremental_ && br->eos_ && src < src_end) { |
1245 | RestoreState(dec); |
1246 | } else if (!br->eos_) { |
1247 | // Process the remaining rows corresponding to last row-block. |
1248 | if (process_func != NULL) { |
1249 | process_func(dec, row > last_row ? last_row : row); |
1250 | } |
1251 | dec->status_ = VP8_STATUS_OK; |
1252 | dec->last_pixel_ = (int)(src - data); // end-of-scan marker |
1253 | } else { |
1254 | // if not incremental, and we are past the end of buffer (eos_=1), then this |
1255 | // is a real bitstream error. |
1256 | goto Error; |
1257 | } |
1258 | return 1; |
1259 | |
1260 | Error: |
1261 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
1262 | return 0; |
1263 | } |
1264 | |
1265 | // ----------------------------------------------------------------------------- |
1266 | // VP8LTransform |
1267 | |
1268 | static void ClearTransform(VP8LTransform* const transform) { |
1269 | WebPSafeFree(transform->data_); |
1270 | transform->data_ = NULL; |
1271 | } |
1272 | |
1273 | // For security reason, we need to remap the color map to span |
1274 | // the total possible bundled values, and not just the num_colors. |
1275 | static int ExpandColorMap(int num_colors, VP8LTransform* const transform) { |
1276 | int i; |
1277 | const int final_num_colors = 1 << (8 >> transform->bits_); |
1278 | uint32_t* const new_color_map = |
1279 | (uint32_t*)WebPSafeMalloc((uint64_t)final_num_colors, |
1280 | sizeof(*new_color_map)); |
1281 | if (new_color_map == NULL) { |
1282 | return 0; |
1283 | } else { |
1284 | uint8_t* const data = (uint8_t*)transform->data_; |
1285 | uint8_t* const new_data = (uint8_t*)new_color_map; |
1286 | new_color_map[0] = transform->data_[0]; |
1287 | for (i = 4; i < 4 * num_colors; ++i) { |
1288 | // Equivalent to VP8LAddPixels(), on a byte-basis. |
1289 | new_data[i] = (data[i] + new_data[i - 4]) & 0xff; |
1290 | } |
1291 | for (; i < 4 * final_num_colors; ++i) { |
1292 | new_data[i] = 0; // black tail. |
1293 | } |
1294 | WebPSafeFree(transform->data_); |
1295 | transform->data_ = new_color_map; |
1296 | } |
1297 | return 1; |
1298 | } |
1299 | |
1300 | static int ReadTransform(int* const xsize, int const* ysize, |
1301 | VP8LDecoder* const dec) { |
1302 | int ok = 1; |
1303 | VP8LBitReader* const br = &dec->br_; |
1304 | VP8LTransform* transform = &dec->transforms_[dec->next_transform_]; |
1305 | const VP8LImageTransformType type = |
1306 | (VP8LImageTransformType)VP8LReadBits(br, 2); |
1307 | |
1308 | // Each transform type can only be present once in the stream. |
1309 | if (dec->transforms_seen_ & (1U << type)) { |
1310 | return 0; // Already there, let's not accept the second same transform. |
1311 | } |
1312 | dec->transforms_seen_ |= (1U << type); |
1313 | |
1314 | transform->type_ = type; |
1315 | transform->xsize_ = *xsize; |
1316 | transform->ysize_ = *ysize; |
1317 | transform->data_ = NULL; |
1318 | ++dec->next_transform_; |
1319 | assert(dec->next_transform_ <= NUM_TRANSFORMS); |
1320 | |
1321 | switch (type) { |
1322 | case PREDICTOR_TRANSFORM: |
1323 | case CROSS_COLOR_TRANSFORM: |
1324 | transform->bits_ = VP8LReadBits(br, 3) + 2; |
1325 | ok = DecodeImageStream(VP8LSubSampleSize(transform->xsize_, |
1326 | transform->bits_), |
1327 | VP8LSubSampleSize(transform->ysize_, |
1328 | transform->bits_), |
1329 | 0, dec, &transform->data_); |
1330 | break; |
1331 | case COLOR_INDEXING_TRANSFORM: { |
1332 | const int num_colors = VP8LReadBits(br, 8) + 1; |
1333 | const int bits = (num_colors > 16) ? 0 |
1334 | : (num_colors > 4) ? 1 |
1335 | : (num_colors > 2) ? 2 |
1336 | : 3; |
1337 | *xsize = VP8LSubSampleSize(transform->xsize_, bits); |
1338 | transform->bits_ = bits; |
1339 | ok = DecodeImageStream(num_colors, 1, 0, dec, &transform->data_); |
1340 | ok = ok && ExpandColorMap(num_colors, transform); |
1341 | break; |
1342 | } |
1343 | case SUBTRACT_GREEN_TRANSFORM: |
1344 | break; |
1345 | default: |
1346 | assert(0); // can't happen |
1347 | break; |
1348 | } |
1349 | |
1350 | return ok; |
1351 | } |
1352 | |
1353 | // ----------------------------------------------------------------------------- |
1354 | // VP8LMetadata |
1355 | |
1356 | static void InitMetadata(VP8LMetadata* const hdr) { |
1357 | assert(hdr != NULL); |
1358 | memset(hdr, 0, sizeof(*hdr)); |
1359 | } |
1360 | |
1361 | static void ClearMetadata(VP8LMetadata* const hdr) { |
1362 | assert(hdr != NULL); |
1363 | |
1364 | WebPSafeFree(hdr->huffman_image_); |
1365 | VP8LHuffmanTablesDeallocate(&hdr->huffman_tables_); |
1366 | VP8LHtreeGroupsFree(hdr->htree_groups_); |
1367 | VP8LColorCacheClear(&hdr->color_cache_); |
1368 | VP8LColorCacheClear(&hdr->saved_color_cache_); |
1369 | InitMetadata(hdr); |
1370 | } |
1371 | |
1372 | // ----------------------------------------------------------------------------- |
1373 | // VP8LDecoder |
1374 | |
1375 | VP8LDecoder* VP8LNew(void) { |
1376 | VP8LDecoder* const dec = (VP8LDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec)); |
1377 | if (dec == NULL) return NULL; |
1378 | dec->status_ = VP8_STATUS_OK; |
1379 | dec->state_ = READ_DIM; |
1380 | |
1381 | VP8LDspInit(); // Init critical function pointers. |
1382 | |
1383 | return dec; |
1384 | } |
1385 | |
1386 | void VP8LClear(VP8LDecoder* const dec) { |
1387 | int i; |
1388 | if (dec == NULL) return; |
1389 | ClearMetadata(&dec->hdr_); |
1390 | |
1391 | WebPSafeFree(dec->pixels_); |
1392 | dec->pixels_ = NULL; |
1393 | for (i = 0; i < dec->next_transform_; ++i) { |
1394 | ClearTransform(&dec->transforms_[i]); |
1395 | } |
1396 | dec->next_transform_ = 0; |
1397 | dec->transforms_seen_ = 0; |
1398 | |
1399 | WebPSafeFree(dec->rescaler_memory); |
1400 | dec->rescaler_memory = NULL; |
1401 | |
1402 | dec->output_ = NULL; // leave no trace behind |
1403 | } |
1404 | |
1405 | void VP8LDelete(VP8LDecoder* const dec) { |
1406 | if (dec != NULL) { |
1407 | VP8LClear(dec); |
1408 | WebPSafeFree(dec); |
1409 | } |
1410 | } |
1411 | |
1412 | static void UpdateDecoder(VP8LDecoder* const dec, int width, int height) { |
1413 | VP8LMetadata* const hdr = &dec->hdr_; |
1414 | const int num_bits = hdr->huffman_subsample_bits_; |
1415 | dec->width_ = width; |
1416 | dec->height_ = height; |
1417 | |
1418 | hdr->huffman_xsize_ = VP8LSubSampleSize(width, num_bits); |
1419 | hdr->huffman_mask_ = (num_bits == 0) ? ~0 : (1 << num_bits) - 1; |
1420 | } |
1421 | |
1422 | static int DecodeImageStream(int xsize, int ysize, |
1423 | int is_level0, |
1424 | VP8LDecoder* const dec, |
1425 | uint32_t** const decoded_data) { |
1426 | int ok = 1; |
1427 | int transform_xsize = xsize; |
1428 | int transform_ysize = ysize; |
1429 | VP8LBitReader* const br = &dec->br_; |
1430 | VP8LMetadata* const hdr = &dec->hdr_; |
1431 | uint32_t* data = NULL; |
1432 | int color_cache_bits = 0; |
1433 | |
1434 | // Read the transforms (may recurse). |
1435 | if (is_level0) { |
1436 | while (ok && VP8LReadBits(br, 1)) { |
1437 | ok = ReadTransform(&transform_xsize, &transform_ysize, dec); |
1438 | } |
1439 | } |
1440 | |
1441 | // Color cache |
1442 | if (ok && VP8LReadBits(br, 1)) { |
1443 | color_cache_bits = VP8LReadBits(br, 4); |
1444 | ok = (color_cache_bits >= 1 && color_cache_bits <= MAX_CACHE_BITS); |
1445 | if (!ok) { |
1446 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
1447 | goto End; |
1448 | } |
1449 | } |
1450 | |
1451 | // Read the Huffman codes (may recurse). |
1452 | ok = ok && ReadHuffmanCodes(dec, transform_xsize, transform_ysize, |
1453 | color_cache_bits, is_level0); |
1454 | if (!ok) { |
1455 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
1456 | goto End; |
1457 | } |
1458 | |
1459 | // Finish setting up the color-cache |
1460 | if (color_cache_bits > 0) { |
1461 | hdr->color_cache_size_ = 1 << color_cache_bits; |
1462 | if (!VP8LColorCacheInit(&hdr->color_cache_, color_cache_bits)) { |
1463 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
1464 | ok = 0; |
1465 | goto End; |
1466 | } |
1467 | } else { |
1468 | hdr->color_cache_size_ = 0; |
1469 | } |
1470 | UpdateDecoder(dec, transform_xsize, transform_ysize); |
1471 | |
1472 | if (is_level0) { // level 0 complete |
1473 | dec->state_ = READ_HDR; |
1474 | goto End; |
1475 | } |
1476 | |
1477 | { |
1478 | const uint64_t total_size = (uint64_t)transform_xsize * transform_ysize; |
1479 | data = (uint32_t*)WebPSafeMalloc(total_size, sizeof(*data)); |
1480 | if (data == NULL) { |
1481 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
1482 | ok = 0; |
1483 | goto End; |
1484 | } |
1485 | } |
1486 | |
1487 | // Use the Huffman trees to decode the LZ77 encoded data. |
1488 | ok = DecodeImageData(dec, data, transform_xsize, transform_ysize, |
1489 | transform_ysize, NULL); |
1490 | ok = ok && !br->eos_; |
1491 | |
1492 | End: |
1493 | if (!ok) { |
1494 | WebPSafeFree(data); |
1495 | ClearMetadata(hdr); |
1496 | } else { |
1497 | if (decoded_data != NULL) { |
1498 | *decoded_data = data; |
1499 | } else { |
1500 | // We allocate image data in this function only for transforms. At level 0 |
1501 | // (that is: not the transforms), we shouldn't have allocated anything. |
1502 | assert(data == NULL); |
1503 | assert(is_level0); |
1504 | } |
1505 | dec->last_pixel_ = 0; // Reset for future DECODE_DATA_FUNC() calls. |
1506 | if (!is_level0) ClearMetadata(hdr); // Clean up temporary data behind. |
1507 | } |
1508 | return ok; |
1509 | } |
1510 | |
1511 | //------------------------------------------------------------------------------ |
1512 | // Allocate internal buffers dec->pixels_ and dec->argb_cache_. |
1513 | static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) { |
1514 | const uint64_t num_pixels = (uint64_t)dec->width_ * dec->height_; |
1515 | // Scratch buffer corresponding to top-prediction row for transforming the |
1516 | // first row in the row-blocks. Not needed for paletted alpha. |
1517 | const uint64_t cache_top_pixels = (uint16_t)final_width; |
1518 | // Scratch buffer for temporary BGRA storage. Not needed for paletted alpha. |
1519 | const uint64_t cache_pixels = (uint64_t)final_width * NUM_ARGB_CACHE_ROWS; |
1520 | const uint64_t total_num_pixels = |
1521 | num_pixels + cache_top_pixels + cache_pixels; |
1522 | |
1523 | assert(dec->width_ <= final_width); |
1524 | dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint32_t)); |
1525 | if (dec->pixels_ == NULL) { |
1526 | dec->argb_cache_ = NULL; // for soundness |
1527 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
1528 | return 0; |
1529 | } |
1530 | dec->argb_cache_ = dec->pixels_ + num_pixels + cache_top_pixels; |
1531 | return 1; |
1532 | } |
1533 | |
1534 | static int AllocateInternalBuffers8b(VP8LDecoder* const dec) { |
1535 | const uint64_t total_num_pixels = (uint64_t)dec->width_ * dec->height_; |
1536 | dec->argb_cache_ = NULL; // for soundness |
1537 | dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint8_t)); |
1538 | if (dec->pixels_ == NULL) { |
1539 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
1540 | return 0; |
1541 | } |
1542 | return 1; |
1543 | } |
1544 | |
1545 | //------------------------------------------------------------------------------ |
1546 | |
1547 | // Special row-processing that only stores the alpha data. |
1548 | static void (VP8LDecoder* const dec, int last_row) { |
1549 | int cur_row = dec->last_row_; |
1550 | int num_rows = last_row - cur_row; |
1551 | const uint32_t* in = dec->pixels_ + dec->width_ * cur_row; |
1552 | |
1553 | assert(last_row <= dec->io_->crop_bottom); |
1554 | while (num_rows > 0) { |
1555 | const int num_rows_to_process = |
1556 | (num_rows > NUM_ARGB_CACHE_ROWS) ? NUM_ARGB_CACHE_ROWS : num_rows; |
1557 | // Extract alpha (which is stored in the green plane). |
1558 | ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; |
1559 | uint8_t* const output = alph_dec->output_; |
1560 | const int width = dec->io_->width; // the final width (!= dec->width_) |
1561 | const int cache_pixs = width * num_rows_to_process; |
1562 | uint8_t* const dst = output + width * cur_row; |
1563 | const uint32_t* const src = dec->argb_cache_; |
1564 | ApplyInverseTransforms(dec, cur_row, num_rows_to_process, in); |
1565 | WebPExtractGreen(src, dst, cache_pixs); |
1566 | AlphaApplyFilter(alph_dec, |
1567 | cur_row, cur_row + num_rows_to_process, dst, width); |
1568 | num_rows -= num_rows_to_process; |
1569 | in += num_rows_to_process * dec->width_; |
1570 | cur_row += num_rows_to_process; |
1571 | } |
1572 | assert(cur_row == last_row); |
1573 | dec->last_row_ = dec->last_out_row_ = last_row; |
1574 | } |
1575 | |
1576 | int (ALPHDecoder* const alph_dec, |
1577 | const uint8_t* const data, size_t data_size) { |
1578 | int ok = 0; |
1579 | VP8LDecoder* dec = VP8LNew(); |
1580 | |
1581 | if (dec == NULL) return 0; |
1582 | |
1583 | assert(alph_dec != NULL); |
1584 | |
1585 | dec->width_ = alph_dec->width_; |
1586 | dec->height_ = alph_dec->height_; |
1587 | dec->io_ = &alph_dec->io_; |
1588 | dec->io_->opaque = alph_dec; |
1589 | dec->io_->width = alph_dec->width_; |
1590 | dec->io_->height = alph_dec->height_; |
1591 | |
1592 | dec->status_ = VP8_STATUS_OK; |
1593 | VP8LInitBitReader(&dec->br_, data, data_size); |
1594 | |
1595 | if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, 1, dec, NULL)) { |
1596 | goto Err; |
1597 | } |
1598 | |
1599 | // Special case: if alpha data uses only the color indexing transform and |
1600 | // doesn't use color cache (a frequent case), we will use DecodeAlphaData() |
1601 | // method that only needs allocation of 1 byte per pixel (alpha channel). |
1602 | if (dec->next_transform_ == 1 && |
1603 | dec->transforms_[0].type_ == COLOR_INDEXING_TRANSFORM && |
1604 | Is8bOptimizable(&dec->hdr_)) { |
1605 | alph_dec->use_8b_decode_ = 1; |
1606 | ok = AllocateInternalBuffers8b(dec); |
1607 | } else { |
1608 | // Allocate internal buffers (note that dec->width_ may have changed here). |
1609 | alph_dec->use_8b_decode_ = 0; |
1610 | ok = AllocateInternalBuffers32b(dec, alph_dec->width_); |
1611 | } |
1612 | |
1613 | if (!ok) goto Err; |
1614 | |
1615 | // Only set here, once we are sure it is valid (to avoid thread races). |
1616 | alph_dec->vp8l_dec_ = dec; |
1617 | return 1; |
1618 | |
1619 | Err: |
1620 | VP8LDelete(dec); |
1621 | return 0; |
1622 | } |
1623 | |
1624 | int VP8LDecodeAlphaImageStream(ALPHDecoder* const alph_dec, int last_row) { |
1625 | VP8LDecoder* const dec = alph_dec->vp8l_dec_; |
1626 | assert(dec != NULL); |
1627 | assert(last_row <= dec->height_); |
1628 | |
1629 | if (dec->last_row_ >= last_row) { |
1630 | return 1; // done |
1631 | } |
1632 | |
1633 | if (!alph_dec->use_8b_decode_) WebPInitAlphaProcessing(); |
1634 | |
1635 | // Decode (with special row processing). |
1636 | return alph_dec->use_8b_decode_ ? |
1637 | DecodeAlphaData(dec, (uint8_t*)dec->pixels_, dec->width_, dec->height_, |
1638 | last_row) : |
1639 | DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, |
1640 | last_row, ExtractAlphaRows); |
1641 | } |
1642 | |
1643 | //------------------------------------------------------------------------------ |
1644 | |
1645 | int (VP8LDecoder* const dec, VP8Io* const io) { |
1646 | int width, height, has_alpha; |
1647 | |
1648 | if (dec == NULL) return 0; |
1649 | if (io == NULL) { |
1650 | dec->status_ = VP8_STATUS_INVALID_PARAM; |
1651 | return 0; |
1652 | } |
1653 | |
1654 | dec->io_ = io; |
1655 | dec->status_ = VP8_STATUS_OK; |
1656 | VP8LInitBitReader(&dec->br_, io->data, io->data_size); |
1657 | if (!ReadImageInfo(&dec->br_, &width, &height, &has_alpha)) { |
1658 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
1659 | goto Error; |
1660 | } |
1661 | dec->state_ = READ_DIM; |
1662 | io->width = width; |
1663 | io->height = height; |
1664 | |
1665 | if (!DecodeImageStream(width, height, 1, dec, NULL)) goto Error; |
1666 | return 1; |
1667 | |
1668 | Error: |
1669 | VP8LClear(dec); |
1670 | assert(dec->status_ != VP8_STATUS_OK); |
1671 | return 0; |
1672 | } |
1673 | |
1674 | int VP8LDecodeImage(VP8LDecoder* const dec) { |
1675 | VP8Io* io = NULL; |
1676 | WebPDecParams* params = NULL; |
1677 | |
1678 | if (dec == NULL) return 0; |
1679 | |
1680 | assert(dec->hdr_.huffman_tables_.root.start != NULL); |
1681 | assert(dec->hdr_.htree_groups_ != NULL); |
1682 | assert(dec->hdr_.num_htree_groups_ > 0); |
1683 | |
1684 | io = dec->io_; |
1685 | assert(io != NULL); |
1686 | params = (WebPDecParams*)io->opaque; |
1687 | assert(params != NULL); |
1688 | |
1689 | // Initialization. |
1690 | if (dec->state_ != READ_DATA) { |
1691 | dec->output_ = params->output; |
1692 | assert(dec->output_ != NULL); |
1693 | |
1694 | if (!WebPIoInitFromOptions(params->options, io, MODE_BGRA)) { |
1695 | dec->status_ = VP8_STATUS_INVALID_PARAM; |
1696 | goto Err; |
1697 | } |
1698 | |
1699 | if (!AllocateInternalBuffers32b(dec, io->width)) goto Err; |
1700 | |
1701 | #if !defined(WEBP_REDUCE_SIZE) |
1702 | if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err; |
1703 | #else |
1704 | if (io->use_scaling) { |
1705 | dec->status_ = VP8_STATUS_INVALID_PARAM; |
1706 | goto Err; |
1707 | } |
1708 | #endif |
1709 | if (io->use_scaling || WebPIsPremultipliedMode(dec->output_->colorspace)) { |
1710 | // need the alpha-multiply functions for premultiplied output or rescaling |
1711 | WebPInitAlphaProcessing(); |
1712 | } |
1713 | |
1714 | if (!WebPIsRGBMode(dec->output_->colorspace)) { |
1715 | WebPInitConvertARGBToYUV(); |
1716 | if (dec->output_->u.YUVA.a != NULL) WebPInitAlphaProcessing(); |
1717 | } |
1718 | if (dec->incremental_) { |
1719 | if (dec->hdr_.color_cache_size_ > 0 && |
1720 | dec->hdr_.saved_color_cache_.colors_ == NULL) { |
1721 | if (!VP8LColorCacheInit(&dec->hdr_.saved_color_cache_, |
1722 | dec->hdr_.color_cache_.hash_bits_)) { |
1723 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
1724 | goto Err; |
1725 | } |
1726 | } |
1727 | } |
1728 | dec->state_ = READ_DATA; |
1729 | } |
1730 | |
1731 | // Decode. |
1732 | if (!DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, |
1733 | io->crop_bottom, ProcessRows)) { |
1734 | goto Err; |
1735 | } |
1736 | |
1737 | params->last_y = dec->last_out_row_; |
1738 | return 1; |
1739 | |
1740 | Err: |
1741 | VP8LClear(dec); |
1742 | assert(dec->status_ != VP8_STATUS_OK); |
1743 | return 0; |
1744 | } |
1745 | |
1746 | //------------------------------------------------------------------------------ |
1747 | |