1 /* Copyright 2014 Google Inc. All Rights Reserved.
2 
3    Distributed under MIT license.
4    See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
5 */
6 
7 /* Brotli bit stream functions to support the low level format. There are no
8    compression algorithms here, just the right ordering of bits to match the
9    specs. */
10 
11 #include "./brotli_bit_stream.h"
12 
13 #include <string.h>  /* memcpy, memset */
14 
15 #include "../common/constants.h"
16 #include <brotli/types.h>
17 #include "./context.h"
18 #include "./entropy_encode.h"
19 #include "./entropy_encode_static.h"
20 #include "./fast_log.h"
21 #include "./memory.h"
22 #include "./port.h"
23 #include "./write_bits.h"
24 
25 #if defined(__cplusplus) || defined(c_plusplus)
26 extern "C" {
27 #endif
28 
29 #define MAX_HUFFMAN_TREE_SIZE (2 * BROTLI_NUM_COMMAND_SYMBOLS + 1)
30 /* The size of Huffman dictionary for distances assuming that NPOSTFIX = 0 and
31  NDIRECT = 0. */
32 #define SIMPLE_DISTANCE_ALPHABET_SIZE (BROTLI_NUM_DISTANCE_SHORT_CODES + \
33                                        (2 * BROTLI_MAX_DISTANCE_BITS))
34 /* SIMPLE_DISTANCE_ALPHABET_SIZE == 64 */
35 #define SIMPLE_DISTANCE_ALPHABET_BITS 6
36 
37 /* Represents the range of values belonging to a prefix code:
38    [offset, offset + 2^nbits) */
39 typedef struct PrefixCodeRange {
40   uint32_t offset;
41   uint32_t nbits;
42 } PrefixCodeRange;
43 
44 static const PrefixCodeRange
45     kBlockLengthPrefixCode[BROTLI_NUM_BLOCK_LEN_SYMBOLS] = {
46   { 1, 2}, { 5, 2}, { 9, 2}, {13, 2}, {17, 3}, { 25, 3}, { 33, 3},
47   {41, 3}, {49, 4}, {65, 4}, {81, 4}, {97, 4}, {113, 5}, {145, 5},
48   {177, 5}, { 209,  5}, { 241,  6}, { 305,  6}, { 369,  7}, {  497,  8},
49   {753, 9}, {1265, 10}, {2289, 11}, {4337, 12}, {8433, 13}, {16625, 24}
50 };
51 
BlockLengthPrefixCode(uint32_t len)52 static BROTLI_INLINE uint32_t BlockLengthPrefixCode(uint32_t len) {
53   uint32_t code = (len >= 177) ? (len >= 753 ? 20 : 14) : (len >= 41 ? 7 : 0);
54   while (code < (BROTLI_NUM_BLOCK_LEN_SYMBOLS - 1) &&
55       len >= kBlockLengthPrefixCode[code + 1].offset) ++code;
56   return code;
57 }
58 
GetBlockLengthPrefixCode(uint32_t len,size_t * code,uint32_t * n_extra,uint32_t * extra)59 static BROTLI_INLINE void GetBlockLengthPrefixCode(uint32_t len, size_t* code,
60     uint32_t* n_extra, uint32_t* extra) {
61   *code = BlockLengthPrefixCode(len);
62   *n_extra = kBlockLengthPrefixCode[*code].nbits;
63   *extra = len - kBlockLengthPrefixCode[*code].offset;
64 }
65 
66 typedef struct BlockTypeCodeCalculator {
67   size_t last_type;
68   size_t second_last_type;
69 } BlockTypeCodeCalculator;
70 
InitBlockTypeCodeCalculator(BlockTypeCodeCalculator * self)71 static void InitBlockTypeCodeCalculator(BlockTypeCodeCalculator* self) {
72   self->last_type = 1;
73   self->second_last_type = 0;
74 }
75 
NextBlockTypeCode(BlockTypeCodeCalculator * calculator,uint8_t type)76 static BROTLI_INLINE size_t NextBlockTypeCode(
77     BlockTypeCodeCalculator* calculator, uint8_t type) {
78   size_t type_code = (type == calculator->last_type + 1) ? 1u :
79       (type == calculator->second_last_type) ? 0u : type + 2u;
80   calculator->second_last_type = calculator->last_type;
81   calculator->last_type = type;
82   return type_code;
83 }
84 
85 /* |nibblesbits| represents the 2 bits to encode MNIBBLES (0-3)
86    REQUIRES: length > 0
87    REQUIRES: length <= (1 << 24) */
BrotliEncodeMlen(size_t length,uint64_t * bits,size_t * numbits,uint64_t * nibblesbits)88 static void BrotliEncodeMlen(size_t length, uint64_t* bits,
89                              size_t* numbits, uint64_t* nibblesbits) {
90   size_t lg = (length == 1) ? 1 : Log2FloorNonZero((uint32_t)(length - 1)) + 1;
91   size_t mnibbles = (lg < 16 ? 16 : (lg + 3)) / 4;
92   assert(length > 0);
93   assert(length <= (1 << 24));
94   assert(lg <= 24);
95   *nibblesbits = mnibbles - 4;
96   *numbits = mnibbles * 4;
97   *bits = length - 1;
98 }
99 
StoreCommandExtra(const Command * cmd,size_t * storage_ix,uint8_t * storage)100 static BROTLI_INLINE void StoreCommandExtra(
101     const Command* cmd, size_t* storage_ix, uint8_t* storage) {
102   uint32_t copylen_code = CommandCopyLenCode(cmd);
103   uint16_t inscode = GetInsertLengthCode(cmd->insert_len_);
104   uint16_t copycode = GetCopyLengthCode(copylen_code);
105   uint32_t insnumextra = GetInsertExtra(inscode);
106   uint64_t insextraval = cmd->insert_len_ - GetInsertBase(inscode);
107   uint64_t copyextraval = copylen_code - GetCopyBase(copycode);
108   uint64_t bits = (copyextraval << insnumextra) | insextraval;
109   BrotliWriteBits(
110       insnumextra + GetCopyExtra(copycode), bits, storage_ix, storage);
111 }
112 
113 /* Data structure that stores almost everything that is needed to encode each
114    block switch command. */
115 typedef struct BlockSplitCode {
116   BlockTypeCodeCalculator type_code_calculator;
117   uint8_t type_depths[BROTLI_MAX_BLOCK_TYPE_SYMBOLS];
118   uint16_t type_bits[BROTLI_MAX_BLOCK_TYPE_SYMBOLS];
119   uint8_t length_depths[BROTLI_NUM_BLOCK_LEN_SYMBOLS];
120   uint16_t length_bits[BROTLI_NUM_BLOCK_LEN_SYMBOLS];
121 } BlockSplitCode;
122 
123 /* Stores a number between 0 and 255. */
StoreVarLenUint8(size_t n,size_t * storage_ix,uint8_t * storage)124 static void StoreVarLenUint8(size_t n, size_t* storage_ix, uint8_t* storage) {
125   if (n == 0) {
126     BrotliWriteBits(1, 0, storage_ix, storage);
127   } else {
128     size_t nbits = Log2FloorNonZero(n);
129     BrotliWriteBits(1, 1, storage_ix, storage);
130     BrotliWriteBits(3, nbits, storage_ix, storage);
131     BrotliWriteBits(nbits, n - ((size_t)1 << nbits), storage_ix, storage);
132   }
133 }
134 
135 /* Stores the compressed meta-block header.
136    REQUIRES: length > 0
137    REQUIRES: length <= (1 << 24) */
StoreCompressedMetaBlockHeader(BROTLI_BOOL is_final_block,size_t length,size_t * storage_ix,uint8_t * storage)138 static void StoreCompressedMetaBlockHeader(BROTLI_BOOL is_final_block,
139                                            size_t length,
140                                            size_t* storage_ix,
141                                            uint8_t* storage) {
142   uint64_t lenbits;
143   size_t nlenbits;
144   uint64_t nibblesbits;
145 
146   /* Write ISLAST bit. */
147   BrotliWriteBits(1, (uint64_t)is_final_block, storage_ix, storage);
148   /* Write ISEMPTY bit. */
149   if (is_final_block) {
150     BrotliWriteBits(1, 0, storage_ix, storage);
151   }
152 
153   BrotliEncodeMlen(length, &lenbits, &nlenbits, &nibblesbits);
154   BrotliWriteBits(2, nibblesbits, storage_ix, storage);
155   BrotliWriteBits(nlenbits, lenbits, storage_ix, storage);
156 
157   if (!is_final_block) {
158     /* Write ISUNCOMPRESSED bit. */
159     BrotliWriteBits(1, 0, storage_ix, storage);
160   }
161 }
162 
163 /* Stores the uncompressed meta-block header.
164    REQUIRES: length > 0
165    REQUIRES: length <= (1 << 24) */
BrotliStoreUncompressedMetaBlockHeader(size_t length,size_t * storage_ix,uint8_t * storage)166 static void BrotliStoreUncompressedMetaBlockHeader(size_t length,
167                                                    size_t* storage_ix,
168                                                    uint8_t* storage) {
169   uint64_t lenbits;
170   size_t nlenbits;
171   uint64_t nibblesbits;
172 
173   /* Write ISLAST bit.
174      Uncompressed block cannot be the last one, so set to 0. */
175   BrotliWriteBits(1, 0, storage_ix, storage);
176   BrotliEncodeMlen(length, &lenbits, &nlenbits, &nibblesbits);
177   BrotliWriteBits(2, nibblesbits, storage_ix, storage);
178   BrotliWriteBits(nlenbits, lenbits, storage_ix, storage);
179   /* Write ISUNCOMPRESSED bit. */
180   BrotliWriteBits(1, 1, storage_ix, storage);
181 }
182 
BrotliStoreHuffmanTreeOfHuffmanTreeToBitMask(const int num_codes,const uint8_t * code_length_bitdepth,size_t * storage_ix,uint8_t * storage)183 static void BrotliStoreHuffmanTreeOfHuffmanTreeToBitMask(
184     const int num_codes, const uint8_t* code_length_bitdepth,
185     size_t* storage_ix, uint8_t* storage) {
186   static const uint8_t kStorageOrder[BROTLI_CODE_LENGTH_CODES] = {
187     1, 2, 3, 4, 0, 5, 17, 6, 16, 7, 8, 9, 10, 11, 12, 13, 14, 15
188   };
189   /* The bit lengths of the Huffman code over the code length alphabet
190      are compressed with the following static Huffman code:
191        Symbol   Code
192        ------   ----
193        0          00
194        1        1110
195        2         110
196        3          01
197        4          10
198        5        1111 */
199   static const uint8_t kHuffmanBitLengthHuffmanCodeSymbols[6] = {
200      0, 7, 3, 2, 1, 15
201   };
202   static const uint8_t kHuffmanBitLengthHuffmanCodeBitLengths[6] = {
203     2, 4, 3, 2, 2, 4
204   };
205 
206   size_t skip_some = 0;  /* skips none. */
207 
208   /* Throw away trailing zeros: */
209   size_t codes_to_store = BROTLI_CODE_LENGTH_CODES;
210   if (num_codes > 1) {
211     for (; codes_to_store > 0; --codes_to_store) {
212       if (code_length_bitdepth[kStorageOrder[codes_to_store - 1]] != 0) {
213         break;
214       }
215     }
216   }
217   if (code_length_bitdepth[kStorageOrder[0]] == 0 &&
218       code_length_bitdepth[kStorageOrder[1]] == 0) {
219     skip_some = 2;  /* skips two. */
220     if (code_length_bitdepth[kStorageOrder[2]] == 0) {
221       skip_some = 3;  /* skips three. */
222     }
223   }
224   BrotliWriteBits(2, skip_some, storage_ix, storage);
225   {
226     size_t i;
227     for (i = skip_some; i < codes_to_store; ++i) {
228       size_t l = code_length_bitdepth[kStorageOrder[i]];
229       BrotliWriteBits(kHuffmanBitLengthHuffmanCodeBitLengths[l],
230           kHuffmanBitLengthHuffmanCodeSymbols[l], storage_ix, storage);
231     }
232   }
233 }
234 
BrotliStoreHuffmanTreeToBitMask(const size_t huffman_tree_size,const uint8_t * huffman_tree,const uint8_t * huffman_tree_extra_bits,const uint8_t * code_length_bitdepth,const uint16_t * code_length_bitdepth_symbols,size_t * BROTLI_RESTRICT storage_ix,uint8_t * BROTLI_RESTRICT storage)235 static void BrotliStoreHuffmanTreeToBitMask(
236     const size_t huffman_tree_size, const uint8_t* huffman_tree,
237     const uint8_t* huffman_tree_extra_bits, const uint8_t* code_length_bitdepth,
238     const uint16_t* code_length_bitdepth_symbols,
239     size_t* BROTLI_RESTRICT storage_ix, uint8_t* BROTLI_RESTRICT storage) {
240   size_t i;
241   for (i = 0; i < huffman_tree_size; ++i) {
242     size_t ix = huffman_tree[i];
243     BrotliWriteBits(code_length_bitdepth[ix], code_length_bitdepth_symbols[ix],
244                     storage_ix, storage);
245     /* Extra bits */
246     switch (ix) {
247       case BROTLI_REPEAT_PREVIOUS_CODE_LENGTH:
248         BrotliWriteBits(2, huffman_tree_extra_bits[i], storage_ix, storage);
249         break;
250       case BROTLI_REPEAT_ZERO_CODE_LENGTH:
251         BrotliWriteBits(3, huffman_tree_extra_bits[i], storage_ix, storage);
252         break;
253     }
254   }
255 }
256 
StoreSimpleHuffmanTree(const uint8_t * depths,size_t symbols[4],size_t num_symbols,size_t max_bits,size_t * storage_ix,uint8_t * storage)257 static void StoreSimpleHuffmanTree(const uint8_t* depths,
258                                    size_t symbols[4],
259                                    size_t num_symbols,
260                                    size_t max_bits,
261                                    size_t *storage_ix, uint8_t *storage) {
262   /* value of 1 indicates a simple Huffman code */
263   BrotliWriteBits(2, 1, storage_ix, storage);
264   BrotliWriteBits(2, num_symbols - 1, storage_ix, storage);  /* NSYM - 1 */
265 
266   {
267     /* Sort */
268     size_t i;
269     for (i = 0; i < num_symbols; i++) {
270       size_t j;
271       for (j = i + 1; j < num_symbols; j++) {
272         if (depths[symbols[j]] < depths[symbols[i]]) {
273           BROTLI_SWAP(size_t, symbols, j, i);
274         }
275       }
276     }
277   }
278 
279   if (num_symbols == 2) {
280     BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
281     BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
282   } else if (num_symbols == 3) {
283     BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
284     BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
285     BrotliWriteBits(max_bits, symbols[2], storage_ix, storage);
286   } else {
287     BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
288     BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
289     BrotliWriteBits(max_bits, symbols[2], storage_ix, storage);
290     BrotliWriteBits(max_bits, symbols[3], storage_ix, storage);
291     /* tree-select */
292     BrotliWriteBits(1, depths[symbols[0]] == 1 ? 1 : 0, storage_ix, storage);
293   }
294 }
295 
296 /* num = alphabet size
297    depths = symbol depths */
BrotliStoreHuffmanTree(const uint8_t * depths,size_t num,HuffmanTree * tree,size_t * storage_ix,uint8_t * storage)298 void BrotliStoreHuffmanTree(const uint8_t* depths, size_t num,
299                             HuffmanTree* tree,
300                             size_t *storage_ix, uint8_t *storage) {
301   /* Write the Huffman tree into the brotli-representation.
302      The command alphabet is the largest, so this allocation will fit all
303      alphabets. */
304   uint8_t huffman_tree[BROTLI_NUM_COMMAND_SYMBOLS];
305   uint8_t huffman_tree_extra_bits[BROTLI_NUM_COMMAND_SYMBOLS];
306   size_t huffman_tree_size = 0;
307   uint8_t code_length_bitdepth[BROTLI_CODE_LENGTH_CODES] = { 0 };
308   uint16_t code_length_bitdepth_symbols[BROTLI_CODE_LENGTH_CODES];
309   uint32_t huffman_tree_histogram[BROTLI_CODE_LENGTH_CODES] = { 0 };
310   size_t i;
311   int num_codes = 0;
312   size_t code = 0;
313 
314   assert(num <= BROTLI_NUM_COMMAND_SYMBOLS);
315 
316   BrotliWriteHuffmanTree(depths, num, &huffman_tree_size, huffman_tree,
317                          huffman_tree_extra_bits);
318 
319   /* Calculate the statistics of the Huffman tree in brotli-representation. */
320   for (i = 0; i < huffman_tree_size; ++i) {
321     ++huffman_tree_histogram[huffman_tree[i]];
322   }
323 
324   for (i = 0; i < BROTLI_CODE_LENGTH_CODES; ++i) {
325     if (huffman_tree_histogram[i]) {
326       if (num_codes == 0) {
327         code = i;
328         num_codes = 1;
329       } else if (num_codes == 1) {
330         num_codes = 2;
331         break;
332       }
333     }
334   }
335 
336   /* Calculate another Huffman tree to use for compressing both the
337      earlier Huffman tree with. */
338   BrotliCreateHuffmanTree(huffman_tree_histogram, BROTLI_CODE_LENGTH_CODES,
339                           5, tree, code_length_bitdepth);
340   BrotliConvertBitDepthsToSymbols(code_length_bitdepth,
341                                   BROTLI_CODE_LENGTH_CODES,
342                                   code_length_bitdepth_symbols);
343 
344   /* Now, we have all the data, let's start storing it */
345   BrotliStoreHuffmanTreeOfHuffmanTreeToBitMask(num_codes, code_length_bitdepth,
346                                                storage_ix, storage);
347 
348   if (num_codes == 1) {
349     code_length_bitdepth[code] = 0;
350   }
351 
352   /* Store the real Huffman tree now. */
353   BrotliStoreHuffmanTreeToBitMask(huffman_tree_size,
354                                   huffman_tree,
355                                   huffman_tree_extra_bits,
356                                   code_length_bitdepth,
357                                   code_length_bitdepth_symbols,
358                                   storage_ix, storage);
359 }
360 
361 /* Builds a Huffman tree from histogram[0:length] into depth[0:length] and
362    bits[0:length] and stores the encoded tree to the bit stream. */
BuildAndStoreHuffmanTree(const uint32_t * histogram,const size_t length,HuffmanTree * tree,uint8_t * depth,uint16_t * bits,size_t * storage_ix,uint8_t * storage)363 static void BuildAndStoreHuffmanTree(const uint32_t *histogram,
364                                      const size_t length,
365                                      HuffmanTree* tree,
366                                      uint8_t* depth,
367                                      uint16_t* bits,
368                                      size_t* storage_ix,
369                                      uint8_t* storage) {
370   size_t count = 0;
371   size_t s4[4] = { 0 };
372   size_t i;
373   size_t max_bits = 0;
374   for (i = 0; i < length; i++) {
375     if (histogram[i]) {
376       if (count < 4) {
377         s4[count] = i;
378       } else if (count > 4) {
379         break;
380       }
381       count++;
382     }
383   }
384 
385   {
386     size_t max_bits_counter = length - 1;
387     while (max_bits_counter) {
388       max_bits_counter >>= 1;
389       ++max_bits;
390     }
391   }
392 
393   if (count <= 1) {
394     BrotliWriteBits(4, 1, storage_ix, storage);
395     BrotliWriteBits(max_bits, s4[0], storage_ix, storage);
396     depth[s4[0]] = 0;
397     bits[s4[0]] = 0;
398     return;
399   }
400 
401   memset(depth, 0, length * sizeof(depth[0]));
402   BrotliCreateHuffmanTree(histogram, length, 15, tree, depth);
403   BrotliConvertBitDepthsToSymbols(depth, length, bits);
404 
405   if (count <= 4) {
406     StoreSimpleHuffmanTree(depth, s4, count, max_bits, storage_ix, storage);
407   } else {
408     BrotliStoreHuffmanTree(depth, length, tree, storage_ix, storage);
409   }
410 }
411 
SortHuffmanTree(const HuffmanTree * v0,const HuffmanTree * v1)412 static BROTLI_INLINE BROTLI_BOOL SortHuffmanTree(
413     const HuffmanTree* v0, const HuffmanTree* v1) {
414   return TO_BROTLI_BOOL(v0->total_count_ < v1->total_count_);
415 }
416 
BrotliBuildAndStoreHuffmanTreeFast(MemoryManager * m,const uint32_t * histogram,const size_t histogram_total,const size_t max_bits,uint8_t * depth,uint16_t * bits,size_t * storage_ix,uint8_t * storage)417 void BrotliBuildAndStoreHuffmanTreeFast(MemoryManager* m,
418                                         const uint32_t* histogram,
419                                         const size_t histogram_total,
420                                         const size_t max_bits,
421                                         uint8_t* depth, uint16_t* bits,
422                                         size_t* storage_ix,
423                                         uint8_t* storage) {
424   size_t count = 0;
425   size_t symbols[4] = { 0 };
426   size_t length = 0;
427   size_t total = histogram_total;
428   while (total != 0) {
429     if (histogram[length]) {
430       if (count < 4) {
431         symbols[count] = length;
432       }
433       ++count;
434       total -= histogram[length];
435     }
436     ++length;
437   }
438 
439   if (count <= 1) {
440     BrotliWriteBits(4, 1, storage_ix, storage);
441     BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
442     depth[symbols[0]] = 0;
443     bits[symbols[0]] = 0;
444     return;
445   }
446 
447   memset(depth, 0, length * sizeof(depth[0]));
448   {
449     const size_t max_tree_size = 2 * length + 1;
450     HuffmanTree* tree = BROTLI_ALLOC(m, HuffmanTree, max_tree_size);
451     uint32_t count_limit;
452     if (BROTLI_IS_OOM(m)) return;
453     for (count_limit = 1; ; count_limit *= 2) {
454       HuffmanTree* node = tree;
455       size_t l;
456       for (l = length; l != 0;) {
457         --l;
458         if (histogram[l]) {
459           if (BROTLI_PREDICT_TRUE(histogram[l] >= count_limit)) {
460             InitHuffmanTree(node, histogram[l], -1, (int16_t)l);
461           } else {
462             InitHuffmanTree(node, count_limit, -1, (int16_t)l);
463           }
464           ++node;
465         }
466       }
467       {
468         const int n = (int)(node - tree);
469         HuffmanTree sentinel;
470         int i = 0;      /* Points to the next leaf node. */
471         int j = n + 1;  /* Points to the next non-leaf node. */
472         int k;
473 
474         SortHuffmanTreeItems(tree, (size_t)n, SortHuffmanTree);
475         /* The nodes are:
476            [0, n): the sorted leaf nodes that we start with.
477            [n]: we add a sentinel here.
478            [n + 1, 2n): new parent nodes are added here, starting from
479                         (n+1). These are naturally in ascending order.
480            [2n]: we add a sentinel at the end as well.
481            There will be (2n+1) elements at the end. */
482         InitHuffmanTree(&sentinel, BROTLI_UINT32_MAX, -1, -1);
483         *node++ = sentinel;
484         *node++ = sentinel;
485 
486         for (k = n - 1; k > 0; --k) {
487           int left, right;
488           if (tree[i].total_count_ <= tree[j].total_count_) {
489             left = i;
490             ++i;
491           } else {
492             left = j;
493             ++j;
494           }
495           if (tree[i].total_count_ <= tree[j].total_count_) {
496             right = i;
497             ++i;
498           } else {
499             right = j;
500             ++j;
501           }
502           /* The sentinel node becomes the parent node. */
503           node[-1].total_count_ =
504               tree[left].total_count_ + tree[right].total_count_;
505           node[-1].index_left_ = (int16_t)left;
506           node[-1].index_right_or_value_ = (int16_t)right;
507           /* Add back the last sentinel node. */
508           *node++ = sentinel;
509         }
510         if (BrotliSetDepth(2 * n - 1, tree, depth, 14)) {
511           /* We need to pack the Huffman tree in 14 bits. If this was not
512              successful, add fake entities to the lowest values and retry. */
513           break;
514         }
515       }
516     }
517     BROTLI_FREE(m, tree);
518   }
519   BrotliConvertBitDepthsToSymbols(depth, length, bits);
520   if (count <= 4) {
521     size_t i;
522     /* value of 1 indicates a simple Huffman code */
523     BrotliWriteBits(2, 1, storage_ix, storage);
524     BrotliWriteBits(2, count - 1, storage_ix, storage);  /* NSYM - 1 */
525 
526     /* Sort */
527     for (i = 0; i < count; i++) {
528       size_t j;
529       for (j = i + 1; j < count; j++) {
530         if (depth[symbols[j]] < depth[symbols[i]]) {
531           BROTLI_SWAP(size_t, symbols, j, i);
532         }
533       }
534     }
535 
536     if (count == 2) {
537       BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
538       BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
539     } else if (count == 3) {
540       BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
541       BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
542       BrotliWriteBits(max_bits, symbols[2], storage_ix, storage);
543     } else {
544       BrotliWriteBits(max_bits, symbols[0], storage_ix, storage);
545       BrotliWriteBits(max_bits, symbols[1], storage_ix, storage);
546       BrotliWriteBits(max_bits, symbols[2], storage_ix, storage);
547       BrotliWriteBits(max_bits, symbols[3], storage_ix, storage);
548       /* tree-select */
549       BrotliWriteBits(1, depth[symbols[0]] == 1 ? 1 : 0, storage_ix, storage);
550     }
551   } else {
552     uint8_t previous_value = 8;
553     size_t i;
554     /* Complex Huffman Tree */
555     StoreStaticCodeLengthCode(storage_ix, storage);
556 
557     /* Actual RLE coding. */
558     for (i = 0; i < length;) {
559       const uint8_t value = depth[i];
560       size_t reps = 1;
561       size_t k;
562       for (k = i + 1; k < length && depth[k] == value; ++k) {
563         ++reps;
564       }
565       i += reps;
566       if (value == 0) {
567         BrotliWriteBits(kZeroRepsDepth[reps], kZeroRepsBits[reps],
568                         storage_ix, storage);
569       } else {
570         if (previous_value != value) {
571           BrotliWriteBits(kCodeLengthDepth[value], kCodeLengthBits[value],
572                           storage_ix, storage);
573           --reps;
574         }
575         if (reps < 3) {
576           while (reps != 0) {
577             reps--;
578             BrotliWriteBits(kCodeLengthDepth[value], kCodeLengthBits[value],
579                             storage_ix, storage);
580           }
581         } else {
582           reps -= 3;
583           BrotliWriteBits(kNonZeroRepsDepth[reps], kNonZeroRepsBits[reps],
584                           storage_ix, storage);
585         }
586         previous_value = value;
587       }
588     }
589   }
590 }
591 
IndexOf(const uint8_t * v,size_t v_size,uint8_t value)592 static size_t IndexOf(const uint8_t* v, size_t v_size, uint8_t value) {
593   size_t i = 0;
594   for (; i < v_size; ++i) {
595     if (v[i] == value) return i;
596   }
597   return i;
598 }
599 
MoveToFront(uint8_t * v,size_t index)600 static void MoveToFront(uint8_t* v, size_t index) {
601   uint8_t value = v[index];
602   size_t i;
603   for (i = index; i != 0; --i) {
604     v[i] = v[i - 1];
605   }
606   v[0] = value;
607 }
608 
MoveToFrontTransform(const uint32_t * BROTLI_RESTRICT v_in,const size_t v_size,uint32_t * v_out)609 static void MoveToFrontTransform(const uint32_t* BROTLI_RESTRICT v_in,
610                                  const size_t v_size,
611                                  uint32_t* v_out) {
612   size_t i;
613   uint8_t mtf[256];
614   uint32_t max_value;
615   if (v_size == 0) {
616     return;
617   }
618   max_value = v_in[0];
619   for (i = 1; i < v_size; ++i) {
620     if (v_in[i] > max_value) max_value = v_in[i];
621   }
622   assert(max_value < 256u);
623   for (i = 0; i <= max_value; ++i) {
624     mtf[i] = (uint8_t)i;
625   }
626   {
627     size_t mtf_size = max_value + 1;
628     for (i = 0; i < v_size; ++i) {
629       size_t index = IndexOf(mtf, mtf_size, (uint8_t)v_in[i]);
630       assert(index < mtf_size);
631       v_out[i] = (uint32_t)index;
632       MoveToFront(mtf, index);
633     }
634   }
635 }
636 
637 /* Finds runs of zeros in v[0..in_size) and replaces them with a prefix code of
638    the run length plus extra bits (lower 9 bits is the prefix code and the rest
639    are the extra bits). Non-zero values in v[] are shifted by
640    *max_length_prefix. Will not create prefix codes bigger than the initial
641    value of *max_run_length_prefix. The prefix code of run length L is simply
642    Log2Floor(L) and the number of extra bits is the same as the prefix code. */
RunLengthCodeZeros(const size_t in_size,uint32_t * BROTLI_RESTRICT v,size_t * BROTLI_RESTRICT out_size,uint32_t * BROTLI_RESTRICT max_run_length_prefix)643 static void RunLengthCodeZeros(const size_t in_size,
644     uint32_t* BROTLI_RESTRICT v, size_t* BROTLI_RESTRICT out_size,
645     uint32_t* BROTLI_RESTRICT max_run_length_prefix) {
646   uint32_t max_reps = 0;
647   size_t i;
648   uint32_t max_prefix;
649   for (i = 0; i < in_size;) {
650     uint32_t reps = 0;
651     for (; i < in_size && v[i] != 0; ++i) ;
652     for (; i < in_size && v[i] == 0; ++i) {
653       ++reps;
654     }
655     max_reps = BROTLI_MAX(uint32_t, reps, max_reps);
656   }
657   max_prefix = max_reps > 0 ? Log2FloorNonZero(max_reps) : 0;
658   max_prefix = BROTLI_MIN(uint32_t, max_prefix, *max_run_length_prefix);
659   *max_run_length_prefix = max_prefix;
660   *out_size = 0;
661   for (i = 0; i < in_size;) {
662     assert(*out_size <= i);
663     if (v[i] != 0) {
664       v[*out_size] = v[i] + *max_run_length_prefix;
665       ++i;
666       ++(*out_size);
667     } else {
668       uint32_t reps = 1;
669       size_t k;
670       for (k = i + 1; k < in_size && v[k] == 0; ++k) {
671         ++reps;
672       }
673       i += reps;
674       while (reps != 0) {
675         if (reps < (2u << max_prefix)) {
676           uint32_t run_length_prefix = Log2FloorNonZero(reps);
677           const uint32_t extra_bits = reps - (1u << run_length_prefix);
678           v[*out_size] = run_length_prefix + (extra_bits << 9);
679           ++(*out_size);
680           break;
681         } else {
682           const uint32_t extra_bits = (1u << max_prefix) - 1u;
683           v[*out_size] = max_prefix + (extra_bits << 9);
684           reps -= (2u << max_prefix) - 1u;
685           ++(*out_size);
686         }
687       }
688     }
689   }
690 }
691 
692 #define SYMBOL_BITS 9
693 
EncodeContextMap(MemoryManager * m,const uint32_t * context_map,size_t context_map_size,size_t num_clusters,HuffmanTree * tree,size_t * storage_ix,uint8_t * storage)694 static void EncodeContextMap(MemoryManager* m,
695                              const uint32_t* context_map,
696                              size_t context_map_size,
697                              size_t num_clusters,
698                              HuffmanTree* tree,
699                              size_t* storage_ix, uint8_t* storage) {
700   size_t i;
701   uint32_t* rle_symbols;
702   uint32_t max_run_length_prefix = 6;
703   size_t num_rle_symbols = 0;
704   uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
705   static const uint32_t kSymbolMask = (1u << SYMBOL_BITS) - 1u;
706   uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
707   uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
708 
709   StoreVarLenUint8(num_clusters - 1, storage_ix, storage);
710 
711   if (num_clusters == 1) {
712     return;
713   }
714 
715   rle_symbols = BROTLI_ALLOC(m, uint32_t, context_map_size);
716   if (BROTLI_IS_OOM(m)) return;
717   MoveToFrontTransform(context_map, context_map_size, rle_symbols);
718   RunLengthCodeZeros(context_map_size, rle_symbols,
719                      &num_rle_symbols, &max_run_length_prefix);
720   memset(histogram, 0, sizeof(histogram));
721   for (i = 0; i < num_rle_symbols; ++i) {
722     ++histogram[rle_symbols[i] & kSymbolMask];
723   }
724   {
725     BROTLI_BOOL use_rle = TO_BROTLI_BOOL(max_run_length_prefix > 0);
726     BrotliWriteBits(1, (uint64_t)use_rle, storage_ix, storage);
727     if (use_rle) {
728       BrotliWriteBits(4, max_run_length_prefix - 1, storage_ix, storage);
729     }
730   }
731   BuildAndStoreHuffmanTree(histogram, num_clusters + max_run_length_prefix,
732                            tree, depths, bits, storage_ix, storage);
733   for (i = 0; i < num_rle_symbols; ++i) {
734     const uint32_t rle_symbol = rle_symbols[i] & kSymbolMask;
735     const uint32_t extra_bits_val = rle_symbols[i] >> SYMBOL_BITS;
736     BrotliWriteBits(depths[rle_symbol], bits[rle_symbol], storage_ix, storage);
737     if (rle_symbol > 0 && rle_symbol <= max_run_length_prefix) {
738       BrotliWriteBits(rle_symbol, extra_bits_val, storage_ix, storage);
739     }
740   }
741   BrotliWriteBits(1, 1, storage_ix, storage);  /* use move-to-front */
742   BROTLI_FREE(m, rle_symbols);
743 }
744 
745 /* Stores the block switch command with index block_ix to the bit stream. */
StoreBlockSwitch(BlockSplitCode * code,const uint32_t block_len,const uint8_t block_type,BROTLI_BOOL is_first_block,size_t * storage_ix,uint8_t * storage)746 static BROTLI_INLINE void StoreBlockSwitch(BlockSplitCode* code,
747                                            const uint32_t block_len,
748                                            const uint8_t block_type,
749                                            BROTLI_BOOL is_first_block,
750                                            size_t* storage_ix,
751                                            uint8_t* storage) {
752   size_t typecode = NextBlockTypeCode(&code->type_code_calculator, block_type);
753   size_t lencode;
754   uint32_t len_nextra;
755   uint32_t len_extra;
756   if (!is_first_block) {
757     BrotliWriteBits(code->type_depths[typecode], code->type_bits[typecode],
758                     storage_ix, storage);
759   }
760   GetBlockLengthPrefixCode(block_len, &lencode, &len_nextra, &len_extra);
761 
762   BrotliWriteBits(code->length_depths[lencode], code->length_bits[lencode],
763                   storage_ix, storage);
764   BrotliWriteBits(len_nextra, len_extra, storage_ix, storage);
765 }
766 
767 /* Builds a BlockSplitCode data structure from the block split given by the
768    vector of block types and block lengths and stores it to the bit stream. */
BuildAndStoreBlockSplitCode(const uint8_t * types,const uint32_t * lengths,const size_t num_blocks,const size_t num_types,HuffmanTree * tree,BlockSplitCode * code,size_t * storage_ix,uint8_t * storage)769 static void BuildAndStoreBlockSplitCode(const uint8_t* types,
770                                         const uint32_t* lengths,
771                                         const size_t num_blocks,
772                                         const size_t num_types,
773                                         HuffmanTree* tree,
774                                         BlockSplitCode* code,
775                                         size_t* storage_ix,
776                                         uint8_t* storage) {
777   uint32_t type_histo[BROTLI_MAX_BLOCK_TYPE_SYMBOLS];
778   uint32_t length_histo[BROTLI_NUM_BLOCK_LEN_SYMBOLS];
779   size_t i;
780   BlockTypeCodeCalculator type_code_calculator;
781   memset(type_histo, 0, (num_types + 2) * sizeof(type_histo[0]));
782   memset(length_histo, 0, sizeof(length_histo));
783   InitBlockTypeCodeCalculator(&type_code_calculator);
784   for (i = 0; i < num_blocks; ++i) {
785     size_t type_code = NextBlockTypeCode(&type_code_calculator, types[i]);
786     if (i != 0) ++type_histo[type_code];
787     ++length_histo[BlockLengthPrefixCode(lengths[i])];
788   }
789   StoreVarLenUint8(num_types - 1, storage_ix, storage);
790   if (num_types > 1) {  /* TODO: else? could StoreBlockSwitch occur? */
791     BuildAndStoreHuffmanTree(&type_histo[0], num_types + 2, tree,
792                              &code->type_depths[0], &code->type_bits[0],
793                              storage_ix, storage);
794     BuildAndStoreHuffmanTree(&length_histo[0], BROTLI_NUM_BLOCK_LEN_SYMBOLS,
795                              tree, &code->length_depths[0],
796                              &code->length_bits[0], storage_ix, storage);
797     StoreBlockSwitch(code, lengths[0], types[0], 1, storage_ix, storage);
798   }
799 }
800 
801 /* Stores a context map where the histogram type is always the block type. */
StoreTrivialContextMap(size_t num_types,size_t context_bits,HuffmanTree * tree,size_t * storage_ix,uint8_t * storage)802 static void StoreTrivialContextMap(size_t num_types,
803                                    size_t context_bits,
804                                    HuffmanTree* tree,
805                                    size_t* storage_ix,
806                                    uint8_t* storage) {
807   StoreVarLenUint8(num_types - 1, storage_ix, storage);
808   if (num_types > 1) {
809     size_t repeat_code = context_bits - 1u;
810     size_t repeat_bits = (1u << repeat_code) - 1u;
811     size_t alphabet_size = num_types + repeat_code;
812     uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
813     uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
814     uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
815     size_t i;
816     memset(histogram, 0, alphabet_size * sizeof(histogram[0]));
817     /* Write RLEMAX. */
818     BrotliWriteBits(1, 1, storage_ix, storage);
819     BrotliWriteBits(4, repeat_code - 1, storage_ix, storage);
820     histogram[repeat_code] = (uint32_t)num_types;
821     histogram[0] = 1;
822     for (i = context_bits; i < alphabet_size; ++i) {
823       histogram[i] = 1;
824     }
825     BuildAndStoreHuffmanTree(histogram, alphabet_size, tree,
826                              depths, bits, storage_ix, storage);
827     for (i = 0; i < num_types; ++i) {
828       size_t code = (i == 0 ? 0 : i + context_bits - 1);
829       BrotliWriteBits(depths[code], bits[code], storage_ix, storage);
830       BrotliWriteBits(
831           depths[repeat_code], bits[repeat_code], storage_ix, storage);
832       BrotliWriteBits(repeat_code, repeat_bits, storage_ix, storage);
833     }
834     /* Write IMTF (inverse-move-to-front) bit. */
835     BrotliWriteBits(1, 1, storage_ix, storage);
836   }
837 }
838 
839 /* Manages the encoding of one block category (literal, command or distance). */
840 typedef struct BlockEncoder {
841   size_t alphabet_size_;
842   size_t num_block_types_;
843   const uint8_t* block_types_;  /* Not owned. */
844   const uint32_t* block_lengths_;  /* Not owned. */
845   size_t num_blocks_;
846   BlockSplitCode block_split_code_;
847   size_t block_ix_;
848   size_t block_len_;
849   size_t entropy_ix_;
850   uint8_t* depths_;
851   uint16_t* bits_;
852 } BlockEncoder;
853 
InitBlockEncoder(BlockEncoder * self,size_t alphabet_size,size_t num_block_types,const uint8_t * block_types,const uint32_t * block_lengths,const size_t num_blocks)854 static void InitBlockEncoder(BlockEncoder* self, size_t alphabet_size,
855     size_t num_block_types, const uint8_t* block_types,
856     const uint32_t* block_lengths, const size_t num_blocks) {
857   self->alphabet_size_ = alphabet_size;
858   self->num_block_types_ = num_block_types;
859   self->block_types_ = block_types;
860   self->block_lengths_ = block_lengths;
861   self->num_blocks_ = num_blocks;
862   InitBlockTypeCodeCalculator(&self->block_split_code_.type_code_calculator);
863   self->block_ix_ = 0;
864   self->block_len_ = num_blocks == 0 ? 0 : block_lengths[0];
865   self->entropy_ix_ = 0;
866   self->depths_ = 0;
867   self->bits_ = 0;
868 }
869 
CleanupBlockEncoder(MemoryManager * m,BlockEncoder * self)870 static void CleanupBlockEncoder(MemoryManager* m, BlockEncoder* self) {
871   BROTLI_FREE(m, self->depths_);
872   BROTLI_FREE(m, self->bits_);
873 }
874 
875 /* Creates entropy codes of block lengths and block types and stores them
876    to the bit stream. */
BuildAndStoreBlockSwitchEntropyCodes(BlockEncoder * self,HuffmanTree * tree,size_t * storage_ix,uint8_t * storage)877 static void BuildAndStoreBlockSwitchEntropyCodes(BlockEncoder* self,
878     HuffmanTree* tree, size_t* storage_ix, uint8_t* storage) {
879   BuildAndStoreBlockSplitCode(self->block_types_, self->block_lengths_,
880       self->num_blocks_, self->num_block_types_, tree, &self->block_split_code_,
881       storage_ix, storage);
882 }
883 
884 /* Stores the next symbol with the entropy code of the current block type.
885    Updates the block type and block length at block boundaries. */
StoreSymbol(BlockEncoder * self,size_t symbol,size_t * storage_ix,uint8_t * storage)886 static void StoreSymbol(BlockEncoder* self, size_t symbol, size_t* storage_ix,
887     uint8_t* storage) {
888   if (self->block_len_ == 0) {
889     size_t block_ix = ++self->block_ix_;
890     uint32_t block_len = self->block_lengths_[block_ix];
891     uint8_t block_type = self->block_types_[block_ix];
892     self->block_len_ = block_len;
893     self->entropy_ix_ = block_type * self->alphabet_size_;
894     StoreBlockSwitch(&self->block_split_code_, block_len, block_type, 0,
895         storage_ix, storage);
896   }
897   --self->block_len_;
898   {
899     size_t ix = self->entropy_ix_ + symbol;
900     BrotliWriteBits(self->depths_[ix], self->bits_[ix], storage_ix, storage);
901   }
902 }
903 
904 /* Stores the next symbol with the entropy code of the current block type and
905    context value.
906    Updates the block type and block length at block boundaries. */
StoreSymbolWithContext(BlockEncoder * self,size_t symbol,size_t context,const uint32_t * context_map,size_t * storage_ix,uint8_t * storage,const size_t context_bits)907 static void StoreSymbolWithContext(BlockEncoder* self, size_t symbol,
908     size_t context, const uint32_t* context_map, size_t* storage_ix,
909     uint8_t* storage, const size_t context_bits) {
910   if (self->block_len_ == 0) {
911     size_t block_ix = ++self->block_ix_;
912     uint32_t block_len = self->block_lengths_[block_ix];
913     uint8_t block_type = self->block_types_[block_ix];
914     self->block_len_ = block_len;
915     self->entropy_ix_ = (size_t)block_type << context_bits;
916     StoreBlockSwitch(&self->block_split_code_, block_len, block_type, 0,
917         storage_ix, storage);
918   }
919   --self->block_len_;
920   {
921     size_t histo_ix = context_map[self->entropy_ix_ + context];
922     size_t ix = histo_ix * self->alphabet_size_ + symbol;
923     BrotliWriteBits(self->depths_[ix], self->bits_[ix], storage_ix, storage);
924   }
925 }
926 
927 #define FN(X) X ## Literal
928 /* NOLINTNEXTLINE(build/include) */
929 #include "./block_encoder_inc.h"
930 #undef FN
931 
932 #define FN(X) X ## Command
933 /* NOLINTNEXTLINE(build/include) */
934 #include "./block_encoder_inc.h"
935 #undef FN
936 
937 #define FN(X) X ## Distance
938 /* NOLINTNEXTLINE(build/include) */
939 #include "./block_encoder_inc.h"
940 #undef FN
941 
JumpToByteBoundary(size_t * storage_ix,uint8_t * storage)942 static void JumpToByteBoundary(size_t* storage_ix, uint8_t* storage) {
943   *storage_ix = (*storage_ix + 7u) & ~7u;
944   storage[*storage_ix >> 3] = 0;
945 }
946 
BrotliStoreMetaBlock(MemoryManager * m,const uint8_t * input,size_t start_pos,size_t length,size_t mask,uint8_t prev_byte,uint8_t prev_byte2,BROTLI_BOOL is_last,uint32_t num_direct_distance_codes,uint32_t distance_postfix_bits,ContextType literal_context_mode,const Command * commands,size_t n_commands,const MetaBlockSplit * mb,size_t * storage_ix,uint8_t * storage)947 void BrotliStoreMetaBlock(MemoryManager* m,
948                           const uint8_t* input,
949                           size_t start_pos,
950                           size_t length,
951                           size_t mask,
952                           uint8_t prev_byte,
953                           uint8_t prev_byte2,
954                           BROTLI_BOOL is_last,
955                           uint32_t num_direct_distance_codes,
956                           uint32_t distance_postfix_bits,
957                           ContextType literal_context_mode,
958                           const Command *commands,
959                           size_t n_commands,
960                           const MetaBlockSplit* mb,
961                           size_t *storage_ix,
962                           uint8_t *storage) {
963   size_t pos = start_pos;
964   size_t i;
965   size_t num_distance_codes =
966       BROTLI_NUM_DISTANCE_SHORT_CODES + num_direct_distance_codes +
967       (48u << distance_postfix_bits);
968   HuffmanTree* tree;
969   BlockEncoder literal_enc;
970   BlockEncoder command_enc;
971   BlockEncoder distance_enc;
972 
973   StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
974 
975   tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE);
976   if (BROTLI_IS_OOM(m)) return;
977   InitBlockEncoder(&literal_enc, 256, mb->literal_split.num_types,
978       mb->literal_split.types, mb->literal_split.lengths,
979       mb->literal_split.num_blocks);
980   InitBlockEncoder(&command_enc, BROTLI_NUM_COMMAND_SYMBOLS,
981       mb->command_split.num_types, mb->command_split.types,
982       mb->command_split.lengths, mb->command_split.num_blocks);
983   InitBlockEncoder(&distance_enc, num_distance_codes,
984       mb->distance_split.num_types, mb->distance_split.types,
985       mb->distance_split.lengths, mb->distance_split.num_blocks);
986 
987   BuildAndStoreBlockSwitchEntropyCodes(&literal_enc, tree, storage_ix, storage);
988   BuildAndStoreBlockSwitchEntropyCodes(&command_enc, tree, storage_ix, storage);
989   BuildAndStoreBlockSwitchEntropyCodes(
990       &distance_enc, tree, storage_ix, storage);
991 
992   BrotliWriteBits(2, distance_postfix_bits, storage_ix, storage);
993   BrotliWriteBits(4, num_direct_distance_codes >> distance_postfix_bits,
994                   storage_ix, storage);
995   for (i = 0; i < mb->literal_split.num_types; ++i) {
996     BrotliWriteBits(2, literal_context_mode, storage_ix, storage);
997   }
998 
999   if (mb->literal_context_map_size == 0) {
1000     StoreTrivialContextMap(mb->literal_histograms_size,
1001         BROTLI_LITERAL_CONTEXT_BITS, tree, storage_ix, storage);
1002   } else {
1003     EncodeContextMap(m,
1004         mb->literal_context_map, mb->literal_context_map_size,
1005         mb->literal_histograms_size, tree, storage_ix, storage);
1006     if (BROTLI_IS_OOM(m)) return;
1007   }
1008 
1009   if (mb->distance_context_map_size == 0) {
1010     StoreTrivialContextMap(mb->distance_histograms_size,
1011         BROTLI_DISTANCE_CONTEXT_BITS, tree, storage_ix, storage);
1012   } else {
1013     EncodeContextMap(m,
1014         mb->distance_context_map, mb->distance_context_map_size,
1015         mb->distance_histograms_size, tree, storage_ix, storage);
1016     if (BROTLI_IS_OOM(m)) return;
1017   }
1018 
1019   BuildAndStoreEntropyCodesLiteral(m, &literal_enc, mb->literal_histograms,
1020       mb->literal_histograms_size, tree, storage_ix, storage);
1021   if (BROTLI_IS_OOM(m)) return;
1022   BuildAndStoreEntropyCodesCommand(m, &command_enc, mb->command_histograms,
1023       mb->command_histograms_size, tree, storage_ix, storage);
1024   if (BROTLI_IS_OOM(m)) return;
1025   BuildAndStoreEntropyCodesDistance(m, &distance_enc, mb->distance_histograms,
1026       mb->distance_histograms_size, tree, storage_ix, storage);
1027   if (BROTLI_IS_OOM(m)) return;
1028   BROTLI_FREE(m, tree);
1029 
1030   for (i = 0; i < n_commands; ++i) {
1031     const Command cmd = commands[i];
1032     size_t cmd_code = cmd.cmd_prefix_;
1033     StoreSymbol(&command_enc, cmd_code, storage_ix, storage);
1034     StoreCommandExtra(&cmd, storage_ix, storage);
1035     if (mb->literal_context_map_size == 0) {
1036       size_t j;
1037       for (j = cmd.insert_len_; j != 0; --j) {
1038         StoreSymbol(&literal_enc, input[pos & mask], storage_ix, storage);
1039         ++pos;
1040       }
1041     } else {
1042       size_t j;
1043       for (j = cmd.insert_len_; j != 0; --j) {
1044         size_t context = Context(prev_byte, prev_byte2, literal_context_mode);
1045         uint8_t literal = input[pos & mask];
1046         StoreSymbolWithContext(&literal_enc, literal, context,
1047             mb->literal_context_map, storage_ix, storage,
1048             BROTLI_LITERAL_CONTEXT_BITS);
1049         prev_byte2 = prev_byte;
1050         prev_byte = literal;
1051         ++pos;
1052       }
1053     }
1054     pos += CommandCopyLen(&cmd);
1055     if (CommandCopyLen(&cmd)) {
1056       prev_byte2 = input[(pos - 2) & mask];
1057       prev_byte = input[(pos - 1) & mask];
1058       if (cmd.cmd_prefix_ >= 128) {
1059         size_t dist_code = cmd.dist_prefix_;
1060         uint32_t distnumextra = cmd.dist_extra_ >> 24;
1061         uint64_t distextra = cmd.dist_extra_ & 0xffffff;
1062         if (mb->distance_context_map_size == 0) {
1063           StoreSymbol(&distance_enc, dist_code, storage_ix, storage);
1064         } else {
1065           size_t context = CommandDistanceContext(&cmd);
1066           StoreSymbolWithContext(&distance_enc, dist_code, context,
1067               mb->distance_context_map, storage_ix, storage,
1068               BROTLI_DISTANCE_CONTEXT_BITS);
1069         }
1070         BrotliWriteBits(distnumextra, distextra, storage_ix, storage);
1071       }
1072     }
1073   }
1074   CleanupBlockEncoder(m, &distance_enc);
1075   CleanupBlockEncoder(m, &command_enc);
1076   CleanupBlockEncoder(m, &literal_enc);
1077   if (is_last) {
1078     JumpToByteBoundary(storage_ix, storage);
1079   }
1080 }
1081 
BuildHistograms(const uint8_t * input,size_t start_pos,size_t mask,const Command * commands,size_t n_commands,HistogramLiteral * lit_histo,HistogramCommand * cmd_histo,HistogramDistance * dist_histo)1082 static void BuildHistograms(const uint8_t* input,
1083                             size_t start_pos,
1084                             size_t mask,
1085                             const Command *commands,
1086                             size_t n_commands,
1087                             HistogramLiteral* lit_histo,
1088                             HistogramCommand* cmd_histo,
1089                             HistogramDistance* dist_histo) {
1090   size_t pos = start_pos;
1091   size_t i;
1092   for (i = 0; i < n_commands; ++i) {
1093     const Command cmd = commands[i];
1094     size_t j;
1095     HistogramAddCommand(cmd_histo, cmd.cmd_prefix_);
1096     for (j = cmd.insert_len_; j != 0; --j) {
1097       HistogramAddLiteral(lit_histo, input[pos & mask]);
1098       ++pos;
1099     }
1100     pos += CommandCopyLen(&cmd);
1101     if (CommandCopyLen(&cmd) && cmd.cmd_prefix_ >= 128) {
1102       HistogramAddDistance(dist_histo, cmd.dist_prefix_);
1103     }
1104   }
1105 }
1106 
StoreDataWithHuffmanCodes(const uint8_t * input,size_t start_pos,size_t mask,const Command * commands,size_t n_commands,const uint8_t * lit_depth,const uint16_t * lit_bits,const uint8_t * cmd_depth,const uint16_t * cmd_bits,const uint8_t * dist_depth,const uint16_t * dist_bits,size_t * storage_ix,uint8_t * storage)1107 static void StoreDataWithHuffmanCodes(const uint8_t* input,
1108                                       size_t start_pos,
1109                                       size_t mask,
1110                                       const Command *commands,
1111                                       size_t n_commands,
1112                                       const uint8_t* lit_depth,
1113                                       const uint16_t* lit_bits,
1114                                       const uint8_t* cmd_depth,
1115                                       const uint16_t* cmd_bits,
1116                                       const uint8_t* dist_depth,
1117                                       const uint16_t* dist_bits,
1118                                       size_t* storage_ix,
1119                                       uint8_t* storage) {
1120   size_t pos = start_pos;
1121   size_t i;
1122   for (i = 0; i < n_commands; ++i) {
1123     const Command cmd = commands[i];
1124     const size_t cmd_code = cmd.cmd_prefix_;
1125     size_t j;
1126     BrotliWriteBits(
1127         cmd_depth[cmd_code], cmd_bits[cmd_code], storage_ix, storage);
1128     StoreCommandExtra(&cmd, storage_ix, storage);
1129     for (j = cmd.insert_len_; j != 0; --j) {
1130       const uint8_t literal = input[pos & mask];
1131       BrotliWriteBits(
1132           lit_depth[literal], lit_bits[literal], storage_ix, storage);
1133       ++pos;
1134     }
1135     pos += CommandCopyLen(&cmd);
1136     if (CommandCopyLen(&cmd) && cmd.cmd_prefix_ >= 128) {
1137       const size_t dist_code = cmd.dist_prefix_;
1138       const uint32_t distnumextra = cmd.dist_extra_ >> 24;
1139       const uint32_t distextra = cmd.dist_extra_ & 0xffffff;
1140       BrotliWriteBits(dist_depth[dist_code], dist_bits[dist_code],
1141                       storage_ix, storage);
1142       BrotliWriteBits(distnumextra, distextra, storage_ix, storage);
1143     }
1144   }
1145 }
1146 
BrotliStoreMetaBlockTrivial(MemoryManager * m,const uint8_t * input,size_t start_pos,size_t length,size_t mask,BROTLI_BOOL is_last,const Command * commands,size_t n_commands,size_t * storage_ix,uint8_t * storage)1147 void BrotliStoreMetaBlockTrivial(MemoryManager* m,
1148                                  const uint8_t* input,
1149                                  size_t start_pos,
1150                                  size_t length,
1151                                  size_t mask,
1152                                  BROTLI_BOOL is_last,
1153                                  const Command *commands,
1154                                  size_t n_commands,
1155                                  size_t *storage_ix,
1156                                  uint8_t *storage) {
1157   HistogramLiteral lit_histo;
1158   HistogramCommand cmd_histo;
1159   HistogramDistance dist_histo;
1160   uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
1161   uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
1162   uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS];
1163   uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS];
1164   uint8_t dist_depth[SIMPLE_DISTANCE_ALPHABET_SIZE];
1165   uint16_t dist_bits[SIMPLE_DISTANCE_ALPHABET_SIZE];
1166   HuffmanTree* tree;
1167 
1168   StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
1169 
1170   HistogramClearLiteral(&lit_histo);
1171   HistogramClearCommand(&cmd_histo);
1172   HistogramClearDistance(&dist_histo);
1173 
1174   BuildHistograms(input, start_pos, mask, commands, n_commands,
1175                   &lit_histo, &cmd_histo, &dist_histo);
1176 
1177   BrotliWriteBits(13, 0, storage_ix, storage);
1178 
1179   tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE);
1180   if (BROTLI_IS_OOM(m)) return;
1181   BuildAndStoreHuffmanTree(lit_histo.data_, BROTLI_NUM_LITERAL_SYMBOLS, tree,
1182                            lit_depth, lit_bits,
1183                            storage_ix, storage);
1184   BuildAndStoreHuffmanTree(cmd_histo.data_, BROTLI_NUM_COMMAND_SYMBOLS, tree,
1185                            cmd_depth, cmd_bits,
1186                            storage_ix, storage);
1187   BuildAndStoreHuffmanTree(dist_histo.data_, SIMPLE_DISTANCE_ALPHABET_SIZE,
1188                            tree,
1189                            dist_depth, dist_bits,
1190                            storage_ix, storage);
1191   BROTLI_FREE(m, tree);
1192   StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
1193                             n_commands, lit_depth, lit_bits,
1194                             cmd_depth, cmd_bits,
1195                             dist_depth, dist_bits,
1196                             storage_ix, storage);
1197   if (is_last) {
1198     JumpToByteBoundary(storage_ix, storage);
1199   }
1200 }
1201 
BrotliStoreMetaBlockFast(MemoryManager * m,const uint8_t * input,size_t start_pos,size_t length,size_t mask,BROTLI_BOOL is_last,const Command * commands,size_t n_commands,size_t * storage_ix,uint8_t * storage)1202 void BrotliStoreMetaBlockFast(MemoryManager* m,
1203                               const uint8_t* input,
1204                               size_t start_pos,
1205                               size_t length,
1206                               size_t mask,
1207                               BROTLI_BOOL is_last,
1208                               const Command *commands,
1209                               size_t n_commands,
1210                               size_t *storage_ix,
1211                               uint8_t *storage) {
1212   StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
1213 
1214   BrotliWriteBits(13, 0, storage_ix, storage);
1215 
1216   if (n_commands <= 128) {
1217     uint32_t histogram[BROTLI_NUM_LITERAL_SYMBOLS] = { 0 };
1218     size_t pos = start_pos;
1219     size_t num_literals = 0;
1220     size_t i;
1221     uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
1222     uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
1223     for (i = 0; i < n_commands; ++i) {
1224       const Command cmd = commands[i];
1225       size_t j;
1226       for (j = cmd.insert_len_; j != 0; --j) {
1227         ++histogram[input[pos & mask]];
1228         ++pos;
1229       }
1230       num_literals += cmd.insert_len_;
1231       pos += CommandCopyLen(&cmd);
1232     }
1233     BrotliBuildAndStoreHuffmanTreeFast(m, histogram, num_literals,
1234                                        /* max_bits = */ 8,
1235                                        lit_depth, lit_bits,
1236                                        storage_ix, storage);
1237     if (BROTLI_IS_OOM(m)) return;
1238     StoreStaticCommandHuffmanTree(storage_ix, storage);
1239     StoreStaticDistanceHuffmanTree(storage_ix, storage);
1240     StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
1241                               n_commands, lit_depth, lit_bits,
1242                               kStaticCommandCodeDepth,
1243                               kStaticCommandCodeBits,
1244                               kStaticDistanceCodeDepth,
1245                               kStaticDistanceCodeBits,
1246                               storage_ix, storage);
1247   } else {
1248     HistogramLiteral lit_histo;
1249     HistogramCommand cmd_histo;
1250     HistogramDistance dist_histo;
1251     uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
1252     uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
1253     uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS];
1254     uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS];
1255     uint8_t dist_depth[SIMPLE_DISTANCE_ALPHABET_SIZE];
1256     uint16_t dist_bits[SIMPLE_DISTANCE_ALPHABET_SIZE];
1257     HistogramClearLiteral(&lit_histo);
1258     HistogramClearCommand(&cmd_histo);
1259     HistogramClearDistance(&dist_histo);
1260     BuildHistograms(input, start_pos, mask, commands, n_commands,
1261                     &lit_histo, &cmd_histo, &dist_histo);
1262     BrotliBuildAndStoreHuffmanTreeFast(m, lit_histo.data_,
1263                                        lit_histo.total_count_,
1264                                        /* max_bits = */ 8,
1265                                        lit_depth, lit_bits,
1266                                        storage_ix, storage);
1267     if (BROTLI_IS_OOM(m)) return;
1268     BrotliBuildAndStoreHuffmanTreeFast(m, cmd_histo.data_,
1269                                        cmd_histo.total_count_,
1270                                        /* max_bits = */ 10,
1271                                        cmd_depth, cmd_bits,
1272                                        storage_ix, storage);
1273     if (BROTLI_IS_OOM(m)) return;
1274     BrotliBuildAndStoreHuffmanTreeFast(m, dist_histo.data_,
1275                                        dist_histo.total_count_,
1276                                        /* max_bits = */
1277                                        SIMPLE_DISTANCE_ALPHABET_BITS,
1278                                        dist_depth, dist_bits,
1279                                        storage_ix, storage);
1280     if (BROTLI_IS_OOM(m)) return;
1281     StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
1282                               n_commands, lit_depth, lit_bits,
1283                               cmd_depth, cmd_bits,
1284                               dist_depth, dist_bits,
1285                               storage_ix, storage);
1286   }
1287 
1288   if (is_last) {
1289     JumpToByteBoundary(storage_ix, storage);
1290   }
1291 }
1292 
1293 /* This is for storing uncompressed blocks (simple raw storage of
1294    bytes-as-bytes). */
BrotliStoreUncompressedMetaBlock(BROTLI_BOOL is_final_block,const uint8_t * BROTLI_RESTRICT input,size_t position,size_t mask,size_t len,size_t * BROTLI_RESTRICT storage_ix,uint8_t * BROTLI_RESTRICT storage)1295 void BrotliStoreUncompressedMetaBlock(BROTLI_BOOL is_final_block,
1296                                       const uint8_t * BROTLI_RESTRICT input,
1297                                       size_t position, size_t mask,
1298                                       size_t len,
1299                                       size_t * BROTLI_RESTRICT storage_ix,
1300                                       uint8_t * BROTLI_RESTRICT storage) {
1301   size_t masked_pos = position & mask;
1302   BrotliStoreUncompressedMetaBlockHeader(len, storage_ix, storage);
1303   JumpToByteBoundary(storage_ix, storage);
1304 
1305   if (masked_pos + len > mask + 1) {
1306     size_t len1 = mask + 1 - masked_pos;
1307     memcpy(&storage[*storage_ix >> 3], &input[masked_pos], len1);
1308     *storage_ix += len1 << 3;
1309     len -= len1;
1310     masked_pos = 0;
1311   }
1312   memcpy(&storage[*storage_ix >> 3], &input[masked_pos], len);
1313   *storage_ix += len << 3;
1314 
1315   /* We need to clear the next 4 bytes to continue to be
1316      compatible with BrotliWriteBits. */
1317   BrotliWriteBitsPrepareStorage(*storage_ix, storage);
1318 
1319   /* Since the uncompressed block itself may not be the final block, add an
1320      empty one after this. */
1321   if (is_final_block) {
1322     BrotliWriteBits(1, 1, storage_ix, storage);  /* islast */
1323     BrotliWriteBits(1, 1, storage_ix, storage);  /* isempty */
1324     JumpToByteBoundary(storage_ix, storage);
1325   }
1326 }
1327 
BrotliStoreSyncMetaBlock(size_t * BROTLI_RESTRICT storage_ix,uint8_t * BROTLI_RESTRICT storage)1328 void BrotliStoreSyncMetaBlock(size_t* BROTLI_RESTRICT storage_ix,
1329                               uint8_t* BROTLI_RESTRICT storage) {
1330   /* Empty metadata meta-block bit pattern:
1331        1 bit:  is_last (0)
1332        2 bits: num nibbles (3)
1333        1 bit:  reserved (0)
1334        2 bits: metadata length bytes (0) */
1335   BrotliWriteBits(6, 6, storage_ix, storage);
1336   JumpToByteBoundary(storage_ix, storage);
1337 }
1338 
1339 #if defined(__cplusplus) || defined(c_plusplus)
1340 }  /* extern "C" */
1341 #endif
1342