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 "./alphai.h"
18#include "./vp8li.h"
19#include "../dsp/dsp.h"
20#include "../dsp/lossless.h"
21#include "../dsp/yuv.h"
22#include "../utils/huffman.h"
23#include "../utils/utils.h"
24
25#define NUM_ARGB_CACHE_ROWS          16
26
27static const int kCodeLengthLiterals = 16;
28static const int kCodeLengthRepeatCode = 16;
29static const int kCodeLengthExtraBits[3] = { 2, 3, 7 };
30static const int kCodeLengthRepeatOffsets[3] = { 3, 3, 11 };
31
32// -----------------------------------------------------------------------------
33//  Five Huffman codes are used at each meta code:
34//  1. green + length prefix codes + color cache codes,
35//  2. alpha,
36//  3. red,
37//  4. blue, and,
38//  5. distance prefix codes.
39typedef enum {
40  GREEN = 0,
41  RED   = 1,
42  BLUE  = 2,
43  ALPHA = 3,
44  DIST  = 4
45} HuffIndex;
46
47static const uint16_t kAlphabetSize[HUFFMAN_CODES_PER_META_CODE] = {
48  NUM_LITERAL_CODES + NUM_LENGTH_CODES,
49  NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES,
50  NUM_DISTANCE_CODES
51};
52
53
54#define NUM_CODE_LENGTH_CODES       19
55static const uint8_t kCodeLengthCodeOrder[NUM_CODE_LENGTH_CODES] = {
56  17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
57};
58
59#define CODE_TO_PLANE_CODES        120
60static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = {
61  0x18, 0x07, 0x17, 0x19, 0x28, 0x06, 0x27, 0x29, 0x16, 0x1a,
62  0x26, 0x2a, 0x38, 0x05, 0x37, 0x39, 0x15, 0x1b, 0x36, 0x3a,
63  0x25, 0x2b, 0x48, 0x04, 0x47, 0x49, 0x14, 0x1c, 0x35, 0x3b,
64  0x46, 0x4a, 0x24, 0x2c, 0x58, 0x45, 0x4b, 0x34, 0x3c, 0x03,
65  0x57, 0x59, 0x13, 0x1d, 0x56, 0x5a, 0x23, 0x2d, 0x44, 0x4c,
66  0x55, 0x5b, 0x33, 0x3d, 0x68, 0x02, 0x67, 0x69, 0x12, 0x1e,
67  0x66, 0x6a, 0x22, 0x2e, 0x54, 0x5c, 0x43, 0x4d, 0x65, 0x6b,
68  0x32, 0x3e, 0x78, 0x01, 0x77, 0x79, 0x53, 0x5d, 0x11, 0x1f,
69  0x64, 0x6c, 0x42, 0x4e, 0x76, 0x7a, 0x21, 0x2f, 0x75, 0x7b,
70  0x31, 0x3f, 0x63, 0x6d, 0x52, 0x5e, 0x00, 0x74, 0x7c, 0x41,
71  0x4f, 0x10, 0x20, 0x62, 0x6e, 0x30, 0x73, 0x7d, 0x51, 0x5f,
72  0x40, 0x72, 0x7e, 0x61, 0x6f, 0x50, 0x71, 0x7f, 0x60, 0x70
73};
74
75static int DecodeImageStream(int xsize, int ysize,
76                             int is_level0,
77                             VP8LDecoder* const dec,
78                             uint32_t** const decoded_data);
79
80//------------------------------------------------------------------------------
81
82int VP8LCheckSignature(const uint8_t* const data, size_t size) {
83  return (size >= VP8L_FRAME_HEADER_SIZE &&
84          data[0] == VP8L_MAGIC_BYTE &&
85          (data[4] >> 5) == 0);  // version
86}
87
88static int ReadImageInfo(VP8LBitReader* const br,
89                         int* const width, int* const height,
90                         int* const has_alpha) {
91  if (VP8LReadBits(br, 8) != VP8L_MAGIC_BYTE) return 0;
92  *width = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1;
93  *height = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1;
94  *has_alpha = VP8LReadBits(br, 1);
95  if (VP8LReadBits(br, VP8L_VERSION_BITS) != 0) return 0;
96  return 1;
97}
98
99int VP8LGetInfo(const uint8_t* data, size_t data_size,
100                int* const width, int* const height, int* const has_alpha) {
101  if (data == NULL || data_size < VP8L_FRAME_HEADER_SIZE) {
102    return 0;         // not enough data
103  } else if (!VP8LCheckSignature(data, data_size)) {
104    return 0;         // bad signature
105  } else {
106    int w, h, a;
107    VP8LBitReader br;
108    VP8LInitBitReader(&br, data, data_size);
109    if (!ReadImageInfo(&br, &w, &h, &a)) {
110      return 0;
111    }
112    if (width != NULL) *width = w;
113    if (height != NULL) *height = h;
114    if (has_alpha != NULL) *has_alpha = a;
115    return 1;
116  }
117}
118
119//------------------------------------------------------------------------------
120
121static WEBP_INLINE int GetCopyDistance(int distance_symbol,
122                                       VP8LBitReader* const br) {
123  int extra_bits, offset;
124  if (distance_symbol < 4) {
125    return distance_symbol + 1;
126  }
127  extra_bits = (distance_symbol - 2) >> 1;
128  offset = (2 + (distance_symbol & 1)) << extra_bits;
129  return offset + VP8LReadBits(br, extra_bits) + 1;
130}
131
132static WEBP_INLINE int GetCopyLength(int length_symbol,
133                                     VP8LBitReader* const br) {
134  // Length and distance prefixes are encoded the same way.
135  return GetCopyDistance(length_symbol, br);
136}
137
138static WEBP_INLINE int PlaneCodeToDistance(int xsize, int plane_code) {
139  if (plane_code > CODE_TO_PLANE_CODES) {
140    return plane_code - CODE_TO_PLANE_CODES;
141  } else {
142    const int dist_code = kCodeToPlane[plane_code - 1];
143    const int yoffset = dist_code >> 4;
144    const int xoffset = 8 - (dist_code & 0xf);
145    const int dist = yoffset * xsize + xoffset;
146    return (dist >= 1) ? dist : 1;  // dist<1 can happen if xsize is very small
147  }
148}
149
150//------------------------------------------------------------------------------
151// Decodes the next Huffman code from bit-stream.
152// FillBitWindow(br) needs to be called at minimum every second call
153// to ReadSymbol, in order to pre-fetch enough bits.
154static WEBP_INLINE int ReadSymbol(const HuffmanTree* tree,
155                                  VP8LBitReader* const br) {
156  const HuffmanTreeNode* node = tree->root_;
157  uint32_t bits = VP8LPrefetchBits(br);
158  int bitpos = br->bit_pos_;
159  // Check if we find the bit combination from the Huffman lookup table.
160  const int lut_ix = bits & (HUFF_LUT - 1);
161  const int lut_bits = tree->lut_bits_[lut_ix];
162  if (lut_bits <= HUFF_LUT_BITS) {
163    VP8LSetBitPos(br, bitpos + lut_bits);
164    return tree->lut_symbol_[lut_ix];
165  }
166  node += tree->lut_jump_[lut_ix];
167  bitpos += HUFF_LUT_BITS;
168  bits >>= HUFF_LUT_BITS;
169
170  // Decode the value from a binary tree.
171  assert(node != NULL);
172  do {
173    node = HuffmanTreeNextNode(node, bits & 1);
174    bits >>= 1;
175    ++bitpos;
176  } while (HuffmanTreeNodeIsNotLeaf(node));
177  VP8LSetBitPos(br, bitpos);
178  return node->symbol_;
179}
180
181static int ReadHuffmanCodeLengths(
182    VP8LDecoder* const dec, const int* const code_length_code_lengths,
183    int num_symbols, int* const code_lengths) {
184  int ok = 0;
185  VP8LBitReader* const br = &dec->br_;
186  int symbol;
187  int max_symbol;
188  int prev_code_len = DEFAULT_CODE_LENGTH;
189  HuffmanTree tree;
190  int huff_codes[NUM_CODE_LENGTH_CODES] = { 0 };
191
192  if (!VP8LHuffmanTreeBuildImplicit(&tree, code_length_code_lengths,
193                                    huff_codes, NUM_CODE_LENGTH_CODES)) {
194    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
195    return 0;
196  }
197
198  if (VP8LReadBits(br, 1)) {    // use length
199    const int length_nbits = 2 + 2 * VP8LReadBits(br, 3);
200    max_symbol = 2 + VP8LReadBits(br, length_nbits);
201    if (max_symbol > num_symbols) {
202      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
203      goto End;
204    }
205  } else {
206    max_symbol = num_symbols;
207  }
208
209  symbol = 0;
210  while (symbol < num_symbols) {
211    int code_len;
212    if (max_symbol-- == 0) break;
213    VP8LFillBitWindow(br);
214    code_len = ReadSymbol(&tree, br);
215    if (code_len < kCodeLengthLiterals) {
216      code_lengths[symbol++] = code_len;
217      if (code_len != 0) prev_code_len = code_len;
218    } else {
219      const int use_prev = (code_len == kCodeLengthRepeatCode);
220      const int slot = code_len - kCodeLengthLiterals;
221      const int extra_bits = kCodeLengthExtraBits[slot];
222      const int repeat_offset = kCodeLengthRepeatOffsets[slot];
223      int repeat = VP8LReadBits(br, extra_bits) + repeat_offset;
224      if (symbol + repeat > num_symbols) {
225        dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
226        goto End;
227      } else {
228        const int length = use_prev ? prev_code_len : 0;
229        while (repeat-- > 0) code_lengths[symbol++] = length;
230      }
231    }
232  }
233  ok = 1;
234
235 End:
236  VP8LHuffmanTreeFree(&tree);
237  if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
238  return ok;
239}
240
241// 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman
242// tree.
243static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
244                           int* const code_lengths, int* const huff_codes,
245                           HuffmanTree* const tree) {
246  int ok = 0;
247  VP8LBitReader* const br = &dec->br_;
248  const int simple_code = VP8LReadBits(br, 1);
249
250  if (simple_code) {  // Read symbols, codes & code lengths directly.
251    int symbols[2];
252    int codes[2];
253    const int num_symbols = VP8LReadBits(br, 1) + 1;
254    const int first_symbol_len_code = VP8LReadBits(br, 1);
255    // The first code is either 1 bit or 8 bit code.
256    symbols[0] = VP8LReadBits(br, (first_symbol_len_code == 0) ? 1 : 8);
257    codes[0] = 0;
258    code_lengths[0] = num_symbols - 1;
259    // The second code (if present), is always 8 bit long.
260    if (num_symbols == 2) {
261      symbols[1] = VP8LReadBits(br, 8);
262      codes[1] = 1;
263      code_lengths[1] = num_symbols - 1;
264    }
265    ok = VP8LHuffmanTreeBuildExplicit(tree, code_lengths, codes, symbols,
266                                      alphabet_size, num_symbols);
267  } else {  // Decode Huffman-coded code lengths.
268    int i;
269    int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 };
270    const int num_codes = VP8LReadBits(br, 4) + 4;
271    if (num_codes > NUM_CODE_LENGTH_CODES) {
272      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
273      return 0;
274    }
275
276    memset(code_lengths, 0, alphabet_size * sizeof(*code_lengths));
277
278    for (i = 0; i < num_codes; ++i) {
279      code_length_code_lengths[kCodeLengthCodeOrder[i]] = VP8LReadBits(br, 3);
280    }
281    ok = ReadHuffmanCodeLengths(dec, code_length_code_lengths, alphabet_size,
282                                code_lengths);
283    ok = ok && VP8LHuffmanTreeBuildImplicit(tree, code_lengths, huff_codes,
284                                            alphabet_size);
285  }
286  ok = ok && !br->error_;
287  if (!ok) {
288    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
289    return 0;
290  }
291  return 1;
292}
293
294static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
295                            int color_cache_bits, int allow_recursion) {
296  int i, j;
297  VP8LBitReader* const br = &dec->br_;
298  VP8LMetadata* const hdr = &dec->hdr_;
299  uint32_t* huffman_image = NULL;
300  HTreeGroup* htree_groups = NULL;
301  int num_htree_groups = 1;
302  int max_alphabet_size = 0;
303  int* code_lengths = NULL;
304  int* huff_codes = NULL;
305
306  if (allow_recursion && VP8LReadBits(br, 1)) {
307    // use meta Huffman codes.
308    const int huffman_precision = VP8LReadBits(br, 3) + 2;
309    const int huffman_xsize = VP8LSubSampleSize(xsize, huffman_precision);
310    const int huffman_ysize = VP8LSubSampleSize(ysize, huffman_precision);
311    const int huffman_pixs = huffman_xsize * huffman_ysize;
312    if (!DecodeImageStream(huffman_xsize, huffman_ysize, 0, dec,
313                           &huffman_image)) {
314      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
315      goto Error;
316    }
317    hdr->huffman_subsample_bits_ = huffman_precision;
318    for (i = 0; i < huffman_pixs; ++i) {
319      // The huffman data is stored in red and green bytes.
320      const int group = (huffman_image[i] >> 8) & 0xffff;
321      huffman_image[i] = group;
322      if (group >= num_htree_groups) {
323        num_htree_groups = group + 1;
324      }
325    }
326  }
327
328  if (br->error_) goto Error;
329
330  // Find maximum alphabet size for the htree group.
331  for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
332    int alphabet_size = kAlphabetSize[j];
333    if (j == 0 && color_cache_bits > 0) {
334      alphabet_size += 1 << color_cache_bits;
335    }
336    if (max_alphabet_size < alphabet_size) {
337      max_alphabet_size = alphabet_size;
338    }
339  }
340
341  htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
342  code_lengths =
343      (int*)WebPSafeCalloc((uint64_t)max_alphabet_size, sizeof(*code_lengths));
344  huff_codes =
345      (int*)WebPSafeMalloc((uint64_t)max_alphabet_size, sizeof(*huff_codes));
346
347  if (htree_groups == NULL || code_lengths == NULL || huff_codes == NULL) {
348    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
349    goto Error;
350  }
351
352  for (i = 0; i < num_htree_groups; ++i) {
353    HuffmanTree* const htrees = htree_groups[i].htrees_;
354    for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
355      int alphabet_size = kAlphabetSize[j];
356      HuffmanTree* const htree = htrees + j;
357      if (j == 0 && color_cache_bits > 0) {
358        alphabet_size += 1 << color_cache_bits;
359      }
360      if (!ReadHuffmanCode(alphabet_size, dec, code_lengths, huff_codes,
361                           htree)) {
362        goto Error;
363      }
364    }
365  }
366  WebPSafeFree(huff_codes);
367  WebPSafeFree(code_lengths);
368
369  // All OK. Finalize pointers and return.
370  hdr->huffman_image_ = huffman_image;
371  hdr->num_htree_groups_ = num_htree_groups;
372  hdr->htree_groups_ = htree_groups;
373  return 1;
374
375 Error:
376  WebPSafeFree(huff_codes);
377  WebPSafeFree(code_lengths);
378  WebPSafeFree(huffman_image);
379  VP8LHtreeGroupsFree(htree_groups, num_htree_groups);
380  return 0;
381}
382
383//------------------------------------------------------------------------------
384// Scaling.
385
386static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
387  const int num_channels = 4;
388  const int in_width = io->mb_w;
389  const int out_width = io->scaled_width;
390  const int in_height = io->mb_h;
391  const int out_height = io->scaled_height;
392  const uint64_t work_size = 2 * num_channels * (uint64_t)out_width;
393  int32_t* work;        // Rescaler work area.
394  const uint64_t scaled_data_size = num_channels * (uint64_t)out_width;
395  uint32_t* scaled_data;  // Temporary storage for scaled BGRA data.
396  const uint64_t memory_size = sizeof(*dec->rescaler) +
397                               work_size * sizeof(*work) +
398                               scaled_data_size * sizeof(*scaled_data);
399  uint8_t* memory = (uint8_t*)WebPSafeCalloc(memory_size, sizeof(*memory));
400  if (memory == NULL) {
401    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
402    return 0;
403  }
404  assert(dec->rescaler_memory == NULL);
405  dec->rescaler_memory = memory;
406
407  dec->rescaler = (WebPRescaler*)memory;
408  memory += sizeof(*dec->rescaler);
409  work = (int32_t*)memory;
410  memory += work_size * sizeof(*work);
411  scaled_data = (uint32_t*)memory;
412
413  WebPRescalerInit(dec->rescaler, in_width, in_height, (uint8_t*)scaled_data,
414                   out_width, out_height, 0, num_channels,
415                   in_width, out_width, in_height, out_height, work);
416  return 1;
417}
418
419//------------------------------------------------------------------------------
420// Export to ARGB
421
422// We have special "export" function since we need to convert from BGRA
423static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace,
424                  int rgba_stride, uint8_t* const rgba) {
425  uint32_t* const src = (uint32_t*)rescaler->dst;
426  const int dst_width = rescaler->dst_width;
427  int num_lines_out = 0;
428  while (WebPRescalerHasPendingOutput(rescaler)) {
429    uint8_t* const dst = rgba + num_lines_out * rgba_stride;
430    WebPRescalerExportRow(rescaler, 0);
431    WebPMultARGBRow(src, dst_width, 1);
432    VP8LConvertFromBGRA(src, dst_width, colorspace, dst);
433    ++num_lines_out;
434  }
435  return num_lines_out;
436}
437
438// Emit scaled rows.
439static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec,
440                                uint8_t* in, int in_stride, int mb_h,
441                                uint8_t* const out, int out_stride) {
442  const WEBP_CSP_MODE colorspace = dec->output_->colorspace;
443  int num_lines_in = 0;
444  int num_lines_out = 0;
445  while (num_lines_in < mb_h) {
446    uint8_t* const row_in = in + num_lines_in * in_stride;
447    uint8_t* const row_out = out + num_lines_out * out_stride;
448    const int lines_left = mb_h - num_lines_in;
449    const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left);
450    assert(needed_lines > 0 && needed_lines <= lines_left);
451    WebPMultARGBRows(row_in, in_stride,
452                     dec->rescaler->src_width, needed_lines, 0);
453    WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride);
454    num_lines_in += needed_lines;
455    num_lines_out += Export(dec->rescaler, colorspace, out_stride, row_out);
456  }
457  return num_lines_out;
458}
459
460// Emit rows without any scaling.
461static int EmitRows(WEBP_CSP_MODE colorspace,
462                    const uint8_t* row_in, int in_stride,
463                    int mb_w, int mb_h,
464                    uint8_t* const out, int out_stride) {
465  int lines = mb_h;
466  uint8_t* row_out = out;
467  while (lines-- > 0) {
468    VP8LConvertFromBGRA((const uint32_t*)row_in, mb_w, colorspace, row_out);
469    row_in += in_stride;
470    row_out += out_stride;
471  }
472  return mb_h;  // Num rows out == num rows in.
473}
474
475//------------------------------------------------------------------------------
476// Export to YUVA
477
478// TODO(skal): should be in yuv.c
479static void ConvertToYUVA(const uint32_t* const src, int width, int y_pos,
480                          const WebPDecBuffer* const output) {
481  const WebPYUVABuffer* const buf = &output->u.YUVA;
482  // first, the luma plane
483  {
484    int i;
485    uint8_t* const y = buf->y + y_pos * buf->y_stride;
486    for (i = 0; i < width; ++i) {
487      const uint32_t p = src[i];
488      y[i] = VP8RGBToY((p >> 16) & 0xff, (p >> 8) & 0xff, (p >> 0) & 0xff,
489                       YUV_HALF);
490    }
491  }
492
493  // then U/V planes
494  {
495    uint8_t* const u = buf->u + (y_pos >> 1) * buf->u_stride;
496    uint8_t* const v = buf->v + (y_pos >> 1) * buf->v_stride;
497    const int uv_width = width >> 1;
498    int i;
499    for (i = 0; i < uv_width; ++i) {
500      const uint32_t v0 = src[2 * i + 0];
501      const uint32_t v1 = src[2 * i + 1];
502      // VP8RGBToU/V expects four accumulated pixels. Hence we need to
503      // scale r/g/b value by a factor 2. We just shift v0/v1 one bit less.
504      const int r = ((v0 >> 15) & 0x1fe) + ((v1 >> 15) & 0x1fe);
505      const int g = ((v0 >>  7) & 0x1fe) + ((v1 >>  7) & 0x1fe);
506      const int b = ((v0 <<  1) & 0x1fe) + ((v1 <<  1) & 0x1fe);
507      if (!(y_pos & 1)) {  // even lines: store values
508        u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
509        v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
510      } else {             // odd lines: average with previous values
511        const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
512        const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
513        // Approximated average-of-four. But it's an acceptable diff.
514        u[i] = (u[i] + tmp_u + 1) >> 1;
515        v[i] = (v[i] + tmp_v + 1) >> 1;
516      }
517    }
518    if (width & 1) {       // last pixel
519      const uint32_t v0 = src[2 * i + 0];
520      const int r = (v0 >> 14) & 0x3fc;
521      const int g = (v0 >>  6) & 0x3fc;
522      const int b = (v0 <<  2) & 0x3fc;
523      if (!(y_pos & 1)) {  // even lines
524        u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
525        v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
526      } else {             // odd lines (note: we could just skip this)
527        const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
528        const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
529        u[i] = (u[i] + tmp_u + 1) >> 1;
530        v[i] = (v[i] + tmp_v + 1) >> 1;
531      }
532    }
533  }
534  // Lastly, store alpha if needed.
535  if (buf->a != NULL) {
536    int i;
537    uint8_t* const a = buf->a + y_pos * buf->a_stride;
538    for (i = 0; i < width; ++i) a[i] = (src[i] >> 24);
539  }
540}
541
542static int ExportYUVA(const VP8LDecoder* const dec, int y_pos) {
543  WebPRescaler* const rescaler = dec->rescaler;
544  uint32_t* const src = (uint32_t*)rescaler->dst;
545  const int dst_width = rescaler->dst_width;
546  int num_lines_out = 0;
547  while (WebPRescalerHasPendingOutput(rescaler)) {
548    WebPRescalerExportRow(rescaler, 0);
549    WebPMultARGBRow(src, dst_width, 1);
550    ConvertToYUVA(src, dst_width, y_pos, dec->output_);
551    ++y_pos;
552    ++num_lines_out;
553  }
554  return num_lines_out;
555}
556
557static int EmitRescaledRowsYUVA(const VP8LDecoder* const dec,
558                                uint8_t* in, int in_stride, int mb_h) {
559  int num_lines_in = 0;
560  int y_pos = dec->last_out_row_;
561  while (num_lines_in < mb_h) {
562    const int lines_left = mb_h - num_lines_in;
563    const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left);
564    WebPMultARGBRows(in, in_stride, dec->rescaler->src_width, needed_lines, 0);
565    WebPRescalerImport(dec->rescaler, lines_left, in, in_stride);
566    num_lines_in += needed_lines;
567    in += needed_lines * in_stride;
568    y_pos += ExportYUVA(dec, y_pos);
569  }
570  return y_pos;
571}
572
573static int EmitRowsYUVA(const VP8LDecoder* const dec,
574                        const uint8_t* in, int in_stride,
575                        int mb_w, int num_rows) {
576  int y_pos = dec->last_out_row_;
577  while (num_rows-- > 0) {
578    ConvertToYUVA((const uint32_t*)in, mb_w, y_pos, dec->output_);
579    in += in_stride;
580    ++y_pos;
581  }
582  return y_pos;
583}
584
585//------------------------------------------------------------------------------
586// Cropping.
587
588// Sets io->mb_y, io->mb_h & io->mb_w according to start row, end row and
589// crop options. Also updates the input data pointer, so that it points to the
590// start of the cropped window. Note that pixels are in ARGB format even if
591// 'in_data' is uint8_t*.
592// Returns true if the crop window is not empty.
593static int SetCropWindow(VP8Io* const io, int y_start, int y_end,
594                         uint8_t** const in_data, int pixel_stride) {
595  assert(y_start < y_end);
596  assert(io->crop_left < io->crop_right);
597  if (y_end > io->crop_bottom) {
598    y_end = io->crop_bottom;  // make sure we don't overflow on last row.
599  }
600  if (y_start < io->crop_top) {
601    const int delta = io->crop_top - y_start;
602    y_start = io->crop_top;
603    *in_data += delta * pixel_stride;
604  }
605  if (y_start >= y_end) return 0;  // Crop window is empty.
606
607  *in_data += io->crop_left * sizeof(uint32_t);
608
609  io->mb_y = y_start - io->crop_top;
610  io->mb_w = io->crop_right - io->crop_left;
611  io->mb_h = y_end - y_start;
612  return 1;  // Non-empty crop window.
613}
614
615//------------------------------------------------------------------------------
616
617static WEBP_INLINE int GetMetaIndex(
618    const uint32_t* const image, int xsize, int bits, int x, int y) {
619  if (bits == 0) return 0;
620  return image[xsize * (y >> bits) + (x >> bits)];
621}
622
623static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr,
624                                                   int x, int y) {
625  const int meta_index = GetMetaIndex(hdr->huffman_image_, hdr->huffman_xsize_,
626                                      hdr->huffman_subsample_bits_, x, y);
627  assert(meta_index < hdr->num_htree_groups_);
628  return hdr->htree_groups_ + meta_index;
629}
630
631//------------------------------------------------------------------------------
632// Main loop, with custom row-processing function
633
634typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row);
635
636static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows,
637                                   const uint32_t* const rows) {
638  int n = dec->next_transform_;
639  const int cache_pixs = dec->width_ * num_rows;
640  const int start_row = dec->last_row_;
641  const int end_row = start_row + num_rows;
642  const uint32_t* rows_in = rows;
643  uint32_t* const rows_out = dec->argb_cache_;
644
645  // Inverse transforms.
646  // TODO: most transforms only need to operate on the cropped region only.
647  memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out));
648  while (n-- > 0) {
649    VP8LTransform* const transform = &dec->transforms_[n];
650    VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out);
651    rows_in = rows_out;
652  }
653}
654
655// Special method for paletted alpha data.
656static void ApplyInverseTransformsAlpha(VP8LDecoder* const dec, int num_rows,
657                                        const uint8_t* const rows) {
658  const int start_row = dec->last_row_;
659  const int end_row = start_row + num_rows;
660  const uint8_t* rows_in = rows;
661  uint8_t* rows_out = (uint8_t*)dec->io_->opaque + dec->io_->width * start_row;
662  VP8LTransform* const transform = &dec->transforms_[0];
663  assert(dec->next_transform_ == 1);
664  assert(transform->type_ == COLOR_INDEXING_TRANSFORM);
665  VP8LColorIndexInverseTransformAlpha(transform, start_row, end_row, rows_in,
666                                      rows_out);
667}
668
669// Processes (transforms, scales & color-converts) the rows decoded after the
670// last call.
671static void ProcessRows(VP8LDecoder* const dec, int row) {
672  const uint32_t* const rows = dec->pixels_ + dec->width_ * dec->last_row_;
673  const int num_rows = row - dec->last_row_;
674
675  if (num_rows <= 0) return;  // Nothing to be done.
676  ApplyInverseTransforms(dec, num_rows, rows);
677
678  // Emit output.
679  {
680    VP8Io* const io = dec->io_;
681    uint8_t* rows_data = (uint8_t*)dec->argb_cache_;
682    const int in_stride = io->width * sizeof(uint32_t);  // in unit of RGBA
683    if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) {
684      // Nothing to output (this time).
685    } else {
686      const WebPDecBuffer* const output = dec->output_;
687      if (output->colorspace < MODE_YUV) {  // convert to RGBA
688        const WebPRGBABuffer* const buf = &output->u.RGBA;
689        uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride;
690        const int num_rows_out = io->use_scaling ?
691            EmitRescaledRowsRGBA(dec, rows_data, in_stride, io->mb_h,
692                                 rgba, buf->stride) :
693            EmitRows(output->colorspace, rows_data, in_stride,
694                     io->mb_w, io->mb_h, rgba, buf->stride);
695        // Update 'last_out_row_'.
696        dec->last_out_row_ += num_rows_out;
697      } else {                              // convert to YUVA
698        dec->last_out_row_ = io->use_scaling ?
699            EmitRescaledRowsYUVA(dec, rows_data, in_stride, io->mb_h) :
700            EmitRowsYUVA(dec, rows_data, in_stride, io->mb_w, io->mb_h);
701      }
702      assert(dec->last_out_row_ <= output->height);
703    }
704  }
705
706  // Update 'last_row_'.
707  dec->last_row_ = row;
708  assert(dec->last_row_ <= dec->height_);
709}
710
711// Row-processing for the special case when alpha data contains only one
712// transform (color indexing), and trivial non-green literals.
713static int Is8bOptimizable(const VP8LMetadata* const hdr) {
714  int i;
715  if (hdr->color_cache_size_ > 0) return 0;
716  // When the Huffman tree contains only one symbol, we can skip the
717  // call to ReadSymbol() for red/blue/alpha channels.
718  for (i = 0; i < hdr->num_htree_groups_; ++i) {
719    const HuffmanTree* const htrees = hdr->htree_groups_[i].htrees_;
720    if (htrees[RED].num_nodes_ > 1) return 0;
721    if (htrees[BLUE].num_nodes_ > 1) return 0;
722    if (htrees[ALPHA].num_nodes_ > 1) return 0;
723  }
724  return 1;
725}
726
727static void ExtractPalettedAlphaRows(VP8LDecoder* const dec, int row) {
728  const int num_rows = row - dec->last_row_;
729  const uint8_t* const in =
730      (uint8_t*)dec->pixels_ + dec->width_ * dec->last_row_;
731  if (num_rows > 0) {
732    ApplyInverseTransformsAlpha(dec, num_rows, in);
733  }
734  dec->last_row_ = dec->last_out_row_ = row;
735}
736
737static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data,
738                           int width, int height, int last_row) {
739  int ok = 1;
740  int row = dec->last_pixel_ / width;
741  int col = dec->last_pixel_ % width;
742  VP8LBitReader* const br = &dec->br_;
743  VP8LMetadata* const hdr = &dec->hdr_;
744  const HTreeGroup* htree_group = GetHtreeGroupForPos(hdr, col, row);
745  int pos = dec->last_pixel_;         // current position
746  const int end = width * height;     // End of data
747  const int last = width * last_row;  // Last pixel to decode
748  const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES;
749  const int mask = hdr->huffman_mask_;
750  assert(htree_group != NULL);
751  assert(pos < end);
752  assert(last_row <= height);
753  assert(Is8bOptimizable(hdr));
754
755  while (!br->eos_ && pos < last) {
756    int code;
757    // Only update when changing tile.
758    if ((col & mask) == 0) {
759      htree_group = GetHtreeGroupForPos(hdr, col, row);
760    }
761    VP8LFillBitWindow(br);
762    code = ReadSymbol(&htree_group->htrees_[GREEN], br);
763    if (code < NUM_LITERAL_CODES) {  // Literal
764      data[pos] = code;
765      ++pos;
766      ++col;
767      if (col >= width) {
768        col = 0;
769        ++row;
770        if (row % NUM_ARGB_CACHE_ROWS == 0) {
771          ExtractPalettedAlphaRows(dec, row);
772        }
773      }
774    } else if (code < len_code_limit) {  // Backward reference
775      int dist_code, dist;
776      const int length_sym = code - NUM_LITERAL_CODES;
777      const int length = GetCopyLength(length_sym, br);
778      const int dist_symbol = ReadSymbol(&htree_group->htrees_[DIST], br);
779      VP8LFillBitWindow(br);
780      dist_code = GetCopyDistance(dist_symbol, br);
781      dist = PlaneCodeToDistance(width, dist_code);
782      if (pos >= dist && end - pos >= length) {
783        int i;
784        for (i = 0; i < length; ++i) data[pos + i] = data[pos + i - dist];
785      } else {
786        ok = 0;
787        goto End;
788      }
789      pos += length;
790      col += length;
791      while (col >= width) {
792        col -= width;
793        ++row;
794        if (row % NUM_ARGB_CACHE_ROWS == 0) {
795          ExtractPalettedAlphaRows(dec, row);
796        }
797      }
798      if (pos < last && (col & mask)) {
799        htree_group = GetHtreeGroupForPos(hdr, col, row);
800      }
801    } else {  // Not reached
802      ok = 0;
803      goto End;
804    }
805    assert(br->eos_ == VP8LIsEndOfStream(br));
806    ok = !br->error_;
807    if (!ok) goto End;
808  }
809  // Process the remaining rows corresponding to last row-block.
810  ExtractPalettedAlphaRows(dec, row);
811
812 End:
813  if (br->error_ || !ok || (br->eos_ && pos < end)) {
814    ok = 0;
815    dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED
816                            : VP8_STATUS_BITSTREAM_ERROR;
817  } else {
818    dec->last_pixel_ = (int)pos;
819    if (pos == end) dec->state_ = READ_DATA;
820  }
821  return ok;
822}
823
824static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
825                           int width, int height, int last_row,
826                           ProcessRowsFunc process_func) {
827  int ok = 1;
828  int row = dec->last_pixel_ / width;
829  int col = dec->last_pixel_ % width;
830  VP8LBitReader* const br = &dec->br_;
831  VP8LMetadata* const hdr = &dec->hdr_;
832  HTreeGroup* htree_group = GetHtreeGroupForPos(hdr, col, row);
833  uint32_t* src = data + dec->last_pixel_;
834  uint32_t* last_cached = src;
835  uint32_t* const src_end = data + width * height;     // End of data
836  uint32_t* const src_last = data + width * last_row;  // Last pixel to decode
837  const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES;
838  const int color_cache_limit = len_code_limit + hdr->color_cache_size_;
839  VP8LColorCache* const color_cache =
840      (hdr->color_cache_size_ > 0) ? &hdr->color_cache_ : NULL;
841  const int mask = hdr->huffman_mask_;
842  assert(htree_group != NULL);
843  assert(src < src_end);
844  assert(src_last <= src_end);
845
846  while (!br->eos_ && src < src_last) {
847    int code;
848    // Only update when changing tile. Note we could use this test:
849    // if "((((prev_col ^ col) | prev_row ^ row)) > mask)" -> tile changed
850    // but that's actually slower and needs storing the previous col/row.
851    if ((col & mask) == 0) {
852      htree_group = GetHtreeGroupForPos(hdr, col, row);
853    }
854    VP8LFillBitWindow(br);
855    code = ReadSymbol(&htree_group->htrees_[GREEN], br);
856    if (code < NUM_LITERAL_CODES) {  // Literal
857      int red, green, blue, alpha;
858      red = ReadSymbol(&htree_group->htrees_[RED], br);
859      green = code;
860      VP8LFillBitWindow(br);
861      blue = ReadSymbol(&htree_group->htrees_[BLUE], br);
862      alpha = ReadSymbol(&htree_group->htrees_[ALPHA], br);
863      *src = ((uint32_t)alpha << 24) | (red << 16) | (green << 8) | blue;
864    AdvanceByOne:
865      ++src;
866      ++col;
867      if (col >= width) {
868        col = 0;
869        ++row;
870        if ((row % NUM_ARGB_CACHE_ROWS == 0) && (process_func != NULL)) {
871          process_func(dec, row);
872        }
873        if (color_cache != NULL) {
874          while (last_cached < src) {
875            VP8LColorCacheInsert(color_cache, *last_cached++);
876          }
877        }
878      }
879    } else if (code < len_code_limit) {  // Backward reference
880      int dist_code, dist;
881      const int length_sym = code - NUM_LITERAL_CODES;
882      const int length = GetCopyLength(length_sym, br);
883      const int dist_symbol = ReadSymbol(&htree_group->htrees_[DIST], br);
884      VP8LFillBitWindow(br);
885      dist_code = GetCopyDistance(dist_symbol, br);
886      dist = PlaneCodeToDistance(width, dist_code);
887      if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) {
888        ok = 0;
889        goto End;
890      } else {
891        int i;
892        for (i = 0; i < length; ++i) src[i] = src[i - dist];
893        src += length;
894      }
895      col += length;
896      while (col >= width) {
897        col -= width;
898        ++row;
899        if ((row % NUM_ARGB_CACHE_ROWS == 0) && (process_func != NULL)) {
900          process_func(dec, row);
901        }
902      }
903      if (src < src_end) {
904        if (col & mask) htree_group = GetHtreeGroupForPos(hdr, col, row);
905        if (color_cache != NULL) {
906          while (last_cached < src) {
907            VP8LColorCacheInsert(color_cache, *last_cached++);
908          }
909        }
910      }
911    } else if (code < color_cache_limit) {  // Color cache
912      const int key = code - len_code_limit;
913      assert(color_cache != NULL);
914      while (last_cached < src) {
915        VP8LColorCacheInsert(color_cache, *last_cached++);
916      }
917      *src = VP8LColorCacheLookup(color_cache, key);
918      goto AdvanceByOne;
919    } else {  // Not reached
920      ok = 0;
921      goto End;
922    }
923    assert(br->eos_ == VP8LIsEndOfStream(br));
924    ok = !br->error_;
925    if (!ok) goto End;
926  }
927  // Process the remaining rows corresponding to last row-block.
928  if (process_func != NULL) process_func(dec, row);
929
930 End:
931  if (br->error_ || !ok || (br->eos_ && src < src_end)) {
932    ok = 0;
933    dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED
934                            : VP8_STATUS_BITSTREAM_ERROR;
935  } else {
936    dec->last_pixel_ = (int)(src - data);
937    if (src == src_end) dec->state_ = READ_DATA;
938  }
939  return ok;
940}
941
942// -----------------------------------------------------------------------------
943// VP8LTransform
944
945static void ClearTransform(VP8LTransform* const transform) {
946  WebPSafeFree(transform->data_);
947  transform->data_ = NULL;
948}
949
950// For security reason, we need to remap the color map to span
951// the total possible bundled values, and not just the num_colors.
952static int ExpandColorMap(int num_colors, VP8LTransform* const transform) {
953  int i;
954  const int final_num_colors = 1 << (8 >> transform->bits_);
955  uint32_t* const new_color_map =
956      (uint32_t*)WebPSafeMalloc((uint64_t)final_num_colors,
957                                sizeof(*new_color_map));
958  if (new_color_map == NULL) {
959    return 0;
960  } else {
961    uint8_t* const data = (uint8_t*)transform->data_;
962    uint8_t* const new_data = (uint8_t*)new_color_map;
963    new_color_map[0] = transform->data_[0];
964    for (i = 4; i < 4 * num_colors; ++i) {
965      // Equivalent to AddPixelEq(), on a byte-basis.
966      new_data[i] = (data[i] + new_data[i - 4]) & 0xff;
967    }
968    for (; i < 4 * final_num_colors; ++i)
969      new_data[i] = 0;  // black tail.
970    WebPSafeFree(transform->data_);
971    transform->data_ = new_color_map;
972  }
973  return 1;
974}
975
976static int ReadTransform(int* const xsize, int const* ysize,
977                         VP8LDecoder* const dec) {
978  int ok = 1;
979  VP8LBitReader* const br = &dec->br_;
980  VP8LTransform* transform = &dec->transforms_[dec->next_transform_];
981  const VP8LImageTransformType type =
982      (VP8LImageTransformType)VP8LReadBits(br, 2);
983
984  // Each transform type can only be present once in the stream.
985  if (dec->transforms_seen_ & (1U << type)) {
986    return 0;  // Already there, let's not accept the second same transform.
987  }
988  dec->transforms_seen_ |= (1U << type);
989
990  transform->type_ = type;
991  transform->xsize_ = *xsize;
992  transform->ysize_ = *ysize;
993  transform->data_ = NULL;
994  ++dec->next_transform_;
995  assert(dec->next_transform_ <= NUM_TRANSFORMS);
996
997  switch (type) {
998    case PREDICTOR_TRANSFORM:
999    case CROSS_COLOR_TRANSFORM:
1000      transform->bits_ = VP8LReadBits(br, 3) + 2;
1001      ok = DecodeImageStream(VP8LSubSampleSize(transform->xsize_,
1002                                               transform->bits_),
1003                             VP8LSubSampleSize(transform->ysize_,
1004                                               transform->bits_),
1005                             0, dec, &transform->data_);
1006      break;
1007    case COLOR_INDEXING_TRANSFORM: {
1008       const int num_colors = VP8LReadBits(br, 8) + 1;
1009       const int bits = (num_colors > 16) ? 0
1010                      : (num_colors > 4) ? 1
1011                      : (num_colors > 2) ? 2
1012                      : 3;
1013       *xsize = VP8LSubSampleSize(transform->xsize_, bits);
1014       transform->bits_ = bits;
1015       ok = DecodeImageStream(num_colors, 1, 0, dec, &transform->data_);
1016       ok = ok && ExpandColorMap(num_colors, transform);
1017      break;
1018    }
1019    case SUBTRACT_GREEN:
1020      break;
1021    default:
1022      assert(0);    // can't happen
1023      break;
1024  }
1025
1026  return ok;
1027}
1028
1029// -----------------------------------------------------------------------------
1030// VP8LMetadata
1031
1032static void InitMetadata(VP8LMetadata* const hdr) {
1033  assert(hdr);
1034  memset(hdr, 0, sizeof(*hdr));
1035}
1036
1037static void ClearMetadata(VP8LMetadata* const hdr) {
1038  assert(hdr);
1039
1040  WebPSafeFree(hdr->huffman_image_);
1041  VP8LHtreeGroupsFree(hdr->htree_groups_, hdr->num_htree_groups_);
1042  VP8LColorCacheClear(&hdr->color_cache_);
1043  InitMetadata(hdr);
1044}
1045
1046// -----------------------------------------------------------------------------
1047// VP8LDecoder
1048
1049VP8LDecoder* VP8LNew(void) {
1050  VP8LDecoder* const dec = (VP8LDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec));
1051  if (dec == NULL) return NULL;
1052  dec->status_ = VP8_STATUS_OK;
1053  dec->action_ = READ_DIM;
1054  dec->state_ = READ_DIM;
1055
1056  VP8LDspInit();  // Init critical function pointers.
1057
1058  return dec;
1059}
1060
1061void VP8LClear(VP8LDecoder* const dec) {
1062  int i;
1063  if (dec == NULL) return;
1064  ClearMetadata(&dec->hdr_);
1065
1066  WebPSafeFree(dec->pixels_);
1067  dec->pixels_ = NULL;
1068  for (i = 0; i < dec->next_transform_; ++i) {
1069    ClearTransform(&dec->transforms_[i]);
1070  }
1071  dec->next_transform_ = 0;
1072  dec->transforms_seen_ = 0;
1073
1074  WebPSafeFree(dec->rescaler_memory);
1075  dec->rescaler_memory = NULL;
1076
1077  dec->output_ = NULL;   // leave no trace behind
1078}
1079
1080void VP8LDelete(VP8LDecoder* const dec) {
1081  if (dec != NULL) {
1082    VP8LClear(dec);
1083    WebPSafeFree(dec);
1084  }
1085}
1086
1087static void UpdateDecoder(VP8LDecoder* const dec, int width, int height) {
1088  VP8LMetadata* const hdr = &dec->hdr_;
1089  const int num_bits = hdr->huffman_subsample_bits_;
1090  dec->width_ = width;
1091  dec->height_ = height;
1092
1093  hdr->huffman_xsize_ = VP8LSubSampleSize(width, num_bits);
1094  hdr->huffman_mask_ = (num_bits == 0) ? ~0 : (1 << num_bits) - 1;
1095}
1096
1097static int DecodeImageStream(int xsize, int ysize,
1098                             int is_level0,
1099                             VP8LDecoder* const dec,
1100                             uint32_t** const decoded_data) {
1101  int ok = 1;
1102  int transform_xsize = xsize;
1103  int transform_ysize = ysize;
1104  VP8LBitReader* const br = &dec->br_;
1105  VP8LMetadata* const hdr = &dec->hdr_;
1106  uint32_t* data = NULL;
1107  int color_cache_bits = 0;
1108
1109  // Read the transforms (may recurse).
1110  if (is_level0) {
1111    while (ok && VP8LReadBits(br, 1)) {
1112      ok = ReadTransform(&transform_xsize, &transform_ysize, dec);
1113    }
1114  }
1115
1116  // Color cache
1117  if (ok && VP8LReadBits(br, 1)) {
1118    color_cache_bits = VP8LReadBits(br, 4);
1119    ok = (color_cache_bits >= 1 && color_cache_bits <= MAX_CACHE_BITS);
1120    if (!ok) {
1121      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
1122      goto End;
1123    }
1124  }
1125
1126  // Read the Huffman codes (may recurse).
1127  ok = ok && ReadHuffmanCodes(dec, transform_xsize, transform_ysize,
1128                              color_cache_bits, is_level0);
1129  if (!ok) {
1130    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
1131    goto End;
1132  }
1133
1134  // Finish setting up the color-cache
1135  if (color_cache_bits > 0) {
1136    hdr->color_cache_size_ = 1 << color_cache_bits;
1137    if (!VP8LColorCacheInit(&hdr->color_cache_, color_cache_bits)) {
1138      dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
1139      ok = 0;
1140      goto End;
1141    }
1142  } else {
1143    hdr->color_cache_size_ = 0;
1144  }
1145  UpdateDecoder(dec, transform_xsize, transform_ysize);
1146
1147  if (is_level0) {   // level 0 complete
1148    dec->state_ = READ_HDR;
1149    goto End;
1150  }
1151
1152  {
1153    const uint64_t total_size = (uint64_t)transform_xsize * transform_ysize;
1154    data = (uint32_t*)WebPSafeMalloc(total_size, sizeof(*data));
1155    if (data == NULL) {
1156      dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
1157      ok = 0;
1158      goto End;
1159    }
1160  }
1161
1162  // Use the Huffman trees to decode the LZ77 encoded data.
1163  ok = DecodeImageData(dec, data, transform_xsize, transform_ysize,
1164                       transform_ysize, NULL);
1165  ok = ok && !br->error_;
1166
1167 End:
1168
1169  if (!ok) {
1170    WebPSafeFree(data);
1171    ClearMetadata(hdr);
1172    // If not enough data (br.eos_) resulted in BIT_STREAM_ERROR, update the
1173    // status appropriately.
1174    if (dec->status_ == VP8_STATUS_BITSTREAM_ERROR && dec->br_.eos_) {
1175      dec->status_ = VP8_STATUS_SUSPENDED;
1176    }
1177  } else {
1178    if (decoded_data != NULL) {
1179      *decoded_data = data;
1180    } else {
1181      // We allocate image data in this function only for transforms. At level 0
1182      // (that is: not the transforms), we shouldn't have allocated anything.
1183      assert(data == NULL);
1184      assert(is_level0);
1185    }
1186    dec->last_pixel_ = 0;  // Reset for future DECODE_DATA_FUNC() calls.
1187    if (!is_level0) ClearMetadata(hdr);  // Clean up temporary data behind.
1188  }
1189  return ok;
1190}
1191
1192//------------------------------------------------------------------------------
1193// Allocate internal buffers dec->pixels_ and dec->argb_cache_.
1194static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) {
1195  const uint64_t num_pixels = (uint64_t)dec->width_ * dec->height_;
1196  // Scratch buffer corresponding to top-prediction row for transforming the
1197  // first row in the row-blocks. Not needed for paletted alpha.
1198  const uint64_t cache_top_pixels = (uint16_t)final_width;
1199  // Scratch buffer for temporary BGRA storage. Not needed for paletted alpha.
1200  const uint64_t cache_pixels = (uint64_t)final_width * NUM_ARGB_CACHE_ROWS;
1201  const uint64_t total_num_pixels =
1202      num_pixels + cache_top_pixels + cache_pixels;
1203
1204  assert(dec->width_ <= final_width);
1205  dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint32_t));
1206  if (dec->pixels_ == NULL) {
1207    dec->argb_cache_ = NULL;    // for sanity check
1208    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
1209    return 0;
1210  }
1211  dec->argb_cache_ = dec->pixels_ + num_pixels + cache_top_pixels;
1212  return 1;
1213}
1214
1215static int AllocateInternalBuffers8b(VP8LDecoder* const dec) {
1216  const uint64_t total_num_pixels = (uint64_t)dec->width_ * dec->height_;
1217  dec->argb_cache_ = NULL;    // for sanity check
1218  dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint8_t));
1219  if (dec->pixels_ == NULL) {
1220    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
1221    return 0;
1222  }
1223  return 1;
1224}
1225
1226//------------------------------------------------------------------------------
1227
1228// Special row-processing that only stores the alpha data.
1229static void ExtractAlphaRows(VP8LDecoder* const dec, int row) {
1230  const int num_rows = row - dec->last_row_;
1231  const uint32_t* const in = dec->pixels_ + dec->width_ * dec->last_row_;
1232
1233  if (num_rows <= 0) return;  // Nothing to be done.
1234  ApplyInverseTransforms(dec, num_rows, in);
1235
1236  // Extract alpha (which is stored in the green plane).
1237  {
1238    const int width = dec->io_->width;      // the final width (!= dec->width_)
1239    const int cache_pixs = width * num_rows;
1240    uint8_t* const dst = (uint8_t*)dec->io_->opaque + width * dec->last_row_;
1241    const uint32_t* const src = dec->argb_cache_;
1242    int i;
1243    for (i = 0; i < cache_pixs; ++i) dst[i] = (src[i] >> 8) & 0xff;
1244  }
1245  dec->last_row_ = dec->last_out_row_ = row;
1246}
1247
1248int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,
1249                          const uint8_t* const data, size_t data_size,
1250                          uint8_t* const output) {
1251  int ok = 0;
1252  VP8LDecoder* dec;
1253  VP8Io* io;
1254  assert(alph_dec != NULL);
1255  alph_dec->vp8l_dec_ = VP8LNew();
1256  if (alph_dec->vp8l_dec_ == NULL) return 0;
1257  dec = alph_dec->vp8l_dec_;
1258
1259  dec->width_ = alph_dec->width_;
1260  dec->height_ = alph_dec->height_;
1261  dec->io_ = &alph_dec->io_;
1262  io = dec->io_;
1263
1264  VP8InitIo(io);
1265  WebPInitCustomIo(NULL, io);  // Just a sanity Init. io won't be used.
1266  io->opaque = output;
1267  io->width = alph_dec->width_;
1268  io->height = alph_dec->height_;
1269
1270  dec->status_ = VP8_STATUS_OK;
1271  VP8LInitBitReader(&dec->br_, data, data_size);
1272
1273  dec->action_ = READ_HDR;
1274  if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, 1, dec, NULL)) {
1275    goto Err;
1276  }
1277
1278  // Special case: if alpha data uses only the color indexing transform and
1279  // doesn't use color cache (a frequent case), we will use DecodeAlphaData()
1280  // method that only needs allocation of 1 byte per pixel (alpha channel).
1281  if (dec->next_transform_ == 1 &&
1282      dec->transforms_[0].type_ == COLOR_INDEXING_TRANSFORM &&
1283      Is8bOptimizable(&dec->hdr_)) {
1284    alph_dec->use_8b_decode = 1;
1285    ok = AllocateInternalBuffers8b(dec);
1286  } else {
1287    // Allocate internal buffers (note that dec->width_ may have changed here).
1288    alph_dec->use_8b_decode = 0;
1289    ok = AllocateInternalBuffers32b(dec, alph_dec->width_);
1290  }
1291
1292  if (!ok) goto Err;
1293
1294  dec->action_ = READ_DATA;
1295  return 1;
1296
1297 Err:
1298  VP8LDelete(alph_dec->vp8l_dec_);
1299  alph_dec->vp8l_dec_ = NULL;
1300  return 0;
1301}
1302
1303int VP8LDecodeAlphaImageStream(ALPHDecoder* const alph_dec, int last_row) {
1304  VP8LDecoder* const dec = alph_dec->vp8l_dec_;
1305  assert(dec != NULL);
1306  assert(dec->action_ == READ_DATA);
1307  assert(last_row <= dec->height_);
1308
1309  if (dec->last_pixel_ == dec->width_ * dec->height_) {
1310    return 1;  // done
1311  }
1312
1313  // Decode (with special row processing).
1314  return alph_dec->use_8b_decode ?
1315      DecodeAlphaData(dec, (uint8_t*)dec->pixels_, dec->width_, dec->height_,
1316                      last_row) :
1317      DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_,
1318                      last_row, ExtractAlphaRows);
1319}
1320
1321//------------------------------------------------------------------------------
1322
1323int VP8LDecodeHeader(VP8LDecoder* const dec, VP8Io* const io) {
1324  int width, height, has_alpha;
1325
1326  if (dec == NULL) return 0;
1327  if (io == NULL) {
1328    dec->status_ = VP8_STATUS_INVALID_PARAM;
1329    return 0;
1330  }
1331
1332  dec->io_ = io;
1333  dec->status_ = VP8_STATUS_OK;
1334  VP8LInitBitReader(&dec->br_, io->data, io->data_size);
1335  if (!ReadImageInfo(&dec->br_, &width, &height, &has_alpha)) {
1336    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
1337    goto Error;
1338  }
1339  dec->state_ = READ_DIM;
1340  io->width = width;
1341  io->height = height;
1342
1343  dec->action_ = READ_HDR;
1344  if (!DecodeImageStream(width, height, 1, dec, NULL)) goto Error;
1345  return 1;
1346
1347 Error:
1348  VP8LClear(dec);
1349  assert(dec->status_ != VP8_STATUS_OK);
1350  return 0;
1351}
1352
1353int VP8LDecodeImage(VP8LDecoder* const dec) {
1354  VP8Io* io = NULL;
1355  WebPDecParams* params = NULL;
1356
1357  // Sanity checks.
1358  if (dec == NULL) return 0;
1359
1360  dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
1361  assert(dec->hdr_.htree_groups_ != NULL);
1362  assert(dec->hdr_.num_htree_groups_ > 0);
1363
1364  io = dec->io_;
1365  assert(io != NULL);
1366  params = (WebPDecParams*)io->opaque;
1367  assert(params != NULL);
1368  dec->output_ = params->output;
1369  assert(dec->output_ != NULL);
1370
1371  // Initialization.
1372  if (!WebPIoInitFromOptions(params->options, io, MODE_BGRA)) {
1373    dec->status_ = VP8_STATUS_INVALID_PARAM;
1374    goto Err;
1375  }
1376
1377  if (!AllocateInternalBuffers32b(dec, io->width)) goto Err;
1378
1379  if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err;
1380
1381  if (io->use_scaling || WebPIsPremultipliedMode(dec->output_->colorspace)) {
1382    // need the alpha-multiply functions for premultiplied output or rescaling
1383    WebPInitAlphaProcessing();
1384  }
1385
1386  // Decode.
1387  dec->action_ = READ_DATA;
1388  if (!DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_,
1389                       dec->height_, ProcessRows)) {
1390    goto Err;
1391  }
1392
1393  // Cleanup.
1394  params->last_y = dec->last_out_row_;
1395  VP8LClear(dec);
1396  return 1;
1397
1398 Err:
1399  VP8LClear(dec);
1400  assert(dec->status_ != VP8_STATUS_OK);
1401  return 0;
1402}
1403
1404//------------------------------------------------------------------------------
1405