1 | // Copyright 2011 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 | // VP8Iterator: block iterator |
11 | // |
12 | // Author: Skal (pascal.massimino@gmail.com) |
13 | |
14 | #include <string.h> |
15 | |
16 | #include "./vp8i_enc.h" |
17 | |
18 | //------------------------------------------------------------------------------ |
19 | // VP8Iterator |
20 | //------------------------------------------------------------------------------ |
21 | |
22 | static void InitLeft(VP8EncIterator* const it) { |
23 | it->y_left_[-1] = it->u_left_[-1] = it->v_left_[-1] = |
24 | (it->y_ > 0) ? 129 : 127; |
25 | memset(it->y_left_, 129, 16); |
26 | memset(it->u_left_, 129, 8); |
27 | memset(it->v_left_, 129, 8); |
28 | it->left_nz_[8] = 0; |
29 | } |
30 | |
31 | static void InitTop(VP8EncIterator* const it) { |
32 | const VP8Encoder* const enc = it->enc_; |
33 | const size_t top_size = enc->mb_w_ * 16; |
34 | memset(enc->y_top_, 127, 2 * top_size); |
35 | memset(enc->nz_, 0, enc->mb_w_ * sizeof(*enc->nz_)); |
36 | } |
37 | |
38 | void VP8IteratorSetRow(VP8EncIterator* const it, int y) { |
39 | VP8Encoder* const enc = it->enc_; |
40 | it->x_ = 0; |
41 | it->y_ = y; |
42 | it->bw_ = &enc->parts_[y & (enc->num_parts_ - 1)]; |
43 | it->preds_ = enc->preds_ + y * 4 * enc->preds_w_; |
44 | it->nz_ = enc->nz_; |
45 | it->mb_ = enc->mb_info_ + y * enc->mb_w_; |
46 | it->y_top_ = enc->y_top_; |
47 | it->uv_top_ = enc->uv_top_; |
48 | InitLeft(it); |
49 | } |
50 | |
51 | void VP8IteratorReset(VP8EncIterator* const it) { |
52 | VP8Encoder* const enc = it->enc_; |
53 | VP8IteratorSetRow(it, 0); |
54 | VP8IteratorSetCountDown(it, enc->mb_w_ * enc->mb_h_); // default |
55 | InitTop(it); |
56 | memset(it->bit_count_, 0, sizeof(it->bit_count_)); |
57 | it->do_trellis_ = 0; |
58 | } |
59 | |
60 | void VP8IteratorSetCountDown(VP8EncIterator* const it, int count_down) { |
61 | it->count_down_ = it->count_down0_ = count_down; |
62 | } |
63 | |
64 | int VP8IteratorIsDone(const VP8EncIterator* const it) { |
65 | return (it->count_down_ <= 0); |
66 | } |
67 | |
68 | void VP8IteratorInit(VP8Encoder* const enc, VP8EncIterator* const it) { |
69 | it->enc_ = enc; |
70 | it->yuv_in_ = (uint8_t*)WEBP_ALIGN(it->yuv_mem_); |
71 | it->yuv_out_ = it->yuv_in_ + YUV_SIZE_ENC; |
72 | it->yuv_out2_ = it->yuv_out_ + YUV_SIZE_ENC; |
73 | it->yuv_p_ = it->yuv_out2_ + YUV_SIZE_ENC; |
74 | it->lf_stats_ = enc->lf_stats_; |
75 | it->percent0_ = enc->percent_; |
76 | it->y_left_ = (uint8_t*)WEBP_ALIGN(it->yuv_left_mem_ + 1); |
77 | it->u_left_ = it->y_left_ + 16 + 16; |
78 | it->v_left_ = it->u_left_ + 16; |
79 | VP8IteratorReset(it); |
80 | } |
81 | |
82 | int VP8IteratorProgress(const VP8EncIterator* const it, int delta) { |
83 | VP8Encoder* const enc = it->enc_; |
84 | if (delta && enc->pic_->progress_hook != NULL) { |
85 | const int done = it->count_down0_ - it->count_down_; |
86 | const int percent = (it->count_down0_ <= 0) |
87 | ? it->percent0_ |
88 | : it->percent0_ + delta * done / it->count_down0_; |
89 | return WebPReportProgress(enc->pic_, percent, &enc->percent_); |
90 | } |
91 | return 1; |
92 | } |
93 | |
94 | //------------------------------------------------------------------------------ |
95 | // Import the source samples into the cache. Takes care of replicating |
96 | // boundary pixels if necessary. |
97 | |
98 | static WEBP_INLINE int MinSize(int a, int b) { return (a < b) ? a : b; } |
99 | |
100 | static void ImportBlock(const uint8_t* src, int src_stride, |
101 | uint8_t* dst, int w, int h, int size) { |
102 | int i; |
103 | for (i = 0; i < h; ++i) { |
104 | memcpy(dst, src, w); |
105 | if (w < size) { |
106 | memset(dst + w, dst[w - 1], size - w); |
107 | } |
108 | dst += BPS; |
109 | src += src_stride; |
110 | } |
111 | for (i = h; i < size; ++i) { |
112 | memcpy(dst, dst - BPS, size); |
113 | dst += BPS; |
114 | } |
115 | } |
116 | |
117 | static void ImportLine(const uint8_t* src, int src_stride, |
118 | uint8_t* dst, int len, int total_len) { |
119 | int i; |
120 | for (i = 0; i < len; ++i, src += src_stride) dst[i] = *src; |
121 | for (; i < total_len; ++i) dst[i] = dst[len - 1]; |
122 | } |
123 | |
124 | void VP8IteratorImport(VP8EncIterator* const it, uint8_t* tmp_32) { |
125 | const VP8Encoder* const enc = it->enc_; |
126 | const int x = it->x_, y = it->y_; |
127 | const WebPPicture* const pic = enc->pic_; |
128 | const uint8_t* const ysrc = pic->y + (y * pic->y_stride + x) * 16; |
129 | const uint8_t* const usrc = pic->u + (y * pic->uv_stride + x) * 8; |
130 | const uint8_t* const vsrc = pic->v + (y * pic->uv_stride + x) * 8; |
131 | const int w = MinSize(pic->width - x * 16, 16); |
132 | const int h = MinSize(pic->height - y * 16, 16); |
133 | const int uv_w = (w + 1) >> 1; |
134 | const int uv_h = (h + 1) >> 1; |
135 | |
136 | ImportBlock(ysrc, pic->y_stride, it->yuv_in_ + Y_OFF_ENC, w, h, 16); |
137 | ImportBlock(usrc, pic->uv_stride, it->yuv_in_ + U_OFF_ENC, uv_w, uv_h, 8); |
138 | ImportBlock(vsrc, pic->uv_stride, it->yuv_in_ + V_OFF_ENC, uv_w, uv_h, 8); |
139 | |
140 | if (tmp_32 == NULL) return; |
141 | |
142 | // Import source (uncompressed) samples into boundary. |
143 | if (x == 0) { |
144 | InitLeft(it); |
145 | } else { |
146 | if (y == 0) { |
147 | it->y_left_[-1] = it->u_left_[-1] = it->v_left_[-1] = 127; |
148 | } else { |
149 | it->y_left_[-1] = ysrc[- 1 - pic->y_stride]; |
150 | it->u_left_[-1] = usrc[- 1 - pic->uv_stride]; |
151 | it->v_left_[-1] = vsrc[- 1 - pic->uv_stride]; |
152 | } |
153 | ImportLine(ysrc - 1, pic->y_stride, it->y_left_, h, 16); |
154 | ImportLine(usrc - 1, pic->uv_stride, it->u_left_, uv_h, 8); |
155 | ImportLine(vsrc - 1, pic->uv_stride, it->v_left_, uv_h, 8); |
156 | } |
157 | |
158 | it->y_top_ = tmp_32 + 0; |
159 | it->uv_top_ = tmp_32 + 16; |
160 | if (y == 0) { |
161 | memset(tmp_32, 127, 32 * sizeof(*tmp_32)); |
162 | } else { |
163 | ImportLine(ysrc - pic->y_stride, 1, tmp_32, w, 16); |
164 | ImportLine(usrc - pic->uv_stride, 1, tmp_32 + 16, uv_w, 8); |
165 | ImportLine(vsrc - pic->uv_stride, 1, tmp_32 + 16 + 8, uv_w, 8); |
166 | } |
167 | } |
168 | |
169 | //------------------------------------------------------------------------------ |
170 | // Copy back the compressed samples into user space if requested. |
171 | |
172 | static void ExportBlock(const uint8_t* src, uint8_t* dst, int dst_stride, |
173 | int w, int h) { |
174 | while (h-- > 0) { |
175 | memcpy(dst, src, w); |
176 | dst += dst_stride; |
177 | src += BPS; |
178 | } |
179 | } |
180 | |
181 | void VP8IteratorExport(const VP8EncIterator* const it) { |
182 | const VP8Encoder* const enc = it->enc_; |
183 | if (enc->config_->show_compressed) { |
184 | const int x = it->x_, y = it->y_; |
185 | const uint8_t* const ysrc = it->yuv_out_ + Y_OFF_ENC; |
186 | const uint8_t* const usrc = it->yuv_out_ + U_OFF_ENC; |
187 | const uint8_t* const vsrc = it->yuv_out_ + V_OFF_ENC; |
188 | const WebPPicture* const pic = enc->pic_; |
189 | uint8_t* const ydst = pic->y + (y * pic->y_stride + x) * 16; |
190 | uint8_t* const udst = pic->u + (y * pic->uv_stride + x) * 8; |
191 | uint8_t* const vdst = pic->v + (y * pic->uv_stride + x) * 8; |
192 | int w = (pic->width - x * 16); |
193 | int h = (pic->height - y * 16); |
194 | |
195 | if (w > 16) w = 16; |
196 | if (h > 16) h = 16; |
197 | |
198 | // Luma plane |
199 | ExportBlock(ysrc, ydst, pic->y_stride, w, h); |
200 | |
201 | { // U/V planes |
202 | const int uv_w = (w + 1) >> 1; |
203 | const int uv_h = (h + 1) >> 1; |
204 | ExportBlock(usrc, udst, pic->uv_stride, uv_w, uv_h); |
205 | ExportBlock(vsrc, vdst, pic->uv_stride, uv_w, uv_h); |
206 | } |
207 | } |
208 | } |
209 | |
210 | //------------------------------------------------------------------------------ |
211 | // Non-zero contexts setup/teardown |
212 | |
213 | // Nz bits: |
214 | // 0 1 2 3 Y |
215 | // 4 5 6 7 |
216 | // 8 9 10 11 |
217 | // 12 13 14 15 |
218 | // 16 17 U |
219 | // 18 19 |
220 | // 20 21 V |
221 | // 22 23 |
222 | // 24 DC-intra16 |
223 | |
224 | // Convert packed context to byte array |
225 | #define BIT(nz, n) (!!((nz) & (1 << (n)))) |
226 | |
227 | void VP8IteratorNzToBytes(VP8EncIterator* const it) { |
228 | const int tnz = it->nz_[0], lnz = it->nz_[-1]; |
229 | int* const top_nz = it->top_nz_; |
230 | int* const left_nz = it->left_nz_; |
231 | |
232 | // Top-Y |
233 | top_nz[0] = BIT(tnz, 12); |
234 | top_nz[1] = BIT(tnz, 13); |
235 | top_nz[2] = BIT(tnz, 14); |
236 | top_nz[3] = BIT(tnz, 15); |
237 | // Top-U |
238 | top_nz[4] = BIT(tnz, 18); |
239 | top_nz[5] = BIT(tnz, 19); |
240 | // Top-V |
241 | top_nz[6] = BIT(tnz, 22); |
242 | top_nz[7] = BIT(tnz, 23); |
243 | // DC |
244 | top_nz[8] = BIT(tnz, 24); |
245 | |
246 | // left-Y |
247 | left_nz[0] = BIT(lnz, 3); |
248 | left_nz[1] = BIT(lnz, 7); |
249 | left_nz[2] = BIT(lnz, 11); |
250 | left_nz[3] = BIT(lnz, 15); |
251 | // left-U |
252 | left_nz[4] = BIT(lnz, 17); |
253 | left_nz[5] = BIT(lnz, 19); |
254 | // left-V |
255 | left_nz[6] = BIT(lnz, 21); |
256 | left_nz[7] = BIT(lnz, 23); |
257 | // left-DC is special, iterated separately |
258 | } |
259 | |
260 | void VP8IteratorBytesToNz(VP8EncIterator* const it) { |
261 | uint32_t nz = 0; |
262 | const int* const top_nz = it->top_nz_; |
263 | const int* const left_nz = it->left_nz_; |
264 | // top |
265 | nz |= (top_nz[0] << 12) | (top_nz[1] << 13); |
266 | nz |= (top_nz[2] << 14) | (top_nz[3] << 15); |
267 | nz |= (top_nz[4] << 18) | (top_nz[5] << 19); |
268 | nz |= (top_nz[6] << 22) | (top_nz[7] << 23); |
269 | nz |= (top_nz[8] << 24); // we propagate the _top_ bit, esp. for intra4 |
270 | // left |
271 | nz |= (left_nz[0] << 3) | (left_nz[1] << 7); |
272 | nz |= (left_nz[2] << 11); |
273 | nz |= (left_nz[4] << 17) | (left_nz[6] << 21); |
274 | |
275 | *it->nz_ = nz; |
276 | } |
277 | |
278 | #undef BIT |
279 | |
280 | //------------------------------------------------------------------------------ |
281 | // Advance to the next position, doing the bookkeeping. |
282 | |
283 | void VP8IteratorSaveBoundary(VP8EncIterator* const it) { |
284 | VP8Encoder* const enc = it->enc_; |
285 | const int x = it->x_, y = it->y_; |
286 | const uint8_t* const ysrc = it->yuv_out_ + Y_OFF_ENC; |
287 | const uint8_t* const uvsrc = it->yuv_out_ + U_OFF_ENC; |
288 | if (x < enc->mb_w_ - 1) { // left |
289 | int i; |
290 | for (i = 0; i < 16; ++i) { |
291 | it->y_left_[i] = ysrc[15 + i * BPS]; |
292 | } |
293 | for (i = 0; i < 8; ++i) { |
294 | it->u_left_[i] = uvsrc[7 + i * BPS]; |
295 | it->v_left_[i] = uvsrc[15 + i * BPS]; |
296 | } |
297 | // top-left (before 'top'!) |
298 | it->y_left_[-1] = it->y_top_[15]; |
299 | it->u_left_[-1] = it->uv_top_[0 + 7]; |
300 | it->v_left_[-1] = it->uv_top_[8 + 7]; |
301 | } |
302 | if (y < enc->mb_h_ - 1) { // top |
303 | memcpy(it->y_top_, ysrc + 15 * BPS, 16); |
304 | memcpy(it->uv_top_, uvsrc + 7 * BPS, 8 + 8); |
305 | } |
306 | } |
307 | |
308 | int VP8IteratorNext(VP8EncIterator* const it) { |
309 | if (++it->x_ == it->enc_->mb_w_) { |
310 | VP8IteratorSetRow(it, ++it->y_); |
311 | } else { |
312 | it->preds_ += 4; |
313 | it->mb_ += 1; |
314 | it->nz_ += 1; |
315 | it->y_top_ += 16; |
316 | it->uv_top_ += 16; |
317 | } |
318 | return (0 < --it->count_down_); |
319 | } |
320 | |
321 | //------------------------------------------------------------------------------ |
322 | // Helper function to set mode properties |
323 | |
324 | void VP8SetIntra16Mode(const VP8EncIterator* const it, int mode) { |
325 | uint8_t* preds = it->preds_; |
326 | int y; |
327 | for (y = 0; y < 4; ++y) { |
328 | memset(preds, mode, 4); |
329 | preds += it->enc_->preds_w_; |
330 | } |
331 | it->mb_->type_ = 1; |
332 | } |
333 | |
334 | void VP8SetIntra4Mode(const VP8EncIterator* const it, const uint8_t* modes) { |
335 | uint8_t* preds = it->preds_; |
336 | int y; |
337 | for (y = 4; y > 0; --y) { |
338 | memcpy(preds, modes, 4 * sizeof(*modes)); |
339 | preds += it->enc_->preds_w_; |
340 | modes += 4; |
341 | } |
342 | it->mb_->type_ = 0; |
343 | } |
344 | |
345 | void VP8SetIntraUVMode(const VP8EncIterator* const it, int mode) { |
346 | it->mb_->uv_mode_ = mode; |
347 | } |
348 | |
349 | void VP8SetSkip(const VP8EncIterator* const it, int skip) { |
350 | it->mb_->skip_ = skip; |
351 | } |
352 | |
353 | void VP8SetSegment(const VP8EncIterator* const it, int segment) { |
354 | it->mb_->segment_ = segment; |
355 | } |
356 | |
357 | //------------------------------------------------------------------------------ |
358 | // Intra4x4 sub-blocks iteration |
359 | // |
360 | // We store and update the boundary samples into an array of 37 pixels. They |
361 | // are updated as we iterate and reconstructs each intra4x4 blocks in turn. |
362 | // The position of the samples has the following snake pattern: |
363 | // |
364 | // 16|17 18 19 20|21 22 23 24|25 26 27 28|29 30 31 32|33 34 35 36 <- Top-right |
365 | // --+-----------+-----------+-----------+-----------+ |
366 | // 15| 19| 23| 27| 31| |
367 | // 14| 18| 22| 26| 30| |
368 | // 13| 17| 21| 25| 29| |
369 | // 12|13 14 15 16|17 18 19 20|21 22 23 24|25 26 27 28| |
370 | // --+-----------+-----------+-----------+-----------+ |
371 | // 11| 15| 19| 23| 27| |
372 | // 10| 14| 18| 22| 26| |
373 | // 9| 13| 17| 21| 25| |
374 | // 8| 9 10 11 12|13 14 15 16|17 18 19 20|21 22 23 24| |
375 | // --+-----------+-----------+-----------+-----------+ |
376 | // 7| 11| 15| 19| 23| |
377 | // 6| 10| 14| 18| 22| |
378 | // 5| 9| 13| 17| 21| |
379 | // 4| 5 6 7 8| 9 10 11 12|13 14 15 16|17 18 19 20| |
380 | // --+-----------+-----------+-----------+-----------+ |
381 | // 3| 7| 11| 15| 19| |
382 | // 2| 6| 10| 14| 18| |
383 | // 1| 5| 9| 13| 17| |
384 | // 0| 1 2 3 4| 5 6 7 8| 9 10 11 12|13 14 15 16| |
385 | // --+-----------+-----------+-----------+-----------+ |
386 | |
387 | // Array to record the position of the top sample to pass to the prediction |
388 | // functions in dsp.c. |
389 | static const uint8_t VP8TopLeftI4[16] = { |
390 | 17, 21, 25, 29, |
391 | 13, 17, 21, 25, |
392 | 9, 13, 17, 21, |
393 | 5, 9, 13, 17 |
394 | }; |
395 | |
396 | void VP8IteratorStartI4(VP8EncIterator* const it) { |
397 | const VP8Encoder* const enc = it->enc_; |
398 | int i; |
399 | |
400 | it->i4_ = 0; // first 4x4 sub-block |
401 | it->i4_top_ = it->i4_boundary_ + VP8TopLeftI4[0]; |
402 | |
403 | // Import the boundary samples |
404 | for (i = 0; i < 17; ++i) { // left |
405 | it->i4_boundary_[i] = it->y_left_[15 - i]; |
406 | } |
407 | for (i = 0; i < 16; ++i) { // top |
408 | it->i4_boundary_[17 + i] = it->y_top_[i]; |
409 | } |
410 | // top-right samples have a special case on the far right of the picture |
411 | if (it->x_ < enc->mb_w_ - 1) { |
412 | for (i = 16; i < 16 + 4; ++i) { |
413 | it->i4_boundary_[17 + i] = it->y_top_[i]; |
414 | } |
415 | } else { // else, replicate the last valid pixel four times |
416 | for (i = 16; i < 16 + 4; ++i) { |
417 | it->i4_boundary_[17 + i] = it->i4_boundary_[17 + 15]; |
418 | } |
419 | } |
420 | VP8IteratorNzToBytes(it); // import the non-zero context |
421 | } |
422 | |
423 | int VP8IteratorRotateI4(VP8EncIterator* const it, |
424 | const uint8_t* const yuv_out) { |
425 | const uint8_t* const blk = yuv_out + VP8Scan[it->i4_]; |
426 | uint8_t* const top = it->i4_top_; |
427 | int i; |
428 | |
429 | // Update the cache with 7 fresh samples |
430 | for (i = 0; i <= 3; ++i) { |
431 | top[-4 + i] = blk[i + 3 * BPS]; // store future top samples |
432 | } |
433 | if ((it->i4_ & 3) != 3) { // if not on the right sub-blocks #3, #7, #11, #15 |
434 | for (i = 0; i <= 2; ++i) { // store future left samples |
435 | top[i] = blk[3 + (2 - i) * BPS]; |
436 | } |
437 | } else { // else replicate top-right samples, as says the specs. |
438 | for (i = 0; i <= 3; ++i) { |
439 | top[i] = top[i + 4]; |
440 | } |
441 | } |
442 | // move pointers to next sub-block |
443 | ++it->i4_; |
444 | if (it->i4_ == 16) { // we're done |
445 | return 0; |
446 | } |
447 | |
448 | it->i4_top_ = it->i4_boundary_ + VP8TopLeftI4[it->i4_]; |
449 | return 1; |
450 | } |
451 | |
452 | //------------------------------------------------------------------------------ |
453 | |
454 | |