1 /*
2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <math.h>
12 #include <stdlib.h>
13 #include <stdio.h>
14
15 #include "./vpx_config.h"
16 #include "./vpx_scale_rtcd.h"
17 #include "./vp9_rtcd.h"
18
19 #include "vpx_dsp/vpx_dsp_common.h"
20 #include "vpx_ports/mem.h"
21 #include "vpx_ports/system_state.h"
22 #include "vpx_scale/vpx_scale.h"
23 #include "vpx_scale/yv12config.h"
24
25 #include "vp9/common/vp9_onyxc_int.h"
26 #include "vp9/common/vp9_postproc.h"
27 #include "vp9/common/vp9_textblit.h"
28
29 #if CONFIG_VP9_POSTPROC
30 static const int16_t kernel5[] = {
31 1, 1, 4, 1, 1
32 };
33
34 const int16_t vp9_rv[] = {
35 8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
36 0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
37 10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
38 8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
39 8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
40 1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
41 3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
42 11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
43 14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
44 4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
45 7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
46 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
47 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
48 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
49 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
50 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
51 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
52 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
53 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
54 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
55 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
56 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
57 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
58 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
59 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
60 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
61 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
62 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
63 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
64 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
65 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
66 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
67 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
68 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
69 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
70 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
71 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
72 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
73 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
74 3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
75 11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
76 14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
77 5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
78 0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
79 };
80
81 static const uint8_t q_diff_thresh = 20;
82 static const uint8_t last_q_thresh = 170;
83
vp9_post_proc_down_and_across_c(const uint8_t * src_ptr,uint8_t * dst_ptr,int src_pixels_per_line,int dst_pixels_per_line,int rows,int cols,int flimit)84 void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
85 uint8_t *dst_ptr,
86 int src_pixels_per_line,
87 int dst_pixels_per_line,
88 int rows,
89 int cols,
90 int flimit) {
91 uint8_t const *p_src;
92 uint8_t *p_dst;
93 int row, col, i, v, kernel;
94 int pitch = src_pixels_per_line;
95 uint8_t d[8];
96 (void)dst_pixels_per_line;
97
98 for (row = 0; row < rows; row++) {
99 /* post_proc_down for one row */
100 p_src = src_ptr;
101 p_dst = dst_ptr;
102
103 for (col = 0; col < cols; col++) {
104 kernel = 4;
105 v = p_src[col];
106
107 for (i = -2; i <= 2; i++) {
108 if (abs(v - p_src[col + i * pitch]) > flimit)
109 goto down_skip_convolve;
110
111 kernel += kernel5[2 + i] * p_src[col + i * pitch];
112 }
113
114 v = (kernel >> 3);
115 down_skip_convolve:
116 p_dst[col] = v;
117 }
118
119 /* now post_proc_across */
120 p_src = dst_ptr;
121 p_dst = dst_ptr;
122
123 for (i = 0; i < 8; i++)
124 d[i] = p_src[i];
125
126 for (col = 0; col < cols; col++) {
127 kernel = 4;
128 v = p_src[col];
129
130 d[col & 7] = v;
131
132 for (i = -2; i <= 2; i++) {
133 if (abs(v - p_src[col + i]) > flimit)
134 goto across_skip_convolve;
135
136 kernel += kernel5[2 + i] * p_src[col + i];
137 }
138
139 d[col & 7] = (kernel >> 3);
140 across_skip_convolve:
141
142 if (col >= 2)
143 p_dst[col - 2] = d[(col - 2) & 7];
144 }
145
146 /* handle the last two pixels */
147 p_dst[col - 2] = d[(col - 2) & 7];
148 p_dst[col - 1] = d[(col - 1) & 7];
149
150
151 /* next row */
152 src_ptr += pitch;
153 dst_ptr += pitch;
154 }
155 }
156
157 #if CONFIG_VP9_HIGHBITDEPTH
vp9_highbd_post_proc_down_and_across_c(const uint16_t * src_ptr,uint16_t * dst_ptr,int src_pixels_per_line,int dst_pixels_per_line,int rows,int cols,int flimit)158 void vp9_highbd_post_proc_down_and_across_c(const uint16_t *src_ptr,
159 uint16_t *dst_ptr,
160 int src_pixels_per_line,
161 int dst_pixels_per_line,
162 int rows,
163 int cols,
164 int flimit) {
165 uint16_t const *p_src;
166 uint16_t *p_dst;
167 int row, col, i, v, kernel;
168 int pitch = src_pixels_per_line;
169 uint16_t d[8];
170
171 for (row = 0; row < rows; row++) {
172 // post_proc_down for one row.
173 p_src = src_ptr;
174 p_dst = dst_ptr;
175
176 for (col = 0; col < cols; col++) {
177 kernel = 4;
178 v = p_src[col];
179
180 for (i = -2; i <= 2; i++) {
181 if (abs(v - p_src[col + i * pitch]) > flimit)
182 goto down_skip_convolve;
183
184 kernel += kernel5[2 + i] * p_src[col + i * pitch];
185 }
186
187 v = (kernel >> 3);
188
189 down_skip_convolve:
190 p_dst[col] = v;
191 }
192
193 /* now post_proc_across */
194 p_src = dst_ptr;
195 p_dst = dst_ptr;
196
197 for (i = 0; i < 8; i++)
198 d[i] = p_src[i];
199
200 for (col = 0; col < cols; col++) {
201 kernel = 4;
202 v = p_src[col];
203
204 d[col & 7] = v;
205
206 for (i = -2; i <= 2; i++) {
207 if (abs(v - p_src[col + i]) > flimit)
208 goto across_skip_convolve;
209
210 kernel += kernel5[2 + i] * p_src[col + i];
211 }
212
213 d[col & 7] = (kernel >> 3);
214
215 across_skip_convolve:
216 if (col >= 2)
217 p_dst[col - 2] = d[(col - 2) & 7];
218 }
219
220 /* handle the last two pixels */
221 p_dst[col - 2] = d[(col - 2) & 7];
222 p_dst[col - 1] = d[(col - 1) & 7];
223
224
225 /* next row */
226 src_ptr += pitch;
227 dst_ptr += dst_pixels_per_line;
228 }
229 }
230 #endif // CONFIG_VP9_HIGHBITDEPTH
231
q2mbl(int x)232 static int q2mbl(int x) {
233 if (x < 20) x = 20;
234
235 x = 50 + (x - 50) * 10 / 8;
236 return x * x / 3;
237 }
238
vp9_mbpost_proc_across_ip_c(uint8_t * src,int pitch,int rows,int cols,int flimit)239 void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
240 int rows, int cols, int flimit) {
241 int r, c, i;
242 uint8_t *s = src;
243 uint8_t d[16];
244
245 for (r = 0; r < rows; r++) {
246 int sumsq = 0;
247 int sum = 0;
248
249 for (i = -8; i <= 6; i++) {
250 sumsq += s[i] * s[i];
251 sum += s[i];
252 d[i + 8] = 0;
253 }
254
255 for (c = 0; c < cols + 8; c++) {
256 int x = s[c + 7] - s[c - 8];
257 int y = s[c + 7] + s[c - 8];
258
259 sum += x;
260 sumsq += x * y;
261
262 d[c & 15] = s[c];
263
264 if (sumsq * 15 - sum * sum < flimit) {
265 d[c & 15] = (8 + sum + s[c]) >> 4;
266 }
267
268 s[c - 8] = d[(c - 8) & 15];
269 }
270 s += pitch;
271 }
272 }
273
274 #if CONFIG_VP9_HIGHBITDEPTH
vp9_highbd_mbpost_proc_across_ip_c(uint16_t * src,int pitch,int rows,int cols,int flimit)275 void vp9_highbd_mbpost_proc_across_ip_c(uint16_t *src, int pitch,
276 int rows, int cols, int flimit) {
277 int r, c, i;
278
279 uint16_t *s = src;
280 uint16_t d[16];
281
282
283 for (r = 0; r < rows; r++) {
284 int sumsq = 0;
285 int sum = 0;
286
287 for (i = -8; i <= 6; i++) {
288 sumsq += s[i] * s[i];
289 sum += s[i];
290 d[i + 8] = 0;
291 }
292
293 for (c = 0; c < cols + 8; c++) {
294 int x = s[c + 7] - s[c - 8];
295 int y = s[c + 7] + s[c - 8];
296
297 sum += x;
298 sumsq += x * y;
299
300 d[c & 15] = s[c];
301
302 if (sumsq * 15 - sum * sum < flimit) {
303 d[c & 15] = (8 + sum + s[c]) >> 4;
304 }
305
306 s[c - 8] = d[(c - 8) & 15];
307 }
308
309 s += pitch;
310 }
311 }
312 #endif // CONFIG_VP9_HIGHBITDEPTH
313
vp9_mbpost_proc_down_c(uint8_t * dst,int pitch,int rows,int cols,int flimit)314 void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
315 int rows, int cols, int flimit) {
316 int r, c, i;
317 const short *rv3 = &vp9_rv[63 & rand()]; // NOLINT
318
319 for (c = 0; c < cols; c++) {
320 uint8_t *s = &dst[c];
321 int sumsq = 0;
322 int sum = 0;
323 uint8_t d[16];
324 const int16_t *rv2 = rv3 + ((c * 17) & 127);
325
326 for (i = -8; i <= 6; i++) {
327 sumsq += s[i * pitch] * s[i * pitch];
328 sum += s[i * pitch];
329 }
330
331 for (r = 0; r < rows + 8; r++) {
332 sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
333 sum += s[7 * pitch] - s[-8 * pitch];
334 d[r & 15] = s[0];
335
336 if (sumsq * 15 - sum * sum < flimit) {
337 d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
338 }
339
340 s[-8 * pitch] = d[(r - 8) & 15];
341 s += pitch;
342 }
343 }
344 }
345
346 #if CONFIG_VP9_HIGHBITDEPTH
vp9_highbd_mbpost_proc_down_c(uint16_t * dst,int pitch,int rows,int cols,int flimit)347 void vp9_highbd_mbpost_proc_down_c(uint16_t *dst, int pitch,
348 int rows, int cols, int flimit) {
349 int r, c, i;
350 const int16_t *rv3 = &vp9_rv[63 & rand()]; // NOLINT
351
352 for (c = 0; c < cols; c++) {
353 uint16_t *s = &dst[c];
354 int sumsq = 0;
355 int sum = 0;
356 uint16_t d[16];
357 const int16_t *rv2 = rv3 + ((c * 17) & 127);
358
359 for (i = -8; i <= 6; i++) {
360 sumsq += s[i * pitch] * s[i * pitch];
361 sum += s[i * pitch];
362 }
363
364 for (r = 0; r < rows + 8; r++) {
365 sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
366 sum += s[7 * pitch] - s[-8 * pitch];
367 d[r & 15] = s[0];
368
369 if (sumsq * 15 - sum * sum < flimit) {
370 d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
371 }
372
373 s[-8 * pitch] = d[(r - 8) & 15];
374 s += pitch;
375 }
376 }
377 }
378 #endif // CONFIG_VP9_HIGHBITDEPTH
379
deblock_and_de_macro_block(YV12_BUFFER_CONFIG * source,YV12_BUFFER_CONFIG * post,int q,int low_var_thresh,int flag)380 static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG *source,
381 YV12_BUFFER_CONFIG *post,
382 int q,
383 int low_var_thresh,
384 int flag) {
385 double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
386 int ppl = (int)(level + .5);
387 (void) low_var_thresh;
388 (void) flag;
389
390 #if CONFIG_VP9_HIGHBITDEPTH
391 if (source->flags & YV12_FLAG_HIGHBITDEPTH) {
392 vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->y_buffer),
393 CONVERT_TO_SHORTPTR(post->y_buffer),
394 source->y_stride, post->y_stride,
395 source->y_height, source->y_width,
396 ppl);
397
398 vp9_highbd_mbpost_proc_across_ip(CONVERT_TO_SHORTPTR(post->y_buffer),
399 post->y_stride, post->y_height,
400 post->y_width, q2mbl(q));
401
402 vp9_highbd_mbpost_proc_down(CONVERT_TO_SHORTPTR(post->y_buffer),
403 post->y_stride, post->y_height,
404 post->y_width, q2mbl(q));
405
406 vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->u_buffer),
407 CONVERT_TO_SHORTPTR(post->u_buffer),
408 source->uv_stride, post->uv_stride,
409 source->uv_height, source->uv_width,
410 ppl);
411 vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->v_buffer),
412 CONVERT_TO_SHORTPTR(post->v_buffer),
413 source->uv_stride, post->uv_stride,
414 source->uv_height, source->uv_width,
415 ppl);
416 } else {
417 vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
418 source->y_stride, post->y_stride,
419 source->y_height, source->y_width, ppl);
420
421 vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
422 post->y_width, q2mbl(q));
423
424 vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
425 post->y_width, q2mbl(q));
426
427 vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
428 source->uv_stride, post->uv_stride,
429 source->uv_height, source->uv_width, ppl);
430 vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
431 source->uv_stride, post->uv_stride,
432 source->uv_height, source->uv_width, ppl);
433 }
434 #else
435 vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
436 source->y_stride, post->y_stride,
437 source->y_height, source->y_width, ppl);
438
439 vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
440 post->y_width, q2mbl(q));
441
442 vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
443 post->y_width, q2mbl(q));
444
445 vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
446 source->uv_stride, post->uv_stride,
447 source->uv_height, source->uv_width, ppl);
448 vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
449 source->uv_stride, post->uv_stride,
450 source->uv_height, source->uv_width, ppl);
451 #endif // CONFIG_VP9_HIGHBITDEPTH
452 }
453
vp9_deblock(const YV12_BUFFER_CONFIG * src,YV12_BUFFER_CONFIG * dst,int q)454 void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
455 int q) {
456 const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
457 + 0.0065 + 0.5);
458 int i;
459
460 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
461 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
462 const int src_widths[3] = {src->y_width, src->uv_width, src->uv_width};
463 const int src_heights[3] = {src->y_height, src->uv_height, src->uv_height};
464
465 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
466 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
467
468 for (i = 0; i < MAX_MB_PLANE; ++i) {
469 #if CONFIG_VP9_HIGHBITDEPTH
470 assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
471 (dst->flags & YV12_FLAG_HIGHBITDEPTH));
472 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
473 vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(srcs[i]),
474 CONVERT_TO_SHORTPTR(dsts[i]),
475 src_strides[i], dst_strides[i],
476 src_heights[i], src_widths[i], ppl);
477 } else {
478 vp9_post_proc_down_and_across(srcs[i], dsts[i],
479 src_strides[i], dst_strides[i],
480 src_heights[i], src_widths[i], ppl);
481 }
482 #else
483 vp9_post_proc_down_and_across(srcs[i], dsts[i],
484 src_strides[i], dst_strides[i],
485 src_heights[i], src_widths[i], ppl);
486 #endif // CONFIG_VP9_HIGHBITDEPTH
487 }
488 }
489
vp9_denoise(const YV12_BUFFER_CONFIG * src,YV12_BUFFER_CONFIG * dst,int q)490 void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
491 int q) {
492 const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
493 + 0.0065 + 0.5);
494 int i;
495
496 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
497 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
498 const int src_widths[3] = {src->y_width, src->uv_width, src->uv_width};
499 const int src_heights[3] = {src->y_height, src->uv_height, src->uv_height};
500
501 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
502 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
503
504 for (i = 0; i < MAX_MB_PLANE; ++i) {
505 const int src_stride = src_strides[i];
506 const int src_width = src_widths[i] - 4;
507 const int src_height = src_heights[i] - 4;
508 const int dst_stride = dst_strides[i];
509
510 #if CONFIG_VP9_HIGHBITDEPTH
511 assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
512 (dst->flags & YV12_FLAG_HIGHBITDEPTH));
513 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
514 const uint16_t *const src_plane = CONVERT_TO_SHORTPTR(
515 srcs[i] + 2 * src_stride + 2);
516 uint16_t *const dst_plane = CONVERT_TO_SHORTPTR(
517 dsts[i] + 2 * dst_stride + 2);
518 vp9_highbd_post_proc_down_and_across(src_plane, dst_plane, src_stride,
519 dst_stride, src_height, src_width,
520 ppl);
521 } else {
522 const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
523 uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
524
525 vp9_post_proc_down_and_across(src_plane, dst_plane, src_stride,
526 dst_stride, src_height, src_width, ppl);
527 }
528 #else
529 const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
530 uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
531 vp9_post_proc_down_and_across(src_plane, dst_plane, src_stride, dst_stride,
532 src_height, src_width, ppl);
533 #endif
534 }
535 }
536
gaussian(double sigma,double mu,double x)537 static double gaussian(double sigma, double mu, double x) {
538 return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
539 (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
540 }
541
fillrd(struct postproc_state * state,int q,int a)542 static void fillrd(struct postproc_state *state, int q, int a) {
543 char char_dist[300];
544
545 double sigma;
546 int ai = a, qi = q, i;
547
548 vpx_clear_system_state();
549
550 sigma = ai + .5 + .6 * (63 - qi) / 63.0;
551
552 /* set up a lookup table of 256 entries that matches
553 * a gaussian distribution with sigma determined by q.
554 */
555 {
556 int next, j;
557
558 next = 0;
559
560 for (i = -32; i < 32; i++) {
561 int a_i = (int)(0.5 + 256 * gaussian(sigma, 0, i));
562
563 if (a_i) {
564 for (j = 0; j < a_i; j++) {
565 char_dist[next + j] = (char) i;
566 }
567
568 next = next + j;
569 }
570 }
571
572 for (; next < 256; next++)
573 char_dist[next] = 0;
574 }
575
576 for (i = 0; i < 3072; i++) {
577 state->noise[i] = char_dist[rand() & 0xff]; // NOLINT
578 }
579
580 for (i = 0; i < 16; i++) {
581 state->blackclamp[i] = -char_dist[0];
582 state->whiteclamp[i] = -char_dist[0];
583 state->bothclamp[i] = -2 * char_dist[0];
584 }
585
586 state->last_q = q;
587 state->last_noise = a;
588 }
589
vp9_plane_add_noise_c(uint8_t * start,char * noise,char blackclamp[16],char whiteclamp[16],char bothclamp[16],unsigned int width,unsigned int height,int pitch)590 void vp9_plane_add_noise_c(uint8_t *start, char *noise,
591 char blackclamp[16],
592 char whiteclamp[16],
593 char bothclamp[16],
594 unsigned int width, unsigned int height, int pitch) {
595 unsigned int i, j;
596
597 // TODO(jbb): why does simd code use both but c doesn't, normalize and
598 // fix..
599 (void) bothclamp;
600 for (i = 0; i < height; i++) {
601 uint8_t *pos = start + i * pitch;
602 char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
603
604 for (j = 0; j < width; j++) {
605 if (pos[j] < blackclamp[0])
606 pos[j] = blackclamp[0];
607
608 if (pos[j] > 255 + whiteclamp[0])
609 pos[j] = 255 + whiteclamp[0];
610
611 pos[j] += ref[j];
612 }
613 }
614 }
615
swap_mi_and_prev_mi(VP9_COMMON * cm)616 static void swap_mi_and_prev_mi(VP9_COMMON *cm) {
617 // Current mip will be the prev_mip for the next frame.
618 MODE_INFO *temp = cm->postproc_state.prev_mip;
619 cm->postproc_state.prev_mip = cm->mip;
620 cm->mip = temp;
621
622 // Update the upper left visible macroblock ptrs.
623 cm->mi = cm->mip + cm->mi_stride + 1;
624 cm->postproc_state.prev_mi = cm->postproc_state.prev_mip + cm->mi_stride + 1;
625 }
626
vp9_post_proc_frame(struct VP9Common * cm,YV12_BUFFER_CONFIG * dest,vp9_ppflags_t * ppflags)627 int vp9_post_proc_frame(struct VP9Common *cm,
628 YV12_BUFFER_CONFIG *dest, vp9_ppflags_t *ppflags) {
629 const int q = VPXMIN(105, cm->lf.filter_level * 2);
630 const int flags = ppflags->post_proc_flag;
631 YV12_BUFFER_CONFIG *const ppbuf = &cm->post_proc_buffer;
632 struct postproc_state *const ppstate = &cm->postproc_state;
633
634 if (!cm->frame_to_show)
635 return -1;
636
637 if (!flags) {
638 *dest = *cm->frame_to_show;
639 return 0;
640 }
641
642 vpx_clear_system_state();
643
644 // Alloc memory for prev_mip in the first frame.
645 if (cm->current_video_frame == 1) {
646 cm->postproc_state.last_base_qindex = cm->base_qindex;
647 cm->postproc_state.last_frame_valid = 1;
648 ppstate->prev_mip = vpx_calloc(cm->mi_alloc_size, sizeof(*cm->mip));
649 if (!ppstate->prev_mip) {
650 return 1;
651 }
652 ppstate->prev_mi = ppstate->prev_mip + cm->mi_stride + 1;
653 memset(ppstate->prev_mip, 0,
654 cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
655 }
656
657 // Allocate post_proc_buffer_int if needed.
658 if ((flags & VP9D_MFQE) && !cm->post_proc_buffer_int.buffer_alloc) {
659 if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
660 const int width = ALIGN_POWER_OF_TWO(cm->width, 4);
661 const int height = ALIGN_POWER_OF_TWO(cm->height, 4);
662
663 if (vpx_alloc_frame_buffer(&cm->post_proc_buffer_int, width, height,
664 cm->subsampling_x, cm->subsampling_y,
665 #if CONFIG_VP9_HIGHBITDEPTH
666 cm->use_highbitdepth,
667 #endif // CONFIG_VP9_HIGHBITDEPTH
668 VP9_ENC_BORDER_IN_PIXELS,
669 cm->byte_alignment) < 0) {
670 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
671 "Failed to allocate MFQE framebuffer");
672 }
673
674 // Ensure that postproc is set to all 0s so that post proc
675 // doesn't pull random data in from edge.
676 memset(cm->post_proc_buffer_int.buffer_alloc, 128,
677 cm->post_proc_buffer.frame_size);
678 }
679 }
680
681 if (vpx_realloc_frame_buffer(&cm->post_proc_buffer, cm->width, cm->height,
682 cm->subsampling_x, cm->subsampling_y,
683 #if CONFIG_VP9_HIGHBITDEPTH
684 cm->use_highbitdepth,
685 #endif
686 VP9_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
687 NULL, NULL, NULL) < 0)
688 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
689 "Failed to allocate post-processing buffer");
690
691 if ((flags & VP9D_MFQE) && cm->current_video_frame >= 2 &&
692 cm->postproc_state.last_frame_valid && cm->bit_depth == 8 &&
693 cm->postproc_state.last_base_qindex <= last_q_thresh &&
694 cm->base_qindex - cm->postproc_state.last_base_qindex >= q_diff_thresh) {
695 vp9_mfqe(cm);
696 // TODO(jackychen): Consider whether enable deblocking by default
697 // if mfqe is enabled. Need to take both the quality and the speed
698 // into consideration.
699 if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
700 vp8_yv12_copy_frame(ppbuf, &cm->post_proc_buffer_int);
701 }
702 if ((flags & VP9D_DEMACROBLOCK) && cm->post_proc_buffer_int.buffer_alloc) {
703 deblock_and_de_macro_block(&cm->post_proc_buffer_int, ppbuf,
704 q + (ppflags->deblocking_level - 5) * 10,
705 1, 0);
706 } else if (flags & VP9D_DEBLOCK) {
707 vp9_deblock(&cm->post_proc_buffer_int, ppbuf, q);
708 } else {
709 vp8_yv12_copy_frame(&cm->post_proc_buffer_int, ppbuf);
710 }
711 } else if (flags & VP9D_DEMACROBLOCK) {
712 deblock_and_de_macro_block(cm->frame_to_show, ppbuf,
713 q + (ppflags->deblocking_level - 5) * 10, 1, 0);
714 } else if (flags & VP9D_DEBLOCK) {
715 vp9_deblock(cm->frame_to_show, ppbuf, q);
716 } else {
717 vp8_yv12_copy_frame(cm->frame_to_show, ppbuf);
718 }
719
720 cm->postproc_state.last_base_qindex = cm->base_qindex;
721 cm->postproc_state.last_frame_valid = 1;
722
723 if (flags & VP9D_ADDNOISE) {
724 const int noise_level = ppflags->noise_level;
725 if (ppstate->last_q != q ||
726 ppstate->last_noise != noise_level) {
727 fillrd(ppstate, 63 - q, noise_level);
728 }
729
730 vp9_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
731 ppstate->whiteclamp, ppstate->bothclamp,
732 ppbuf->y_width, ppbuf->y_height, ppbuf->y_stride);
733 }
734
735 *dest = *ppbuf;
736
737 /* handle problem with extending borders */
738 dest->y_width = cm->width;
739 dest->y_height = cm->height;
740 dest->uv_width = dest->y_width >> cm->subsampling_x;
741 dest->uv_height = dest->y_height >> cm->subsampling_y;
742
743 swap_mi_and_prev_mi(cm);
744 return 0;
745 }
746 #endif // CONFIG_VP9_POSTPROC
747