1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include <assert.h>
13 #include <math.h>
14 #include <string.h>
15
16 #include "config/aom_scale_rtcd.h"
17
18 #include "aom/aom_integer.h"
19 #include "av1/common/cdef.h"
20 #include "av1/common/cdef_block.h"
21 #include "av1/common/onyxc_int.h"
22 #include "av1/common/reconinter.h"
23
sb_all_skip(const AV1_COMMON * const cm,int mi_row,int mi_col)24 int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col) {
25 int maxc, maxr;
26 int skip = 1;
27 maxc = cm->mi_cols - mi_col;
28 maxr = cm->mi_rows - mi_row;
29
30 maxr = AOMMIN(maxr, MI_SIZE_64X64);
31 maxc = AOMMIN(maxc, MI_SIZE_64X64);
32
33 for (int r = 0; r < maxr; r++) {
34 for (int c = 0; c < maxc; c++) {
35 skip =
36 skip &&
37 cm->mi_grid_visible[(mi_row + r) * cm->mi_stride + mi_col + c]->skip;
38 }
39 }
40 return skip;
41 }
42
is_8x8_block_skip(MB_MODE_INFO ** grid,int mi_row,int mi_col,int mi_stride)43 static int is_8x8_block_skip(MB_MODE_INFO **grid, int mi_row, int mi_col,
44 int mi_stride) {
45 int is_skip = 1;
46 for (int r = 0; r < mi_size_high[BLOCK_8X8]; ++r)
47 for (int c = 0; c < mi_size_wide[BLOCK_8X8]; ++c)
48 is_skip &= grid[(mi_row + r) * mi_stride + (mi_col + c)]->skip;
49
50 return is_skip;
51 }
52
sb_compute_cdef_list(const AV1_COMMON * const cm,int mi_row,int mi_col,cdef_list * dlist,BLOCK_SIZE bs)53 int sb_compute_cdef_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
54 cdef_list *dlist, BLOCK_SIZE bs) {
55 MB_MODE_INFO **grid = cm->mi_grid_visible;
56 int maxc = cm->mi_cols - mi_col;
57 int maxr = cm->mi_rows - mi_row;
58
59 if (bs == BLOCK_128X128 || bs == BLOCK_128X64)
60 maxc = AOMMIN(maxc, MI_SIZE_128X128);
61 else
62 maxc = AOMMIN(maxc, MI_SIZE_64X64);
63 if (bs == BLOCK_128X128 || bs == BLOCK_64X128)
64 maxr = AOMMIN(maxr, MI_SIZE_128X128);
65 else
66 maxr = AOMMIN(maxr, MI_SIZE_64X64);
67
68 const int r_step = mi_size_high[BLOCK_8X8];
69 const int c_step = mi_size_wide[BLOCK_8X8];
70 const int r_shift = (r_step == 2);
71 const int c_shift = (c_step == 2);
72
73 assert(r_step == 1 || r_step == 2);
74 assert(c_step == 1 || c_step == 2);
75
76 int count = 0;
77
78 for (int r = 0; r < maxr; r += r_step) {
79 for (int c = 0; c < maxc; c += c_step) {
80 if (!is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride)) {
81 dlist[count].by = r >> r_shift;
82 dlist[count].bx = c >> c_shift;
83 count++;
84 }
85 }
86 }
87 return count;
88 }
89
copy_rect8_8bit_to_16bit_c(uint16_t * dst,int dstride,const uint8_t * src,int sstride,int v,int h)90 void copy_rect8_8bit_to_16bit_c(uint16_t *dst, int dstride, const uint8_t *src,
91 int sstride, int v, int h) {
92 for (int i = 0; i < v; i++) {
93 for (int j = 0; j < h; j++) {
94 dst[i * dstride + j] = src[i * sstride + j];
95 }
96 }
97 }
98
copy_rect8_16bit_to_16bit_c(uint16_t * dst,int dstride,const uint16_t * src,int sstride,int v,int h)99 void copy_rect8_16bit_to_16bit_c(uint16_t *dst, int dstride,
100 const uint16_t *src, int sstride, int v,
101 int h) {
102 for (int i = 0; i < v; i++) {
103 for (int j = 0; j < h; j++) {
104 dst[i * dstride + j] = src[i * sstride + j];
105 }
106 }
107 }
108
copy_sb8_16(AV1_COMMON * cm,uint16_t * dst,int dstride,const uint8_t * src,int src_voffset,int src_hoffset,int sstride,int vsize,int hsize)109 static void copy_sb8_16(AV1_COMMON *cm, uint16_t *dst, int dstride,
110 const uint8_t *src, int src_voffset, int src_hoffset,
111 int sstride, int vsize, int hsize) {
112 if (cm->seq_params.use_highbitdepth) {
113 const uint16_t *base =
114 &CONVERT_TO_SHORTPTR(src)[src_voffset * sstride + src_hoffset];
115 copy_rect8_16bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
116 } else {
117 const uint8_t *base = &src[src_voffset * sstride + src_hoffset];
118 copy_rect8_8bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
119 }
120 }
121
fill_rect(uint16_t * dst,int dstride,int v,int h,uint16_t x)122 static INLINE void fill_rect(uint16_t *dst, int dstride, int v, int h,
123 uint16_t x) {
124 for (int i = 0; i < v; i++) {
125 for (int j = 0; j < h; j++) {
126 dst[i * dstride + j] = x;
127 }
128 }
129 }
130
copy_rect(uint16_t * dst,int dstride,const uint16_t * src,int sstride,int v,int h)131 static INLINE void copy_rect(uint16_t *dst, int dstride, const uint16_t *src,
132 int sstride, int v, int h) {
133 for (int i = 0; i < v; i++) {
134 for (int j = 0; j < h; j++) {
135 dst[i * dstride + j] = src[i * sstride + j];
136 }
137 }
138 }
139
av1_cdef_frame(YV12_BUFFER_CONFIG * frame,AV1_COMMON * cm,MACROBLOCKD * xd)140 void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
141 MACROBLOCKD *xd) {
142 const CdefInfo *const cdef_info = &cm->cdef_info;
143 const int num_planes = av1_num_planes(cm);
144 DECLARE_ALIGNED(16, uint16_t, src[CDEF_INBUF_SIZE]);
145 uint16_t *linebuf[3];
146 uint16_t *colbuf[3];
147 cdef_list dlist[MI_SIZE_64X64 * MI_SIZE_64X64];
148 unsigned char *row_cdef, *prev_row_cdef, *curr_row_cdef;
149 int cdef_count;
150 int dir[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
151 int var[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
152 int mi_wide_l2[3];
153 int mi_high_l2[3];
154 int xdec[3];
155 int ydec[3];
156 int coeff_shift = AOMMAX(cm->seq_params.bit_depth - 8, 0);
157 const int nvfb = (cm->mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
158 const int nhfb = (cm->mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
159 av1_setup_dst_planes(xd->plane, cm->seq_params.sb_size, frame, 0, 0, 0,
160 num_planes);
161 row_cdef = aom_malloc(sizeof(*row_cdef) * (nhfb + 2) * 2);
162 memset(row_cdef, 1, sizeof(*row_cdef) * (nhfb + 2) * 2);
163 prev_row_cdef = row_cdef + 1;
164 curr_row_cdef = prev_row_cdef + nhfb + 2;
165 for (int pli = 0; pli < num_planes; pli++) {
166 xdec[pli] = xd->plane[pli].subsampling_x;
167 ydec[pli] = xd->plane[pli].subsampling_y;
168 mi_wide_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_x;
169 mi_high_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_y;
170 }
171 const int stride = (cm->mi_cols << MI_SIZE_LOG2) + 2 * CDEF_HBORDER;
172 for (int pli = 0; pli < num_planes; pli++) {
173 linebuf[pli] = aom_malloc(sizeof(*linebuf) * CDEF_VBORDER * stride);
174 colbuf[pli] =
175 aom_malloc(sizeof(*colbuf) *
176 ((CDEF_BLOCKSIZE << mi_high_l2[pli]) + 2 * CDEF_VBORDER) *
177 CDEF_HBORDER);
178 }
179 for (int fbr = 0; fbr < nvfb; fbr++) {
180 for (int pli = 0; pli < num_planes; pli++) {
181 const int block_height =
182 (MI_SIZE_64X64 << mi_high_l2[pli]) + 2 * CDEF_VBORDER;
183 fill_rect(colbuf[pli], CDEF_HBORDER, block_height, CDEF_HBORDER,
184 CDEF_VERY_LARGE);
185 }
186 int cdef_left = 1;
187 for (int fbc = 0; fbc < nhfb; fbc++) {
188 int level, sec_strength;
189 int uv_level, uv_sec_strength;
190 int nhb, nvb;
191 int cstart = 0;
192 curr_row_cdef[fbc] = 0;
193 if (cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
194 MI_SIZE_64X64 * fbc] == NULL ||
195 cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
196 MI_SIZE_64X64 * fbc]
197 ->cdef_strength == -1) {
198 cdef_left = 0;
199 continue;
200 }
201 if (!cdef_left) cstart = -CDEF_HBORDER;
202 nhb = AOMMIN(MI_SIZE_64X64, cm->mi_cols - MI_SIZE_64X64 * fbc);
203 nvb = AOMMIN(MI_SIZE_64X64, cm->mi_rows - MI_SIZE_64X64 * fbr);
204 int frame_top, frame_left, frame_bottom, frame_right;
205
206 int mi_row = MI_SIZE_64X64 * fbr;
207 int mi_col = MI_SIZE_64X64 * fbc;
208 // for the current filter block, it's top left corner mi structure (mi_tl)
209 // is first accessed to check whether the top and left boundaries are
210 // frame boundaries. Then bottom-left and top-right mi structures are
211 // accessed to check whether the bottom and right boundaries
212 // (respectively) are frame boundaries.
213 //
214 // Note that we can't just check the bottom-right mi structure - eg. if
215 // we're at the right-hand edge of the frame but not the bottom, then
216 // the bottom-right mi is NULL but the bottom-left is not.
217 frame_top = (mi_row == 0) ? 1 : 0;
218 frame_left = (mi_col == 0) ? 1 : 0;
219
220 if (fbr != nvfb - 1)
221 frame_bottom = (mi_row + MI_SIZE_64X64 == cm->mi_rows) ? 1 : 0;
222 else
223 frame_bottom = 1;
224
225 if (fbc != nhfb - 1)
226 frame_right = (mi_col + MI_SIZE_64X64 == cm->mi_cols) ? 1 : 0;
227 else
228 frame_right = 1;
229
230 const int mbmi_cdef_strength =
231 cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
232 MI_SIZE_64X64 * fbc]
233 ->cdef_strength;
234 level =
235 cdef_info->cdef_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
236 sec_strength =
237 cdef_info->cdef_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
238 sec_strength += sec_strength == 3;
239 uv_level =
240 cdef_info->cdef_uv_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
241 uv_sec_strength =
242 cdef_info->cdef_uv_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
243 uv_sec_strength += uv_sec_strength == 3;
244 if ((level == 0 && sec_strength == 0 && uv_level == 0 &&
245 uv_sec_strength == 0) ||
246 (cdef_count = sb_compute_cdef_list(cm, fbr * MI_SIZE_64X64,
247 fbc * MI_SIZE_64X64, dlist,
248 BLOCK_64X64)) == 0) {
249 cdef_left = 0;
250 continue;
251 }
252
253 curr_row_cdef[fbc] = 1;
254 for (int pli = 0; pli < num_planes; pli++) {
255 int coffset;
256 int rend, cend;
257 int pri_damping = cdef_info->cdef_pri_damping;
258 int sec_damping = cdef_info->cdef_sec_damping;
259 int hsize = nhb << mi_wide_l2[pli];
260 int vsize = nvb << mi_high_l2[pli];
261
262 if (pli) {
263 level = uv_level;
264 sec_strength = uv_sec_strength;
265 }
266
267 if (fbc == nhfb - 1)
268 cend = hsize;
269 else
270 cend = hsize + CDEF_HBORDER;
271
272 if (fbr == nvfb - 1)
273 rend = vsize;
274 else
275 rend = vsize + CDEF_VBORDER;
276
277 coffset = fbc * MI_SIZE_64X64 << mi_wide_l2[pli];
278 if (fbc == nhfb - 1) {
279 /* On the last superblock column, fill in the right border with
280 CDEF_VERY_LARGE to avoid filtering with the outside. */
281 fill_rect(&src[cend + CDEF_HBORDER], CDEF_BSTRIDE,
282 rend + CDEF_VBORDER, hsize + CDEF_HBORDER - cend,
283 CDEF_VERY_LARGE);
284 }
285 if (fbr == nvfb - 1) {
286 /* On the last superblock row, fill in the bottom border with
287 CDEF_VERY_LARGE to avoid filtering with the outside. */
288 fill_rect(&src[(rend + CDEF_VBORDER) * CDEF_BSTRIDE], CDEF_BSTRIDE,
289 CDEF_VBORDER, hsize + 2 * CDEF_HBORDER, CDEF_VERY_LARGE);
290 }
291 /* Copy in the pixels we need from the current superblock for
292 deringing.*/
293 copy_sb8_16(cm,
294 &src[CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER + cstart],
295 CDEF_BSTRIDE, xd->plane[pli].dst.buf,
296 (MI_SIZE_64X64 << mi_high_l2[pli]) * fbr, coffset + cstart,
297 xd->plane[pli].dst.stride, rend, cend - cstart);
298 if (!prev_row_cdef[fbc]) {
299 copy_sb8_16(cm, &src[CDEF_HBORDER], CDEF_BSTRIDE,
300 xd->plane[pli].dst.buf,
301 (MI_SIZE_64X64 << mi_high_l2[pli]) * fbr - CDEF_VBORDER,
302 coffset, xd->plane[pli].dst.stride, CDEF_VBORDER, hsize);
303 } else if (fbr > 0) {
304 copy_rect(&src[CDEF_HBORDER], CDEF_BSTRIDE, &linebuf[pli][coffset],
305 stride, CDEF_VBORDER, hsize);
306 } else {
307 fill_rect(&src[CDEF_HBORDER], CDEF_BSTRIDE, CDEF_VBORDER, hsize,
308 CDEF_VERY_LARGE);
309 }
310 if (!prev_row_cdef[fbc - 1]) {
311 copy_sb8_16(cm, src, CDEF_BSTRIDE, xd->plane[pli].dst.buf,
312 (MI_SIZE_64X64 << mi_high_l2[pli]) * fbr - CDEF_VBORDER,
313 coffset - CDEF_HBORDER, xd->plane[pli].dst.stride,
314 CDEF_VBORDER, CDEF_HBORDER);
315 } else if (fbr > 0 && fbc > 0) {
316 copy_rect(src, CDEF_BSTRIDE, &linebuf[pli][coffset - CDEF_HBORDER],
317 stride, CDEF_VBORDER, CDEF_HBORDER);
318 } else {
319 fill_rect(src, CDEF_BSTRIDE, CDEF_VBORDER, CDEF_HBORDER,
320 CDEF_VERY_LARGE);
321 }
322 if (!prev_row_cdef[fbc + 1]) {
323 copy_sb8_16(cm, &src[CDEF_HBORDER + (nhb << mi_wide_l2[pli])],
324 CDEF_BSTRIDE, xd->plane[pli].dst.buf,
325 (MI_SIZE_64X64 << mi_high_l2[pli]) * fbr - CDEF_VBORDER,
326 coffset + hsize, xd->plane[pli].dst.stride, CDEF_VBORDER,
327 CDEF_HBORDER);
328 } else if (fbr > 0 && fbc < nhfb - 1) {
329 copy_rect(&src[hsize + CDEF_HBORDER], CDEF_BSTRIDE,
330 &linebuf[pli][coffset + hsize], stride, CDEF_VBORDER,
331 CDEF_HBORDER);
332 } else {
333 fill_rect(&src[hsize + CDEF_HBORDER], CDEF_BSTRIDE, CDEF_VBORDER,
334 CDEF_HBORDER, CDEF_VERY_LARGE);
335 }
336 if (cdef_left) {
337 /* If we deringed the superblock on the left then we need to copy in
338 saved pixels. */
339 copy_rect(src, CDEF_BSTRIDE, colbuf[pli], CDEF_HBORDER,
340 rend + CDEF_VBORDER, CDEF_HBORDER);
341 }
342 /* Saving pixels in case we need to dering the superblock on the
343 right. */
344 copy_rect(colbuf[pli], CDEF_HBORDER, src + hsize, CDEF_BSTRIDE,
345 rend + CDEF_VBORDER, CDEF_HBORDER);
346 copy_sb8_16(
347 cm, &linebuf[pli][coffset], stride, xd->plane[pli].dst.buf,
348 (MI_SIZE_64X64 << mi_high_l2[pli]) * (fbr + 1) - CDEF_VBORDER,
349 coffset, xd->plane[pli].dst.stride, CDEF_VBORDER, hsize);
350
351 if (frame_top) {
352 fill_rect(src, CDEF_BSTRIDE, CDEF_VBORDER, hsize + 2 * CDEF_HBORDER,
353 CDEF_VERY_LARGE);
354 }
355 if (frame_left) {
356 fill_rect(src, CDEF_BSTRIDE, vsize + 2 * CDEF_VBORDER, CDEF_HBORDER,
357 CDEF_VERY_LARGE);
358 }
359 if (frame_bottom) {
360 fill_rect(&src[(vsize + CDEF_VBORDER) * CDEF_BSTRIDE], CDEF_BSTRIDE,
361 CDEF_VBORDER, hsize + 2 * CDEF_HBORDER, CDEF_VERY_LARGE);
362 }
363 if (frame_right) {
364 fill_rect(&src[hsize + CDEF_HBORDER], CDEF_BSTRIDE,
365 vsize + 2 * CDEF_VBORDER, CDEF_HBORDER, CDEF_VERY_LARGE);
366 }
367
368 if (cm->seq_params.use_highbitdepth) {
369 cdef_filter_fb(
370 NULL,
371 &CONVERT_TO_SHORTPTR(
372 xd->plane[pli]
373 .dst.buf)[xd->plane[pli].dst.stride *
374 (MI_SIZE_64X64 * fbr << mi_high_l2[pli]) +
375 (fbc * MI_SIZE_64X64 << mi_wide_l2[pli])],
376 xd->plane[pli].dst.stride,
377 &src[CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER], xdec[pli],
378 ydec[pli], dir, NULL, var, pli, dlist, cdef_count, level,
379 sec_strength, pri_damping, sec_damping, coeff_shift);
380 } else {
381 cdef_filter_fb(
382 &xd->plane[pli]
383 .dst.buf[xd->plane[pli].dst.stride *
384 (MI_SIZE_64X64 * fbr << mi_high_l2[pli]) +
385 (fbc * MI_SIZE_64X64 << mi_wide_l2[pli])],
386 NULL, xd->plane[pli].dst.stride,
387 &src[CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER], xdec[pli],
388 ydec[pli], dir, NULL, var, pli, dlist, cdef_count, level,
389 sec_strength, pri_damping, sec_damping, coeff_shift);
390 }
391 }
392 cdef_left = 1;
393 }
394 {
395 unsigned char *tmp = prev_row_cdef;
396 prev_row_cdef = curr_row_cdef;
397 curr_row_cdef = tmp;
398 }
399 }
400 aom_free(row_cdef);
401 for (int pli = 0; pli < num_planes; pli++) {
402 aom_free(linebuf[pli]);
403 aom_free(colbuf[pli]);
404 }
405 }
406