Line data Source code
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 <math.h>
13 :
14 : #include "./aom_config.h"
15 : #include "./aom_dsp_rtcd.h"
16 : #include "av1/common/av1_loopfilter.h"
17 : #include "av1/common/onyxc_int.h"
18 : #include "av1/common/reconinter.h"
19 : #include "aom_dsp/aom_dsp_common.h"
20 : #include "aom_mem/aom_mem.h"
21 : #include "aom_ports/mem.h"
22 :
23 : #include "av1/common/seg_common.h"
24 :
25 : #define PARALLEL_DEBLOCKING_15TAPLUMAONLY 1
26 :
27 : // 64 bit masks for left transform size. Each 1 represents a position where
28 : // we should apply a loop filter across the left border of an 8x8 block
29 : // boundary.
30 : //
31 : // In the case of TX_16X16-> ( in low order byte first we end up with
32 : // a mask that looks like this
33 : //
34 : // 10101010
35 : // 10101010
36 : // 10101010
37 : // 10101010
38 : // 10101010
39 : // 10101010
40 : // 10101010
41 : // 10101010
42 : //
43 : // A loopfilter should be applied to every other 8x8 horizontally.
44 : static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
45 : #if CONFIG_CHROMA_2X2
46 : 0xffffffffffffffffULL, // TX_2X2
47 : #endif
48 : 0xffffffffffffffffULL, // TX_4X4
49 : 0xffffffffffffffffULL, // TX_8x8
50 : 0x5555555555555555ULL, // TX_16x16
51 : 0x1111111111111111ULL, // TX_32x32
52 : #if CONFIG_TX64X64
53 : 0x0101010101010101ULL, // TX_64x64
54 : #endif // CONFIG_TX64X64
55 : };
56 :
57 : // 64 bit masks for above transform size. Each 1 represents a position where
58 : // we should apply a loop filter across the top border of an 8x8 block
59 : // boundary.
60 : //
61 : // In the case of TX_32x32 -> ( in low order byte first we end up with
62 : // a mask that looks like this
63 : //
64 : // 11111111
65 : // 00000000
66 : // 00000000
67 : // 00000000
68 : // 11111111
69 : // 00000000
70 : // 00000000
71 : // 00000000
72 : //
73 : // A loopfilter should be applied to every other 4 the row vertically.
74 : static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
75 : #if CONFIG_CHROMA_2X2
76 : 0xffffffffffffffffULL, // TX_4X4
77 : #endif
78 : 0xffffffffffffffffULL, // TX_4X4
79 : 0xffffffffffffffffULL, // TX_8x8
80 : 0x00ff00ff00ff00ffULL, // TX_16x16
81 : 0x000000ff000000ffULL, // TX_32x32
82 : #if CONFIG_TX64X64
83 : 0x00000000000000ffULL, // TX_64x64
84 : #endif // CONFIG_TX64X64
85 : };
86 :
87 : // 64 bit masks for prediction sizes (left). Each 1 represents a position
88 : // where left border of an 8x8 block. These are aligned to the right most
89 : // appropriate bit, and then shifted into place.
90 : //
91 : // In the case of TX_16x32 -> ( low order byte first ) we end up with
92 : // a mask that looks like this :
93 : //
94 : // 10000000
95 : // 10000000
96 : // 10000000
97 : // 10000000
98 : // 00000000
99 : // 00000000
100 : // 00000000
101 : // 00000000
102 : static const uint64_t left_prediction_mask[BLOCK_SIZES] = {
103 : #if CONFIG_CB4X4
104 : 0x0000000000000001ULL, // BLOCK_2X2,
105 : 0x0000000000000001ULL, // BLOCK_2X4,
106 : 0x0000000000000001ULL, // BLOCK_4X2,
107 : #endif
108 : 0x0000000000000001ULL, // BLOCK_4X4,
109 : 0x0000000000000001ULL, // BLOCK_4X8,
110 : 0x0000000000000001ULL, // BLOCK_8X4,
111 : 0x0000000000000001ULL, // BLOCK_8X8,
112 : 0x0000000000000101ULL, // BLOCK_8X16,
113 : 0x0000000000000001ULL, // BLOCK_16X8,
114 : 0x0000000000000101ULL, // BLOCK_16X16,
115 : 0x0000000001010101ULL, // BLOCK_16X32,
116 : 0x0000000000000101ULL, // BLOCK_32X16,
117 : 0x0000000001010101ULL, // BLOCK_32X32,
118 : 0x0101010101010101ULL, // BLOCK_32X64,
119 : 0x0000000001010101ULL, // BLOCK_64X32,
120 : 0x0101010101010101ULL, // BLOCK_64X64
121 : };
122 :
123 : // 64 bit mask to shift and set for each prediction size.
124 : static const uint64_t above_prediction_mask[BLOCK_SIZES] = {
125 : #if CONFIG_CB4X4
126 : 0x0000000000000001ULL, // BLOCK_2X2
127 : 0x0000000000000001ULL, // BLOCK_2X4
128 : 0x0000000000000001ULL, // BLOCK_4X2
129 : #endif
130 : 0x0000000000000001ULL, // BLOCK_4X4
131 : 0x0000000000000001ULL, // BLOCK_4X8
132 : 0x0000000000000001ULL, // BLOCK_8X4
133 : 0x0000000000000001ULL, // BLOCK_8X8
134 : 0x0000000000000001ULL, // BLOCK_8X16,
135 : 0x0000000000000003ULL, // BLOCK_16X8
136 : 0x0000000000000003ULL, // BLOCK_16X16
137 : 0x0000000000000003ULL, // BLOCK_16X32,
138 : 0x000000000000000fULL, // BLOCK_32X16,
139 : 0x000000000000000fULL, // BLOCK_32X32,
140 : 0x000000000000000fULL, // BLOCK_32X64,
141 : 0x00000000000000ffULL, // BLOCK_64X32,
142 : 0x00000000000000ffULL, // BLOCK_64X64
143 : };
144 : // 64 bit mask to shift and set for each prediction size. A bit is set for
145 : // each 8x8 block that would be in the left most block of the given block
146 : // size in the 64x64 block.
147 : static const uint64_t size_mask[BLOCK_SIZES] = {
148 : #if CONFIG_CB4X4
149 : 0x0000000000000001ULL, // BLOCK_2X2
150 : 0x0000000000000001ULL, // BLOCK_2X4
151 : 0x0000000000000001ULL, // BLOCK_4X2
152 : #endif
153 : 0x0000000000000001ULL, // BLOCK_4X4
154 : 0x0000000000000001ULL, // BLOCK_4X8
155 : 0x0000000000000001ULL, // BLOCK_8X4
156 : 0x0000000000000001ULL, // BLOCK_8X8
157 : 0x0000000000000101ULL, // BLOCK_8X16,
158 : 0x0000000000000003ULL, // BLOCK_16X8
159 : 0x0000000000000303ULL, // BLOCK_16X16
160 : 0x0000000003030303ULL, // BLOCK_16X32,
161 : 0x0000000000000f0fULL, // BLOCK_32X16,
162 : 0x000000000f0f0f0fULL, // BLOCK_32X32,
163 : 0x0f0f0f0f0f0f0f0fULL, // BLOCK_32X64,
164 : 0x00000000ffffffffULL, // BLOCK_64X32,
165 : 0xffffffffffffffffULL, // BLOCK_64X64
166 : };
167 :
168 : // These are used for masking the left and above 32x32 borders.
169 : static const uint64_t left_border = 0x1111111111111111ULL;
170 : static const uint64_t above_border = 0x000000ff000000ffULL;
171 :
172 : // 16 bit masks for uv transform sizes.
173 : static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
174 : #if CONFIG_CHROMA_2X2
175 : 0xffff, // TX_2X2
176 : #endif
177 : 0xffff, // TX_4X4
178 : 0xffff, // TX_8x8
179 : 0x5555, // TX_16x16
180 : 0x1111, // TX_32x32
181 : #if CONFIG_TX64X64
182 : 0x0101, // TX_64x64, never used
183 : #endif // CONFIG_TX64X64
184 : };
185 :
186 : static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
187 : #if CONFIG_CHROMA_2X2
188 : 0xffff, // TX_2X2
189 : #endif
190 : 0xffff, // TX_4X4
191 : 0xffff, // TX_8x8
192 : 0x0f0f, // TX_16x16
193 : 0x000f, // TX_32x32
194 : #if CONFIG_TX64X64
195 : 0x0003, // TX_64x64, never used
196 : #endif // CONFIG_TX64X64
197 : };
198 :
199 : // 16 bit left mask to shift and set for each uv prediction size.
200 : static const uint16_t left_prediction_mask_uv[BLOCK_SIZES] = {
201 : #if CONFIG_CB4X4
202 : 0x0001, // BLOCK_2X2,
203 : 0x0001, // BLOCK_2X4,
204 : 0x0001, // BLOCK_4X2,
205 : #endif
206 : 0x0001, // BLOCK_4X4,
207 : 0x0001, // BLOCK_4X8,
208 : 0x0001, // BLOCK_8X4,
209 : 0x0001, // BLOCK_8X8,
210 : 0x0001, // BLOCK_8X16,
211 : 0x0001, // BLOCK_16X8,
212 : 0x0001, // BLOCK_16X16,
213 : 0x0011, // BLOCK_16X32,
214 : 0x0001, // BLOCK_32X16,
215 : 0x0011, // BLOCK_32X32,
216 : 0x1111, // BLOCK_32X64
217 : 0x0011, // BLOCK_64X32,
218 : 0x1111, // BLOCK_64X64
219 : };
220 : // 16 bit above mask to shift and set for uv each prediction size.
221 : static const uint16_t above_prediction_mask_uv[BLOCK_SIZES] = {
222 : #if CONFIG_CB4X4
223 : 0x0001, // BLOCK_2X2
224 : 0x0001, // BLOCK_2X4
225 : 0x0001, // BLOCK_4X2
226 : #endif
227 : 0x0001, // BLOCK_4X4
228 : 0x0001, // BLOCK_4X8
229 : 0x0001, // BLOCK_8X4
230 : 0x0001, // BLOCK_8X8
231 : 0x0001, // BLOCK_8X16,
232 : 0x0001, // BLOCK_16X8
233 : 0x0001, // BLOCK_16X16
234 : 0x0001, // BLOCK_16X32,
235 : 0x0003, // BLOCK_32X16,
236 : 0x0003, // BLOCK_32X32,
237 : 0x0003, // BLOCK_32X64,
238 : 0x000f, // BLOCK_64X32,
239 : 0x000f, // BLOCK_64X64
240 : };
241 :
242 : // 64 bit mask to shift and set for each uv prediction size
243 : static const uint16_t size_mask_uv[BLOCK_SIZES] = {
244 : #if CONFIG_CB4X4
245 : 0x0001, // BLOCK_2X2
246 : 0x0001, // BLOCK_2X4
247 : 0x0001, // BLOCK_4X2
248 : #endif
249 : 0x0001, // BLOCK_4X4
250 : 0x0001, // BLOCK_4X8
251 : 0x0001, // BLOCK_8X4
252 : 0x0001, // BLOCK_8X8
253 : 0x0001, // BLOCK_8X16,
254 : 0x0001, // BLOCK_16X8
255 : 0x0001, // BLOCK_16X16
256 : 0x0011, // BLOCK_16X32,
257 : 0x0003, // BLOCK_32X16,
258 : 0x0033, // BLOCK_32X32,
259 : 0x3333, // BLOCK_32X64,
260 : 0x00ff, // BLOCK_64X32,
261 : 0xffff, // BLOCK_64X64
262 : };
263 : static const uint16_t left_border_uv = 0x1111;
264 : static const uint16_t above_border_uv = 0x000f;
265 :
266 : static const int mode_lf_lut[] = {
267 : 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // INTRA_MODES
268 : #if CONFIG_ALT_INTRA
269 : 0,
270 : #if CONFIG_SMOOTH_HV
271 : 0, 0,
272 : #endif // CONFIG_SMOOTH_HV
273 : #endif // CONFIG_ALT_INTRA
274 : 1, 1, 0, 1, // INTER_MODES (ZEROMV == 0)
275 : #if CONFIG_EXT_INTER
276 : 1, 1, 1, 1, 1, 1, 0, 1 // INTER_COMPOUND_MODES (ZERO_ZEROMV == 0)
277 : #endif // CONFIG_EXT_INTER
278 : };
279 :
280 0 : static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
281 : int lvl;
282 :
283 : // For each possible value for the loop filter fill out limits
284 0 : for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++) {
285 : // Set loop filter parameters that control sharpness.
286 0 : int block_inside_limit = lvl >> ((sharpness_lvl > 0) + (sharpness_lvl > 4));
287 :
288 0 : if (sharpness_lvl > 0) {
289 0 : if (block_inside_limit > (9 - sharpness_lvl))
290 0 : block_inside_limit = (9 - sharpness_lvl);
291 : }
292 :
293 0 : if (block_inside_limit < 1) block_inside_limit = 1;
294 :
295 0 : memset(lfi->lfthr[lvl].lim, block_inside_limit, SIMD_WIDTH);
296 0 : memset(lfi->lfthr[lvl].mblim, (2 * (lvl + 2) + block_inside_limit),
297 : SIMD_WIDTH);
298 : }
299 0 : }
300 : #if CONFIG_EXT_DELTA_Q
301 0 : static uint8_t get_filter_level(const AV1_COMMON *cm,
302 : const loop_filter_info_n *lfi_n,
303 : const MB_MODE_INFO *mbmi) {
304 : #if CONFIG_SUPERTX
305 : const int segment_id = AOMMIN(mbmi->segment_id, mbmi->segment_id_supertx);
306 : assert(
307 : IMPLIES(supertx_enabled(mbmi), mbmi->segment_id_supertx != MAX_SEGMENTS));
308 : assert(IMPLIES(supertx_enabled(mbmi),
309 : mbmi->segment_id_supertx <= mbmi->segment_id));
310 : #else
311 0 : const int segment_id = mbmi->segment_id;
312 : #endif // CONFIG_SUPERTX
313 0 : if (cm->delta_lf_present_flag) {
314 0 : int lvl_seg = clamp(mbmi->current_delta_lf_from_base + cm->lf.filter_level,
315 : 0, MAX_LOOP_FILTER);
316 0 : const int scale = 1 << (lvl_seg >> 5);
317 0 : if (segfeature_active(&cm->seg, segment_id, SEG_LVL_ALT_LF)) {
318 0 : const int data = get_segdata(&cm->seg, segment_id, SEG_LVL_ALT_LF);
319 0 : lvl_seg =
320 0 : clamp(cm->seg.abs_delta == SEGMENT_ABSDATA ? data : lvl_seg + data, 0,
321 : MAX_LOOP_FILTER);
322 : }
323 :
324 0 : if (cm->lf.mode_ref_delta_enabled) {
325 0 : lvl_seg += cm->lf.ref_deltas[mbmi->ref_frame[0]] * scale;
326 0 : if (mbmi->ref_frame[0] > INTRA_FRAME)
327 0 : lvl_seg += cm->lf.mode_deltas[mode_lf_lut[mbmi->mode]] * scale;
328 0 : lvl_seg = clamp(lvl_seg, 0, MAX_LOOP_FILTER);
329 : }
330 0 : return lvl_seg;
331 : } else {
332 0 : return lfi_n->lvl[segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
333 : }
334 : }
335 : #else
336 : static uint8_t get_filter_level(const loop_filter_info_n *lfi_n,
337 : const MB_MODE_INFO *mbmi) {
338 : #if CONFIG_SUPERTX
339 : const int segment_id = AOMMIN(mbmi->segment_id, mbmi->segment_id_supertx);
340 : assert(
341 : IMPLIES(supertx_enabled(mbmi), mbmi->segment_id_supertx != MAX_SEGMENTS));
342 : assert(IMPLIES(supertx_enabled(mbmi),
343 : mbmi->segment_id_supertx <= mbmi->segment_id));
344 : #else
345 : const int segment_id = mbmi->segment_id;
346 : #endif // CONFIG_SUPERTX
347 : return lfi_n->lvl[segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
348 : }
349 : #endif
350 :
351 0 : void av1_loop_filter_init(AV1_COMMON *cm) {
352 : assert(MB_MODE_COUNT == NELEMENTS(mode_lf_lut));
353 0 : loop_filter_info_n *lfi = &cm->lf_info;
354 0 : struct loopfilter *lf = &cm->lf;
355 : int lvl;
356 :
357 : // init limits for given sharpness
358 0 : update_sharpness(lfi, lf->sharpness_level);
359 0 : lf->last_sharpness_level = lf->sharpness_level;
360 :
361 : // init hev threshold const vectors
362 0 : for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++)
363 0 : memset(lfi->lfthr[lvl].hev_thr, (lvl >> 4), SIMD_WIDTH);
364 0 : }
365 :
366 0 : void av1_loop_filter_frame_init(AV1_COMMON *cm, int default_filt_lvl) {
367 : int seg_id;
368 : // n_shift is the multiplier for lf_deltas
369 : // the multiplier is 1 for when filter_lvl is between 0 and 31;
370 : // 2 when filter_lvl is between 32 and 63
371 0 : const int scale = 1 << (default_filt_lvl >> 5);
372 0 : loop_filter_info_n *const lfi = &cm->lf_info;
373 0 : struct loopfilter *const lf = &cm->lf;
374 0 : const struct segmentation *const seg = &cm->seg;
375 :
376 : // update limits if sharpness has changed
377 0 : if (lf->last_sharpness_level != lf->sharpness_level) {
378 0 : update_sharpness(lfi, lf->sharpness_level);
379 0 : lf->last_sharpness_level = lf->sharpness_level;
380 : }
381 :
382 0 : for (seg_id = 0; seg_id < MAX_SEGMENTS; seg_id++) {
383 0 : int lvl_seg = default_filt_lvl;
384 0 : if (segfeature_active(seg, seg_id, SEG_LVL_ALT_LF)) {
385 0 : const int data = get_segdata(seg, seg_id, SEG_LVL_ALT_LF);
386 0 : lvl_seg = clamp(
387 0 : seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data, 0,
388 : MAX_LOOP_FILTER);
389 : }
390 :
391 0 : if (!lf->mode_ref_delta_enabled) {
392 : // we could get rid of this if we assume that deltas are set to
393 : // zero when not in use; encoder always uses deltas
394 0 : memset(lfi->lvl[seg_id], lvl_seg, sizeof(lfi->lvl[seg_id]));
395 : } else {
396 : int ref, mode;
397 0 : const int intra_lvl = lvl_seg + lf->ref_deltas[INTRA_FRAME] * scale;
398 0 : lfi->lvl[seg_id][INTRA_FRAME][0] = clamp(intra_lvl, 0, MAX_LOOP_FILTER);
399 :
400 0 : for (ref = LAST_FRAME; ref < TOTAL_REFS_PER_FRAME; ++ref) {
401 0 : for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
402 0 : const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
403 0 : lf->mode_deltas[mode] * scale;
404 0 : lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER);
405 : }
406 : }
407 : }
408 : }
409 0 : }
410 :
411 0 : static void filter_selectively_vert_row2(int subsampling_factor, uint8_t *s,
412 : int pitch, unsigned int mask_16x16_l,
413 : unsigned int mask_8x8_l,
414 : unsigned int mask_4x4_l,
415 : unsigned int mask_4x4_int_l,
416 : const loop_filter_info_n *lfi_n,
417 : const uint8_t *lfl) {
418 0 : const int mask_shift = subsampling_factor ? 4 : 8;
419 0 : const int mask_cutoff = subsampling_factor ? 0xf : 0xff;
420 0 : const int lfl_forward = subsampling_factor ? 4 : 8;
421 :
422 0 : unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff;
423 0 : unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff;
424 0 : unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff;
425 0 : unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff;
426 0 : unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff;
427 0 : unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff;
428 0 : unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff;
429 0 : unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff;
430 : unsigned int mask;
431 :
432 0 : for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 |
433 0 : mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
434 0 : mask; mask >>= 1) {
435 0 : const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
436 0 : const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);
437 :
438 0 : if (mask & 1) {
439 0 : if ((mask_16x16_0 | mask_16x16_1) & 1) {
440 0 : if ((mask_16x16_0 & mask_16x16_1) & 1) {
441 0 : aom_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
442 0 : lfi0->hev_thr);
443 0 : } else if (mask_16x16_0 & 1) {
444 0 : aom_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
445 : } else {
446 0 : aom_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
447 0 : lfi1->hev_thr);
448 : }
449 : }
450 :
451 0 : if ((mask_8x8_0 | mask_8x8_1) & 1) {
452 0 : if ((mask_8x8_0 & mask_8x8_1) & 1) {
453 0 : aom_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
454 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
455 0 : lfi1->hev_thr);
456 0 : } else if (mask_8x8_0 & 1) {
457 0 : aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
458 : } else {
459 0 : aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
460 0 : lfi1->hev_thr);
461 : }
462 : }
463 :
464 0 : if ((mask_4x4_0 | mask_4x4_1) & 1) {
465 0 : if ((mask_4x4_0 & mask_4x4_1) & 1) {
466 0 : aom_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
467 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
468 0 : lfi1->hev_thr);
469 0 : } else if (mask_4x4_0 & 1) {
470 0 : aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
471 : } else {
472 0 : aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
473 0 : lfi1->hev_thr);
474 : }
475 : }
476 :
477 0 : if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
478 0 : if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
479 0 : aom_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
480 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
481 0 : lfi1->hev_thr);
482 0 : } else if (mask_4x4_int_0 & 1) {
483 0 : aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
484 0 : lfi0->hev_thr);
485 : } else {
486 0 : aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
487 0 : lfi1->hev_thr);
488 : }
489 : }
490 : }
491 :
492 0 : s += 8;
493 0 : lfl += 1;
494 0 : mask_16x16_0 >>= 1;
495 0 : mask_8x8_0 >>= 1;
496 0 : mask_4x4_0 >>= 1;
497 0 : mask_4x4_int_0 >>= 1;
498 0 : mask_16x16_1 >>= 1;
499 0 : mask_8x8_1 >>= 1;
500 0 : mask_4x4_1 >>= 1;
501 0 : mask_4x4_int_1 >>= 1;
502 : }
503 0 : }
504 :
505 : #if CONFIG_HIGHBITDEPTH
506 0 : static void highbd_filter_selectively_vert_row2(
507 : int subsampling_factor, uint16_t *s, int pitch, unsigned int mask_16x16_l,
508 : unsigned int mask_8x8_l, unsigned int mask_4x4_l,
509 : unsigned int mask_4x4_int_l, const loop_filter_info_n *lfi_n,
510 : const uint8_t *lfl, int bd) {
511 0 : const int mask_shift = subsampling_factor ? 4 : 8;
512 0 : const int mask_cutoff = subsampling_factor ? 0xf : 0xff;
513 0 : const int lfl_forward = subsampling_factor ? 4 : 8;
514 :
515 0 : unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff;
516 0 : unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff;
517 0 : unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff;
518 0 : unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff;
519 0 : unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff;
520 0 : unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff;
521 0 : unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff;
522 0 : unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff;
523 : unsigned int mask;
524 :
525 0 : for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 |
526 0 : mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
527 0 : mask; mask >>= 1) {
528 0 : const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
529 0 : const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);
530 :
531 0 : if (mask & 1) {
532 0 : if ((mask_16x16_0 | mask_16x16_1) & 1) {
533 0 : if ((mask_16x16_0 & mask_16x16_1) & 1) {
534 0 : aom_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
535 0 : lfi0->hev_thr, bd);
536 0 : } else if (mask_16x16_0 & 1) {
537 0 : aom_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
538 0 : lfi0->hev_thr, bd);
539 : } else {
540 0 : aom_highbd_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim,
541 0 : lfi1->lim, lfi1->hev_thr, bd);
542 : }
543 : }
544 :
545 0 : if ((mask_8x8_0 | mask_8x8_1) & 1) {
546 0 : if ((mask_8x8_0 & mask_8x8_1) & 1) {
547 0 : aom_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
548 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
549 0 : lfi1->hev_thr, bd);
550 0 : } else if (mask_8x8_0 & 1) {
551 0 : aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
552 0 : lfi0->hev_thr, bd);
553 : } else {
554 0 : aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
555 0 : lfi1->lim, lfi1->hev_thr, bd);
556 : }
557 : }
558 :
559 0 : if ((mask_4x4_0 | mask_4x4_1) & 1) {
560 0 : if ((mask_4x4_0 & mask_4x4_1) & 1) {
561 0 : aom_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
562 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
563 0 : lfi1->hev_thr, bd);
564 0 : } else if (mask_4x4_0 & 1) {
565 0 : aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
566 0 : lfi0->hev_thr, bd);
567 : } else {
568 0 : aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
569 0 : lfi1->lim, lfi1->hev_thr, bd);
570 : }
571 : }
572 :
573 0 : if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
574 0 : if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
575 0 : aom_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
576 0 : lfi0->hev_thr, lfi1->mblim, lfi1->lim,
577 0 : lfi1->hev_thr, bd);
578 0 : } else if (mask_4x4_int_0 & 1) {
579 0 : aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
580 0 : lfi0->hev_thr, bd);
581 : } else {
582 0 : aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
583 0 : lfi1->lim, lfi1->hev_thr, bd);
584 : }
585 : }
586 : }
587 :
588 0 : s += 8;
589 0 : lfl += 1;
590 0 : mask_16x16_0 >>= 1;
591 0 : mask_8x8_0 >>= 1;
592 0 : mask_4x4_0 >>= 1;
593 0 : mask_4x4_int_0 >>= 1;
594 0 : mask_16x16_1 >>= 1;
595 0 : mask_8x8_1 >>= 1;
596 0 : mask_4x4_1 >>= 1;
597 0 : mask_4x4_int_1 >>= 1;
598 : }
599 0 : }
600 : #endif // CONFIG_HIGHBITDEPTH
601 :
602 0 : static void filter_selectively_horiz(
603 : uint8_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
604 : unsigned int mask_4x4, unsigned int mask_4x4_int,
605 : const loop_filter_info_n *lfi_n, const uint8_t *lfl) {
606 : unsigned int mask;
607 : int count;
608 :
609 0 : for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
610 0 : mask >>= count) {
611 0 : const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
612 :
613 0 : count = 1;
614 0 : if (mask & 1) {
615 0 : if (mask_16x16 & 1) {
616 0 : if ((mask_16x16 & 3) == 3) {
617 0 : aom_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
618 0 : lfi->hev_thr);
619 0 : count = 2;
620 : } else {
621 0 : aom_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
622 0 : lfi->hev_thr);
623 : }
624 0 : } else if (mask_8x8 & 1) {
625 0 : if ((mask_8x8 & 3) == 3) {
626 : // Next block's thresholds.
627 0 : const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
628 :
629 0 : aom_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
630 0 : lfi->hev_thr, lfin->mblim, lfin->lim,
631 0 : lfin->hev_thr);
632 :
633 0 : if ((mask_4x4_int & 3) == 3) {
634 0 : aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
635 0 : lfi->lim, lfi->hev_thr, lfin->mblim,
636 0 : lfin->lim, lfin->hev_thr);
637 : } else {
638 0 : if (mask_4x4_int & 1)
639 0 : aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
640 0 : lfi->hev_thr);
641 0 : else if (mask_4x4_int & 2)
642 0 : aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
643 0 : lfin->lim, lfin->hev_thr);
644 : }
645 0 : count = 2;
646 : } else {
647 0 : aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
648 :
649 0 : if (mask_4x4_int & 1)
650 0 : aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
651 0 : lfi->hev_thr);
652 : }
653 0 : } else if (mask_4x4 & 1) {
654 0 : if ((mask_4x4 & 3) == 3) {
655 : // Next block's thresholds.
656 0 : const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
657 :
658 0 : aom_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
659 0 : lfi->hev_thr, lfin->mblim, lfin->lim,
660 0 : lfin->hev_thr);
661 0 : if ((mask_4x4_int & 3) == 3) {
662 0 : aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
663 0 : lfi->lim, lfi->hev_thr, lfin->mblim,
664 0 : lfin->lim, lfin->hev_thr);
665 : } else {
666 0 : if (mask_4x4_int & 1)
667 0 : aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
668 0 : lfi->hev_thr);
669 0 : else if (mask_4x4_int & 2)
670 0 : aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
671 0 : lfin->lim, lfin->hev_thr);
672 : }
673 0 : count = 2;
674 : } else {
675 0 : aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
676 :
677 0 : if (mask_4x4_int & 1)
678 0 : aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
679 0 : lfi->hev_thr);
680 : }
681 0 : } else if (mask_4x4_int & 1) {
682 0 : aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
683 0 : lfi->hev_thr);
684 : }
685 : }
686 0 : s += 8 * count;
687 0 : lfl += count;
688 0 : mask_16x16 >>= count;
689 0 : mask_8x8 >>= count;
690 0 : mask_4x4 >>= count;
691 0 : mask_4x4_int >>= count;
692 : }
693 0 : }
694 :
695 : #if CONFIG_HIGHBITDEPTH
696 0 : static void highbd_filter_selectively_horiz(
697 : uint16_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
698 : unsigned int mask_4x4, unsigned int mask_4x4_int,
699 : const loop_filter_info_n *lfi_n, const uint8_t *lfl, int bd) {
700 : unsigned int mask;
701 : int count;
702 :
703 0 : for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
704 0 : mask >>= count) {
705 0 : const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
706 :
707 0 : count = 1;
708 0 : if (mask & 1) {
709 0 : if (mask_16x16 & 1) {
710 0 : if ((mask_16x16 & 3) == 3) {
711 0 : aom_highbd_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
712 0 : lfi->hev_thr, bd);
713 0 : count = 2;
714 : } else {
715 0 : aom_highbd_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
716 0 : lfi->hev_thr, bd);
717 : }
718 0 : } else if (mask_8x8 & 1) {
719 0 : if ((mask_8x8 & 3) == 3) {
720 : // Next block's thresholds.
721 0 : const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
722 :
723 0 : aom_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
724 0 : lfi->hev_thr, lfin->mblim, lfin->lim,
725 0 : lfin->hev_thr, bd);
726 :
727 0 : if ((mask_4x4_int & 3) == 3) {
728 0 : aom_highbd_lpf_horizontal_4_dual(
729 0 : s + 4 * pitch, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
730 0 : lfin->mblim, lfin->lim, lfin->hev_thr, bd);
731 : } else {
732 0 : if (mask_4x4_int & 1) {
733 0 : aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
734 0 : lfi->lim, lfi->hev_thr, bd);
735 0 : } else if (mask_4x4_int & 2) {
736 0 : aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
737 0 : lfin->lim, lfin->hev_thr, bd);
738 : }
739 : }
740 0 : count = 2;
741 : } else {
742 0 : aom_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
743 0 : lfi->hev_thr, bd);
744 :
745 0 : if (mask_4x4_int & 1) {
746 0 : aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
747 0 : lfi->lim, lfi->hev_thr, bd);
748 : }
749 : }
750 0 : } else if (mask_4x4 & 1) {
751 0 : if ((mask_4x4 & 3) == 3) {
752 : // Next block's thresholds.
753 0 : const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
754 :
755 0 : aom_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
756 0 : lfi->hev_thr, lfin->mblim, lfin->lim,
757 0 : lfin->hev_thr, bd);
758 0 : if ((mask_4x4_int & 3) == 3) {
759 0 : aom_highbd_lpf_horizontal_4_dual(
760 0 : s + 4 * pitch, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
761 0 : lfin->mblim, lfin->lim, lfin->hev_thr, bd);
762 : } else {
763 0 : if (mask_4x4_int & 1) {
764 0 : aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
765 0 : lfi->lim, lfi->hev_thr, bd);
766 0 : } else if (mask_4x4_int & 2) {
767 0 : aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
768 0 : lfin->lim, lfin->hev_thr, bd);
769 : }
770 : }
771 0 : count = 2;
772 : } else {
773 0 : aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
774 0 : lfi->hev_thr, bd);
775 :
776 0 : if (mask_4x4_int & 1) {
777 0 : aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
778 0 : lfi->lim, lfi->hev_thr, bd);
779 : }
780 : }
781 0 : } else if (mask_4x4_int & 1) {
782 0 : aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
783 0 : lfi->hev_thr, bd);
784 : }
785 : }
786 0 : s += 8 * count;
787 0 : lfl += count;
788 0 : mask_16x16 >>= count;
789 0 : mask_8x8 >>= count;
790 0 : mask_4x4 >>= count;
791 0 : mask_4x4_int >>= count;
792 : }
793 0 : }
794 : #endif // CONFIG_HIGHBITDEPTH
795 :
796 : // This function ors into the current lfm structure, where to do loop
797 : // filters for the specific mi we are looking at. It uses information
798 : // including the block_size_type (32x16, 32x32, etc.), the transform size,
799 : // whether there were any coefficients encoded, and the loop filter strength
800 : // block we are currently looking at. Shift is used to position the
801 : // 1's we produce.
802 : // TODO(JBB) Need another function for different resolution color..
803 0 : static void build_masks(AV1_COMMON *const cm,
804 : const loop_filter_info_n *const lfi_n,
805 : const MODE_INFO *mi, const int shift_y,
806 : const int shift_uv, LOOP_FILTER_MASK *lfm) {
807 0 : const MB_MODE_INFO *mbmi = &mi->mbmi;
808 0 : const BLOCK_SIZE block_size = mbmi->sb_type;
809 : // TODO(debargha): Check if masks can be setup correctly when
810 : // rectangular transfroms are used with the EXT_TX expt.
811 0 : const TX_SIZE tx_size_y = txsize_sqr_map[mbmi->tx_size];
812 0 : const TX_SIZE tx_size_y_left = txsize_horz_map[mbmi->tx_size];
813 0 : const TX_SIZE tx_size_y_above = txsize_vert_map[mbmi->tx_size];
814 0 : const TX_SIZE tx_size_uv =
815 0 : txsize_sqr_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
816 0 : const TX_SIZE tx_size_uv_left =
817 0 : txsize_horz_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
818 0 : const TX_SIZE tx_size_uv_above =
819 0 : txsize_vert_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
820 : #if CONFIG_EXT_DELTA_Q
821 0 : const int filter_level = get_filter_level(cm, lfi_n, mbmi);
822 : #else
823 : const int filter_level = get_filter_level(lfi_n, mbmi);
824 : (void)cm;
825 : #endif
826 0 : uint64_t *const left_y = &lfm->left_y[tx_size_y_left];
827 0 : uint64_t *const above_y = &lfm->above_y[tx_size_y_above];
828 0 : uint64_t *const int_4x4_y = &lfm->int_4x4_y;
829 0 : uint16_t *const left_uv = &lfm->left_uv[tx_size_uv_left];
830 0 : uint16_t *const above_uv = &lfm->above_uv[tx_size_uv_above];
831 0 : uint16_t *const int_4x4_uv = &lfm->left_int_4x4_uv;
832 : int i;
833 :
834 : // If filter level is 0 we don't loop filter.
835 0 : if (!filter_level) {
836 0 : return;
837 : } else {
838 0 : const int w = num_8x8_blocks_wide_lookup[block_size];
839 0 : const int h = num_8x8_blocks_high_lookup[block_size];
840 0 : const int row = (shift_y >> MAX_MIB_SIZE_LOG2);
841 0 : const int col = shift_y - (row << MAX_MIB_SIZE_LOG2);
842 :
843 0 : for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
844 : }
845 :
846 : // These set 1 in the current block size for the block size edges.
847 : // For instance if the block size is 32x16, we'll set:
848 : // above = 1111
849 : // 0000
850 : // and
851 : // left = 1000
852 : // = 1000
853 : // NOTE : In this example the low bit is left most ( 1000 ) is stored as
854 : // 1, not 8...
855 : //
856 : // U and V set things on a 16 bit scale.
857 : //
858 0 : *above_y |= above_prediction_mask[block_size] << shift_y;
859 0 : *above_uv |= above_prediction_mask_uv[block_size] << shift_uv;
860 0 : *left_y |= left_prediction_mask[block_size] << shift_y;
861 0 : *left_uv |= left_prediction_mask_uv[block_size] << shift_uv;
862 :
863 : // If the block has no coefficients and is not intra we skip applying
864 : // the loop filter on block edges.
865 0 : if (mbmi->skip && is_inter_block(mbmi)) return;
866 :
867 : // Here we are adding a mask for the transform size. The transform
868 : // size mask is set to be correct for a 64x64 prediction block size. We
869 : // mask to match the size of the block we are working on and then shift it
870 : // into place..
871 0 : *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y_above])
872 0 : << shift_y;
873 0 : *above_uv |=
874 0 : (size_mask_uv[block_size] & above_64x64_txform_mask_uv[tx_size_uv_above])
875 0 : << shift_uv;
876 :
877 0 : *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y_left])
878 0 : << shift_y;
879 0 : *left_uv |=
880 0 : (size_mask_uv[block_size] & left_64x64_txform_mask_uv[tx_size_uv_left])
881 0 : << shift_uv;
882 :
883 : // Here we are trying to determine what to do with the internal 4x4 block
884 : // boundaries. These differ from the 4x4 boundaries on the outside edge of
885 : // an 8x8 in that the internal ones can be skipped and don't depend on
886 : // the prediction block size.
887 0 : if (tx_size_y == TX_4X4)
888 0 : *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffffULL) << shift_y;
889 :
890 0 : if (tx_size_uv == TX_4X4)
891 0 : *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv;
892 : }
893 :
894 : // This function does the same thing as the one above with the exception that
895 : // it only affects the y masks. It exists because for blocks < 16x16 in size,
896 : // we only update u and v masks on the first block.
897 0 : static void build_y_mask(AV1_COMMON *const cm,
898 : const loop_filter_info_n *const lfi_n,
899 : const MODE_INFO *mi, const int shift_y,
900 : #if CONFIG_SUPERTX
901 : int supertx_enabled,
902 : #endif // CONFIG_SUPERTX
903 : LOOP_FILTER_MASK *lfm) {
904 0 : const MB_MODE_INFO *mbmi = &mi->mbmi;
905 0 : const TX_SIZE tx_size_y = txsize_sqr_map[mbmi->tx_size];
906 0 : const TX_SIZE tx_size_y_left = txsize_horz_map[mbmi->tx_size];
907 0 : const TX_SIZE tx_size_y_above = txsize_vert_map[mbmi->tx_size];
908 : #if CONFIG_SUPERTX
909 : const BLOCK_SIZE block_size =
910 : supertx_enabled ? (BLOCK_SIZE)(3 * tx_size_y) : mbmi->sb_type;
911 : #else
912 0 : const BLOCK_SIZE block_size = mbmi->sb_type;
913 : #endif
914 : #if CONFIG_EXT_DELTA_Q
915 0 : const int filter_level = get_filter_level(cm, lfi_n, mbmi);
916 : #else
917 : const int filter_level = get_filter_level(lfi_n, mbmi);
918 : (void)cm;
919 : #endif
920 0 : uint64_t *const left_y = &lfm->left_y[tx_size_y_left];
921 0 : uint64_t *const above_y = &lfm->above_y[tx_size_y_above];
922 0 : uint64_t *const int_4x4_y = &lfm->int_4x4_y;
923 : int i;
924 :
925 0 : if (!filter_level) {
926 0 : return;
927 : } else {
928 0 : const int w = num_8x8_blocks_wide_lookup[block_size];
929 0 : const int h = num_8x8_blocks_high_lookup[block_size];
930 0 : const int row = (shift_y >> MAX_MIB_SIZE_LOG2);
931 0 : const int col = shift_y - (row << MAX_MIB_SIZE_LOG2);
932 :
933 0 : for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
934 : }
935 :
936 0 : *above_y |= above_prediction_mask[block_size] << shift_y;
937 0 : *left_y |= left_prediction_mask[block_size] << shift_y;
938 :
939 0 : if (mbmi->skip && is_inter_block(mbmi)) return;
940 :
941 0 : *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y_above])
942 0 : << shift_y;
943 :
944 0 : *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y_left])
945 0 : << shift_y;
946 :
947 0 : if (tx_size_y == TX_4X4)
948 0 : *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffffULL) << shift_y;
949 : }
950 :
951 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
952 : // This function update the bit masks for the entire 64x64 region represented
953 : // by mi_row, mi_col. In case one of the edge is a tile boundary, loop filtering
954 : // for that edge is disabled. This function only check the tile boundary info
955 : // for the top left corner mi to determine the boundary information for the
956 : // top and left edge of the whole super block
957 0 : static void update_tile_boundary_filter_mask(AV1_COMMON *const cm,
958 : const int mi_row, const int mi_col,
959 : LOOP_FILTER_MASK *lfm) {
960 : int i;
961 0 : MODE_INFO *const mi = cm->mi + mi_row * cm->mi_stride + mi_col;
962 :
963 0 : if (mi->mbmi.boundary_info & TILE_LEFT_BOUNDARY) {
964 0 : for (i = 0; i <= TX_32X32; i++) {
965 0 : lfm->left_y[i] &= 0xfefefefefefefefeULL;
966 0 : lfm->left_uv[i] &= 0xeeee;
967 : }
968 : }
969 :
970 0 : if (mi->mbmi.boundary_info & TILE_ABOVE_BOUNDARY) {
971 0 : for (i = 0; i <= TX_32X32; i++) {
972 0 : lfm->above_y[i] &= 0xffffffffffffff00ULL;
973 0 : lfm->above_uv[i] &= 0xfff0;
974 : }
975 : }
976 0 : }
977 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
978 :
979 : // This function sets up the bit masks for the entire 64x64 region represented
980 : // by mi_row, mi_col.
981 : // TODO(JBB): This function only works for yv12.
982 0 : void av1_setup_mask(AV1_COMMON *const cm, const int mi_row, const int mi_col,
983 : MODE_INFO **mi, const int mode_info_stride,
984 : LOOP_FILTER_MASK *lfm) {
985 : int idx_32, idx_16, idx_8;
986 0 : const loop_filter_info_n *const lfi_n = &cm->lf_info;
987 0 : MODE_INFO **mip = mi;
988 0 : MODE_INFO **mip2 = mi;
989 :
990 : // These are offsets to the next mi in the 64x64 block. It is what gets
991 : // added to the mi ptr as we go through each loop. It helps us to avoid
992 : // setting up special row and column counters for each index. The last step
993 : // brings us out back to the starting position.
994 0 : const int offset_32[] = { 4, (mode_info_stride << 2) - 4, 4,
995 0 : -(mode_info_stride << 2) - 4 };
996 0 : const int offset_16[] = { 2, (mode_info_stride << 1) - 2, 2,
997 0 : -(mode_info_stride << 1) - 2 };
998 0 : const int offset[] = { 1, mode_info_stride - 1, 1, -mode_info_stride - 1 };
999 :
1000 : // Following variables represent shifts to position the current block
1001 : // mask over the appropriate block. A shift of 36 to the left will move
1002 : // the bits for the final 32 by 32 block in the 64x64 up 4 rows and left
1003 : // 4 rows to the appropriate spot.
1004 0 : const int shift_32_y[] = { 0, 4, 32, 36 };
1005 0 : const int shift_16_y[] = { 0, 2, 16, 18 };
1006 0 : const int shift_8_y[] = { 0, 1, 8, 9 };
1007 0 : const int shift_32_uv[] = { 0, 2, 8, 10 };
1008 0 : const int shift_16_uv[] = { 0, 1, 4, 5 };
1009 : int i;
1010 0 : const int max_rows = AOMMIN(cm->mi_rows - mi_row, MAX_MIB_SIZE);
1011 0 : const int max_cols = AOMMIN(cm->mi_cols - mi_col, MAX_MIB_SIZE);
1012 : #if CONFIG_EXT_PARTITION
1013 : assert(0 && "Not yet updated");
1014 : #endif // CONFIG_EXT_PARTITION
1015 :
1016 0 : av1_zero(*lfm);
1017 0 : assert(mip[0] != NULL);
1018 :
1019 : // TODO(jimbankoski): Try moving most of the following code into decode
1020 : // loop and storing lfm in the mbmi structure so that we don't have to go
1021 : // through the recursive loop structure multiple times.
1022 0 : switch (mip[0]->mbmi.sb_type) {
1023 0 : case BLOCK_64X64: build_masks(cm, lfi_n, mip[0], 0, 0, lfm); break;
1024 0 : case BLOCK_64X32: build_masks(cm, lfi_n, mip[0], 0, 0, lfm);
1025 : #if CONFIG_SUPERTX && CONFIG_TX64X64
1026 : if (supertx_enabled(&mip[0]->mbmi)) break;
1027 : #endif // CONFIG_SUPERTX && CONFIG_TX64X64
1028 0 : mip2 = mip + mode_info_stride * 4;
1029 0 : if (4 >= max_rows) break;
1030 0 : build_masks(cm, lfi_n, mip2[0], 32, 8, lfm);
1031 0 : break;
1032 0 : case BLOCK_32X64: build_masks(cm, lfi_n, mip[0], 0, 0, lfm);
1033 : #if CONFIG_SUPERTX && CONFIG_TX64X64
1034 : if (supertx_enabled(&mip[0]->mbmi)) break;
1035 : #endif // CONFIG_SUPERTX && CONFIG_TX64X64
1036 0 : mip2 = mip + 4;
1037 0 : if (4 >= max_cols) break;
1038 0 : build_masks(cm, lfi_n, mip2[0], 4, 2, lfm);
1039 0 : break;
1040 : default:
1041 : #if CONFIG_SUPERTX && CONFIG_TX64X64
1042 : if (mip[0]->mbmi.tx_size == TX_64X64) {
1043 : build_masks(cm, lfi_n, mip[0], 0, 0, lfm);
1044 : } else {
1045 : #endif // CONFIG_SUPERTX && CONFIG_TX64X64
1046 0 : for (idx_32 = 0; idx_32 < 4; mip += offset_32[idx_32], ++idx_32) {
1047 0 : const int shift_y_32 = shift_32_y[idx_32];
1048 0 : const int shift_uv_32 = shift_32_uv[idx_32];
1049 0 : const int mi_32_col_offset = ((idx_32 & 1) << 2);
1050 0 : const int mi_32_row_offset = ((idx_32 >> 1) << 2);
1051 0 : if (mi_32_col_offset >= max_cols || mi_32_row_offset >= max_rows)
1052 0 : continue;
1053 0 : switch (mip[0]->mbmi.sb_type) {
1054 : case BLOCK_32X32:
1055 0 : build_masks(cm, lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1056 0 : break;
1057 : case BLOCK_32X16:
1058 0 : build_masks(cm, lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1059 : #if CONFIG_SUPERTX
1060 : if (supertx_enabled(&mip[0]->mbmi)) break;
1061 : #endif
1062 0 : if (mi_32_row_offset + 2 >= max_rows) continue;
1063 0 : mip2 = mip + mode_info_stride * 2;
1064 0 : build_masks(cm, lfi_n, mip2[0], shift_y_32 + 16, shift_uv_32 + 4,
1065 : lfm);
1066 0 : break;
1067 : case BLOCK_16X32:
1068 0 : build_masks(cm, lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1069 : #if CONFIG_SUPERTX
1070 : if (supertx_enabled(&mip[0]->mbmi)) break;
1071 : #endif
1072 0 : if (mi_32_col_offset + 2 >= max_cols) continue;
1073 0 : mip2 = mip + 2;
1074 0 : build_masks(cm, lfi_n, mip2[0], shift_y_32 + 2, shift_uv_32 + 1,
1075 : lfm);
1076 0 : break;
1077 : default:
1078 : #if CONFIG_SUPERTX
1079 : if (mip[0]->mbmi.tx_size == TX_32X32) {
1080 : build_masks(cm, lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1081 : break;
1082 : }
1083 : #endif
1084 0 : for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
1085 0 : const int shift_y_32_16 = shift_y_32 + shift_16_y[idx_16];
1086 0 : const int shift_uv_32_16 = shift_uv_32 + shift_16_uv[idx_16];
1087 0 : const int mi_16_col_offset =
1088 0 : mi_32_col_offset + ((idx_16 & 1) << 1);
1089 0 : const int mi_16_row_offset =
1090 0 : mi_32_row_offset + ((idx_16 >> 1) << 1);
1091 :
1092 0 : if (mi_16_col_offset >= max_cols ||
1093 : mi_16_row_offset >= max_rows)
1094 0 : continue;
1095 :
1096 0 : switch (mip[0]->mbmi.sb_type) {
1097 : case BLOCK_16X16:
1098 0 : build_masks(cm, lfi_n, mip[0], shift_y_32_16,
1099 : shift_uv_32_16, lfm);
1100 0 : break;
1101 : case BLOCK_16X8:
1102 : #if CONFIG_SUPERTX
1103 : if (supertx_enabled(&mip[0]->mbmi)) break;
1104 : #endif
1105 0 : build_masks(cm, lfi_n, mip[0], shift_y_32_16,
1106 : shift_uv_32_16, lfm);
1107 0 : if (mi_16_row_offset + 1 >= max_rows) continue;
1108 0 : mip2 = mip + mode_info_stride;
1109 0 : build_y_mask(cm, lfi_n, mip2[0], shift_y_32_16 + 8,
1110 : #if CONFIG_SUPERTX
1111 : 0,
1112 : #endif
1113 : lfm);
1114 0 : break;
1115 : case BLOCK_8X16:
1116 : #if CONFIG_SUPERTX
1117 : if (supertx_enabled(&mip[0]->mbmi)) break;
1118 : #endif
1119 0 : build_masks(cm, lfi_n, mip[0], shift_y_32_16,
1120 : shift_uv_32_16, lfm);
1121 0 : if (mi_16_col_offset + 1 >= max_cols) continue;
1122 0 : mip2 = mip + 1;
1123 0 : build_y_mask(cm, lfi_n, mip2[0], shift_y_32_16 + 1,
1124 : #if CONFIG_SUPERTX
1125 : 0,
1126 : #endif
1127 : lfm);
1128 0 : break;
1129 : default: {
1130 0 : const int shift_y_32_16_8_zero =
1131 0 : shift_y_32_16 + shift_8_y[0];
1132 : #if CONFIG_SUPERTX
1133 : if (mip[0]->mbmi.tx_size == TX_16X16) {
1134 : build_masks(cm, lfi_n, mip[0], shift_y_32_16_8_zero,
1135 : shift_uv_32_16, lfm);
1136 : break;
1137 : }
1138 : #endif
1139 0 : build_masks(cm, lfi_n, mip[0], shift_y_32_16_8_zero,
1140 : shift_uv_32_16, lfm);
1141 0 : mip += offset[0];
1142 0 : for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
1143 0 : const int shift_y_32_16_8 =
1144 0 : shift_y_32_16 + shift_8_y[idx_8];
1145 0 : const int mi_8_col_offset =
1146 0 : mi_16_col_offset + ((idx_8 & 1));
1147 0 : const int mi_8_row_offset =
1148 0 : mi_16_row_offset + ((idx_8 >> 1));
1149 :
1150 0 : if (mi_8_col_offset >= max_cols ||
1151 : mi_8_row_offset >= max_rows)
1152 0 : continue;
1153 0 : build_y_mask(cm, lfi_n, mip[0], shift_y_32_16_8,
1154 : #if CONFIG_SUPERTX
1155 : supertx_enabled(&mip[0]->mbmi),
1156 : #endif
1157 : lfm);
1158 : }
1159 0 : break;
1160 : }
1161 : }
1162 : }
1163 0 : break;
1164 : }
1165 : }
1166 : #if CONFIG_SUPERTX && CONFIG_TX64X64
1167 : }
1168 : #endif // CONFIG_SUPERTX && CONFIG_TX64X64
1169 0 : break;
1170 : }
1171 : // The largest loopfilter we have is 16x16 so we use the 16x16 mask
1172 : // for 32x32 transforms also.
1173 0 : lfm->left_y[TX_16X16] |= lfm->left_y[TX_32X32];
1174 0 : lfm->above_y[TX_16X16] |= lfm->above_y[TX_32X32];
1175 0 : lfm->left_uv[TX_16X16] |= lfm->left_uv[TX_32X32];
1176 0 : lfm->above_uv[TX_16X16] |= lfm->above_uv[TX_32X32];
1177 :
1178 : // We do at least 8 tap filter on every 32x32 even if the transform size
1179 : // is 4x4. So if the 4x4 is set on a border pixel add it to the 8x8 and
1180 : // remove it from the 4x4.
1181 0 : lfm->left_y[TX_8X8] |= lfm->left_y[TX_4X4] & left_border;
1182 0 : lfm->left_y[TX_4X4] &= ~left_border;
1183 0 : lfm->above_y[TX_8X8] |= lfm->above_y[TX_4X4] & above_border;
1184 0 : lfm->above_y[TX_4X4] &= ~above_border;
1185 0 : lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_4X4] & left_border_uv;
1186 0 : lfm->left_uv[TX_4X4] &= ~left_border_uv;
1187 0 : lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_4X4] & above_border_uv;
1188 0 : lfm->above_uv[TX_4X4] &= ~above_border_uv;
1189 :
1190 : // We do some special edge handling.
1191 0 : if (mi_row + MAX_MIB_SIZE > cm->mi_rows) {
1192 0 : const uint64_t rows = cm->mi_rows - mi_row;
1193 :
1194 : // Each pixel inside the border gets a 1,
1195 0 : const uint64_t mask_y = (((uint64_t)1 << (rows << MAX_MIB_SIZE_LOG2)) - 1);
1196 0 : const uint16_t mask_uv =
1197 0 : (((uint16_t)1 << (((rows + 1) >> 1) << (MAX_MIB_SIZE_LOG2 - 1))) - 1);
1198 :
1199 : // Remove values completely outside our border.
1200 0 : for (i = 0; i < TX_32X32; i++) {
1201 0 : lfm->left_y[i] &= mask_y;
1202 0 : lfm->above_y[i] &= mask_y;
1203 0 : lfm->left_uv[i] &= mask_uv;
1204 0 : lfm->above_uv[i] &= mask_uv;
1205 : }
1206 0 : lfm->int_4x4_y &= mask_y;
1207 0 : lfm->above_int_4x4_uv = lfm->left_int_4x4_uv & mask_uv;
1208 :
1209 : // We don't apply a wide loop filter on the last uv block row. If set
1210 : // apply the shorter one instead.
1211 0 : if (rows == 1) {
1212 0 : lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16];
1213 0 : lfm->above_uv[TX_16X16] = 0;
1214 : }
1215 0 : if (rows == 5) {
1216 0 : lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16] & 0xff00;
1217 0 : lfm->above_uv[TX_16X16] &= ~(lfm->above_uv[TX_16X16] & 0xff00);
1218 : }
1219 : } else {
1220 0 : lfm->above_int_4x4_uv = lfm->left_int_4x4_uv;
1221 : }
1222 :
1223 0 : if (mi_col + MAX_MIB_SIZE > cm->mi_cols) {
1224 0 : const uint64_t columns = cm->mi_cols - mi_col;
1225 :
1226 : // Each pixel inside the border gets a 1, the multiply copies the border
1227 : // to where we need it.
1228 0 : const uint64_t mask_y = (((1 << columns) - 1)) * 0x0101010101010101ULL;
1229 0 : const uint16_t mask_uv = ((1 << ((columns + 1) >> 1)) - 1) * 0x1111;
1230 :
1231 : // Internal edges are not applied on the last column of the image so
1232 : // we mask 1 more for the internal edges
1233 0 : const uint16_t mask_uv_int = ((1 << (columns >> 1)) - 1) * 0x1111;
1234 :
1235 : // Remove the bits outside the image edge.
1236 0 : for (i = 0; i < TX_32X32; i++) {
1237 0 : lfm->left_y[i] &= mask_y;
1238 0 : lfm->above_y[i] &= mask_y;
1239 0 : lfm->left_uv[i] &= mask_uv;
1240 0 : lfm->above_uv[i] &= mask_uv;
1241 : }
1242 0 : lfm->int_4x4_y &= mask_y;
1243 0 : lfm->left_int_4x4_uv &= mask_uv_int;
1244 :
1245 : // We don't apply a wide loop filter on the last uv column. If set
1246 : // apply the shorter one instead.
1247 0 : if (columns == 1) {
1248 0 : lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_16X16];
1249 0 : lfm->left_uv[TX_16X16] = 0;
1250 : }
1251 0 : if (columns == 5) {
1252 0 : lfm->left_uv[TX_8X8] |= (lfm->left_uv[TX_16X16] & 0xcccc);
1253 0 : lfm->left_uv[TX_16X16] &= ~(lfm->left_uv[TX_16X16] & 0xcccc);
1254 : }
1255 : }
1256 : // We don't apply a loop filter on the first column in the image, mask that
1257 : // out.
1258 0 : if (mi_col == 0) {
1259 0 : for (i = 0; i < TX_32X32; i++) {
1260 0 : lfm->left_y[i] &= 0xfefefefefefefefeULL;
1261 0 : lfm->left_uv[i] &= 0xeeee;
1262 : }
1263 : }
1264 :
1265 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
1266 0 : if (av1_disable_loopfilter_on_tile_boundary(cm)) {
1267 0 : update_tile_boundary_filter_mask(cm, mi_row, mi_col, lfm);
1268 : }
1269 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
1270 :
1271 : // Assert if we try to apply 2 different loop filters at the same position.
1272 0 : assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_8X8]));
1273 0 : assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_4X4]));
1274 0 : assert(!(lfm->left_y[TX_8X8] & lfm->left_y[TX_4X4]));
1275 0 : assert(!(lfm->int_4x4_y & lfm->left_y[TX_16X16]));
1276 0 : assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_8X8]));
1277 0 : assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_4X4]));
1278 0 : assert(!(lfm->left_uv[TX_8X8] & lfm->left_uv[TX_4X4]));
1279 0 : assert(!(lfm->left_int_4x4_uv & lfm->left_uv[TX_16X16]));
1280 0 : assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_8X8]));
1281 0 : assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_4X4]));
1282 0 : assert(!(lfm->above_y[TX_8X8] & lfm->above_y[TX_4X4]));
1283 0 : assert(!(lfm->int_4x4_y & lfm->above_y[TX_16X16]));
1284 0 : assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_8X8]));
1285 0 : assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_4X4]));
1286 0 : assert(!(lfm->above_uv[TX_8X8] & lfm->above_uv[TX_4X4]));
1287 0 : assert(!(lfm->above_int_4x4_uv & lfm->above_uv[TX_16X16]));
1288 0 : }
1289 :
1290 0 : static void filter_selectively_vert(
1291 : uint8_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
1292 : unsigned int mask_4x4, unsigned int mask_4x4_int,
1293 : const loop_filter_info_n *lfi_n, const uint8_t *lfl) {
1294 : unsigned int mask;
1295 :
1296 0 : for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
1297 0 : mask >>= 1) {
1298 0 : const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
1299 :
1300 0 : if (mask & 1) {
1301 0 : if (mask_16x16 & 1) {
1302 0 : aom_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
1303 0 : } else if (mask_8x8 & 1) {
1304 0 : aom_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
1305 0 : } else if (mask_4x4 & 1) {
1306 0 : aom_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
1307 : }
1308 : }
1309 0 : if (mask_4x4_int & 1)
1310 0 : aom_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
1311 0 : s += 8;
1312 0 : lfl += 1;
1313 0 : mask_16x16 >>= 1;
1314 0 : mask_8x8 >>= 1;
1315 0 : mask_4x4 >>= 1;
1316 0 : mask_4x4_int >>= 1;
1317 : }
1318 0 : }
1319 :
1320 : #if CONFIG_HIGHBITDEPTH
1321 0 : static void highbd_filter_selectively_vert(
1322 : uint16_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
1323 : unsigned int mask_4x4, unsigned int mask_4x4_int,
1324 : const loop_filter_info_n *lfi_n, const uint8_t *lfl, int bd) {
1325 : unsigned int mask;
1326 :
1327 0 : for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
1328 0 : mask >>= 1) {
1329 0 : const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
1330 :
1331 0 : if (mask & 1) {
1332 0 : if (mask_16x16 & 1) {
1333 0 : aom_highbd_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
1334 : bd);
1335 0 : } else if (mask_8x8 & 1) {
1336 0 : aom_highbd_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
1337 : bd);
1338 0 : } else if (mask_4x4 & 1) {
1339 0 : aom_highbd_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
1340 : bd);
1341 : }
1342 : }
1343 0 : if (mask_4x4_int & 1)
1344 0 : aom_highbd_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim,
1345 0 : lfi->hev_thr, bd);
1346 0 : s += 8;
1347 0 : lfl += 1;
1348 0 : mask_16x16 >>= 1;
1349 0 : mask_8x8 >>= 1;
1350 0 : mask_4x4 >>= 1;
1351 0 : mask_4x4_int >>= 1;
1352 : }
1353 0 : }
1354 : #endif // CONFIG_HIGHBITDEPTH
1355 :
1356 : typedef struct {
1357 : unsigned int m16x16;
1358 : unsigned int m8x8;
1359 : unsigned int m4x4;
1360 : } FilterMasks;
1361 :
1362 : // Get filter level and masks for the given row index 'idx_r'. (Only used for
1363 : // the non420 case).
1364 : // Note: 'row_masks_ptr' and/or 'col_masks_ptr' can be passed NULL.
1365 0 : static void get_filter_level_and_masks_non420(
1366 : AV1_COMMON *const cm, const struct macroblockd_plane *const plane, int pl,
1367 : MODE_INFO **mib, int mi_row, int mi_col, int idx_r, uint8_t *const lfl_r,
1368 : unsigned int *const mask_4x4_int_r_ptr,
1369 : unsigned int *const mask_4x4_int_c_ptr, FilterMasks *const row_masks_ptr,
1370 : FilterMasks *const col_masks_ptr) {
1371 0 : const int ss_x = plane->subsampling_x;
1372 0 : const int ss_y = plane->subsampling_y;
1373 0 : const int col_step = mi_size_wide[BLOCK_8X8] << ss_x;
1374 : FilterMasks row_masks, col_masks;
1375 0 : memset(&row_masks, 0, sizeof(row_masks));
1376 0 : memset(&col_masks, 0, sizeof(col_masks));
1377 0 : unsigned int mask_4x4_int_r = 0, mask_4x4_int_c = 0;
1378 0 : const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
1379 :
1380 : // Determine the vertical edges that need filtering
1381 : int idx_c;
1382 0 : for (idx_c = 0; idx_c < cm->mib_size && mi_col + idx_c < cm->mi_cols;
1383 0 : idx_c += col_step) {
1384 0 : const MODE_INFO *mi = mib[idx_r * cm->mi_stride + idx_c];
1385 0 : const MB_MODE_INFO *mbmi = &mi[0].mbmi;
1386 0 : const BLOCK_SIZE sb_type = mbmi->sb_type;
1387 0 : const int skip_this = mbmi->skip && is_inter_block(mbmi);
1388 : // Map index to 8x8 unit
1389 0 : const int c = idx_c >> mi_width_log2_lookup[BLOCK_8X8];
1390 :
1391 0 : const int blk_row = r & (num_8x8_blocks_high_lookup[sb_type] - 1);
1392 0 : const int blk_col = c & (num_8x8_blocks_wide_lookup[sb_type] - 1);
1393 :
1394 : // left edge of current unit is block/partition edge -> no skip
1395 0 : const int block_edge_left =
1396 0 : (num_4x4_blocks_wide_lookup[sb_type] > 1) ? !blk_col : 1;
1397 0 : const int skip_this_c = skip_this && !block_edge_left;
1398 : // top edge of current unit is block/partition edge -> no skip
1399 0 : const int block_edge_above =
1400 0 : (num_4x4_blocks_high_lookup[sb_type] > 1) ? !blk_row : 1;
1401 0 : const int skip_this_r = skip_this && !block_edge_above;
1402 :
1403 0 : TX_SIZE tx_size = (plane->plane_type == PLANE_TYPE_UV)
1404 : ? get_uv_tx_size(mbmi, plane)
1405 0 : : mbmi->tx_size;
1406 :
1407 0 : const int skip_border_4x4_c =
1408 0 : ss_x && mi_col + idx_c >= cm->mi_cols - mi_size_wide[BLOCK_8X8];
1409 0 : const int skip_border_4x4_r =
1410 0 : ss_y && mi_row + idx_r >= cm->mi_rows - mi_size_high[BLOCK_8X8];
1411 :
1412 0 : int tx_size_mask = 0;
1413 0 : const int c_step = (c >> ss_x);
1414 0 : const int r_step = (r >> ss_y);
1415 0 : const int col_mask = 1 << c_step;
1416 :
1417 : #if CONFIG_VAR_TX
1418 0 : if (is_inter_block(mbmi) && !mbmi->skip) {
1419 0 : const int tx_row_idx =
1420 0 : (blk_row * mi_size_high[BLOCK_8X8] << TX_UNIT_HIGH_LOG2) >> 1;
1421 0 : const int tx_col_idx =
1422 0 : (blk_col * mi_size_wide[BLOCK_8X8] << TX_UNIT_WIDE_LOG2) >> 1;
1423 0 : const BLOCK_SIZE bsize =
1424 0 : AOMMAX(BLOCK_4X4, get_plane_block_size(mbmi->sb_type, plane));
1425 0 : const TX_SIZE mb_tx_size = mbmi->inter_tx_size[tx_row_idx][tx_col_idx];
1426 0 : tx_size = (plane->plane_type == PLANE_TYPE_UV)
1427 : ? uv_txsize_lookup[bsize][mb_tx_size][0][0]
1428 0 : : mb_tx_size;
1429 : }
1430 : #endif
1431 :
1432 : // Filter level can vary per MI
1433 : #if CONFIG_EXT_DELTA_Q
1434 0 : if (!(lfl_r[c_step] = get_filter_level(cm, &cm->lf_info, mbmi))) continue;
1435 : #else
1436 : if (!(lfl_r[c_step] = get_filter_level(&cm->lf_info, mbmi))) continue;
1437 : #endif
1438 :
1439 : #if CONFIG_VAR_TX
1440 : TX_SIZE tx_size_r, tx_size_c;
1441 :
1442 0 : const int tx_wide =
1443 0 : AOMMIN(tx_size_wide[tx_size],
1444 : tx_size_wide[cm->top_txfm_context[pl][(mi_col + idx_c)
1445 : << TX_UNIT_WIDE_LOG2]]);
1446 0 : const int tx_high = AOMMIN(
1447 : tx_size_high[tx_size],
1448 : tx_size_high[cm->left_txfm_context[pl][((mi_row + idx_r) & MAX_MIB_MASK)
1449 : << TX_UNIT_HIGH_LOG2]]);
1450 :
1451 0 : tx_size_c = get_sqr_tx_size(tx_wide);
1452 0 : tx_size_r = get_sqr_tx_size(tx_high);
1453 :
1454 0 : memset(cm->top_txfm_context[pl] + ((mi_col + idx_c) << TX_UNIT_WIDE_LOG2),
1455 0 : tx_size, mi_size_wide[BLOCK_8X8] << TX_UNIT_WIDE_LOG2);
1456 0 : memset(cm->left_txfm_context[pl] +
1457 0 : (((mi_row + idx_r) & MAX_MIB_MASK) << TX_UNIT_HIGH_LOG2),
1458 0 : tx_size, mi_size_high[BLOCK_8X8] << TX_UNIT_HIGH_LOG2);
1459 : #else
1460 : TX_SIZE tx_size_c = txsize_horz_map[tx_size];
1461 : TX_SIZE tx_size_r = txsize_vert_map[tx_size];
1462 : (void)pl;
1463 : #endif // CONFIG_VAR_TX
1464 :
1465 0 : if (tx_size_c == TX_32X32)
1466 0 : tx_size_mask = 3;
1467 0 : else if (tx_size_c == TX_16X16)
1468 0 : tx_size_mask = 1;
1469 : else
1470 0 : tx_size_mask = 0;
1471 :
1472 : // Build masks based on the transform size of each block
1473 : // handle vertical mask
1474 0 : if (tx_size_c == TX_32X32) {
1475 0 : if (!skip_this_c && (c_step & tx_size_mask) == 0) {
1476 0 : if (!skip_border_4x4_c)
1477 0 : col_masks.m16x16 |= col_mask;
1478 : else
1479 0 : col_masks.m8x8 |= col_mask;
1480 : }
1481 0 : } else if (tx_size_c == TX_16X16) {
1482 0 : if (!skip_this_c && (c_step & tx_size_mask) == 0) {
1483 0 : if (!skip_border_4x4_c)
1484 0 : col_masks.m16x16 |= col_mask;
1485 : else
1486 0 : col_masks.m8x8 |= col_mask;
1487 : }
1488 : } else {
1489 : // force 8x8 filtering on 32x32 boundaries
1490 0 : if (!skip_this_c && (c_step & tx_size_mask) == 0) {
1491 0 : if (tx_size_c == TX_8X8 || ((c >> ss_x) & 3) == 0)
1492 0 : col_masks.m8x8 |= col_mask;
1493 : else
1494 0 : col_masks.m4x4 |= col_mask;
1495 : }
1496 :
1497 0 : if (!skip_this && tx_size_c < TX_8X8 && !skip_border_4x4_c &&
1498 0 : (c_step & tx_size_mask) == 0)
1499 0 : mask_4x4_int_c |= col_mask;
1500 : }
1501 :
1502 0 : if (tx_size_r == TX_32X32)
1503 0 : tx_size_mask = 3;
1504 0 : else if (tx_size_r == TX_16X16)
1505 0 : tx_size_mask = 1;
1506 : else
1507 0 : tx_size_mask = 0;
1508 :
1509 : // set horizontal mask
1510 0 : if (tx_size_r == TX_32X32) {
1511 0 : if (!skip_this_r && (r_step & tx_size_mask) == 0) {
1512 0 : if (!skip_border_4x4_r)
1513 0 : row_masks.m16x16 |= col_mask;
1514 : else
1515 0 : row_masks.m8x8 |= col_mask;
1516 : }
1517 0 : } else if (tx_size_r == TX_16X16) {
1518 0 : if (!skip_this_r && (r_step & tx_size_mask) == 0) {
1519 0 : if (!skip_border_4x4_r)
1520 0 : row_masks.m16x16 |= col_mask;
1521 : else
1522 0 : row_masks.m8x8 |= col_mask;
1523 : }
1524 : } else {
1525 : // force 8x8 filtering on 32x32 boundaries
1526 0 : if (!skip_this_r && (r_step & tx_size_mask) == 0) {
1527 0 : if (tx_size_r == TX_8X8 || (r_step & 3) == 0)
1528 0 : row_masks.m8x8 |= col_mask;
1529 : else
1530 0 : row_masks.m4x4 |= col_mask;
1531 : }
1532 :
1533 0 : if (!skip_this && tx_size_r < TX_8X8 && !skip_border_4x4_r &&
1534 0 : ((r >> ss_y) & tx_size_mask) == 0)
1535 0 : mask_4x4_int_r |= col_mask;
1536 : }
1537 : }
1538 :
1539 0 : if (row_masks_ptr) *row_masks_ptr = row_masks;
1540 0 : if (col_masks_ptr) *col_masks_ptr = col_masks;
1541 0 : if (mask_4x4_int_c_ptr) *mask_4x4_int_c_ptr = mask_4x4_int_c;
1542 0 : if (mask_4x4_int_r_ptr) *mask_4x4_int_r_ptr = mask_4x4_int_r;
1543 0 : }
1544 :
1545 0 : void av1_filter_block_plane_non420_ver(AV1_COMMON *const cm,
1546 : struct macroblockd_plane *plane,
1547 : MODE_INFO **mib, int mi_row, int mi_col,
1548 : int pl) {
1549 0 : const int ss_y = plane->subsampling_y;
1550 0 : const int row_step = mi_size_high[BLOCK_8X8] << ss_y;
1551 0 : struct buf_2d *const dst = &plane->dst;
1552 0 : uint8_t *const dst0 = dst->buf;
1553 0 : uint8_t lfl[MAX_MIB_SIZE][MAX_MIB_SIZE] = { { 0 } };
1554 :
1555 : int idx_r;
1556 0 : for (idx_r = 0; idx_r < cm->mib_size && mi_row + idx_r < cm->mi_rows;
1557 0 : idx_r += row_step) {
1558 : unsigned int mask_4x4_int;
1559 : FilterMasks col_masks;
1560 0 : const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
1561 0 : get_filter_level_and_masks_non420(cm, plane, pl, mib, mi_row, mi_col, idx_r,
1562 : &lfl[r][0], NULL, &mask_4x4_int, NULL,
1563 : &col_masks);
1564 :
1565 : // Disable filtering on the leftmost column or tile boundary
1566 0 : unsigned int border_mask = ~(mi_col == 0);
1567 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
1568 0 : if (av1_disable_loopfilter_on_tile_boundary(cm) &&
1569 0 : ((mib[0]->mbmi.boundary_info & TILE_LEFT_BOUNDARY) != 0)) {
1570 0 : border_mask = 0xfffffffe;
1571 : }
1572 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
1573 :
1574 : #if CONFIG_HIGHBITDEPTH
1575 0 : if (cm->use_highbitdepth)
1576 0 : highbd_filter_selectively_vert(
1577 0 : CONVERT_TO_SHORTPTR(dst->buf), dst->stride,
1578 0 : col_masks.m16x16 & border_mask, col_masks.m8x8 & border_mask,
1579 0 : col_masks.m4x4 & border_mask, mask_4x4_int, &cm->lf_info, &lfl[r][0],
1580 0 : (int)cm->bit_depth);
1581 : else
1582 : #endif // CONFIG_HIGHBITDEPTH
1583 0 : filter_selectively_vert(
1584 0 : dst->buf, dst->stride, col_masks.m16x16 & border_mask,
1585 0 : col_masks.m8x8 & border_mask, col_masks.m4x4 & border_mask,
1586 0 : mask_4x4_int, &cm->lf_info, &lfl[r][0]);
1587 0 : dst->buf += 8 * dst->stride;
1588 : }
1589 :
1590 : // Now do horizontal pass
1591 0 : dst->buf = dst0;
1592 0 : }
1593 :
1594 0 : void av1_filter_block_plane_non420_hor(AV1_COMMON *const cm,
1595 : struct macroblockd_plane *plane,
1596 : MODE_INFO **mib, int mi_row, int mi_col,
1597 : int pl) {
1598 0 : const int ss_y = plane->subsampling_y;
1599 0 : const int row_step = mi_size_high[BLOCK_8X8] << ss_y;
1600 0 : struct buf_2d *const dst = &plane->dst;
1601 0 : uint8_t *const dst0 = dst->buf;
1602 : FilterMasks row_masks_array[MAX_MIB_SIZE];
1603 0 : unsigned int mask_4x4_int[MAX_MIB_SIZE] = { 0 };
1604 0 : uint8_t lfl[MAX_MIB_SIZE][MAX_MIB_SIZE] = { { 0 } };
1605 : int idx_r;
1606 0 : for (idx_r = 0; idx_r < cm->mib_size && mi_row + idx_r < cm->mi_rows;
1607 0 : idx_r += row_step) {
1608 0 : const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
1609 0 : get_filter_level_and_masks_non420(cm, plane, pl, mib, mi_row, mi_col, idx_r,
1610 0 : &lfl[r][0], mask_4x4_int + r, NULL,
1611 0 : row_masks_array + r, NULL);
1612 : }
1613 0 : for (idx_r = 0; idx_r < cm->mib_size && mi_row + idx_r < cm->mi_rows;
1614 0 : idx_r += row_step) {
1615 0 : const int r = idx_r >> mi_width_log2_lookup[BLOCK_8X8];
1616 : FilterMasks row_masks;
1617 :
1618 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
1619 : // Disable filtering on the abovemost row or tile boundary
1620 0 : const MODE_INFO *mi = cm->mi + (mi_row + r) * cm->mi_stride;
1621 0 : if ((av1_disable_loopfilter_on_tile_boundary(cm) &&
1622 0 : (mi->mbmi.boundary_info & TILE_ABOVE_BOUNDARY)) ||
1623 0 : (mi_row + idx_r == 0)) {
1624 0 : memset(&row_masks, 0, sizeof(row_masks));
1625 : #else
1626 : if (mi_row + idx_r == 0) {
1627 : memset(&row_masks, 0, sizeof(row_masks));
1628 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
1629 : } else {
1630 0 : memcpy(&row_masks, row_masks_array + r, sizeof(row_masks));
1631 : }
1632 : #if CONFIG_HIGHBITDEPTH
1633 0 : if (cm->use_highbitdepth)
1634 0 : highbd_filter_selectively_horiz(
1635 0 : CONVERT_TO_SHORTPTR(dst->buf), dst->stride, row_masks.m16x16,
1636 0 : row_masks.m8x8, row_masks.m4x4, mask_4x4_int[r], &cm->lf_info,
1637 0 : &lfl[r][0], (int)cm->bit_depth);
1638 : else
1639 : #endif // CONFIG_HIGHBITDEPTH
1640 0 : filter_selectively_horiz(dst->buf, dst->stride, row_masks.m16x16,
1641 : row_masks.m8x8, row_masks.m4x4, mask_4x4_int[r],
1642 0 : &cm->lf_info, &lfl[r][0]);
1643 0 : dst->buf += 8 * dst->stride;
1644 : }
1645 0 : dst->buf = dst0;
1646 0 : }
1647 :
1648 0 : void av1_filter_block_plane_ss00_ver(AV1_COMMON *const cm,
1649 : struct macroblockd_plane *const plane,
1650 : int mi_row, LOOP_FILTER_MASK *lfm) {
1651 0 : struct buf_2d *const dst = &plane->dst;
1652 0 : uint8_t *const dst0 = dst->buf;
1653 : int r;
1654 0 : uint64_t mask_16x16 = lfm->left_y[TX_16X16];
1655 0 : uint64_t mask_8x8 = lfm->left_y[TX_8X8];
1656 0 : uint64_t mask_4x4 = lfm->left_y[TX_4X4];
1657 0 : uint64_t mask_4x4_int = lfm->int_4x4_y;
1658 :
1659 0 : assert(plane->subsampling_x == 0 && plane->subsampling_y == 0);
1660 :
1661 : // Vertical pass: do 2 rows at one time
1662 0 : for (r = 0; r < cm->mib_size && mi_row + r < cm->mi_rows; r += 2) {
1663 0 : unsigned int mask_16x16_l = mask_16x16 & 0xffff;
1664 0 : unsigned int mask_8x8_l = mask_8x8 & 0xffff;
1665 0 : unsigned int mask_4x4_l = mask_4x4 & 0xffff;
1666 0 : unsigned int mask_4x4_int_l = mask_4x4_int & 0xffff;
1667 :
1668 : // Disable filtering on the leftmost column.
1669 : #if CONFIG_HIGHBITDEPTH
1670 0 : if (cm->use_highbitdepth)
1671 0 : highbd_filter_selectively_vert_row2(
1672 0 : plane->subsampling_x, CONVERT_TO_SHORTPTR(dst->buf), dst->stride,
1673 0 : mask_16x16_l, mask_8x8_l, mask_4x4_l, mask_4x4_int_l, &cm->lf_info,
1674 0 : &lfm->lfl_y[r][0], (int)cm->bit_depth);
1675 : else
1676 : #endif // CONFIG_HIGHBITDEPTH
1677 0 : filter_selectively_vert_row2(
1678 : plane->subsampling_x, dst->buf, dst->stride, mask_16x16_l, mask_8x8_l,
1679 0 : mask_4x4_l, mask_4x4_int_l, &cm->lf_info, &lfm->lfl_y[r][0]);
1680 :
1681 0 : dst->buf += 2 * MI_SIZE * dst->stride;
1682 0 : mask_16x16 >>= 2 * MI_SIZE;
1683 0 : mask_8x8 >>= 2 * MI_SIZE;
1684 0 : mask_4x4 >>= 2 * MI_SIZE;
1685 0 : mask_4x4_int >>= 2 * MI_SIZE;
1686 : }
1687 :
1688 : // Horizontal pass
1689 0 : dst->buf = dst0;
1690 0 : }
1691 :
1692 0 : void av1_filter_block_plane_ss00_hor(AV1_COMMON *const cm,
1693 : struct macroblockd_plane *const plane,
1694 : int mi_row, LOOP_FILTER_MASK *lfm) {
1695 0 : struct buf_2d *const dst = &plane->dst;
1696 0 : uint8_t *const dst0 = dst->buf;
1697 : int r;
1698 0 : uint64_t mask_16x16 = lfm->above_y[TX_16X16];
1699 0 : uint64_t mask_8x8 = lfm->above_y[TX_8X8];
1700 0 : uint64_t mask_4x4 = lfm->above_y[TX_4X4];
1701 0 : uint64_t mask_4x4_int = lfm->int_4x4_y;
1702 :
1703 0 : assert(plane->subsampling_x == 0 && plane->subsampling_y == 0);
1704 :
1705 0 : for (r = 0; r < cm->mib_size && mi_row + r < cm->mi_rows; r++) {
1706 : unsigned int mask_16x16_r;
1707 : unsigned int mask_8x8_r;
1708 : unsigned int mask_4x4_r;
1709 :
1710 0 : if (mi_row + r == 0) {
1711 0 : mask_16x16_r = 0;
1712 0 : mask_8x8_r = 0;
1713 0 : mask_4x4_r = 0;
1714 : } else {
1715 0 : mask_16x16_r = mask_16x16 & 0xff;
1716 0 : mask_8x8_r = mask_8x8 & 0xff;
1717 0 : mask_4x4_r = mask_4x4 & 0xff;
1718 : }
1719 :
1720 : #if CONFIG_HIGHBITDEPTH
1721 0 : if (cm->use_highbitdepth)
1722 0 : highbd_filter_selectively_horiz(
1723 0 : CONVERT_TO_SHORTPTR(dst->buf), dst->stride, mask_16x16_r, mask_8x8_r,
1724 0 : mask_4x4_r, mask_4x4_int & 0xff, &cm->lf_info, &lfm->lfl_y[r][0],
1725 0 : (int)cm->bit_depth);
1726 : else
1727 : #endif // CONFIG_HIGHBITDEPTH
1728 0 : filter_selectively_horiz(dst->buf, dst->stride, mask_16x16_r, mask_8x8_r,
1729 0 : mask_4x4_r, mask_4x4_int & 0xff, &cm->lf_info,
1730 0 : &lfm->lfl_y[r][0]);
1731 :
1732 0 : dst->buf += MI_SIZE * dst->stride;
1733 0 : mask_16x16 >>= MI_SIZE;
1734 0 : mask_8x8 >>= MI_SIZE;
1735 0 : mask_4x4 >>= MI_SIZE;
1736 0 : mask_4x4_int >>= MI_SIZE;
1737 : }
1738 : // restore the buf pointer in case there is additional filter pass.
1739 0 : dst->buf = dst0;
1740 0 : }
1741 :
1742 0 : void av1_filter_block_plane_ss11_ver(AV1_COMMON *const cm,
1743 : struct macroblockd_plane *const plane,
1744 : int mi_row, LOOP_FILTER_MASK *lfm) {
1745 0 : struct buf_2d *const dst = &plane->dst;
1746 0 : uint8_t *const dst0 = dst->buf;
1747 : int r, c;
1748 :
1749 0 : uint16_t mask_16x16 = lfm->left_uv[TX_16X16];
1750 0 : uint16_t mask_8x8 = lfm->left_uv[TX_8X8];
1751 0 : uint16_t mask_4x4 = lfm->left_uv[TX_4X4];
1752 0 : uint16_t mask_4x4_int = lfm->left_int_4x4_uv;
1753 :
1754 0 : assert(plane->subsampling_x == 1 && plane->subsampling_y == 1);
1755 0 : assert(plane->plane_type == PLANE_TYPE_UV);
1756 0 : memset(lfm->lfl_uv, 0, sizeof(lfm->lfl_uv));
1757 :
1758 : // Vertical pass: do 2 rows at one time
1759 0 : for (r = 0; r < cm->mib_size && mi_row + r < cm->mi_rows; r += 4) {
1760 0 : for (c = 0; c < (cm->mib_size >> 1); c++) {
1761 0 : lfm->lfl_uv[r >> 1][c] = lfm->lfl_y[r][c << 1];
1762 0 : lfm->lfl_uv[(r + 2) >> 1][c] = lfm->lfl_y[r + 2][c << 1];
1763 : }
1764 :
1765 : {
1766 0 : unsigned int mask_16x16_l = mask_16x16 & 0xff;
1767 0 : unsigned int mask_8x8_l = mask_8x8 & 0xff;
1768 0 : unsigned int mask_4x4_l = mask_4x4 & 0xff;
1769 0 : unsigned int mask_4x4_int_l = mask_4x4_int & 0xff;
1770 :
1771 : // Disable filtering on the leftmost column.
1772 : #if CONFIG_HIGHBITDEPTH
1773 0 : if (cm->use_highbitdepth)
1774 0 : highbd_filter_selectively_vert_row2(
1775 0 : plane->subsampling_x, CONVERT_TO_SHORTPTR(dst->buf), dst->stride,
1776 0 : mask_16x16_l, mask_8x8_l, mask_4x4_l, mask_4x4_int_l, &cm->lf_info,
1777 0 : &lfm->lfl_uv[r >> 1][0], (int)cm->bit_depth);
1778 : else
1779 : #endif // CONFIG_HIGHBITDEPTH
1780 0 : filter_selectively_vert_row2(plane->subsampling_x, dst->buf,
1781 : dst->stride, mask_16x16_l, mask_8x8_l,
1782 0 : mask_4x4_l, mask_4x4_int_l, &cm->lf_info,
1783 0 : &lfm->lfl_uv[r >> 1][0]);
1784 :
1785 0 : dst->buf += 2 * MI_SIZE * dst->stride;
1786 0 : mask_16x16 >>= MI_SIZE;
1787 0 : mask_8x8 >>= MI_SIZE;
1788 0 : mask_4x4 >>= MI_SIZE;
1789 0 : mask_4x4_int >>= MI_SIZE;
1790 : }
1791 : }
1792 :
1793 : // Horizontal pass
1794 0 : dst->buf = dst0;
1795 0 : }
1796 :
1797 0 : void av1_filter_block_plane_ss11_hor(AV1_COMMON *const cm,
1798 : struct macroblockd_plane *const plane,
1799 : int mi_row, LOOP_FILTER_MASK *lfm) {
1800 0 : struct buf_2d *const dst = &plane->dst;
1801 0 : uint8_t *const dst0 = dst->buf;
1802 : int r, c;
1803 0 : uint64_t mask_16x16 = lfm->above_uv[TX_16X16];
1804 0 : uint64_t mask_8x8 = lfm->above_uv[TX_8X8];
1805 0 : uint64_t mask_4x4 = lfm->above_uv[TX_4X4];
1806 0 : uint64_t mask_4x4_int = lfm->above_int_4x4_uv;
1807 :
1808 0 : assert(plane->subsampling_x == 1 && plane->subsampling_y == 1);
1809 0 : memset(lfm->lfl_uv, 0, sizeof(lfm->lfl_uv));
1810 :
1811 : // re-porpulate the filter level for uv, same as the code for vertical
1812 : // filter in av1_filter_block_plane_ss11_ver
1813 0 : for (r = 0; r < cm->mib_size && mi_row + r < cm->mi_rows; r += 4) {
1814 0 : for (c = 0; c < (cm->mib_size >> 1); c++) {
1815 0 : lfm->lfl_uv[r >> 1][c] = lfm->lfl_y[r][c << 1];
1816 0 : lfm->lfl_uv[(r + 2) >> 1][c] = lfm->lfl_y[r + 2][c << 1];
1817 : }
1818 : }
1819 :
1820 0 : for (r = 0; r < cm->mib_size && mi_row + r < cm->mi_rows; r += 2) {
1821 0 : const int skip_border_4x4_r = mi_row + r == cm->mi_rows - 1;
1822 0 : const unsigned int mask_4x4_int_r =
1823 0 : skip_border_4x4_r ? 0 : (mask_4x4_int & 0xf);
1824 : unsigned int mask_16x16_r;
1825 : unsigned int mask_8x8_r;
1826 : unsigned int mask_4x4_r;
1827 :
1828 0 : if (mi_row + r == 0) {
1829 0 : mask_16x16_r = 0;
1830 0 : mask_8x8_r = 0;
1831 0 : mask_4x4_r = 0;
1832 : } else {
1833 0 : mask_16x16_r = mask_16x16 & 0xf;
1834 0 : mask_8x8_r = mask_8x8 & 0xf;
1835 0 : mask_4x4_r = mask_4x4 & 0xf;
1836 : }
1837 :
1838 : #if CONFIG_HIGHBITDEPTH
1839 0 : if (cm->use_highbitdepth)
1840 0 : highbd_filter_selectively_horiz(
1841 0 : CONVERT_TO_SHORTPTR(dst->buf), dst->stride, mask_16x16_r, mask_8x8_r,
1842 0 : mask_4x4_r, mask_4x4_int_r, &cm->lf_info, &lfm->lfl_uv[r >> 1][0],
1843 0 : (int)cm->bit_depth);
1844 : else
1845 : #endif // CONFIG_HIGHBITDEPTH
1846 0 : filter_selectively_horiz(dst->buf, dst->stride, mask_16x16_r, mask_8x8_r,
1847 0 : mask_4x4_r, mask_4x4_int_r, &cm->lf_info,
1848 0 : &lfm->lfl_uv[r >> 1][0]);
1849 :
1850 0 : dst->buf += MI_SIZE * dst->stride;
1851 0 : mask_16x16 >>= MI_SIZE / 2;
1852 0 : mask_8x8 >>= MI_SIZE / 2;
1853 0 : mask_4x4 >>= MI_SIZE / 2;
1854 0 : mask_4x4_int >>= MI_SIZE / 2;
1855 : }
1856 : // restore the buf pointer in case there is additional filter pass.
1857 0 : dst->buf = dst0;
1858 0 : }
1859 :
1860 : #if CONFIG_PARALLEL_DEBLOCKING
1861 : typedef enum EDGE_DIR { VERT_EDGE = 0, HORZ_EDGE = 1, NUM_EDGE_DIRS } EDGE_DIR;
1862 : static const uint32_t av1_prediction_masks[NUM_EDGE_DIRS][BLOCK_SIZES] = {
1863 : // mask for vertical edges filtering
1864 : {
1865 : #if CONFIG_CB4X4
1866 : 2 - 1, // BLOCK_2X2
1867 : 2 - 1, // BLOCK_2X4
1868 : 4 - 1, // BLOCK_4X2
1869 : #endif // CONFIG_CB4X4
1870 : 4 - 1, // BLOCK_4X4
1871 : 4 - 1, // BLOCK_4X8
1872 : 8 - 1, // BLOCK_8X4
1873 : 8 - 1, // BLOCK_8X8
1874 : 8 - 1, // BLOCK_8X16
1875 : 16 - 1, // BLOCK_16X8
1876 : 16 - 1, // BLOCK_16X16
1877 : 16 - 1, // BLOCK_16X32
1878 : 32 - 1, // BLOCK_32X16
1879 : 32 - 1, // BLOCK_32X32
1880 : 32 - 1, // BLOCK_32X64
1881 : 64 - 1, // BLOCK_64X32
1882 : 64 - 1, // BLOCK_64X64
1883 : #if CONFIG_EXT_PARTITION
1884 : 64 - 1, // BLOCK_64X128
1885 : 128 - 1, // BLOCK_128X64
1886 : 128 - 1 // BLOCK_128X128
1887 : #endif // CONFIG_EXT_PARTITION
1888 : },
1889 : // mask for horizontal edges filtering
1890 : {
1891 : #if CONFIG_CB4X4
1892 : 2 - 1, // BLOCK_2X2
1893 : 4 - 1, // BLOCK_2X4
1894 : 2 - 1, // BLOCK_4X2
1895 : #endif // CONFIG_CB4X4
1896 : 4 - 1, // BLOCK_4X4
1897 : 8 - 1, // BLOCK_4X8
1898 : 4 - 1, // BLOCK_8X4
1899 : 8 - 1, // BLOCK_8X8
1900 : 16 - 1, // BLOCK_8X16
1901 : 8 - 1, // BLOCK_16X8
1902 : 16 - 1, // BLOCK_16X16
1903 : 32 - 1, // BLOCK_16X32
1904 : 16 - 1, // BLOCK_32X16
1905 : 32 - 1, // BLOCK_32X32
1906 : 64 - 1, // BLOCK_32X64
1907 : 32 - 1, // BLOCK_64X32
1908 : 64 - 1, // BLOCK_64X64
1909 : #if CONFIG_EXT_PARTITION
1910 : 128 - 1, // BLOCK_64X128
1911 : 64 - 1, // BLOCK_128X64
1912 : 128 - 1 // BLOCK_128X128
1913 : #endif // CONFIG_EXT_PARTITION
1914 : },
1915 : };
1916 :
1917 : static const uint32_t av1_transform_masks[NUM_EDGE_DIRS][TX_SIZES_ALL] = {
1918 : {
1919 : #if CONFIG_CHROMA_2X2
1920 : 2 - 1, // TX_2X2
1921 : #endif
1922 : 4 - 1, // TX_4X4
1923 : 8 - 1, // TX_8X8
1924 : 16 - 1, // TX_16X16
1925 : 32 - 1, // TX_32X32
1926 : #if CONFIG_TX64X64
1927 : 64 - 1, // TX_64X64
1928 : #endif // CONFIG_TX64X64
1929 : 4 - 1, // TX_4X8
1930 : 8 - 1, // TX_8X4
1931 : 8 - 1, // TX_8X16
1932 : 16 - 1, // TX_16X8
1933 : 16 - 1, // TX_16X32
1934 : 32 - 1, // TX_32X16
1935 : 4 - 1, // TX_4X16
1936 : 16 - 1, // TX_16X4
1937 : 8 - 1, // TX_8X32
1938 : 32 - 1 // TX_32X8
1939 : },
1940 : {
1941 : #if CONFIG_CHROMA_2X2
1942 : 2 - 1, // TX_2X2
1943 : #endif
1944 : 4 - 1, // TX_4X4
1945 : 8 - 1, // TX_8X8
1946 : 16 - 1, // TX_16X16
1947 : 32 - 1, // TX_32X32
1948 : #if CONFIG_TX64X64
1949 : 64 - 1, // TX_64X64
1950 : #endif // CONFIG_TX64X64
1951 : 8 - 1, // TX_4X8
1952 : 4 - 1, // TX_8X4
1953 : 16 - 1, // TX_8X16
1954 : 8 - 1, // TX_16X8
1955 : 32 - 1, // TX_16X32
1956 : 16 - 1, // TX_32X16
1957 : 16 - 1, // TX_4X16
1958 : 4 - 1, // TX_16X4
1959 : 32 - 1, // TX_8X32
1960 : 8 - 1 // TX_32X8
1961 : }
1962 : };
1963 :
1964 : static TX_SIZE av1_get_transform_size(const MODE_INFO *const pCurr,
1965 : const EDGE_DIR edgeDir,
1966 : const uint32_t scaleHorz,
1967 : const uint32_t scaleVert) {
1968 : const BLOCK_SIZE bs = pCurr->mbmi.sb_type;
1969 : TX_SIZE txSize;
1970 : // since in case of chrominance or non-square transorm need to convert
1971 : // transform size into transform size in particular direction.
1972 : txSize = uv_txsize_lookup[bs][pCurr->mbmi.tx_size][scaleHorz][scaleVert];
1973 : if (VERT_EDGE == edgeDir) {
1974 : txSize = txsize_horz_map[txSize];
1975 : } else {
1976 : txSize = txsize_vert_map[txSize];
1977 : }
1978 : return txSize;
1979 : }
1980 :
1981 : typedef struct AV1_DEBLOCKING_PARAMETERS {
1982 : // length of the filter applied to the outer edge
1983 : uint32_t filterLength;
1984 : // length of the filter applied to the inner edge
1985 : uint32_t filterLengthInternal;
1986 : // deblocking limits
1987 : const uint8_t *lim;
1988 : const uint8_t *mblim;
1989 : const uint8_t *hev_thr;
1990 : } AV1_DEBLOCKING_PARAMETERS;
1991 :
1992 : static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
1993 : const MODE_INFO **const ppCurr,
1994 : const ptrdiff_t modeStep,
1995 : const AV1_COMMON *const cm,
1996 : const EDGE_DIR edgeDir, const uint32_t x,
1997 : const uint32_t y, const uint32_t width,
1998 : const uint32_t height, const uint32_t scaleHorz,
1999 : const uint32_t scaleVert) {
2000 : // reset to initial values
2001 : pParams->filterLength = 0;
2002 : pParams->filterLengthInternal = 0;
2003 : // no deblocking is required
2004 : if ((width <= x) || (height <= y)) {
2005 : return;
2006 : }
2007 : #if CONFIG_EXT_PARTITION
2008 : // not sure if changes are required.
2009 : assert(0 && "Not yet updated");
2010 : #endif // CONFIG_EXT_PARTITION
2011 :
2012 : {
2013 : const TX_SIZE ts =
2014 : av1_get_transform_size(ppCurr[0], edgeDir, scaleHorz, scaleVert);
2015 : #if CONFIG_EXT_DELTA_Q
2016 : const uint32_t currLevel =
2017 : get_filter_level(cm, &cm->lf_info, &ppCurr[0]->mbmi);
2018 : #else
2019 : const uint32_t currLevel = get_filter_level(&cm->lf_info, &ppCurr[0]->mbmi);
2020 : #endif // CONFIG_EXT_DELTA_Q
2021 :
2022 : const int currSkipped =
2023 : ppCurr[0]->mbmi.skip && is_inter_block(&ppCurr[0]->mbmi);
2024 : const uint32_t coord = (VERT_EDGE == edgeDir) ? (x) : (y);
2025 : uint32_t level = currLevel;
2026 : // prepare outer edge parameters. deblock the edge if it's an edge of a TU
2027 : if (coord) {
2028 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
2029 : if (!av1_disable_loopfilter_on_tile_boundary(cm) ||
2030 : ((VERT_EDGE == edgeDir) &&
2031 : (0 == (ppCurr[0]->mbmi.boundary_info & TILE_LEFT_BOUNDARY))) ||
2032 : ((HORZ_EDGE == edgeDir) &&
2033 : (0 == (ppCurr[0]->mbmi.boundary_info & TILE_ABOVE_BOUNDARY))))
2034 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
2035 : {
2036 : const int32_t tuEdge =
2037 : (coord & av1_transform_masks[edgeDir][ts]) ? (0) : (1);
2038 : if (tuEdge) {
2039 : const MODE_INFO *const pPrev = *(ppCurr - modeStep);
2040 : const TX_SIZE pvTs =
2041 : av1_get_transform_size(pPrev, edgeDir, scaleHorz, scaleVert);
2042 : #if CONFIG_EXT_DELTA_Q
2043 : const uint32_t pvLvl =
2044 : get_filter_level(cm, &cm->lf_info, &pPrev->mbmi);
2045 : #else
2046 : const uint32_t pvLvl = get_filter_level(&cm->lf_info, &pPrev->mbmi);
2047 : #endif // CONFIG_EXT_DELTA_Q
2048 :
2049 : const int pvSkip = pPrev->mbmi.skip && is_inter_block(&pPrev->mbmi);
2050 : const int32_t puEdge =
2051 : (coord &
2052 : av1_prediction_masks[edgeDir]
2053 : [ss_size_lookup[ppCurr[0]->mbmi.sb_type]
2054 : [scaleHorz][scaleVert]])
2055 : ? (0)
2056 : : (1);
2057 : // if the current and the previous blocks are skipped,
2058 : // deblock the edge if the edge belongs to a PU's edge only.
2059 : if ((currLevel || pvLvl) && (!pvSkip || !currSkipped || puEdge)) {
2060 : #if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2061 : const TX_SIZE minTs = AOMMIN(ts, pvTs);
2062 : if (TX_4X4 >= minTs) {
2063 : pParams->filterLength = 4;
2064 : } else if (TX_8X8 == minTs) {
2065 : pParams->filterLength = 8;
2066 : } else {
2067 : pParams->filterLength = 16;
2068 : #if PARALLEL_DEBLOCKING_15TAPLUMAONLY
2069 : // No wide filtering for chroma plane
2070 : if (scaleHorz || scaleVert) {
2071 : pParams->filterLength = 8;
2072 : }
2073 : #endif
2074 : }
2075 : #else
2076 : pParams->filterLength = (TX_4X4 >= AOMMIN(ts, pvTs)) ? (4) : (8);
2077 :
2078 : #endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2079 :
2080 : // update the level if the current block is skipped,
2081 : // but the previous one is not
2082 : level = (currLevel) ? (currLevel) : (pvLvl);
2083 : }
2084 : }
2085 : }
2086 :
2087 : #if !CONFIG_CB4X4
2088 : // prepare internal edge parameters
2089 : if (currLevel && !currSkipped) {
2090 : pParams->filterLengthInternal = (TX_4X4 >= ts) ? (4) : (0);
2091 : }
2092 : #endif
2093 :
2094 : // prepare common parameters
2095 : if (pParams->filterLength || pParams->filterLengthInternal) {
2096 : const loop_filter_thresh *const limits = cm->lf_info.lfthr + level;
2097 : pParams->lim = limits->lim;
2098 : pParams->mblim = limits->mblim;
2099 : pParams->hev_thr = limits->hev_thr;
2100 : }
2101 : }
2102 : }
2103 : }
2104 :
2105 : static void av1_filter_block_plane_vert(const AV1_COMMON *const cm,
2106 : const MACROBLOCKD_PLANE *const pPlane,
2107 : const MODE_INFO **ppModeInfo,
2108 : const ptrdiff_t modeStride,
2109 : const uint32_t cuX,
2110 : const uint32_t cuY) {
2111 : const int col_step = MI_SIZE >> MI_SIZE_LOG2;
2112 : const int row_step = MI_SIZE >> MI_SIZE_LOG2;
2113 : const uint32_t scaleHorz = pPlane->subsampling_x;
2114 : const uint32_t scaleVert = pPlane->subsampling_y;
2115 : const uint32_t width = pPlane->dst.width;
2116 : const uint32_t height = pPlane->dst.height;
2117 : uint8_t *const pDst = pPlane->dst.buf;
2118 : const int dstStride = pPlane->dst.stride;
2119 : for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += row_step) {
2120 : uint8_t *p = pDst + y * MI_SIZE * dstStride;
2121 : for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += col_step) {
2122 : // inner loop always filter vertical edges in a MI block. If MI size
2123 : // is 8x8, it will filter the vertical edge aligned with a 8x8 block.
2124 : // If 4x4 trasnform is used, it will then filter the internal edge
2125 : // aligned with a 4x4 block
2126 : const MODE_INFO **const pCurr =
2127 : ppModeInfo + (y << scaleVert) * modeStride + (x << scaleHorz);
2128 : AV1_DEBLOCKING_PARAMETERS params;
2129 : memset(¶ms, 0, sizeof(params));
2130 : set_lpf_parameters(¶ms, pCurr, ((ptrdiff_t)1 << scaleHorz), cm,
2131 : VERT_EDGE, cuX + x * MI_SIZE, cuY + y * MI_SIZE, width,
2132 : height, scaleHorz, scaleVert);
2133 : switch (params.filterLength) {
2134 : // apply 4-tap filtering
2135 : case 4:
2136 : #if CONFIG_HIGHBITDEPTH
2137 : if (cm->use_highbitdepth)
2138 : aom_highbd_lpf_vertical_4_c(CONVERT_TO_SHORTPTR(p), dstStride,
2139 : params.mblim, params.lim,
2140 : params.hev_thr, cm->bit_depth);
2141 : else
2142 : #endif // CONFIG_HIGHBITDEPTH
2143 : aom_lpf_vertical_4_c(p, dstStride, params.mblim, params.lim,
2144 : params.hev_thr);
2145 : break;
2146 : // apply 8-tap filtering
2147 : case 8:
2148 : #if CONFIG_HIGHBITDEPTH
2149 : if (cm->use_highbitdepth)
2150 : aom_highbd_lpf_vertical_8_c(CONVERT_TO_SHORTPTR(p), dstStride,
2151 : params.mblim, params.lim,
2152 : params.hev_thr, cm->bit_depth);
2153 : else
2154 : #endif // CONFIG_HIGHBITDEPTH
2155 : aom_lpf_vertical_8_c(p, dstStride, params.mblim, params.lim,
2156 : params.hev_thr);
2157 : break;
2158 : #if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2159 : // apply 16-tap filtering
2160 : case 16:
2161 : #if CONFIG_HIGHBITDEPTH
2162 : if (cm->use_highbitdepth)
2163 : aom_highbd_lpf_vertical_16_c(CONVERT_TO_SHORTPTR(p), dstStride,
2164 : params.mblim, params.lim,
2165 : params.hev_thr, cm->bit_depth);
2166 : else
2167 : #endif // CONFIG_HIGHBITDEPTH
2168 : aom_lpf_vertical_16_c(p, dstStride, params.mblim, params.lim,
2169 : params.hev_thr);
2170 : break;
2171 : #endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2172 : // no filtering
2173 : default: break;
2174 : }
2175 : // process the internal edge
2176 : if (params.filterLengthInternal) {
2177 : #if CONFIG_HIGHBITDEPTH
2178 : if (cm->use_highbitdepth)
2179 : aom_highbd_lpf_vertical_4_c(CONVERT_TO_SHORTPTR(p + 4), dstStride,
2180 : params.mblim, params.lim, params.hev_thr,
2181 : cm->bit_depth);
2182 : else
2183 : #endif // CONFIG_HIGHBITDEPTH
2184 : aom_lpf_vertical_4_c(p + 4, dstStride, params.mblim, params.lim,
2185 : params.hev_thr);
2186 : }
2187 : // advance the destination pointer
2188 : p += MI_SIZE;
2189 : }
2190 : }
2191 : }
2192 :
2193 : static void av1_filter_block_plane_horz(const AV1_COMMON *const cm,
2194 : const MACROBLOCKD_PLANE *const pPlane,
2195 : const MODE_INFO **ppModeInfo,
2196 : const ptrdiff_t modeStride,
2197 : const uint32_t cuX,
2198 : const uint32_t cuY) {
2199 : const int col_step = MI_SIZE >> MI_SIZE_LOG2;
2200 : const int row_step = MI_SIZE >> MI_SIZE_LOG2;
2201 : const uint32_t scaleHorz = pPlane->subsampling_x;
2202 : const uint32_t scaleVert = pPlane->subsampling_y;
2203 : const uint32_t width = pPlane->dst.width;
2204 : const uint32_t height = pPlane->dst.height;
2205 : uint8_t *const pDst = pPlane->dst.buf;
2206 : const int dstStride = pPlane->dst.stride;
2207 : for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += row_step) {
2208 : uint8_t *p = pDst + y * MI_SIZE * dstStride;
2209 : for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += col_step) {
2210 : // inner loop always filter vertical edges in a MI block. If MI size
2211 : // is 8x8, it will first filter the vertical edge aligned with a 8x8
2212 : // block. If 4x4 trasnform is used, it will then filter the internal
2213 : // edge aligned with a 4x4 block
2214 : const MODE_INFO **const pCurr =
2215 : ppModeInfo + (y << scaleVert) * modeStride + (x << scaleHorz);
2216 : AV1_DEBLOCKING_PARAMETERS params;
2217 : memset(¶ms, 0, sizeof(params));
2218 : set_lpf_parameters(¶ms, pCurr, (modeStride << scaleVert), cm,
2219 : HORZ_EDGE, cuX + x * MI_SIZE, cuY + y * MI_SIZE, width,
2220 : height, scaleHorz, scaleVert);
2221 : switch (params.filterLength) {
2222 : // apply 4-tap filtering
2223 : case 4:
2224 : #if CONFIG_HIGHBITDEPTH
2225 : if (cm->use_highbitdepth)
2226 : aom_highbd_lpf_horizontal_4_c(CONVERT_TO_SHORTPTR(p), dstStride,
2227 : params.mblim, params.lim,
2228 : params.hev_thr, cm->bit_depth);
2229 : else
2230 : #endif // CONFIG_HIGHBITDEPTH
2231 : aom_lpf_horizontal_4_c(p, dstStride, params.mblim, params.lim,
2232 : params.hev_thr);
2233 : break;
2234 : // apply 8-tap filtering
2235 : case 8:
2236 : #if CONFIG_HIGHBITDEPTH
2237 : if (cm->use_highbitdepth)
2238 : aom_highbd_lpf_horizontal_8_c(CONVERT_TO_SHORTPTR(p), dstStride,
2239 : params.mblim, params.lim,
2240 : params.hev_thr, cm->bit_depth);
2241 : else
2242 : #endif // CONFIG_HIGHBITDEPTH
2243 : aom_lpf_horizontal_8_c(p, dstStride, params.mblim, params.lim,
2244 : params.hev_thr);
2245 : break;
2246 : #if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2247 : // apply 16-tap filtering
2248 : case 16:
2249 : #if CONFIG_HIGHBITDEPTH
2250 : if (cm->use_highbitdepth)
2251 : aom_highbd_lpf_horizontal_edge_16_c(
2252 : CONVERT_TO_SHORTPTR(p), dstStride, params.mblim, params.lim,
2253 : params.hev_thr, cm->bit_depth);
2254 : else
2255 : #endif // CONFIG_HIGHBITDEPTH
2256 : aom_lpf_horizontal_edge_16_c(p, dstStride, params.mblim, params.lim,
2257 : params.hev_thr);
2258 : break;
2259 : #endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
2260 : // no filtering
2261 : default: break;
2262 : }
2263 : // process the internal edge
2264 : if (params.filterLengthInternal) {
2265 : #if CONFIG_HIGHBITDEPTH
2266 : if (cm->use_highbitdepth)
2267 : aom_highbd_lpf_horizontal_4_c(CONVERT_TO_SHORTPTR(p + 4 * dstStride),
2268 : dstStride, params.mblim, params.lim,
2269 : params.hev_thr, cm->bit_depth);
2270 : else
2271 : #endif // CONFIG_HIGHBITDEPTH
2272 : aom_lpf_horizontal_4_c(p + 4 * dstStride, dstStride, params.mblim,
2273 : params.lim, params.hev_thr);
2274 : }
2275 : // advance the destination pointer
2276 : p += MI_SIZE;
2277 : }
2278 : }
2279 : }
2280 : #endif // CONFIG_PARALLEL_DEBLOCKING
2281 :
2282 0 : void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
2283 : struct macroblockd_plane planes[MAX_MB_PLANE],
2284 : int start, int stop, int y_only) {
2285 0 : const int num_planes = y_only ? 1 : MAX_MB_PLANE;
2286 : int mi_row, mi_col;
2287 :
2288 : #if CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES || \
2289 : CONFIG_CB4X4
2290 :
2291 : #if !CONFIG_PARALLEL_DEBLOCKING
2292 : #if CONFIG_VAR_TX
2293 0 : for (int i = 0; i < MAX_MB_PLANE; ++i)
2294 0 : memset(cm->top_txfm_context[i], TX_32X32, cm->mi_cols << TX_UNIT_WIDE_LOG2);
2295 : #endif // CONFIG_VAR_TX
2296 0 : for (mi_row = start; mi_row < stop; mi_row += cm->mib_size) {
2297 0 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2298 : #if CONFIG_VAR_TX
2299 0 : for (int i = 0; i < MAX_MB_PLANE; ++i)
2300 0 : memset(cm->left_txfm_context[i], TX_32X32, MAX_MIB_SIZE
2301 0 : << TX_UNIT_WIDE_LOG2);
2302 : #endif // CONFIG_VAR_TX
2303 0 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += cm->mib_size) {
2304 : int plane;
2305 :
2306 0 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2307 :
2308 0 : for (plane = 0; plane < num_planes; ++plane) {
2309 0 : av1_filter_block_plane_non420_ver(cm, &planes[plane], mi + mi_col,
2310 : mi_row, mi_col, plane);
2311 0 : av1_filter_block_plane_non420_hor(cm, &planes[plane], mi + mi_col,
2312 : mi_row, mi_col, plane);
2313 : }
2314 : }
2315 : }
2316 : #else
2317 :
2318 : #if CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
2319 : assert(0 && "Not yet updated. ToDo as next steps");
2320 : #endif // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
2321 :
2322 : for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
2323 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2324 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
2325 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2326 : // filter all vertical edges in every 64x64 super block
2327 : for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
2328 : const int32_t scaleHorz = planes[planeIdx].subsampling_x;
2329 : const int32_t scaleVert = planes[planeIdx].subsampling_y;
2330 : av1_filter_block_plane_vert(
2331 : cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
2332 : cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
2333 : (mi_row * MI_SIZE) >> scaleVert);
2334 : }
2335 : }
2336 : }
2337 : for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
2338 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2339 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
2340 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2341 : // filter all horizontal edges in every 64x64 super block
2342 : for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
2343 : const int32_t scaleHorz = planes[planeIdx].subsampling_x;
2344 : const int32_t scaleVert = planes[planeIdx].subsampling_y;
2345 : av1_filter_block_plane_horz(
2346 : cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
2347 : cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
2348 : (mi_row * MI_SIZE) >> scaleVert);
2349 : }
2350 : }
2351 : }
2352 : #endif // CONFIG_PARALLEL_DEBLOCKING
2353 :
2354 : #else // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
2355 :
2356 : #if CONFIG_PARALLEL_DEBLOCKING
2357 : for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
2358 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2359 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
2360 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2361 : // filter all vertical edges in every 64x64 super block
2362 : for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
2363 : const int32_t scaleHorz = planes[planeIdx].subsampling_x;
2364 : const int32_t scaleVert = planes[planeIdx].subsampling_y;
2365 : av1_filter_block_plane_vert(
2366 : cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
2367 : cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
2368 : (mi_row * MI_SIZE) >> scaleVert);
2369 : }
2370 : }
2371 : }
2372 : for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
2373 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2374 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
2375 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2376 : // filter all horizontal edges in every 64x64 super block
2377 : for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
2378 : const int32_t scaleHorz = planes[planeIdx].subsampling_x;
2379 : const int32_t scaleVert = planes[planeIdx].subsampling_y;
2380 : av1_filter_block_plane_horz(
2381 : cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
2382 : cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
2383 : (mi_row * MI_SIZE) >> scaleVert);
2384 : }
2385 : }
2386 : }
2387 : #else // CONFIG_PARALLEL_DEBLOCKING
2388 : enum lf_path path;
2389 : LOOP_FILTER_MASK lfm;
2390 :
2391 : if (y_only)
2392 : path = LF_PATH_444;
2393 : else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
2394 : path = LF_PATH_420;
2395 : else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
2396 : path = LF_PATH_444;
2397 : else
2398 : path = LF_PATH_SLOW;
2399 :
2400 : for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
2401 : MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
2402 : for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
2403 : int plane;
2404 :
2405 : av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
2406 :
2407 : // TODO(JBB): Make setup_mask work for non 420.
2408 : av1_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride, &lfm);
2409 :
2410 : av1_filter_block_plane_ss00_ver(cm, &planes[0], mi_row, &lfm);
2411 : av1_filter_block_plane_ss00_hor(cm, &planes[0], mi_row, &lfm);
2412 : for (plane = 1; plane < num_planes; ++plane) {
2413 : switch (path) {
2414 : case LF_PATH_420:
2415 : av1_filter_block_plane_ss11_ver(cm, &planes[plane], mi_row, &lfm);
2416 : av1_filter_block_plane_ss11_hor(cm, &planes[plane], mi_row, &lfm);
2417 : break;
2418 : case LF_PATH_444:
2419 : av1_filter_block_plane_ss00_ver(cm, &planes[plane], mi_row, &lfm);
2420 : av1_filter_block_plane_ss00_hor(cm, &planes[plane], mi_row, &lfm);
2421 : break;
2422 : case LF_PATH_SLOW:
2423 : av1_filter_block_plane_non420_ver(cm, &planes[plane], mi + mi_col,
2424 : mi_row, mi_col, plane);
2425 : av1_filter_block_plane_non420_hor(cm, &planes[plane], mi + mi_col,
2426 : mi_row, mi_col, plane);
2427 :
2428 : break;
2429 : }
2430 : }
2431 : }
2432 : }
2433 : #endif // CONFIG_PARALLEL_DEBLOCKING
2434 : #endif // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
2435 0 : }
2436 :
2437 0 : void av1_loop_filter_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
2438 : MACROBLOCKD *xd, int frame_filter_level, int y_only,
2439 : int partial_frame) {
2440 : int start_mi_row, end_mi_row, mi_rows_to_filter;
2441 : #if CONFIG_EXT_DELTA_Q
2442 0 : int orig_filter_level = cm->lf.filter_level;
2443 : #endif
2444 0 : if (!frame_filter_level) return;
2445 0 : start_mi_row = 0;
2446 0 : mi_rows_to_filter = cm->mi_rows;
2447 0 : if (partial_frame && cm->mi_rows > 8) {
2448 0 : start_mi_row = cm->mi_rows >> 1;
2449 0 : start_mi_row &= 0xfffffff8;
2450 0 : mi_rows_to_filter = AOMMAX(cm->mi_rows / 8, 8);
2451 : }
2452 0 : end_mi_row = start_mi_row + mi_rows_to_filter;
2453 0 : av1_loop_filter_frame_init(cm, frame_filter_level);
2454 : #if CONFIG_EXT_DELTA_Q
2455 0 : cm->lf.filter_level = frame_filter_level;
2456 : #endif
2457 0 : av1_loop_filter_rows(frame, cm, xd->plane, start_mi_row, end_mi_row, y_only);
2458 : #if CONFIG_EXT_DELTA_Q
2459 0 : cm->lf.filter_level = orig_filter_level;
2460 : #endif
2461 : }
2462 :
2463 0 : void av1_loop_filter_data_reset(
2464 : LFWorkerData *lf_data, YV12_BUFFER_CONFIG *frame_buffer,
2465 : struct AV1Common *cm, const struct macroblockd_plane planes[MAX_MB_PLANE]) {
2466 0 : lf_data->frame_buffer = frame_buffer;
2467 0 : lf_data->cm = cm;
2468 0 : lf_data->start = 0;
2469 0 : lf_data->stop = 0;
2470 0 : lf_data->y_only = 0;
2471 0 : memcpy(lf_data->planes, planes, sizeof(lf_data->planes));
2472 0 : }
2473 :
2474 0 : int av1_loop_filter_worker(LFWorkerData *const lf_data, void *unused) {
2475 : (void)unused;
2476 0 : av1_loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
2477 : lf_data->start, lf_data->stop, lf_data->y_only);
2478 0 : return 1;
2479 : }
|
