FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/vp9recon.c
Date: 2024-04-24 13:31:03
Exec Total Coverage
Lines: 274 288 95.1%
Functions: 11 11 100.0%
Branches: 297 322 92.2%

Line Branch Exec Source
1 /*
2 * VP9 compatible video decoder
3 *
4 * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
5 * Copyright (C) 2013 Clément Bœsch <u pkh me>
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 #include "libavutil/avassert.h"
25 #include "libavutil/frame.h"
26 #include "libavutil/mem_internal.h"
27
28 #include "progressframe.h"
29 #include "videodsp.h"
30 #include "vp9data.h"
31 #include "vp9dec.h"
32
33 1535849 static av_always_inline int check_intra_mode(VP9TileData *td, int mode, uint8_t **a,
34 uint8_t *dst_edge, ptrdiff_t stride_edge,
35 uint8_t *dst_inner, ptrdiff_t stride_inner,
36 uint8_t *l, int col, int x, int w,
37 int row, int y, enum TxfmMode tx,
38 int p, int ss_h, int ss_v, int bytesperpixel)
39 {
40 1535849 const VP9Context *s = td->s;
41
4/4
✓ Branch 0 taken 58564 times.
✓ Branch 1 taken 1477285 times.
✓ Branch 2 taken 17329 times.
✓ Branch 3 taken 41235 times.
1535849 int have_top = row > 0 || y > 0;
42
4/4
✓ Branch 0 taken 52903 times.
✓ Branch 1 taken 1482946 times.
✓ Branch 2 taken 17408 times.
✓ Branch 3 taken 35495 times.
1535849 int have_left = col > td->tile_col_start || x > 0;
43 1535849 int have_right = x < w - 1;
44 1535849 int bpp = s->s.h.bpp;
45 static const uint8_t mode_conv[10][2 /* have_left */][2 /* have_top */] = {
46 [VERT_PRED] = { { DC_127_PRED, VERT_PRED },
47 { DC_127_PRED, VERT_PRED } },
48 [HOR_PRED] = { { DC_129_PRED, DC_129_PRED },
49 { HOR_PRED, HOR_PRED } },
50 [DC_PRED] = { { DC_128_PRED, TOP_DC_PRED },
51 { LEFT_DC_PRED, DC_PRED } },
52 [DIAG_DOWN_LEFT_PRED] = { { DC_127_PRED, DIAG_DOWN_LEFT_PRED },
53 { DC_127_PRED, DIAG_DOWN_LEFT_PRED } },
54 [DIAG_DOWN_RIGHT_PRED] = { { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED },
55 { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED } },
56 [VERT_RIGHT_PRED] = { { VERT_RIGHT_PRED, VERT_RIGHT_PRED },
57 { VERT_RIGHT_PRED, VERT_RIGHT_PRED } },
58 [HOR_DOWN_PRED] = { { HOR_DOWN_PRED, HOR_DOWN_PRED },
59 { HOR_DOWN_PRED, HOR_DOWN_PRED } },
60 [VERT_LEFT_PRED] = { { DC_127_PRED, VERT_LEFT_PRED },
61 { DC_127_PRED, VERT_LEFT_PRED } },
62 [HOR_UP_PRED] = { { DC_129_PRED, DC_129_PRED },
63 { HOR_UP_PRED, HOR_UP_PRED } },
64 [TM_VP8_PRED] = { { DC_129_PRED, VERT_PRED },
65 { HOR_PRED, TM_VP8_PRED } },
66 };
67 static const struct {
68 uint8_t needs_left:1;
69 uint8_t needs_top:1;
70 uint8_t needs_topleft:1;
71 uint8_t needs_topright:1;
72 uint8_t invert_left:1;
73 } edges[N_INTRA_PRED_MODES] = {
74 [VERT_PRED] = { .needs_top = 1 },
75 [HOR_PRED] = { .needs_left = 1 },
76 [DC_PRED] = { .needs_top = 1, .needs_left = 1 },
77 [DIAG_DOWN_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
78 [DIAG_DOWN_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1,
79 .needs_topleft = 1 },
80 [VERT_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1,
81 .needs_topleft = 1 },
82 [HOR_DOWN_PRED] = { .needs_left = 1, .needs_top = 1,
83 .needs_topleft = 1 },
84 [VERT_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
85 [HOR_UP_PRED] = { .needs_left = 1, .invert_left = 1 },
86 [TM_VP8_PRED] = { .needs_left = 1, .needs_top = 1,
87 .needs_topleft = 1 },
88 [LEFT_DC_PRED] = { .needs_left = 1 },
89 [TOP_DC_PRED] = { .needs_top = 1 },
90 [DC_128_PRED] = { 0 },
91 [DC_127_PRED] = { 0 },
92 [DC_129_PRED] = { 0 }
93 };
94
95 av_assert2(mode >= 0 && mode < 10);
96 1535849 mode = mode_conv[mode][have_left][have_top];
97
2/2
✓ Branch 0 taken 1078515 times.
✓ Branch 1 taken 457334 times.
1535849 if (edges[mode].needs_top) {
98 uint8_t *top, *topleft;
99 1078515 int n_px_need = 4 << tx, n_px_have = (((s->cols - col) << !ss_h) - x) * 4;
100 1078515 int n_px_need_tr = 0;
101
102
6/6
✓ Branch 0 taken 795515 times.
✓ Branch 1 taken 283000 times.
✓ Branch 2 taken 57929 times.
✓ Branch 3 taken 737586 times.
✓ Branch 4 taken 22995 times.
✓ Branch 5 taken 34934 times.
1078515 if (tx == TX_4X4 && edges[mode].needs_topright && have_right)
103 22995 n_px_need_tr = 4;
104
105 // if top of sb64-row, use s->intra_pred_data[] instead of
106 // dst[-stride] for intra prediction (it contains pre- instead of
107 // post-loopfilter data)
108
2/2
✓ Branch 0 taken 1074180 times.
✓ Branch 1 taken 4335 times.
1078515 if (have_top) {
109
2/2
✓ Branch 0 taken 114535 times.
✓ Branch 1 taken 101130 times.
215665 top = !(row & 7) && !y ?
110
2/2
✓ Branch 0 taken 215665 times.
✓ Branch 1 taken 858515 times.
2148360 s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
111
2/2
✓ Branch 0 taken 648210 times.
✓ Branch 1 taken 311435 times.
959645 y == 0 ? &dst_edge[-stride_edge] : &dst_inner[-stride_inner];
112
2/2
✓ Branch 0 taken 1042933 times.
✓ Branch 1 taken 31247 times.
1074180 if (have_left)
113
2/2
✓ Branch 0 taken 109440 times.
✓ Branch 1 taken 98834 times.
1251207 topleft = !(row & 7) && !y ?
114
2/2
✓ Branch 0 taken 208274 times.
✓ Branch 1 taken 834659 times.
2085866 s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
115
4/4
✓ Branch 0 taken 304747 times.
✓ Branch 1 taken 628746 times.
✓ Branch 2 taken 136405 times.
✓ Branch 3 taken 168342 times.
933493 y == 0 || x == 0 ? &dst_edge[-stride_edge] :
116 168342 &dst_inner[-stride_inner];
117 }
118
119
2/2
✓ Branch 0 taken 1074180 times.
✓ Branch 1 taken 4335 times.
1078515 if (have_top &&
120
8/8
✓ Branch 0 taken 269669 times.
✓ Branch 1 taken 804511 times.
✓ Branch 2 taken 266887 times.
✓ Branch 3 taken 2782 times.
✓ Branch 4 taken 266872 times.
✓ Branch 5 taken 15 times.
✓ Branch 6 taken 789266 times.
✓ Branch 7 taken 282117 times.
1074180 (!edges[mode].needs_topleft || (have_left && top == topleft)) &&
121
4/4
✓ Branch 0 taken 57929 times.
✓ Branch 1 taken 731337 times.
✓ Branch 2 taken 22995 times.
✓ Branch 3 taken 34934 times.
789266 (tx != TX_4X4 || !edges[mode].needs_topright || have_right) &&
122
2/2
✓ Branch 0 taken 1036295 times.
✓ Branch 1 taken 154 times.
1036449 n_px_need + n_px_need_tr <= n_px_have) {
123 1036295 *a = top;
124 } else {
125
2/2
✓ Branch 0 taken 37885 times.
✓ Branch 1 taken 4335 times.
42220 if (have_top) {
126
2/2
✓ Branch 0 taken 37747 times.
✓ Branch 1 taken 138 times.
37885 if (n_px_need <= n_px_have) {
127 37747 memcpy(*a, top, n_px_need * bytesperpixel);
128 } else {
129 #define memset_bpp(c, i1, v, i2, num) do { \
130 if (bytesperpixel == 1) { \
131 memset(&(c)[(i1)], (v)[(i2)], (num)); \
132 } else { \
133 int n, val = AV_RN16A(&(v)[(i2) * 2]); \
134 for (n = 0; n < (num); n++) { \
135 AV_WN16A(&(c)[((i1) + n) * 2], val); \
136 } \
137 } \
138 } while (0)
139 138 memcpy(*a, top, n_px_have * bytesperpixel);
140
1/4
✓ Branch 0 taken 138 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
138 memset_bpp(*a, n_px_have, (*a), n_px_have - 1, n_px_need - n_px_have);
141 }
142 } else {
143 #define memset_val(c, val, num) do { \
144 if (bytesperpixel == 1) { \
145 memset((c), (val), (num)); \
146 } else { \
147 int n; \
148 for (n = 0; n < (num); n++) { \
149 AV_WN16A(&(c)[n * 2], (val)); \
150 } \
151 } \
152 } while (0)
153
4/4
✓ Branch 0 taken 4127 times.
✓ Branch 1 taken 208 times.
✓ Branch 2 taken 848 times.
✓ Branch 3 taken 208 times.
5183 memset_val(*a, (128 << (bpp - 8)) - 1, n_px_need);
154 }
155
2/2
✓ Branch 0 taken 7168 times.
✓ Branch 1 taken 35052 times.
42220 if (edges[mode].needs_topleft) {
156
4/4
✓ Branch 0 taken 4137 times.
✓ Branch 1 taken 3031 times.
✓ Branch 2 taken 51 times.
✓ Branch 3 taken 4086 times.
7168 if (have_left && have_top) {
157 #define assign_bpp(c, i1, v, i2) do { \
158 if (bytesperpixel == 1) { \
159 (c)[(i1)] = (v)[(i2)]; \
160 } else { \
161 AV_COPY16(&(c)[(i1) * 2], &(v)[(i2) * 2]); \
162 } \
163 } while (0)
164
1/2
✓ Branch 0 taken 51 times.
✗ Branch 1 not taken.
51 assign_bpp(*a, -1, topleft, -1);
165 } else {
166 #define assign_val(c, i, v) do { \
167 if (bytesperpixel == 1) { \
168 (c)[(i)] = (v); \
169 } else { \
170 AV_WN16A(&(c)[(i) * 2], (v)); \
171 } \
172 } while (0)
173
6/6
✓ Branch 0 taken 6818 times.
✓ Branch 1 taken 299 times.
✓ Branch 2 taken 2691 times.
✓ Branch 3 taken 4127 times.
✓ Branch 4 taken 91 times.
✓ Branch 5 taken 208 times.
7117 assign_val((*a), -1, (128 << (bpp - 8)) + (have_top ? +1 : -1));
174 }
175 }
176
4/4
✓ Branch 0 taken 41199 times.
✓ Branch 1 taken 1021 times.
✓ Branch 2 taken 34950 times.
✓ Branch 3 taken 6249 times.
42220 if (tx == TX_4X4 && edges[mode].needs_topright) {
177
3/4
✓ Branch 0 taken 34950 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 16 times.
✓ Branch 3 taken 34934 times.
34950 if (have_top && have_right &&
178
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 16 times.
16 n_px_need + n_px_need_tr <= n_px_have) {
179 memcpy(&(*a)[4 * bytesperpixel], &top[4 * bytesperpixel], 4 * bytesperpixel);
180 } else {
181
4/4
✓ Branch 0 taken 33220 times.
✓ Branch 1 taken 1730 times.
✓ Branch 2 taken 6920 times.
✓ Branch 3 taken 1730 times.
41870 memset_bpp(*a, 4, *a, 3, 4);
182 }
183 }
184 }
185 }
186
2/2
✓ Branch 0 taken 1267290 times.
✓ Branch 1 taken 268559 times.
1535849 if (edges[mode].needs_left) {
187
2/2
✓ Branch 0 taken 1264259 times.
✓ Branch 1 taken 3031 times.
1267290 if (have_left) {
188 1264259 int n_px_need = 4 << tx, i, n_px_have = (((s->rows - row) << !ss_v) - y) * 4;
189
2/2
✓ Branch 0 taken 861511 times.
✓ Branch 1 taken 402748 times.
1264259 uint8_t *dst = x == 0 ? dst_edge : dst_inner;
190
2/2
✓ Branch 0 taken 861511 times.
✓ Branch 1 taken 402748 times.
1264259 ptrdiff_t stride = x == 0 ? stride_edge : stride_inner;
191
192
2/2
✓ Branch 0 taken 63656 times.
✓ Branch 1 taken 1200603 times.
1264259 if (edges[mode].invert_left) {
193
2/2
✓ Branch 0 taken 63648 times.
✓ Branch 1 taken 8 times.
63656 if (n_px_need <= n_px_have) {
194
2/2
✓ Branch 0 taken 319740 times.
✓ Branch 1 taken 63648 times.
383388 for (i = 0; i < n_px_need; i++)
195
2/2
✓ Branch 0 taken 306388 times.
✓ Branch 1 taken 13352 times.
319740 assign_bpp(l, i, &dst[i * stride], -1);
196 } else {
197
2/2
✓ Branch 0 taken 96 times.
✓ Branch 1 taken 8 times.
104 for (i = 0; i < n_px_have; i++)
198
1/2
✓ Branch 0 taken 96 times.
✗ Branch 1 not taken.
96 assign_bpp(l, i, &dst[i * stride], -1);
199
1/4
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
8 memset_bpp(l, n_px_have, l, n_px_have - 1, n_px_need - n_px_have);
200 }
201 } else {
202
2/2
✓ Branch 0 taken 1199057 times.
✓ Branch 1 taken 1546 times.
1200603 if (n_px_need <= n_px_have) {
203
2/2
✓ Branch 0 taken 6636792 times.
✓ Branch 1 taken 1199057 times.
7835849 for (i = 0; i < n_px_need; i++)
204
2/2
✓ Branch 0 taken 6003888 times.
✓ Branch 1 taken 632904 times.
6636792 assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
205 } else {
206
2/2
✓ Branch 0 taken 9144 times.
✓ Branch 1 taken 1546 times.
10690 for (i = 0; i < n_px_have; i++)
207
2/2
✓ Branch 0 taken 6648 times.
✓ Branch 1 taken 2496 times.
9144 assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
208
4/4
✓ Branch 0 taken 1312 times.
✓ Branch 1 taken 234 times.
✓ Branch 2 taken 2496 times.
✓ Branch 3 taken 234 times.
4042 memset_bpp(l, 0, l, n_px_need - n_px_have, n_px_need - n_px_have);
209 }
210 }
211 } else {
212
4/4
✓ Branch 0 taken 2930 times.
✓ Branch 1 taken 101 times.
✓ Branch 2 taken 852 times.
✓ Branch 3 taken 101 times.
3883 memset_val(l, (128 << (bpp - 8)) + 1, 4 << tx);
213 }
214 }
215
216 1535849 return mode;
217 }
218
219 284950 static av_always_inline void intra_recon(VP9TileData *td, ptrdiff_t y_off,
220 ptrdiff_t uv_off, int bytesperpixel)
221 {
222 284950 const VP9Context *s = td->s;
223 284950 VP9Block *b = td->b;
224 284950 int row = td->row, col = td->col;
225 284950 int w4 = ff_vp9_bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
226 284950 int h4 = ff_vp9_bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
227 284950 int end_x = FFMIN(2 * (s->cols - col), w4);
228 284950 int end_y = FFMIN(2 * (s->rows - row), h4);
229 284950 int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
230 284950 int uvstep1d = 1 << b->uvtx, p;
231 284950 uint8_t *dst = td->dst[0], *dst_r = s->s.frames[CUR_FRAME].tf.f->data[0] + y_off;
232 284950 LOCAL_ALIGNED_32(uint8_t, a_buf, [96]);
233 284950 LOCAL_ALIGNED_32(uint8_t, l, [64]);
234
235
2/2
✓ Branch 0 taken 444628 times.
✓ Branch 1 taken 284950 times.
729578 for (n = 0, y = 0; y < end_y; y += step1d) {
236 444628 uint8_t *ptr = dst, *ptr_r = dst_r;
237
2/2
✓ Branch 0 taken 832605 times.
✓ Branch 1 taken 444628 times.
1277233 for (x = 0; x < end_x; x += step1d, ptr += 4 * step1d * bytesperpixel,
238 832605 ptr_r += 4 * step1d * bytesperpixel, n += step) {
239
1/2
✓ Branch 0 taken 345912 times.
✗ Branch 1 not taken.
345912 int mode = b->mode[b->bs > BS_8x8 && b->tx == TX_4X4 ?
240
2/2
✓ Branch 0 taken 345912 times.
✓ Branch 1 taken 486693 times.
1178517 y * 2 + x : 0];
241 832605 uint8_t *a = &a_buf[32];
242 832605 enum TxfmType txtp = ff_vp9_intra_txfm_type[mode];
243
4/4
✓ Branch 0 taken 724884 times.
✓ Branch 1 taken 107721 times.
✓ Branch 2 taken 42491 times.
✓ Branch 3 taken 682393 times.
832605 int eob = b->skip ? 0 : b->tx > TX_8X8 ? AV_RN16A(&td->eob[n]) : td->eob[n];
244
245 832605 mode = check_intra_mode(td, mode, &a, ptr_r,
246 832605 s->s.frames[CUR_FRAME].tf.f->linesize[0],
247 ptr, td->y_stride, l,
248 col, x, w4, row, y, b->tx, 0, 0, 0, bytesperpixel);
249 832605 s->dsp.intra_pred[b->tx][mode](ptr, td->y_stride, l, a);
250
2/2
✓ Branch 0 taken 575338 times.
✓ Branch 1 taken 257267 times.
832605 if (eob)
251 575338 s->dsp.itxfm_add[tx][txtp](ptr, td->y_stride,
252 575338 td->block + 16 * n * bytesperpixel, eob);
253 }
254 444628 dst_r += 4 * step1d * s->s.frames[CUR_FRAME].tf.f->linesize[0];
255 444628 dst += 4 * step1d * td->y_stride;
256 }
257
258 // U/V
259 284950 w4 >>= s->ss_h;
260 284950 end_x >>= s->ss_h;
261 284950 end_y >>= s->ss_v;
262 284950 step = 1 << (b->uvtx * 2);
263
2/2
✓ Branch 0 taken 569900 times.
✓ Branch 1 taken 284950 times.
854850 for (p = 0; p < 2; p++) {
264 569900 dst = td->dst[1 + p];
265 569900 dst_r = s->s.frames[CUR_FRAME].tf.f->data[1 + p] + uv_off;
266
2/2
✓ Branch 0 taken 613900 times.
✓ Branch 1 taken 569900 times.
1183800 for (n = 0, y = 0; y < end_y; y += uvstep1d) {
267 613900 uint8_t *ptr = dst, *ptr_r = dst_r;
268
2/2
✓ Branch 0 taken 703244 times.
✓ Branch 1 taken 613900 times.
1317144 for (x = 0; x < end_x; x += uvstep1d, ptr += 4 * uvstep1d * bytesperpixel,
269 703244 ptr_r += 4 * uvstep1d * bytesperpixel, n += step) {
270 703244 int mode = b->uvmode;
271 703244 uint8_t *a = &a_buf[32];
272
4/4
✓ Branch 0 taken 583454 times.
✓ Branch 1 taken 119790 times.
✓ Branch 2 taken 16086 times.
✓ Branch 3 taken 567368 times.
703244 int eob = b->skip ? 0 : b->uvtx > TX_8X8 ? AV_RN16A(&td->uveob[p][n]) : td->uveob[p][n];
273
274 703244 mode = check_intra_mode(td, mode, &a, ptr_r,
275 703244 s->s.frames[CUR_FRAME].tf.f->linesize[1],
276 ptr, td->uv_stride, l, col, x, w4, row, y,
277 703244 b->uvtx, p + 1, s->ss_h, s->ss_v, bytesperpixel);
278 703244 s->dsp.intra_pred[b->uvtx][mode](ptr, td->uv_stride, l, a);
279
2/2
✓ Branch 0 taken 234008 times.
✓ Branch 1 taken 469236 times.
703244 if (eob)
280 234008 s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, td->uv_stride,
281 234008 td->uvblock[p] + 16 * n * bytesperpixel, eob);
282 }
283 613900 dst_r += 4 * uvstep1d * s->s.frames[CUR_FRAME].tf.f->linesize[1];
284 613900 dst += 4 * uvstep1d * td->uv_stride;
285 }
286 }
287 284950 }
288
289 260921 void ff_vp9_intra_recon_8bpp(VP9TileData *td, ptrdiff_t y_off, ptrdiff_t uv_off)
290 {
291 260921 intra_recon(td, y_off, uv_off, 1);
292 260921 }
293
294 24029 void ff_vp9_intra_recon_16bpp(VP9TileData *td, ptrdiff_t y_off, ptrdiff_t uv_off)
295 {
296 24029 intra_recon(td, y_off, uv_off, 2);
297 24029 }
298
299 642439 static av_always_inline void mc_luma_unscaled(VP9TileData *td, const vp9_mc_func (*mc)[2],
300 uint8_t *dst, ptrdiff_t dst_stride,
301 const uint8_t *ref, ptrdiff_t ref_stride,
302 const ProgressFrame *ref_frame,
303 ptrdiff_t y, ptrdiff_t x, const VP9mv *mv,
304 int bw, int bh, int w, int h, int bytesperpixel)
305 {
306 642439 const VP9Context *s = td->s;
307 642439 int mx = mv->x, my = mv->y, th;
308
309 642439 y += my >> 3;
310 642439 x += mx >> 3;
311 642439 ref += y * ref_stride + x * bytesperpixel;
312 642439 mx &= 7;
313 642439 my &= 7;
314 // FIXME bilinear filter only needs 0/1 pixels, not 3/4
315 // we use +7 because the last 7 pixels of each sbrow can be changed in
316 // the longest loopfilter of the next sbrow
317
2/2
✓ Branch 0 taken 439226 times.
✓ Branch 1 taken 203213 times.
642439 th = (y + bh + 4 * !!my + 7) >> 6;
318 642439 ff_progress_frame_await(ref_frame, FFMAX(th, 0));
319 // The arm/aarch64 _hv filters read one more row than what actually is
320 // needed, so switch to emulated edge one pixel sooner vertically
321 // (!!my * 5) than horizontally (!!mx * 4).
322
8/8
✓ Branch 0 taken 467327 times.
✓ Branch 1 taken 175112 times.
✓ Branch 2 taken 637433 times.
✓ Branch 3 taken 5006 times.
✓ Branch 4 taken 436615 times.
✓ Branch 5 taken 200818 times.
✓ Branch 6 taken 620001 times.
✓ Branch 7 taken 17432 times.
642439 if (x < !!mx * 3 || y < !!my * 3 ||
323
8/8
✓ Branch 0 taken 448749 times.
✓ Branch 1 taken 171252 times.
✓ Branch 2 taken 593770 times.
✓ Branch 3 taken 26231 times.
✓ Branch 4 taken 404486 times.
✓ Branch 5 taken 189284 times.
✓ Branch 6 taken 20872 times.
✓ Branch 7 taken 572898 times.
620001 x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
324 417246 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
325
4/4
✓ Branch 0 taken 48794 times.
✓ Branch 1 taken 20747 times.
✓ Branch 2 taken 50515 times.
✓ Branch 3 taken 19026 times.
69541 ref - !!my * 3 * ref_stride - !!mx * 3 * bytesperpixel,
326 160, ref_stride,
327
2/2
✓ Branch 0 taken 50515 times.
✓ Branch 1 taken 19026 times.
69541 bw + !!mx * 7, bh + !!my * 7,
328
6/6
✓ Branch 0 taken 48794 times.
✓ Branch 1 taken 20747 times.
✓ Branch 2 taken 50515 times.
✓ Branch 3 taken 19026 times.
✓ Branch 4 taken 48794 times.
✓ Branch 5 taken 20747 times.
69541 x - !!mx * 3, y - !!my * 3, w, h);
329
4/4
✓ Branch 0 taken 48794 times.
✓ Branch 1 taken 20747 times.
✓ Branch 2 taken 50515 times.
✓ Branch 3 taken 19026 times.
69541 ref = td->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
330 69541 ref_stride = 160;
331 }
332
2/2
✓ Branch 0 taken 467327 times.
✓ Branch 1 taken 175112 times.
642439 mc[!!mx][!!my](dst, dst_stride, ref, ref_stride, bh, mx << 1, my << 1);
333 642439 }
334
335 561917 static av_always_inline void mc_chroma_unscaled(VP9TileData *td, const vp9_mc_func (*mc)[2],
336 uint8_t *dst_u, uint8_t *dst_v,
337 ptrdiff_t dst_stride,
338 const uint8_t *ref_u, ptrdiff_t src_stride_u,
339 const uint8_t *ref_v, ptrdiff_t src_stride_v,
340 const ProgressFrame *ref_frame,
341 ptrdiff_t y, ptrdiff_t x, const VP9mv *mv,
342 int bw, int bh, int w, int h, int bytesperpixel)
343 {
344 561917 const VP9Context *s = td->s;
345
4/4
✓ Branch 0 taken 20366 times.
✓ Branch 1 taken 541551 times.
✓ Branch 2 taken 16495 times.
✓ Branch 3 taken 545422 times.
561917 int mx = mv->x * (1 << !s->ss_h), my = mv->y * (1 << !s->ss_v), th;
346
347 561917 y += my >> 4;
348 561917 x += mx >> 4;
349 561917 ref_u += y * src_stride_u + x * bytesperpixel;
350 561917 ref_v += y * src_stride_v + x * bytesperpixel;
351 561917 mx &= 15;
352 561917 my &= 15;
353 // FIXME bilinear filter only needs 0/1 pixels, not 3/4
354 // we use +7 because the last 7 pixels of each sbrow can be changed in
355 // the longest loopfilter of the next sbrow
356
2/2
✓ Branch 0 taken 437482 times.
✓ Branch 1 taken 124435 times.
561917 th = (y + bh + 4 * !!my + 7) >> (6 - s->ss_v);
357 561917 ff_progress_frame_await(ref_frame, FFMAX(th, 0));
358 // The arm/aarch64 _hv filters read one more row than what actually is
359 // needed, so switch to emulated edge one pixel sooner vertically
360 // (!!my * 5) than horizontally (!!mx * 4).
361
8/8
✓ Branch 0 taken 464134 times.
✓ Branch 1 taken 97783 times.
✓ Branch 2 taken 549715 times.
✓ Branch 3 taken 12202 times.
✓ Branch 4 taken 428439 times.
✓ Branch 5 taken 121276 times.
✓ Branch 6 taken 532220 times.
✓ Branch 7 taken 17495 times.
561917 if (x < !!mx * 3 || y < !!my * 3 ||
362
8/8
✓ Branch 0 taken 436501 times.
✓ Branch 1 taken 95719 times.
✓ Branch 2 taken 501373 times.
✓ Branch 3 taken 30847 times.
✓ Branch 4 taken 389197 times.
✓ Branch 5 taken 112176 times.
✓ Branch 6 taken 25025 times.
✓ Branch 7 taken 476348 times.
532220 x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
363 513414 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
364
4/4
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
85569 ref_u - !!my * 3 * src_stride_u - !!mx * 3 * bytesperpixel,
365 160, src_stride_u,
366
2/2
✓ Branch 0 taken 71537 times.
✓ Branch 1 taken 14032 times.
85569 bw + !!mx * 7, bh + !!my * 7,
367
6/6
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
✓ Branch 4 taken 68164 times.
✓ Branch 5 taken 17405 times.
85569 x - !!mx * 3, y - !!my * 3, w, h);
368
4/4
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
85569 ref_u = td->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
369
2/2
✓ Branch 0 taken 71537 times.
✓ Branch 1 taken 14032 times.
85569 mc[!!mx][!!my](dst_u, dst_stride, ref_u, 160, bh, mx, my);
370
371 513414 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
372
4/4
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
85569 ref_v - !!my * 3 * src_stride_v - !!mx * 3 * bytesperpixel,
373 160, src_stride_v,
374
2/2
✓ Branch 0 taken 71537 times.
✓ Branch 1 taken 14032 times.
85569 bw + !!mx * 7, bh + !!my * 7,
375
6/6
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
✓ Branch 4 taken 68164 times.
✓ Branch 5 taken 17405 times.
85569 x - !!mx * 3, y - !!my * 3, w, h);
376
4/4
✓ Branch 0 taken 68164 times.
✓ Branch 1 taken 17405 times.
✓ Branch 2 taken 71537 times.
✓ Branch 3 taken 14032 times.
85569 ref_v = td->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
377
2/2
✓ Branch 0 taken 71537 times.
✓ Branch 1 taken 14032 times.
85569 mc[!!mx][!!my](dst_v, dst_stride, ref_v, 160, bh, mx, my);
378 } else {
379
2/2
✓ Branch 0 taken 392597 times.
✓ Branch 1 taken 83751 times.
476348 mc[!!mx][!!my](dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my);
380
2/2
✓ Branch 0 taken 392597 times.
✓ Branch 1 taken 83751 times.
476348 mc[!!mx][!!my](dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my);
381 }
382 561917 }
383
384 #define mc_luma_dir(td, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
385 px, py, pw, ph, bw, bh, w, h, i) \
386 mc_luma_unscaled(td, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
387 mv, bw, bh, w, h, bytesperpixel)
388 #define mc_chroma_dir(td, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
389 row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
390 mc_chroma_unscaled(td, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
391 row, col, mv, bw, bh, w, h, bytesperpixel)
392 #define SCALED 0
393 #define FN(x) x##_8bpp
394 #define BYTES_PER_PIXEL 1
395 #include "vp9_mc_template.c"
396 #undef FN
397 #undef BYTES_PER_PIXEL
398 #define FN(x) x##_16bpp
399 #define BYTES_PER_PIXEL 2
400 #include "vp9_mc_template.c"
401 #undef mc_luma_dir
402 #undef mc_chroma_dir
403 #undef FN
404 #undef BYTES_PER_PIXEL
405 #undef SCALED
406
407 689 static av_always_inline void mc_luma_scaled(VP9TileData *td, vp9_scaled_mc_func smc,
408 const vp9_mc_func (*mc)[2],
409 uint8_t *dst, ptrdiff_t dst_stride,
410 const uint8_t *ref, ptrdiff_t ref_stride,
411 const ProgressFrame *ref_frame,
412 ptrdiff_t y, ptrdiff_t x, const VP9mv *in_mv,
413 int px, int py, int pw, int ph,
414 int bw, int bh, int w, int h, int bytesperpixel,
415 const uint16_t *scale, const uint8_t *step)
416 {
417 689 const VP9Context *s = td->s;
418
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 689 times.
689 if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
419 s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
420 mc_luma_unscaled(td, mc, dst, dst_stride, ref, ref_stride, ref_frame,
421 y, x, in_mv, bw, bh, w, h, bytesperpixel);
422 } else {
423 #define scale_mv(n, dim) (((int64_t)(n) * scale[dim]) >> 14)
424 int mx, my;
425 int refbw_m1, refbh_m1;
426 int th;
427 VP9mv mv;
428
429 689 mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
430 689 mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
431 // BUG libvpx seems to scale the two components separately. This introduces
432 // rounding errors but we have to reproduce them to be exactly compatible
433 // with the output from libvpx...
434 689 mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
435 689 my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
436
437 689 y = my >> 4;
438 689 x = mx >> 4;
439 689 ref += y * ref_stride + x * bytesperpixel;
440 689 mx &= 15;
441 689 my &= 15;
442 689 refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
443 689 refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
444 // FIXME bilinear filter only needs 0/1 pixels, not 3/4
445 // we use +7 because the last 7 pixels of each sbrow can be changed in
446 // the longest loopfilter of the next sbrow
447 689 th = (y + refbh_m1 + 4 + 7) >> 6;
448 689 ff_progress_frame_await(ref_frame, FFMAX(th, 0));
449 // The arm/aarch64 _hv filters read one more row than what actually is
450 // needed, so switch to emulated edge one pixel sooner vertically
451 // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
452
8/8
✓ Branch 0 taken 649 times.
✓ Branch 1 taken 40 times.
✓ Branch 2 taken 607 times.
✓ Branch 3 taken 42 times.
✓ Branch 4 taken 586 times.
✓ Branch 5 taken 21 times.
✓ Branch 6 taken 31 times.
✓ Branch 7 taken 555 times.
689 if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
453 134 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
454 134 ref - 3 * ref_stride - 3 * bytesperpixel,
455 288, ref_stride,
456 refbw_m1 + 8, refbh_m1 + 8,
457 134 x - 3, y - 3, w, h);
458 134 ref = td->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
459 134 ref_stride = 288;
460 }
461 689 smc(dst, dst_stride, ref, ref_stride, bh, mx, my, step[0], step[1]);
462 }
463 689 }
464
465 689 static av_always_inline void mc_chroma_scaled(VP9TileData *td, vp9_scaled_mc_func smc,
466 const vp9_mc_func (*mc)[2],
467 uint8_t *dst_u, uint8_t *dst_v,
468 ptrdiff_t dst_stride,
469 const uint8_t *ref_u, ptrdiff_t src_stride_u,
470 const uint8_t *ref_v, ptrdiff_t src_stride_v,
471 const ProgressFrame *ref_frame,
472 ptrdiff_t y, ptrdiff_t x, const VP9mv *in_mv,
473 int px, int py, int pw, int ph,
474 int bw, int bh, int w, int h, int bytesperpixel,
475 const uint16_t *scale, const uint8_t *step)
476 {
477 689 const VP9Context *s = td->s;
478
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 689 times.
689 if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
479 s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
480 mc_chroma_unscaled(td, mc, dst_u, dst_v, dst_stride, ref_u, src_stride_u,
481 ref_v, src_stride_v, ref_frame,
482 y, x, in_mv, bw, bh, w, h, bytesperpixel);
483 } else {
484 int mx, my;
485 int refbw_m1, refbh_m1;
486 int th;
487 VP9mv mv;
488
489
1/2
✓ Branch 0 taken 689 times.
✗ Branch 1 not taken.
689 if (s->ss_h) {
490 // BUG https://code.google.com/p/webm/issues/detail?id=820
491 689 mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 16, (s->cols * 4 - x + px + 3) * 16);
492 689 mx = scale_mv(mv.x, 0) + (scale_mv(x * 16, 0) & ~15) + (scale_mv(x * 32, 0) & 15);
493 } else {
494 mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
495 mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
496 }
497
1/2
✓ Branch 0 taken 689 times.
✗ Branch 1 not taken.
689 if (s->ss_v) {
498 // BUG https://code.google.com/p/webm/issues/detail?id=820
499 689 mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 16, (s->rows * 4 - y + py + 3) * 16);
500 689 my = scale_mv(mv.y, 1) + (scale_mv(y * 16, 1) & ~15) + (scale_mv(y * 32, 1) & 15);
501 } else {
502 mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
503 my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
504 }
505 #undef scale_mv
506 689 y = my >> 4;
507 689 x = mx >> 4;
508 689 ref_u += y * src_stride_u + x * bytesperpixel;
509 689 ref_v += y * src_stride_v + x * bytesperpixel;
510 689 mx &= 15;
511 689 my &= 15;
512 689 refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
513 689 refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
514 // FIXME bilinear filter only needs 0/1 pixels, not 3/4
515 // we use +7 because the last 7 pixels of each sbrow can be changed in
516 // the longest loopfilter of the next sbrow
517 689 th = (y + refbh_m1 + 4 + 7) >> (6 - s->ss_v);
518 689 ff_progress_frame_await(ref_frame, FFMAX(th, 0));
519 // The arm/aarch64 _hv filters read one more row than what actually is
520 // needed, so switch to emulated edge one pixel sooner vertically
521 // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
522
8/8
✓ Branch 0 taken 633 times.
✓ Branch 1 taken 56 times.
✓ Branch 2 taken 583 times.
✓ Branch 3 taken 50 times.
✓ Branch 4 taken 553 times.
✓ Branch 5 taken 30 times.
✓ Branch 6 taken 41 times.
✓ Branch 7 taken 512 times.
689 if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
523 177 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
524 177 ref_u - 3 * src_stride_u - 3 * bytesperpixel,
525 288, src_stride_u,
526 refbw_m1 + 8, refbh_m1 + 8,
527 177 x - 3, y - 3, w, h);
528 177 ref_u = td->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
529 177 smc(dst_u, dst_stride, ref_u, 288, bh, mx, my, step[0], step[1]);
530
531 177 s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
532 177 ref_v - 3 * src_stride_v - 3 * bytesperpixel,
533 288, src_stride_v,
534 refbw_m1 + 8, refbh_m1 + 8,
535 177 x - 3, y - 3, w, h);
536 177 ref_v = td->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
537 177 smc(dst_v, dst_stride, ref_v, 288, bh, mx, my, step[0], step[1]);
538 } else {
539 512 smc(dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my, step[0], step[1]);
540 512 smc(dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my, step[0], step[1]);
541 }
542 }
543 689 }
544
545 #define mc_luma_dir(td, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
546 px, py, pw, ph, bw, bh, w, h, i) \
547 mc_luma_scaled(td, s->dsp.s##mc, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
548 mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
549 s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
550 #define mc_chroma_dir(td, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
551 row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
552 mc_chroma_scaled(td, s->dsp.s##mc, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
553 row, col, mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
554 s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
555 #define SCALED 1
556 #define FN(x) x##_scaled_8bpp
557 #define BYTES_PER_PIXEL 1
558 #include "vp9_mc_template.c"
559 #undef FN
560 #undef BYTES_PER_PIXEL
561 #define FN(x) x##_scaled_16bpp
562 #define BYTES_PER_PIXEL 2
563 #include "vp9_mc_template.c"
564 #undef mc_luma_dir
565 #undef mc_chroma_dir
566 #undef FN
567 #undef BYTES_PER_PIXEL
568 #undef SCALED
569
570 530281 static av_always_inline void inter_recon(VP9TileData *td, int bytesperpixel)
571 {
572 530281 const VP9Context *s = td->s;
573 530281 VP9Block *b = td->b;
574 530281 int row = td->row, col = td->col;
575
576
1/2
✓ Branch 0 taken 530281 times.
✗ Branch 1 not taken.
530281 if (s->mvscale[b->ref[0]][0] == REF_INVALID_SCALE ||
577
3/4
✓ Branch 0 taken 27251 times.
✓ Branch 1 taken 503030 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 27251 times.
530281 (b->comp && s->mvscale[b->ref[1]][0] == REF_INVALID_SCALE)) {
578 if (!s->td->error_info) {
579 s->td->error_info = AVERROR_INVALIDDATA;
580 av_log(NULL, AV_LOG_ERROR, "Bitstream not supported, "
581 "reference frame has invalid dimensions\n");
582 }
583 return;
584 }
585
586
5/6
✓ Branch 0 taken 529592 times.
✓ Branch 1 taken 689 times.
✓ Branch 2 taken 27251 times.
✓ Branch 3 taken 502341 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 27251 times.
530281 if (s->mvscale[b->ref[0]][0] || (b->comp && s->mvscale[b->ref[1]][0])) {
587
1/2
✓ Branch 0 taken 689 times.
✗ Branch 1 not taken.
689 if (bytesperpixel == 1) {
588 689 inter_pred_scaled_8bpp(td);
589 } else {
590 inter_pred_scaled_16bpp(td);
591 }
592 } else {
593
2/2
✓ Branch 0 taken 513805 times.
✓ Branch 1 taken 15787 times.
529592 if (bytesperpixel == 1) {
594 513805 inter_pred_8bpp(td);
595 } else {
596 15787 inter_pred_16bpp(td);
597 }
598 }
599
600
2/2
✓ Branch 0 taken 137847 times.
✓ Branch 1 taken 392434 times.
530281 if (!b->skip) {
601 /* mostly copied intra_recon() */
602
603 137847 int w4 = ff_vp9_bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
604 137847 int h4 = ff_vp9_bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
605 137847 int end_x = FFMIN(2 * (s->cols - col), w4);
606 137847 int end_y = FFMIN(2 * (s->rows - row), h4);
607 137847 int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
608 137847 int uvstep1d = 1 << b->uvtx, p;
609 137847 uint8_t *dst = td->dst[0];
610
611 // y itxfm add
612
2/2
✓ Branch 0 taken 227388 times.
✓ Branch 1 taken 137847 times.
365235 for (n = 0, y = 0; y < end_y; y += step1d) {
613 227388 uint8_t *ptr = dst;
614
2/2
✓ Branch 0 taken 422951 times.
✓ Branch 1 taken 227388 times.
650339 for (x = 0; x < end_x; x += step1d,
615 422951 ptr += 4 * step1d * bytesperpixel, n += step) {
616
2/2
✓ Branch 0 taken 17838 times.
✓ Branch 1 taken 405113 times.
422951 int eob = b->tx > TX_8X8 ? AV_RN16A(&td->eob[n]) : td->eob[n];
617
618
2/2
✓ Branch 0 taken 240652 times.
✓ Branch 1 taken 182299 times.
422951 if (eob)
619 240652 s->dsp.itxfm_add[tx][DCT_DCT](ptr, td->y_stride,
620 240652 td->block + 16 * n * bytesperpixel, eob);
621 }
622 227388 dst += 4 * td->y_stride * step1d;
623 }
624
625 // uv itxfm add
626 137847 end_x >>= s->ss_h;
627 137847 end_y >>= s->ss_v;
628 137847 step = 1 << (b->uvtx * 2);
629
2/2
✓ Branch 0 taken 275694 times.
✓ Branch 1 taken 137847 times.
413541 for (p = 0; p < 2; p++) {
630 275694 dst = td->dst[p + 1];
631
2/2
✓ Branch 0 taken 300834 times.
✓ Branch 1 taken 275694 times.
576528 for (n = 0, y = 0; y < end_y; y += uvstep1d) {
632 300834 uint8_t *ptr = dst;
633
2/2
✓ Branch 0 taken 342698 times.
✓ Branch 1 taken 300834 times.
643532 for (x = 0; x < end_x; x += uvstep1d,
634 342698 ptr += 4 * uvstep1d * bytesperpixel, n += step) {
635
2/2
✓ Branch 0 taken 7772 times.
✓ Branch 1 taken 334926 times.
342698 int eob = b->uvtx > TX_8X8 ? AV_RN16A(&td->uveob[p][n]) : td->uveob[p][n];
636
637
2/2
✓ Branch 0 taken 66756 times.
✓ Branch 1 taken 275942 times.
342698 if (eob)
638 66756 s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, td->uv_stride,
639 66756 td->uvblock[p] + 16 * n * bytesperpixel, eob);
640 }
641 300834 dst += 4 * uvstep1d * td->uv_stride;
642 }
643 }
644 }
645 }
646
647 514494 void ff_vp9_inter_recon_8bpp(VP9TileData *td)
648 {
649 514494 inter_recon(td, 1);
650 514494 }
651
652 15787 void ff_vp9_inter_recon_16bpp(VP9TileData *td)
653 {
654 15787 inter_recon(td, 2);
655 15787 }
656