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 "avcodec.h" | ||
25 | #include "vp9dec.h" | ||
26 | |||
27 | 149298 | static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v, | |
28 | uint8_t *lvl, uint8_t (*mask)[4], | ||
29 | uint8_t *dst, ptrdiff_t ls) | ||
30 | { | ||
31 | 149298 | int y, x, bytesperpixel = s->bytesperpixel; | |
32 | |||
33 | // filter edges between columns (e.g. block1 | block2) | ||
34 |
2/2✓ Branch 0 taken 401440 times.
✓ Branch 1 taken 149298 times.
|
550738 | for (y = 0; y < 8; y += 2 << ss_v, dst += 16 * ls, lvl += 16 << ss_v) { |
35 | 401440 | uint8_t *ptr = dst, *l = lvl, *hmask1 = mask[y], *hmask2 = mask[y + 1 + ss_v]; | |
36 | 401440 | unsigned hm1 = hmask1[0] | hmask1[1] | hmask1[2], hm13 = hmask1[3]; | |
37 | 401440 | unsigned hm2 = hmask2[1] | hmask2[2], hm23 = hmask2[3]; | |
38 | 401440 | unsigned hm = hm1 | hm2 | hm13 | hm23; | |
39 | |||
40 |
2/2✓ Branch 0 taken 1790348 times.
✓ Branch 1 taken 401440 times.
|
2191788 | for (x = 1; hm & ~(x - 1); x <<= 1, ptr += 8 * bytesperpixel >> ss_h) { |
41 |
4/4✓ Branch 0 taken 287136 times.
✓ Branch 1 taken 1503212 times.
✓ Branch 2 taken 233721 times.
✓ Branch 3 taken 53415 times.
|
1790348 | if (col || x > 1) { |
42 |
2/2✓ Branch 0 taken 1177739 times.
✓ Branch 1 taken 559194 times.
|
1736933 | if (hm1 & x) { |
43 | 1177739 | int L = *l, H = L >> 4; | |
44 | 1177739 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
45 | |||
46 |
2/2✓ Branch 0 taken 345292 times.
✓ Branch 1 taken 832447 times.
|
1177739 | if (hmask1[0] & x) { |
47 |
2/2✓ Branch 0 taken 343504 times.
✓ Branch 1 taken 1788 times.
|
345292 | if (hmask2[0] & x) { |
48 | av_assert2(l[8 << ss_v] == L); | ||
49 | 343504 | s->dsp.loop_filter_16[0](ptr, ls, E, I, H); | |
50 | } else { | ||
51 | 1788 | s->dsp.loop_filter_8[2][0](ptr, ls, E, I, H); | |
52 | } | ||
53 |
2/2✓ Branch 0 taken 766320 times.
✓ Branch 1 taken 66127 times.
|
832447 | } else if (hm2 & x) { |
54 | 766320 | L = l[8 << ss_v]; | |
55 | 766320 | H |= (L >> 4) << 8; | |
56 | 766320 | E |= s->filter_lut.mblim_lut[L] << 8; | |
57 | 766320 | I |= s->filter_lut.lim_lut[L] << 8; | |
58 | 766320 | s->dsp.loop_filter_mix2[!!(hmask1[1] & x)] | |
59 | 766320 | [!!(hmask2[1] & x)] | |
60 | 766320 | [0](ptr, ls, E, I, H); | |
61 | } else { | ||
62 | 66127 | s->dsp.loop_filter_8[!!(hmask1[1] & x)] | |
63 | 66127 | [0](ptr, ls, E, I, H); | |
64 | } | ||
65 |
2/2✓ Branch 0 taken 33532 times.
✓ Branch 1 taken 525662 times.
|
559194 | } else if (hm2 & x) { |
66 | 33532 | int L = l[8 << ss_v], H = L >> 4; | |
67 | 33532 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
68 | |||
69 | 33532 | s->dsp.loop_filter_8[!!(hmask2[1] & x)] | |
70 | 33532 | [0](ptr + 8 * ls, ls, E, I, H); | |
71 | } | ||
72 | } | ||
73 |
2/2✓ Branch 0 taken 867642 times.
✓ Branch 1 taken 922706 times.
|
1790348 | if (ss_h) { |
74 |
2/2✓ Branch 0 taken 362726 times.
✓ Branch 1 taken 504916 times.
|
867642 | if (x & 0xAA) |
75 | 362726 | l += 2; | |
76 | } else { | ||
77 |
2/2✓ Branch 0 taken 115223 times.
✓ Branch 1 taken 807483 times.
|
922706 | if (hm13 & x) { |
78 | 115223 | int L = *l, H = L >> 4; | |
79 | 115223 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
80 | |||
81 |
2/2✓ Branch 0 taken 66821 times.
✓ Branch 1 taken 48402 times.
|
115223 | if (hm23 & x) { |
82 | 66821 | L = l[8 << ss_v]; | |
83 | 66821 | H |= (L >> 4) << 8; | |
84 | 66821 | E |= s->filter_lut.mblim_lut[L] << 8; | |
85 | 66821 | I |= s->filter_lut.lim_lut[L] << 8; | |
86 | 66821 | s->dsp.loop_filter_mix2[0][0][0](ptr + 4 * bytesperpixel, ls, E, I, H); | |
87 | } else { | ||
88 | 48402 | s->dsp.loop_filter_8[0][0](ptr + 4 * bytesperpixel, ls, E, I, H); | |
89 | } | ||
90 |
2/2✓ Branch 0 taken 39533 times.
✓ Branch 1 taken 767950 times.
|
807483 | } else if (hm23 & x) { |
91 | 39533 | int L = l[8 << ss_v], H = L >> 4; | |
92 | 39533 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
93 | |||
94 | 39533 | s->dsp.loop_filter_8[0][0](ptr + 8 * ls + 4 * bytesperpixel, ls, E, I, H); | |
95 | } | ||
96 | 922706 | l++; | |
97 | } | ||
98 | } | ||
99 | } | ||
100 | 149298 | } | |
101 | |||
102 | 149298 | static av_always_inline void filter_plane_rows(VP9Context *s, int row, int ss_h, int ss_v, | |
103 | uint8_t *lvl, uint8_t (*mask)[4], | ||
104 | uint8_t *dst, ptrdiff_t ls) | ||
105 | { | ||
106 | 149298 | int y, x, bytesperpixel = s->bytesperpixel; | |
107 | |||
108 | // block1 | ||
109 | // filter edges between rows (e.g. ------) | ||
110 | // block2 | ||
111 |
2/2✓ Branch 0 taken 1194384 times.
✓ Branch 1 taken 149298 times.
|
1343682 | for (y = 0; y < 8; y++, dst += 8 * ls >> ss_v) { |
112 | 1194384 | uint8_t *ptr = dst, *l = lvl, *vmask = mask[y]; | |
113 | 1194384 | unsigned vm = vmask[0] | vmask[1] | vmask[2], vm3 = vmask[3]; | |
114 | |||
115 |
2/2✓ Branch 0 taken 1385779 times.
✓ Branch 1 taken 1194384 times.
|
2580163 | for (x = 1; vm & ~(x - 1); x <<= (2 << ss_h), ptr += 16 * bytesperpixel, l += 2 << ss_h) { |
116 |
4/4✓ Branch 0 taken 293184 times.
✓ Branch 1 taken 1092595 times.
✓ Branch 2 taken 224709 times.
✓ Branch 3 taken 68475 times.
|
1385779 | if (row || y) { |
117 |
2/2✓ Branch 0 taken 1169997 times.
✓ Branch 1 taken 147307 times.
|
1317304 | if (vm & x) { |
118 | 1169997 | int L = *l, H = L >> 4; | |
119 | 1169997 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
120 | |||
121 |
2/2✓ Branch 0 taken 347705 times.
✓ Branch 1 taken 822292 times.
|
1169997 | if (vmask[0] & x) { |
122 |
2/2✓ Branch 0 taken 347386 times.
✓ Branch 1 taken 319 times.
|
347705 | if (vmask[0] & (x << (1 + ss_h))) { |
123 | av_assert2(l[1 + ss_h] == L); | ||
124 | 347386 | s->dsp.loop_filter_16[1](ptr, ls, E, I, H); | |
125 | } else { | ||
126 | 319 | s->dsp.loop_filter_8[2][1](ptr, ls, E, I, H); | |
127 | } | ||
128 |
2/2✓ Branch 0 taken 754663 times.
✓ Branch 1 taken 67629 times.
|
822292 | } else if (vm & (x << (1 + ss_h))) { |
129 | 754663 | L = l[1 + ss_h]; | |
130 | 754663 | H |= (L >> 4) << 8; | |
131 | 754663 | E |= s->filter_lut.mblim_lut[L] << 8; | |
132 | 754663 | I |= s->filter_lut.lim_lut[L] << 8; | |
133 | 754663 | s->dsp.loop_filter_mix2[!!(vmask[1] & x)] | |
134 | 754663 | [!!(vmask[1] & (x << (1 + ss_h)))] | |
135 | 754663 | [1](ptr, ls, E, I, H); | |
136 | } else { | ||
137 | 67629 | s->dsp.loop_filter_8[!!(vmask[1] & x)] | |
138 | 67629 | [1](ptr, ls, E, I, H); | |
139 | } | ||
140 |
2/2✓ Branch 0 taken 35358 times.
✓ Branch 1 taken 111949 times.
|
147307 | } else if (vm & (x << (1 + ss_h))) { |
141 | 35358 | int L = l[1 + ss_h], H = L >> 4; | |
142 | 35358 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
143 | |||
144 | 35358 | s->dsp.loop_filter_8[!!(vmask[1] & (x << (1 + ss_h)))] | |
145 | 35358 | [1](ptr + 8 * bytesperpixel, ls, E, I, H); | |
146 | } | ||
147 | } | ||
148 |
2/2✓ Branch 0 taken 754557 times.
✓ Branch 1 taken 631222 times.
|
1385779 | if (!ss_v) { |
149 |
2/2✓ Branch 0 taken 110039 times.
✓ Branch 1 taken 644518 times.
|
754557 | if (vm3 & x) { |
150 | 110039 | int L = *l, H = L >> 4; | |
151 | 110039 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
152 | |||
153 |
2/2✓ Branch 0 taken 65543 times.
✓ Branch 1 taken 44496 times.
|
110039 | if (vm3 & (x << (1 + ss_h))) { |
154 | 65543 | L = l[1 + ss_h]; | |
155 | 65543 | H |= (L >> 4) << 8; | |
156 | 65543 | E |= s->filter_lut.mblim_lut[L] << 8; | |
157 | 65543 | I |= s->filter_lut.lim_lut[L] << 8; | |
158 | 65543 | s->dsp.loop_filter_mix2[0][0][1](ptr + ls * 4, ls, E, I, H); | |
159 | } else { | ||
160 | 44496 | s->dsp.loop_filter_8[0][1](ptr + ls * 4, ls, E, I, H); | |
161 | } | ||
162 |
2/2✓ Branch 0 taken 42225 times.
✓ Branch 1 taken 602293 times.
|
644518 | } else if (vm3 & (x << (1 + ss_h))) { |
163 | 42225 | int L = l[1 + ss_h], H = L >> 4; | |
164 | 42225 | int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L]; | |
165 | |||
166 | 42225 | s->dsp.loop_filter_8[0][1](ptr + ls * 4 + 8 * bytesperpixel, ls, E, I, H); | |
167 | } | ||
168 | } | ||
169 | } | ||
170 |
2/2✓ Branch 0 taken 783008 times.
✓ Branch 1 taken 411376 times.
|
1194384 | if (ss_v) { |
171 |
2/2✓ Branch 0 taken 391504 times.
✓ Branch 1 taken 391504 times.
|
783008 | if (y & 1) |
172 | 391504 | lvl += 16; | |
173 | } else { | ||
174 | 411376 | lvl += 8; | |
175 | } | ||
176 | } | ||
177 | 149298 | } | |
178 | |||
179 | 49766 | void ff_vp9_loopfilter_sb(AVCodecContext *avctx, VP9Filter *lflvl, | |
180 | int row, int col, ptrdiff_t yoff, ptrdiff_t uvoff) | ||
181 | { | ||
182 | 49766 | VP9Context *s = avctx->priv_data; | |
183 | 49766 | AVFrame *f = s->s.frames[CUR_FRAME].tf.f; | |
184 | 49766 | uint8_t *dst = f->data[0] + yoff; | |
185 | 49766 | ptrdiff_t ls_y = f->linesize[0], ls_uv = f->linesize[1]; | |
186 | 49766 | uint8_t (*uv_masks)[8][4] = lflvl->mask[s->ss_h | s->ss_v]; | |
187 | int p; | ||
188 | |||
189 | /* FIXME: In how far can we interleave the v/h loopfilter calls? E.g. | ||
190 | * if you think of them as acting on a 8x8 block max, we can interleave | ||
191 | * each v/h within the single x loop, but that only works if we work on | ||
192 | * 8 pixel blocks, and we won't always do that (we want at least 16px | ||
193 | * to use SSE2 optimizations, perhaps 32 for AVX2) */ | ||
194 | |||
195 | 49766 | filter_plane_cols(s, col, 0, 0, lflvl->level, lflvl->mask[0][0], dst, ls_y); | |
196 | 49766 | filter_plane_rows(s, row, 0, 0, lflvl->level, lflvl->mask[0][1], dst, ls_y); | |
197 | |||
198 |
2/2✓ Branch 0 taken 99532 times.
✓ Branch 1 taken 49766 times.
|
149298 | for (p = 0; p < 2; p++) { |
199 | 99532 | dst = f->data[1 + p] + uvoff; | |
200 | 99532 | filter_plane_cols(s, col, s->ss_h, s->ss_v, lflvl->level, uv_masks[0], dst, ls_uv); | |
201 | 99532 | filter_plane_rows(s, row, s->ss_h, s->ss_v, lflvl->level, uv_masks[1], dst, ls_uv); | |
202 | } | ||
203 | 49766 | } | |
204 |