Mercurial > libavcodec.hg
annotate vp8dsp.c @ 12435:fe78a4548d12 libavcodec
Put ff_ prefix on non-static {put_signed,put,add}_pixels_clamped_mmx()
functions.
author | rbultje |
---|---|
date | Mon, 30 Aug 2010 16:22:27 +0000 |
parents | b4c63ffd959b |
children |
rev | line source |
---|---|
11921 | 1 /** |
2 * VP8 compatible video decoder | |
3 * | |
4 * Copyright (C) 2010 David Conrad | |
5 * Copyright (C) 2010 Ronald S. Bultje | |
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 "dsputil.h" | |
25 #include "vp8dsp.h" | |
26 | |
27 // TODO: Maybe add dequant | |
28 static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16]) | |
29 { | |
30 int i, t0, t1, t2, t3; | |
31 | |
32 for (i = 0; i < 4; i++) { | |
33 t0 = dc[0*4+i] + dc[3*4+i]; | |
34 t1 = dc[1*4+i] + dc[2*4+i]; | |
35 t2 = dc[1*4+i] - dc[2*4+i]; | |
36 t3 = dc[0*4+i] - dc[3*4+i]; | |
37 | |
38 dc[0*4+i] = t0 + t1; | |
39 dc[1*4+i] = t3 + t2; | |
40 dc[2*4+i] = t0 - t1; | |
41 dc[3*4+i] = t3 - t2; | |
42 } | |
43 | |
44 for (i = 0; i < 4; i++) { | |
45 t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding | |
46 t1 = dc[i*4+1] + dc[i*4+2]; | |
47 t2 = dc[i*4+1] - dc[i*4+2]; | |
48 t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding | |
12340
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12241
diff
changeset
|
49 dc[i*4+0] = 0; |
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12241
diff
changeset
|
50 dc[i*4+1] = 0; |
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12241
diff
changeset
|
51 dc[i*4+2] = 0; |
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12241
diff
changeset
|
52 dc[i*4+3] = 0; |
11921 | 53 |
12342 | 54 block[i][0][0] = (t0 + t1) >> 3; |
55 block[i][1][0] = (t3 + t2) >> 3; | |
56 block[i][2][0] = (t0 - t1) >> 3; | |
57 block[i][3][0] = (t3 - t2) >> 3; | |
11921 | 58 } |
59 } | |
60 | |
12342 | 61 static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16]) |
62 { | |
63 int i, val = (dc[0] + 3) >> 3; | |
64 dc[0] = 0; | |
65 | |
66 for (i = 0; i < 4; i++) { | |
67 block[i][0][0] = val; | |
68 block[i][1][0] = val; | |
69 block[i][2][0] = val; | |
70 block[i][3][0] = val; | |
71 } | |
72 } | |
11921 | 73 |
74 #define MUL_20091(a) ((((a)*20091) >> 16) + (a)) | |
75 #define MUL_35468(a) (((a)*35468) >> 16) | |
76 | |
77 static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride) | |
78 { | |
79 int i, t0, t1, t2, t3; | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
80 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
11921 | 81 DCTELEM tmp[16]; |
82 | |
83 for (i = 0; i < 4; i++) { | |
84 t0 = block[0*4+i] + block[2*4+i]; | |
85 t1 = block[0*4+i] - block[2*4+i]; | |
86 t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]); | |
87 t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]); | |
12235
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12194
diff
changeset
|
88 block[0*4+i] = 0; |
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12194
diff
changeset
|
89 block[1*4+i] = 0; |
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12194
diff
changeset
|
90 block[2*4+i] = 0; |
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12194
diff
changeset
|
91 block[3*4+i] = 0; |
11921 | 92 |
93 tmp[i*4+0] = t0 + t3; | |
94 tmp[i*4+1] = t1 + t2; | |
95 tmp[i*4+2] = t1 - t2; | |
96 tmp[i*4+3] = t0 - t3; | |
97 } | |
98 | |
99 for (i = 0; i < 4; i++) { | |
100 t0 = tmp[0*4+i] + tmp[2*4+i]; | |
101 t1 = tmp[0*4+i] - tmp[2*4+i]; | |
102 t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]); | |
103 t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]); | |
104 | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
105 dst[0] = cm[dst[0] + ((t0 + t3 + 4) >> 3)]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
106 dst[1] = cm[dst[1] + ((t1 + t2 + 4) >> 3)]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
107 dst[2] = cm[dst[2] + ((t1 - t2 + 4) >> 3)]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
108 dst[3] = cm[dst[3] + ((t0 - t3 + 4) >> 3)]; |
11921 | 109 dst += stride; |
110 } | |
111 } | |
112 | |
113 static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride) | |
114 { | |
115 int i, dc = (block[0] + 4) >> 3; | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
116 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP + dc; |
12235
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12194
diff
changeset
|
117 block[0] = 0; |
11921 | 118 |
119 for (i = 0; i < 4; i++) { | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
120 dst[0] = cm[dst[0]]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
121 dst[1] = cm[dst[1]]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
122 dst[2] = cm[dst[2]]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
123 dst[3] = cm[dst[3]]; |
11921 | 124 dst += stride; |
125 } | |
126 } | |
127 | |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
128 static void vp8_idct_dc_add4uv_c(uint8_t *dst, DCTELEM block[4][16], int stride) |
12238 | 129 { |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
130 vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
131 vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
132 vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
133 vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
134 } |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
135 |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
136 static void vp8_idct_dc_add4y_c(uint8_t *dst, DCTELEM block[4][16], int stride) |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
137 { |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
138 vp8_idct_dc_add_c(dst+ 0, block[0], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
139 vp8_idct_dc_add_c(dst+ 4, block[1], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
140 vp8_idct_dc_add_c(dst+ 8, block[2], stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
141 vp8_idct_dc_add_c(dst+12, block[3], stride); |
12238 | 142 } |
11921 | 143 |
144 // because I like only having two parameters to pass functions... | |
145 #define LOAD_PIXELS\ | |
146 int av_unused p3 = p[-4*stride];\ | |
147 int av_unused p2 = p[-3*stride];\ | |
148 int av_unused p1 = p[-2*stride];\ | |
149 int av_unused p0 = p[-1*stride];\ | |
150 int av_unused q0 = p[ 0*stride];\ | |
151 int av_unused q1 = p[ 1*stride];\ | |
152 int av_unused q2 = p[ 2*stride];\ | |
153 int av_unused q3 = p[ 3*stride]; | |
154 | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
155 #define clip_int8(n) (cm[n+0x80]-0x80) |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
156 |
11921 | 157 static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap) |
158 { | |
159 LOAD_PIXELS | |
160 int a, f1, f2; | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
161 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
11921 | 162 |
163 a = 3*(q0 - p0); | |
164 | |
165 if (is4tap) | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
166 a += clip_int8(p1 - q1); |
11921 | 167 |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
168 a = clip_int8(a); |
11921 | 169 |
170 // We deviate from the spec here with c(a+3) >> 3 | |
171 // since that's what libvpx does. | |
172 f1 = FFMIN(a+4, 127) >> 3; | |
173 f2 = FFMIN(a+3, 127) >> 3; | |
174 | |
175 // Despite what the spec says, we do need to clamp here to | |
176 // be bitexact with libvpx. | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
177 p[-1*stride] = cm[p0 + f2]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
178 p[ 0*stride] = cm[q0 - f1]; |
11921 | 179 |
180 // only used for _inner on blocks without high edge variance | |
181 if (!is4tap) { | |
182 a = (f1+1)>>1; | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
183 p[-2*stride] = cm[p1 + a]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
184 p[ 1*stride] = cm[q1 - a]; |
11921 | 185 } |
186 } | |
187 | |
188 static av_always_inline int simple_limit(uint8_t *p, int stride, int flim) | |
189 { | |
190 LOAD_PIXELS | |
191 return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim; | |
192 } | |
193 | |
194 /** | |
195 * E - limit at the macroblock edge | |
196 * I - limit for interior difference | |
197 */ | |
198 static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I) | |
199 { | |
200 LOAD_PIXELS | |
12081
812e23197d64
VP8: Move calculation of outer filter limit out of dsp functions for normal
conrad
parents:
12011
diff
changeset
|
201 return simple_limit(p, stride, E) |
11921 | 202 && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I |
203 && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I; | |
204 } | |
205 | |
206 // high edge variance | |
207 static av_always_inline int hev(uint8_t *p, int stride, int thresh) | |
208 { | |
209 LOAD_PIXELS | |
210 return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh; | |
211 } | |
212 | |
213 static av_always_inline void filter_mbedge(uint8_t *p, int stride) | |
214 { | |
215 int a0, a1, a2, w; | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
216 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
11921 | 217 |
218 LOAD_PIXELS | |
219 | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
220 w = clip_int8(p1-q1); |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
221 w = clip_int8(w + 3*(q0-p0)); |
11921 | 222 |
223 a0 = (27*w + 63) >> 7; | |
224 a1 = (18*w + 63) >> 7; | |
225 a2 = ( 9*w + 63) >> 7; | |
226 | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
227 p[-3*stride] = cm[p2 + a2]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
228 p[-2*stride] = cm[p1 + a1]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
229 p[-1*stride] = cm[p0 + a0]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
230 p[ 0*stride] = cm[q0 - a0]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
231 p[ 1*stride] = cm[q1 - a1]; |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
232 p[ 2*stride] = cm[q2 - a2]; |
11921 | 233 } |
234 | |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
235 #define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
236 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\ |
11921 | 237 int flim_E, int flim_I, int hev_thresh)\ |
238 {\ | |
239 int i;\ | |
240 \ | |
241 for (i = 0; i < size; i++)\ | |
242 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\ | |
243 if (hev(dst+i*stridea, strideb, hev_thresh))\ | |
244 filter_common(dst+i*stridea, strideb, 1);\ | |
245 else\ | |
246 filter_mbedge(dst+i*stridea, strideb);\ | |
247 }\ | |
248 }\ | |
249 \ | |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
250 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\ |
11921 | 251 int flim_E, int flim_I, int hev_thresh)\ |
252 {\ | |
12008 | 253 int i;\ |
11921 | 254 \ |
255 for (i = 0; i < size; i++)\ | |
256 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\ | |
12008 | 257 int hv = hev(dst+i*stridea, strideb, hev_thresh);\ |
258 if (hv) \ | |
259 filter_common(dst+i*stridea, strideb, 1);\ | |
260 else \ | |
261 filter_common(dst+i*stridea, strideb, 0);\ | |
11921 | 262 }\ |
263 } | |
264 | |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
265 LOOP_FILTER(v, 16, 1, stride,) |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
266 LOOP_FILTER(h, 16, stride, 1,) |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
267 |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
268 #define UV_LOOP_FILTER(dir, stridea, strideb) \ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
269 LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
270 static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, int stride,\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
271 int fE, int fI, int hev_thresh)\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
272 {\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
273 vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
274 vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
275 }\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
276 static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, int stride,\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
277 int fE, int fI, int hev_thresh)\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
278 {\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
279 vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
280 vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\ |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
281 } |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
282 |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
283 UV_LOOP_FILTER(v, 1, stride) |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
284 UV_LOOP_FILTER(h, stride, 1) |
11921 | 285 |
286 static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim) | |
287 { | |
288 int i; | |
289 | |
290 for (i = 0; i < 16; i++) | |
291 if (simple_limit(dst+i, stride, flim)) | |
292 filter_common(dst+i, stride, 1); | |
293 } | |
294 | |
295 static void vp8_h_loop_filter_simple_c(uint8_t *dst, int stride, int flim) | |
296 { | |
297 int i; | |
298 | |
299 for (i = 0; i < 16; i++) | |
300 if (simple_limit(dst+i*stride, 1, flim)) | |
301 filter_common(dst+i*stride, 1, 1); | |
302 } | |
303 | |
304 static const uint8_t subpel_filters[7][6] = { | |
305 { 0, 6, 123, 12, 1, 0 }, | |
306 { 2, 11, 108, 36, 8, 1 }, | |
307 { 0, 9, 93, 50, 6, 0 }, | |
308 { 3, 16, 77, 77, 16, 3 }, | |
309 { 0, 6, 50, 93, 9, 0 }, | |
310 { 1, 8, 36, 108, 11, 2 }, | |
311 { 0, 1, 12, 123, 6, 0 }, | |
312 }; | |
313 | |
11950 | 314 #define PUT_PIXELS(WIDTH) \ |
315 static void put_vp8_pixels ## WIDTH ##_c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int x, int y) { \ | |
11956 | 316 int i; \ |
317 for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \ | |
11950 | 318 memcpy(dst, src, WIDTH); \ |
319 } \ | |
320 } | |
321 | |
322 PUT_PIXELS(16) | |
323 PUT_PIXELS(8) | |
324 PUT_PIXELS(4) | |
11921 | 325 |
326 #define FILTER_6TAP(src, F, stride) \ | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
327 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \ |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
328 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7] |
11921 | 329 |
330 #define FILTER_4TAP(src, F, stride) \ | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
331 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \ |
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
332 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7] |
11921 | 333 |
334 #define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \ | |
11950 | 335 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \ |
11921 | 336 { \ |
337 const uint8_t *filter = subpel_filters[mx-1]; \ | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
338 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \ |
11921 | 339 int x, y; \ |
340 \ | |
341 for (y = 0; y < h; y++) { \ | |
342 for (x = 0; x < SIZE; x++) \ | |
343 dst[x] = FILTER(src, filter, 1); \ | |
11950 | 344 dst += dststride; \ |
345 src += srcstride; \ | |
11921 | 346 } \ |
347 } | |
348 #define VP8_EPEL_V(SIZE, FILTER, FILTERNAME) \ | |
11950 | 349 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \ |
11921 | 350 { \ |
351 const uint8_t *filter = subpel_filters[my-1]; \ | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
352 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \ |
11921 | 353 int x, y; \ |
354 \ | |
355 for (y = 0; y < h; y++) { \ | |
356 for (x = 0; x < SIZE; x++) \ | |
11950 | 357 dst[x] = FILTER(src, filter, srcstride); \ |
358 dst += dststride; \ | |
359 src += srcstride; \ | |
11921 | 360 } \ |
361 } | |
362 #define VP8_EPEL_HV(SIZE, FILTERX, FILTERY, FILTERNAME) \ | |
11950 | 363 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \ |
11921 | 364 { \ |
365 const uint8_t *filter = subpel_filters[mx-1]; \ | |
12007
ec7be1d7d5b4
Use crop table in C implementations of VP8 DSP functions.
darkshikari
parents:
11985
diff
changeset
|
366 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \ |
11921 | 367 int x, y; \ |
368 uint8_t tmp_array[(2*SIZE+5)*SIZE]; \ | |
369 uint8_t *tmp = tmp_array; \ | |
11950 | 370 src -= 2*srcstride; \ |
11921 | 371 \ |
372 for (y = 0; y < h+5; y++) { \ | |
373 for (x = 0; x < SIZE; x++) \ | |
374 tmp[x] = FILTERX(src, filter, 1); \ | |
375 tmp += SIZE; \ | |
11950 | 376 src += srcstride; \ |
11921 | 377 } \ |
378 \ | |
379 tmp = tmp_array + 2*SIZE; \ | |
380 filter = subpel_filters[my-1]; \ | |
381 \ | |
382 for (y = 0; y < h; y++) { \ | |
383 for (x = 0; x < SIZE; x++) \ | |
384 dst[x] = FILTERY(tmp, filter, SIZE); \ | |
11950 | 385 dst += dststride; \ |
11921 | 386 tmp += SIZE; \ |
387 } \ | |
388 } | |
389 | |
390 VP8_EPEL_H(16, FILTER_4TAP, h4) | |
391 VP8_EPEL_H(8, FILTER_4TAP, h4) | |
392 VP8_EPEL_H(4, FILTER_4TAP, h4) | |
393 VP8_EPEL_H(16, FILTER_6TAP, h6) | |
394 VP8_EPEL_H(8, FILTER_6TAP, h6) | |
395 VP8_EPEL_H(4, FILTER_6TAP, h6) | |
396 VP8_EPEL_V(16, FILTER_4TAP, v4) | |
397 VP8_EPEL_V(8, FILTER_4TAP, v4) | |
398 VP8_EPEL_V(4, FILTER_4TAP, v4) | |
399 VP8_EPEL_V(16, FILTER_6TAP, v6) | |
400 VP8_EPEL_V(8, FILTER_6TAP, v6) | |
401 VP8_EPEL_V(4, FILTER_6TAP, v6) | |
402 VP8_EPEL_HV(16, FILTER_4TAP, FILTER_4TAP, h4v4) | |
403 VP8_EPEL_HV(8, FILTER_4TAP, FILTER_4TAP, h4v4) | |
404 VP8_EPEL_HV(4, FILTER_4TAP, FILTER_4TAP, h4v4) | |
405 VP8_EPEL_HV(16, FILTER_4TAP, FILTER_6TAP, h4v6) | |
406 VP8_EPEL_HV(8, FILTER_4TAP, FILTER_6TAP, h4v6) | |
407 VP8_EPEL_HV(4, FILTER_4TAP, FILTER_6TAP, h4v6) | |
408 VP8_EPEL_HV(16, FILTER_6TAP, FILTER_4TAP, h6v4) | |
409 VP8_EPEL_HV(8, FILTER_6TAP, FILTER_4TAP, h6v4) | |
410 VP8_EPEL_HV(4, FILTER_6TAP, FILTER_4TAP, h6v4) | |
411 VP8_EPEL_HV(16, FILTER_6TAP, FILTER_6TAP, h6v6) | |
412 VP8_EPEL_HV(8, FILTER_6TAP, FILTER_6TAP, h6v6) | |
413 VP8_EPEL_HV(4, FILTER_6TAP, FILTER_6TAP, h6v6) | |
414 | |
11974 | 415 #define VP8_BILINEAR(SIZE) \ |
416 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \ | |
417 { \ | |
418 int a = 8-mx, b = mx; \ | |
419 int x, y; \ | |
420 \ | |
421 for (y = 0; y < h; y++) { \ | |
422 for (x = 0; x < SIZE; x++) \ | |
423 dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \ | |
424 dst += stride; \ | |
425 src += stride; \ | |
426 } \ | |
427 } \ | |
428 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \ | |
429 { \ | |
430 int c = 8-my, d = my; \ | |
431 int x, y; \ | |
432 \ | |
433 for (y = 0; y < h; y++) { \ | |
434 for (x = 0; x < SIZE; x++) \ | |
435 dst[x] = (c*src[x] + d*src[x+stride] + 4) >> 3; \ | |
436 dst += stride; \ | |
437 src += stride; \ | |
438 } \ | |
439 } \ | |
440 \ | |
441 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \ | |
442 { \ | |
443 int a = 8-mx, b = mx; \ | |
444 int c = 8-my, d = my; \ | |
445 int x, y; \ | |
446 uint8_t tmp_array[(2*SIZE+1)*SIZE]; \ | |
447 uint8_t *tmp = tmp_array; \ | |
448 \ | |
449 for (y = 0; y < h+1; y++) { \ | |
450 for (x = 0; x < SIZE; x++) \ | |
451 tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \ | |
452 tmp += SIZE; \ | |
453 src += stride; \ | |
454 } \ | |
455 \ | |
456 tmp = tmp_array; \ | |
457 \ | |
458 for (y = 0; y < h; y++) { \ | |
459 for (x = 0; x < SIZE; x++) \ | |
460 dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \ | |
461 dst += stride; \ | |
462 tmp += SIZE; \ | |
463 } \ | |
464 } | |
465 | |
466 VP8_BILINEAR(16) | |
467 VP8_BILINEAR(8) | |
468 VP8_BILINEAR(4) | |
469 | |
11921 | 470 #define VP8_MC_FUNC(IDX, SIZE) \ |
11950 | 471 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \ |
11921 | 472 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \ |
473 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \ | |
474 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \ | |
475 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \ | |
476 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \ | |
477 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \ | |
478 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \ | |
479 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c | |
480 | |
11974 | 481 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \ |
482 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \ | |
483 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \ | |
484 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \ | |
485 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \ | |
486 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \ | |
487 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \ | |
488 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \ | |
489 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \ | |
490 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c | |
491 | |
11921 | 492 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp) |
493 { | |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
494 dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c; |
12342 | 495 dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c; |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
496 dsp->vp8_idct_add = vp8_idct_add_c; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
497 dsp->vp8_idct_dc_add = vp8_idct_dc_add_c; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
498 dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
499 dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c; |
11921 | 500 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
501 dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
502 dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
503 dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
504 dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c; |
11921 | 505 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
506 dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
507 dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
508 dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12081
diff
changeset
|
509 dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c; |
11921 | 510 |
511 dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c; | |
512 dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c; | |
513 | |
514 VP8_MC_FUNC(0, 16); | |
515 VP8_MC_FUNC(1, 8); | |
516 VP8_MC_FUNC(2, 4); | |
11974 | 517 |
518 VP8_BILINEAR_MC_FUNC(0, 16); | |
519 VP8_BILINEAR_MC_FUNC(1, 8); | |
520 VP8_BILINEAR_MC_FUNC(2, 4); | |
11975 | 521 |
11985 | 522 if (HAVE_MMX) |
11975 | 523 ff_vp8dsp_init_x86(dsp); |
12011 | 524 if (HAVE_ALTIVEC) |
525 ff_vp8dsp_init_altivec(dsp); | |
11921 | 526 } |