2 * Copyright (C) 2010 David Conrad
3 * Copyright (C) 2010 Ronald S. Bultje
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * VP8 compatible video decoder
29 #include "libavutil/common.h"
31 // TODO: Maybe add dequant
32 static void vp8_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
34 int i, t0, t1, t2, t3;
36 for (i = 0; i < 4; i++) {
37 t0 = dc[0*4+i] + dc[3*4+i];
38 t1 = dc[1*4+i] + dc[2*4+i];
39 t2 = dc[1*4+i] - dc[2*4+i];
40 t3 = dc[0*4+i] - dc[3*4+i];
48 for (i = 0; i < 4; i++) {
49 t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding
50 t1 = dc[i*4+1] + dc[i*4+2];
51 t2 = dc[i*4+1] - dc[i*4+2];
52 t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
58 block[i][0][0] = (t0 + t1) >> 3;
59 block[i][1][0] = (t3 + t2) >> 3;
60 block[i][2][0] = (t0 - t1) >> 3;
61 block[i][3][0] = (t3 - t2) >> 3;
65 static void vp8_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
67 int i, val = (dc[0] + 3) >> 3;
70 for (i = 0; i < 4; i++) {
78 #define MUL_20091(a) ((((a)*20091) >> 16) + (a))
79 #define MUL_35468(a) (((a)*35468) >> 16)
81 static void vp8_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
83 int i, t0, t1, t2, t3;
86 for (i = 0; i < 4; i++) {
87 t0 = block[0*4+i] + block[2*4+i];
88 t1 = block[0*4+i] - block[2*4+i];
89 t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
90 t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
102 for (i = 0; i < 4; i++) {
103 t0 = tmp[0*4+i] + tmp[2*4+i];
104 t1 = tmp[0*4+i] - tmp[2*4+i];
105 t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
106 t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
108 dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
109 dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
110 dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
111 dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
116 static void vp8_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
118 int i, dc = (block[0] + 4) >> 3;
121 for (i = 0; i < 4; i++) {
122 dst[0] = av_clip_uint8(dst[0] + dc);
123 dst[1] = av_clip_uint8(dst[1] + dc);
124 dst[2] = av_clip_uint8(dst[2] + dc);
125 dst[3] = av_clip_uint8(dst[3] + dc);
130 static void vp8_idct_dc_add4uv_c(uint8_t *dst, int16_t block[4][16], ptrdiff_t stride)
132 vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride);
133 vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride);
134 vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride);
135 vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride);
138 static void vp8_idct_dc_add4y_c(uint8_t *dst, int16_t block[4][16], ptrdiff_t stride)
140 vp8_idct_dc_add_c(dst+ 0, block[0], stride);
141 vp8_idct_dc_add_c(dst+ 4, block[1], stride);
142 vp8_idct_dc_add_c(dst+ 8, block[2], stride);
143 vp8_idct_dc_add_c(dst+12, block[3], stride);
146 // because I like only having two parameters to pass functions...
148 int av_unused p3 = p[-4*stride];\
149 int av_unused p2 = p[-3*stride];\
150 int av_unused p1 = p[-2*stride];\
151 int av_unused p0 = p[-1*stride];\
152 int av_unused q0 = p[ 0*stride];\
153 int av_unused q1 = p[ 1*stride];\
154 int av_unused q2 = p[ 2*stride];\
155 int av_unused q3 = p[ 3*stride];
157 #define clip_int8(n) (cm[n+0x80]-0x80)
159 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride, int is4tap)
163 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
168 a += clip_int8(p1 - q1);
172 // We deviate from the spec here with c(a+3) >> 3
173 // since that's what libvpx does.
174 f1 = FFMIN(a+4, 127) >> 3;
175 f2 = FFMIN(a+3, 127) >> 3;
177 // Despite what the spec says, we do need to clamp here to
178 // be bitexact with libvpx.
179 p[-1*stride] = cm[p0 + f2];
180 p[ 0*stride] = cm[q0 - f1];
182 // only used for _inner on blocks without high edge variance
185 p[-2*stride] = cm[p1 + a];
186 p[ 1*stride] = cm[q1 - a];
190 static av_always_inline int simple_limit(uint8_t *p, ptrdiff_t stride, int flim)
193 return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
197 * E - limit at the macroblock edge
198 * I - limit for interior difference
200 static av_always_inline int normal_limit(uint8_t *p, ptrdiff_t stride, int E, int I)
203 return simple_limit(p, stride, E)
204 && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
205 && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
208 // high edge variance
209 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh)
212 return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
215 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride)
218 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
222 w = clip_int8(p1-q1);
223 w = clip_int8(w + 3*(q0-p0));
225 a0 = (27*w + 63) >> 7;
226 a1 = (18*w + 63) >> 7;
227 a2 = ( 9*w + 63) >> 7;
229 p[-3*stride] = cm[p2 + a2];
230 p[-2*stride] = cm[p1 + a1];
231 p[-1*stride] = cm[p0 + a0];
232 p[ 0*stride] = cm[q0 - a0];
233 p[ 1*stride] = cm[q1 - a1];
234 p[ 2*stride] = cm[q2 - a2];
237 #define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
238 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, ptrdiff_t stride,\
239 int flim_E, int flim_I, int hev_thresh)\
243 for (i = 0; i < size; i++)\
244 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
245 if (hev(dst+i*stridea, strideb, hev_thresh))\
246 filter_common(dst+i*stridea, strideb, 1);\
248 filter_mbedge(dst+i*stridea, strideb);\
252 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, ptrdiff_t stride,\
253 int flim_E, int flim_I, int hev_thresh)\
257 for (i = 0; i < size; i++)\
258 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
259 int hv = hev(dst+i*stridea, strideb, hev_thresh);\
261 filter_common(dst+i*stridea, strideb, 1);\
263 filter_common(dst+i*stridea, strideb, 0);\
267 LOOP_FILTER(v, 16, 1, stride,)
268 LOOP_FILTER(h, 16, stride, 1,)
270 #define UV_LOOP_FILTER(dir, stridea, strideb) \
271 LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
272 static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, ptrdiff_t stride,\
273 int fE, int fI, int hev_thresh)\
275 vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
276 vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
278 static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, ptrdiff_t stride,\
279 int fE, int fI, int hev_thresh)\
281 vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
282 vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
285 UV_LOOP_FILTER(v, 1, stride)
286 UV_LOOP_FILTER(h, stride, 1)
288 static void vp8_v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, int flim)
292 for (i = 0; i < 16; i++)
293 if (simple_limit(dst+i, stride, flim))
294 filter_common(dst+i, stride, 1);
297 static void vp8_h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, int flim)
301 for (i = 0; i < 16; i++)
302 if (simple_limit(dst+i*stride, 1, flim))
303 filter_common(dst+i*stride, 1, 1);
306 static const uint8_t subpel_filters[7][6] = {
307 { 0, 6, 123, 12, 1, 0 },
308 { 2, 11, 108, 36, 8, 1 },
309 { 0, 9, 93, 50, 6, 0 },
310 { 3, 16, 77, 77, 16, 3 },
311 { 0, 6, 50, 93, 9, 0 },
312 { 1, 8, 36, 108, 11, 2 },
313 { 0, 1, 12, 123, 6, 0 },
316 #define PUT_PIXELS(WIDTH) \
317 static void put_vp8_pixels ## WIDTH ##_c(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, int h, int x, int y) { \
319 for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \
320 memcpy(dst, src, WIDTH); \
328 #define FILTER_6TAP(src, F, stride) \
329 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
330 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
332 #define FILTER_4TAP(src, F, stride) \
333 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
334 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
336 #define VP8_EPEL_H(SIZE, TAPS) \
337 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, int h, int mx, int my) \
339 const uint8_t *filter = subpel_filters[mx-1]; \
340 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
343 for (y = 0; y < h; y++) { \
344 for (x = 0; x < SIZE; x++) \
345 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1); \
350 #define VP8_EPEL_V(SIZE, TAPS) \
351 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, int h, int mx, int my) \
353 const uint8_t *filter = subpel_filters[my-1]; \
354 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
357 for (y = 0; y < h; y++) { \
358 for (x = 0; x < SIZE; x++) \
359 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride); \
364 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS) \
365 static void put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, int h, int mx, int my) \
367 const uint8_t *filter = subpel_filters[mx-1]; \
368 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
370 uint8_t tmp_array[(2*SIZE+VTAPS-1)*SIZE]; \
371 uint8_t *tmp = tmp_array; \
372 src -= (2-(VTAPS==4))*srcstride; \
374 for (y = 0; y < h+VTAPS-1; y++) { \
375 for (x = 0; x < SIZE; x++) \
376 tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1); \
381 tmp = tmp_array + (2-(VTAPS==4))*SIZE; \
382 filter = subpel_filters[my-1]; \
384 for (y = 0; y < h; y++) { \
385 for (x = 0; x < SIZE; x++) \
386 dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE); \
404 VP8_EPEL_HV(16, 4, 4)
407 VP8_EPEL_HV(16, 4, 6)
410 VP8_EPEL_HV(16, 6, 4)
413 VP8_EPEL_HV(16, 6, 6)
417 #define VP8_BILINEAR(SIZE) \
418 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) \
420 int a = 8-mx, b = mx; \
423 for (y = 0; y < h; y++) { \
424 for (x = 0; x < SIZE; x++) \
425 dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
430 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) \
432 int c = 8-my, d = my; \
435 for (y = 0; y < h; y++) { \
436 for (x = 0; x < SIZE; x++) \
437 dst[x] = (c*src[x] + d*src[x+sstride] + 4) >> 3; \
443 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) \
445 int a = 8-mx, b = mx; \
446 int c = 8-my, d = my; \
448 uint8_t tmp_array[(2*SIZE+1)*SIZE]; \
449 uint8_t *tmp = tmp_array; \
451 for (y = 0; y < h+1; y++) { \
452 for (x = 0; x < SIZE; x++) \
453 tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
460 for (y = 0; y < h; y++) { \
461 for (x = 0; x < SIZE; x++) \
462 dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \
472 #define VP8_MC_FUNC(IDX, SIZE) \
473 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
474 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
475 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
476 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
477 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
478 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
479 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
480 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
481 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
483 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \
484 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
485 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
486 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
487 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
488 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
489 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
490 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
491 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
492 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
494 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
496 dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
497 dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
498 dsp->vp8_idct_add = vp8_idct_add_c;
499 dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
500 dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c;
501 dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
503 dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
504 dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
505 dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
506 dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
508 dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
509 dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
510 dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
511 dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
513 dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
514 dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
520 VP8_BILINEAR_MC_FUNC(0, 16);
521 VP8_BILINEAR_MC_FUNC(1, 8);
522 VP8_BILINEAR_MC_FUNC(2, 4);
525 ff_vp8dsp_init_arm(dsp);
527 ff_vp8dsp_init_ppc(dsp);
529 ff_vp8dsp_init_x86(dsp);