2 * Copyright (C) 2010 David Conrad
3 * Copyright (C) 2010 Ronald S. Bultje
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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
30 // TODO: Maybe add dequant
31 static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
33 int i, t0, t1, t2, t3;
35 for (i = 0; i < 4; i++) {
36 t0 = dc[0*4+i] + dc[3*4+i];
37 t1 = dc[1*4+i] + dc[2*4+i];
38 t2 = dc[1*4+i] - dc[2*4+i];
39 t3 = dc[0*4+i] - dc[3*4+i];
47 for (i = 0; i < 4; i++) {
48 t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding
49 t1 = dc[i*4+1] + dc[i*4+2];
50 t2 = dc[i*4+1] - dc[i*4+2];
51 t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
57 block[i][0][0] = (t0 + t1) >> 3;
58 block[i][1][0] = (t3 + t2) >> 3;
59 block[i][2][0] = (t0 - t1) >> 3;
60 block[i][3][0] = (t3 - t2) >> 3;
64 static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16])
66 int i, val = (dc[0] + 3) >> 3;
69 for (i = 0; i < 4; i++) {
77 #define MUL_20091(a) ((((a)*20091) >> 16) + (a))
78 #define MUL_35468(a) (((a)*35468) >> 16)
80 static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], ptrdiff_t stride)
82 int i, t0, t1, t2, t3;
85 for (i = 0; i < 4; i++) {
86 t0 = block[0*4+i] + block[2*4+i];
87 t1 = block[0*4+i] - block[2*4+i];
88 t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
89 t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
101 for (i = 0; i < 4; i++) {
102 t0 = tmp[0*4+i] + tmp[2*4+i];
103 t1 = tmp[0*4+i] - tmp[2*4+i];
104 t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
105 t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
107 dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
108 dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
109 dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
110 dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
115 static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], ptrdiff_t stride)
117 int i, dc = (block[0] + 4) >> 3;
120 for (i = 0; i < 4; i++) {
121 dst[0] = av_clip_uint8(dst[0] + dc);
122 dst[1] = av_clip_uint8(dst[1] + dc);
123 dst[2] = av_clip_uint8(dst[2] + dc);
124 dst[3] = av_clip_uint8(dst[3] + dc);
129 static void vp8_idct_dc_add4uv_c(uint8_t *dst, DCTELEM block[4][16], ptrdiff_t stride)
131 vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride);
132 vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride);
133 vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride);
134 vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride);
137 static void vp8_idct_dc_add4y_c(uint8_t *dst, DCTELEM block[4][16], ptrdiff_t stride)
139 vp8_idct_dc_add_c(dst+ 0, block[0], stride);
140 vp8_idct_dc_add_c(dst+ 4, block[1], stride);
141 vp8_idct_dc_add_c(dst+ 8, block[2], stride);
142 vp8_idct_dc_add_c(dst+12, block[3], stride);
145 // because I like only having two parameters to pass functions...
147 int av_unused p3 = p[-4*stride];\
148 int av_unused p2 = p[-3*stride];\
149 int av_unused p1 = p[-2*stride];\
150 int av_unused p0 = p[-1*stride];\
151 int av_unused q0 = p[ 0*stride];\
152 int av_unused q1 = p[ 1*stride];\
153 int av_unused q2 = p[ 2*stride];\
154 int av_unused q3 = p[ 3*stride];
156 #define clip_int8(n) (cm[n+0x80]-0x80)
158 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride, int is4tap)
162 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
167 a += clip_int8(p1 - q1);
171 // We deviate from the spec here with c(a+3) >> 3
172 // since that's what libvpx does.
173 f1 = FFMIN(a+4, 127) >> 3;
174 f2 = FFMIN(a+3, 127) >> 3;
176 // Despite what the spec says, we do need to clamp here to
177 // be bitexact with libvpx.
178 p[-1*stride] = cm[p0 + f2];
179 p[ 0*stride] = cm[q0 - f1];
181 // only used for _inner on blocks without high edge variance
184 p[-2*stride] = cm[p1 + a];
185 p[ 1*stride] = cm[q1 - a];
189 static av_always_inline int simple_limit(uint8_t *p, ptrdiff_t stride, int flim)
192 return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
196 * E - limit at the macroblock edge
197 * I - limit for interior difference
199 static av_always_inline int normal_limit(uint8_t *p, ptrdiff_t stride, int E, int I)
202 return simple_limit(p, stride, E)
203 && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
204 && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
207 // high edge variance
208 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh)
211 return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
214 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride)
217 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
221 w = clip_int8(p1-q1);
222 w = clip_int8(w + 3*(q0-p0));
224 a0 = (27*w + 63) >> 7;
225 a1 = (18*w + 63) >> 7;
226 a2 = ( 9*w + 63) >> 7;
228 p[-3*stride] = cm[p2 + a2];
229 p[-2*stride] = cm[p1 + a1];
230 p[-1*stride] = cm[p0 + a0];
231 p[ 0*stride] = cm[q0 - a0];
232 p[ 1*stride] = cm[q1 - a1];
233 p[ 2*stride] = cm[q2 - a2];
236 #define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
237 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, ptrdiff_t stride,\
238 int flim_E, int flim_I, int hev_thresh)\
242 for (i = 0; i < size; i++)\
243 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
244 if (hev(dst+i*stridea, strideb, hev_thresh))\
245 filter_common(dst+i*stridea, strideb, 1);\
247 filter_mbedge(dst+i*stridea, strideb);\
251 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, ptrdiff_t stride,\
252 int flim_E, int flim_I, int hev_thresh)\
256 for (i = 0; i < size; i++)\
257 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
258 int hv = hev(dst+i*stridea, strideb, hev_thresh);\
260 filter_common(dst+i*stridea, strideb, 1);\
262 filter_common(dst+i*stridea, strideb, 0);\
266 LOOP_FILTER(v, 16, 1, stride,)
267 LOOP_FILTER(h, 16, stride, 1,)
269 #define UV_LOOP_FILTER(dir, stridea, strideb) \
270 LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
271 static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, ptrdiff_t stride,\
272 int fE, int fI, int hev_thresh)\
274 vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
275 vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
277 static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, ptrdiff_t stride,\
278 int fE, int fI, int hev_thresh)\
280 vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
281 vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
284 UV_LOOP_FILTER(v, 1, stride)
285 UV_LOOP_FILTER(h, stride, 1)
287 static void vp8_v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, int flim)
291 for (i = 0; i < 16; i++)
292 if (simple_limit(dst+i, stride, flim))
293 filter_common(dst+i, stride, 1);
296 static void vp8_h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, int flim)
300 for (i = 0; i < 16; i++)
301 if (simple_limit(dst+i*stride, 1, flim))
302 filter_common(dst+i*stride, 1, 1);
305 static const uint8_t subpel_filters[7][6] = {
306 { 0, 6, 123, 12, 1, 0 },
307 { 2, 11, 108, 36, 8, 1 },
308 { 0, 9, 93, 50, 6, 0 },
309 { 3, 16, 77, 77, 16, 3 },
310 { 0, 6, 50, 93, 9, 0 },
311 { 1, 8, 36, 108, 11, 2 },
312 { 0, 1, 12, 123, 6, 0 },
315 #define PUT_PIXELS(WIDTH) \
316 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) { \
318 for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \
319 memcpy(dst, src, WIDTH); \
327 #define FILTER_6TAP(src, F, stride) \
328 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
329 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
331 #define FILTER_4TAP(src, F, stride) \
332 cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
333 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
335 #define VP8_EPEL_H(SIZE, TAPS) \
336 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) \
338 const uint8_t *filter = subpel_filters[mx-1]; \
339 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
342 for (y = 0; y < h; y++) { \
343 for (x = 0; x < SIZE; x++) \
344 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1); \
349 #define VP8_EPEL_V(SIZE, TAPS) \
350 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) \
352 const uint8_t *filter = subpel_filters[my-1]; \
353 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
356 for (y = 0; y < h; y++) { \
357 for (x = 0; x < SIZE; x++) \
358 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride); \
363 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS) \
364 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) \
366 const uint8_t *filter = subpel_filters[mx-1]; \
367 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
369 uint8_t tmp_array[(2*SIZE+VTAPS-1)*SIZE]; \
370 uint8_t *tmp = tmp_array; \
371 src -= (2-(VTAPS==4))*srcstride; \
373 for (y = 0; y < h+VTAPS-1; y++) { \
374 for (x = 0; x < SIZE; x++) \
375 tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1); \
380 tmp = tmp_array + (2-(VTAPS==4))*SIZE; \
381 filter = subpel_filters[my-1]; \
383 for (y = 0; y < h; y++) { \
384 for (x = 0; x < SIZE; x++) \
385 dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE); \
403 VP8_EPEL_HV(16, 4, 4)
406 VP8_EPEL_HV(16, 4, 6)
409 VP8_EPEL_HV(16, 6, 4)
412 VP8_EPEL_HV(16, 6, 6)
416 #define VP8_BILINEAR(SIZE) \
417 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s2, int h, int mx, int my) \
419 int a = 8-mx, b = mx; \
422 for (y = 0; y < h; y++) { \
423 for (x = 0; x < SIZE; x++) \
424 dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
429 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s2, int h, int mx, int my) \
431 int c = 8-my, d = my; \
434 for (y = 0; y < h; y++) { \
435 for (x = 0; x < SIZE; x++) \
436 dst[x] = (c*src[x] + d*src[x+stride] + 4) >> 3; \
442 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s2, int h, int mx, int my) \
444 int a = 8-mx, b = mx; \
445 int c = 8-my, d = my; \
447 uint8_t tmp_array[(2*SIZE+1)*SIZE]; \
448 uint8_t *tmp = tmp_array; \
450 for (y = 0; y < h+1; y++) { \
451 for (x = 0; x < SIZE; x++) \
452 tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
459 for (y = 0; y < h; y++) { \
460 for (x = 0; x < SIZE; x++) \
461 dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \
471 #define VP8_MC_FUNC(IDX, SIZE) \
472 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
473 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
474 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
475 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
476 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
477 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
478 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
479 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
480 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
482 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \
483 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
484 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
485 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
486 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
487 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
488 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
489 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
490 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
491 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
493 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
495 dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
496 dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
497 dsp->vp8_idct_add = vp8_idct_add_c;
498 dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
499 dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c;
500 dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
502 dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
503 dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
504 dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
505 dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
507 dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
508 dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
509 dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
510 dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
512 dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
513 dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
519 VP8_BILINEAR_MC_FUNC(0, 16);
520 VP8_BILINEAR_MC_FUNC(1, 8);
521 VP8_BILINEAR_MC_FUNC(2, 4);
524 ff_vp8dsp_init_x86(dsp);
526 ff_vp8dsp_init_altivec(dsp);
528 ff_vp8dsp_init_arm(dsp);