Native VP8 decoder.

Patch by David Conrad <lessen42 gmail com> and myself.

Originally committed as revision 23719 to svn://svn.ffmpeg.org/ffmpeg/trunk

doc/general.texi

@@ -441,8 +441,8 @@ following image formats are supported:
     @tab fourcc: VP50
 @item On2 VP6                @tab     @tab  X
     @tab fourcc: VP60,VP61,VP62
-@item VP8                    @tab  X  @tab  X
-    @tab fourcc: VP80, de/encoding supported through external library libvpx
+@item VP8                    @tab  E  @tab  X
+    @tab fourcc: VP80, encoding supported through external library libvpx
 @item planar RGB             @tab     @tab  X
     @tab fourcc: 8BPS
 @item Q-team QPEG            @tab     @tab  X

libavcodec/Makefile

@@ -375,6 +375,8 @@ OBJS-$(CONFIG_VP5_DECODER)             += vp5.o vp56.o vp56data.o vp56dsp.o \
                                           vp3dsp.o
 OBJS-$(CONFIG_VP6_DECODER)             += vp6.o vp56.o vp56data.o vp56dsp.o \
                                           vp3dsp.o vp6dsp.o huffman.o
+OBJS-$(CONFIG_VP8_DECODER)             += vp8.o vp8dsp.o vp56.o vp56data.o \
+                                          h264pred.o
 OBJS-$(CONFIG_VQA_DECODER)             += vqavideo.o
 OBJS-$(CONFIG_WAVPACK_DECODER)         += wavpack.o
 OBJS-$(CONFIG_WMAPRO_DECODER)          += wmaprodec.o wma.o

libavcodec/allcodecs.c

@@ -201,6 +201,7 @@ void avcodec_register_all(void)
     REGISTER_DECODER (VP6, vp6);
     REGISTER_DECODER (VP6A, vp6a);
     REGISTER_DECODER (VP6F, vp6f);
+    REGISTER_DECODER (VP8, vp8);
     REGISTER_DECODER (VQA, vqa);
     REGISTER_ENCDEC  (WMV1, wmv1);
     REGISTER_ENCDEC  (WMV2, wmv2);

libavcodec/avcodec.h

@@ -30,8 +30,8 @@
 #include "libavutil/avutil.h"
 
 #define LIBAVCODEC_VERSION_MAJOR 52
-#define LIBAVCODEC_VERSION_MINOR 77
-#define LIBAVCODEC_VERSION_MICRO  1
+#define LIBAVCODEC_VERSION_MINOR 78
+#define LIBAVCODEC_VERSION_MICRO  0
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \

libavcodec/dsputil.c

@@ -39,6 +39,7 @@
 #include "ac3dec.h"
 #include "vorbis.h"
 #include "png.h"
+#include "vp8dsp.h"
 
 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, };
 uint32_t ff_squareTbl[512] = {0, };
@@ -2656,6 +2657,18 @@ static void avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){
 }
 #endif /* CONFIG_RV40_DECODER */
 
+#if CONFIG_VP8_DECODER
+void ff_put_vp8_pixels16_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) {
+    put_pixels16_c(dst, src, stride, h);
+}
+void ff_put_vp8_pixels8_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) {
+    put_pixels8_c(dst, src, stride, h);
+}
+void ff_put_vp8_pixels4_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) {
+    put_pixels4_c(dst, src, stride, h);
+}
+#endif
+
 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){
     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
     int i;

libavcodec/vp56.h

@@ -237,6 +237,12 @@ static inline int vp56_rac_get(VP56RangeCoder *c)
     return bit;
 }
 
+// rounding is different than vp56_rac_get, is vp56_rac_get wrong?
+static inline int vp8_rac_get(VP56RangeCoder *c)
+{
+    return vp56_rac_get_prob(c, 128);
+}
+
 static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
 {
     int value = 0;
@@ -248,12 +254,46 @@ static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
     return value;
 }
 
+static inline int vp8_rac_get_uint(VP56RangeCoder *c, int bits)
+{
+    int value = 0;
+
+    while (bits--) {
+        value = (value << 1) | vp8_rac_get(c);
+    }
+
+    return value;
+}
+
+// fixme: add 1 bit to all the calls to this?
+static inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
+{
+    int v;
+
+    if (!vp8_rac_get(c))
+        return 0;
+
+    v = vp8_rac_get_uint(c, bits);
+
+    if (vp8_rac_get(c))
+        v = -v;
+
+    return v;
+}
+
+// P(7)
 static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
 {
     int v = vp56_rac_gets(c, 7) << 1;
     return v + !v;
 }
 
+static inline int vp8_rac_get_nn(VP56RangeCoder *c)
+{
+    int v = vp8_rac_get_uint(c, 7) << 1;
+    return v + !v;
+}
+
 static inline int vp56_rac_get_tree(VP56RangeCoder *c,
                                     const VP56Tree *tree,
                                     const uint8_t *probs)
@@ -267,4 +307,39 @@ static inline int vp56_rac_get_tree(VP56RangeCoder *c,
     return -tree->val;
 }
 
+/**
+ * This is identical to vp8_rac_get_tree except for the possibility of starting
+ * on a node other than the root node, needed for coeff decode where this is
+ * used to save a bit after a 0 token (by disallowing EOB to immediately follow.)
+ */
+static inline int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2],
+                                               const uint8_t *probs, int i)
+{
+    do {
+        i = tree[i][vp56_rac_get_prob(c, probs[i])];
+    } while (i > 0);
+
+    return -i;
+}
+
+// how probabilities are associated with decisions is different I think
+// well, the new scheme fits in the old but this way has one fewer branches per decision
+static inline int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2],
+                                   const uint8_t *probs)
+{
+    return vp8_rac_get_tree_with_offset(c, tree, probs, 0);
+}
+
+// DCTextra
+static inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob)
+{
+    int v = 0;
+
+    do {
+        v = (v<<1) + vp56_rac_get_prob(c, *prob++);
+    } while (*prob);
+
+    return v;
+}
+
 #endif /* AVCODEC_VP56_H */

libavcodec/vp8dsp.c

+/**
+ * VP8 compatible video decoder
+ *
+ * Copyright (C) 2010 David Conrad
+ * Copyright (C) 2010 Ronald S. Bultje
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "dsputil.h"
+#include "vp8dsp.h"
+
+// TODO: Maybe add dequant
+static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
+{
+    int i, t0, t1, t2, t3;
+
+    for (i = 0; i < 4; i++) {
+        t0 = dc[0*4+i] + dc[3*4+i];
+        t1 = dc[1*4+i] + dc[2*4+i];
+        t2 = dc[1*4+i] - dc[2*4+i];
+        t3 = dc[0*4+i] - dc[3*4+i];
+
+        dc[0*4+i] = t0 + t1;
+        dc[1*4+i] = t3 + t2;
+        dc[2*4+i] = t0 - t1;
+        dc[3*4+i] = t3 - t2;
+    }
+
+    for (i = 0; i < 4; i++) {
+        t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding
+        t1 = dc[i*4+1] + dc[i*4+2];
+        t2 = dc[i*4+1] - dc[i*4+2];
+        t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
+
+        *block[i][0] = (t0 + t1) >> 3;
+        *block[i][1] = (t3 + t2) >> 3;
+        *block[i][2] = (t0 - t1) >> 3;
+        *block[i][3] = (t3 - t2) >> 3;
+    }
+}
+
+
+#define MUL_20091(a) ((((a)*20091) >> 16) + (a))
+#define MUL_35468(a)  (((a)*35468) >> 16)
+
+static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
+{
+    int i, t0, t1, t2, t3;
+    DCTELEM tmp[16];
+
+    for (i = 0; i < 4; i++) {
+        t0 = block[0*4+i] + block[2*4+i];
+        t1 = block[0*4+i] - block[2*4+i];
+        t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
+        t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
+
+        tmp[i*4+0] = t0 + t3;
+        tmp[i*4+1] = t1 + t2;
+        tmp[i*4+2] = t1 - t2;
+        tmp[i*4+3] = t0 - t3;
+    }
+
+    for (i = 0; i < 4; i++) {
+        t0 = tmp[0*4+i] + tmp[2*4+i];
+        t1 = tmp[0*4+i] - tmp[2*4+i];
+        t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
+        t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
+
+        dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
+        dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
+        dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
+        dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
+        dst += stride;
+    }
+}
+
+static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
+{
+    int i, dc = (block[0] + 4) >> 3;
+
+    for (i = 0; i < 4; i++) {
+        dst[0] = av_clip_uint8(dst[0] + dc);
+        dst[1] = av_clip_uint8(dst[1] + dc);
+        dst[2] = av_clip_uint8(dst[2] + dc);
+        dst[3] = av_clip_uint8(dst[3] + dc);
+        dst += stride;
+    }
+}
+
+
+// because I like only having two parameters to pass functions...
+#define LOAD_PIXELS\
+    int av_unused p3 = p[-4*stride];\
+    int av_unused p2 = p[-3*stride];\
+    int av_unused p1 = p[-2*stride];\
+    int av_unused p0 = p[-1*stride];\
+    int av_unused q0 = p[ 0*stride];\
+    int av_unused q1 = p[ 1*stride];\
+    int av_unused q2 = p[ 2*stride];\
+    int av_unused q3 = p[ 3*stride];
+
+static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
+{
+    LOAD_PIXELS
+    int a, f1, f2;
+
+    a = 3*(q0 - p0);
+
+    if (is4tap)
+        a += av_clip_int8(p1 - q1);
+
+    a = av_clip_int8(a);
+
+    // We deviate from the spec here with c(a+3) >> 3
+    // since that's what libvpx does.
+    f1 = FFMIN(a+4, 127) >> 3;
+    f2 = FFMIN(a+3, 127) >> 3;
+
+    // Despite what the spec says, we do need to clamp here to
+    // be bitexact with libvpx.
+    p[-1*stride] = av_clip_uint8(p0 + f2);
+    p[ 0*stride] = av_clip_uint8(q0 - f1);
+
+    // only used for _inner on blocks without high edge variance
+    if (!is4tap) {
+        a = (f1+1)>>1;
+        p[-2*stride] = av_clip_uint8(p1 + a);
+        p[ 1*stride] = av_clip_uint8(q1 - a);
+    }
+}
+
+static av_always_inline int simple_limit(uint8_t *p, int stride, int flim)
+{
+    LOAD_PIXELS
+    return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
+}
+
+/**
+ * E - limit at the macroblock edge
+ * I - limit for interior difference
+ */
+static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I)
+{
+    LOAD_PIXELS
+    return simple_limit(p, stride, 2*E+I)
+        && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
+        && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
+}
+
+// high edge variance
+static av_always_inline int hev(uint8_t *p, int stride, int thresh)
+{
+    LOAD_PIXELS
+    return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
+}
+
+static av_always_inline void filter_mbedge(uint8_t *p, int stride)
+{
+    int a0, a1, a2, w;
+
+    LOAD_PIXELS
+
+    w = av_clip_int8(p1-q1);
+    w = av_clip_int8(w + 3*(q0-p0));
+
+    a0 = (27*w + 63) >> 7;
+    a1 = (18*w + 63) >> 7;
+    a2 = ( 9*w + 63) >> 7;
+
+    p[-3*stride] = av_clip_uint8(p2 + a2);
+    p[-2*stride] = av_clip_uint8(p1 + a1);
+    p[-1*stride] = av_clip_uint8(p0 + a0);
+    p[ 0*stride] = av_clip_uint8(q0 - a0);
+    p[ 1*stride] = av_clip_uint8(q1 - a1);
+    p[ 2*stride] = av_clip_uint8(q2 - a2);
+}
+
+#define LOOP_FILTER(dir, size, stridea, strideb) \
+static void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
+                                     int flim_E, int flim_I, int hev_thresh)\
+{\
+    int i;\
+\
+    for (i = 0; i < size; i++)\
+        if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
+            if (hev(dst+i*stridea, strideb, hev_thresh))\
+                filter_common(dst+i*stridea, strideb, 1);\
+            else\
+                filter_mbedge(dst+i*stridea, strideb);\
+        }\
+}\
+\
+static void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
+                                      int flim_E, int flim_I, int hev_thresh)\
+{\
+    int i, hv;\
+\
+    for (i = 0; i < size; i++)\
+        if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
+            hv = hev(dst+i*stridea, strideb, hev_thresh);\
+            filter_common(dst+i*stridea, strideb, hv);\
+        }\
+}
+
+LOOP_FILTER(v, 16, 1, stride)
+LOOP_FILTER(h, 16, stride, 1)
+LOOP_FILTER(v,  8, 1, stride)
+LOOP_FILTER(h,  8, stride, 1)
+
+static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
+{
+    int i;
+
+    for (i = 0; i < 16; i++)
+        if (simple_limit(dst+i, stride, flim))
+            filter_common(dst+i, stride, 1);
+}
+
+static void vp8_h_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
+{
+    int i;
+
+    for (i = 0; i < 16; i++)
+        if (simple_limit(dst+i*stride, 1, flim))
+            filter_common(dst+i*stride, 1, 1);
+}
+
+static const uint8_t subpel_filters[7][6] = {
+    { 0,   6, 123,  12,   1,   0 },
+    { 2,  11, 108,  36,   8,   1 },
+    { 0,   9,  93,  50,   6,   0 },
+    { 3,  16,  77,  77,  16,   3 },
+    { 0,   6,  50,  93,   9,   0 },
+    { 1,   8,  36, 108,  11,   2 },
+    { 0,   1,  12, 123,   6,   0 },
+};
+
+
+#define FILTER_6TAP(src, F, stride) \
+    av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
+                   F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7)
+
+#define FILTER_4TAP(src, F, stride) \
+    av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
+                   F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7)
+
+#define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \
+static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, uint8_t *src, int stride, int h, int mx, int my) \
+{ \
+    const uint8_t *filter = subpel_filters[mx-1]; \
+    int x, y; \
+\
+    for (y = 0; y < h; y++) { \
+        for (x = 0; x < SIZE; x++) \
+            dst[x] = FILTER(src, filter, 1); \
+        dst += stride; \
+        src += stride; \
+    } \
+}
+#define VP8_EPEL_V(SIZE, FILTER, FILTERNAME) \
+static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, uint8_t *src, int stride, int h, int mx, int my) \
+{ \
+    const uint8_t *filter = subpel_filters[my-1]; \
+    int x, y; \
+\
+    for (y = 0; y < h; y++) { \
+        for (x = 0; x < SIZE; x++) \
+            dst[x] = FILTER(src, filter, stride); \
+        dst += stride; \
+        src += stride; \
+    } \
+}
+#define VP8_EPEL_HV(SIZE, FILTERX, FILTERY, FILTERNAME) \
+static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, uint8_t *src, int stride, int h, int mx, int my) \
+{ \
+    const uint8_t *filter = subpel_filters[mx-1]; \
+    int x, y; \
+    uint8_t tmp_array[(2*SIZE+5)*SIZE]; \
+    uint8_t *tmp = tmp_array; \
+    src -= 2*stride; \
+\
+    for (y = 0; y < h+5; y++) { \
+        for (x = 0; x < SIZE; x++) \
+            tmp[x] = FILTERX(src, filter, 1); \
+        tmp += SIZE; \
+        src += stride; \
+    } \
+\
+    tmp = tmp_array + 2*SIZE; \
+    filter = subpel_filters[my-1]; \
+\
+    for (y = 0; y < h; y++) { \
+        for (x = 0; x < SIZE; x++) \
+            dst[x] = FILTERY(tmp, filter, SIZE); \
+        dst += stride; \
+        tmp += SIZE; \
+    } \
+}
+
+VP8_EPEL_H(16, FILTER_4TAP, h4)
+VP8_EPEL_H(8,  FILTER_4TAP, h4)
+VP8_EPEL_H(4,  FILTER_4TAP, h4)
+VP8_EPEL_H(16, FILTER_6TAP, h6)
+VP8_EPEL_H(8,  FILTER_6TAP, h6)
+VP8_EPEL_H(4,  FILTER_6TAP, h6)
+VP8_EPEL_V(16, FILTER_4TAP, v4)
+VP8_EPEL_V(8,  FILTER_4TAP, v4)
+VP8_EPEL_V(4,  FILTER_4TAP, v4)
+VP8_EPEL_V(16, FILTER_6TAP, v6)
+VP8_EPEL_V(8,  FILTER_6TAP, v6)
+VP8_EPEL_V(4,  FILTER_6TAP, v6)
+VP8_EPEL_HV(16, FILTER_4TAP, FILTER_4TAP, h4v4)
+VP8_EPEL_HV(8,  FILTER_4TAP, FILTER_4TAP, h4v4)
+VP8_EPEL_HV(4,  FILTER_4TAP, FILTER_4TAP, h4v4)
+VP8_EPEL_HV(16, FILTER_4TAP, FILTER_6TAP, h4v6)
+VP8_EPEL_HV(8,  FILTER_4TAP, FILTER_6TAP, h4v6)
+VP8_EPEL_HV(4,  FILTER_4TAP, FILTER_6TAP, h4v6)
+VP8_EPEL_HV(16, FILTER_6TAP, FILTER_4TAP, h6v4)
+VP8_EPEL_HV(8,  FILTER_6TAP, FILTER_4TAP, h6v4)
+VP8_EPEL_HV(4,  FILTER_6TAP, FILTER_4TAP, h6v4)
+VP8_EPEL_HV(16, FILTER_6TAP, FILTER_6TAP, h6v6)
+VP8_EPEL_HV(8,  FILTER_6TAP, FILTER_6TAP, h6v6)
+VP8_EPEL_HV(4,  FILTER_6TAP, FILTER_6TAP, h6v6)
+
+#define VP8_MC_FUNC(IDX, SIZE) \
+    dsp->put_vp8_epel_pixels_tab[IDX][0][0] = ff_put_vp8_pixels ## SIZE ## _c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
+    dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
+
+av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
+{
+    dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
+    dsp->vp8_idct_add    = vp8_idct_add_c;
+    dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
+
+    dsp->vp8_v_loop_filter16 = vp8_v_loop_filter16_c;
+    dsp->vp8_h_loop_filter16 = vp8_h_loop_filter16_c;
+    dsp->vp8_v_loop_filter8  = vp8_v_loop_filter8_c;
+    dsp->vp8_h_loop_filter8  = vp8_h_loop_filter8_c;
+
+    dsp->vp8_v_loop_filter16_inner = vp8_v_loop_filter16_inner_c;
+    dsp->vp8_h_loop_filter16_inner = vp8_h_loop_filter16_inner_c;
+    dsp->vp8_v_loop_filter8_inner  = vp8_v_loop_filter8_inner_c;
+    dsp->vp8_h_loop_filter8_inner  = vp8_h_loop_filter8_inner_c;
+
+    dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
+    dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
+
+    VP8_MC_FUNC(0, 16);
+    VP8_MC_FUNC(1, 8);
+    VP8_MC_FUNC(2, 4);
+}

libavcodec/vp8dsp.h

+/**
+ * VP8 compatible video decoder
+ *
+ * Copyright (C) 2010 David Conrad
+ * Copyright (C) 2010 Ronald S. Bultje
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+
+#ifndef AVCODEC_VP8DSP_H
+#define AVCODEC_VP8DSP_H
+
+#include "dsputil.h"
+
+typedef struct VP8DSPContext {
+    void (*vp8_luma_dc_wht)(DCTELEM block[4][4][16], DCTELEM dc[16]);
+    void (*vp8_idct_add)(uint8_t *dst, DCTELEM block[16], int stride);
+    void (*vp8_idct_dc_add)(uint8_t *dst, DCTELEM block[16], int stride);
+
+    // loop filter applied to edges between macroblocks
+    void (*vp8_v_loop_filter16)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter16)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter8)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter8)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+
+    // loop filter applied to inner macroblock edges
+    void (*vp8_v_loop_filter16_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter16_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter8_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter8_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+
+    void (*vp8_v_loop_filter_simple)(uint8_t *dst, int stride, int flim);
+    void (*vp8_h_loop_filter_simple)(uint8_t *dst, int stride, int flim);
+
+    /**
+     * first dimension: width>>3, height is assumed equal to width
+     * second dimension: 0 if no vertical interpolation is needed;
+     *                   1 4-tap vertical interpolation filter (my & 1)
+     *                   2 6-tap vertical interpolation filter (!(my & 1))
+     * third dimension: same as second dimention, for horizontal interpolation
+     * so something like put_vp8_epel_pixels_tab[width>>3][2*!!my-(my&1)][2*!!mx-(mx&1)](..., mx, my)
+     */
+    h264_chroma_mc_func put_vp8_epel_pixels_tab[3][3][3];
+} VP8DSPContext;
+
+void ff_put_vp8_pixels16_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y);
+void ff_put_vp8_pixels8_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y);
+void ff_put_vp8_pixels4_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y);
+
+void ff_vp8dsp_init(VP8DSPContext *c);
+
+#endif /* AVCODEC_VP8DSP_H */