ppc: Rename H.264 optimization template file for consistency.
[ffmpeg.git] / libavcodec / vp3.c
index a2a3974..373f5e0 100644 (file)
@@ -1,20 +1,20 @@
 /*
  * Copyright (C) 2003-2004 the ffmpeg project
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav 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,
+ * Libav 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
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 #include <stdlib.h>
 #include <string.h>
 
-#include "libavcore/imgutils.h"
+#include "libavutil/imgutils.h"
 #include "avcodec.h"
+#include "internal.h"
 #include "dsputil.h"
 #include "get_bits.h"
 
 #include "vp3data.h"
 #include "xiph.h"
+#include "thread.h"
 
 #define FRAGMENT_PIXELS 8
 
-static av_cold int vp3_decode_end(AVCodecContext *avctx);
-
 //FIXME split things out into their own arrays
 typedef struct Vp3Fragment {
     int16_t dc;
@@ -224,7 +224,7 @@ typedef struct Vp3DecodeContext {
 
     /* these arrays need to be on 16-byte boundaries since SSE2 operations
      * index into them */
-    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
+    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     ///< qmat[qpi][is_inter][plane]
 
     /* This table contains superblock_count * 16 entries. Each set of 16
      * numbers corresponds to the fragment indexes 0..15 of the superblock.
@@ -236,8 +236,7 @@ typedef struct Vp3DecodeContext {
      * is coded. */
     unsigned char *macroblock_coding;
 
-    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
-    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
+    uint8_t *edge_emu_buffer;
 
     /* Huffman decode */
     int hti;
@@ -254,6 +253,63 @@ typedef struct Vp3DecodeContext {
  * VP3 specific functions
  ************************************************************************/
 
+static void vp3_decode_flush(AVCodecContext *avctx)
+{
+    Vp3DecodeContext *s = avctx->priv_data;
+
+    if (s->golden_frame.data[0]) {
+        if (s->golden_frame.data[0] == s->last_frame.data[0])
+            memset(&s->last_frame, 0, sizeof(AVFrame));
+        if (s->current_frame.data[0] == s->golden_frame.data[0])
+            memset(&s->current_frame, 0, sizeof(AVFrame));
+        ff_thread_release_buffer(avctx, &s->golden_frame);
+    }
+    if (s->last_frame.data[0]) {
+        if (s->current_frame.data[0] == s->last_frame.data[0])
+            memset(&s->current_frame, 0, sizeof(AVFrame));
+        ff_thread_release_buffer(avctx, &s->last_frame);
+    }
+    if (s->current_frame.data[0])
+        ff_thread_release_buffer(avctx, &s->current_frame);
+}
+
+static av_cold int vp3_decode_end(AVCodecContext *avctx)
+{
+    Vp3DecodeContext *s = avctx->priv_data;
+    int i;
+
+    av_free(s->superblock_coding);
+    av_free(s->all_fragments);
+    av_free(s->coded_fragment_list[0]);
+    av_free(s->dct_tokens_base);
+    av_free(s->superblock_fragments);
+    av_free(s->macroblock_coding);
+    av_free(s->motion_val[0]);
+    av_free(s->motion_val[1]);
+    av_free(s->edge_emu_buffer);
+
+    if (avctx->internal->is_copy)
+        return 0;
+
+    for (i = 0; i < 16; i++) {
+        ff_free_vlc(&s->dc_vlc[i]);
+        ff_free_vlc(&s->ac_vlc_1[i]);
+        ff_free_vlc(&s->ac_vlc_2[i]);
+        ff_free_vlc(&s->ac_vlc_3[i]);
+        ff_free_vlc(&s->ac_vlc_4[i]);
+    }
+
+    ff_free_vlc(&s->superblock_run_length_vlc);
+    ff_free_vlc(&s->fragment_run_length_vlc);
+    ff_free_vlc(&s->mode_code_vlc);
+    ff_free_vlc(&s->motion_vector_vlc);
+
+    /* release all frames */
+    vp3_decode_flush(avctx);
+
+    return 0;
+}
+
 /*
  * This function sets up all of the various blocks mappings:
  * superblocks <-> fragments, macroblocks <-> fragments,
@@ -324,8 +380,6 @@ static void init_dequantizer(Vp3DecodeContext *s, int qpi)
             s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
         }
     }
-
-    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
 }
 
 /*
@@ -342,6 +396,7 @@ static void init_loop_filter(Vp3DecodeContext *s)
     int value;
 
     filter_limit = s->filter_limit_values[s->qps[0]];
+    assert(filter_limit < 128);
 
     /* set up the bounding values */
     memset(s->bounding_values_array, 0, 256 * sizeof(int));
@@ -892,7 +947,7 @@ static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
             /* decode a VLC into a token */
             token = get_vlc2(gb, vlc_table, 11, 3);
             /* use the token to get a zero run, a coefficient, and an eob run */
-            if (token <= 6) {
+            if ((unsigned) token <= 6U) {
                 eob_run = eob_run_base[token];
                 if (eob_run_get_bits[token])
                     eob_run += get_bits(gb, eob_run_get_bits[token]);
@@ -910,7 +965,7 @@ static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
                     coeff_i        += eob_run;
                     eob_run = 0;
                 }
-            } else {
+            } else if (token >= 0) {
                 bits_to_get = coeff_get_bits[token];
                 if (bits_to_get)
                     bits_to_get = get_bits(gb, bits_to_get);
@@ -944,6 +999,10 @@ static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
                 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
                     s->num_coded_frags[plane][i]--;
                 coeff_i++;
+            } else {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Invalid token %d\n", token);
+                return -1;
             }
     }
 
@@ -993,6 +1052,8 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
     /* unpack the Y plane DC coefficients */
     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
         0, residual_eob_run);
+    if (residual_eob_run < 0)
+        return residual_eob_run;
 
     /* reverse prediction of the Y-plane DC coefficients */
     reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
@@ -1000,8 +1061,12 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
     /* unpack the C plane DC coefficients */
     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
         1, residual_eob_run);
+    if (residual_eob_run < 0)
+        return residual_eob_run;
     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
         2, residual_eob_run);
+    if (residual_eob_run < 0)
+        return residual_eob_run;
 
     /* reverse prediction of the C-plane DC coefficients */
     if (!(s->avctx->flags & CODEC_FLAG_GRAY))
@@ -1038,11 +1103,17 @@ static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
     for (i = 1; i <= 63; i++) {
             residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
                 0, residual_eob_run);
+            if (residual_eob_run < 0)
+                return residual_eob_run;
 
             residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
                 1, residual_eob_run);
+            if (residual_eob_run < 0)
+                return residual_eob_run;
             residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
                 2, residual_eob_run);
+            if (residual_eob_run < 0)
+                return residual_eob_run;
     }
 
     return 0;
@@ -1293,6 +1364,10 @@ static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
         case 1: // zero run
             s->dct_tokens[plane][i]++;
             i += (token >> 2) & 0x7f;
+            if (i > 63) {
+                av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
+                return i;
+            }
             block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
             i++;
             break;
@@ -1304,6 +1379,8 @@ static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
             return i;
         }
     } while (i < 64);
+    // return value is expected to be a valid level
+    i--;
 end:
     // the actual DC+prediction is in the fragment structure
     block[0] = frag->dc * s->qmat[0][inter][plane][0];
@@ -1315,8 +1392,17 @@ end:
  */
 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
 {
-    int h, cy;
-    int offset[4];
+    int h, cy, i;
+    int offset[AV_NUM_DATA_POINTERS];
+
+    if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
+        int y_flipped = s->flipped_image ? s->avctx->height-y : y;
+
+        // At the end of the frame, report INT_MAX instead of the height of the frame.
+        // This makes the other threads' ff_thread_await_progress() calls cheaper, because
+        // they don't have to clip their values.
+        ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
+    }
 
     if(s->avctx->draw_horiz_band==NULL)
         return;
@@ -1333,19 +1419,42 @@ static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
     offset[0] = s->current_frame.linesize[0]*y;
     offset[1] = s->current_frame.linesize[1]*cy;
     offset[2] = s->current_frame.linesize[2]*cy;
-    offset[3] = 0;
+    for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
+        offset[i] = 0;
 
     emms_c();
     s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
 }
 
+/**
+ * Wait for the reference frame of the current fragment.
+ * The progress value is in luma pixel rows.
+ */
+static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
+{
+    AVFrame *ref_frame;
+    int ref_row;
+    int border = motion_y&1;
+
+    if (fragment->coding_method == MODE_USING_GOLDEN ||
+        fragment->coding_method == MODE_GOLDEN_MV)
+        ref_frame = &s->golden_frame;
+    else
+        ref_frame = &s->last_frame;
+
+    ref_row = y + (motion_y>>1);
+    ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
+
+    ff_thread_await_progress(ref_frame, ref_row, 0);
+}
+
 /*
  * Perform the final rendering for a particular slice of data.
  * The slice number ranges from 0..(c_superblock_height - 1).
  */
 static void render_slice(Vp3DecodeContext *s, int slice)
 {
-    int x, y, i, j;
+    int x, y, i, j, fragment;
     LOCAL_ALIGNED_16(DCTELEM, block, [64]);
     int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
     int motion_halfpel_index;
@@ -1371,15 +1480,12 @@ static void render_slice(Vp3DecodeContext *s, int slice)
         int fragment_width    = s->fragment_width[!!plane];
         int fragment_height   = s->fragment_height[!!plane];
         int fragment_start    = s->fragment_start[plane];
+        int do_await          = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
 
         if (!s->flipped_image) stride = -stride;
         if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
             continue;
 
-
-        if(FFABS(stride) > 2048)
-            return; //various tables are fixed size
-
         /* for each superblock row in the slice (both of them)... */
         for (; sb_y < slice_height; sb_y++) {
 
@@ -1390,8 +1496,9 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                 for (j = 0; j < 16; j++) {
                     x = 4*sb_x + hilbert_offset[j][0];
                     y = 4*sb_y + hilbert_offset[j][1];
+                    fragment = y*fragment_width + x;
 
-                    i = fragment_start + y*fragment_width + x;
+                    i = fragment_start + fragment;
 
                     // bounds check
                     if (x >= fragment_width || y >= fragment_height)
@@ -1399,6 +1506,9 @@ static void render_slice(Vp3DecodeContext *s, int slice)
 
                 first_pixel = 8*y*stride + 8*x;
 
+                if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
+                    await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
+
                 /* transform if this block was coded */
                 if (s->all_fragments[i].coding_method != MODE_COPY) {
                     if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
@@ -1415,8 +1525,8 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                     if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
                         (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
                         int src_x, src_y;
-                        motion_x = motion_val[y*fragment_width + x][0];
-                        motion_y = motion_val[y*fragment_width + x][1];
+                        motion_x = motion_val[fragment][0];
+                        motion_y = motion_val[fragment][1];
 
                         src_x= (motion_x>>1) + 8*x;
                         src_y= (motion_y>>1) + 8*y;
@@ -1429,10 +1539,9 @@ static void render_slice(Vp3DecodeContext *s, int slice)
 
                         if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
                             uint8_t *temp= s->edge_emu_buffer;
-                            if(stride<0) temp -= 9*stride;
-                            else temp += 9*stride;
+                            if(stride<0) temp -= 8*stride;
 
-                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
+                            s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
                             motion_source= temp;
                         }
                     }
@@ -1464,7 +1573,10 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                     /* invert DCT and place (or add) in final output */
 
                     if (s->all_fragments[i].coding_method == MODE_INTRA) {
-                        vp3_dequant(s, s->all_fragments + i, plane, 0, block);
+                        int index;
+                        index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
+                        if (index > 63)
+                            continue;
                         if(s->avctx->idct_algo!=FF_IDCT_VP3)
                             block[0] += 128<<3;
                         s->dsp.idct_put(
@@ -1472,7 +1584,10 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                             stride,
                             block);
                     } else {
-                        if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
+                        int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
+                        if (index > 63)
+                            continue;
+                        if (index > 0) {
                         s->dsp.idct_add(
                             output_plane + first_pixel,
                             stride,
@@ -1510,9 +1625,38 @@ static void render_slice(Vp3DecodeContext *s, int slice)
     vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
 }
 
-/*
- * This is the ffmpeg/libavcodec API init function.
- */
+/// Allocate tables for per-frame data in Vp3DecodeContext
+static av_cold int allocate_tables(AVCodecContext *avctx)
+{
+    Vp3DecodeContext *s = avctx->priv_data;
+    int y_fragment_count, c_fragment_count;
+
+    y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
+    c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
+
+    s->superblock_coding = av_malloc(s->superblock_count);
+    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
+    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
+    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
+    s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
+    s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
+
+    /* work out the block mapping tables */
+    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
+    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
+
+    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
+        !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
+        !s->motion_val[0] || !s->motion_val[1]) {
+        vp3_decode_end(avctx);
+        return -1;
+    }
+
+    init_block_mapping(s);
+
+    return 0;
+}
+
 static av_cold int vp3_decode_init(AVCodecContext *avctx)
 {
     Vp3DecodeContext *s = avctx->priv_data;
@@ -1534,7 +1678,7 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
     avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
     if(avctx->idct_algo==FF_IDCT_AUTO)
         avctx->idct_algo=FF_IDCT_VP3;
-    dsputil_init(&s->dsp, avctx);
+    ff_dsputil_init(&s->dsp, avctx);
 
     ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
 
@@ -1559,7 +1703,6 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
     s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
     s->u_superblock_start = s->y_superblock_count;
     s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
-    s->superblock_coding = av_malloc(s->superblock_count);
 
     s->macroblock_width = (s->width + 15) / 16;
     s->macroblock_height = (s->height + 15) / 16;
@@ -1577,18 +1720,6 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
     s->fragment_start[1] = y_fragment_count;
     s->fragment_start[2] = y_fragment_count + c_fragment_count;
 
-    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
-    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
-    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
-    s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
-    s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
-
-    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
-        !s->coded_fragment_list[0] || !s->motion_val[0] || !s->motion_val[1]) {
-        vp3_decode_end(avctx);
-        return -1;
-    }
-
     if (!s->theora_tables)
     {
         for (i = 0; i < 64; i++) {
@@ -1688,31 +1819,95 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
         &motion_vector_vlc_table[0][1], 2, 1,
         &motion_vector_vlc_table[0][0], 2, 1, 0);
 
-    /* work out the block mapping tables */
-    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
-    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
-    if (!s->superblock_fragments || !s->macroblock_coding) {
-        vp3_decode_end(avctx);
-        return -1;
-    }
-    init_block_mapping(s);
-
     for (i = 0; i < 3; i++) {
         s->current_frame.data[i] = NULL;
         s->last_frame.data[i] = NULL;
         s->golden_frame.data[i] = NULL;
     }
 
-    return 0;
+    return allocate_tables(avctx);
 
 vlc_fail:
     av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
     return -1;
 }
 
-/*
- * This is the ffmpeg/libavcodec API frame decode function.
- */
+/// Release and shuffle frames after decode finishes
+static void update_frames(AVCodecContext *avctx)
+{
+    Vp3DecodeContext *s = avctx->priv_data;
+
+    /* release the last frame, if it is allocated and if it is not the
+     * golden frame */
+    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
+        ff_thread_release_buffer(avctx, &s->last_frame);
+
+    /* shuffle frames (last = current) */
+    s->last_frame= s->current_frame;
+
+    if (s->keyframe) {
+        if (s->golden_frame.data[0])
+            ff_thread_release_buffer(avctx, &s->golden_frame);
+        s->golden_frame = s->current_frame;
+        s->last_frame.type = FF_BUFFER_TYPE_COPY;
+    }
+
+    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
+}
+
+static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
+{
+    Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
+    int qps_changed = 0, i, err;
+
+#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
+
+    if (!s1->current_frame.data[0]
+        ||s->width != s1->width
+        ||s->height!= s1->height) {
+        if (s != s1)
+            copy_fields(s, s1, golden_frame, current_frame);
+        return -1;
+    }
+
+    if (s != s1) {
+        // init tables if the first frame hasn't been decoded
+        if (!s->current_frame.data[0]) {
+            int y_fragment_count, c_fragment_count;
+            s->avctx = dst;
+            err = allocate_tables(dst);
+            if (err)
+                return err;
+            y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
+            c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
+            memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
+            memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
+        }
+
+        // copy previous frame data
+        copy_fields(s, s1, golden_frame, dsp);
+
+        // copy qscale data if necessary
+        for (i = 0; i < 3; i++) {
+            if (s->qps[i] != s1->qps[1]) {
+                qps_changed = 1;
+                memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
+            }
+        }
+
+        if (s->qps[0] != s1->qps[0])
+            memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
+
+        if (qps_changed)
+            copy_fields(s, s1, qps, superblock_count);
+#undef copy_fields
+    }
+
+    update_frames(dst);
+
+    return 0;
+}
+
 static int vp3_decode_frame(AVCodecContext *avctx,
                             void *data, int *data_size,
                             AVPacket *avpkt)
@@ -1764,12 +1959,15 @@ static int vp3_decode_frame(AVCodecContext *avctx,
         return buf_size;
 
     s->current_frame.reference = 3;
-    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
-    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
+    s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
+    if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
         goto error;
     }
 
+    if (!s->edge_emu_buffer)
+        s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
+
     if (s->keyframe) {
         if (!s->theora)
         {
@@ -1793,20 +1991,19 @@ static int vp3_decode_frame(AVCodecContext *avctx,
             av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
 
             s->golden_frame.reference = 3;
-            s->golden_frame.pict_type = FF_I_TYPE;
-            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
+            s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
+            if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
                 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
                 goto error;
             }
             s->last_frame = s->golden_frame;
             s->last_frame.type = FF_BUFFER_TYPE_COPY;
+            ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
         }
     }
 
-    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
-    s->current_frame.qstride= 0;
-
     memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
+    ff_thread_finish_setup(avctx);
 
     if (unpack_superblocks(s, &gb)){
         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
@@ -1851,70 +2048,18 @@ static int vp3_decode_frame(AVCodecContext *avctx,
     *data_size=sizeof(AVFrame);
     *(AVFrame*)data= s->current_frame;
 
-    /* release the last frame, if it is allocated and if it is not the
-     * golden frame */
-    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
-        avctx->release_buffer(avctx, &s->last_frame);
-
-    /* shuffle frames (last = current) */
-    s->last_frame= s->current_frame;
-
-    if (s->keyframe) {
-        if (s->golden_frame.data[0])
-            avctx->release_buffer(avctx, &s->golden_frame);
-        s->golden_frame = s->current_frame;
-        s->last_frame.type = FF_BUFFER_TYPE_COPY;
-    }
-
-    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
+    if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
+        update_frames(avctx);
 
     return buf_size;
 
 error:
-    if (s->current_frame.data[0])
-        avctx->release_buffer(avctx, &s->current_frame);
-    return -1;
-}
-
-/*
- * This is the ffmpeg/libavcodec API module cleanup function.
- */
-static av_cold int vp3_decode_end(AVCodecContext *avctx)
-{
-    Vp3DecodeContext *s = avctx->priv_data;
-    int i;
-
-    av_free(s->superblock_coding);
-    av_free(s->all_fragments);
-    av_free(s->coded_fragment_list[0]);
-    av_free(s->dct_tokens_base);
-    av_free(s->superblock_fragments);
-    av_free(s->macroblock_coding);
-    av_free(s->motion_val[0]);
-    av_free(s->motion_val[1]);
-
-    for (i = 0; i < 16; i++) {
-        free_vlc(&s->dc_vlc[i]);
-        free_vlc(&s->ac_vlc_1[i]);
-        free_vlc(&s->ac_vlc_2[i]);
-        free_vlc(&s->ac_vlc_3[i]);
-        free_vlc(&s->ac_vlc_4[i]);
-    }
-
-    free_vlc(&s->superblock_run_length_vlc);
-    free_vlc(&s->fragment_run_length_vlc);
-    free_vlc(&s->mode_code_vlc);
-    free_vlc(&s->motion_vector_vlc);
+    ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
 
-    /* release all frames */
-    if (s->golden_frame.data[0])
-        avctx->release_buffer(avctx, &s->golden_frame);
-    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
-        avctx->release_buffer(avctx, &s->last_frame);
-    /* no need to release the current_frame since it will always be pointing
-     * to the same frame as either the golden or last frame */
+    if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
+        avctx->release_buffer(avctx, &s->current_frame);
 
-    return 0;
+    return -1;
 }
 
 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
@@ -1951,6 +2096,23 @@ static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
     return 0;
 }
 
+static int vp3_init_thread_copy(AVCodecContext *avctx)
+{
+    Vp3DecodeContext *s = avctx->priv_data;
+
+    s->superblock_coding      = NULL;
+    s->all_fragments          = NULL;
+    s->coded_fragment_list[0] = NULL;
+    s->dct_tokens_base        = NULL;
+    s->superblock_fragments   = NULL;
+    s->macroblock_coding      = NULL;
+    s->motion_val[0]          = NULL;
+    s->motion_val[1]          = NULL;
+    s->edge_emu_buffer        = NULL;
+
+    return 0;
+}
+
 #if CONFIG_THEORA_DECODER
 static const enum PixelFormat theora_pix_fmts[4] = {
     PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
@@ -2166,7 +2328,7 @@ static av_cold int theora_decode_init(AVCodecContext *avctx)
         return -1;
     }
 
-    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
+    if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
                               42, header_start, header_len) < 0) {
         av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
         return -1;
@@ -2212,31 +2374,35 @@ static av_cold int theora_decode_init(AVCodecContext *avctx)
     return vp3_decode_init(avctx);
 }
 
-AVCodec theora_decoder = {
-    "theora",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_THEORA,
-    sizeof(Vp3DecodeContext),
-    theora_decode_init,
-    NULL,
-    vp3_decode_end,
-    vp3_decode_frame,
-    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
-    NULL,
-    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
+AVCodec ff_theora_decoder = {
+    .name                  = "theora",
+    .type                  = AVMEDIA_TYPE_VIDEO,
+    .id                    = CODEC_ID_THEORA,
+    .priv_data_size        = sizeof(Vp3DecodeContext),
+    .init                  = theora_decode_init,
+    .close                 = vp3_decode_end,
+    .decode                = vp3_decode_frame,
+    .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
+                             CODEC_CAP_FRAME_THREADS,
+    .flush                 = vp3_decode_flush,
+    .long_name             = NULL_IF_CONFIG_SMALL("Theora"),
+    .init_thread_copy      = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
+    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
 };
 #endif
 
-AVCodec vp3_decoder = {
-    "vp3",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_VP3,
-    sizeof(Vp3DecodeContext),
-    vp3_decode_init,
-    NULL,
-    vp3_decode_end,
-    vp3_decode_frame,
-    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
-    NULL,
-    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
+AVCodec ff_vp3_decoder = {
+    .name                  = "vp3",
+    .type                  = AVMEDIA_TYPE_VIDEO,
+    .id                    = CODEC_ID_VP3,
+    .priv_data_size        = sizeof(Vp3DecodeContext),
+    .init                  = vp3_decode_init,
+    .close                 = vp3_decode_end,
+    .decode                = vp3_decode_frame,
+    .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
+                             CODEC_CAP_FRAME_THREADS,
+    .flush                 = vp3_decode_flush,
+    .long_name             = NULL_IF_CONFIG_SMALL("On2 VP3"),
+    .init_thread_copy      = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
+    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
 };