* Copyright (C) 2010 Ronald S. Bultje
* Copyright (C) 2010 Jason Garrett-Glaser
*
- * 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 "libavutil/imgutils.h"
#include "avcodec.h"
-#include "vp56.h"
+#include "vp8.h"
#include "vp8data.h"
-#include "vp8dsp.h"
-#include "h264pred.h"
#include "rectangle.h"
+#include "thread.h"
#if ARCH_ARM
# include "arm/vp8.h"
#endif
-typedef struct {
- uint8_t filter_level;
- uint8_t inner_limit;
- uint8_t inner_filter;
-} VP8FilterStrength;
-
-typedef struct {
- uint8_t skip;
- // todo: make it possible to check for at least (i4x4 or split_mv)
- // in one op. are others needed?
- uint8_t mode;
- uint8_t ref_frame;
- uint8_t partitioning;
- VP56mv mv;
- VP56mv bmv[16];
-} VP8Macroblock;
-
-typedef struct {
- AVCodecContext *avctx;
- DSPContext dsp;
- VP8DSPContext vp8dsp;
- H264PredContext hpc;
- vp8_mc_func put_pixels_tab[3][3][3];
- AVFrame frames[4];
- AVFrame *framep[4];
- uint8_t *edge_emu_buffer;
- VP56RangeCoder c; ///< header context, includes mb modes and motion vectors
- int profile;
-
- int mb_width; /* number of horizontal MB */
- int mb_height; /* number of vertical MB */
- int linesize;
- int uvlinesize;
-
- int keyframe;
- int invisible;
- int update_last; ///< update VP56_FRAME_PREVIOUS with the current one
- int update_golden; ///< VP56_FRAME_NONE if not updated, or which frame to copy if so
- int update_altref;
- int deblock_filter;
-
- /**
- * If this flag is not set, all the probability updates
- * are discarded after this frame is decoded.
- */
- int update_probabilities;
-
- /**
- * All coefficients are contained in separate arith coding contexts.
- * There can be 1, 2, 4, or 8 of these after the header context.
- */
- int num_coeff_partitions;
- VP56RangeCoder coeff_partition[8];
-
- VP8Macroblock *macroblocks;
- VP8Macroblock *macroblocks_base;
- VP8FilterStrength *filter_strength;
-
- uint8_t *intra4x4_pred_mode_top;
- uint8_t intra4x4_pred_mode_left[4];
- uint8_t *segmentation_map;
-
- /**
- * Cache of the top row needed for intra prediction
- * 16 for luma, 8 for each chroma plane
- */
- uint8_t (*top_border)[16+8+8];
-
- /**
- * For coeff decode, we need to know whether the above block had non-zero
- * coefficients. This means for each macroblock, we need data for 4 luma
- * blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9
- * per macroblock. We keep the last row in top_nnz.
- */
- uint8_t (*top_nnz)[9];
- DECLARE_ALIGNED(8, uint8_t, left_nnz)[9];
-
- /**
- * This is the index plus one of the last non-zero coeff
- * for each of the blocks in the current macroblock.
- * So, 0 -> no coeffs
- * 1 -> dc-only (special transform)
- * 2+-> full transform
- */
- DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4];
- DECLARE_ALIGNED(16, DCTELEM, block)[6][4][16];
- DECLARE_ALIGNED(16, DCTELEM, block_dc)[16];
- uint8_t intra4x4_pred_mode_mb[16];
-
- int chroma_pred_mode; ///< 8x8c pred mode of the current macroblock
- int segment; ///< segment of the current macroblock
- VP56mv mv_min;
- VP56mv mv_max;
-
- int mbskip_enabled;
- int sign_bias[4]; ///< one state [0, 1] per ref frame type
- int ref_count[3];
-
- /**
- * Base parameters for segmentation, i.e. per-macroblock parameters.
- * These must be kept unchanged even if segmentation is not used for
- * a frame, since the values persist between interframes.
- */
- struct {
- int enabled;
- int absolute_vals;
- int update_map;
- int8_t base_quant[4];
- int8_t filter_level[4]; ///< base loop filter level
- } segmentation;
-
- /**
- * Macroblocks can have one of 4 different quants in a frame when
- * segmentation is enabled.
- * If segmentation is disabled, only the first segment's values are used.
- */
- struct {
- // [0] - DC qmul [1] - AC qmul
- int16_t luma_qmul[2];
- int16_t luma_dc_qmul[2]; ///< luma dc-only block quant
- int16_t chroma_qmul[2];
- } qmat[4];
-
- struct {
- int simple;
- int level;
- int sharpness;
- } filter;
-
- struct {
- int enabled; ///< whether each mb can have a different strength based on mode/ref
-
- /**
- * filter strength adjustment for the following macroblock modes:
- * [0-3] - i16x16 (always zero)
- * [4] - i4x4
- * [5] - zero mv
- * [6] - inter modes except for zero or split mv
- * [7] - split mv
- * i16x16 modes never have any adjustment
- */
- int8_t mode[VP8_MVMODE_SPLIT+1];
-
- /**
- * filter strength adjustment for macroblocks that reference:
- * [0] - intra / VP56_FRAME_CURRENT
- * [1] - VP56_FRAME_PREVIOUS
- * [2] - VP56_FRAME_GOLDEN
- * [3] - altref / VP56_FRAME_GOLDEN2
- */
- int8_t ref[4];
- } lf_delta;
-
- /**
- * These are all of the updatable probabilities for binary decisions.
- * They are only implictly reset on keyframes, making it quite likely
- * for an interframe to desync if a prior frame's header was corrupt
- * or missing outright!
- */
- struct {
- uint8_t segmentid[3];
- uint8_t mbskip;
- uint8_t intra;
- uint8_t last;
- uint8_t golden;
- uint8_t pred16x16[4];
- uint8_t pred8x8c[3];
- /* Padded to allow overreads */
- uint8_t token[4][17][3][NUM_DCT_TOKENS-1];
- uint8_t mvc[2][19];
- } prob[2];
-} VP8Context;
-
static void vp8_decode_flush(AVCodecContext *avctx)
{
VP8Context *s = avctx->priv_data;
int i;
- for (i = 0; i < 4; i++)
- if (s->frames[i].data[0])
- avctx->release_buffer(avctx, &s->frames[i]);
+ if (!avctx->is_copy) {
+ for (i = 0; i < 5; i++)
+ if (s->frames[i].data[0])
+ ff_thread_release_buffer(avctx, &s->frames[i]);
+ }
memset(s->framep, 0, sizeof(s->framep));
av_freep(&s->macroblocks_base);
static int update_dimensions(VP8Context *s, int width, int height)
{
- if (av_image_check_size(width, height, 0, s->avctx))
- return AVERROR_INVALIDDATA;
+ if (width != s->avctx->width ||
+ height != s->avctx->height) {
+ if (av_image_check_size(width, height, 0, s->avctx))
+ return AVERROR_INVALIDDATA;
- vp8_decode_flush(s->avctx);
+ vp8_decode_flush(s->avctx);
- avcodec_set_dimensions(s->avctx, width, height);
+ avcodec_set_dimensions(s->avctx, width, height);
+ }
s->mb_width = (s->avctx->coded_width +15) / 16;
s->mb_height = (s->avctx->coded_height+15) / 16;
} else
base_qi = yac_qi;
- s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip(base_qi + ydc_delta , 0, 127)];
- s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip(base_qi , 0, 127)];
- s->qmat[i].luma_dc_qmul[0] = 2 * vp8_dc_qlookup[av_clip(base_qi + y2dc_delta, 0, 127)];
- s->qmat[i].luma_dc_qmul[1] = 155 * vp8_ac_qlookup[av_clip(base_qi + y2ac_delta, 0, 127)] / 100;
- s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip(base_qi + uvdc_delta, 0, 127)];
- s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip(base_qi + uvac_delta, 0, 127)];
+ s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + ydc_delta , 7)];
+ s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi , 7)];
+ s->qmat[i].luma_dc_qmul[0] = 2 * vp8_dc_qlookup[av_clip_uintp2(base_qi + y2dc_delta, 7)];
+ s->qmat[i].luma_dc_qmul[1] = 155 * vp8_ac_qlookup[av_clip_uintp2(base_qi + y2ac_delta, 7)] / 100;
+ s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + uvdc_delta, 7)];
+ s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + uvac_delta, 7)];
s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8);
s->qmat[i].chroma_qmul[0] = FFMIN(s->qmat[i].chroma_qmul[0], 132);
mb - 1 /* left */,
mb + 1 /* top-left */ };
enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
- enum { EDGE_TOP, EDGE_LEFT, EDGE_TOPLEFT };
+ enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
int idx = CNT_ZERO;
int cur_sign_bias = s->sign_bias[mb->ref_frame];
- int *sign_bias = s->sign_bias;
+ int8_t *sign_bias = s->sign_bias;
VP56mv near_mv[4];
uint8_t cnt[4] = { 0 };
VP56RangeCoder *c = &s->c;
AV_ZERO32(&near_mv[0]);
AV_ZERO32(&near_mv[1]);
- AV_ZERO32(&near_mv[2]);
/* Process MB on top, left and top-left */
#define MV_EDGE_CHECK(n)\
mb->mode = VP8_MVMODE_MV;
/* If we have three distinct MVs, merge first and last if they're the same */
- if (cnt[CNT_SPLITMV] && AV_RN32A(&near_mv[1+EDGE_TOP]) == AV_RN32A(&near_mv[1+EDGE_TOPLEFT]))
+ if (cnt[CNT_SPLITMV] && AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT]))
cnt[CNT_NEAREST] += 1;
/* Swap near and nearest if necessary */
/* Choose the best mv out of 0,0 and the nearest mv */
clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
- cnt[CNT_SPLITMV] = ((mb_edge[EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) +
- (mb_edge[EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 +
- (mb_edge[EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
+ cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) +
+ (mb_edge[VP8_EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 +
+ (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
mb->mode = VP8_MVMODE_SPLIT;
}
static av_always_inline
-void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment)
+void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment, uint8_t *ref)
{
VP56RangeCoder *c = &s->c;
if (s->segmentation.update_map)
*segment = vp8_rac_get_tree(c, vp8_segmentid_tree, s->prob->segmentid);
+ else
+ *segment = ref ? *ref : *segment;
s->segment = *segment;
mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0;
* otherwise, the index of the last coeff decoded plus one
*/
static int decode_block_coeffs_internal(VP56RangeCoder *c, DCTELEM block[16],
- uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+ uint8_t probs[16][3][NUM_DCT_TOKENS-1],
int i, uint8_t *token_prob, int16_t qmul[2])
{
goto skip_eob;
static av_always_inline
int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
- uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+ uint8_t probs[16][3][NUM_DCT_TOKENS-1],
int i, int zero_nhood, int16_t qmul[2])
{
uint8_t *token_prob = probs[i][zero_nhood];
* @param s VP8 decoding context
* @param luma 1 for luma (Y) planes, 0 for chroma (Cb/Cr) planes
* @param dst target buffer for block data at block position
- * @param src reference picture buffer at origin (0, 0)
+ * @param ref reference picture buffer at origin (0, 0)
* @param mv motion vector (relative to block position) to get pixel data from
* @param x_off horizontal position of block from origin (0, 0)
* @param y_off vertical position of block from origin (0, 0)
* @param mc_func motion compensation function pointers (bilinear or sixtap MC)
*/
static av_always_inline
-void vp8_mc_luma(VP8Context *s, uint8_t *dst, uint8_t *src, const VP56mv *mv,
+void vp8_mc_luma(VP8Context *s, uint8_t *dst, AVFrame *ref, const VP56mv *mv,
int x_off, int y_off, int block_w, int block_h,
int width, int height, int linesize,
vp8_mc_func mc_func[3][3])
{
+ uint8_t *src = ref->data[0];
+
if (AV_RN32A(mv)) {
int mx = (mv->x << 1)&7, mx_idx = subpel_idx[0][mx];
y_off += mv->y >> 2;
// edge emulation
+ ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4, 0);
src += y_off * linesize + x_off;
if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
src = s->edge_emu_buffer + mx_idx + linesize * my_idx;
}
mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my);
- } else
+ } else {
+ ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0);
mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, linesize, block_h, 0, 0);
+ }
}
static av_always_inline
-void vp8_mc_chroma(VP8Context *s, uint8_t *dst1, uint8_t *dst2, uint8_t *src1,
- uint8_t *src2, const VP56mv *mv, int x_off, int y_off,
+void vp8_mc_chroma(VP8Context *s, uint8_t *dst1, uint8_t *dst2, AVFrame *ref,
+ const VP56mv *mv, int x_off, int y_off,
int block_w, int block_h, int width, int height, int linesize,
vp8_mc_func mc_func[3][3])
{
+ uint8_t *src1 = ref->data[1], *src2 = ref->data[2];
+
if (AV_RN32A(mv)) {
int mx = mv->x&7, mx_idx = subpel_idx[0][mx];
int my = mv->y&7, my_idx = subpel_idx[0][my];
// edge emulation
src1 += y_off * linesize + x_off;
src2 += y_off * linesize + x_off;
+ ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0);
if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize,
mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
}
} else {
+ ff_thread_await_progress(ref, (3 + y_off + block_h) >> 3, 0);
mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0);
mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0);
}
/* Y */
vp8_mc_luma(s, dst[0] + by_off * s->linesize + bx_off,
- ref_frame->data[0], mv, x_off + bx_off, y_off + by_off,
+ ref_frame, mv, x_off + bx_off, y_off + by_off,
block_w, block_h, width, height, s->linesize,
s->put_pixels_tab[block_w == 8]);
width >>= 1; height >>= 1;
block_w >>= 1; block_h >>= 1;
vp8_mc_chroma(s, dst[1] + by_off * s->uvlinesize + bx_off,
- dst[2] + by_off * s->uvlinesize + bx_off, ref_frame->data[1],
- ref_frame->data[2], &uvmv, x_off + bx_off, y_off + by_off,
+ dst[2] + by_off * s->uvlinesize + bx_off, ref_frame,
+ &uvmv, x_off + bx_off, y_off + by_off,
block_w, block_h, width, height, s->uvlinesize,
s->put_pixels_tab[1 + (block_w == 4)]);
}
int my = (mb->mv.y>>2) + y_off;
uint8_t **src= s->framep[ref]->data;
int off= mx + (my + (mb_x&3)*4)*s->linesize + 64;
+ /* For threading, a ff_thread_await_progress here might be useful, but
+ * it actually slows down the decoder. Since a bad prefetch doesn't
+ * generate bad decoder output, we don't run it here. */
s->dsp.prefetch(src[0]+off, s->linesize, 4);
off= (mx>>1) + ((my>>1) + (mb_x&7))*s->uvlinesize + 64;
s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
for (y = 0; y < 4; y++) {
for (x = 0; x < 4; x++) {
vp8_mc_luma(s, dst[0] + 4*y*s->linesize + x*4,
- ref->data[0], &bmv[4*y + x],
+ ref, &bmv[4*y + x],
4*x + x_off, 4*y + y_off, 4, 4,
width, height, s->linesize,
s->put_pixels_tab[2]);
uvmv.y &= ~7;
}
vp8_mc_chroma(s, dst[1] + 4*y*s->uvlinesize + x*4,
- dst[2] + 4*y*s->uvlinesize + x*4,
- ref->data[1], ref->data[2], &uvmv,
+ dst[2] + 4*y*s->uvlinesize + x*4, ref, &uvmv,
4*x + x_off, 4*y + y_off, 4, 4,
width, height, s->uvlinesize,
s->put_pixels_tab[2]);
s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
nnz4 >>= 8;
if (!nnz4)
- break;
+ goto chroma_idct_end;
}
ch_dst += 4*s->uvlinesize;
}
s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, s->block[4+ch], s->uvlinesize);
}
}
+chroma_idct_end: ;
}
}
filter_level += s->lf_delta.mode[mb->mode];
}
-/* Like av_clip for inputs 0 and max, where max is equal to (2^n-1) */
-#define POW2CLIP(x,max) (((x) & ~max) ? (-(x))>>31 & max : (x));
- filter_level = POW2CLIP(filter_level, 63);
+ filter_level = av_clip_uintp2(filter_level, 6);
interior_limit = filter_level;
if (s->filter.sharpness) {
}
}
-static void filter_mb_row(VP8Context *s, int mb_y)
+static void filter_mb_row(VP8Context *s, AVFrame *curframe, int mb_y)
{
VP8FilterStrength *f = s->filter_strength;
uint8_t *dst[3] = {
- s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize,
- s->framep[VP56_FRAME_CURRENT]->data[1] + 8*mb_y*s->uvlinesize,
- s->framep[VP56_FRAME_CURRENT]->data[2] + 8*mb_y*s->uvlinesize
+ curframe->data[0] + 16*mb_y*s->linesize,
+ curframe->data[1] + 8*mb_y*s->uvlinesize,
+ curframe->data[2] + 8*mb_y*s->uvlinesize
};
int mb_x;
}
}
-static void filter_mb_row_simple(VP8Context *s, int mb_y)
+static void filter_mb_row_simple(VP8Context *s, AVFrame *curframe, int mb_y)
{
VP8FilterStrength *f = s->filter_strength;
- uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
+ uint8_t *dst = curframe->data[0] + 16*mb_y*s->linesize;
int mb_x;
for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
VP8Context *s = avctx->priv_data;
int ret, mb_x, mb_y, i, y, referenced;
enum AVDiscard skip_thresh;
- AVFrame *av_uninit(curframe);
+ AVFrame *av_uninit(curframe), *prev_frame = s->framep[VP56_FRAME_CURRENT];
if ((ret = decode_frame_header(s, avpkt->data, avpkt->size)) < 0)
return ret;
}
s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;
- for (i = 0; i < 4; i++)
- if (&s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
+ // release no longer referenced frames
+ for (i = 0; i < 5; i++)
+ if (s->frames[i].data[0] &&
+ &s->frames[i] != prev_frame &&
+ &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
+ &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] &&
+ &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2])
+ ff_thread_release_buffer(avctx, &s->frames[i]);
+
+ // find a free buffer
+ for (i = 0; i < 5; i++)
+ if (&s->frames[i] != prev_frame &&
+ &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN] &&
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) {
curframe = s->framep[VP56_FRAME_CURRENT] = &s->frames[i];
break;
}
+ if (i == 5) {
+ av_log(avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
+ abort();
+ }
if (curframe->data[0])
- avctx->release_buffer(avctx, curframe);
+ ff_thread_release_buffer(avctx, curframe);
curframe->key_frame = s->keyframe;
- curframe->pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
+ curframe->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
curframe->reference = referenced ? 3 : 0;
- if ((ret = avctx->get_buffer(avctx, curframe))) {
+ curframe->ref_index[0] = s->segmentation_map;
+ if ((ret = ff_thread_get_buffer(avctx, curframe))) {
av_log(avctx, AV_LOG_ERROR, "get_buffer() failed!\n");
return ret;
}
+ // check if golden and altref are swapped
+ if (s->update_altref != VP56_FRAME_NONE) {
+ s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[s->update_altref];
+ } else {
+ s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[VP56_FRAME_GOLDEN2];
+ }
+ if (s->update_golden != VP56_FRAME_NONE) {
+ s->next_framep[VP56_FRAME_GOLDEN] = s->framep[s->update_golden];
+ } else {
+ s->next_framep[VP56_FRAME_GOLDEN] = s->framep[VP56_FRAME_GOLDEN];
+ }
+ if (s->update_last) {
+ s->next_framep[VP56_FRAME_PREVIOUS] = curframe;
+ } else {
+ s->next_framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_PREVIOUS];
+ }
+ s->next_framep[VP56_FRAME_CURRENT] = curframe;
+
+ ff_thread_finish_setup(avctx);
+
// Given that arithmetic probabilities are updated every frame, it's quite likely
// that the values we have on a random interframe are complete junk if we didn't
// start decode on a keyframe. So just don't display anything rather than junk.
if (s->keyframe)
memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4);
- #define MARGIN (16 << 2)
+#define MARGIN (16 << 2)
s->mv_min.y = -MARGIN;
s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
s->mv_min.x = -MARGIN;
s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
+ if (prev_frame && s->segmentation.enabled && !s->segmentation.update_map)
+ ff_thread_await_progress(prev_frame, mb_y, 0);
for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
/* Prefetch the current frame, 4 MBs ahead */
s->dsp.prefetch(dst[0] + (mb_x&3)*4*s->linesize + 64, s->linesize, 4);
s->dsp.prefetch(dst[1] + (mb_x&7)*s->uvlinesize + 64, dst[2] - dst[1], 2);
- decode_mb_mode(s, mb, mb_x, mb_y, s->segmentation_map + mb_xy);
+ decode_mb_mode(s, mb, mb_x, mb_y, s->segmentation_map + mb_xy,
+ prev_frame ? prev_frame->ref_index[0] + mb_xy : NULL);
prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
}
if (s->deblock_filter) {
if (s->filter.simple)
- filter_mb_row_simple(s, mb_y);
+ filter_mb_row_simple(s, curframe, mb_y);
else
- filter_mb_row(s, mb_y);
+ filter_mb_row(s, curframe, mb_y);
}
s->mv_min.y -= 64;
s->mv_max.y -= 64;
+
+ ff_thread_report_progress(curframe, mb_y, 0);
}
+ ff_thread_report_progress(curframe, INT_MAX, 0);
skip_decode:
// if future frames don't use the updated probabilities,
// reset them to the values we saved
if (!s->update_probabilities)
s->prob[0] = s->prob[1];
- // check if golden and altref are swapped
- if (s->update_altref == VP56_FRAME_GOLDEN &&
- s->update_golden == VP56_FRAME_GOLDEN2)
- FFSWAP(AVFrame *, s->framep[VP56_FRAME_GOLDEN], s->framep[VP56_FRAME_GOLDEN2]);
- else {
- if (s->update_altref != VP56_FRAME_NONE)
- s->framep[VP56_FRAME_GOLDEN2] = s->framep[s->update_altref];
-
- if (s->update_golden != VP56_FRAME_NONE)
- s->framep[VP56_FRAME_GOLDEN] = s->framep[s->update_golden];
- }
-
- if (s->update_last) // move cur->prev
- s->framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_CURRENT];
-
- // release no longer referenced frames
- for (i = 0; i < 4; i++)
- if (s->frames[i].data[0] &&
- &s->frames[i] != s->framep[VP56_FRAME_CURRENT] &&
- &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
- &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] &&
- &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2])
- avctx->release_buffer(avctx, &s->frames[i]);
+ memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
if (!s->invisible) {
- *(AVFrame*)data = *s->framep[VP56_FRAME_CURRENT];
+ *(AVFrame*)data = *curframe;
*data_size = sizeof(AVFrame);
}
avctx->pix_fmt = PIX_FMT_YUV420P;
dsputil_init(&s->dsp, avctx);
- ff_h264_pred_init(&s->hpc, CODEC_ID_VP8);
+ ff_h264_pred_init(&s->hpc, CODEC_ID_VP8, 8);
ff_vp8dsp_init(&s->vp8dsp);
return 0;
return 0;
}
+static av_cold int vp8_decode_init_thread_copy(AVCodecContext *avctx)
+{
+ VP8Context *s = avctx->priv_data;
+
+ s->avctx = avctx;
+
+ return 0;
+}
+
+#define REBASE(pic) \
+ pic ? pic - &s_src->frames[0] + &s->frames[0] : NULL
+
+static int vp8_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
+{
+ VP8Context *s = dst->priv_data, *s_src = src->priv_data;
+
+ s->prob[0] = s_src->prob[!s_src->update_probabilities];
+ s->segmentation = s_src->segmentation;
+ s->lf_delta = s_src->lf_delta;
+ memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias));
+
+ memcpy(&s->frames, &s_src->frames, sizeof(s->frames));
+ s->framep[0] = REBASE(s_src->next_framep[0]);
+ s->framep[1] = REBASE(s_src->next_framep[1]);
+ s->framep[2] = REBASE(s_src->next_framep[2]);
+ s->framep[3] = REBASE(s_src->next_framep[3]);
+
+ return 0;
+}
+
AVCodec ff_vp8_decoder = {
"vp8",
AVMEDIA_TYPE_VIDEO,
NULL,
vp8_decode_free,
vp8_decode_frame,
- CODEC_CAP_DR1,
+ CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
.flush = vp8_decode_flush,
.long_name = NULL_IF_CONFIG_SMALL("On2 VP8"),
+ .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy),
+ .update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context),
};