-/**
+/*
* VP8 compatible video decoder
*
* Copyright (C) 2010 David Conrad
* Copyright (C) 2010 Ronald S. Bultje
* Copyright (C) 2010 Jason Garrett-Glaser
+ * Copyright (C) 2012 Daniel Kang
*
- * 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 "internal.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
-
- 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 free_buffers(VP8Context *s)
+{
+ int i;
+ if (s->thread_data)
+ for (i = 0; i < MAX_THREADS; i++) {
+ av_freep(&s->thread_data[i].filter_strength);
+ av_freep(&s->thread_data[i].edge_emu_buffer);
+ }
+ av_freep(&s->thread_data);
+ av_freep(&s->macroblocks_base);
+ av_freep(&s->intra4x4_pred_mode_top);
+ av_freep(&s->top_nnz);
+ av_freep(&s->top_border);
-static void vp8_decode_flush(AVCodecContext *avctx)
+ s->macroblocks = NULL;
+}
+
+static int vp8_alloc_frame(VP8Context *s, AVFrame *f)
+{
+ int ret;
+ if ((ret = ff_thread_get_buffer(s->avctx, f)) < 0)
+ return ret;
+ if (s->num_maps_to_be_freed && !s->maps_are_invalid) {
+ f->ref_index[0] = s->segmentation_maps[--s->num_maps_to_be_freed];
+ } else if (!(f->ref_index[0] = av_mallocz(s->mb_width * s->mb_height))) {
+ ff_thread_release_buffer(s->avctx, f);
+ return AVERROR(ENOMEM);
+ }
+ return 0;
+}
+
+static void vp8_release_frame(VP8Context *s, AVFrame *f, int prefer_delayed_free, int can_direct_free)
+{
+ if (f->ref_index[0]) {
+ if (prefer_delayed_free) {
+ /* Upon a size change, we want to free the maps but other threads may still
+ * be using them, so queue them. Upon a seek, all threads are inactive so
+ * we want to cache one to prevent re-allocation in the next decoding
+ * iteration, but the rest we can free directly. */
+ int max_queued_maps = can_direct_free ? 1 : FF_ARRAY_ELEMS(s->segmentation_maps);
+ if (s->num_maps_to_be_freed < max_queued_maps) {
+ s->segmentation_maps[s->num_maps_to_be_freed++] = f->ref_index[0];
+ } else if (can_direct_free) /* vp8_decode_flush(), but our queue is full */ {
+ av_free(f->ref_index[0]);
+ } /* else: MEMLEAK (should never happen, but better that than crash) */
+ f->ref_index[0] = NULL;
+ } else /* vp8_decode_free() */ {
+ av_free(f->ref_index[0]);
+ }
+ }
+ ff_thread_release_buffer(s->avctx, f);
+}
+
+static void vp8_decode_flush_impl(AVCodecContext *avctx,
+ int prefer_delayed_free, int can_direct_free, int free_mem)
{
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->internal->is_copy) {
+ for (i = 0; i < 5; i++)
+ if (s->frames[i].data[0])
+ vp8_release_frame(s, &s->frames[i], prefer_delayed_free, can_direct_free);
+ }
memset(s->framep, 0, sizeof(s->framep));
- av_freep(&s->macroblocks_base);
- av_freep(&s->filter_strength);
- av_freep(&s->intra4x4_pred_mode_top);
- av_freep(&s->top_nnz);
- av_freep(&s->edge_emu_buffer);
- av_freep(&s->top_border);
- av_freep(&s->segmentation_map);
+ if (free_mem) {
+ free_buffers(s);
+ s->maps_are_invalid = 1;
+ }
+}
- s->macroblocks = NULL;
+static void vp8_decode_flush(AVCodecContext *avctx)
+{
+ vp8_decode_flush_impl(avctx, 1, 1, 0);
}
static int update_dimensions(VP8Context *s, int width, int height)
{
- if (av_image_check_size(width, height, 0, s->avctx))
- return AVERROR_INVALIDDATA;
+ AVCodecContext *avctx = s->avctx;
+ int i;
- vp8_decode_flush(s->avctx);
+ if (width != s->avctx->width ||
+ height != s->avctx->height) {
+ if (av_image_check_size(width, height, 0, s->avctx))
+ return AVERROR_INVALIDDATA;
- avcodec_set_dimensions(s->avctx, width, height);
+ vp8_decode_flush_impl(s->avctx, 1, 0, 1);
+
+ 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;
- s->macroblocks_base = av_mallocz((s->mb_width+s->mb_height*2+1)*sizeof(*s->macroblocks));
- s->filter_strength = av_mallocz(s->mb_width*sizeof(*s->filter_strength));
- s->intra4x4_pred_mode_top = av_mallocz(s->mb_width*4);
- s->top_nnz = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
- s->top_border = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));
- s->segmentation_map = av_mallocz(s->mb_width*s->mb_height);
+ s->mb_layout = (avctx->active_thread_type == FF_THREAD_SLICE) && (FFMIN(s->num_coeff_partitions, avctx->thread_count) > 1);
+ if (!s->mb_layout) { // Frame threading and one thread
+ s->macroblocks_base = av_mallocz((s->mb_width+s->mb_height*2+1)*sizeof(*s->macroblocks));
+ s->intra4x4_pred_mode_top = av_mallocz(s->mb_width*4);
+ }
+ else // Sliced threading
+ s->macroblocks_base = av_mallocz((s->mb_width+2)*(s->mb_height+2)*sizeof(*s->macroblocks));
+ s->top_nnz = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
+ s->top_border = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));
+ s->thread_data = av_mallocz(MAX_THREADS*sizeof(VP8ThreadData));
+
+ for (i = 0; i < MAX_THREADS; i++) {
+ s->thread_data[i].filter_strength = av_mallocz(s->mb_width*sizeof(*s->thread_data[0].filter_strength));
+#if HAVE_THREADS
+ pthread_mutex_init(&s->thread_data[i].lock, NULL);
+ pthread_cond_init(&s->thread_data[i].cond, NULL);
+#endif
+ }
- if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_top ||
- !s->top_nnz || !s->top_border || !s->segmentation_map)
+ if (!s->macroblocks_base || !s->top_nnz || !s->top_border ||
+ (!s->intra4x4_pred_mode_top && !s->mb_layout))
return AVERROR(ENOMEM);
s->macroblocks = s->macroblocks_base + 1;
VP56RangeCoder *c = &s->c;
int i;
- for (i = 0; i < 4; i++)
- s->lf_delta.ref[i] = vp8_rac_get_sint(c, 6);
+ for (i = 0; i < 4; i++) {
+ if (vp8_rac_get(c)) {
+ s->lf_delta.ref[i] = vp8_rac_get_uint(c, 6);
- for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++)
- s->lf_delta.mode[i] = vp8_rac_get_sint(c, 6);
+ if (vp8_rac_get(c))
+ s->lf_delta.ref[i] = -s->lf_delta.ref[i];
+ }
+ }
+
+ for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++) {
+ if (vp8_rac_get(c)) {
+ s->lf_delta.mode[i] = vp8_rac_get_uint(c, 6);
+
+ if (vp8_rac_get(c))
+ s->lf_delta.mode[i] = -s->lf_delta.mode[i];
+ }
+ }
}
static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
} 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)];
+ /* 101581>>16 is equivalent to 155/100 */
+ s->qmat[i].luma_dc_qmul[1] = (101581 * vp8_ac_qlookup[av_clip_uintp2(base_qi + y2ac_delta, 7)]) >> 16;
+ 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);
memcpy(s->prob->pred8x8c , vp8_pred8x8c_prob_inter , sizeof(s->prob->pred8x8c));
memcpy(s->prob->mvc , vp8_mv_default_prob , sizeof(s->prob->mvc));
memset(&s->segmentation, 0, sizeof(s->segmentation));
- }
-
- if (!s->macroblocks_base || /* first frame */
- width != s->avctx->width || height != s->avctx->height) {
- if ((ret = update_dimensions(s, width, height) < 0))
- return ret;
+ memset(&s->lf_delta, 0, sizeof(s->lf_delta));
}
ff_vp56_init_range_decoder(c, buf, header_size);
return AVERROR_INVALIDDATA;
}
+ if (!s->macroblocks_base || /* first frame */
+ width != s->avctx->width || height != s->avctx->height) {
+ if ((ret = update_dimensions(s, width, height)) < 0)
+ return ret;
+ }
+
get_quants(s);
if (!s->keyframe) {
return 0;
}
-static av_always_inline
-void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src, int mb_x, int mb_y)
+static av_always_inline void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src)
{
-#define MARGIN (16 << 2)
- dst->x = av_clip(src->x, -((mb_x << 6) + MARGIN),
- ((s->mb_width - 1 - mb_x) << 6) + MARGIN);
- dst->y = av_clip(src->y, -((mb_y << 6) + MARGIN),
- ((s->mb_height - 1 - mb_y) << 6) + MARGIN);
+ dst->x = av_clip(src->x, s->mv_min.x, s->mv_max.x);
+ dst->y = av_clip(src->y, s->mv_min.y, s->mv_max.y);
}
/**
* @returns the number of motion vectors parsed (2, 4 or 16)
*/
static av_always_inline
-int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
+int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb, int layout)
{
int part_idx;
int n, num;
- VP8Macroblock *top_mb = &mb[2];
+ VP8Macroblock *top_mb;
VP8Macroblock *left_mb = &mb[-1];
const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning],
- *mbsplits_top = vp8_mbsplits[top_mb->partitioning],
+ *mbsplits_top,
*mbsplits_cur, *firstidx;
- VP56mv *top_mv = top_mb->bmv;
+ VP56mv *top_mv;
VP56mv *left_mv = left_mb->bmv;
VP56mv *cur_mv = mb->bmv;
+ if (!layout) // layout is inlined, s->mb_layout is not
+ top_mb = &mb[2];
+ else
+ top_mb = &mb[-s->mb_width-1];
+ mbsplits_top = vp8_mbsplits[top_mb->partitioning];
+ top_mv = top_mb->bmv;
+
if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) {
if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) {
part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]);
}
static av_always_inline
-void decode_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y)
+void decode_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int layout)
{
- VP8Macroblock *mb_edge[3] = { mb + 2 /* top */,
+ VP8Macroblock *mb_edge[3] = { 0 /* top */,
mb - 1 /* left */,
- mb + 1 /* top-left */ };
+ 0 /* 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;
+ if (!layout) { // layout is inlined (s->mb_layout is not)
+ mb_edge[0] = mb + 2;
+ mb_edge[2] = mb + 1;
+ }
+ else {
+ mb_edge[0] = mb - s->mb_width-1;
+ mb_edge[2] = mb - s->mb_width-2;
+ }
+
AV_ZERO32(&near_mv[0]);
AV_ZERO32(&near_mv[1]);
AV_ZERO32(&near_mv[2]);
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 */
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {
/* 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])], mb_x, mb_y);
- 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);
+ clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
+ 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;
- mb->mv = mb->bmv[decode_splitmvs(s, c, mb) - 1];
+ mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout) - 1];
} else {
mb->mv.y += read_mv_component(c, s->prob->mvc[0]);
mb->mv.x += read_mv_component(c, s->prob->mvc[1]);
mb->bmv[0] = mb->mv;
}
} else {
- clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR], mb_x, mb_y);
+ clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR]);
mb->bmv[0] = mb->mv;
}
} else {
- clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST], mb_x, mb_y);
+ clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST]);
mb->bmv[0] = mb->mv;
}
} else {
}
static av_always_inline
-void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c,
- int mb_x, int keyframe)
+void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
+ int mb_x, int keyframe, int layout)
{
- uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
+ uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
+
+ if (layout == 1) {
+ VP8Macroblock *mb_top = mb - s->mb_width - 1;
+ memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4);
+ }
if (keyframe) {
int x, y;
- uint8_t* const top = s->intra4x4_pred_mode_top + 4 * mb_x;
+ uint8_t* top;
uint8_t* const left = s->intra4x4_pred_mode_left;
+ if (layout == 1)
+ top = mb->intra4x4_pred_mode_top;
+ else
+ top = s->intra4x4_pred_mode_top + 4 * mb_x;
for (y = 0; y < 4; y++) {
for (x = 0; x < 4; x++) {
const uint8_t *ctx;
}
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, int layout)
{
VP56RangeCoder *c = &s->c;
if (s->segmentation.update_map)
*segment = vp8_rac_get_tree(c, vp8_segmentid_tree, s->prob->segmentid);
- s->segment = *segment;
+ else if (s->segmentation.enabled)
+ *segment = ref ? *ref : *segment;
+ mb->segment = *segment;
mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0;
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra, vp8_pred16x16_prob_intra);
if (mb->mode == MODE_I4x4) {
- decode_intra4x4_modes(s, c, mb_x, 1);
+ decode_intra4x4_modes(s, c, mb, mb_x, 1, layout);
} else {
const uint32_t modes = vp8_pred4x4_mode[mb->mode] * 0x01010101u;
- AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
- AV_WN32A(s->intra4x4_pred_mode_left, modes);
+ if (s->mb_layout == 1)
+ AV_WN32A(mb->intra4x4_pred_mode_top, modes);
+ else
+ AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
+ AV_WN32A( s->intra4x4_pred_mode_left, modes);
}
- s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra);
+ mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra);
mb->ref_frame = VP56_FRAME_CURRENT;
} else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) {
// inter MB, 16.2
s->ref_count[mb->ref_frame-1]++;
// motion vectors, 16.3
- decode_mvs(s, mb, mb_x, mb_y);
+ decode_mvs(s, mb, mb_x, mb_y, layout);
} else {
// intra MB, 16.1
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
if (mb->mode == MODE_I4x4)
- decode_intra4x4_modes(s, c, mb_x, 0);
+ decode_intra4x4_modes(s, c, mb, mb_x, 0, layout);
- s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c);
+ mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c);
mb->ref_frame = VP56_FRAME_CURRENT;
mb->partitioning = VP8_SPLITMVMODE_NONE;
AV_ZERO32(&mb->bmv[0]);
#ifndef decode_block_coeffs_internal
/**
- * @param c arithmetic bitstream reader context
+ * @param r arithmetic bitstream reader context
* @param block destination for block coefficients
* @param probs probabilities to use when reading trees from the bitstream
* @param i initial coeff index, 0 unless a separate DC block is coded
- * @param zero_nhood the initial prediction context for number of surrounding
- * all-zero blocks (only left/top, so 0-2)
* @param qmul array holding the dc/ac dequant factor at position 0/1
* @return 0 if no coeffs were decoded
* 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],
+static int decode_block_coeffs_internal(VP56RangeCoder *r, DCTELEM block[16],
+ uint8_t probs[16][3][NUM_DCT_TOKENS-1],
int i, uint8_t *token_prob, int16_t qmul[2])
{
+ VP56RangeCoder c = *r;
goto skip_eob;
do {
int coeff;
- if (!vp56_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
- return i;
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB
+ break;
skip_eob:
- if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0
if (++i == 16)
- return i; // invalid input; blocks should end with EOB
+ break; // invalid input; blocks should end with EOB
token_prob = probs[i][0];
goto skip_eob;
}
- if (!vp56_rac_get_prob_branchy(c, token_prob[2])) { // DCT_1
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1
coeff = 1;
token_prob = probs[i+1][1];
} else {
- if (!vp56_rac_get_prob_branchy(c, token_prob[3])) { // DCT 2,3,4
- coeff = vp56_rac_get_prob_branchy(c, token_prob[4]);
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
+ coeff = vp56_rac_get_prob_branchy(&c, token_prob[4]);
if (coeff)
- coeff += vp56_rac_get_prob(c, token_prob[5]);
+ coeff += vp56_rac_get_prob(&c, token_prob[5]);
coeff += 2;
} else {
// DCT_CAT*
- if (!vp56_rac_get_prob_branchy(c, token_prob[6])) {
- if (!vp56_rac_get_prob_branchy(c, token_prob[7])) { // DCT_CAT1
- coeff = 5 + vp56_rac_get_prob(c, vp8_dct_cat1_prob[0]);
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[6])) {
+ if (!vp56_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1
+ coeff = 5 + vp56_rac_get_prob(&c, vp8_dct_cat1_prob[0]);
} else { // DCT_CAT2
coeff = 7;
- coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[0]) << 1;
- coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[1]);
+ coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
+ coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[1]);
}
} else { // DCT_CAT3 and up
- int a = vp56_rac_get_prob(c, token_prob[8]);
- int b = vp56_rac_get_prob(c, token_prob[9+a]);
+ int a = vp56_rac_get_prob(&c, token_prob[8]);
+ int b = vp56_rac_get_prob(&c, token_prob[9+a]);
int cat = (a<<1) + b;
coeff = 3 + (8<<cat);
- coeff += vp8_rac_get_coeff(c, ff_vp8_dct_cat_prob[cat]);
+ coeff += vp8_rac_get_coeff(&c, ff_vp8_dct_cat_prob[cat]);
}
}
token_prob = probs[i+1][2];
}
- block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
+ block[zigzag_scan[i]] = (vp8_rac_get(&c) ? -coeff : coeff) * qmul[!!i];
} while (++i < 16);
+ *r = c;
return i;
}
#endif
+/**
+ * @param c arithmetic bitstream reader context
+ * @param block destination for block coefficients
+ * @param probs probabilities to use when reading trees from the bitstream
+ * @param i initial coeff index, 0 unless a separate DC block is coded
+ * @param zero_nhood the initial prediction context for number of surrounding
+ * all-zero blocks (only left/top, so 0-2)
+ * @param qmul array holding the dc/ac dequant factor at position 0/1
+ * @return 0 if no coeffs were decoded
+ * otherwise, the index of the last coeff decoded plus one
+ */
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];
}
static av_always_inline
-void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
+void decode_mb_coeffs(VP8Context *s, VP8ThreadData *td, VP56RangeCoder *c, VP8Macroblock *mb,
uint8_t t_nnz[9], uint8_t l_nnz[9])
{
int i, x, y, luma_start = 0, luma_ctx = 3;
int nnz_pred, nnz, nnz_total = 0;
- int segment = s->segment;
+ int segment = mb->segment;
int block_dc = 0;
if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
nnz_pred = t_nnz[8] + l_nnz[8];
// decode DC values and do hadamard
- nnz = decode_block_coeffs(c, s->block_dc, s->prob->token[1], 0, nnz_pred,
+ nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0, nnz_pred,
s->qmat[segment].luma_dc_qmul);
l_nnz[8] = t_nnz[8] = !!nnz;
if (nnz) {
nnz_total += nnz;
block_dc = 1;
if (nnz == 1)
- s->vp8dsp.vp8_luma_dc_wht_dc(s->block, s->block_dc);
+ s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc);
else
- s->vp8dsp.vp8_luma_dc_wht(s->block, s->block_dc);
+ s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc);
}
luma_start = 1;
luma_ctx = 0;
for (y = 0; y < 4; y++)
for (x = 0; x < 4; x++) {
nnz_pred = l_nnz[y] + t_nnz[x];
- nnz = decode_block_coeffs(c, s->block[y][x], s->prob->token[luma_ctx], luma_start,
+ nnz = decode_block_coeffs(c, td->block[y][x], s->prob->token[luma_ctx], luma_start,
nnz_pred, s->qmat[segment].luma_qmul);
// nnz+block_dc may be one more than the actual last index, but we don't care
- s->non_zero_count_cache[y][x] = nnz + block_dc;
+ td->non_zero_count_cache[y][x] = nnz + block_dc;
t_nnz[x] = l_nnz[y] = !!nnz;
nnz_total += nnz;
}
for (y = 0; y < 2; y++)
for (x = 0; x < 2; x++) {
nnz_pred = l_nnz[i+2*y] + t_nnz[i+2*x];
- nnz = decode_block_coeffs(c, s->block[i][(y<<1)+x], s->prob->token[2], 0,
+ nnz = decode_block_coeffs(c, td->block[i][(y<<1)+x], s->prob->token[2], 0,
nnz_pred, s->qmat[segment].chroma_qmul);
- s->non_zero_count_cache[i][(y<<1)+x] = nnz;
+ td->non_zero_count_cache[i][(y<<1)+x] = nnz;
t_nnz[i+2*x] = l_nnz[i+2*y] = !!nnz;
nnz_total += nnz;
}
}
static av_always_inline
-void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
- int mb_x, int mb_y)
+void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
+ VP8Macroblock *mb, int mb_x, int mb_y)
{
AVCodecContext *avctx = s->avctx;
- int x, y, mode, nnz, tr;
+ int x, y, mode, nnz;
+ uint32_t tr;
// for the first row, we need to run xchg_mb_border to init the top edge to 127
// otherwise, skip it if we aren't going to deblock
- if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 1);
s->hpc.pred16x16[mode](dst[0], s->linesize);
} else {
uint8_t *ptr = dst[0];
- uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
+ uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
uint8_t tr_top[4] = { 127, 127, 127, 127 };
// all blocks on the right edge of the macroblock use bottom edge
// from the top macroblock
if (!(!mb_y && avctx->flags & CODEC_FLAG_EMU_EDGE) &&
mb_x == s->mb_width-1) {
- tr = tr_right[-1]*0x01010101;
+ tr = tr_right[-1]*0x01010101u;
tr_right = (uint8_t *)&tr;
}
if (mb->skip)
- AV_ZERO128(s->non_zero_count_cache);
+ AV_ZERO128(td->non_zero_count_cache);
for (y = 0; y < 4; y++) {
uint8_t *topright = ptr + 4 - s->linesize;
AV_COPY32(ptr+4*x+s->linesize*3, copy_dst+36);
}
- nnz = s->non_zero_count_cache[y][x];
+ nnz = td->non_zero_count_cache[y][x];
if (nnz) {
if (nnz == 1)
- s->vp8dsp.vp8_idct_dc_add(ptr+4*x, s->block[y][x], s->linesize);
+ s->vp8dsp.vp8_idct_dc_add(ptr+4*x, td->block[y][x], s->linesize);
else
- s->vp8dsp.vp8_idct_add(ptr+4*x, s->block[y][x], s->linesize);
+ s->vp8dsp.vp8_idct_add(ptr+4*x, td->block[y][x], s->linesize);
}
topright += 4;
}
}
if (avctx->flags & CODEC_FLAG_EMU_EDGE) {
- mode = check_intra_pred8x8_mode_emuedge(s->chroma_pred_mode, mb_x, mb_y);
+ mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode, mb_x, mb_y);
} else {
- mode = check_intra_pred8x8_mode(s->chroma_pred_mode, mb_x, mb_y);
+ mode = check_intra_pred8x8_mode(mb->chroma_pred_mode, mb_x, mb_y);
}
s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
s->hpc.pred8x8[mode](dst[2], s->uvlinesize);
- if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 0);
};
/**
- * Generic MC function.
+ * luma MC function
*
* @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, VP8ThreadData *td, 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]) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize,
+ s->dsp.emulated_edge_mc(td->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize,
block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
x_off - mx_idx, y_off - my_idx, width, height);
- src = s->edge_emu_buffer + mx_idx + linesize * my_idx;
+ src = td->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);
+ }
}
+/**
+ * chroma MC function
+ *
+ * @param s VP8 decoding context
+ * @param dst1 target buffer for block data at block position (U plane)
+ * @param dst2 target buffer for block data at block position (V plane)
+ * @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 block_w width of block (16, 8 or 4)
+ * @param block_h height of block (always same as block_w)
+ * @param width width of src/dst plane data
+ * @param height height of src/dst plane data
+ * @param linesize size of a single line of plane data, including padding
+ * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
+ */
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, VP8ThreadData *td, 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,
+ s->dsp.emulated_edge_mc(td->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize,
block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
x_off - mx_idx, y_off - my_idx, width, height);
- src1 = s->edge_emu_buffer + mx_idx + linesize * my_idx;
+ src1 = td->edge_emu_buffer + mx_idx + linesize * my_idx;
mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
- s->dsp.emulated_edge_mc(s->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize,
+ s->dsp.emulated_edge_mc(td->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize,
block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
x_off - mx_idx, y_off - my_idx, width, height);
- src2 = s->edge_emu_buffer + mx_idx + linesize * my_idx;
+ src2 = td->edge_emu_buffer + mx_idx + linesize * my_idx;
mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
} else {
mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
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);
}
}
static av_always_inline
-void vp8_mc_part(VP8Context *s, uint8_t *dst[3],
+void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
AVFrame *ref_frame, int x_off, int y_off,
int bx_off, int by_off,
int block_w, int block_h,
VP56mv uvmv = *mv;
/* 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,
+ vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_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]);
bx_off >>= 1; by_off >>= 1;
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,
+ vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_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);
* Apply motion vectors to prediction buffer, chapter 18.
*/
static av_always_inline
-void inter_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
- int mb_x, int mb_y)
+void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
+ VP8Macroblock *mb, int mb_x, int mb_y)
{
int x_off = mb_x << 4, y_off = mb_y << 4;
int width = 16*s->mb_width, height = 16*s->mb_height;
switch (mb->partitioning) {
case VP8_SPLITMVMODE_NONE:
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 0, 16, 16, width, height, &mb->mv);
break;
case VP8_SPLITMVMODE_4x4: {
/* Y */
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],
+ vp8_mc_luma(s, td, dst[0] + 4*y*s->linesize + x*4,
+ 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.x &= ~7;
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,
+ vp8_mc_chroma(s, td, dst[1] + 4*y*s->uvlinesize + x*4,
+ 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]);
break;
}
case VP8_SPLITMVMODE_16x8:
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 0, 16, 8, width, height, &bmv[0]);
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 8, 16, 8, width, height, &bmv[1]);
break;
case VP8_SPLITMVMODE_8x16:
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 0, 8, 16, width, height, &bmv[0]);
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
8, 0, 8, 16, width, height, &bmv[1]);
break;
case VP8_SPLITMVMODE_8x8:
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 0, 8, 8, width, height, &bmv[0]);
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
8, 0, 8, 8, width, height, &bmv[1]);
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
0, 8, 8, 8, width, height, &bmv[2]);
- vp8_mc_part(s, dst, ref, x_off, y_off,
+ vp8_mc_part(s, td, dst, ref, x_off, y_off,
8, 8, 8, 8, width, height, &bmv[3]);
break;
}
}
-static av_always_inline void idct_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb)
+static av_always_inline void idct_mb(VP8Context *s, VP8ThreadData *td,
+ uint8_t *dst[3], VP8Macroblock *mb)
{
int x, y, ch;
if (mb->mode != MODE_I4x4) {
uint8_t *y_dst = dst[0];
for (y = 0; y < 4; y++) {
- uint32_t nnz4 = AV_RL32(s->non_zero_count_cache[y]);
+ uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]);
if (nnz4) {
if (nnz4&~0x01010101) {
for (x = 0; x < 4; x++) {
if ((uint8_t)nnz4 == 1)
- s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, s->block[y][x], s->linesize);
+ s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, td->block[y][x], s->linesize);
else if((uint8_t)nnz4 > 1)
- s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
+ s->vp8dsp.vp8_idct_add(y_dst+4*x, td->block[y][x], s->linesize);
nnz4 >>= 8;
if (!nnz4)
break;
}
} else {
- s->vp8dsp.vp8_idct_dc_add4y(y_dst, s->block[y], s->linesize);
+ s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize);
}
}
y_dst += 4*s->linesize;
}
for (ch = 0; ch < 2; ch++) {
- uint32_t nnz4 = AV_RL32(s->non_zero_count_cache[4+ch]);
+ uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4+ch]);
if (nnz4) {
uint8_t *ch_dst = dst[1+ch];
if (nnz4&~0x01010101) {
for (y = 0; y < 2; y++) {
for (x = 0; x < 2; x++) {
if ((uint8_t)nnz4 == 1)
- s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
+ s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, td->block[4+ch][(y<<1)+x], s->uvlinesize);
else if((uint8_t)nnz4 > 1)
- s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
+ s->vp8dsp.vp8_idct_add(ch_dst+4*x, td->block[4+ch][(y<<1)+x], s->uvlinesize);
nnz4 >>= 8;
if (!nnz4)
- break;
+ goto chroma_idct_end;
}
ch_dst += 4*s->uvlinesize;
}
} else {
- s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, s->block[4+ch], s->uvlinesize);
+ s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4+ch], s->uvlinesize);
}
}
+chroma_idct_end: ;
}
}
int interior_limit, filter_level;
if (s->segmentation.enabled) {
- filter_level = s->segmentation.filter_level[s->segment];
+ filter_level = s->segmentation.filter_level[mb->segment];
if (!s->segmentation.absolute_vals)
filter_level += s->filter.level;
} else
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 release_queued_segmaps(VP8Context *s, int is_close)
+{
+ int leave_behind = is_close ? 0 : !s->maps_are_invalid;
+ while (s->num_maps_to_be_freed > leave_behind)
+ av_freep(&s->segmentation_maps[--s->num_maps_to_be_freed]);
+ s->maps_are_invalid = 0;
+}
+
+#define MARGIN (16 << 2)
+static void vp8_decode_mv_mb_modes(AVCodecContext *avctx, AVFrame *curframe,
+ AVFrame *prev_frame)
+{
+ VP8Context *s = avctx->priv_data;
+ int mb_x, mb_y;
+
+ s->mv_min.y = -MARGIN;
+ s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
+ for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+ VP8Macroblock *mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1);
+ int mb_xy = mb_y*s->mb_width;
+
+ AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);
+
+ s->mv_min.x = -MARGIN;
+ s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
+ for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
+ if (mb_y == 0)
+ AV_WN32A((mb-s->mb_width-1)->intra4x4_pred_mode_top, DC_PRED*0x01010101);
+ decode_mb_mode(s, mb, mb_x, mb_y, curframe->ref_index[0] + mb_xy,
+ prev_frame && prev_frame->ref_index[0] ? prev_frame->ref_index[0] + mb_xy : NULL, 1);
+ s->mv_min.x -= 64;
+ s->mv_max.x -= 64;
+ }
+ s->mv_min.y -= 64;
+ s->mv_max.y -= 64;
+ }
+}
+
+#if HAVE_THREADS
+#define check_thread_pos(td, otd, mb_x_check, mb_y_check)\
+ do {\
+ int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF);\
+ if (otd->thread_mb_pos < tmp) {\
+ pthread_mutex_lock(&otd->lock);\
+ td->wait_mb_pos = tmp;\
+ do {\
+ if (otd->thread_mb_pos >= tmp)\
+ break;\
+ pthread_cond_wait(&otd->cond, &otd->lock);\
+ } while (1);\
+ td->wait_mb_pos = INT_MAX;\
+ pthread_mutex_unlock(&otd->lock);\
+ }\
+ } while(0);
+
+#define update_pos(td, mb_y, mb_x)\
+ do {\
+ int pos = (mb_y << 16) | (mb_x & 0xFFFF);\
+ int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE) && (num_jobs > 1);\
+ int is_null = (next_td == NULL) || (prev_td == NULL);\
+ int pos_check = (is_null) ? 1 :\
+ (next_td != td && pos >= next_td->wait_mb_pos) ||\
+ (prev_td != td && pos >= prev_td->wait_mb_pos);\
+ td->thread_mb_pos = pos;\
+ if (sliced_threading && pos_check) {\
+ pthread_mutex_lock(&td->lock);\
+ pthread_cond_broadcast(&td->cond);\
+ pthread_mutex_unlock(&td->lock);\
+ }\
+ } while(0);
+#else
+#define check_thread_pos(td, otd, mb_x_check, mb_y_check)
+#define update_pos(td, mb_y, mb_x)
+#endif
+
+static void vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
+ int jobnr, int threadnr)
{
- VP8FilterStrength *f = s->filter_strength;
+ VP8Context *s = avctx->priv_data;
+ VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr];
+ int mb_y = td->thread_mb_pos>>16;
+ int i, y, mb_x, mb_xy = mb_y*s->mb_width;
+ int num_jobs = s->num_jobs;
+ AVFrame *curframe = s->curframe, *prev_frame = s->prev_frame;
+ VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)];
+ VP8Macroblock *mb;
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;
+ if (mb_y == 0) prev_td = td;
+ else prev_td = &s->thread_data[(jobnr + num_jobs - 1)%num_jobs];
+ if (mb_y == s->mb_height-1) next_td = td;
+ else next_td = &s->thread_data[(jobnr + 1)%num_jobs];
+ if (s->mb_layout == 1)
+ mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1);
+ else {
+ mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
+ memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
+ AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);
+ }
+
+ memset(td->left_nnz, 0, sizeof(td->left_nnz));
+ // left edge of 129 for intra prediction
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
+ for (i = 0; i < 3; i++)
+ for (y = 0; y < 16>>!!i; y++)
+ dst[i][y*curframe->linesize[i]-1] = 129;
+ if (mb_y == 1) {
+ s->top_border[0][15] = s->top_border[0][23] = s->top_border[0][31] = 129;
+ }
+ }
+
+ s->mv_min.x = -MARGIN;
+ s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
+
+ for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
+ // Wait for previous thread to read mb_x+2, and reach mb_y-1.
+ if (prev_td != td) {
+ if (threadnr != 0) {
+ check_thread_pos(td, prev_td, mb_x+1, mb_y-1);
+ } else {
+ check_thread_pos(td, prev_td, (s->mb_width+3) + (mb_x+1), mb_y-1);
+ }
+ }
+
+ 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);
+
+ if (!s->mb_layout)
+ decode_mb_mode(s, mb, mb_x, mb_y, curframe->ref_index[0] + mb_xy,
+ prev_frame && prev_frame->ref_index[0] ? prev_frame->ref_index[0] + mb_xy : NULL, 0);
+
+ prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
+
+ if (!mb->skip)
+ decode_mb_coeffs(s, td, c, mb, s->top_nnz[mb_x], td->left_nnz);
+
+ if (mb->mode <= MODE_I4x4)
+ intra_predict(s, td, dst, mb, mb_x, mb_y);
+ else
+ inter_predict(s, td, dst, mb, mb_x, mb_y);
+
+ prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);
+
+ if (!mb->skip) {
+ idct_mb(s, td, dst, mb);
+ } else {
+ AV_ZERO64(td->left_nnz);
+ AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
+
+ // Reset DC block predictors if they would exist if the mb had coefficients
+ if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
+ td->left_nnz[8] = 0;
+ s->top_nnz[mb_x][8] = 0;
+ }
+ }
+
+ if (s->deblock_filter)
+ filter_level_for_mb(s, mb, &td->filter_strength[mb_x]);
+
+ if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs-1) {
+ if (s->filter.simple)
+ backup_mb_border(s->top_border[mb_x+1], dst[0], NULL, NULL, s->linesize, 0, 1);
+ else
+ backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
+ }
+
+ prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2);
- for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
- backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
- filter_mb(s, dst, f++, mb_x, mb_y);
dst[0] += 16;
dst[1] += 8;
dst[2] += 8;
+ s->mv_min.x -= 64;
+ s->mv_max.x -= 64;
+
+ if (mb_x == s->mb_width+1) {
+ update_pos(td, mb_y, s->mb_width+3);
+ } else {
+ update_pos(td, mb_y, mb_x);
+ }
}
}
-static void filter_mb_row_simple(VP8Context *s, int mb_y)
+static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata,
+ int jobnr, int threadnr)
{
- VP8FilterStrength *f = s->filter_strength;
- uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
- int mb_x;
+ VP8Context *s = avctx->priv_data;
+ VP8ThreadData *td = &s->thread_data[threadnr];
+ int mb_x, mb_y = td->thread_mb_pos>>16, num_jobs = s->num_jobs;
+ AVFrame *curframe = s->curframe;
+ VP8Macroblock *mb;
+ VP8ThreadData *prev_td, *next_td;
+ uint8_t *dst[3] = {
+ curframe->data[0] + 16*mb_y*s->linesize,
+ curframe->data[1] + 8*mb_y*s->uvlinesize,
+ curframe->data[2] + 8*mb_y*s->uvlinesize
+ };
+
+ if (s->mb_layout == 1)
+ mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1);
+ else
+ mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
+
+ if (mb_y == 0) prev_td = td;
+ else prev_td = &s->thread_data[(jobnr + num_jobs - 1)%num_jobs];
+ if (mb_y == s->mb_height-1) next_td = td;
+ else next_td = &s->thread_data[(jobnr + 1)%num_jobs];
+
+ for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) {
+ VP8FilterStrength *f = &td->filter_strength[mb_x];
+ if (prev_td != td) {
+ check_thread_pos(td, prev_td, (mb_x+1) + (s->mb_width+3), mb_y-1);
+ }
+ if (next_td != td)
+ if (next_td != &s->thread_data[0]) {
+ check_thread_pos(td, next_td, mb_x+1, mb_y+1);
+ }
+
+ if (num_jobs == 1) {
+ if (s->filter.simple)
+ backup_mb_border(s->top_border[mb_x+1], dst[0], NULL, NULL, s->linesize, 0, 1);
+ else
+ backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
+ }
- for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
- backup_mb_border(s->top_border[mb_x+1], dst, NULL, NULL, s->linesize, 0, 1);
- filter_mb_simple(s, dst, f++, mb_x, mb_y);
- dst += 16;
+ if (s->filter.simple)
+ filter_mb_simple(s, dst[0], f, mb_x, mb_y);
+ else
+ filter_mb(s, dst, f, mb_x, mb_y);
+ dst[0] += 16;
+ dst[1] += 8;
+ dst[2] += 8;
+
+ update_pos(td, mb_y, (s->mb_width+3) + mb_x);
}
}
+static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
+ int jobnr, int threadnr)
+{
+ VP8Context *s = avctx->priv_data;
+ VP8ThreadData *td = &s->thread_data[jobnr];
+ VP8ThreadData *next_td = NULL, *prev_td = NULL;
+ AVFrame *curframe = s->curframe;
+ int mb_y, num_jobs = s->num_jobs;
+ td->thread_nr = threadnr;
+ for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) {
+ if (mb_y >= s->mb_height) break;
+ td->thread_mb_pos = mb_y<<16;
+ vp8_decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr);
+ if (s->deblock_filter)
+ vp8_filter_mb_row(avctx, tdata, jobnr, threadnr);
+ update_pos(td, mb_y, INT_MAX & 0xFFFF);
+
+ s->mv_min.y -= 64;
+ s->mv_max.y -= 64;
+
+ if (avctx->active_thread_type == FF_THREAD_FRAME)
+ ff_thread_report_progress(curframe, mb_y, 0);
+ }
+
+ return 0;
+}
+
static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
AVPacket *avpkt)
{
VP8Context *s = avctx->priv_data;
- int ret, mb_x, mb_y, i, y, referenced;
+ int ret, i, referenced, num_jobs;
enum AVDiscard skip_thresh;
- AVFrame *av_uninit(curframe);
+ AVFrame *av_uninit(curframe), *prev_frame;
+
+ release_queued_segmaps(s, 0);
if ((ret = decode_frame_header(s, avpkt->data, avpkt->size)) < 0)
- return ret;
+ goto err;
+
+ prev_frame = s->framep[VP56_FRAME_CURRENT];
referenced = s->update_last || s->update_golden == VP56_FRAME_CURRENT
|| s->update_altref == VP56_FRAME_CURRENT;
if (avctx->skip_frame >= skip_thresh) {
s->invisible = 1;
+ memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
goto skip_decode;
}
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])
+ vp8_release_frame(s, &s->frames[i], 1, 0);
+
+ // 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 (curframe->data[0])
- avctx->release_buffer(avctx, curframe);
-
- curframe->key_frame = s->keyframe;
- curframe->pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
- curframe->reference = referenced ? 3 : 0;
- if ((ret = avctx->get_buffer(avctx, curframe))) {
- av_log(avctx, AV_LOG_ERROR, "get_buffer() failed!\n");
- return ret;
+ if (i == 5) {
+ av_log(avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
+ abort();
}
+ if (curframe->data[0])
+ vp8_release_frame(s, curframe, 1, 0);
// 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
!s->framep[VP56_FRAME_GOLDEN] ||
!s->framep[VP56_FRAME_GOLDEN2])) {
av_log(avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n");
- return AVERROR_INVALIDDATA;
+ ret = AVERROR_INVALIDDATA;
+ goto err;
+ }
+
+ curframe->key_frame = s->keyframe;
+ curframe->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
+ curframe->reference = referenced ? 3 : 0;
+ if ((ret = vp8_alloc_frame(s, curframe))) {
+ av_log(avctx, AV_LOG_ERROR, "get_buffer() failed!\n");
+ goto err;
+ }
+
+ // 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);
s->linesize = curframe->linesize[0];
s->uvlinesize = curframe->linesize[1];
- if (!s->edge_emu_buffer)
- s->edge_emu_buffer = av_malloc(21*s->linesize);
+ if (!s->thread_data[0].edge_emu_buffer)
+ for (i = 0; i < MAX_THREADS; i++)
+ s->thread_data[i].edge_emu_buffer = av_malloc(21*s->linesize);
memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz));
-
/* Zero macroblock structures for top/top-left prediction from outside the frame. */
- memset(s->macroblocks + s->mb_height*2 - 1, 0, (s->mb_width+1)*sizeof(*s->macroblocks));
+ if (!s->mb_layout)
+ memset(s->macroblocks + s->mb_height*2 - 1, 0, (s->mb_width+1)*sizeof(*s->macroblocks));
+ if (!s->mb_layout && s->keyframe)
+ memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4);
// top edge of 127 for intra prediction
if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
s->top_border[0][15] = s->top_border[0][23] = 127;
- memset(s->top_border[1]-1, 127, s->mb_width*sizeof(*s->top_border)+1);
+ s->top_border[0][31] = 127;
+ memset(s->top_border[1], 127, s->mb_width*sizeof(*s->top_border));
}
memset(s->ref_count, 0, sizeof(s->ref_count));
- if (s->keyframe)
- memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4);
- for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
- VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)];
- VP8Macroblock *mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
- int mb_xy = mb_y*s->mb_width;
- uint8_t *dst[3] = {
- curframe->data[0] + 16*mb_y*s->linesize,
- curframe->data[1] + 8*mb_y*s->uvlinesize,
- curframe->data[2] + 8*mb_y*s->uvlinesize
- };
-
- memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
- memset(s->left_nnz, 0, sizeof(s->left_nnz));
- AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);
- // left edge of 129 for intra prediction
- if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
- for (i = 0; i < 3; i++)
- for (y = 0; y < 16>>!!i; y++)
- dst[i][y*curframe->linesize[i]-1] = 129;
- if (mb_y == 1) // top left edge is also 129
- s->top_border[0][15] = s->top_border[0][23] = s->top_border[0][31] = 129;
- }
-
- 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);
-
- prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
+ // Make sure the previous frame has read its segmentation map,
+ // if we re-use the same map.
+ if (prev_frame && s->segmentation.enabled && !s->segmentation.update_map)
+ ff_thread_await_progress(prev_frame, 1, 0);
- if (!mb->skip)
- decode_mb_coeffs(s, c, mb, s->top_nnz[mb_x], s->left_nnz);
+ if (s->mb_layout == 1)
+ vp8_decode_mv_mb_modes(avctx, curframe, prev_frame);
- if (mb->mode <= MODE_I4x4)
- intra_predict(s, dst, mb, mb_x, mb_y);
- else
- inter_predict(s, dst, mb, mb_x, mb_y);
-
- prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);
-
- if (!mb->skip) {
- idct_mb(s, dst, mb);
- } else {
- AV_ZERO64(s->left_nnz);
- AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
-
- // Reset DC block predictors if they would exist if the mb had coefficients
- if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
- s->left_nnz[8] = 0;
- s->top_nnz[mb_x][8] = 0;
- }
- }
-
- if (s->deblock_filter)
- filter_level_for_mb(s, mb, &s->filter_strength[mb_x]);
-
- prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2);
-
- dst[0] += 16;
- dst[1] += 8;
- dst[2] += 8;
- }
- if (s->deblock_filter) {
- if (s->filter.simple)
- filter_mb_row_simple(s, mb_y);
- else
- filter_mb_row(s, mb_y);
- }
+ if (avctx->active_thread_type == FF_THREAD_FRAME)
+ num_jobs = 1;
+ else
+ num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count);
+ s->num_jobs = num_jobs;
+ s->curframe = curframe;
+ s->prev_frame = prev_frame;
+ s->mv_min.y = -MARGIN;
+ s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
+ for (i = 0; i < MAX_THREADS; i++) {
+ s->thread_data[i].thread_mb_pos = 0;
+ s->thread_data[i].wait_mb_pos = INT_MAX;
}
+ avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL, num_jobs);
+
+ ff_thread_report_progress(curframe, INT_MAX, 0);
+ memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
skip_decode:
// if future frames don't use the updated probabilities,
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]);
-
if (!s->invisible) {
- *(AVFrame*)data = *s->framep[VP56_FRAME_CURRENT];
+ *(AVFrame*)data = *curframe;
*data_size = sizeof(AVFrame);
}
return avpkt->size;
+err:
+ memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
+ return ret;
}
static av_cold int vp8_decode_init(AVCodecContext *avctx)
VP8Context *s = avctx->priv_data;
s->avctx = avctx;
- avctx->pix_fmt = PIX_FMT_YUV420P;
+ avctx->pix_fmt = AV_PIX_FMT_YUV420P;
- dsputil_init(&s->dsp, avctx);
- ff_h264_pred_init(&s->hpc, CODEC_ID_VP8);
+ ff_dsputil_init(&s->dsp, avctx);
+ ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1);
ff_vp8dsp_init(&s->vp8dsp);
return 0;
static av_cold int vp8_decode_free(AVCodecContext *avctx)
{
- vp8_decode_flush(avctx);
+ vp8_decode_flush_impl(avctx, 0, 1, 1);
+ release_queued_segmaps(avctx->priv_data, 1);
+ 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;
+
+ if (s->macroblocks_base &&
+ (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) {
+ free_buffers(s);
+ s->maps_are_invalid = 1;
+ s->mb_width = s_src->mb_width;
+ s->mb_height = s_src->mb_height;
+ }
+
+ 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,
- CODEC_ID_VP8,
- sizeof(VP8Context),
- vp8_decode_init,
- NULL,
- vp8_decode_free,
- vp8_decode_frame,
- CODEC_CAP_DR1,
- .flush = vp8_decode_flush,
- .long_name = NULL_IF_CONFIG_SMALL("On2 VP8"),
+ .name = "vp8",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .id = AV_CODEC_ID_VP8,
+ .priv_data_size = sizeof(VP8Context),
+ .init = vp8_decode_init,
+ .close = vp8_decode_free,
+ .decode = vp8_decode_frame,
+ .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS | CODEC_CAP_SLICE_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),
};
projects / ffmpeg / blobdiff