32f3c541 |
/* |
89f2f5db |
* VP7/VP8 compatible video decoder |
3b636f21 |
*
* Copyright (C) 2010 David Conrad
* Copyright (C) 2010 Ronald S. Bultje |
79793f83 |
* Copyright (C) 2010 Fiona Glaser |
951455c1 |
* Copyright (C) 2012 Daniel Kang |
89f2f5db |
* Copyright (C) 2014 Peter Ross |
3b636f21 |
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
|
737eb597 |
#include "libavutil/imgutils.h" |
53c20f17 |
|
3b636f21 |
#include "avcodec.h" |
f3a29b75 |
#include "internal.h" |
f4d581cd |
#include "mathops.h" |
3b636f21 |
#include "rectangle.h" |
4773d904 |
#include "thread.h" |
53c20f17 |
#include "vp8.h"
#include "vp8data.h" |
3b636f21 |
|
a7878c9f |
#if ARCH_ARM
# include "arm/vp8.h"
#endif
|
89f2f5db |
#if CONFIG_VP7_DECODER && CONFIG_VP8_DECODER
#define VPX(vp7, f) (vp7 ? vp7_ ## f : vp8_ ## f)
#elif CONFIG_VP7_DECODER
#define VPX(vp7, f) vp7_ ## f
#else // CONFIG_VP8_DECODER
#define VPX(vp7, f) vp8_ ## f
#endif
|
56535793 |
static void free_buffers(VP8Context *s)
{ |
951455c1 |
int i;
if (s->thread_data)
for (i = 0; i < MAX_THREADS; i++) { |
65340c97 |
#if HAVE_THREADS
pthread_cond_destroy(&s->thread_data[i].cond);
pthread_mutex_destroy(&s->thread_data[i].lock);
#endif |
951455c1 |
av_freep(&s->thread_data[i].filter_strength);
}
av_freep(&s->thread_data); |
56535793 |
av_freep(&s->macroblocks_base);
av_freep(&s->intra4x4_pred_mode_top);
av_freep(&s->top_nnz);
av_freep(&s->top_border);
s->macroblocks = NULL;
}
|
759001c5 |
static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref) |
ce42a048 |
{
int ret; |
759001c5 |
if ((ret = ff_thread_get_buffer(s->avctx, &f->tf,
ref ? AV_GET_BUFFER_FLAG_REF : 0)) < 0) |
ce42a048 |
return ret; |
759001c5 |
if (!(f->seg_map = av_buffer_allocz(s->mb_width * s->mb_height))) {
ff_thread_release_buffer(s->avctx, &f->tf); |
ce42a048 |
return AVERROR(ENOMEM);
}
return 0;
}
|
759001c5 |
static void vp8_release_frame(VP8Context *s, VP8Frame *f) |
ce42a048 |
{ |
759001c5 |
av_buffer_unref(&f->seg_map);
ff_thread_release_buffer(s->avctx, &f->tf);
}
|
89f2f5db |
#if CONFIG_VP8_DECODER |
759001c5 |
static int vp8_ref_frame(VP8Context *s, VP8Frame *dst, VP8Frame *src)
{
int ret;
vp8_release_frame(s, dst);
if ((ret = ff_thread_ref_frame(&dst->tf, &src->tf)) < 0)
return ret;
if (src->seg_map &&
!(dst->seg_map = av_buffer_ref(src->seg_map))) {
vp8_release_frame(s, dst);
return AVERROR(ENOMEM); |
ce42a048 |
} |
759001c5 |
return 0; |
ce42a048 |
} |
ac4b32df |
#endif /* CONFIG_VP8_DECODER */ |
ce42a048 |
|
759001c5 |
static void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem) |
3b636f21 |
{
VP8Context *s = avctx->priv_data;
int i;
|
759001c5 |
for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
vp8_release_frame(s, &s->frames[i]); |
3b636f21 |
memset(s->framep, 0, sizeof(s->framep));
|
759001c5 |
if (free_mem) |
bfa0f965 |
free_buffers(s); |
ce42a048 |
}
static void vp8_decode_flush(AVCodecContext *avctx)
{ |
759001c5 |
vp8_decode_flush_impl(avctx, 0); |
3b636f21 |
}
|
ac4b32df |
static VP8Frame *vp8_find_free_buffer(VP8Context *s) |
7d4c0220 |
{
VP8Frame *frame = NULL;
int i;
// find a free buffer
for (i = 0; i < 5; i++) |
ac4b32df |
if (&s->frames[i] != s->framep[VP56_FRAME_CURRENT] && |
7d4c0220 |
&s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] && |
ac4b32df |
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN] && |
7d4c0220 |
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) {
frame = &s->frames[i];
break;
}
if (i == 5) {
av_log(s->avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
abort();
}
if (frame->tf.f->data[0])
vp8_release_frame(s, frame);
return frame;
}
|
ac4b32df |
static av_always_inline
int update_dimensions(VP8Context *s, int width, int height, int is_vp7) |
3b636f21 |
{ |
951455c1 |
AVCodecContext *avctx = s->avctx; |
757d5e8e |
int i, ret; |
951455c1 |
|
4d870010 |
if (width != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base || |
4773d904 |
height != s->avctx->height) { |
759001c5 |
vp8_decode_flush_impl(s->avctx, 1); |
3b636f21 |
|
757d5e8e |
ret = ff_set_dimensions(s->avctx, width, height);
if (ret < 0)
return ret; |
4773d904 |
} |
3b636f21 |
|
53c20f17 |
s->mb_width = (s->avctx->coded_width + 15) / 16;
s->mb_height = (s->avctx->coded_height + 15) / 16; |
3b636f21 |
|
ac4b32df |
s->mb_layout = is_vp7 || avctx->active_thread_type == FF_THREAD_SLICE && |
dabea74d |
avctx->thread_count > 1; |
951455c1 |
if (!s->mb_layout) { // Frame threading and one thread |
53c20f17 |
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)); |
3b636f21 |
|
014b6b41 |
if (!s->macroblocks_base || !s->top_nnz || !s->top_border ||
!s->thread_data || (!s->intra4x4_pred_mode_top && !s->mb_layout)) {
free_buffers(s);
return AVERROR(ENOMEM);
}
|
951455c1 |
for (i = 0; i < MAX_THREADS; i++) { |
53c20f17 |
s->thread_data[i].filter_strength =
av_mallocz(s->mb_width * sizeof(*s->thread_data[0].filter_strength)); |
014b6b41 |
if (!s->thread_data[i].filter_strength) {
free_buffers(s);
return AVERROR(ENOMEM);
} |
25f056e6 |
#if HAVE_THREADS |
951455c1 |
pthread_mutex_init(&s->thread_data[i].lock, NULL);
pthread_cond_init(&s->thread_data[i].cond, NULL); |
25f056e6 |
#endif |
951455c1 |
} |
3b636f21 |
|
53c20f17 |
s->macroblocks = s->macroblocks_base + 1; |
3b636f21 |
return 0;
}
|
ac4b32df |
static int vp7_update_dimensions(VP8Context *s, int width, int height)
{
return update_dimensions(s, width, height, IS_VP7);
}
static int vp8_update_dimensions(VP8Context *s, int width, int height)
{
return update_dimensions(s, width, height, IS_VP8);
}
|
89f2f5db |
|
3b636f21 |
static void parse_segment_info(VP8Context *s)
{
VP56RangeCoder *c = &s->c;
int i;
s->segmentation.update_map = vp8_rac_get(c);
if (vp8_rac_get(c)) { // update segment feature data
s->segmentation.absolute_vals = vp8_rac_get(c);
for (i = 0; i < 4; i++)
s->segmentation.base_quant[i] = vp8_rac_get_sint(c, 7);
for (i = 0; i < 4; i++)
s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6);
}
if (s->segmentation.update_map)
for (i = 0; i < 3; i++)
s->prob->segmentid[i] = vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255;
}
static void update_lf_deltas(VP8Context *s)
{
VP56RangeCoder *c = &s->c;
int i;
|
14ba7472 |
for (i = 0; i < 4; i++) {
if (vp8_rac_get(c)) {
s->lf_delta.ref[i] = vp8_rac_get_uint(c, 6); |
3b636f21 |
|
14ba7472 |
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];
}
} |
3b636f21 |
}
static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
{
const uint8_t *sizes = buf;
int i;
s->num_coeff_partitions = 1 << vp8_rac_get_uint(&s->c, 2);
|
53c20f17 |
buf += 3 * (s->num_coeff_partitions - 1);
buf_size -= 3 * (s->num_coeff_partitions - 1); |
3b636f21 |
if (buf_size < 0)
return -1;
|
53c20f17 |
for (i = 0; i < s->num_coeff_partitions - 1; i++) {
int size = AV_RL24(sizes + 3 * i); |
3b636f21 |
if (buf_size - size < 0)
return -1;
|
905ef0d0 |
ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, size); |
3b636f21 |
buf += size;
buf_size -= size;
} |
905ef0d0 |
ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size); |
3b636f21 |
return 0;
} |
89f2f5db |
static void vp7_get_quants(VP8Context *s)
{
VP56RangeCoder *c = &s->c;
int yac_qi = vp8_rac_get_uint(c, 7);
int ydc_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
int y2dc_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
int y2ac_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
int uvdc_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
int uvac_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
s->qmat[0].luma_qmul[0] = vp7_ydc_qlookup[ydc_qi];
s->qmat[0].luma_qmul[1] = vp7_yac_qlookup[yac_qi];
s->qmat[0].luma_dc_qmul[0] = vp7_y2dc_qlookup[y2dc_qi];
s->qmat[0].luma_dc_qmul[1] = vp7_y2ac_qlookup[y2ac_qi];
s->qmat[0].chroma_qmul[0] = FFMIN(vp7_ydc_qlookup[uvdc_qi], 132);
s->qmat[0].chroma_qmul[1] = vp7_yac_qlookup[uvac_qi];
} |
3b636f21 |
|
89f2f5db |
static void vp8_get_quants(VP8Context *s) |
3b636f21 |
{
VP56RangeCoder *c = &s->c;
int i, base_qi;
int yac_qi = vp8_rac_get_uint(c, 7);
int ydc_delta = vp8_rac_get_sint(c, 4);
int y2dc_delta = vp8_rac_get_sint(c, 4);
int y2ac_delta = vp8_rac_get_sint(c, 4);
int uvdc_delta = vp8_rac_get_sint(c, 4);
int uvac_delta = vp8_rac_get_sint(c, 4);
for (i = 0; i < 4; i++) {
if (s->segmentation.enabled) {
base_qi = s->segmentation.base_quant[i];
if (!s->segmentation.absolute_vals)
base_qi += yac_qi;
} else
base_qi = yac_qi;
|
53c20f17 |
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] = vp8_dc_qlookup[av_clip_uintp2(base_qi + y2dc_delta, 7)] * 2; |
48098788 |
/* 101581>>16 is equivalent to 155/100 */ |
53c20f17 |
s->qmat[i].luma_dc_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + y2ac_delta, 7)] * 101581 >> 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)]; |
a8ab0ccc |
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); |
3b636f21 |
}
}
/**
* Determine which buffers golden and altref should be updated with after this frame.
* The spec isn't clear here, so I'm going by my understanding of what libvpx does
*
* Intra frames update all 3 references
* Inter frames update VP56_FRAME_PREVIOUS if the update_last flag is set
* If the update (golden|altref) flag is set, it's updated with the current frame
* if update_last is set, and VP56_FRAME_PREVIOUS otherwise.
* If the flag is not set, the number read means:
* 0: no update
* 1: VP56_FRAME_PREVIOUS
* 2: update golden with altref, or update altref with golden
*/
static VP56Frame ref_to_update(VP8Context *s, int update, VP56Frame ref)
{
VP56RangeCoder *c = &s->c;
if (update)
return VP56_FRAME_CURRENT;
switch (vp8_rac_get_uint(c, 2)) {
case 1:
return VP56_FRAME_PREVIOUS;
case 2:
return (ref == VP56_FRAME_GOLDEN) ? VP56_FRAME_GOLDEN2 : VP56_FRAME_GOLDEN;
}
return VP56_FRAME_NONE;
} |
89f2f5db |
static void vp78_reset_probability_tables(VP8Context *s)
{
int i, j;
for (i = 0; i < 4; i++)
for (j = 0; j < 16; j++) |
ac4b32df |
memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]], |
89f2f5db |
sizeof(s->prob->token[i][j]));
}
static void vp78_update_probability_tables(VP8Context *s)
{
VP56RangeCoder *c = &s->c;
int i, j, k, l, m;
for (i = 0; i < 4; i++)
for (j = 0; j < 8; j++)
for (k = 0; k < 3; k++) |
ac4b32df |
for (l = 0; l < NUM_DCT_TOKENS-1; l++) |
89f2f5db |
if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) {
int prob = vp8_rac_get_uint(c, 8);
for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
}
}
|
ac4b32df |
#define VP7_MVC_SIZE 17
#define VP8_MVC_SIZE 19
static void vp78_update_pred16x16_pred8x8_mvc_probabilities(VP8Context *s,
int mvc_size) |
89f2f5db |
{
VP56RangeCoder *c = &s->c;
int i, j;
if (vp8_rac_get(c))
for (i = 0; i < 4; i++)
s->prob->pred16x16[i] = vp8_rac_get_uint(c, 8);
if (vp8_rac_get(c))
for (i = 0; i < 3; i++)
s->prob->pred8x8c[i] = vp8_rac_get_uint(c, 8);
// 17.2 MV probability update
for (i = 0; i < 2; i++) |
ac4b32df |
for (j = 0; j < mvc_size; j++) |
89f2f5db |
if (vp56_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j]))
s->prob->mvc[i][j] = vp8_rac_get_nn(c);
} |
3b636f21 |
static void update_refs(VP8Context *s)
{
VP56RangeCoder *c = &s->c;
int update_golden = vp8_rac_get(c);
int update_altref = vp8_rac_get(c);
s->update_golden = ref_to_update(s, update_golden, VP56_FRAME_GOLDEN);
s->update_altref = ref_to_update(s, update_altref, VP56_FRAME_GOLDEN2);
} |
89f2f5db |
|
26e70fd5 |
static void copy_chroma(AVFrame *dst, AVFrame *src, int width, int height) |
ac4b32df |
{
int i, j;
for (j = 1; j < 3; j++) {
for (i = 0; i < height / 2; i++)
memcpy(dst->data[j] + i * dst->linesize[j],
src->data[j] + i * src->linesize[j], width / 2);
}
}
|
fb61ed1e |
static void fade(uint8_t *dst, int dst_linesize,
const uint8_t *src, int src_linesize,
int width, int height, |
ac4b32df |
int alpha, int beta) |
89f2f5db |
{
int i, j; |
ac4b32df |
for (j = 0; j < height; j++) { |
89f2f5db |
for (i = 0; i < width; i++) { |
fb61ed1e |
uint8_t y = src[j * src_linesize + i];
dst[j * dst_linesize + i] = av_clip_uint8(y + ((y * beta) >> 8) + alpha); |
89f2f5db |
} |
ac4b32df |
}
}
static int vp7_fade_frame(VP8Context *s, VP56RangeCoder *c)
{
int alpha = (int8_t) vp8_rac_get_uint(c, 8);
int beta = (int8_t) vp8_rac_get_uint(c, 8);
int ret;
if (!s->keyframe && (alpha || beta)) {
int width = s->mb_width * 16;
int height = s->mb_height * 16;
AVFrame *src, *dst;
|
fb61ed1e |
if (!s->framep[VP56_FRAME_PREVIOUS] ||
!s->framep[VP56_FRAME_GOLDEN]) {
av_log(s->avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n"); |
ac4b32df |
return AVERROR_INVALIDDATA; |
fb61ed1e |
} |
ac4b32df |
dst =
src = s->framep[VP56_FRAME_PREVIOUS]->tf.f;
/* preserve the golden frame, write a new previous frame */
if (s->framep[VP56_FRAME_GOLDEN] == s->framep[VP56_FRAME_PREVIOUS]) {
s->framep[VP56_FRAME_PREVIOUS] = vp8_find_free_buffer(s);
if ((ret = vp8_alloc_frame(s, s->framep[VP56_FRAME_PREVIOUS], 1)) < 0) |
fb61ed1e |
return ret; |
ac4b32df |
dst = s->framep[VP56_FRAME_PREVIOUS]->tf.f;
|
26e70fd5 |
copy_chroma(dst, src, width, height); |
ac4b32df |
}
|
fb61ed1e |
fade(dst->data[0], dst->linesize[0],
src->data[0], src->linesize[0],
width, height, alpha, beta); |
ac4b32df |
}
return 0; |
89f2f5db |
}
static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
{
VP56RangeCoder *c = &s->c;
int part1_size, hscale, vscale, i, j, ret;
int width = s->avctx->width;
int height = s->avctx->height;
|
7bf96472 |
if (buf_size < 4) {
return AVERROR_INVALIDDATA;
}
|
ac4b32df |
s->profile = (buf[0] >> 1) & 7; |
89f2f5db |
if (s->profile > 1) {
avpriv_request_sample(s->avctx, "Unknown profile %d", s->profile);
return AVERROR_INVALIDDATA;
}
s->keyframe = !(buf[0] & 1);
s->invisible = 0;
part1_size = AV_RL24(buf) >> 4;
|
46f72ea5 |
if (buf_size < 4 - s->profile + part1_size) {
av_log(s->avctx, AV_LOG_ERROR, "Buffer size %d is too small, needed : %d\n", buf_size, 4 - s->profile + part1_size);
return AVERROR_INVALIDDATA;
}
|
89f2f5db |
buf += 4 - s->profile;
buf_size -= 4 - s->profile;
memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
ff_vp56_init_range_decoder(c, buf, part1_size);
buf += part1_size;
buf_size -= part1_size;
/* A. Dimension information (keyframes only) */
if (s->keyframe) {
width = vp8_rac_get_uint(c, 12);
height = vp8_rac_get_uint(c, 12);
hscale = vp8_rac_get_uint(c, 2);
vscale = vp8_rac_get_uint(c, 2);
if (hscale || vscale)
avpriv_request_sample(s->avctx, "Upscaling");
s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
vp78_reset_probability_tables(s); |
ac4b32df |
memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
sizeof(s->prob->pred16x16));
memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
sizeof(s->prob->pred8x8c)); |
89f2f5db |
for (i = 0; i < 2; i++) |
ac4b32df |
memcpy(s->prob->mvc[i], vp7_mv_default_prob[i],
sizeof(vp7_mv_default_prob[i])); |
89f2f5db |
memset(&s->segmentation, 0, sizeof(s->segmentation));
memset(&s->lf_delta, 0, sizeof(s->lf_delta)); |
f4d581cd |
memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan)); |
89f2f5db |
}
if (s->keyframe || s->profile > 0)
memset(s->inter_dc_pred, 0 , sizeof(s->inter_dc_pred));
/* B. Decoding information for all four macroblock-level features */
for (i = 0; i < 4; i++) {
s->feature_enabled[i] = vp8_rac_get(c);
if (s->feature_enabled[i]) {
s->feature_present_prob[i] = vp8_rac_get_uint(c, 8);
for (j = 0; j < 3; j++) |
ac4b32df |
s->feature_index_prob[i][j] =
vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255; |
89f2f5db |
|
28f8114b |
if (vp7_feature_value_size[s->profile][i]) |
89f2f5db |
for (j = 0; j < 4; j++) |
ac4b32df |
s->feature_value[i][j] = |
fb61ed1e |
vp8_rac_get(c) ? vp8_rac_get_uint(c, vp7_feature_value_size[s->profile][i]) : 0; |
89f2f5db |
}
}
|
ac4b32df |
s->segmentation.enabled = 0; |
89f2f5db |
s->segmentation.update_map = 0; |
ac4b32df |
s->lf_delta.enabled = 0; |
89f2f5db |
s->num_coeff_partitions = 1;
ff_vp56_init_range_decoder(&s->coeff_partition[0], buf, buf_size);
if (!s->macroblocks_base || /* first frame */ |
ac4b32df |
width != s->avctx->width || height != s->avctx->height ||
(width + 15) / 16 != s->mb_width || (height + 15) / 16 != s->mb_height) {
if ((ret = vp7_update_dimensions(s, width, height)) < 0) |
89f2f5db |
return ret;
}
/* C. Dequantization indices */
vp7_get_quants(s);
/* D. Golden frame update flag (a Flag) for interframes only */
if (!s->keyframe) {
s->update_golden = vp8_rac_get(c) ? VP56_FRAME_CURRENT : VP56_FRAME_NONE;
s->sign_bias[VP56_FRAME_GOLDEN] = 0;
}
|
ac4b32df |
s->update_last = 1; |
89f2f5db |
s->update_probabilities = 1;
s->fade_present = 1;
if (s->profile > 0) {
s->update_probabilities = vp8_rac_get(c);
if (!s->update_probabilities)
s->prob[1] = s->prob[0];
if (!s->keyframe)
s->fade_present = vp8_rac_get(c);
}
/* E. Fading information for previous frame */
if (s->fade_present && vp8_rac_get(c)) { |
ac4b32df |
if ((ret = vp7_fade_frame(s ,c)) < 0)
return ret; |
89f2f5db |
}
/* F. Loop filter type */
if (!s->profile)
s->filter.simple = vp8_rac_get(c);
/* G. DCT coefficient ordering specification */
if (vp8_rac_get(c))
for (i = 1; i < 16; i++) |
f4d581cd |
s->prob[0].scan[i] = ff_zigzag_scan[vp8_rac_get_uint(c, 4)]; |
89f2f5db |
/* H. Loop filter levels */
if (s->profile > 0)
s->filter.simple = vp8_rac_get(c);
s->filter.level = vp8_rac_get_uint(c, 6);
s->filter.sharpness = vp8_rac_get_uint(c, 3);
/* I. DCT coefficient probability update; 13.3 Token Probability Updates */
vp78_update_probability_tables(s);
s->mbskip_enabled = 0;
/* J. The remaining frame header data occurs ONLY FOR INTERFRAMES */
if (!s->keyframe) {
s->prob->intra = vp8_rac_get_uint(c, 8);
s->prob->last = vp8_rac_get_uint(c, 8); |
ac4b32df |
vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP7_MVC_SIZE); |
89f2f5db |
}
return 0;
} |
3b636f21 |
|
89f2f5db |
static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size) |
3b636f21 |
{
VP56RangeCoder *c = &s->c; |
89f2f5db |
int header_size, hscale, vscale, ret; |
3b636f21 |
int width = s->avctx->width;
int height = s->avctx->height;
|
599d746e |
if (buf_size < 3) {
av_log(s->avctx, AV_LOG_ERROR, "Insufficent data (%d) for header\n", buf_size);
return AVERROR_INVALIDDATA;
}
|
3b636f21 |
s->keyframe = !(buf[0] & 1);
s->profile = (buf[0]>>1) & 7;
s->invisible = !(buf[0] & 0x10); |
06d50ca8 |
header_size = AV_RL24(buf) >> 5; |
3b636f21 |
buf += 3;
buf_size -= 3;
|
0ef1dbed |
if (s->profile > 3)
av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);
if (!s->profile) |
53c20f17 |
memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab,
sizeof(s->put_pixels_tab)); |
0ef1dbed |
else // profile 1-3 use bilinear, 4+ aren't defined so whatever |
53c20f17 |
memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab,
sizeof(s->put_pixels_tab)); |
3b636f21 |
|
53c20f17 |
if (header_size > buf_size - 7 * s->keyframe) { |
3b636f21 |
av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
return AVERROR_INVALIDDATA;
}
if (s->keyframe) { |
06d50ca8 |
if (AV_RL24(buf) != 0x2a019d) { |
53c20f17 |
av_log(s->avctx, AV_LOG_ERROR,
"Invalid start code 0x%x\n", AV_RL24(buf)); |
3b636f21 |
return AVERROR_INVALIDDATA;
} |
53c20f17 |
width = AV_RL16(buf + 3) & 0x3fff;
height = AV_RL16(buf + 5) & 0x3fff;
hscale = buf[4] >> 6;
vscale = buf[6] >> 6; |
3b636f21 |
buf += 7;
buf_size -= 7;
|
92a54426 |
if (hscale || vscale) |
12e25ed2 |
avpriv_request_sample(s->avctx, "Upscaling"); |
92a54426 |
|
3b636f21 |
s->update_golden = s->update_altref = VP56_FRAME_CURRENT; |
89f2f5db |
vp78_reset_probability_tables(s); |
53c20f17 |
memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
sizeof(s->prob->pred16x16));
memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
sizeof(s->prob->pred8x8c));
memcpy(s->prob->mvc, vp8_mv_default_prob,
sizeof(s->prob->mvc)); |
3b636f21 |
memset(&s->segmentation, 0, sizeof(s->segmentation)); |
3cc02527 |
memset(&s->lf_delta, 0, sizeof(s->lf_delta)); |
3b636f21 |
}
|
905ef0d0 |
ff_vp56_init_range_decoder(c, buf, header_size); |
3b636f21 |
buf += header_size;
buf_size -= header_size;
if (s->keyframe) { |
65875a8b |
s->colorspace = vp8_rac_get(c);
if (s->colorspace) |
3b636f21 |
av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n"); |
65875a8b |
s->fullrange = vp8_rac_get(c); |
3b636f21 |
}
if ((s->segmentation.enabled = vp8_rac_get(c)))
parse_segment_info(s);
else
s->segmentation.update_map = 0; // FIXME: move this to some init function?
s->filter.simple = vp8_rac_get(c);
s->filter.level = vp8_rac_get_uint(c, 6);
s->filter.sharpness = vp8_rac_get_uint(c, 3);
if ((s->lf_delta.enabled = vp8_rac_get(c)))
if (vp8_rac_get(c))
update_lf_deltas(s);
if (setup_partitions(s, buf, buf_size)) {
av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n");
return AVERROR_INVALIDDATA;
}
|
951455c1 |
if (!s->macroblocks_base || /* first frame */ |
ae3313e1 |
width != s->avctx->width || height != s->avctx->height ||
(width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) |
ac4b32df |
if ((ret = vp8_update_dimensions(s, width, height)) < 0) |
951455c1 |
return ret;
|
89f2f5db |
vp8_get_quants(s); |
3b636f21 |
if (!s->keyframe) {
update_refs(s);
s->sign_bias[VP56_FRAME_GOLDEN] = vp8_rac_get(c);
s->sign_bias[VP56_FRAME_GOLDEN2 /* altref */] = vp8_rac_get(c);
}
// if we aren't saving this frame's probabilities for future frames,
// make a copy of the current probabilities
if (!(s->update_probabilities = vp8_rac_get(c)))
s->prob[1] = s->prob[0];
s->update_last = s->keyframe || vp8_rac_get(c);
|
89f2f5db |
vp78_update_probability_tables(s); |
3b636f21 |
if ((s->mbskip_enabled = vp8_rac_get(c))) |
a8ab0ccc |
s->prob->mbskip = vp8_rac_get_uint(c, 8); |
3b636f21 |
if (!s->keyframe) { |
a8ab0ccc |
s->prob->intra = vp8_rac_get_uint(c, 8);
s->prob->last = vp8_rac_get_uint(c, 8);
s->prob->golden = vp8_rac_get_uint(c, 8); |
ac4b32df |
vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP8_MVC_SIZE); |
3b636f21 |
}
return 0;
}
|
53c20f17 |
static av_always_inline
void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src) |
3b636f21 |
{ |
6fdbaa2b |
dst->x = av_clip(src->x, av_clip(s->mv_min.x, INT16_MIN, INT16_MAX),
av_clip(s->mv_max.x, INT16_MIN, INT16_MAX));
dst->y = av_clip(src->y, av_clip(s->mv_min.y, INT16_MIN, INT16_MAX),
av_clip(s->mv_max.y, INT16_MIN, INT16_MAX)); |
3b636f21 |
}
/**
* Motion vector coding, 17.1.
*/ |
89f2f5db |
static av_always_inline int read_mv_component(VP56RangeCoder *c, const uint8_t *p, int vp7) |
3b636f21 |
{ |
ca18a478 |
int bit, x = 0; |
3b636f21 |
|
7697cdcf |
if (vp56_rac_get_prob_branchy(c, p[0])) { |
3b636f21 |
int i;
for (i = 0; i < 3; i++)
x += vp56_rac_get_prob(c, p[9 + i]) << i; |
89f2f5db |
for (i = (vp7 ? 7 : 9); i > 3; i--) |
3b636f21 |
x += vp56_rac_get_prob(c, p[9 + i]) << i; |
89f2f5db |
if (!(x & (vp7 ? 0xF0 : 0xFFF0)) || vp56_rac_get_prob(c, p[12])) |
3b636f21 |
x += 8; |
ca18a478 |
} else {
// small_mvtree |
53c20f17 |
const uint8_t *ps = p + 2; |
ca18a478 |
bit = vp56_rac_get_prob(c, *ps); |
53c20f17 |
ps += 1 + 3 * bit;
x += 4 * bit; |
ca18a478 |
bit = vp56_rac_get_prob(c, *ps);
ps += 1 + bit; |
53c20f17 |
x += 2 * bit; |
ca18a478 |
x += vp56_rac_get_prob(c, *ps);
} |
3b636f21 |
return (x && vp56_rac_get_prob(c, p[1])) ? -x : x;
}
|
89f2f5db |
static int vp7_read_mv_component(VP56RangeCoder *c, const uint8_t *p)
{
return read_mv_component(c, p, 1);
}
static int vp8_read_mv_component(VP56RangeCoder *c, const uint8_t *p)
{
return read_mv_component(c, p, 0);
}
|
414ac27d |
static av_always_inline |
ac4b32df |
const uint8_t *get_submv_prob(uint32_t left, uint32_t top, int is_vp7) |
3b636f21 |
{ |
ac4b32df |
if (is_vp7)
return vp7_submv_prob;
|
7bf254c4 |
if (left == top) |
53c20f17 |
return vp8_submv_prob[4 - !!left]; |
7bf254c4 |
if (!top) |
3b636f21 |
return vp8_submv_prob[2]; |
53c20f17 |
return vp8_submv_prob[1 - !!left]; |
3b636f21 |
}
/**
* Split motion vector prediction, 16.4. |
7ed06b2b |
* @returns the number of motion vectors parsed (2, 4 or 16) |
3b636f21 |
*/ |
414ac27d |
static av_always_inline |
ac4b32df |
int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
int layout, int is_vp7) |
3b636f21 |
{ |
0908f1b9 |
int part_idx;
int n, num; |
951455c1 |
VP8Macroblock *top_mb; |
7bf254c4 |
VP8Macroblock *left_mb = &mb[-1]; |
53c20f17 |
const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning];
const uint8_t *mbsplits_top, *mbsplits_cur, *firstidx; |
951455c1 |
VP56mv *top_mv; |
c55e0d34 |
VP56mv *left_mv = left_mb->bmv;
VP56mv *cur_mv = mb->bmv; |
3b636f21 |
|
951455c1 |
if (!layout) // layout is inlined, s->mb_layout is not
top_mb = &mb[2];
else |
53c20f17 |
top_mb = &mb[-s->mb_width - 1]; |
951455c1 |
mbsplits_top = vp8_mbsplits[top_mb->partitioning]; |
53c20f17 |
top_mv = top_mb->bmv; |
951455c1 |
|
0908f1b9 |
if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) { |
53c20f17 |
if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) |
0908f1b9 |
part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]); |
53c20f17 |
else |
0908f1b9 |
part_idx = VP8_SPLITMVMODE_8x8;
} else {
part_idx = VP8_SPLITMVMODE_4x4;
}
|
53c20f17 |
num = vp8_mbsplit_count[part_idx];
mbsplits_cur = vp8_mbsplits[part_idx],
firstidx = vp8_mbfirstidx[part_idx]; |
0908f1b9 |
mb->partitioning = part_idx;
|
3b636f21 |
for (n = 0; n < num; n++) { |
7ed06b2b |
int k = firstidx[n]; |
7bf254c4 |
uint32_t left, above; |
7ed06b2b |
const uint8_t *submv_prob;
|
7bf254c4 |
if (!(k & 3))
left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
else |
53c20f17 |
left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]); |
7bf254c4 |
if (k <= 3)
above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
else
above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]); |
7ed06b2b |
|
ac4b32df |
submv_prob = get_submv_prob(left, above, is_vp7); |
3b636f21 |
|
c5dec7f1 |
if (vp56_rac_get_prob_branchy(c, submv_prob[0])) {
if (vp56_rac_get_prob_branchy(c, submv_prob[1])) {
if (vp56_rac_get_prob_branchy(c, submv_prob[2])) { |
ac4b32df |
mb->bmv[n].y = mb->mv.y +
read_mv_component(c, s->prob->mvc[0], is_vp7);
mb->bmv[n].x = mb->mv.x +
read_mv_component(c, s->prob->mvc[1], is_vp7); |
c5dec7f1 |
} else {
AV_ZERO32(&mb->bmv[n]);
}
} else {
AV_WN32A(&mb->bmv[n], above);
}
} else { |
7bf254c4 |
AV_WN32A(&mb->bmv[n], left); |
3b636f21 |
}
} |
7ed06b2b |
return num; |
3b636f21 |
}
|
89f2f5db |
/** |
ac4b32df |
* The vp7 reference decoder uses a padding macroblock column (added to right
* edge of the frame) to guard against illegal macroblock offsets. The
* algorithm has bugs that permit offsets to straddle the padding column.
* This function replicates those bugs.
* |
89f2f5db |
* @param[out] edge_x macroblock x address
* @param[out] edge_y macroblock y address |
ac4b32df |
* |
89f2f5db |
* @return macroblock offset legal (boolean)
*/ |
ac4b32df |
static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width,
int xoffset, int yoffset, int boundary,
int *edge_x, int *edge_y) |
89f2f5db |
{
int vwidth = mb_width + 1;
int new = (mb_y + yoffset) * vwidth + mb_x + xoffset;
if (new < boundary || new % vwidth == vwidth - 1)
return 0;
*edge_y = new / vwidth;
*edge_x = new % vwidth;
return 1;
}
|
ac4b32df |
static const VP56mv *get_bmv_ptr(const VP8Macroblock *mb, int subblock) |
89f2f5db |
{
return &mb->bmv[mb->mode == VP8_MVMODE_SPLIT ? vp8_mbsplits[mb->partitioning][subblock] : 0];
}
|
414ac27d |
static av_always_inline |
ac4b32df |
void vp7_decode_mvs(VP8Context *s, VP8Macroblock *mb,
int mb_x, int mb_y, int layout) |
89f2f5db |
{
VP8Macroblock *mb_edge[12];
enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR };
enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
int idx = CNT_ZERO;
VP56mv near_mv[3];
uint8_t cnt[3] = { 0 };
VP56RangeCoder *c = &s->c;
int i;
AV_ZERO32(&near_mv[0]);
AV_ZERO32(&near_mv[1]);
AV_ZERO32(&near_mv[2]);
for (i = 0; i < VP7_MV_PRED_COUNT; i++) {
const VP7MVPred * pred = &vp7_mv_pred[i];
int edge_x, edge_y;
|
ac4b32df |
if (vp7_calculate_mb_offset(mb_x, mb_y, s->mb_width, pred->xoffset,
pred->yoffset, !s->profile, &edge_x, &edge_y)) {
VP8Macroblock *edge = mb_edge[i] = (s->mb_layout == 1)
? s->macroblocks_base + 1 + edge_x +
(s->mb_width + 1) * (edge_y + 1)
: s->macroblocks + edge_x +
(s->mb_height - edge_y - 1) * 2; |
89f2f5db |
uint32_t mv = AV_RN32A(get_bmv_ptr(edge, vp7_mv_pred[i].subblock));
if (mv) {
if (AV_RN32A(&near_mv[CNT_NEAREST])) {
if (mv == AV_RN32A(&near_mv[CNT_NEAREST])) {
idx = CNT_NEAREST;
} else if (AV_RN32A(&near_mv[CNT_NEAR])) {
if (mv != AV_RN32A(&near_mv[CNT_NEAR]))
continue;
idx = CNT_NEAR;
} else {
AV_WN32A(&near_mv[CNT_NEAR], mv);
idx = CNT_NEAR;
}
} else {
AV_WN32A(&near_mv[CNT_NEAREST], mv);
idx = CNT_NEAREST;
}
} else {
idx = CNT_ZERO;
}
} else {
idx = CNT_ZERO;
}
cnt[idx] += vp7_mv_pred[i].score;
}
mb->partitioning = VP8_SPLITMVMODE_NONE;
if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_ZERO]][0])) {
mb->mode = VP8_MVMODE_MV;
if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAREST]][1])) {
if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][2])) {
if (cnt[CNT_NEAREST] > cnt[CNT_NEAR])
AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : AV_RN32A(&near_mv[CNT_NEAREST]));
else
AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAR] ? 0 : AV_RN32A(&near_mv[CNT_NEAR]));
if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][3])) {
mb->mode = VP8_MVMODE_SPLIT; |
ac4b32df |
mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP7) - 1]; |
89f2f5db |
} else {
mb->mv.y += vp7_read_mv_component(c, s->prob->mvc[0]);
mb->mv.x += vp7_read_mv_component(c, s->prob->mvc[1]);
mb->bmv[0] = mb->mv;
}
} else {
mb->mv = near_mv[CNT_NEAR];
mb->bmv[0] = mb->mv;
}
} else {
mb->mv = near_mv[CNT_NEAREST];
mb->bmv[0] = mb->mv;
}
} else {
mb->mode = VP8_MVMODE_ZERO;
AV_ZERO32(&mb->mv);
mb->bmv[0] = mb->mv;
}
}
static av_always_inline |
ae3313e1 |
void vp8_decode_mvs(VP8Context *s, VP8Macroblock *mb,
int mb_x, int mb_y, int layout) |
f3d09d44 |
{ |
53c20f17 |
VP8Macroblock *mb_edge[3] = { 0 /* top */, |
f3d09d44 |
mb - 1 /* left */, |
53c20f17 |
0 /* top-left */ }; |
f3d09d44 |
enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV }; |
d375c104 |
enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT }; |
f3d09d44 |
int idx = CNT_ZERO;
int cur_sign_bias = s->sign_bias[mb->ref_frame]; |
1eeca886 |
int8_t *sign_bias = s->sign_bias; |
f3d09d44 |
VP56mv near_mv[4];
uint8_t cnt[4] = { 0 };
VP56RangeCoder *c = &s->c;
|
951455c1 |
if (!layout) { // layout is inlined (s->mb_layout is not)
mb_edge[0] = mb + 2;
mb_edge[2] = mb + 1; |
53c20f17 |
} else {
mb_edge[0] = mb - s->mb_width - 1;
mb_edge[2] = mb - s->mb_width - 2; |
951455c1 |
}
|
f3d09d44 |
AV_ZERO32(&near_mv[0]);
AV_ZERO32(&near_mv[1]); |
0f0b5d64 |
AV_ZERO32(&near_mv[2]); |
f3d09d44 |
/* Process MB on top, left and top-left */ |
53c20f17 |
#define MV_EDGE_CHECK(n) \
{ \
VP8Macroblock *edge = mb_edge[n]; \
int edge_ref = edge->ref_frame; \
if (edge_ref != VP56_FRAME_CURRENT) { \
uint32_t mv = AV_RN32A(&edge->mv); \
if (mv) { \
if (cur_sign_bias != sign_bias[edge_ref]) { \
/* SWAR negate of the values in mv. */ \
mv = ~mv; \
mv = ((mv & 0x7fff7fff) + \
0x00010001) ^ (mv & 0x80008000); \
} \
if (!n || mv != AV_RN32A(&near_mv[idx])) \
AV_WN32A(&near_mv[++idx], mv); \
cnt[idx] += 1 + (n != 2); \
} else \
cnt[CNT_ZERO] += 1 + (n != 2); \
} \ |
f3d09d44 |
}
MV_EDGE_CHECK(0)
MV_EDGE_CHECK(1)
MV_EDGE_CHECK(2)
mb->partitioning = VP8_SPLITMVMODE_NONE;
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) {
mb->mode = VP8_MVMODE_MV;
/* If we have three distinct MVs, merge first and last if they're the same */ |
53c20f17 |
if (cnt[CNT_SPLITMV] &&
AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT])) |
f3d09d44 |
cnt[CNT_NEAREST] += 1;
/* Swap near and nearest if necessary */
if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR]);
FFSWAP( VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
}
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
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 */ |
7634771e |
clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]); |
d375c104 |
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); |
f3d09d44 |
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
mb->mode = VP8_MVMODE_SPLIT; |
ac4b32df |
mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP8) - 1]; |
f3d09d44 |
} else { |
ae3313e1 |
mb->mv.y += vp8_read_mv_component(c, s->prob->mvc[0]);
mb->mv.x += vp8_read_mv_component(c, s->prob->mvc[1]); |
f3d09d44 |
mb->bmv[0] = mb->mv;
}
} else { |
7634771e |
clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR]); |
f3d09d44 |
mb->bmv[0] = mb->mv;
}
} else { |
7634771e |
clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST]); |
f3d09d44 |
mb->bmv[0] = mb->mv;
}
} else {
mb->mode = VP8_MVMODE_ZERO;
AV_ZERO32(&mb->mv);
mb->bmv[0] = mb->mv;
}
}
static av_always_inline |
17343e39 |
void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb, |
951455c1 |
int mb_x, int keyframe, int layout) |
3b636f21 |
{ |
17343e39 |
uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
|
f29cdbe1 |
if (layout) { |
951455c1 |
VP8Macroblock *mb_top = mb - s->mb_width - 1;
memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4);
} |
d1c58fce |
if (keyframe) { |
d2840fa4 |
int x, y; |
53c20f17 |
uint8_t *top;
uint8_t *const left = s->intra4x4_pred_mode_left; |
f29cdbe1 |
if (layout) |
951455c1 |
top = mb->intra4x4_pred_mode_top;
else
top = s->intra4x4_pred_mode_top + 4 * mb_x; |
d1c58fce |
for (y = 0; y < 4; y++) {
for (x = 0; x < 4; x++) { |
d2840fa4 |
const uint8_t *ctx; |
53c20f17 |
ctx = vp8_pred4x4_prob_intra[top[x]][left[y]]; |
d2840fa4 |
*intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx); |
53c20f17 |
left[y] = top[x] = *intra4x4; |
d2840fa4 |
intra4x4++; |
3b636f21 |
}
} |
d1c58fce |
} else { |
d2840fa4 |
int i; |
d1c58fce |
for (i = 0; i < 16; i++) |
53c20f17 |
intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree,
vp8_pred4x4_prob_inter); |
3b636f21 |
}
}
|
414ac27d |
static av_always_inline |
951455c1 |
void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, |
ac4b32df |
uint8_t *segment, uint8_t *ref, int layout, int is_vp7) |
3b636f21 |
{
VP56RangeCoder *c = &s->c; |
5afb94c8 |
static const char *vp7_feature_name[] = { "q-index",
"lf-delta",
"partial-golden-update",
"blit-pitch" }; |
ac4b32df |
if (is_vp7) { |
89f2f5db |
int i;
*segment = 0;
for (i = 0; i < 4; i++) {
if (s->feature_enabled[i]) { |
cef99e12 |
if (vp56_rac_get_prob_branchy(c, s->feature_present_prob[i])) { |
ac4b32df |
int index = vp8_rac_get_tree(c, vp7_feature_index_tree,
s->feature_index_prob[i]);
av_log(s->avctx, AV_LOG_WARNING,
"Feature %s present in macroblock (value 0x%x)\n",
vp7_feature_name[i], s->feature_value[i][index]); |
89f2f5db |
} |
ac4b32df |
} |
89f2f5db |
} |
fb61ed1e |
} else if (s->segmentation.update_map) {
int bit = vp56_rac_get_prob(c, s->prob->segmentid[0]);
*segment = vp56_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit;
} else if (s->segmentation.enabled) |
4773d904 |
*segment = ref ? *ref : *segment; |
17343e39 |
mb->segment = *segment; |
3b636f21 |
|
a8ab0ccc |
mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0; |
3b636f21 |
if (s->keyframe) { |
53c20f17 |
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra,
vp8_pred16x16_prob_intra); |
3b636f21 |
if (mb->mode == MODE_I4x4) { |
951455c1 |
decode_intra4x4_modes(s, c, mb, mb_x, 1, layout); |
d2840fa4 |
} else { |
ac4b32df |
const uint32_t modes = (is_vp7 ? vp7_pred4x4_mode
: vp8_pred4x4_mode)[mb->mode] * 0x01010101u; |
f29cdbe1 |
if (s->mb_layout) |
951455c1 |
AV_WN32A(mb->intra4x4_pred_mode_top, modes);
else
AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes); |
53c20f17 |
AV_WN32A(s->intra4x4_pred_mode_left, modes); |
d2840fa4 |
} |
3b636f21 |
|
53c20f17 |
mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree,
vp8_pred8x8c_prob_intra);
mb->ref_frame = VP56_FRAME_CURRENT; |
a8ab0ccc |
} else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) { |
3b636f21 |
// inter MB, 16.2 |
a8ab0ccc |
if (vp56_rac_get_prob_branchy(c, s->prob->last)) |
53c20f17 |
mb->ref_frame = |
ac4b32df |
(!is_vp7 && vp56_rac_get_prob(c, s->prob->golden)) ? VP56_FRAME_GOLDEN2 /* altref */
: VP56_FRAME_GOLDEN; |
3b636f21 |
else
mb->ref_frame = VP56_FRAME_PREVIOUS; |
53c20f17 |
s->ref_count[mb->ref_frame - 1]++; |
3b636f21 |
// motion vectors, 16.3 |
ac4b32df |
if (is_vp7) |
89f2f5db |
vp7_decode_mvs(s, mb, mb_x, mb_y, layout);
else
vp8_decode_mvs(s, mb, mb_x, mb_y, layout); |
3b636f21 |
} else {
// intra MB, 16.1
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
|
158e062c |
if (mb->mode == MODE_I4x4) |
951455c1 |
decode_intra4x4_modes(s, c, mb, mb_x, 0, layout); |
3b636f21 |
|
53c20f17 |
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; |
14767f35 |
AV_ZERO32(&mb->bmv[0]); |
3b636f21 |
}
}
/** |
53c20f17 |
* @param r arithmetic bitstream reader context |
e394953e |
* @param block destination for block coefficients
* @param probs probabilities to use when reading trees from the bitstream |
53c20f17 |
* @param i initial coeff index, 0 unless a separate DC block is coded
* @param qmul array holding the dc/ac dequant factor at position 0/1
* |
3b636f21 |
* @return 0 if no coeffs were decoded
* otherwise, the index of the last coeff decoded plus one
*/ |
89f2f5db |
static av_always_inline
int decode_block_coeffs_internal(VP56RangeCoder *r, int16_t block[16], |
ac4b32df |
uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
int i, uint8_t *token_prob, int16_t qmul[2],
const uint8_t scan[16], int vp7) |
3b636f21 |
{ |
6163d880 |
VP56RangeCoder c = *r; |
afb54a85 |
goto skip_eob; |
fe1b5d97 |
do { |
1e739679 |
int coeff; |
89f2f5db |
restart: |
6163d880 |
if (!vp56_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB
break; |
3b636f21 |
|
fe1b5d97 |
skip_eob: |
6163d880 |
if (!vp56_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0 |
c22b4468 |
if (++i == 16) |
6163d880 |
break; // invalid input; blocks should end with EOB |
370b622a |
token_prob = probs[i][0]; |
89f2f5db |
if (vp7)
goto restart; |
c22b4468 |
goto skip_eob; |
fe1b5d97 |
}
|
6163d880 |
if (!vp56_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1 |
fe1b5d97 |
coeff = 1; |
53c20f17 |
token_prob = probs[i + 1][1]; |
fe1b5d97 |
} else { |
6163d880 |
if (!vp56_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
coeff = vp56_rac_get_prob_branchy(&c, token_prob[4]); |
fe1b5d97 |
if (coeff) |
6163d880 |
coeff += vp56_rac_get_prob(&c, token_prob[5]); |
fe1b5d97 |
coeff += 2;
} else {
// DCT_CAT* |
6163d880 |
if (!vp56_rac_get_prob_branchy(&c, token_prob[6])) {
if (!vp56_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1 |
53c20f17 |
coeff = 5 + vp56_rac_get_prob(&c, vp8_dct_cat1_prob[0]); |
fe1b5d97 |
} else { // DCT_CAT2
coeff = 7; |
6163d880 |
coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[1]); |
fe1b5d97 |
}
} else { // DCT_CAT3 and up |
53c20f17 |
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); |
6163d880 |
coeff += vp8_rac_get_coeff(&c, ff_vp8_dct_cat_prob[cat]); |
fe1b5d97 |
}
} |
53c20f17 |
token_prob = probs[i + 1][2]; |
fe1b5d97 |
} |
89f2f5db |
block[scan[i]] = (vp8_rac_get(&c) ? -coeff : coeff) * qmul[!!i]; |
afb54a85 |
} while (++i < 16); |
fe1b5d97 |
|
6163d880 |
*r = c; |
afb54a85 |
return i; |
3b636f21 |
} |
89f2f5db |
|
ac4b32df |
static av_always_inline
int inter_predict_dc(int16_t block[16], int16_t pred[2]) |
89f2f5db |
{ |
ac4b32df |
int16_t dc = block[0];
int ret = 0;
if (pred[1] > 3) {
dc += pred[0];
ret = 1;
}
if (!pred[0] | !dc | ((int32_t)pred[0] ^ (int32_t)dc) >> 31) {
block[0] = pred[0] = dc;
pred[1] = 0;
} else {
if (pred[0] == dc)
pred[1]++;
block[0] = pred[0] = dc;
}
return ret;
}
static int vp7_decode_block_coeffs_internal(VP56RangeCoder *r,
int16_t block[16],
uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
int i, uint8_t *token_prob,
int16_t qmul[2],
const uint8_t scan[16])
{
return decode_block_coeffs_internal(r, block, probs, i,
token_prob, qmul, scan, IS_VP7); |
89f2f5db |
}
#ifndef vp8_decode_block_coeffs_internal |
ac4b32df |
static int vp8_decode_block_coeffs_internal(VP56RangeCoder *r,
int16_t block[16],
uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
int i, uint8_t *token_prob,
int16_t qmul[2]) |
89f2f5db |
{ |
ac4b32df |
return decode_block_coeffs_internal(r, block, probs, i, |
f4d581cd |
token_prob, qmul, ff_zigzag_scan, IS_VP8); |
89f2f5db |
} |
a7878c9f |
#endif |
3b636f21 |
|
3c432e11 |
/** |
53c20f17 |
* @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 |
3c432e11 |
* @param zero_nhood the initial prediction context for number of surrounding
* all-zero blocks (only left/top, so 0-2) |
53c20f17 |
* @param qmul array holding the dc/ac dequant factor at position 0/1 |
ae3313e1 |
* @param scan scan pattern (VP7 only) |
53c20f17 |
* |
3c432e11 |
* @return 0 if no coeffs were decoded
* otherwise, the index of the last coeff decoded plus one
*/ |
414ac27d |
static av_always_inline |
88bd7fdc |
int decode_block_coeffs(VP56RangeCoder *c, int16_t block[16], |
53c20f17 |
uint8_t probs[16][3][NUM_DCT_TOKENS - 1], |
89f2f5db |
int i, int zero_nhood, int16_t qmul[2],
const uint8_t scan[16], int vp7) |
1e739679 |
{
uint8_t *token_prob = probs[i][zero_nhood];
if (!vp56_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
return 0; |
ac4b32df |
return vp7 ? vp7_decode_block_coeffs_internal(c, block, probs, i,
token_prob, qmul, scan)
: vp8_decode_block_coeffs_internal(c, block, probs, i,
token_prob, qmul); |
1e739679 |
}
static av_always_inline |
53c20f17 |
void decode_mb_coeffs(VP8Context *s, VP8ThreadData *td, VP56RangeCoder *c, |
ae3313e1 |
VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9], |
ac4b32df |
int is_vp7) |
3b636f21 |
{
int i, x, y, luma_start = 0, luma_ctx = 3;
int nnz_pred, nnz, nnz_total = 0; |
17343e39 |
int segment = mb->segment; |
f311208c |
int block_dc = 0; |
3b636f21 |
|
ac4b32df |
if (mb->mode != MODE_I4x4 && (is_vp7 || mb->mode != VP8_MVMODE_SPLIT)) { |
3b636f21 |
nnz_pred = t_nnz[8] + l_nnz[8];
// decode DC values and do hadamard |
53c20f17 |
nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0, |
ae3313e1 |
nnz_pred, s->qmat[segment].luma_dc_qmul, |
f4d581cd |
ff_zigzag_scan, is_vp7); |
3b636f21 |
l_nnz[8] = t_nnz[8] = !!nnz; |
89f2f5db |
|
ac4b32df |
if (is_vp7 && mb->mode > MODE_I4x4) {
nnz |= inter_predict_dc(td->block_dc,
s->inter_dc_pred[mb->ref_frame - 1]);
} |
89f2f5db |
|
f311208c |
if (nnz) {
nnz_total += nnz; |
53c20f17 |
block_dc = 1; |
f311208c |
if (nnz == 1) |
951455c1 |
s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc); |
f311208c |
else |
951455c1 |
s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc); |
f311208c |
} |
3b636f21 |
luma_start = 1; |
53c20f17 |
luma_ctx = 0; |
3b636f21 |
}
// luma blocks
for (y = 0; y < 4; y++)
for (x = 0; x < 4; x++) { |
ffbf0794 |
nnz_pred = l_nnz[y] + t_nnz[x]; |
53c20f17 |
nnz = decode_block_coeffs(c, td->block[y][x],
s->prob->token[luma_ctx],
luma_start, nnz_pred, |
ae3313e1 |
s->qmat[segment].luma_qmul, |
ac4b32df |
s->prob[0].scan, is_vp7); |
53c20f17 |
/* nnz+block_dc may be one more than the actual last index,
* but we don't care */ |
951455c1 |
td->non_zero_count_cache[y][x] = nnz + block_dc; |
3b636f21 |
t_nnz[x] = l_nnz[y] = !!nnz;
nnz_total += nnz;
}
// chroma blocks
// TODO: what to do about dimensions? 2nd dim for luma is x,
// but for chroma it's (y<<1)|x
for (i = 4; i < 6; i++)
for (y = 0; y < 2; y++)
for (x = 0; x < 2; x++) { |
53c20f17 |
nnz_pred = l_nnz[i + 2 * y] + t_nnz[i + 2 * x]; |
ac4b32df |
nnz = decode_block_coeffs(c, td->block[i][(y << 1) + x],
s->prob->token[2], 0, nnz_pred,
s->qmat[segment].chroma_qmul,
s->prob[0].scan, is_vp7); |
53c20f17 |
td->non_zero_count_cache[i][(y << 1) + x] = nnz;
t_nnz[i + 2 * x] = l_nnz[i + 2 * y] = !!nnz; |
ac4b32df |
nnz_total += nnz; |
3b636f21 |
}
// if there were no coded coeffs despite the macroblock not being marked skip,
// we MUST not do the inner loop filter and should not do IDCT
// Since skip isn't used for bitstream prediction, just manually set it.
if (!nnz_total)
mb->skip = 1;
}
|
9ac831c2 |
static av_always_inline |
53c20f17 |
void backup_mb_border(uint8_t *top_border, uint8_t *src_y,
uint8_t *src_cb, uint8_t *src_cr, |
9ac831c2 |
int linesize, int uvlinesize, int simple)
{ |
53c20f17 |
AV_COPY128(top_border, src_y + 15 * linesize); |
9ac831c2 |
if (!simple) { |
53c20f17 |
AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize); |
9ac831c2 |
}
}
static av_always_inline |
53c20f17 |
void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb,
uint8_t *src_cr, int linesize, int uvlinesize, int mb_x,
int mb_y, int mb_width, int simple, int xchg) |
9ac831c2 |
{ |
53c20f17 |
uint8_t *top_border_m1 = top_border - 32; // for TL prediction
src_y -= linesize; |
9ac831c2 |
src_cb -= uvlinesize;
src_cr -= uvlinesize;
|
53c20f17 |
#define XCHG(a, b, xchg) \
do { \
if (xchg) \
AV_SWAP64(b, a); \
else \
AV_COPY64(b, a); \ |
096971e8 |
} while (0) |
9ac831c2 |
|
53c20f17 |
XCHG(top_border_m1 + 8, src_y - 8, xchg);
XCHG(top_border, src_y, xchg);
XCHG(top_border + 8, src_y + 8, 1);
if (mb_x < mb_width - 1)
XCHG(top_border + 32, src_y + 16, 1); |
070ce7ef |
|
9ac831c2 |
// only copy chroma for normal loop filter
// or to initialize the top row to 127
if (!simple || !mb_y) { |
53c20f17 |
XCHG(top_border_m1 + 16, src_cb - 8, xchg);
XCHG(top_border_m1 + 24, src_cr - 8, xchg);
XCHG(top_border + 16, src_cb, 1);
XCHG(top_border + 24, src_cr, 1); |
9ac831c2 |
}
}
|
414ac27d |
static av_always_inline |
ee555de7 |
int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
{ |
53c20f17 |
if (!mb_x) |
ee555de7 |
return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8; |
53c20f17 |
else |
ee555de7 |
return mb_y ? mode : LEFT_DC_PRED8x8;
}
static av_always_inline |
89f2f5db |
int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7) |
ee555de7 |
{ |
53c20f17 |
if (!mb_x) |
89f2f5db |
return mb_y ? VERT_PRED8x8 : (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8); |
53c20f17 |
else |
ee555de7 |
return mb_y ? mode : HOR_PRED8x8;
}
static av_always_inline |
89f2f5db |
int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7) |
ee555de7 |
{
switch (mode) {
case DC_PRED8x8:
return check_dc_pred8x8_mode(mode, mb_x, mb_y);
case VERT_PRED8x8: |
89f2f5db |
return !mb_y ? (vp7 ? DC_128_PRED8x8 : DC_127_PRED8x8) : mode; |
ee555de7 |
case HOR_PRED8x8: |
89f2f5db |
return !mb_x ? (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8) : mode; |
53c20f17 |
case PLANE_PRED8x8: /* TM */ |
89f2f5db |
return check_tm_pred8x8_mode(mode, mb_x, mb_y, vp7); |
ee555de7 |
}
return mode;
}
static av_always_inline |
89f2f5db |
int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7) |
ee555de7 |
{
if (!mb_x) { |
89f2f5db |
return mb_y ? VERT_VP8_PRED : (vp7 ? DC_128_PRED : DC_129_PRED); |
ee555de7 |
} else {
return mb_y ? mode : HOR_VP8_PRED;
}
}
static av_always_inline |
ac4b32df |
int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y,
int *copy_buf, int vp7) |
ee555de7 |
{
switch (mode) {
case VERT_PRED:
if (!mb_x && mb_y) {
*copy_buf = 1;
return mode;
}
/* fall-through */
case DIAG_DOWN_LEFT_PRED:
case VERT_LEFT_PRED: |
89f2f5db |
return !mb_y ? (vp7 ? DC_128_PRED : DC_127_PRED) : mode; |
ee555de7 |
case HOR_PRED:
if (!mb_y) {
*copy_buf = 1;
return mode; |
a71abb71 |
} |
ee555de7 |
/* fall-through */
case HOR_UP_PRED: |
89f2f5db |
return !mb_x ? (vp7 ? DC_128_PRED : DC_129_PRED) : mode; |
ee555de7 |
case TM_VP8_PRED: |
89f2f5db |
return check_tm_pred4x4_mode(mode, mb_x, mb_y, vp7); |
53c20f17 |
case DC_PRED: /* 4x4 DC doesn't use the same "H.264-style" exceptions
* as 16x16/8x8 DC */ |
ee555de7 |
case DIAG_DOWN_RIGHT_PRED:
case VERT_RIGHT_PRED:
case HOR_DOWN_PRED:
if (!mb_y || !mb_x)
*copy_buf = 1;
return mode; |
3b636f21 |
}
return mode;
}
|
414ac27d |
static av_always_inline |
951455c1 |
void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], |
ac4b32df |
VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7) |
3b636f21 |
{ |
bb591566 |
int x, y, mode, nnz;
uint32_t tr; |
3b636f21 |
|
53c20f17 |
/* 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 */ |
ef8c93e2 |
if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0) |
53c20f17 |
xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2], |
9ac831c2 |
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 1);
|
3b636f21 |
if (mb->mode < MODE_I4x4) { |
ac4b32df |
mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y, is_vp7); |
3b636f21 |
s->hpc.pred16x16[mode](dst[0], s->linesize);
} else {
uint8_t *ptr = dst[0]; |
17343e39 |
uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb; |
ac4b32df |
const uint8_t lo = is_vp7 ? 128 : 127;
const uint8_t hi = is_vp7 ? 128 : 129; |
89f2f5db |
uint8_t tr_top[4] = { lo, lo, lo, lo }; |
3b636f21 |
// all blocks on the right edge of the macroblock use bottom edge
// the top macroblock for their topright edge
uint8_t *tr_right = ptr - s->linesize + 16;
// if we're on the right edge of the frame, said edge is extended
// from the top macroblock |
53c20f17 |
if (mb_y && mb_x == s->mb_width - 1) {
tr = tr_right[-1] * 0x01010101u;
tr_right = (uint8_t *) &tr; |
3b636f21 |
}
|
b74f70d6 |
if (mb->skip) |
951455c1 |
AV_ZERO128(td->non_zero_count_cache); |
b74f70d6 |
|
3b636f21 |
for (y = 0; y < 4; y++) {
uint8_t *topright = ptr + 4 - s->linesize;
for (x = 0; x < 4; x++) { |
ee555de7 |
int copy = 0, linesize = s->linesize; |
53c20f17 |
uint8_t *dst = ptr + 4 * x; |
3814f92f |
LOCAL_ALIGNED(4, uint8_t, copy_dst, [5 * 8]); |
ee555de7 |
|
ef8c93e2 |
if ((y == 0 || x == 3) && mb_y == 0) { |
ee555de7 |
topright = tr_top;
} else if (x == 3) |
3b636f21 |
topright = tr_right;
|
ac4b32df |
mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x,
mb_y + y, ©, is_vp7); |
ef8c93e2 |
if (copy) { |
53c20f17 |
dst = copy_dst + 12; |
ef8c93e2 |
linesize = 8;
if (!(mb_y + y)) { |
89f2f5db |
copy_dst[3] = lo; |
ae3313e1 |
AV_WN32A(copy_dst + 4, lo * 0x01010101U); |
ef8c93e2 |
} else { |
53c20f17 |
AV_COPY32(copy_dst + 4, ptr + 4 * x - s->linesize); |
ee555de7 |
if (!(mb_x + x)) { |
89f2f5db |
copy_dst[3] = hi; |
ee555de7 |
} else { |
53c20f17 |
copy_dst[3] = ptr[4 * x - s->linesize - 1]; |
ee555de7 |
}
} |
ef8c93e2 |
if (!(mb_x + x)) {
copy_dst[11] =
copy_dst[19] =
copy_dst[27] = |
89f2f5db |
copy_dst[35] = hi; |
ef8c93e2 |
} else { |
53c20f17 |
copy_dst[11] = ptr[4 * x - 1];
copy_dst[19] = ptr[4 * x + s->linesize - 1];
copy_dst[27] = ptr[4 * x + s->linesize * 2 - 1];
copy_dst[35] = ptr[4 * x + s->linesize * 3 - 1]; |
ef8c93e2 |
} |
ee555de7 |
}
s->hpc.pred4x4[mode](dst, topright, linesize);
if (copy) { |
53c20f17 |
AV_COPY32(ptr + 4 * x, copy_dst + 12);
AV_COPY32(ptr + 4 * x + s->linesize, copy_dst + 20);
AV_COPY32(ptr + 4 * x + s->linesize * 2, copy_dst + 28);
AV_COPY32(ptr + 4 * x + s->linesize * 3, copy_dst + 36); |
ee555de7 |
} |
3b636f21 |
|
951455c1 |
nnz = td->non_zero_count_cache[y][x]; |
3b636f21 |
if (nnz) {
if (nnz == 1) |
53c20f17 |
s->vp8dsp.vp8_idct_dc_add(ptr + 4 * x,
td->block[y][x], s->linesize); |
3b636f21 |
else |
53c20f17 |
s->vp8dsp.vp8_idct_add(ptr + 4 * x,
td->block[y][x], s->linesize); |
3b636f21 |
}
topright += 4;
}
|
53c20f17 |
ptr += 4 * s->linesize; |
d2840fa4 |
intra4x4 += 4; |
3b636f21 |
}
}
|
ac4b32df |
mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode,
mb_x, mb_y, is_vp7); |
3b636f21 |
s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
s->hpc.pred8x8[mode](dst[2], s->uvlinesize); |
9ac831c2 |
|
ef8c93e2 |
if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0) |
53c20f17 |
xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2], |
9ac831c2 |
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 0); |
3b636f21 |
}
|
64233e70 |
static const uint8_t subpel_idx[3][8] = {
{ 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
// also function pointer index
{ 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
{ 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
};
|
3b636f21 |
/** |
3c432e11 |
* luma MC function |
3b636f21 |
* |
53c20f17 |
* @param s VP8 decoding context
* @param dst target buffer for block data at block position
* @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 |
3b636f21 |
* @param linesize size of a single line of plane data, including padding |
53c20f17 |
* @param mc_func motion compensation function pointers (bilinear or sixtap MC) |
3b636f21 |
*/ |
414ac27d |
static av_always_inline |
951455c1 |
void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, |
759001c5 |
ThreadFrame *ref, const VP56mv *mv, |
64233e70 |
int x_off, int y_off, int block_w, int block_h, |
c341f734 |
int width, int height, ptrdiff_t linesize, |
64233e70 |
vp8_mc_func mc_func[3][3]) |
3b636f21 |
{ |
759001c5 |
uint8_t *src = ref->f->data[0]; |
4773d904 |
|
c0498b30 |
if (AV_RN32A(mv)) { |
face578d |
int src_linesize = linesize; |
64233e70 |
|
9b2a964c |
int mx = (mv->x * 2) & 7, mx_idx = subpel_idx[0][mx];
int my = (mv->y * 2) & 7, my_idx = subpel_idx[0][my]; |
64233e70 |
x_off += mv->x >> 2;
y_off += mv->y >> 2; |
c0498b30 |
// edge emulation |
4773d904 |
ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4, 0); |
c0498b30 |
src += y_off * linesize + x_off; |
64233e70 |
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]) { |
458446ac |
s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
src - my_idx * linesize - mx_idx, |
e46ad30a |
EDGE_EMU_LINESIZE, linesize, |
face578d |
block_w + subpel_idx[1][mx],
block_h + subpel_idx[1][my], |
53c20f17 |
x_off - mx_idx, y_off - my_idx,
width, height); |
e46ad30a |
src = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
src_linesize = EDGE_EMU_LINESIZE; |
c0498b30 |
} |
face578d |
mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my); |
4773d904 |
} else {
ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0); |
53c20f17 |
mc_func[0][0](dst, linesize, src + y_off * linesize + x_off,
linesize, block_h, 0, 0); |
4773d904 |
} |
3b636f21 |
}
|
3c432e11 |
/**
* chroma MC function
* |
53c20f17 |
* @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 |
3c432e11 |
* @param linesize size of a single line of plane data, including padding |
53c20f17 |
* @param mc_func motion compensation function pointers (bilinear or sixtap MC) |
3c432e11 |
*/ |
414ac27d |
static av_always_inline |
53c20f17 |
void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1,
uint8_t *dst2, ThreadFrame *ref, const VP56mv *mv,
int x_off, int y_off, int block_w, int block_h,
int width, int height, ptrdiff_t linesize, |
64233e70 |
vp8_mc_func mc_func[3][3])
{ |
759001c5 |
uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2]; |
4773d904 |
|
64233e70 |
if (AV_RN32A(mv)) { |
53c20f17 |
int mx = mv->x & 7, mx_idx = subpel_idx[0][mx];
int my = mv->y & 7, my_idx = subpel_idx[0][my]; |
64233e70 |
x_off += mv->x >> 3;
y_off += mv->y >> 3;
// edge emulation
src1 += y_off * linesize + x_off;
src2 += y_off * linesize + x_off; |
4773d904 |
ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0); |
64233e70 |
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]) { |
458446ac |
s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
src1 - my_idx * linesize - mx_idx, |
e46ad30a |
EDGE_EMU_LINESIZE, linesize, |
face578d |
block_w + subpel_idx[1][mx],
block_h + subpel_idx[1][my], |
8c53d39e |
x_off - mx_idx, y_off - my_idx, width, height); |
e46ad30a |
src1 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
mc_func[my_idx][mx_idx](dst1, linesize, src1, EDGE_EMU_LINESIZE, block_h, mx, my); |
64233e70 |
|
458446ac |
s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
src2 - my_idx * linesize - mx_idx, |
e46ad30a |
EDGE_EMU_LINESIZE, linesize, |
face578d |
block_w + subpel_idx[1][mx],
block_h + subpel_idx[1][my], |
8c53d39e |
x_off - mx_idx, y_off - my_idx, width, height); |
9a082fec |
src2 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx; |
e46ad30a |
mc_func[my_idx][mx_idx](dst2, linesize, src2, EDGE_EMU_LINESIZE, block_h, mx, my); |
64233e70 |
} 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 { |
4773d904 |
ff_thread_await_progress(ref, (3 + y_off + block_h) >> 3, 0); |
64233e70 |
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 |
951455c1 |
void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], |
759001c5 |
ThreadFrame *ref_frame, int x_off, int y_off, |
53c20f17 |
int bx_off, int by_off, int block_w, int block_h, |
414ac27d |
int width, int height, VP56mv *mv) |
7c4dcf81 |
{
VP56mv uvmv = *mv;
/* Y */ |
951455c1 |
vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off, |
4773d904 |
ref_frame, mv, x_off + bx_off, y_off + by_off, |
64233e70 |
block_w, block_h, width, height, s->linesize,
s->put_pixels_tab[block_w == 8]); |
7c4dcf81 |
/* U/V */
if (s->profile == 3) { |
ac4b32df |
/* this block only applies VP8; it is safe to check
* only the profile, as VP7 profile <= 1 */ |
7c4dcf81 |
uvmv.x &= ~7;
uvmv.y &= ~7;
} |
53c20f17 |
x_off >>= 1;
y_off >>= 1;
bx_off >>= 1;
by_off >>= 1;
width >>= 1;
height >>= 1;
block_w >>= 1;
block_h >>= 1; |
951455c1 |
vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off, |
4773d904 |
dst[2] + by_off * s->uvlinesize + bx_off, ref_frame,
&uvmv, x_off + bx_off, y_off + by_off, |
64233e70 |
block_w, block_h, width, height, s->uvlinesize,
s->put_pixels_tab[1 + (block_w == 4)]); |
7c4dcf81 |
}
|
d864dee8 |
/* Fetch pixels for estimated mv 4 macroblocks ahead. |
53c20f17 |
* Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */
static av_always_inline
void prefetch_motion(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
int mb_xy, int ref) |
d864dee8 |
{ |
ef38842f |
/* Don't prefetch refs that haven't been used very often this frame. */ |
53c20f17 |
if (s->ref_count[ref - 1] > (mb_xy >> 5)) { |
c4211046 |
int x_off = mb_x << 4, y_off = mb_y << 4; |
53c20f17 |
int mx = (mb->mv.x >> 2) + x_off + 8;
int my = (mb->mv.y >> 2) + y_off;
uint8_t **src = s->framep[ref]->tf.f->data;
int off = mx + (my + (mb_x & 3) * 4) * s->linesize + 64; |
4773d904 |
/* 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. */ |
53c20f17 |
s->vdsp.prefetch(src[0] + off, s->linesize, 4);
off = (mx >> 1) + ((my >> 1) + (mb_x & 7)) * s->uvlinesize + 64;
s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2); |
c4211046 |
} |
d864dee8 |
}
|
3b636f21 |
/**
* Apply motion vectors to prediction buffer, chapter 18.
*/ |
414ac27d |
static av_always_inline |
951455c1 |
void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
VP8Macroblock *mb, int mb_x, int mb_y) |
3b636f21 |
{
int x_off = mb_x << 4, y_off = mb_y << 4; |
53c20f17 |
int width = 16 * s->mb_width, height = 16 * s->mb_height; |
759001c5 |
ThreadFrame *ref = &s->framep[mb->ref_frame]->tf; |
d292c345 |
VP56mv *bmv = mb->bmv; |
3b636f21 |
|
73be29b0 |
switch (mb->partitioning) {
case VP8_SPLITMVMODE_NONE: |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
7c4dcf81 |
0, 0, 16, 16, width, height, &mb->mv); |
73be29b0 |
break; |
7c4dcf81 |
case VP8_SPLITMVMODE_4x4: { |
3b636f21 |
int x, y; |
7c4dcf81 |
VP56mv uvmv; |
3b636f21 |
/* Y */
for (y = 0; y < 4; y++) {
for (x = 0; x < 4; x++) { |
53c20f17 |
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, |
64233e70 |
width, height, s->linesize,
s->put_pixels_tab[2]); |
3b636f21 |
}
}
/* U/V */ |
53c20f17 |
x_off >>= 1;
y_off >>= 1;
width >>= 1;
height >>= 1; |
3b636f21 |
for (y = 0; y < 2; y++) {
for (x = 0; x < 2; x++) { |
53c20f17 |
uvmv.x = mb->bmv[2 * y * 4 + 2 * x ].x +
mb->bmv[2 * y * 4 + 2 * x + 1].x +
mb->bmv[(2 * y + 1) * 4 + 2 * x ].x +
mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].x;
uvmv.y = mb->bmv[2 * y * 4 + 2 * x ].y +
mb->bmv[2 * y * 4 + 2 * x + 1].y +
mb->bmv[(2 * y + 1) * 4 + 2 * x ].y +
mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].y; |
85dc006b |
uvmv.x = (uvmv.x + 2 + FF_SIGNBIT(uvmv.x)) >> 2;
uvmv.y = (uvmv.y + 2 + FF_SIGNBIT(uvmv.y)) >> 2; |
3b636f21 |
if (s->profile == 3) {
uvmv.x &= ~7;
uvmv.y &= ~7;
} |
53c20f17 |
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, |
64233e70 |
width, height, s->uvlinesize,
s->put_pixels_tab[2]); |
3b636f21 |
}
} |
7c4dcf81 |
break;
}
case VP8_SPLITMVMODE_16x8: |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
0, 0, 16, 8, width, height, &bmv[0]); |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
0, 8, 16, 8, width, height, &bmv[1]); |
7c4dcf81 |
break;
case VP8_SPLITMVMODE_8x16: |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
0, 0, 8, 16, width, height, &bmv[0]); |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
8, 0, 8, 16, width, height, &bmv[1]); |
7c4dcf81 |
break;
case VP8_SPLITMVMODE_8x8: |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
0, 0, 8, 8, width, height, &bmv[0]); |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
8, 0, 8, 8, width, height, &bmv[1]); |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
0, 8, 8, 8, width, height, &bmv[2]); |
951455c1 |
vp8_mc_part(s, td, dst, ref, x_off, y_off, |
d292c345 |
8, 8, 8, 8, width, height, &bmv[3]); |
7c4dcf81 |
break; |
3b636f21 |
}
}
|
53c20f17 |
static av_always_inline
void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], VP8Macroblock *mb) |
3b636f21 |
{ |
3df56f41 |
int x, y, ch; |
3b636f21 |
|
8a467b2d |
if (mb->mode != MODE_I4x4) {
uint8_t *y_dst = dst[0]; |
3b636f21 |
for (y = 0; y < 4; y++) { |
951455c1 |
uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]); |
3df56f41 |
if (nnz4) { |
53c20f17 |
if (nnz4 & ~0x01010101) { |
8a467b2d |
for (x = 0; x < 4; x++) { |
53c20f17 |
if ((uint8_t) nnz4 == 1)
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,
td->block[y][x],
s->linesize); |
62457f90 |
nnz4 >>= 8;
if (!nnz4)
break; |
8a467b2d |
}
} else { |
951455c1 |
s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize); |
3b636f21 |
}
} |
53c20f17 |
y_dst += 4 * s->linesize; |
3b636f21 |
} |
8a467b2d |
} |
3b636f21 |
|
8a467b2d |
for (ch = 0; ch < 2; ch++) { |
53c20f17 |
uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4 + ch]); |
3ae079a3 |
if (nnz4) { |
53c20f17 |
uint8_t *ch_dst = dst[1 + ch];
if (nnz4 & ~0x01010101) { |
3ae079a3 |
for (y = 0; y < 2; y++) {
for (x = 0; x < 2; x++) { |
53c20f17 |
if ((uint8_t) nnz4 == 1)
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,
td->block[4 + ch][(y << 1) + x],
s->uvlinesize); |
62457f90 |
nnz4 >>= 8;
if (!nnz4) |
628b48db |
goto chroma_idct_end; |
8a467b2d |
} |
53c20f17 |
ch_dst += 4 * s->uvlinesize; |
8a467b2d |
} |
3ae079a3 |
} else { |
53c20f17 |
s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4 + ch], s->uvlinesize); |
3b636f21 |
}
} |
53c20f17 |
chroma_idct_end:
; |
3b636f21 |
}
}
|
53c20f17 |
static av_always_inline |
ac4b32df |
void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb,
VP8FilterStrength *f, int is_vp7) |
3b636f21 |
{
int interior_limit, filter_level;
if (s->segmentation.enabled) { |
17343e39 |
filter_level = s->segmentation.filter_level[mb->segment]; |
3b636f21 |
if (!s->segmentation.absolute_vals)
filter_level += s->filter.level;
} else
filter_level = s->filter.level;
if (s->lf_delta.enabled) {
filter_level += s->lf_delta.ref[mb->ref_frame]; |
dd18c9a0 |
filter_level += s->lf_delta.mode[mb->mode]; |
3b636f21 |
} |
a1b227bb |
|
1550f45a |
filter_level = av_clip_uintp2(filter_level, 6); |
3b636f21 |
interior_limit = filter_level;
if (s->filter.sharpness) { |
8a2c99b4 |
interior_limit >>= (s->filter.sharpness + 3) >> 2; |
3b636f21 |
interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
}
interior_limit = FFMAX(interior_limit, 1);
|
968570d6 |
f->filter_level = filter_level;
f->inner_limit = interior_limit; |
ac4b32df |
f->inner_filter = is_vp7 || !mb->skip || mb->mode == MODE_I4x4 || |
53c20f17 |
mb->mode == VP8_MVMODE_SPLIT; |
3b636f21 |
}
|
53c20f17 |
static av_always_inline
void filter_mb(VP8Context *s, uint8_t *dst[3], VP8FilterStrength *f, |
ac4b32df |
int mb_x, int mb_y, int is_vp7) |
3b636f21 |
{ |
89f2f5db |
int mbedge_lim, bedge_lim_y, bedge_lim_uv, hev_thresh; |
968570d6 |
int filter_level = f->filter_level;
int inner_limit = f->inner_limit; |
c55e0d34 |
int inner_filter = f->inner_filter; |
145d3186 |
int linesize = s->linesize;
int uvlinesize = s->uvlinesize; |
79dec154 |
static const uint8_t hev_thresh_lut[2][64] = {
{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3 },
{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2 }
}; |
3b636f21 |
if (!filter_level)
return;
|
ac4b32df |
if (is_vp7) {
bedge_lim_y = filter_level;
bedge_lim_uv = filter_level * 2;
mbedge_lim = filter_level + 2; |
89f2f5db |
} else {
bedge_lim_y = |
ac4b32df |
bedge_lim_uv = filter_level * 2 + inner_limit;
mbedge_lim = bedge_lim_y + 4; |
89f2f5db |
} |
968570d6 |
|
79dec154 |
hev_thresh = hev_thresh_lut[s->keyframe][filter_level]; |
5245c04d |
|
3b636f21 |
if (mb_x) { |
53c20f17 |
s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize, |
3facfc99 |
mbedge_lim, inner_limit, hev_thresh); |
53c20f17 |
s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize, |
3facfc99 |
mbedge_lim, inner_limit, hev_thresh); |
3b636f21 |
}
|
ac4b32df |
#define H_LOOP_FILTER_16Y_INNER(cond) \
if (cond && inner_filter) { \
s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 4, linesize, \
bedge_lim_y, inner_limit, \
hev_thresh); \
s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize, \
bedge_lim_y, inner_limit, \
hev_thresh); \
s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 12, linesize, \
bedge_lim_y, inner_limit, \
hev_thresh); \
s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, \
uvlinesize, bedge_lim_uv, \
inner_limit, hev_thresh); \ |
3b636f21 |
}
|
ac4b32df |
H_LOOP_FILTER_16Y_INNER(!is_vp7) |
89f2f5db |
|
3b636f21 |
if (mb_y) { |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize, |
3facfc99 |
mbedge_lim, inner_limit, hev_thresh); |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize, |
3facfc99 |
mbedge_lim, inner_limit, hev_thresh); |
3b636f21 |
}
|
c55e0d34 |
if (inner_filter) { |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 4 * linesize, |
ae3313e1 |
linesize, bedge_lim_y, |
145d3186 |
inner_limit, hev_thresh); |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 8 * linesize, |
ae3313e1 |
linesize, bedge_lim_y, |
145d3186 |
inner_limit, hev_thresh); |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 12 * linesize, |
ae3313e1 |
linesize, bedge_lim_y, |
145d3186 |
inner_limit, hev_thresh); |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize,
dst[2] + 4 * uvlinesize, |
ae3313e1 |
uvlinesize, bedge_lim_uv, |
3facfc99 |
inner_limit, hev_thresh); |
3b636f21 |
} |
89f2f5db |
|
ac4b32df |
H_LOOP_FILTER_16Y_INNER(is_vp7) |
3b636f21 |
}
|
53c20f17 |
static av_always_inline
void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8FilterStrength *f,
int mb_x, int mb_y) |
3b636f21 |
{ |
968570d6 |
int mbedge_lim, bedge_lim;
int filter_level = f->filter_level; |
53c20f17 |
int inner_limit = f->inner_limit; |
c55e0d34 |
int inner_filter = f->inner_filter; |
53c20f17 |
int linesize = s->linesize; |
3b636f21 |
if (!filter_level)
return;
|
53c20f17 |
bedge_lim = 2 * filter_level + inner_limit; |
79dec154 |
mbedge_lim = bedge_lim + 4; |
3b636f21 |
if (mb_x) |
145d3186 |
s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim); |
c55e0d34 |
if (inner_filter) { |
53c20f17 |
s->vp8dsp.vp8_h_loop_filter_simple(dst + 4, linesize, bedge_lim);
s->vp8dsp.vp8_h_loop_filter_simple(dst + 8, linesize, bedge_lim);
s->vp8dsp.vp8_h_loop_filter_simple(dst + 12, linesize, bedge_lim); |
3b636f21 |
}
if (mb_y) |
145d3186 |
s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim); |
c55e0d34 |
if (inner_filter) { |
53c20f17 |
s->vp8dsp.vp8_v_loop_filter_simple(dst + 4 * linesize, linesize, bedge_lim);
s->vp8dsp.vp8_v_loop_filter_simple(dst + 8 * linesize, linesize, bedge_lim);
s->vp8dsp.vp8_v_loop_filter_simple(dst + 12 * linesize, linesize, bedge_lim); |
3b636f21 |
}
}
|
337ade52 |
#define MARGIN (16 << 2) |
89f2f5db |
static av_always_inline |
ac4b32df |
void vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe,
VP8Frame *prev_frame, int is_vp7) |
337ade52 |
{
VP8Context *s = avctx->priv_data; |
951455c1 |
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++) { |
53c20f17 |
VP8Macroblock *mb = s->macroblocks_base +
((s->mb_width + 1) * (mb_y + 1) + 1);
int mb_xy = mb_y * s->mb_width; |
951455c1 |
|
53c20f17 |
AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101); |
951455c1 |
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) |
53c20f17 |
AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,
DC_PRED * 0x01010101); |
759001c5 |
decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
prev_frame && prev_frame->seg_map ? |
ac4b32df |
prev_frame->seg_map->data + mb_xy : NULL, 1, is_vp7); |
951455c1 |
s->mv_min.x -= 64;
s->mv_max.x -= 64;
}
s->mv_min.y -= 64;
s->mv_max.y -= 64;
}
}
|
ac4b32df |
static void vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, |
89f2f5db |
VP8Frame *prev_frame)
{ |
ac4b32df |
vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP7); |
89f2f5db |
}
|
ac4b32df |
static void vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, |
89f2f5db |
VP8Frame *prev_frame)
{ |
ac4b32df |
vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP8); |
89f2f5db |
}
|
25f056e6 |
#if HAVE_THREADS |
53c20f17 |
#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); \
} \ |
4d028bf2 |
} while (0) |
53c20f17 |
#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); \ |
f929ab05 |
int is_null = !next_td || !prev_td; \ |
53c20f17 |
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); \
} \ |
4d028bf2 |
} while (0) |
25f056e6 |
#else |
0a6b410e |
#define check_thread_pos(td, otd, mb_x_check, mb_y_check) while(0)
#define update_pos(td, mb_y, mb_x) while(0) |
25f056e6 |
#endif |
951455c1 |
|
89f2f5db |
static av_always_inline void decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, |
ac4b32df |
int jobnr, int threadnr, int is_vp7) |
951455c1 |
{
VP8Context *s = avctx->priv_data;
VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr]; |
53c20f17 |
int mb_y = td->thread_mb_pos >> 16;
int mb_x, mb_xy = mb_y * s->mb_width; |
951455c1 |
int num_jobs = s->num_jobs; |
759001c5 |
VP8Frame *curframe = s->curframe, *prev_frame = s->prev_frame; |
53c20f17 |
VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions - 1)]; |
951455c1 |
VP8Macroblock *mb; |
3b636f21 |
uint8_t *dst[3] = { |
53c20f17 |
curframe->tf.f->data[0] + 16 * mb_y * s->linesize,
curframe->tf.f->data[1] + 8 * mb_y * s->uvlinesize,
curframe->tf.f->data[2] + 8 * mb_y * s->uvlinesize |
3b636f21 |
}; |
53c20f17 |
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]; |
951455c1 |
if (s->mb_layout == 1) |
53c20f17 |
mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1); |
951455c1 |
else { |
c329713d |
// 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->tf, mb_y, 0); |
53c20f17 |
mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2; |
951455c1 |
memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock |
53c20f17 |
AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101); |
951455c1 |
}
|
ac4b32df |
if (!is_vp7 || mb_y == 0) |
89f2f5db |
memset(td->left_nnz, 0, sizeof(td->left_nnz)); |
337ade52 |
s->mv_min.x = -MARGIN; |
53c20f17 |
s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN; |
337ade52 |
for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) { |
951455c1 |
// Wait for previous thread to read mb_x+2, and reach mb_y-1.
if (prev_td != td) {
if (threadnr != 0) { |
ac4b32df |
check_thread_pos(td, prev_td,
mb_x + (is_vp7 ? 2 : 1),
mb_y - (is_vp7 ? 2 : 1)); |
951455c1 |
} else { |
53c20f17 |
check_thread_pos(td, prev_td, |
ac4b32df |
mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3,
mb_y - (is_vp7 ? 2 : 1)); |
951455c1 |
}
}
|
53c20f17 |
s->vdsp.prefetch(dst[0] + (mb_x & 3) * 4 * s->linesize + 64,
s->linesize, 4);
s->vdsp.prefetch(dst[1] + (mb_x & 7) * s->uvlinesize + 64,
dst[2] - dst[1], 2); |
337ade52 |
|
951455c1 |
if (!s->mb_layout) |
759001c5 |
decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
prev_frame && prev_frame->seg_map ? |
ac4b32df |
prev_frame->seg_map->data + mb_xy : NULL, 0, is_vp7); |
337ade52 |
prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
if (!mb->skip) |
ac4b32df |
decode_mb_coeffs(s, td, c, mb, s->top_nnz[mb_x], td->left_nnz, is_vp7); |
337ade52 |
if (mb->mode <= MODE_I4x4) |
ac4b32df |
intra_predict(s, td, dst, mb, mb_x, mb_y, is_vp7); |
337ade52 |
else |
951455c1 |
inter_predict(s, td, dst, mb, mb_x, mb_y); |
337ade52 |
prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);
if (!mb->skip) { |
951455c1 |
idct_mb(s, td, dst, mb); |
337ade52 |
} else { |
951455c1 |
AV_ZERO64(td->left_nnz); |
337ade52 |
AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
|
53c20f17 |
/* Reset DC block predictors if they would exist
* if the mb had coefficients */ |
337ade52 |
if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) { |
951455c1 |
td->left_nnz[8] = 0; |
337ade52 |
s->top_nnz[mb_x][8] = 0;
}
}
if (s->deblock_filter) |
ac4b32df |
filter_level_for_mb(s, mb, &td->filter_strength[mb_x], is_vp7); |
951455c1 |
|
53c20f17 |
if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs - 1) { |
951455c1 |
if (s->filter.simple) |
53c20f17 |
backup_mb_border(s->top_border[mb_x + 1], dst[0],
NULL, NULL, s->linesize, 0, 1); |
951455c1 |
else |
53c20f17 |
backup_mb_border(s->top_border[mb_x + 1], dst[0],
dst[1], dst[2], s->linesize, s->uvlinesize, 0); |
951455c1 |
} |
337ade52 |
prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2); |
3b636f21 |
|
53c20f17 |
dst[0] += 16;
dst[1] += 8;
dst[2] += 8; |
337ade52 |
s->mv_min.x -= 64;
s->mv_max.x -= 64; |
951455c1 |
|
53c20f17 |
if (mb_x == s->mb_width + 1) {
update_pos(td, mb_y, s->mb_width + 3); |
951455c1 |
} else {
update_pos(td, mb_y, mb_x);
} |
3b636f21 |
}
}
|
89f2f5db |
static void vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 1);
}
static void vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 0);
}
static av_always_inline void filter_mb_row(AVCodecContext *avctx, void *tdata, |
ac4b32df |
int jobnr, int threadnr, int is_vp7) |
3b636f21 |
{ |
951455c1 |
VP8Context *s = avctx->priv_data;
VP8ThreadData *td = &s->thread_data[threadnr]; |
53c20f17 |
int mb_x, mb_y = td->thread_mb_pos >> 16, num_jobs = s->num_jobs; |
759001c5 |
AVFrame *curframe = s->curframe->tf.f; |
951455c1 |
VP8Macroblock *mb;
VP8ThreadData *prev_td, *next_td;
uint8_t *dst[3] = { |
53c20f17 |
curframe->data[0] + 16 * mb_y * s->linesize,
curframe->data[1] + 8 * mb_y * s->uvlinesize,
curframe->data[2] + 8 * mb_y * s->uvlinesize |
951455c1 |
}; |
3b636f21 |
|
951455c1 |
if (s->mb_layout == 1) |
53c20f17 |
mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1); |
951455c1 |
else |
53c20f17 |
mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2; |
951455c1 |
|
53c20f17 |
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]; |
951455c1 |
for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) {
VP8FilterStrength *f = &td->filter_strength[mb_x]; |
53c20f17 |
if (prev_td != td)
check_thread_pos(td, prev_td,
(mb_x + 1) + (s->mb_width + 3), mb_y - 1); |
951455c1 |
if (next_td != td) |
53c20f17 |
if (next_td != &s->thread_data[0])
check_thread_pos(td, next_td, mb_x + 1, mb_y + 1); |
951455c1 |
if (num_jobs == 1) {
if (s->filter.simple) |
53c20f17 |
backup_mb_border(s->top_border[mb_x + 1], dst[0],
NULL, NULL, s->linesize, 0, 1); |
951455c1 |
else |
53c20f17 |
backup_mb_border(s->top_border[mb_x + 1], dst[0],
dst[1], dst[2], s->linesize, s->uvlinesize, 0); |
951455c1 |
}
|
337ade52 |
if (s->filter.simple) |
951455c1 |
filter_mb_simple(s, dst[0], f, mb_x, mb_y); |
337ade52 |
else |
ac4b32df |
filter_mb(s, dst, f, mb_x, mb_y, is_vp7); |
951455c1 |
dst[0] += 16;
dst[1] += 8;
dst[2] += 8;
|
53c20f17 |
update_pos(td, mb_y, (s->mb_width + 3) + mb_x); |
3b636f21 |
}
}
|
89f2f5db |
static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
filter_mb_row(avctx, tdata, jobnr, threadnr, 1);
}
static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
filter_mb_row(avctx, tdata, jobnr, threadnr, 0);
}
|
ac4b32df |
static av_always_inline
int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
int threadnr, int is_vp7) |
ce42a048 |
{ |
951455c1 |
VP8Context *s = avctx->priv_data;
VP8ThreadData *td = &s->thread_data[jobnr];
VP8ThreadData *next_td = NULL, *prev_td = NULL; |
759001c5 |
VP8Frame *curframe = s->curframe; |
951455c1 |
int mb_y, num_jobs = s->num_jobs; |
53c20f17 |
|
951455c1 |
td->thread_nr = threadnr;
for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) { |
53c20f17 |
if (mb_y >= s->mb_height)
break;
td->thread_mb_pos = mb_y << 16; |
89f2f5db |
s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr); |
951455c1 |
if (s->deblock_filter) |
89f2f5db |
s->filter_mb_row(avctx, tdata, jobnr, threadnr); |
951455c1 |
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) |
759001c5 |
ff_thread_report_progress(&curframe->tf, mb_y, 0); |
337ade52 |
} |
951455c1 |
return 0; |
ce42a048 |
}
|
ac4b32df |
static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP7);
}
static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
int jobnr, int threadnr)
{
return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP8);
}
static av_always_inline
int vp78_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
AVPacket *avpkt, int is_vp7) |
3b636f21 |
{
VP8Context *s = avctx->priv_data; |
951455c1 |
int ret, i, referenced, num_jobs; |
3b636f21 |
enum AVDiscard skip_thresh; |
759001c5 |
VP8Frame *av_uninit(curframe), *prev_frame; |
ce42a048 |
|
ac4b32df |
if (is_vp7)
ret = vp7_decode_frame_header(s, avpkt->data, avpkt->size);
else
ret = vp8_decode_frame_header(s, avpkt->data, avpkt->size);
if (ret < 0) |
fb90785e |
goto err; |
3b636f21 |
|
e02dec25 |
prev_frame = s->framep[VP56_FRAME_CURRENT];
|
53c20f17 |
referenced = s->update_last || s->update_golden == VP56_FRAME_CURRENT ||
s->update_altref == VP56_FRAME_CURRENT; |
3b636f21 |
|
53c20f17 |
skip_thresh = !referenced ? AVDISCARD_NONREF
: !s->keyframe ? AVDISCARD_NONKEY
: AVDISCARD_ALL; |
3b636f21 |
if (avctx->skip_frame >= skip_thresh) {
s->invisible = 1; |
fb90785e |
memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); |
3b636f21 |
goto skip_decode;
} |
9ac831c2 |
s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh; |
3b636f21 |
|
4773d904 |
// release no longer referenced frames
for (i = 0; i < 5; i++) |
759001c5 |
if (s->frames[i].tf.f->data[0] && |
4773d904 |
&s->frames[i] != prev_frame &&
&s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] && |
53c20f17 |
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN] && |
4773d904 |
&s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) |
759001c5 |
vp8_release_frame(s, &s->frames[i]); |
4773d904 |
|
7d4c0220 |
curframe = s->framep[VP56_FRAME_CURRENT] = vp8_find_free_buffer(s); |
3b636f21 |
|
65875a8b |
if (!s->colorspace)
avctx->colorspace = AVCOL_SPC_BT470BG;
if (s->fullrange)
avctx->color_range = AVCOL_RANGE_JPEG;
else
avctx->color_range = AVCOL_RANGE_MPEG;
|
53c20f17 |
/* 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. */ |
fb90785e |
if (!s->keyframe && (!s->framep[VP56_FRAME_PREVIOUS] || |
53c20f17 |
!s->framep[VP56_FRAME_GOLDEN] || |
fb90785e |
!s->framep[VP56_FRAME_GOLDEN2])) { |
53c20f17 |
av_log(avctx, AV_LOG_WARNING,
"Discarding interframe without a prior keyframe!\n"); |
fb90785e |
ret = AVERROR_INVALIDDATA;
goto err;
}
|
759001c5 |
curframe->tf.f->key_frame = s->keyframe; |
53c20f17 |
curframe->tf.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
: AV_PICTURE_TYPE_P; |
1ec94b0f |
if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0) |
fb90785e |
goto err; |
3b636f21 |
|
4773d904 |
// check if golden and altref are swapped |
53c20f17 |
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) |
4773d904 |
s->next_framep[VP56_FRAME_PREVIOUS] = curframe; |
53c20f17 |
else |
4773d904 |
s->next_framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_PREVIOUS]; |
53c20f17 |
s->next_framep[VP56_FRAME_CURRENT] = curframe; |
4773d904 |
|
8c22bea5 |
if (avctx->codec->update_thread_context)
ff_thread_finish_setup(avctx); |
75947915 |
|
759001c5 |
s->linesize = curframe->tf.f->linesize[0];
s->uvlinesize = curframe->tf.f->linesize[1]; |
3b636f21 |
|
53c20f17 |
memset(s->top_nnz, 0, s->mb_width * sizeof(*s->top_nnz));
/* Zero macroblock structures for top/top-left prediction
* from outside the frame. */ |
951455c1 |
if (!s->mb_layout) |
53c20f17 |
memset(s->macroblocks + s->mb_height * 2 - 1, 0,
(s->mb_width + 1) * sizeof(*s->macroblocks)); |
951455c1 |
if (!s->mb_layout && s->keyframe) |
53c20f17 |
memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width * 4); |
c55e0d34 |
|
c4211046 |
memset(s->ref_count, 0, sizeof(s->ref_count)); |
3b636f21 |
|
c329713d |
if (s->mb_layout == 1) {
// 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->tf, 1, 0); |
ac4b32df |
if (is_vp7)
vp7_decode_mv_mb_modes(avctx, curframe, prev_frame);
else
vp8_decode_mv_mb_modes(avctx, curframe, prev_frame); |
c329713d |
} |
3b636f21 |
|
951455c1 |
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; |
53c20f17 |
s->thread_data[i].wait_mb_pos = INT_MAX; |
951455c1 |
} |
ac4b32df |
if (is_vp7)
avctx->execute2(avctx, vp7_decode_mb_row_sliced, s->thread_data, NULL,
num_jobs);
else
avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL,
num_jobs); |
3b636f21 |
|
759001c5 |
ff_thread_report_progress(&curframe->tf, INT_MAX, 0); |
fb90785e |
memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
|
3b636f21 |
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];
if (!s->invisible) { |
759001c5 |
if ((ret = av_frame_ref(data, curframe->tf.f)) < 0)
return ret; |
53c20f17 |
*got_frame = 1; |
3b636f21 |
}
return avpkt->size; |
fb90785e |
err:
memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); |
75947915 |
return ret; |
3b636f21 |
}
|
ac4b32df |
int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
AVPacket *avpkt)
{
return vp78_decode_frame(avctx, data, got_frame, avpkt, IS_VP8);
}
#if CONFIG_VP7_DECODER
static int vp7_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
AVPacket *avpkt)
{
return vp78_decode_frame(avctx, data, got_frame, avpkt, IS_VP7);
}
#endif /* CONFIG_VP7_DECODER */
|
c4bfa098 |
av_cold int ff_vp8_decode_free(AVCodecContext *avctx) |
759001c5 |
{
VP8Context *s = avctx->priv_data;
int i;
|
a84f0e8d |
if (!s)
return 0;
|
759001c5 |
vp8_decode_flush_impl(avctx, 1);
for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
av_frame_free(&s->frames[i].tf.f);
return 0;
}
static av_cold int vp8_init_frames(VP8Context *s)
{
int i;
for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) {
s->frames[i].tf.f = av_frame_alloc();
if (!s->frames[i].tf.f)
return AVERROR(ENOMEM);
}
return 0;
}
|
ac4b32df |
static av_always_inline
int vp78_decode_init(AVCodecContext *avctx, int is_vp7) |
3b636f21 |
{
VP8Context *s = avctx->priv_data; |
759001c5 |
int ret; |
3b636f21 |
s->avctx = avctx; |
89f2f5db |
s->vp7 = avctx->codec->id == AV_CODEC_ID_VP7; |
716d413c |
avctx->pix_fmt = AV_PIX_FMT_YUV420P; |
759001c5 |
avctx->internal->allocate_progress = 1; |
3b636f21 |
|
8c53d39e |
ff_videodsp_init(&s->vdsp, 8); |
89f2f5db |
|
ac4b32df |
ff_vp78dsp_init(&s->vp8dsp);
if (CONFIG_VP7_DECODER && is_vp7) {
ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP7, 8, 1);
ff_vp7dsp_init(&s->vp8dsp); |
fb61ed1e |
s->decode_mb_row_no_filter = vp7_decode_mb_row_no_filter;
s->filter_mb_row = vp7_filter_mb_row; |
ac4b32df |
} else if (CONFIG_VP8_DECODER && !is_vp7) {
ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1);
ff_vp8dsp_init(&s->vp8dsp); |
fb61ed1e |
s->decode_mb_row_no_filter = vp8_decode_mb_row_no_filter;
s->filter_mb_row = vp8_filter_mb_row; |
ac4b32df |
}
/* does not change for VP8 */ |
f4d581cd |
memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan)); |
3b636f21 |
|
759001c5 |
if ((ret = vp8_init_frames(s)) < 0) { |
c4bfa098 |
ff_vp8_decode_free(avctx); |
759001c5 |
return ret;
} |
3b636f21 |
return 0;
}
|
ac4b32df |
#if CONFIG_VP7_DECODER
static int vp7_decode_init(AVCodecContext *avctx)
{
return vp78_decode_init(avctx, IS_VP7);
}
#endif /* CONFIG_VP7_DECODER */
av_cold int ff_vp8_decode_init(AVCodecContext *avctx)
{
return vp78_decode_init(avctx, IS_VP8);
}
|
89f2f5db |
#if CONFIG_VP8_DECODER |
2cbaa078 |
#if HAVE_THREADS |
4773d904 |
static av_cold int vp8_decode_init_thread_copy(AVCodecContext *avctx)
{
VP8Context *s = avctx->priv_data; |
759001c5 |
int ret; |
4773d904 |
s->avctx = avctx;
|
759001c5 |
if ((ret = vp8_init_frames(s)) < 0) { |
c4bfa098 |
ff_vp8_decode_free(avctx); |
759001c5 |
return ret;
}
|
4773d904 |
return 0;
}
|
c89be82c |
#define REBASE(pic) ((pic) ? (pic) - &s_src->frames[0] + &s->frames[0] : NULL) |
4773d904 |
|
53c20f17 |
static int vp8_decode_update_thread_context(AVCodecContext *dst,
const AVCodecContext *src) |
4773d904 |
{
VP8Context *s = dst->priv_data, *s_src = src->priv_data; |
759001c5 |
int i; |
4773d904 |
|
56535793 |
if (s->macroblocks_base &&
(s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) {
free_buffers(s); |
82a0497c |
s->mb_width = s_src->mb_width;
s->mb_height = s_src->mb_height; |
56535793 |
}
|
53c20f17 |
s->prob[0] = s_src->prob[!s_src->update_probabilities]; |
4773d904 |
s->segmentation = s_src->segmentation; |
53c20f17 |
s->lf_delta = s_src->lf_delta; |
4773d904 |
memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias));
|
759001c5 |
for (i = 0; i < FF_ARRAY_ELEMS(s_src->frames); i++) {
if (s_src->frames[i].tf.f->data[0]) {
int ret = vp8_ref_frame(s, &s->frames[i], &s_src->frames[i]);
if (ret < 0)
return ret;
}
}
|
4773d904 |
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;
} |
2cbaa078 |
#endif /* HAVE_THREADS */ |
ac4b32df |
#endif /* CONFIG_VP8_DECODER */ |
4773d904 |
|
89f2f5db |
#if CONFIG_VP7_DECODER
AVCodec ff_vp7_decoder = {
.name = "vp7",
.long_name = NULL_IF_CONFIG_SMALL("On2 VP7"),
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_VP7,
.priv_data_size = sizeof(VP8Context), |
ac4b32df |
.init = vp7_decode_init, |
89f2f5db |
.close = ff_vp8_decode_free, |
ac4b32df |
.decode = vp7_decode_frame, |
def97856 |
.capabilities = AV_CODEC_CAP_DR1, |
89f2f5db |
.flush = vp8_decode_flush,
}; |
ac4b32df |
#endif /* CONFIG_VP7_DECODER */ |
89f2f5db |
#if CONFIG_VP8_DECODER |
d36beb3f |
AVCodec ff_vp8_decoder = { |
00c3b67b |
.name = "vp8", |
b2bed932 |
.long_name = NULL_IF_CONFIG_SMALL("On2 VP8"), |
00c3b67b |
.type = AVMEDIA_TYPE_VIDEO, |
36ef5369 |
.id = AV_CODEC_ID_VP8, |
00c3b67b |
.priv_data_size = sizeof(VP8Context), |
c4bfa098 |
.init = ff_vp8_decode_init,
.close = ff_vp8_decode_free,
.decode = ff_vp8_decode_frame, |
def97856 |
.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
AV_CODEC_CAP_SLICE_THREADS, |
00c3b67b |
.flush = vp8_decode_flush, |
4773d904 |
.init_thread_copy = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy),
.update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context), |
3b636f21 |
}; |
ac4b32df |
#endif /* CONFIG_VP7_DECODER */ |