vpx/vp9/decoder/vp9_decodemv.c

840 lines
31 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <assert.h>
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_mvref_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_seg_common.h"
#include "vp9/decoder/vp9_decodemv.h"
#include "vp9/decoder/vp9_decodeframe.h"
#include "vpx_dsp/vpx_dsp_common.h"
static PREDICTION_MODE read_intra_mode(vpx_reader *r, const vpx_prob *p) {
return (PREDICTION_MODE)vpx_read_tree(r, vp9_intra_mode_tree, p);
}
static PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r, int size_group) {
const PREDICTION_MODE y_mode =
read_intra_mode(r, cm->fc->y_mode_prob[size_group]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->y_mode[size_group][y_mode];
return y_mode;
}
static PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r,
PREDICTION_MODE y_mode) {
const PREDICTION_MODE uv_mode =
read_intra_mode(r, cm->fc->uv_mode_prob[y_mode]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->uv_mode[y_mode][uv_mode];
return uv_mode;
}
static PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r, int ctx) {
const int mode =
vpx_read_tree(r, vp9_inter_mode_tree, cm->fc->inter_mode_probs[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->inter_mode[ctx][mode];
return NEARESTMV + mode;
}
static int read_segment_id(vpx_reader *r, const struct segmentation *seg) {
return vpx_read_tree(r, vp9_segment_tree, seg->tree_probs);
}
static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
TX_SIZE max_tx_size, vpx_reader *r) {
FRAME_COUNTS *counts = xd->counts;
const int ctx = get_tx_size_context(xd);
const vpx_prob *tx_probs = get_tx_probs(max_tx_size, ctx, &cm->fc->tx_probs);
int tx_size = vpx_read(r, tx_probs[0]);
if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) {
tx_size += vpx_read(r, tx_probs[1]);
if (tx_size != TX_8X8 && max_tx_size >= TX_32X32)
tx_size += vpx_read(r, tx_probs[2]);
}
if (counts) ++get_tx_counts(max_tx_size, ctx, &counts->tx)[tx_size];
return (TX_SIZE)tx_size;
}
static INLINE TX_SIZE read_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
int allow_select, vpx_reader *r) {
TX_MODE tx_mode = cm->tx_mode;
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
const TX_SIZE max_tx_size = max_txsize_lookup[bsize];
if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8)
return read_selected_tx_size(cm, xd, max_tx_size, r);
else
return VPXMIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
}
static int dec_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids,
int mi_offset, int x_mis, int y_mis) {
int x, y, segment_id = INT_MAX;
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
segment_id =
VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
return segment_id;
}
static void set_segment_id(VP9_COMMON *cm, int mi_offset, int x_mis, int y_mis,
int segment_id) {
int x, y;
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
}
static void copy_segment_id(const VP9_COMMON *cm,
const uint8_t *last_segment_ids,
uint8_t *current_segment_ids, int mi_offset,
int x_mis, int y_mis) {
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
int x, y;
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
current_segment_ids[mi_offset + y * cm->mi_cols + x] =
last_segment_ids ? last_segment_ids[mi_offset + y * cm->mi_cols + x]
: 0;
}
static int read_intra_segment_id(VP9_COMMON *const cm, int mi_offset, int x_mis,
int y_mis, vpx_reader *r) {
struct segmentation *const seg = &cm->seg;
int segment_id;
if (!seg->enabled) return 0; // Default for disabled segmentation
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
if (!seg->update_map) {
copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
mi_offset, x_mis, y_mis);
return 0;
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
}
segment_id = read_segment_id(r, seg);
set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
return segment_id;
}
static int read_inter_segment_id(VP9_COMMON *const cm, MACROBLOCKD *const xd,
int mi_row, int mi_col, vpx_reader *r,
int x_mis, int y_mis) {
struct segmentation *const seg = &cm->seg;
MODE_INFO *const mi = xd->mi[0];
int predicted_segment_id, segment_id;
const int mi_offset = mi_row * cm->mi_cols + mi_col;
if (!seg->enabled) return 0; // Default for disabled segmentation
predicted_segment_id = cm->last_frame_seg_map
? dec_get_segment_id(cm, cm->last_frame_seg_map,
mi_offset, x_mis, y_mis)
: 0;
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
if (!seg->update_map) {
copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
mi_offset, x_mis, y_mis);
return predicted_segment_id;
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
}
if (seg->temporal_update) {
const vpx_prob pred_prob = vp9_get_pred_prob_seg_id(seg, xd);
mi->seg_id_predicted = vpx_read(r, pred_prob);
segment_id =
mi->seg_id_predicted ? predicted_segment_id : read_segment_id(r, seg);
} else {
segment_id = read_segment_id(r, seg);
}
set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
return segment_id;
}
static int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
vpx_reader *r) {
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
return 1;
} else {
const int ctx = vp9_get_skip_context(xd);
const int skip = vpx_read(r, cm->fc->skip_probs[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->skip[ctx][skip];
return skip;
}
}
static void read_intra_frame_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd, int mi_row,
int mi_col, vpx_reader *r, int x_mis,
int y_mis) {
MODE_INFO *const mi = xd->mi[0];
const MODE_INFO *above_mi = xd->above_mi;
const MODE_INFO *left_mi = xd->left_mi;
const BLOCK_SIZE bsize = mi->sb_type;
int i;
const int mi_offset = mi_row * cm->mi_cols + mi_col;
mi->segment_id = read_intra_segment_id(cm, mi_offset, x_mis, y_mis, r);
mi->skip = read_skip(cm, xd, mi->segment_id, r);
mi->tx_size = read_tx_size(cm, xd, 1, r);
mi->ref_frame[0] = INTRA_FRAME;
mi->ref_frame[1] = NONE;
switch (bsize) {
case BLOCK_4X4:
for (i = 0; i < 4; ++i)
mi->bmi[i].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, i));
mi->mode = mi->bmi[3].as_mode;
break;
case BLOCK_4X8:
mi->bmi[0].as_mode = mi->bmi[2].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
mi->bmi[1].as_mode = mi->bmi[3].as_mode = mi->mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 1));
break;
case BLOCK_8X4:
mi->bmi[0].as_mode = mi->bmi[1].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
mi->bmi[2].as_mode = mi->bmi[3].as_mode = mi->mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 2));
break;
default:
mi->mode = read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
}
mi->uv_mode = read_intra_mode(r, vp9_kf_uv_mode_prob[mi->mode]);
}
2010-05-18 17:58:33 +02:00
static int read_mv_component(vpx_reader *r, const nmv_component *mvcomp,
int usehp) {
int mag, d, fr, hp;
const int sign = vpx_read(r, mvcomp->sign);
const int mv_class = vpx_read_tree(r, vp9_mv_class_tree, mvcomp->classes);
const int class0 = mv_class == MV_CLASS_0;
// Integer part
if (class0) {
d = vpx_read(r, mvcomp->class0[0]);
mag = 0;
} else {
int i;
const int n = mv_class + CLASS0_BITS - 1; // number of bits
d = 0;
for (i = 0; i < n; ++i) d |= vpx_read(r, mvcomp->bits[i]) << i;
mag = CLASS0_SIZE << (mv_class + 2);
}
// Fractional part
fr = vpx_read_tree(r, vp9_mv_fp_tree,
class0 ? mvcomp->class0_fp[d] : mvcomp->fp);
// High precision part (if hp is not used, the default value of the hp is 1)
hp = usehp ? vpx_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp) : 1;
// Result
mag += ((d << 3) | (fr << 1) | hp) + 1;
return sign ? -mag : mag;
}
static INLINE void read_mv(vpx_reader *r, MV *mv, const MV *ref,
const nmv_context *ctx, nmv_context_counts *counts,
int allow_hp) {
const MV_JOINT_TYPE joint_type =
(MV_JOINT_TYPE)vpx_read_tree(r, vp9_mv_joint_tree, ctx->joints);
const int use_hp = allow_hp && use_mv_hp(ref);
MV diff = { 0, 0 };
if (mv_joint_vertical(joint_type))
diff.row = read_mv_component(r, &ctx->comps[0], use_hp);
if (mv_joint_horizontal(joint_type))
diff.col = read_mv_component(r, &ctx->comps[1], use_hp);
vp9_inc_mv(&diff, counts);
mv->row = ref->row + diff.row;
mv->col = ref->col + diff.col;
}
static REFERENCE_MODE read_block_reference_mode(VP9_COMMON *cm,
const MACROBLOCKD *xd,
vpx_reader *r) {
if (cm->reference_mode == REFERENCE_MODE_SELECT) {
const int ctx = vp9_get_reference_mode_context(cm, xd);
const REFERENCE_MODE mode =
(REFERENCE_MODE)vpx_read(r, cm->fc->comp_inter_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->comp_inter[ctx][mode];
return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE
} else {
return cm->reference_mode;
}
}
// Read the referncence frame
static void read_ref_frames(VP9_COMMON *const cm, MACROBLOCKD *const xd,
vpx_reader *r, int segment_id,
MV_REFERENCE_FRAME ref_frame[2]) {
FRAME_CONTEXT *const fc = cm->fc;
FRAME_COUNTS *counts = xd->counts;
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
ref_frame[0] = (MV_REFERENCE_FRAME)get_segdata(&cm->seg, segment_id,
SEG_LVL_REF_FRAME);
ref_frame[1] = NONE;
} else {
const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r);
// FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding
if (mode == COMPOUND_REFERENCE) {
const int idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
const int ctx = vp9_get_pred_context_comp_ref_p(cm, xd);
const int bit = vpx_read(r, fc->comp_ref_prob[ctx]);
if (counts) ++counts->comp_ref[ctx][bit];
ref_frame[idx] = cm->comp_fixed_ref;
ref_frame[!idx] = cm->comp_var_ref[bit];
} else if (mode == SINGLE_REFERENCE) {
const int ctx0 = vp9_get_pred_context_single_ref_p1(xd);
const int bit0 = vpx_read(r, fc->single_ref_prob[ctx0][0]);
if (counts) ++counts->single_ref[ctx0][0][bit0];
if (bit0) {
const int ctx1 = vp9_get_pred_context_single_ref_p2(xd);
const int bit1 = vpx_read(r, fc->single_ref_prob[ctx1][1]);
if (counts) ++counts->single_ref[ctx1][1][bit1];
ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME;
} else {
ref_frame[0] = LAST_FRAME;
}
ref_frame[1] = NONE;
} else {
assert(0 && "Invalid prediction mode.");
}
}
}
2010-05-18 17:58:33 +02:00
static INLINE INTERP_FILTER read_switchable_interp_filter(VP9_COMMON *const cm,
MACROBLOCKD *const xd,
vpx_reader *r) {
const int ctx = get_pred_context_switchable_interp(xd);
const INTERP_FILTER type = (INTERP_FILTER)vpx_read_tree(
r, vp9_switchable_interp_tree, cm->fc->switchable_interp_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->switchable_interp[ctx][type];
return type;
}
static void read_intra_block_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd, MODE_INFO *mi,
vpx_reader *r) {
const BLOCK_SIZE bsize = mi->sb_type;
int i;
switch (bsize) {
case BLOCK_4X4:
for (i = 0; i < 4; ++i)
mi->bmi[i].as_mode = read_intra_mode_y(cm, xd, r, 0);
mi->mode = mi->bmi[3].as_mode;
break;
case BLOCK_4X8:
mi->bmi[0].as_mode = mi->bmi[2].as_mode = read_intra_mode_y(cm, xd, r, 0);
mi->bmi[1].as_mode = mi->bmi[3].as_mode = mi->mode =
read_intra_mode_y(cm, xd, r, 0);
break;
case BLOCK_8X4:
mi->bmi[0].as_mode = mi->bmi[1].as_mode = read_intra_mode_y(cm, xd, r, 0);
mi->bmi[2].as_mode = mi->bmi[3].as_mode = mi->mode =
read_intra_mode_y(cm, xd, r, 0);
break;
default: mi->mode = read_intra_mode_y(cm, xd, r, size_group_lookup[bsize]);
}
mi->uv_mode = read_intra_mode_uv(cm, xd, r, mi->mode);
// Initialize interp_filter here so we do not have to check for inter block
// modes in get_pred_context_switchable_interp()
mi->interp_filter = SWITCHABLE_FILTERS;
mi->ref_frame[0] = INTRA_FRAME;
mi->ref_frame[1] = NONE;
}
static INLINE int is_mv_valid(const MV *mv) {
return mv->row > MV_LOW && mv->row < MV_UPP && mv->col > MV_LOW &&
mv->col < MV_UPP;
}
static INLINE void copy_mv_pair(int_mv *dst, const int_mv *src) {
memcpy(dst, src, sizeof(*dst) * 2);
}
static INLINE void zero_mv_pair(int_mv *dst) {
memset(dst, 0, sizeof(*dst) * 2);
}
static INLINE int assign_mv(VP9_COMMON *cm, MACROBLOCKD *xd,
PREDICTION_MODE mode, int_mv mv[2],
int_mv ref_mv[2], int_mv near_nearest_mv[2],
int is_compound, int allow_hp, vpx_reader *r) {
int i;
int ret = 1;
switch (mode) {
case NEWMV: {
FRAME_COUNTS *counts = xd->counts;
nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL;
for (i = 0; i < 1 + is_compound; ++i) {
read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts,
allow_hp);
ret = ret && is_mv_valid(&mv[i].as_mv);
}
break;
}
case NEARMV:
case NEARESTMV: {
copy_mv_pair(mv, near_nearest_mv);
break;
}
case ZEROMV: {
zero_mv_pair(mv);
break;
}
default: { return 0; }
}
return ret;
}
static int read_is_inter_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
int segment_id, vpx_reader *r) {
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
return get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME;
} else {
const int ctx = get_intra_inter_context(xd);
const int is_inter = vpx_read(r, cm->fc->intra_inter_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts) ++counts->intra_inter[ctx][is_inter];
return is_inter;
}
}
static void dec_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *best_mv,
int refmv_count) {
int i;
// Make sure all the candidates are properly clamped etc
for (i = 0; i < refmv_count; ++i) {
lower_mv_precision(&mvlist[i].as_mv, allow_hp);
*best_mv = mvlist[i];
}
}
// This macro is used to add a motion vector mv_ref list if it isn't
// already in the list. If it's the second motion vector or early_break
// it will also skip all additional processing and jump to Done!
#define ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done) \
do { \
if (refmv_count) { \
if ((mv).as_int != (mv_ref_list)[0].as_int) { \
(mv_ref_list)[(refmv_count)] = (mv); \
refmv_count++; \
goto Done; \
} \
} else { \
(mv_ref_list)[(refmv_count)++] = (mv); \
if (early_break) goto Done; \
} \
} while (0)
// If either reference frame is different, not INTRA, and they
// are different from each other scale and add the mv to our list.
#define IF_DIFF_REF_FRAME_ADD_MV_EB(mbmi, ref_frame, ref_sign_bias, \
refmv_count, mv_ref_list, Done) \
do { \
if (is_inter_block(mbmi)) { \
if ((mbmi)->ref_frame[0] != ref_frame) \
ADD_MV_REF_LIST_EB(scale_mv((mbmi), 0, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, Done); \
if (has_second_ref(mbmi) && (mbmi)->ref_frame[1] != ref_frame && \
(mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \
ADD_MV_REF_LIST_EB(scale_mv((mbmi), 1, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, Done); \
} \
} while (0)
// This function searches the neighborhood of a given MB/SB
// to try and find candidate reference vectors.
static int dec_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd,
PREDICTION_MODE mode, MV_REFERENCE_FRAME ref_frame,
const POSITION *const mv_ref_search,
int_mv *mv_ref_list, int mi_row, int mi_col,
int block, int is_sub8x8) {
const int *ref_sign_bias = cm->ref_frame_sign_bias;
int i, refmv_count = 0;
int different_ref_found = 0;
const MV_REF *const prev_frame_mvs =
cm->use_prev_frame_mvs
? cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col
: NULL;
const TileInfo *const tile = &xd->tile;
// If mode is nearestmv or newmv (uses nearestmv as a reference) then stop
// searching after the first mv is found.
const int early_break = (mode != NEARMV);
// Blank the reference vector list
memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
i = 0;
if (is_sub8x8) {
// If the size < 8x8 we get the mv from the bmi substructure for the
// nearest two blocks.
for (i = 0; i < 2; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate_mi =
xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride];
different_ref_found = 1;
if (candidate_mi->ref_frame[0] == ref_frame)
ADD_MV_REF_LIST_EB(
get_sub_block_mv(candidate_mi, 0, mv_ref->col, block),
refmv_count, mv_ref_list, Done);
else if (candidate_mi->ref_frame[1] == ref_frame)
ADD_MV_REF_LIST_EB(
get_sub_block_mv(candidate_mi, 1, mv_ref->col, block),
refmv_count, mv_ref_list, Done);
}
}
}
// Check the rest of the neighbors in much the same way
// as before except we don't need to keep track of sub blocks or
// mode counts.
for (; i < MVREF_NEIGHBOURS; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate =
xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride];
different_ref_found = 1;
if (candidate->ref_frame[0] == ref_frame)
ADD_MV_REF_LIST_EB(candidate->mv[0], refmv_count, mv_ref_list, Done);
else if (candidate->ref_frame[1] == ref_frame)
ADD_MV_REF_LIST_EB(candidate->mv[1], refmv_count, mv_ref_list, Done);
}
}
// Check the last frame's mode and mv info.
if (prev_frame_mvs) {
if (prev_frame_mvs->ref_frame[0] == ref_frame) {
ADD_MV_REF_LIST_EB(prev_frame_mvs->mv[0], refmv_count, mv_ref_list, Done);
} else if (prev_frame_mvs->ref_frame[1] == ref_frame) {
ADD_MV_REF_LIST_EB(prev_frame_mvs->mv[1], refmv_count, mv_ref_list, Done);
}
}
// Since we couldn't find 2 mvs from the same reference frame
// go back through the neighbors and find motion vectors from
// different reference frames.
if (different_ref_found) {
for (i = 0; i < MVREF_NEIGHBOURS; ++i) {
const POSITION *mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate =
xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride];
// If the candidate is INTRA we don't want to consider its mv.
IF_DIFF_REF_FRAME_ADD_MV_EB(candidate, ref_frame, ref_sign_bias,
refmv_count, mv_ref_list, Done);
}
}
}
// Since we still don't have a candidate we'll try the last frame.
if (prev_frame_mvs) {
if (prev_frame_mvs->ref_frame[0] != ref_frame &&
prev_frame_mvs->ref_frame[0] > INTRA_FRAME) {
int_mv mv = prev_frame_mvs->mv[0];
if (ref_sign_bias[prev_frame_mvs->ref_frame[0]] !=
ref_sign_bias[ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done);
}
if (prev_frame_mvs->ref_frame[1] > INTRA_FRAME &&
prev_frame_mvs->ref_frame[1] != ref_frame &&
prev_frame_mvs->mv[1].as_int != prev_frame_mvs->mv[0].as_int) {
int_mv mv = prev_frame_mvs->mv[1];
if (ref_sign_bias[prev_frame_mvs->ref_frame[1]] !=
ref_sign_bias[ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done);
}
}
if (mode == NEARMV)
refmv_count = MAX_MV_REF_CANDIDATES;
else
// we only care about the nearestmv for the remaining modes
refmv_count = 1;
Done:
// Clamp vectors
for (i = 0; i < refmv_count; ++i) clamp_mv_ref(&mv_ref_list[i].as_mv, xd);
return refmv_count;
}
static void append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
const POSITION *const mv_ref_search,
PREDICTION_MODE b_mode, int block,
int ref, int mi_row, int mi_col,
int_mv *best_sub8x8) {
int_mv mv_list[MAX_MV_REF_CANDIDATES];
MODE_INFO *const mi = xd->mi[0];
b_mode_info *bmi = mi->bmi;
int n;
int refmv_count;
assert(MAX_MV_REF_CANDIDATES == 2);
refmv_count =
dec_find_mv_refs(cm, xd, b_mode, mi->ref_frame[ref], mv_ref_search,
mv_list, mi_row, mi_col, block, 1);
switch (block) {
case 0: best_sub8x8->as_int = mv_list[refmv_count - 1].as_int; break;
case 1:
case 2:
if (b_mode == NEARESTMV) {
best_sub8x8->as_int = bmi[0].as_mv[ref].as_int;
} else {
best_sub8x8->as_int = 0;
for (n = 0; n < refmv_count; ++n)
if (bmi[0].as_mv[ref].as_int != mv_list[n].as_int) {
best_sub8x8->as_int = mv_list[n].as_int;
break;
}
}
break;
case 3:
if (b_mode == NEARESTMV) {
best_sub8x8->as_int = bmi[2].as_mv[ref].as_int;
} else {
int_mv candidates[2 + MAX_MV_REF_CANDIDATES];
candidates[0] = bmi[1].as_mv[ref];
candidates[1] = bmi[0].as_mv[ref];
candidates[2] = mv_list[0];
candidates[3] = mv_list[1];
best_sub8x8->as_int = 0;
for (n = 0; n < 2 + MAX_MV_REF_CANDIDATES; ++n)
if (bmi[2].as_mv[ref].as_int != candidates[n].as_int) {
best_sub8x8->as_int = candidates[n].as_int;
break;
}
}
break;
default: assert(0 && "Invalid block index.");
}
}
static uint8_t get_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd,
const POSITION *const mv_ref_search, int mi_row,
int mi_col) {
int i;
int context_counter = 0;
const TileInfo *const tile = &xd->tile;
// Get mode count from nearest 2 blocks
for (i = 0; i < 2; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate =
xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride];
// Keep counts for entropy encoding.
context_counter += mode_2_counter[candidate->mode];
}
}
return counter_to_context[context_counter];
}
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
static void read_inter_block_mode_info(VP9Decoder *const pbi,
MACROBLOCKD *const xd,
MODE_INFO *const mi, int mi_row,
int mi_col, vpx_reader *r) {
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
VP9_COMMON *const cm = &pbi->common;
const BLOCK_SIZE bsize = mi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
int_mv best_ref_mvs[2];
int ref, is_compound;
uint8_t inter_mode_ctx;
const POSITION *const mv_ref_search = mv_ref_blocks[bsize];
read_ref_frames(cm, xd, r, mi->segment_id, mi->ref_frame);
is_compound = has_second_ref(mi);
inter_mode_ctx = get_mode_context(cm, xd, mv_ref_search, mi_row, mi_col);
if (segfeature_active(&cm->seg, mi->segment_id, SEG_LVL_SKIP)) {
mi->mode = ZEROMV;
if (bsize < BLOCK_8X8) {
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid usage of segement feature on small blocks");
return;
}
} else {
if (bsize >= BLOCK_8X8)
mi->mode = read_inter_mode(cm, xd, r, inter_mode_ctx);
else
// Sub 8x8 blocks use the nearestmv as a ref_mv if the b_mode is NEWMV.
// Setting mode to NEARESTMV forces the search to stop after the nearestmv
// has been found. After b_modes have been read, mode will be overwritten
// by the last b_mode.
mi->mode = NEARESTMV;
if (mi->mode != ZEROMV) {
for (ref = 0; ref < 1 + is_compound; ++ref) {
int_mv tmp_mvs[MAX_MV_REF_CANDIDATES];
const MV_REFERENCE_FRAME frame = mi->ref_frame[ref];
int refmv_count;
refmv_count = dec_find_mv_refs(cm, xd, mi->mode, frame, mv_ref_search,
tmp_mvs, mi_row, mi_col, -1, 0);
dec_find_best_ref_mvs(allow_hp, tmp_mvs, &best_ref_mvs[ref],
refmv_count);
}
}
}
mi->interp_filter = (cm->interp_filter == SWITCHABLE)
? read_switchable_interp_filter(cm, xd, r)
: cm->interp_filter;
if (bsize < BLOCK_8X8) {
const int num_4x4_w = 1 << xd->bmode_blocks_wl;
const int num_4x4_h = 1 << xd->bmode_blocks_hl;
int idx, idy;
PREDICTION_MODE b_mode;
int_mv best_sub8x8[2];
const uint32_t invalid_mv = 0x80008000;
// Initialize the 2nd element as even though it won't be used meaningfully
// if is_compound is false, copying/clamping it may trigger a MSan warning.
best_sub8x8[1].as_int = invalid_mv;
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, xd, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV) {
for (ref = 0; ref < 1 + is_compound; ++ref)
append_sub8x8_mvs_for_idx(cm, xd, mv_ref_search, b_mode, j, ref,
mi_row, mi_col, &best_sub8x8[ref]);
}
if (!assign_mv(cm, xd, b_mode, mi->bmi[j].as_mv, best_ref_mvs,
best_sub8x8, is_compound, allow_hp, r)) {
xd->corrupted |= 1;
break;
}
if (num_4x4_h == 2) mi->bmi[j + 2] = mi->bmi[j];
if (num_4x4_w == 2) mi->bmi[j + 1] = mi->bmi[j];
}
}
mi->mode = b_mode;
copy_mv_pair(mi->mv, mi->bmi[3].as_mv);
} else {
xd->corrupted |= !assign_mv(cm, xd, mi->mode, mi->mv, best_ref_mvs,
best_ref_mvs, is_compound, allow_hp, r);
}
}
2010-05-18 17:58:33 +02:00
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
static void read_inter_frame_mode_info(VP9Decoder *const pbi,
MACROBLOCKD *const xd, int mi_row,
int mi_col, vpx_reader *r, int x_mis,
int y_mis) {
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
VP9_COMMON *const cm = &pbi->common;
MODE_INFO *const mi = xd->mi[0];
int inter_block;
mi->segment_id =
read_inter_segment_id(cm, xd, mi_row, mi_col, r, x_mis, y_mis);
mi->skip = read_skip(cm, xd, mi->segment_id, r);
inter_block = read_is_inter_block(cm, xd, mi->segment_id, r);
mi->tx_size = read_tx_size(cm, xd, !mi->skip || !inter_block, r);
if (inter_block)
read_inter_block_mode_info(pbi, xd, mi, mi_row, mi_col, r);
else
read_intra_block_mode_info(cm, xd, mi, r);
}
static INLINE void copy_ref_frame_pair(MV_REFERENCE_FRAME *dst,
const MV_REFERENCE_FRAME *src) {
memcpy(dst, src, sizeof(*dst) * 2);
}
void vp9_read_mode_info(TileWorkerData *twd, VP9Decoder *const pbi, int mi_row,
int mi_col, int x_mis, int y_mis) {
vpx_reader *r = &twd->bit_reader;
MACROBLOCKD *const xd = &twd->xd;
Try again to merge branch 'frame-parallel' into master branch. In frame parallel decode, libvpx decoder decodes several frames on all cpus in parallel fashion. If not being flushed, it will only return frame when all the cpus are busy. If getting flushed, it will return all the frames in the decoder. Compare with current serial decode mode in which libvpx decoder is idle between decode calls, libvpx decoder is busy between decode calls. Current frame parallel decode will only speed up the decoding for frame parallel encoded videos. For non frame parallel encoded videos, frame parallel decode is slower than serial decode due to lack of loopfilter worker thread. There are still some known issues that need to be addressed. For example: decode frame parallel videos with segmentation enabled is not right sometimes. * frame-parallel: Add error handling for frame parallel decode and unit test for that. Fix a bug in frame parallel decode and add a unit test for that. Add two test vectors to test frame parallel decode. Add key frame seeking to webmdec and webm_video_source. Implement frame parallel decode for VP9. Increase the thread test range to cover 5, 6, 7, 8 threads. Fix a bug in adding frame parallel unit test. Add VP9 frame-parallel unit test. Manually pick "Make the api behavior conform to api spec." from master branch. Move vp9_dec_build_inter_predictors_* to decoder folder. Add segmentation map array for current and last frame segmentation. Include the right header for VP9 worker thread. Move vp9_thread.* to common. ctrl_get_reference does not need user_priv. Seperate the frame buffers from VP9 encoder/decoder structure. Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:""" Conflicts: test/codec_factory.h test/decode_test_driver.cc test/decode_test_driver.h test/invalid_file_test.cc test/test-data.sha1 test/test.mk test/test_vectors.cc vp8/vp8_dx_iface.c vp9/common/vp9_alloccommon.c vp9/common/vp9_entropymode.c vp9/common/vp9_loopfilter_thread.c vp9/common/vp9_loopfilter_thread.h vp9/common/vp9_mvref_common.c vp9/common/vp9_onyxc_int.h vp9/common/vp9_reconinter.c vp9/decoder/vp9_decodeframe.c vp9/decoder/vp9_decodeframe.h vp9/decoder/vp9_decodemv.c vp9/decoder/vp9_decoder.c vp9/decoder/vp9_decoder.h vp9/encoder/vp9_encoder.c vp9/encoder/vp9_pickmode.c vp9/encoder/vp9_rdopt.c vp9/vp9_cx_iface.c vp9/vp9_dx_iface.c This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5. Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-27 21:26:28 +01:00
VP9_COMMON *const cm = &pbi->common;
MODE_INFO *const mi = xd->mi[0];
MV_REF *frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
int w, h;
if (frame_is_intra_only(cm)) {
read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r, x_mis, y_mis);
} else {
read_inter_frame_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
for (h = 0; h < y_mis; ++h) {
for (w = 0; w < x_mis; ++w) {
MV_REF *const mv = frame_mvs + w;
copy_ref_frame_pair(mv->ref_frame, mi->ref_frame);
copy_mv_pair(mv->mv, mi->mv);
}
frame_mvs += cm->mi_cols;
}
}
#if 0 // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH
if ((xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) &&
(xd->above_mi == NULL || xd->left_mi == NULL) &&
!is_inter_block(mi) && need_top_left[mi->uv_mode])
assert(0);
#endif // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH
}