2012-08-20 23:43:34 +02:00
|
|
|
|
2012-01-28 11:07:08 +01:00
|
|
|
/*
|
2012-01-31 13:45:30 +01:00
|
|
|
* Copyright (c) 2012 The WebM project authors. All Rights Reserved.
|
2012-01-28 11:07:08 +01:00
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2013-04-19 23:25:32 +02:00
|
|
|
#include <limits.h>
|
|
|
|
|
2013-01-06 03:20:25 +01:00
|
|
|
#include "vp9/common/vp9_common.h"
|
2012-11-27 22:59:17 +01:00
|
|
|
#include "vp9/common/vp9_pred_common.h"
|
|
|
|
#include "vp9/common/vp9_seg_common.h"
|
Consistently use get_prob(), clip_prob() and newly added clip_pixel().
Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.
Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.
Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
2012-12-10 21:09:07 +01:00
|
|
|
#include "vp9/common/vp9_treecoder.h"
|
2012-01-28 11:07:08 +01:00
|
|
|
|
|
|
|
// TBD prediction functions for various bitstream signals
|
|
|
|
|
|
|
|
// Returns a context number for the given MB prediction signal
|
2012-10-31 01:53:32 +01:00
|
|
|
unsigned char vp9_get_pred_context(const VP9_COMMON *const cm,
|
2012-10-29 14:44:18 +01:00
|
|
|
const MACROBLOCKD *const xd,
|
|
|
|
PRED_ID pred_id) {
|
2012-07-14 00:21:29 +02:00
|
|
|
int pred_context;
|
2013-04-24 01:18:09 +02:00
|
|
|
const MODE_INFO *const mi = xd->mode_info_context;
|
|
|
|
const MODE_INFO *const above_mi = mi - cm->mode_info_stride;
|
|
|
|
const MODE_INFO *const left_mi = mi - 1;
|
2012-07-14 00:21:29 +02:00
|
|
|
// Note:
|
|
|
|
// The mode info data structure has a one element border above and to the
|
|
|
|
// left of the entries correpsonding to real macroblocks.
|
|
|
|
// The prediction flags in these dummy entries are initialised to 0.
|
|
|
|
switch (pred_id) {
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_SEG_ID:
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context = above_mi->mbmi.seg_id_predicted;
|
[WIP] Add column-based tiling.
This patch adds column-based tiling. The idea is to make each tile
independently decodable (after reading the common frame header) and
also independendly encodable (minus within-frame cost adjustments in
the RD loop) to speed-up hardware & software en/decoders if they used
multi-threading. Column-based tiling has the added advantage (over
other tiling methods) that it minimizes realtime use-case latency,
since all threads can start encoding data as soon as the first SB-row
worth of data is available to the encoder.
There is some test code that does random tile ordering in the decoder,
to confirm that each tile is indeed independently decodable from other
tiles in the same frame. At tile edges, all contexts assume default
values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode),
and motion vector search and ordering do not cross tiles in the same
frame.
t log
Tile independence is not maintained between frames ATM, i.e. tile 0 of
frame 1 is free to use motion vectors that point into any tile of frame
0. We support 1 (i.e. no tiling), 2 or 4 column-tiles.
The loopfilter crosses tile boundaries. I discussed this briefly with Aki
and he says that's OK. An in-loop loopfilter would need to do some sync
between tile threads, but that shouldn't be a big issue.
Resuls: with tiling disabled, we go up slightly because of improved edge
use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf,
~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5%
on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is
concentrated in the low-bitrate end of clips, and most of it is because
of the loss of edges at tile boundaries and the resulting loss of intra
predictors.
TODO:
- more tiles (perhaps allow row-based tiling also, and max. 8 tiles)?
- maybe optionally (for EC purposes), motion vectors themselves
should not cross tile edges, or we should emulate such borders as
if they were off-frame, to limit error propagation to within one
tile only. This doesn't have to be the default behaviour but could
be an optional bitstream flag.
Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
|
|
|
if (xd->left_available)
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context += left_mi->mbmi.seg_id_predicted;
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-01-28 11:07:08 +01:00
|
|
|
|
|
|
|
case PRED_REF:
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context = above_mi->mbmi.ref_predicted;
|
[WIP] Add column-based tiling.
This patch adds column-based tiling. The idea is to make each tile
independently decodable (after reading the common frame header) and
also independendly encodable (minus within-frame cost adjustments in
the RD loop) to speed-up hardware & software en/decoders if they used
multi-threading. Column-based tiling has the added advantage (over
other tiling methods) that it minimizes realtime use-case latency,
since all threads can start encoding data as soon as the first SB-row
worth of data is available to the encoder.
There is some test code that does random tile ordering in the decoder,
to confirm that each tile is indeed independently decodable from other
tiles in the same frame. At tile edges, all contexts assume default
values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode),
and motion vector search and ordering do not cross tiles in the same
frame.
t log
Tile independence is not maintained between frames ATM, i.e. tile 0 of
frame 1 is free to use motion vectors that point into any tile of frame
0. We support 1 (i.e. no tiling), 2 or 4 column-tiles.
The loopfilter crosses tile boundaries. I discussed this briefly with Aki
and he says that's OK. An in-loop loopfilter would need to do some sync
between tile threads, but that shouldn't be a big issue.
Resuls: with tiling disabled, we go up slightly because of improved edge
use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf,
~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5%
on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is
concentrated in the low-bitrate end of clips, and most of it is because
of the loss of edges at tile boundaries and the resulting loss of intra
predictors.
TODO:
- more tiles (perhaps allow row-based tiling also, and max. 8 tiles)?
- maybe optionally (for EC purposes), motion vectors themselves
should not cross tile edges, or we should emulate such borders as
if they were off-frame, to limit error propagation to within one
tile only. This doesn't have to be the default behaviour but could
be an optional bitstream flag.
Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
|
|
|
if (xd->left_available)
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context += left_mi->mbmi.ref_predicted;
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-02-02 18:30:27 +01:00
|
|
|
|
2012-02-29 02:25:45 +01:00
|
|
|
case PRED_COMP:
|
2013-04-24 01:18:09 +02:00
|
|
|
if (mi->mbmi.ref_frame == LAST_FRAME)
|
2012-07-14 00:21:29 +02:00
|
|
|
pred_context = 0;
|
|
|
|
else
|
|
|
|
pred_context = 1;
|
|
|
|
break;
|
2012-01-28 11:07:08 +01:00
|
|
|
|
2012-03-19 19:02:04 +01:00
|
|
|
case PRED_MBSKIP:
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context = above_mi->mbmi.mb_skip_coeff;
|
[WIP] Add column-based tiling.
This patch adds column-based tiling. The idea is to make each tile
independently decodable (after reading the common frame header) and
also independendly encodable (minus within-frame cost adjustments in
the RD loop) to speed-up hardware & software en/decoders if they used
multi-threading. Column-based tiling has the added advantage (over
other tiling methods) that it minimizes realtime use-case latency,
since all threads can start encoding data as soon as the first SB-row
worth of data is available to the encoder.
There is some test code that does random tile ordering in the decoder,
to confirm that each tile is indeed independently decodable from other
tiles in the same frame. At tile edges, all contexts assume default
values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode),
and motion vector search and ordering do not cross tiles in the same
frame.
t log
Tile independence is not maintained between frames ATM, i.e. tile 0 of
frame 1 is free to use motion vectors that point into any tile of frame
0. We support 1 (i.e. no tiling), 2 or 4 column-tiles.
The loopfilter crosses tile boundaries. I discussed this briefly with Aki
and he says that's OK. An in-loop loopfilter would need to do some sync
between tile threads, but that shouldn't be a big issue.
Resuls: with tiling disabled, we go up slightly because of improved edge
use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf,
~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5%
on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is
concentrated in the low-bitrate end of clips, and most of it is because
of the loss of edges at tile boundaries and the resulting loss of intra
predictors.
TODO:
- more tiles (perhaps allow row-based tiling also, and max. 8 tiles)?
- maybe optionally (for EC purposes), motion vectors themselves
should not cross tile edges, or we should emulate such borders as
if they were off-frame, to limit error propagation to within one
tile only. This doesn't have to be the default behaviour but could
be an optional bitstream flag.
Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
|
|
|
if (xd->left_available)
|
2013-04-24 01:18:09 +02:00
|
|
|
pred_context += left_mi->mbmi.mb_skip_coeff;
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-03-19 19:02:04 +01:00
|
|
|
|
2013-04-24 01:18:09 +02:00
|
|
|
case PRED_SWITCHABLE_INTERP: {
|
|
|
|
// left
|
|
|
|
const int left_in_image = xd->left_available && left_mi->mbmi.mb_in_image;
|
2013-04-24 21:14:58 +02:00
|
|
|
const int left_mv_pred = is_inter_mode(left_mi->mbmi.mode);
|
2013-04-24 01:18:09 +02:00
|
|
|
const int left_interp = left_in_image && left_mv_pred ?
|
|
|
|
vp9_switchable_interp_map[left_mi->mbmi.interp_filter] :
|
|
|
|
VP9_SWITCHABLE_FILTERS;
|
|
|
|
|
|
|
|
// above
|
2013-04-24 19:45:32 +02:00
|
|
|
const int above_in_image = xd->up_available && above_mi->mbmi.mb_in_image;
|
2013-04-24 21:14:58 +02:00
|
|
|
const int above_mv_pred = is_inter_mode(above_mi->mbmi.mode);
|
2013-04-24 01:18:09 +02:00
|
|
|
const int above_interp = above_in_image && above_mv_pred ?
|
|
|
|
vp9_switchable_interp_map[above_mi->mbmi.interp_filter] :
|
|
|
|
VP9_SWITCHABLE_FILTERS;
|
|
|
|
|
|
|
|
assert(left_interp != -1);
|
|
|
|
assert(above_interp != -1);
|
|
|
|
|
|
|
|
if (left_interp == above_interp)
|
|
|
|
pred_context = left_interp;
|
|
|
|
else if (left_interp == VP9_SWITCHABLE_FILTERS &&
|
|
|
|
above_interp != VP9_SWITCHABLE_FILTERS)
|
|
|
|
pred_context = above_interp;
|
|
|
|
else if (left_interp != VP9_SWITCHABLE_FILTERS &&
|
|
|
|
above_interp == VP9_SWITCHABLE_FILTERS)
|
|
|
|
pred_context = left_interp;
|
|
|
|
else
|
|
|
|
pred_context = VP9_SWITCHABLE_FILTERS;
|
|
|
|
|
2012-07-18 22:43:01 +02:00
|
|
|
break;
|
2013-04-24 01:18:09 +02:00
|
|
|
}
|
2012-07-18 22:43:01 +02:00
|
|
|
|
2012-01-28 11:07:08 +01:00
|
|
|
default:
|
2013-03-05 23:12:16 +01:00
|
|
|
pred_context = 0; // *** add error trap code.
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
|
|
|
}
|
2012-01-28 11:07:08 +01:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
return pred_context;
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// This function returns a context probability for coding a given
|
|
|
|
// prediction signal
|
2012-10-31 22:40:53 +01:00
|
|
|
vp9_prob vp9_get_pred_prob(const VP9_COMMON *const cm,
|
2012-10-29 14:44:18 +01:00
|
|
|
const MACROBLOCKD *const xd,
|
|
|
|
PRED_ID pred_id) {
|
2013-03-05 23:12:16 +01:00
|
|
|
const int pred_context = vp9_get_pred_context(cm, xd, pred_id);
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
switch (pred_id) {
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_SEG_ID:
|
2013-03-05 23:12:16 +01:00
|
|
|
return cm->segment_pred_probs[pred_context];
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_REF:
|
2013-03-05 23:12:16 +01:00
|
|
|
return cm->ref_pred_probs[pred_context];
|
2012-02-29 02:25:45 +01:00
|
|
|
case PRED_COMP:
|
2012-07-14 00:21:29 +02:00
|
|
|
// In keeping with convention elsewhre the probability returned is
|
|
|
|
// the probability of a "0" outcome which in this case means the
|
|
|
|
// probability of comp pred off.
|
2013-03-05 23:12:16 +01:00
|
|
|
return cm->prob_comppred[pred_context];
|
2012-03-19 19:02:04 +01:00
|
|
|
case PRED_MBSKIP:
|
2013-03-05 23:12:16 +01:00
|
|
|
return cm->mbskip_pred_probs[pred_context];
|
2012-01-28 11:07:08 +01:00
|
|
|
default:
|
2013-03-05 23:12:16 +01:00
|
|
|
return 128; // *** add error trap code.
|
2012-07-14 00:21:29 +02:00
|
|
|
}
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
2012-07-18 22:43:01 +02:00
|
|
|
// This function returns a context probability ptr for coding a given
|
|
|
|
// prediction signal
|
2012-10-31 22:40:53 +01:00
|
|
|
const vp9_prob *vp9_get_pred_probs(const VP9_COMMON *const cm,
|
2012-10-29 14:44:18 +01:00
|
|
|
const MACROBLOCKD *const xd,
|
|
|
|
PRED_ID pred_id) {
|
2013-03-05 23:12:16 +01:00
|
|
|
const int pred_context = vp9_get_pred_context(cm, xd, pred_id);
|
2012-07-18 22:43:01 +02:00
|
|
|
|
|
|
|
switch (pred_id) {
|
|
|
|
case PRED_SEG_ID:
|
2013-03-05 23:12:16 +01:00
|
|
|
return &cm->segment_pred_probs[pred_context];
|
2012-07-18 22:43:01 +02:00
|
|
|
case PRED_REF:
|
2013-03-05 23:12:16 +01:00
|
|
|
return &cm->ref_pred_probs[pred_context];
|
2012-07-18 22:43:01 +02:00
|
|
|
case PRED_COMP:
|
|
|
|
// In keeping with convention elsewhre the probability returned is
|
|
|
|
// the probability of a "0" outcome which in this case means the
|
|
|
|
// probability of comp pred off.
|
2013-03-05 23:12:16 +01:00
|
|
|
return &cm->prob_comppred[pred_context];
|
2012-07-18 22:43:01 +02:00
|
|
|
case PRED_MBSKIP:
|
2013-03-05 23:12:16 +01:00
|
|
|
return &cm->mbskip_pred_probs[pred_context];
|
2012-07-18 22:43:01 +02:00
|
|
|
case PRED_SWITCHABLE_INTERP:
|
2013-03-05 23:12:16 +01:00
|
|
|
return &cm->fc.switchable_interp_prob[pred_context][0];
|
2012-07-18 22:43:01 +02:00
|
|
|
default:
|
2013-03-05 23:12:16 +01:00
|
|
|
return NULL; // *** add error trap code.
|
2012-07-18 22:43:01 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-01-28 11:07:08 +01:00
|
|
|
// This function returns the status of the given prediction signal.
|
|
|
|
// I.e. is the predicted value for the given signal correct.
|
2012-10-29 14:44:18 +01:00
|
|
|
unsigned char vp9_get_pred_flag(const MACROBLOCKD *const xd,
|
|
|
|
PRED_ID pred_id) {
|
2012-07-14 00:21:29 +02:00
|
|
|
switch (pred_id) {
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_SEG_ID:
|
2013-03-05 23:12:16 +01:00
|
|
|
return xd->mode_info_context->mbmi.seg_id_predicted;
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_REF:
|
2013-03-05 23:12:16 +01:00
|
|
|
return xd->mode_info_context->mbmi.ref_predicted;
|
2012-03-19 19:02:04 +01:00
|
|
|
case PRED_MBSKIP:
|
2013-03-05 23:12:16 +01:00
|
|
|
return xd->mode_info_context->mbmi.mb_skip_coeff;
|
2012-01-28 11:07:08 +01:00
|
|
|
default:
|
2013-03-05 23:12:16 +01:00
|
|
|
return 0; // *** add error trap code.
|
2012-07-14 00:21:29 +02:00
|
|
|
}
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// This function sets the status of the given prediction signal.
|
|
|
|
// I.e. is the predicted value for the given signal correct.
|
2012-10-29 14:44:18 +01:00
|
|
|
void vp9_set_pred_flag(MACROBLOCKD *const xd,
|
|
|
|
PRED_ID pred_id,
|
|
|
|
unsigned char pred_flag) {
|
2012-10-30 01:58:18 +01:00
|
|
|
const int mis = xd->mode_info_stride;
|
2013-04-11 01:50:01 +02:00
|
|
|
BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
|
2013-04-26 20:57:17 +02:00
|
|
|
const int bh = 1 << mi_height_log2(bsize);
|
|
|
|
const int bw = 1 << mi_width_log2(bsize);
|
2013-04-11 01:50:01 +02:00
|
|
|
#define sub(a, b) (b) < 0 ? (a) + (b) : (a)
|
2013-04-26 20:57:17 +02:00
|
|
|
const int x_mis = sub(bw, xd->mb_to_right_edge >> (3 + LOG2_MI_SIZE));
|
|
|
|
const int y_mis = sub(bh, xd->mb_to_bottom_edge >> (3 + LOG2_MI_SIZE));
|
2013-04-11 01:50:01 +02:00
|
|
|
#undef sub
|
|
|
|
int x, y;
|
2012-10-30 01:58:18 +01:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
switch (pred_id) {
|
2012-01-28 11:07:08 +01:00
|
|
|
case PRED_SEG_ID:
|
2013-04-26 20:57:17 +02:00
|
|
|
for (y = 0; y < y_mis; y++) {
|
|
|
|
for (x = 0; x < x_mis; x++) {
|
|
|
|
xd->mode_info_context[y * mis + x].mbmi.seg_id_predicted = pred_flag;
|
2012-10-30 01:58:18 +01:00
|
|
|
}
|
2012-08-20 23:43:34 +02:00
|
|
|
}
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-01-28 11:07:08 +01:00
|
|
|
|
|
|
|
case PRED_REF:
|
2013-04-26 20:57:17 +02:00
|
|
|
for (y = 0; y < y_mis; y++) {
|
|
|
|
for (x = 0; x < x_mis; x++) {
|
2013-04-11 01:50:01 +02:00
|
|
|
xd->mode_info_context[y * mis + x].mbmi.ref_predicted = pred_flag;
|
2012-10-30 01:58:18 +01:00
|
|
|
}
|
2012-08-20 23:43:34 +02:00
|
|
|
}
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-01-28 11:07:08 +01:00
|
|
|
|
2012-03-19 19:02:04 +01:00
|
|
|
case PRED_MBSKIP:
|
2013-04-26 20:57:17 +02:00
|
|
|
for (y = 0; y < y_mis; y++) {
|
|
|
|
for (x = 0; x < x_mis; x++) {
|
2013-04-11 01:50:01 +02:00
|
|
|
xd->mode_info_context[y * mis + x].mbmi.mb_skip_coeff = pred_flag;
|
2012-10-30 01:58:18 +01:00
|
|
|
}
|
|
|
|
}
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
2012-03-19 19:02:04 +01:00
|
|
|
|
2012-01-28 11:07:08 +01:00
|
|
|
default:
|
2013-03-05 23:12:16 +01:00
|
|
|
// *** add error trap code.
|
2012-07-14 00:21:29 +02:00
|
|
|
break;
|
|
|
|
}
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
// The following contain the guts of the prediction code used to
|
|
|
|
// peredict various bitstream signals.
|
|
|
|
|
|
|
|
// Macroblock segment id prediction function
|
2013-04-26 20:57:17 +02:00
|
|
|
int vp9_get_pred_mi_segid(VP9_COMMON *cm, BLOCK_SIZE_TYPE sb_type,
|
|
|
|
int mi_row, int mi_col) {
|
|
|
|
const int mi_index = mi_row * cm->mi_cols + mi_col;
|
|
|
|
const int bw = 1 << mi_width_log2(sb_type);
|
|
|
|
const int bh = 1 << mi_height_log2(sb_type);
|
|
|
|
const int ymis = MIN(cm->mi_rows - mi_row, bh);
|
|
|
|
const int xmis = MIN(cm->mi_cols - mi_col, bw);
|
|
|
|
int segment_id = INT_MAX;
|
|
|
|
int x, y;
|
|
|
|
|
|
|
|
for (y = 0; y < ymis; y++) {
|
|
|
|
for (x = 0; x < xmis; x++) {
|
|
|
|
const int index = mi_index + (y * cm->mi_cols + x);
|
|
|
|
segment_id = MIN(segment_id, cm->last_frame_seg_map[index]);
|
2012-10-30 01:58:18 +01:00
|
|
|
}
|
|
|
|
}
|
2013-04-26 20:57:17 +02:00
|
|
|
return segment_id;
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
2012-10-31 01:53:32 +01:00
|
|
|
MV_REFERENCE_FRAME vp9_get_pred_ref(const VP9_COMMON *const cm,
|
2012-10-29 14:44:18 +01:00
|
|
|
const MACROBLOCKD *const xd) {
|
2012-07-14 00:21:29 +02:00
|
|
|
MODE_INFO *m = xd->mode_info_context;
|
|
|
|
|
|
|
|
MV_REFERENCE_FRAME left;
|
|
|
|
MV_REFERENCE_FRAME above;
|
|
|
|
MV_REFERENCE_FRAME above_left;
|
|
|
|
MV_REFERENCE_FRAME pred_ref = LAST_FRAME;
|
|
|
|
|
|
|
|
int segment_id = xd->mode_info_context->mbmi.segment_id;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
unsigned char frame_allowed[MAX_REF_FRAMES] = {1, 1, 1, 1};
|
|
|
|
unsigned char ref_score[MAX_REF_FRAMES];
|
|
|
|
unsigned char best_score = 0;
|
|
|
|
unsigned char left_in_image;
|
|
|
|
unsigned char above_in_image;
|
|
|
|
unsigned char above_left_in_image;
|
|
|
|
|
|
|
|
// Is segment coding ennabled
|
2013-03-05 23:12:16 +01:00
|
|
|
int seg_ref_active = vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME);
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// Special case treatment if segment coding is enabled.
|
|
|
|
// Dont allow prediction of a reference frame that the segment
|
|
|
|
// does not allow
|
|
|
|
if (seg_ref_active) {
|
|
|
|
for (i = 0; i < MAX_REF_FRAMES; i++) {
|
|
|
|
frame_allowed[i] =
|
2012-10-30 06:15:27 +01:00
|
|
|
vp9_check_segref(xd, segment_id, i);
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// Score set to 0 if ref frame not allowed
|
|
|
|
ref_score[i] = cm->ref_scores[i] * frame_allowed[i];
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
2012-07-14 00:21:29 +02:00
|
|
|
} else
|
|
|
|
vpx_memcpy(ref_score, cm->ref_scores, sizeof(ref_score));
|
|
|
|
|
|
|
|
// Reference frames used by neighbours
|
|
|
|
left = (m - 1)->mbmi.ref_frame;
|
|
|
|
above = (m - cm->mode_info_stride)->mbmi.ref_frame;
|
|
|
|
above_left = (m - 1 - cm->mode_info_stride)->mbmi.ref_frame;
|
|
|
|
|
|
|
|
// Are neighbours in image
|
[WIP] Add column-based tiling.
This patch adds column-based tiling. The idea is to make each tile
independently decodable (after reading the common frame header) and
also independendly encodable (minus within-frame cost adjustments in
the RD loop) to speed-up hardware & software en/decoders if they used
multi-threading. Column-based tiling has the added advantage (over
other tiling methods) that it minimizes realtime use-case latency,
since all threads can start encoding data as soon as the first SB-row
worth of data is available to the encoder.
There is some test code that does random tile ordering in the decoder,
to confirm that each tile is indeed independently decodable from other
tiles in the same frame. At tile edges, all contexts assume default
values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode),
and motion vector search and ordering do not cross tiles in the same
frame.
t log
Tile independence is not maintained between frames ATM, i.e. tile 0 of
frame 1 is free to use motion vectors that point into any tile of frame
0. We support 1 (i.e. no tiling), 2 or 4 column-tiles.
The loopfilter crosses tile boundaries. I discussed this briefly with Aki
and he says that's OK. An in-loop loopfilter would need to do some sync
between tile threads, but that shouldn't be a big issue.
Resuls: with tiling disabled, we go up slightly because of improved edge
use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf,
~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5%
on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is
concentrated in the low-bitrate end of clips, and most of it is because
of the loss of edges at tile boundaries and the resulting loss of intra
predictors.
TODO:
- more tiles (perhaps allow row-based tiling also, and max. 8 tiles)?
- maybe optionally (for EC purposes), motion vectors themselves
should not cross tile edges, or we should emulate such borders as
if they were off-frame, to limit error propagation to within one
tile only. This doesn't have to be the default behaviour but could
be an optional bitstream flag.
Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
|
|
|
left_in_image = (m - 1)->mbmi.mb_in_image && xd->left_available;
|
2012-07-14 00:21:29 +02:00
|
|
|
above_in_image = (m - cm->mode_info_stride)->mbmi.mb_in_image;
|
[WIP] Add column-based tiling.
This patch adds column-based tiling. The idea is to make each tile
independently decodable (after reading the common frame header) and
also independendly encodable (minus within-frame cost adjustments in
the RD loop) to speed-up hardware & software en/decoders if they used
multi-threading. Column-based tiling has the added advantage (over
other tiling methods) that it minimizes realtime use-case latency,
since all threads can start encoding data as soon as the first SB-row
worth of data is available to the encoder.
There is some test code that does random tile ordering in the decoder,
to confirm that each tile is indeed independently decodable from other
tiles in the same frame. At tile edges, all contexts assume default
values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode),
and motion vector search and ordering do not cross tiles in the same
frame.
t log
Tile independence is not maintained between frames ATM, i.e. tile 0 of
frame 1 is free to use motion vectors that point into any tile of frame
0. We support 1 (i.e. no tiling), 2 or 4 column-tiles.
The loopfilter crosses tile boundaries. I discussed this briefly with Aki
and he says that's OK. An in-loop loopfilter would need to do some sync
between tile threads, but that shouldn't be a big issue.
Resuls: with tiling disabled, we go up slightly because of improved edge
use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf,
~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5%
on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is
concentrated in the low-bitrate end of clips, and most of it is because
of the loss of edges at tile boundaries and the resulting loss of intra
predictors.
TODO:
- more tiles (perhaps allow row-based tiling also, and max. 8 tiles)?
- maybe optionally (for EC purposes), motion vectors themselves
should not cross tile edges, or we should emulate such borders as
if they were off-frame, to limit error propagation to within one
tile only. This doesn't have to be the default behaviour but could
be an optional bitstream flag.
Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
|
|
|
above_left_in_image = (m - 1 - cm->mode_info_stride)->mbmi.mb_in_image &&
|
|
|
|
xd->left_available;
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// Adjust scores for candidate reference frames based on neigbours
|
|
|
|
if (frame_allowed[left] && left_in_image) {
|
|
|
|
ref_score[left] += 16;
|
|
|
|
if (above_left_in_image && (left == above_left))
|
|
|
|
ref_score[left] += 4;
|
|
|
|
}
|
|
|
|
if (frame_allowed[above] && above_in_image) {
|
|
|
|
ref_score[above] += 16;
|
|
|
|
if (above_left_in_image && (above == above_left))
|
|
|
|
ref_score[above] += 4;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now choose the candidate with the highest score
|
|
|
|
for (i = 0; i < MAX_REF_FRAMES; i++) {
|
|
|
|
if (ref_score[i] > best_score) {
|
|
|
|
pred_ref = i;
|
|
|
|
best_score = ref_score[i];
|
2012-01-31 13:45:30 +01:00
|
|
|
}
|
2012-07-14 00:21:29 +02:00
|
|
|
}
|
2012-01-28 11:07:08 +01:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
return pred_ref;
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// Functions to computes a set of modified reference frame probabilities
|
|
|
|
// to use when the prediction of the reference frame value fails
|
2012-10-31 22:40:53 +01:00
|
|
|
void vp9_calc_ref_probs(int *count, vp9_prob *probs) {
|
2013-03-05 23:12:16 +01:00
|
|
|
int tot_count = count[0] + count[1] + count[2] + count[3];
|
Consistently use get_prob(), clip_prob() and newly added clip_pixel().
Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.
Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.
Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
2012-12-10 21:09:07 +01:00
|
|
|
probs[0] = get_prob(count[0], tot_count);
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
tot_count -= count[0];
|
Consistently use get_prob(), clip_prob() and newly added clip_pixel().
Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.
Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.
Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
2012-12-10 21:09:07 +01:00
|
|
|
probs[1] = get_prob(count[1], tot_count);
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
tot_count -= count[1];
|
Consistently use get_prob(), clip_prob() and newly added clip_pixel().
Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.
Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.
Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
2012-12-10 21:09:07 +01:00
|
|
|
probs[2] = get_prob(count[2], tot_count);
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|
|
|
|
|
2012-01-31 13:45:30 +01:00
|
|
|
// Computes a set of modified conditional probabilities for the reference frame
|
|
|
|
// Values willbe set to 0 for reference frame options that are not possible
|
|
|
|
// because wither they were predicted and prediction has failed or because
|
|
|
|
// they are not allowed for a given segment.
|
2012-10-31 01:53:32 +01:00
|
|
|
void vp9_compute_mod_refprobs(VP9_COMMON *const cm) {
|
2012-07-14 00:21:29 +02:00
|
|
|
int norm_cnt[MAX_REF_FRAMES];
|
2013-03-05 23:12:16 +01:00
|
|
|
const int intra_count = cm->prob_intra_coded;
|
|
|
|
const int inter_count = (255 - intra_count);
|
|
|
|
const int last_count = (inter_count * cm->prob_last_coded) / 255;
|
|
|
|
const int gfarf_count = inter_count - last_count;
|
|
|
|
const int gf_count = (gfarf_count * cm->prob_gf_coded) / 255;
|
|
|
|
const int arf_count = gfarf_count - gf_count;
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// Work out modified reference frame probabilities to use where prediction
|
|
|
|
// of the reference frame fails
|
|
|
|
norm_cnt[0] = 0;
|
|
|
|
norm_cnt[1] = last_count;
|
|
|
|
norm_cnt[2] = gf_count;
|
|
|
|
norm_cnt[3] = arf_count;
|
2012-10-29 14:44:18 +01:00
|
|
|
vp9_calc_ref_probs(norm_cnt, cm->mod_refprobs[INTRA_FRAME]);
|
2012-07-14 00:21:29 +02:00
|
|
|
cm->mod_refprobs[INTRA_FRAME][0] = 0; // This branch implicit
|
|
|
|
|
|
|
|
norm_cnt[0] = intra_count;
|
|
|
|
norm_cnt[1] = 0;
|
|
|
|
norm_cnt[2] = gf_count;
|
|
|
|
norm_cnt[3] = arf_count;
|
2012-10-29 14:44:18 +01:00
|
|
|
vp9_calc_ref_probs(norm_cnt, cm->mod_refprobs[LAST_FRAME]);
|
2012-07-14 00:21:29 +02:00
|
|
|
cm->mod_refprobs[LAST_FRAME][1] = 0; // This branch implicit
|
|
|
|
|
|
|
|
norm_cnt[0] = intra_count;
|
|
|
|
norm_cnt[1] = last_count;
|
|
|
|
norm_cnt[2] = 0;
|
|
|
|
norm_cnt[3] = arf_count;
|
2012-10-29 14:44:18 +01:00
|
|
|
vp9_calc_ref_probs(norm_cnt, cm->mod_refprobs[GOLDEN_FRAME]);
|
2012-07-14 00:21:29 +02:00
|
|
|
cm->mod_refprobs[GOLDEN_FRAME][2] = 0; // This branch implicit
|
|
|
|
|
|
|
|
norm_cnt[0] = intra_count;
|
|
|
|
norm_cnt[1] = last_count;
|
|
|
|
norm_cnt[2] = gf_count;
|
|
|
|
norm_cnt[3] = 0;
|
2012-10-29 14:44:18 +01:00
|
|
|
vp9_calc_ref_probs(norm_cnt, cm->mod_refprobs[ALTREF_FRAME]);
|
2012-07-14 00:21:29 +02:00
|
|
|
cm->mod_refprobs[ALTREF_FRAME][2] = 0; // This branch implicit
|
|
|
|
|
|
|
|
// Score the reference frames based on overal frequency.
|
|
|
|
// These scores contribute to the prediction choices.
|
|
|
|
// Max score 17 min 1
|
|
|
|
cm->ref_scores[INTRA_FRAME] = 1 + (intra_count * 16 / 255);
|
|
|
|
cm->ref_scores[LAST_FRAME] = 1 + (last_count * 16 / 255);
|
|
|
|
cm->ref_scores[GOLDEN_FRAME] = 1 + (gf_count * 16 / 255);
|
|
|
|
cm->ref_scores[ALTREF_FRAME] = 1 + (arf_count * 16 / 255);
|
2012-01-28 11:07:08 +01:00
|
|
|
}
|