2010-05-18 17:58:33 +02:00
|
|
|
/*
|
2010-09-09 14:16:39 +02:00
|
|
|
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
|
2010-05-18 17:58:33 +02:00
|
|
|
*
|
2010-06-18 18:39:21 +02:00
|
|
|
* Use of this source code is governed by a BSD-style license
|
2010-06-04 22:19:40 +02:00
|
|
|
* that can be found in the LICENSE file in the root of the source
|
|
|
|
* tree. An additional intellectual property rights grant can be found
|
2010-06-18 18:39:21 +02:00
|
|
|
* in the file PATENTS. All contributing project authors may
|
2010-06-04 22:19:40 +02:00
|
|
|
* be found in the AUTHORS file in the root of the source tree.
|
2010-05-18 17:58:33 +02:00
|
|
|
*/
|
|
|
|
|
2012-11-30 01:36:10 +01:00
|
|
|
#ifndef VP9_ENCODER_VP9_BLOCK_H_
|
|
|
|
#define VP9_ENCODER_VP9_BLOCK_H_
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-11-27 22:59:17 +01:00
|
|
|
#include "vp9/common/vp9_onyx.h"
|
|
|
|
#include "vp9/common/vp9_entropymv.h"
|
|
|
|
#include "vp9/common/vp9_entropy.h"
|
2010-05-18 17:58:33 +02:00
|
|
|
#include "vpx_ports/mem.h"
|
2012-11-27 22:59:17 +01:00
|
|
|
#include "vp9/common/vp9_onyxc_int.h"
|
2010-05-18 17:58:33 +02:00
|
|
|
|
|
|
|
// motion search site
|
2012-07-14 00:21:29 +02:00
|
|
|
typedef struct {
|
|
|
|
MV mv;
|
|
|
|
int offset;
|
2010-05-18 17:58:33 +02:00
|
|
|
} search_site;
|
|
|
|
|
2012-10-28 18:38:23 +01:00
|
|
|
typedef struct block {
|
2012-07-14 00:21:29 +02:00
|
|
|
// 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
|
2012-12-19 00:31:19 +01:00
|
|
|
int16_t *src_diff;
|
|
|
|
int16_t *coeff;
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
|
2012-12-19 00:31:19 +01:00
|
|
|
int16_t *quant;
|
|
|
|
int16_t *quant_fast; // fast quant deprecated for now
|
|
|
|
uint8_t *quant_shift;
|
|
|
|
int16_t *zbin;
|
|
|
|
int16_t *zbin_8x8;
|
|
|
|
int16_t *zbin_16x16;
|
|
|
|
int16_t *zbin_32x32;
|
|
|
|
int16_t *zrun_zbin_boost;
|
|
|
|
int16_t *zrun_zbin_boost_8x8;
|
|
|
|
int16_t *zrun_zbin_boost_16x16;
|
|
|
|
int16_t *zrun_zbin_boost_32x32;
|
|
|
|
int16_t *round;
|
2012-07-14 00:21:29 +02:00
|
|
|
|
|
|
|
// Zbin Over Quant value
|
|
|
|
short zbin_extra;
|
|
|
|
|
2012-12-19 00:31:19 +01:00
|
|
|
uint8_t **base_src;
|
|
|
|
uint8_t **base_second_src;
|
2012-07-14 00:21:29 +02:00
|
|
|
int src;
|
|
|
|
int src_stride;
|
|
|
|
|
2013-01-29 14:33:17 +01:00
|
|
|
int skip_block;
|
2010-05-18 17:58:33 +02:00
|
|
|
} BLOCK;
|
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
typedef struct {
|
|
|
|
int count;
|
|
|
|
struct {
|
|
|
|
B_PREDICTION_MODE mode;
|
|
|
|
int_mv mv;
|
|
|
|
int_mv second_mv;
|
|
|
|
} bmi[16];
|
2010-09-02 22:17:52 +02:00
|
|
|
} PARTITION_INFO;
|
|
|
|
|
2012-04-07 01:38:34 +02:00
|
|
|
// Structure to hold snapshot of coding context during the mode picking process
|
|
|
|
// TODO Do we need all of these?
|
2012-07-14 00:21:29 +02:00
|
|
|
typedef struct {
|
|
|
|
MODE_INFO mic;
|
|
|
|
PARTITION_INFO partition_info;
|
2013-01-07 20:02:14 +01:00
|
|
|
int skip;
|
2012-07-14 00:21:29 +02:00
|
|
|
int_mv best_ref_mv;
|
|
|
|
int_mv second_best_ref_mv;
|
2012-12-04 18:21:05 +01:00
|
|
|
int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
|
2012-07-14 00:21:29 +02:00
|
|
|
int rate;
|
|
|
|
int distortion;
|
2012-08-02 20:01:48 +02:00
|
|
|
int64_t intra_error;
|
2012-07-14 00:21:29 +02:00
|
|
|
int best_mode_index;
|
|
|
|
int rddiv;
|
|
|
|
int rdmult;
|
2012-08-20 23:43:34 +02:00
|
|
|
int hybrid_pred_diff;
|
|
|
|
int comp_pred_diff;
|
|
|
|
int single_pred_diff;
|
2012-10-09 18:18:21 +02:00
|
|
|
int64_t txfm_rd_diff[NB_TXFM_MODES];
|
2012-04-07 01:38:34 +02:00
|
|
|
} PICK_MODE_CONTEXT;
|
|
|
|
|
32x32 transform for superblocks.
This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds
code all over the place to wrap that in the bitstream/encoder/decoder/RD.
Some implementation notes (these probably need careful review):
- token range is extended by 1 bit, since the value range out of this
transform is [-16384,16383].
- the coefficients coming out of the FDCT are manually scaled back by
1 bit, or else they won't fit in int16_t (they are 17 bits). Because
of this, the RD error scoring does not right-shift the MSE score by
two (unlike for 4x4/8x8/16x16).
- to compensate for this loss in precision, the quantizer is halved
also. This is currently a little hacky.
- FDCT and IDCT is double-only right now. Needs a fixed-point impl.
- There are no default probabilities for the 32x32 transform yet; I'm
simply using the 16x16 luma ones. A future commit will add newly
generated probabilities for all transforms.
- No ADST version. I don't think we'll add one for this level; if an
ADST is desired, transform-size selection can scale back to 16x16
or lower, and use an ADST at that level.
Additional notes specific to Debargha's DWT/DCT hybrid:
- coefficient scale is different for the top/left 16x16 (DCT-over-DWT)
block than for the rest (DWT pixel differences) of the block. Therefore,
RD error scoring isn't easily scalable between coefficient and pixel
domain. Thus, unfortunately, we need to compute the RD distortion in
the pixel domain until we figure out how to scale these appropriately.
Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
|
|
|
typedef struct superblock {
|
2012-12-19 00:31:19 +01:00
|
|
|
DECLARE_ALIGNED(16, int16_t, src_diff[32*32+16*16*2]);
|
|
|
|
DECLARE_ALIGNED(16, int16_t, coeff[32*32+16*16*2]);
|
32x32 transform for superblocks.
This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds
code all over the place to wrap that in the bitstream/encoder/decoder/RD.
Some implementation notes (these probably need careful review):
- token range is extended by 1 bit, since the value range out of this
transform is [-16384,16383].
- the coefficients coming out of the FDCT are manually scaled back by
1 bit, or else they won't fit in int16_t (they are 17 bits). Because
of this, the RD error scoring does not right-shift the MSE score by
two (unlike for 4x4/8x8/16x16).
- to compensate for this loss in precision, the quantizer is halved
also. This is currently a little hacky.
- FDCT and IDCT is double-only right now. Needs a fixed-point impl.
- There are no default probabilities for the 32x32 transform yet; I'm
simply using the 16x16 luma ones. A future commit will add newly
generated probabilities for all transforms.
- No ADST version. I don't think we'll add one for this level; if an
ADST is desired, transform-size selection can scale back to 16x16
or lower, and use an ADST at that level.
Additional notes specific to Debargha's DWT/DCT hybrid:
- coefficient scale is different for the top/left 16x16 (DCT-over-DWT)
block than for the rest (DWT pixel differences) of the block. Therefore,
RD error scoring isn't easily scalable between coefficient and pixel
domain. Thus, unfortunately, we need to compute the RD distortion in
the pixel domain until we figure out how to scale these appropriately.
Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
|
|
|
} SUPERBLOCK;
|
|
|
|
|
2012-10-28 18:38:23 +01:00
|
|
|
typedef struct macroblock {
|
2012-12-19 00:31:19 +01:00
|
|
|
DECLARE_ALIGNED(16, int16_t, src_diff[400]); // 16x16 Y 8x8 U 8x8 V 4x4 2nd Y
|
|
|
|
DECLARE_ALIGNED(16, int16_t, coeff[400]); // 16x16 Y 8x8 U 8x8 V 4x4 2nd Y
|
2012-07-14 00:21:29 +02:00
|
|
|
// 16 Y blocks, 4 U blocks, 4 V blocks,
|
|
|
|
// 1 DC 2nd order block each with 16 entries
|
|
|
|
BLOCK block[25];
|
|
|
|
|
32x32 transform for superblocks.
This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds
code all over the place to wrap that in the bitstream/encoder/decoder/RD.
Some implementation notes (these probably need careful review):
- token range is extended by 1 bit, since the value range out of this
transform is [-16384,16383].
- the coefficients coming out of the FDCT are manually scaled back by
1 bit, or else they won't fit in int16_t (they are 17 bits). Because
of this, the RD error scoring does not right-shift the MSE score by
two (unlike for 4x4/8x8/16x16).
- to compensate for this loss in precision, the quantizer is halved
also. This is currently a little hacky.
- FDCT and IDCT is double-only right now. Needs a fixed-point impl.
- There are no default probabilities for the 32x32 transform yet; I'm
simply using the 16x16 luma ones. A future commit will add newly
generated probabilities for all transforms.
- No ADST version. I don't think we'll add one for this level; if an
ADST is desired, transform-size selection can scale back to 16x16
or lower, and use an ADST at that level.
Additional notes specific to Debargha's DWT/DCT hybrid:
- coefficient scale is different for the top/left 16x16 (DCT-over-DWT)
block than for the rest (DWT pixel differences) of the block. Therefore,
RD error scoring isn't easily scalable between coefficient and pixel
domain. Thus, unfortunately, we need to compute the RD distortion in
the pixel domain until we figure out how to scale these appropriately.
Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
|
|
|
SUPERBLOCK sb_coeff_data;
|
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
YV12_BUFFER_CONFIG src;
|
|
|
|
|
|
|
|
MACROBLOCKD e_mbd;
|
|
|
|
PARTITION_INFO *partition_info; /* work pointer */
|
|
|
|
PARTITION_INFO *pi; /* Corresponds to upper left visible macroblock */
|
|
|
|
PARTITION_INFO *pip; /* Base of allocated array */
|
|
|
|
|
|
|
|
search_site *ss;
|
|
|
|
int ss_count;
|
|
|
|
int searches_per_step;
|
|
|
|
|
|
|
|
int errorperbit;
|
|
|
|
int sadperbit16;
|
|
|
|
int sadperbit4;
|
|
|
|
int rddiv;
|
|
|
|
int rdmult;
|
|
|
|
unsigned int *mb_activity_ptr;
|
|
|
|
int *mb_norm_activity_ptr;
|
|
|
|
signed int act_zbin_adj;
|
|
|
|
|
2012-12-05 17:23:38 +01:00
|
|
|
int mv_best_ref_index[MAX_REF_FRAMES];
|
|
|
|
|
2012-07-26 22:42:07 +02:00
|
|
|
int nmvjointcost[MV_JOINTS];
|
|
|
|
int nmvcosts[2][MV_VALS];
|
|
|
|
int *nmvcost[2];
|
|
|
|
int nmvcosts_hp[2][MV_VALS];
|
|
|
|
int *nmvcost_hp[2];
|
2012-11-09 00:44:39 +01:00
|
|
|
int **mvcost;
|
2012-07-26 22:42:07 +02:00
|
|
|
|
|
|
|
int nmvjointsadcost[MV_JOINTS];
|
|
|
|
int nmvsadcosts[2][MV_VALS];
|
|
|
|
int *nmvsadcost[2];
|
|
|
|
int nmvsadcosts_hp[2][MV_VALS];
|
|
|
|
int *nmvsadcost_hp[2];
|
2012-11-09 00:44:39 +01:00
|
|
|
int **mvsadcost;
|
2012-08-03 21:17:18 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
int mbmode_cost[2][MB_MODE_COUNT];
|
|
|
|
int intra_uv_mode_cost[2][MB_MODE_COUNT];
|
2012-10-09 22:19:15 +02:00
|
|
|
int bmode_costs[VP9_KF_BINTRAMODES][VP9_KF_BINTRAMODES][VP9_KF_BINTRAMODES];
|
2012-07-14 00:21:29 +02:00
|
|
|
int i8x8_mode_costs[MB_MODE_COUNT];
|
|
|
|
int inter_bmode_costs[B_MODE_COUNT];
|
2012-10-31 22:40:53 +01:00
|
|
|
int switchable_interp_costs[VP9_SWITCHABLE_FILTERS + 1]
|
|
|
|
[VP9_SWITCHABLE_FILTERS];
|
2012-07-14 00:21:29 +02:00
|
|
|
|
2012-10-22 23:43:01 +02:00
|
|
|
// These define limits to motion vector components to prevent them
|
|
|
|
// from extending outside the UMV borders
|
2012-07-14 00:21:29 +02:00
|
|
|
int mv_col_min;
|
|
|
|
int mv_col_max;
|
|
|
|
int mv_row_min;
|
|
|
|
int mv_row_max;
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
int skip;
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
int encode_breakout;
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
// char * gf_active_ptr;
|
|
|
|
signed char *gf_active_ptr;
|
2010-08-11 17:02:31 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
unsigned char *active_ptr;
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-12-08 01:09:59 +01:00
|
|
|
vp9_coeff_count token_costs[TX_SIZE_MAX_SB][BLOCK_TYPES_4X4];
|
|
|
|
vp9_coeff_count hybrid_token_costs[TX_SIZE_MAX_SB][BLOCK_TYPES_4X4];
|
2012-01-26 23:36:20 +01:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
int optimize;
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-07-14 00:21:29 +02:00
|
|
|
// Structure to hold context for each of the 4 MBs within a SB:
|
|
|
|
// when encoded as 4 independent MBs:
|
2013-01-06 03:20:25 +01:00
|
|
|
PICK_MODE_CONTEXT mb_context[4][4];
|
2012-07-14 00:21:29 +02:00
|
|
|
// when 4 MBs share coding parameters:
|
2013-01-06 03:20:25 +01:00
|
|
|
PICK_MODE_CONTEXT sb32_context[4];
|
|
|
|
PICK_MODE_CONTEXT sb64_context;
|
2012-04-07 01:38:34 +02:00
|
|
|
|
2012-12-19 00:31:19 +01:00
|
|
|
void (*vp9_short_fdct4x4)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*vp9_short_fdct8x4)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*short_walsh4x4)(int16_t *input, int16_t *output, int pitch);
|
2012-10-13 06:41:58 +02:00
|
|
|
void (*quantize_b_4x4)(BLOCK *b, BLOCKD *d);
|
|
|
|
void (*quantize_b_4x4_pair)(BLOCK *b1, BLOCK *b2, BLOCKD *d0, BLOCKD *d1);
|
2012-12-19 00:31:19 +01:00
|
|
|
void (*vp9_short_fdct8x8)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*vp9_short_fdct16x16)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*short_fhaar2x2)(int16_t *input, int16_t *output, int pitch);
|
2012-08-03 02:03:14 +02:00
|
|
|
void (*quantize_b_16x16)(BLOCK *b, BLOCKD *d);
|
2012-07-14 00:21:29 +02:00
|
|
|
void (*quantize_b_8x8)(BLOCK *b, BLOCKD *d);
|
|
|
|
void (*quantize_b_2x2)(BLOCK *b, BLOCKD *d);
|
2010-05-18 17:58:33 +02:00
|
|
|
} MACROBLOCK;
|
|
|
|
|
2012-12-19 00:31:19 +01:00
|
|
|
#endif // VP9_ENCODER_VP9_BLOCK_H_
|