2010-05-18 19:58:33 +04:00
|
|
|
/*
|
2010-09-09 16:16:39 +04:00
|
|
|
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
|
2010-05-18 19:58:33 +04:00
|
|
|
*
|
2010-06-18 20:39:21 +04:00
|
|
|
* Use of this source code is governed by a BSD-style license
|
2010-06-05 00:19:40 +04:00
|
|
|
* 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 20:39:21 +04:00
|
|
|
* in the file PATENTS. All contributing project authors may
|
2010-06-05 00:19:40 +04:00
|
|
|
* be found in the AUTHORS file in the root of the source tree.
|
2010-05-18 19:58:33 +04:00
|
|
|
*/
|
|
|
|
|
|
|
|
|
2012-11-30 04:36:10 +04:00
|
|
|
#ifndef VP9_COMMON_VP9_BLOCKD_H_
|
|
|
|
#define VP9_COMMON_VP9_BLOCKD_H_
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-12-23 19:20:10 +04:00
|
|
|
#include "./vpx_config.h"
|
2010-05-18 19:58:33 +04:00
|
|
|
#include "vpx_scale/yv12config.h"
|
Convert subpixel filters to use convolve framework
Update the code to call the new convolution functions to do subpixel
prediction rather than the existing functions. Remove the old C and
assembly code, since it is unused. This causes a 50% performance
reduction on the decoder, but that will be resolved when the asm for
the new functions is available.
There is no consensus for whether 6-tap or 2-tap predictors will be
supported in the final codec, so these filters are implemented in
terms of the 8-tap code, so that quality testing of these modes
can continue. Implementing the lower complexity algorithms is a
simple exercise, should it be necessary.
This code produces slightly better results in the EIGHTTAP_SMOOTH
case, since the filter is now applied in only one direction when
the subpel motion is only in one direction. Like the previous code,
the filtering is skipped entirely on full-pel MVs. This combination
seems to give the best quality gains, but this may be indicative of a
bug in the encoder's filter selection, since the encoder could
achieve the result of skipping the filtering on full-pel by selecting
one of the other filters. This should be revisited.
Quality gains on derf positive on almost all clips. The only clip
that seemed to be hurt at all datarates was football
(-0.115% PSNR average, -0.587% min). Overall averages 0.375% PSNR,
0.347% SSIM.
Change-Id: I7d469716091b1d89b4b08adde5863999319d69ff
2013-01-29 04:59:03 +04:00
|
|
|
#include "vp9/common/vp9_convolve.h"
|
2012-11-28 22:41:40 +04:00
|
|
|
#include "vp9/common/vp9_mv.h"
|
|
|
|
#include "vp9/common/vp9_treecoder.h"
|
2010-05-18 19:58:33 +04:00
|
|
|
#include "vpx_ports/mem.h"
|
2012-11-28 22:41:40 +04:00
|
|
|
#include "vp9/common/vp9_common.h"
|
2013-04-10 08:28:27 +04:00
|
|
|
#include "vp9/common/vp9_enums.h"
|
2010-05-18 19:58:33 +04:00
|
|
|
|
|
|
|
#define TRUE 1
|
|
|
|
#define FALSE 0
|
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
// #define MODE_STATS
|
2011-08-05 03:30:27 +04:00
|
|
|
|
2010-05-18 19:58:33 +04:00
|
|
|
#define MB_FEATURE_TREE_PROBS 3
|
2012-01-27 22:29:07 +04:00
|
|
|
#define PREDICTION_PROBS 3
|
2010-09-01 04:43:14 +04:00
|
|
|
|
2012-03-19 22:02:04 +04:00
|
|
|
#define MBSKIP_CONTEXTS 3
|
|
|
|
|
2010-05-18 19:58:33 +04:00
|
|
|
#define MAX_MB_SEGMENTS 4
|
|
|
|
|
|
|
|
#define MAX_REF_LF_DELTAS 4
|
|
|
|
#define MAX_MODE_LF_DELTAS 4
|
|
|
|
|
2010-10-28 03:04:02 +04:00
|
|
|
/* Segment Feature Masks */
|
2010-05-18 19:58:33 +04:00
|
|
|
#define SEGMENT_DELTADATA 0
|
|
|
|
#define SEGMENT_ABSDATA 1
|
2012-11-05 16:32:49 +04:00
|
|
|
#define MAX_MV_REFS 9
|
2012-12-04 21:21:05 +04:00
|
|
|
#define MAX_MV_REF_CANDIDATES 4
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2013-02-20 23:36:31 +04:00
|
|
|
typedef enum {
|
2012-10-15 02:29:56 +04:00
|
|
|
PLANE_TYPE_Y_WITH_DC,
|
2013-02-16 00:09:05 +04:00
|
|
|
PLANE_TYPE_UV,
|
2012-10-15 02:29:56 +04:00
|
|
|
} PLANE_TYPE;
|
2011-02-24 00:37:08 +03:00
|
|
|
|
2010-08-31 18:49:57 +04:00
|
|
|
typedef char ENTROPY_CONTEXT;
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef struct {
|
|
|
|
ENTROPY_CONTEXT y1[4];
|
|
|
|
ENTROPY_CONTEXT u[2];
|
|
|
|
ENTROPY_CONTEXT v[2];
|
2010-08-31 18:49:57 +04:00
|
|
|
} ENTROPY_CONTEXT_PLANES;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2013-02-20 23:36:31 +04:00
|
|
|
#define VP9_COMBINEENTROPYCONTEXTS(Dest, A, B) \
|
2012-07-14 02:21:29 +04:00
|
|
|
Dest = ((A)!=0) + ((B)!=0);
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef enum {
|
|
|
|
KEY_FRAME = 0,
|
|
|
|
INTER_FRAME = 1
|
2010-05-18 19:58:33 +04:00
|
|
|
} FRAME_TYPE;
|
|
|
|
|
2013-02-27 23:17:38 +04:00
|
|
|
typedef enum {
|
2013-01-09 02:14:01 +04:00
|
|
|
#if CONFIG_ENABLE_6TAP
|
|
|
|
SIXTAP,
|
|
|
|
#endif
|
|
|
|
EIGHTTAP_SMOOTH,
|
|
|
|
EIGHTTAP,
|
|
|
|
EIGHTTAP_SHARP,
|
|
|
|
BILINEAR,
|
2012-07-19 00:43:01 +04:00
|
|
|
SWITCHABLE /* should be the last one */
|
|
|
|
} INTERPOLATIONFILTERTYPE;
|
|
|
|
|
2013-02-27 23:17:38 +04:00
|
|
|
typedef enum {
|
2012-07-14 02:21:29 +04:00
|
|
|
DC_PRED, /* average of above and left pixels */
|
|
|
|
V_PRED, /* vertical prediction */
|
|
|
|
H_PRED, /* horizontal prediction */
|
|
|
|
D45_PRED, /* Directional 45 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
D135_PRED, /* Directional 135 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
D117_PRED, /* Directional 112 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
D153_PRED, /* Directional 157 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
D27_PRED, /* Directional 22 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
D63_PRED, /* Directional 67 deg prediction [anti-clockwise from 0 deg hor] */
|
|
|
|
TM_PRED, /* Truemotion prediction */
|
2013-04-11 23:16:35 +04:00
|
|
|
I8X8_PRED, /* 8x8 based prediction, each 8x8 has its own mode */
|
|
|
|
I4X4_PRED, /* 4x4 based prediction, each 4x4 has its own mode */
|
2012-07-14 02:21:29 +04:00
|
|
|
NEARESTMV,
|
|
|
|
NEARMV,
|
|
|
|
ZEROMV,
|
|
|
|
NEWMV,
|
|
|
|
SPLITMV,
|
|
|
|
MB_MODE_COUNT
|
2010-05-18 19:58:33 +04:00
|
|
|
} MB_PREDICTION_MODE;
|
|
|
|
|
2011-09-13 14:26:39 +04:00
|
|
|
// Segment level features.
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef enum {
|
|
|
|
SEG_LVL_ALT_Q = 0, // Use alternate Quantizer ....
|
|
|
|
SEG_LVL_ALT_LF = 1, // Use alternate loop filter value...
|
|
|
|
SEG_LVL_REF_FRAME = 2, // Optional Segment reference frame
|
2013-01-28 19:22:53 +04:00
|
|
|
SEG_LVL_SKIP = 3, // Optional Segment (0,0) + skip mode
|
2013-03-19 06:53:02 +04:00
|
|
|
SEG_LVL_MAX = 4 // Number of MB level features supported
|
2011-09-13 14:26:39 +04:00
|
|
|
} SEG_LVL_FEATURES;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2011-11-04 14:59:54 +04:00
|
|
|
// Segment level features.
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef enum {
|
2012-12-07 00:40:57 +04:00
|
|
|
TX_4X4 = 0, // 4x4 dct transform
|
|
|
|
TX_8X8 = 1, // 8x8 dct transform
|
|
|
|
TX_16X16 = 2, // 16x16 dct transform
|
|
|
|
TX_SIZE_MAX_MB = 3, // Number of different transforms available
|
|
|
|
TX_32X32 = TX_SIZE_MAX_MB, // 32x32 dct transform
|
|
|
|
TX_SIZE_MAX_SB, // Number of transforms available to SBs
|
2011-11-04 14:59:54 +04:00
|
|
|
} TX_SIZE;
|
|
|
|
|
2012-06-25 23:26:09 +04:00
|
|
|
typedef enum {
|
|
|
|
DCT_DCT = 0, // DCT in both horizontal and vertical
|
2012-10-22 23:19:00 +04:00
|
|
|
ADST_DCT = 1, // ADST in vertical, DCT in horizontal
|
|
|
|
DCT_ADST = 2, // DCT in vertical, ADST in horizontal
|
2012-06-25 23:26:09 +04:00
|
|
|
ADST_ADST = 3 // ADST in both directions
|
|
|
|
} TX_TYPE;
|
|
|
|
|
2013-04-11 23:16:35 +04:00
|
|
|
#define VP9_YMODES (I4X4_PRED + 1)
|
2012-11-01 01:40:53 +04:00
|
|
|
#define VP9_UV_MODES (TM_PRED + 1)
|
|
|
|
#define VP9_I8X8_MODES (TM_PRED + 1)
|
|
|
|
#define VP9_I32X32_MODES (TM_PRED + 1)
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-11-01 01:40:53 +04:00
|
|
|
#define VP9_MVREFS (1 + SPLITMV - NEARESTMV)
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2013-02-14 02:22:15 +04:00
|
|
|
#define WHT_UPSCALE_FACTOR 2
|
2012-11-25 07:33:58 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef enum {
|
|
|
|
B_DC_PRED, /* average of above and left pixels */
|
2013-04-12 02:52:28 +04:00
|
|
|
B_V_PRED, /* vertical prediction */
|
|
|
|
B_H_PRED, /* horizontal prediction */
|
|
|
|
B_D45_PRED,
|
|
|
|
B_D135_PRED,
|
|
|
|
B_D117_PRED,
|
|
|
|
B_D153_PRED,
|
|
|
|
B_D27_PRED,
|
|
|
|
B_D63_PRED,
|
2012-07-14 02:21:29 +04:00
|
|
|
B_TM_PRED,
|
2012-10-10 00:19:15 +04:00
|
|
|
#if CONFIG_NEWBINTRAMODES
|
|
|
|
B_CONTEXT_PRED,
|
|
|
|
#endif
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
LEFT4X4,
|
|
|
|
ABOVE4X4,
|
|
|
|
ZERO4X4,
|
|
|
|
NEW4X4,
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
B_MODE_COUNT
|
2010-05-18 19:58:33 +04:00
|
|
|
} B_PREDICTION_MODE;
|
|
|
|
|
2012-10-10 00:19:15 +04:00
|
|
|
#define VP9_BINTRAMODES (LEFT4X4)
|
2012-11-01 01:40:53 +04:00
|
|
|
#define VP9_SUBMVREFS (1 + NEW4X4 - LEFT4X4)
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-10-10 00:19:15 +04:00
|
|
|
#if CONFIG_NEWBINTRAMODES
|
2013-04-11 23:16:35 +04:00
|
|
|
/* The number of I4X4_PRED intra modes that are replaced by B_CONTEXT_PRED */
|
2012-10-10 00:19:15 +04:00
|
|
|
#define CONTEXT_PRED_REPLACEMENTS 0
|
|
|
|
#define VP9_KF_BINTRAMODES (VP9_BINTRAMODES - 1)
|
|
|
|
#define VP9_NKF_BINTRAMODES (VP9_BINTRAMODES - CONTEXT_PRED_REPLACEMENTS)
|
|
|
|
#else
|
|
|
|
#define VP9_KF_BINTRAMODES (VP9_BINTRAMODES) /* 10 */
|
|
|
|
#define VP9_NKF_BINTRAMODES (VP9_BINTRAMODES) /* 10 */
|
|
|
|
#endif
|
|
|
|
|
2012-10-22 22:25:48 +04:00
|
|
|
typedef enum {
|
|
|
|
PARTITIONING_16X8 = 0,
|
|
|
|
PARTITIONING_8X16,
|
|
|
|
PARTITIONING_8X8,
|
|
|
|
PARTITIONING_4X4,
|
|
|
|
NB_PARTITIONINGS,
|
|
|
|
} SPLITMV_PARTITIONING_TYPE;
|
|
|
|
|
2010-05-18 19:58:33 +04:00
|
|
|
/* For keyframes, intra block modes are predicted by the (already decoded)
|
|
|
|
modes for the Y blocks to the left and above us; for interframes, there
|
|
|
|
is a single probability table. */
|
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
union b_mode_info {
|
|
|
|
struct {
|
|
|
|
B_PREDICTION_MODE first;
|
2012-10-10 00:19:15 +04:00
|
|
|
#if CONFIG_NEWBINTRAMODES
|
|
|
|
B_PREDICTION_MODE context;
|
2012-02-29 05:12:08 +04:00
|
|
|
#endif
|
2012-07-14 02:21:29 +04:00
|
|
|
} as_mode;
|
2013-02-09 07:46:36 +04:00
|
|
|
int_mv as_mv[2]; // first, second inter predictor motion vectors
|
2011-05-26 23:13:00 +04:00
|
|
|
};
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef enum {
|
2012-11-07 18:50:25 +04:00
|
|
|
NONE = -1,
|
2012-07-14 02:21:29 +04:00
|
|
|
INTRA_FRAME = 0,
|
|
|
|
LAST_FRAME = 1,
|
|
|
|
GOLDEN_FRAME = 2,
|
|
|
|
ALTREF_FRAME = 3,
|
|
|
|
MAX_REF_FRAMES = 4
|
2010-05-18 19:58:33 +04:00
|
|
|
} MV_REFERENCE_FRAME;
|
|
|
|
|
2013-04-10 08:28:27 +04:00
|
|
|
static INLINE int mb_width_log2(BLOCK_SIZE_TYPE sb_type) {
|
|
|
|
switch (sb_type) {
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB16X32:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_MB16X16: return 0;
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB32X16:
|
|
|
|
case BLOCK_SIZE_SB32X64:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_SB32X32: return 1;
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB64X32:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_SB64X64: return 2;
|
|
|
|
default: assert(0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static INLINE int mb_height_log2(BLOCK_SIZE_TYPE sb_type) {
|
|
|
|
switch (sb_type) {
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB32X16:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_MB16X16: return 0;
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB16X32:
|
|
|
|
case BLOCK_SIZE_SB64X32:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_SB32X32: return 1;
|
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB32X64:
|
|
|
|
#endif
|
|
|
|
case BLOCK_SIZE_SB64X64: return 2;
|
|
|
|
default: assert(0);
|
|
|
|
}
|
|
|
|
}
|
2013-01-06 06:20:25 +04:00
|
|
|
|
2013-04-11 23:12:11 +04:00
|
|
|
// parse block dimension in the unit of 4x4 blocks
|
|
|
|
static INLINE int b_width_log2(BLOCK_SIZE_TYPE sb_type) {
|
|
|
|
return mb_width_log2(sb_type) + 2;
|
|
|
|
}
|
|
|
|
|
|
|
|
static INLINE int b_height_log2(BLOCK_SIZE_TYPE sb_type) {
|
|
|
|
return mb_height_log2(sb_type) + 2;
|
|
|
|
}
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
typedef enum {
|
|
|
|
BLOCK_4X4_LG2 = 0,
|
|
|
|
BLOCK_8X8_LG2 = 2,
|
|
|
|
BLOCK_16X16_LG2 = 4,
|
|
|
|
BLOCK_32X32_LG2 = 6,
|
|
|
|
BLOCK_64X64_LG2 = 8
|
|
|
|
} BLOCK_SIZE_LG2;
|
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef struct {
|
|
|
|
MB_PREDICTION_MODE mode, uv_mode;
|
2012-11-07 18:50:25 +04:00
|
|
|
#if CONFIG_COMP_INTERINTRA_PRED
|
|
|
|
MB_PREDICTION_MODE interintra_mode, interintra_uv_mode;
|
2012-02-29 05:12:08 +04:00
|
|
|
#endif
|
2012-07-14 02:21:29 +04:00
|
|
|
MV_REFERENCE_FRAME ref_frame, second_ref_frame;
|
|
|
|
TX_SIZE txfm_size;
|
2012-08-10 03:07:41 +04:00
|
|
|
int_mv mv[2]; // for each reference frame used
|
2012-12-04 21:21:05 +04:00
|
|
|
int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
|
2012-11-09 22:52:08 +04:00
|
|
|
int_mv best_mv, best_second_mv;
|
|
|
|
#if CONFIG_NEW_MVREF
|
|
|
|
int best_index, best_second_index;
|
|
|
|
#endif
|
2012-08-24 18:44:01 +04:00
|
|
|
|
2012-11-12 19:09:25 +04:00
|
|
|
int mb_mode_context[MAX_REF_FRAMES];
|
|
|
|
|
2012-10-22 22:25:48 +04:00
|
|
|
SPLITMV_PARTITIONING_TYPE partitioning;
|
2012-07-14 02:21:29 +04:00
|
|
|
unsigned char mb_skip_coeff; /* does this mb has coefficients at all, 1=no coefficients, 0=need decode tokens */
|
|
|
|
unsigned char need_to_clamp_mvs;
|
|
|
|
unsigned char need_to_clamp_secondmv;
|
|
|
|
unsigned char segment_id; /* Which set of segmentation parameters should be used for this MB */
|
|
|
|
|
|
|
|
// Flags used for prediction status of various bistream signals
|
|
|
|
unsigned char seg_id_predicted;
|
|
|
|
unsigned char ref_predicted;
|
|
|
|
|
|
|
|
// Indicates if the mb is part of the image (1) vs border (0)
|
|
|
|
// This can be useful in determining whether the MB provides
|
|
|
|
// a valid predictor
|
|
|
|
unsigned char mb_in_image;
|
2012-01-28 14:07:08 +04:00
|
|
|
|
2013-01-08 22:29:22 +04:00
|
|
|
INTERPOLATIONFILTERTYPE interp_filter;
|
2012-06-26 03:23:58 +04:00
|
|
|
|
2013-01-06 06:20:25 +04:00
|
|
|
BLOCK_SIZE_TYPE sb_type;
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
uint16_t nzcs[256+64*2];
|
|
|
|
#endif
|
2010-05-18 19:58:33 +04:00
|
|
|
} MB_MODE_INFO;
|
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
typedef struct {
|
|
|
|
MB_MODE_INFO mbmi;
|
|
|
|
union b_mode_info bmi[16];
|
2010-05-18 19:58:33 +04:00
|
|
|
} MODE_INFO;
|
|
|
|
|
2012-10-28 21:38:23 +04:00
|
|
|
typedef struct blockd {
|
2012-12-19 03:31:19 +04:00
|
|
|
uint8_t *predictor;
|
|
|
|
int16_t *diff;
|
|
|
|
int16_t *dequant;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries */
|
2012-12-19 03:31:19 +04:00
|
|
|
uint8_t **base_pre;
|
|
|
|
uint8_t **base_second_pre;
|
2012-07-14 02:21:29 +04:00
|
|
|
int pre;
|
|
|
|
int pre_stride;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-12-19 03:31:19 +04:00
|
|
|
uint8_t **base_dst;
|
2012-07-14 02:21:29 +04:00
|
|
|
int dst;
|
|
|
|
int dst_stride;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
union b_mode_info bmi;
|
2010-05-18 19:58:33 +04:00
|
|
|
} BLOCKD;
|
|
|
|
|
2013-02-09 05:49:44 +04:00
|
|
|
struct scale_factors {
|
|
|
|
int x_num;
|
|
|
|
int x_den;
|
|
|
|
int x_offset_q4;
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 08:55:14 +04:00
|
|
|
int x_step_q4;
|
2013-02-09 05:49:44 +04:00
|
|
|
int y_num;
|
|
|
|
int y_den;
|
|
|
|
int y_offset_q4;
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 08:55:14 +04:00
|
|
|
int y_step_q4;
|
Implicit weighted prediction experiment
Adds an experiment to use a weighted prediction of two INTER
predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8),
(1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly
based on consistency of the predictors to the already
reconstructed pixels to the top and left of the current macroblock
or superblock.
Currently the weighting is not applied to SPLITMV modes, which
default to the usual (1/2, 1/2) weighting. However the code is in
place controlled by a macro. The same weighting is used for Y and
UV components, where the weight is derived from analyzing the Y
component only.
Results (over compound inter-intra experiment)
derf: +0.18%
yt: +0.34%
hd: +0.49%
stdhd: +0.23%
The experiment suggests bigger benefit for explicitly signaled weights.
Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-13 01:21:08 +04:00
|
|
|
#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
|
|
|
|
convolve_fn_t predict[2][2][8]; // horiz, vert, weight (0 - 7)
|
|
|
|
#else
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 08:55:14 +04:00
|
|
|
convolve_fn_t predict[2][2][2]; // horiz, vert, avg
|
Implicit weighted prediction experiment
Adds an experiment to use a weighted prediction of two INTER
predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8),
(1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly
based on consistency of the predictors to the already
reconstructed pixels to the top and left of the current macroblock
or superblock.
Currently the weighting is not applied to SPLITMV modes, which
default to the usual (1/2, 1/2) weighting. However the code is in
place controlled by a macro. The same weighting is used for Y and
UV components, where the weight is derived from analyzing the Y
component only.
Results (over compound inter-intra experiment)
derf: +0.18%
yt: +0.34%
hd: +0.49%
stdhd: +0.23%
The experiment suggests bigger benefit for explicitly signaled weights.
Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-13 01:21:08 +04:00
|
|
|
#endif
|
2013-02-09 05:49:44 +04:00
|
|
|
};
|
|
|
|
|
2013-04-03 01:50:40 +04:00
|
|
|
enum { MAX_MB_PLANE = 3 };
|
|
|
|
|
|
|
|
struct mb_plane {
|
|
|
|
DECLARE_ALIGNED(16, int16_t, qcoeff[64 * 64]);
|
|
|
|
DECLARE_ALIGNED(16, int16_t, dqcoeff[64 * 64]);
|
2013-04-04 23:03:27 +04:00
|
|
|
DECLARE_ALIGNED(16, uint16_t, eobs[256]);
|
2013-04-06 02:54:59 +04:00
|
|
|
PLANE_TYPE plane_type;
|
|
|
|
int subsampling_x;
|
|
|
|
int subsampling_y;
|
2013-04-03 01:50:40 +04:00
|
|
|
};
|
|
|
|
|
|
|
|
#define BLOCK_OFFSET(x, i, n) ((x) + (i) * (n))
|
|
|
|
|
|
|
|
#define MB_SUBBLOCK_FIELD(x, field, i) (\
|
|
|
|
((i) < 16) ? BLOCK_OFFSET((x)->plane[0].field, (i), 16) : \
|
|
|
|
((i) < 20) ? BLOCK_OFFSET((x)->plane[1].field, ((i) - 16), 16) : \
|
|
|
|
BLOCK_OFFSET((x)->plane[2].field, ((i) - 20), 16))
|
|
|
|
|
2012-10-28 21:38:23 +04:00
|
|
|
typedef struct macroblockd {
|
2013-03-05 02:12:17 +04:00
|
|
|
DECLARE_ALIGNED(16, int16_t, diff[64*64+32*32*2]); /* from idct diff */
|
|
|
|
DECLARE_ALIGNED(16, uint8_t, predictor[384]); // unused for superblocks
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
DECLARE_ALIGNED(16, uint16_t, nzcs[256+64*2]);
|
|
|
|
#endif
|
2013-04-03 01:50:40 +04:00
|
|
|
struct mb_plane plane[MAX_MB_PLANE];
|
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
2012-12-08 02:45:05 +04:00
|
|
|
|
2013-02-15 22:15:42 +04:00
|
|
|
/* 16 Y blocks, 4 U, 4 V, each with 16 entries. */
|
|
|
|
BLOCKD block[24];
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
YV12_BUFFER_CONFIG pre; /* Filtered copy of previous frame reconstruction */
|
2013-02-07 22:09:05 +04:00
|
|
|
YV12_BUFFER_CONFIG second_pre;
|
2012-07-14 02:21:29 +04:00
|
|
|
YV12_BUFFER_CONFIG dst;
|
2013-02-09 05:49:44 +04:00
|
|
|
struct scale_factors scale_factor[2];
|
|
|
|
struct scale_factors scale_factor_uv[2];
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
MODE_INFO *prev_mode_info_context;
|
|
|
|
MODE_INFO *mode_info_context;
|
|
|
|
int mode_info_stride;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
FRAME_TYPE frame_type;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
int up_available;
|
|
|
|
int left_available;
|
[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 21:35:28 +04:00
|
|
|
int right_available;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2013-02-15 22:15:42 +04:00
|
|
|
/* Y,U,V */
|
2012-07-14 02:21:29 +04:00
|
|
|
ENTROPY_CONTEXT_PLANES *above_context;
|
|
|
|
ENTROPY_CONTEXT_PLANES *left_context;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* 0 indicates segmentation at MB level is not enabled. Otherwise the individual bits indicate which features are active. */
|
|
|
|
unsigned char segmentation_enabled;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* 0 (do not update) 1 (update) the macroblock segmentation map. */
|
|
|
|
unsigned char update_mb_segmentation_map;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* 0 (do not update) 1 (update) the macroblock segmentation feature data. */
|
|
|
|
unsigned char update_mb_segmentation_data;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* 0 (do not update) 1 (update) the macroblock segmentation feature data. */
|
|
|
|
unsigned char mb_segment_abs_delta;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* Per frame flags that define which MB level features (such as quantizer or loop filter level) */
|
|
|
|
/* are enabled and when enabled the proabilities used to decode the per MB flags in MB_MODE_INFO */
|
2011-11-15 15:13:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
// Probability Tree used to code Segment number
|
2012-11-01 01:40:53 +04:00
|
|
|
vp9_prob mb_segment_tree_probs[MB_FEATURE_TREE_PROBS];
|
2013-01-30 21:30:46 +04:00
|
|
|
vp9_prob mb_segment_mispred_tree_probs[MAX_MB_SEGMENTS];
|
2011-11-15 15:13:33 +04:00
|
|
|
|
2012-10-25 16:58:21 +04:00
|
|
|
#if CONFIG_NEW_MVREF
|
2012-12-10 16:38:48 +04:00
|
|
|
vp9_prob mb_mv_ref_probs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES-1];
|
2012-10-25 16:58:21 +04:00
|
|
|
#endif
|
2011-09-13 14:26:39 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
// Segment features
|
|
|
|
signed char segment_feature_data[MAX_MB_SEGMENTS][SEG_LVL_MAX];
|
|
|
|
unsigned int segment_feature_mask[MAX_MB_SEGMENTS];
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
/* mode_based Loop filter adjustment */
|
|
|
|
unsigned char mode_ref_lf_delta_enabled;
|
|
|
|
unsigned char mode_ref_lf_delta_update;
|
|
|
|
|
|
|
|
/* Delta values have the range +/- MAX_LOOP_FILTER */
|
2013-04-11 23:16:35 +04:00
|
|
|
/* 0 = Intra, Last, GF, ARF */
|
|
|
|
signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
|
|
|
|
/* 0 = Intra, Last, GF, ARF */
|
|
|
|
signed char ref_lf_deltas[MAX_REF_LF_DELTAS];
|
|
|
|
/* 0 = I4X4_PRED, ZERO_MV, MV, SPLIT */
|
|
|
|
signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
|
|
|
|
/* 0 = I4X4_PRED, ZERO_MV, MV, SPLIT */
|
|
|
|
signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];
|
2012-07-14 02:21:29 +04:00
|
|
|
|
|
|
|
/* Distance of MB away from frame edges */
|
|
|
|
int mb_to_left_edge;
|
|
|
|
int mb_to_right_edge;
|
|
|
|
int mb_to_top_edge;
|
|
|
|
int mb_to_bottom_edge;
|
|
|
|
|
|
|
|
unsigned int frames_since_golden;
|
|
|
|
unsigned int frames_till_alt_ref_frame;
|
2012-11-25 07:33:58 +04:00
|
|
|
|
2013-02-12 03:58:22 +04:00
|
|
|
int lossless;
|
2012-11-25 07:33:58 +04:00
|
|
|
/* Inverse transform function pointers. */
|
2013-02-12 09:14:46 +04:00
|
|
|
void (*inv_txm4x4_1)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*inv_txm4x4)(int16_t *input, int16_t *output, int pitch);
|
|
|
|
void (*itxm_add)(int16_t *input, const int16_t *dq,
|
2013-03-01 01:01:41 +04:00
|
|
|
uint8_t *pred, uint8_t *output, int pitch, int stride, int eob);
|
2013-02-12 09:14:46 +04:00
|
|
|
void (*itxm_add_y_block)(int16_t *q, const int16_t *dq,
|
2013-04-11 03:46:52 +04:00
|
|
|
uint8_t *pre, int pre_stride, uint8_t *dst, int stride,
|
|
|
|
struct macroblockd *xd);
|
2013-02-12 09:14:46 +04:00
|
|
|
void (*itxm_add_uv_block)(int16_t *q, const int16_t *dq,
|
2013-04-11 03:46:52 +04:00
|
|
|
uint8_t *pre, int pre_stride, uint8_t *dst, int stride,
|
2013-04-03 01:50:40 +04:00
|
|
|
uint16_t *eobs);
|
2012-11-25 07:33:58 +04:00
|
|
|
|
Convert subpixel filters to use convolve framework
Update the code to call the new convolution functions to do subpixel
prediction rather than the existing functions. Remove the old C and
assembly code, since it is unused. This causes a 50% performance
reduction on the decoder, but that will be resolved when the asm for
the new functions is available.
There is no consensus for whether 6-tap or 2-tap predictors will be
supported in the final codec, so these filters are implemented in
terms of the 8-tap code, so that quality testing of these modes
can continue. Implementing the lower complexity algorithms is a
simple exercise, should it be necessary.
This code produces slightly better results in the EIGHTTAP_SMOOTH
case, since the filter is now applied in only one direction when
the subpel motion is only in one direction. Like the previous code,
the filtering is skipped entirely on full-pel MVs. This combination
seems to give the best quality gains, but this may be indicative of a
bug in the encoder's filter selection, since the encoder could
achieve the result of skipping the filtering on full-pel by selecting
one of the other filters. This should be revisited.
Quality gains on derf positive on almost all clips. The only clip
that seemed to be hurt at all datarates was football
(-0.115% PSNR average, -0.587% min). Overall averages 0.375% PSNR,
0.347% SSIM.
Change-Id: I7d469716091b1d89b4b08adde5863999319d69ff
2013-01-29 04:59:03 +04:00
|
|
|
struct subpix_fn_table subpix;
|
2012-11-25 07:33:58 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
int allow_high_precision_mv;
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
int corrupted;
|
2010-12-16 18:46:31 +03:00
|
|
|
|
2013-01-06 06:20:25 +04:00
|
|
|
int sb_index;
|
2012-07-14 02:21:29 +04:00
|
|
|
int mb_index; // Index of the MB in the SB (0..3)
|
2012-06-25 23:26:09 +04:00
|
|
|
int q_index;
|
|
|
|
|
2010-05-18 19:58:33 +04:00
|
|
|
} MACROBLOCKD;
|
|
|
|
|
2013-02-06 00:37:13 +04:00
|
|
|
#define ACTIVE_HT 110 // quantization stepsize threshold
|
2012-10-16 03:41:41 +04:00
|
|
|
|
2013-02-06 00:37:13 +04:00
|
|
|
#define ACTIVE_HT8 300
|
2012-10-16 03:41:41 +04:00
|
|
|
|
2013-02-17 02:08:36 +04:00
|
|
|
#define ACTIVE_HT16 300
|
2012-10-16 03:41:41 +04:00
|
|
|
|
|
|
|
// convert MB_PREDICTION_MODE to B_PREDICTION_MODE
|
|
|
|
static B_PREDICTION_MODE pred_mode_conv(MB_PREDICTION_MODE mode) {
|
|
|
|
switch (mode) {
|
2013-02-27 23:17:38 +04:00
|
|
|
case DC_PRED: return B_DC_PRED;
|
2013-04-12 02:52:28 +04:00
|
|
|
case V_PRED: return B_V_PRED;
|
|
|
|
case H_PRED: return B_H_PRED;
|
2013-02-27 23:17:38 +04:00
|
|
|
case TM_PRED: return B_TM_PRED;
|
2013-04-12 02:52:28 +04:00
|
|
|
case D45_PRED: return B_D45_PRED;
|
|
|
|
case D135_PRED: return B_D135_PRED;
|
|
|
|
case D117_PRED: return B_D117_PRED;
|
|
|
|
case D153_PRED: return B_D153_PRED;
|
|
|
|
case D27_PRED: return B_D27_PRED;
|
|
|
|
case D63_PRED: return B_D63_PRED;
|
2013-02-27 23:17:38 +04:00
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
return B_MODE_COUNT; // Dummy value
|
2012-10-16 03:41:41 +04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-08-02 20:07:33 +04:00
|
|
|
// transform mapping
|
2012-10-16 03:41:41 +04:00
|
|
|
static TX_TYPE txfm_map(B_PREDICTION_MODE bmode) {
|
2012-08-02 20:07:33 +04:00
|
|
|
switch (bmode) {
|
|
|
|
case B_TM_PRED :
|
2013-04-12 02:52:28 +04:00
|
|
|
case B_D135_PRED :
|
2013-02-27 23:17:38 +04:00
|
|
|
return ADST_ADST;
|
2012-08-02 20:07:33 +04:00
|
|
|
|
2013-04-12 02:52:28 +04:00
|
|
|
case B_V_PRED :
|
|
|
|
case B_D117_PRED :
|
2013-02-27 23:17:38 +04:00
|
|
|
return ADST_DCT;
|
2012-08-02 20:07:33 +04:00
|
|
|
|
2013-04-12 02:52:28 +04:00
|
|
|
case B_H_PRED :
|
|
|
|
case B_D153_PRED :
|
|
|
|
case B_D27_PRED :
|
2013-02-27 23:17:38 +04:00
|
|
|
return DCT_ADST;
|
2012-08-02 20:07:33 +04:00
|
|
|
|
2012-10-10 00:19:15 +04:00
|
|
|
#if CONFIG_NEWBINTRAMODES
|
|
|
|
case B_CONTEXT_PRED:
|
|
|
|
assert(0);
|
|
|
|
break;
|
|
|
|
#endif
|
|
|
|
|
2013-02-27 23:17:38 +04:00
|
|
|
default:
|
|
|
|
return DCT_DCT;
|
2012-08-02 20:07:33 +04:00
|
|
|
}
|
2012-10-16 03:41:41 +04:00
|
|
|
}
|
|
|
|
|
2013-03-05 02:12:17 +04:00
|
|
|
extern const uint8_t vp9_block2left[TX_SIZE_MAX_MB][24];
|
|
|
|
extern const uint8_t vp9_block2above[TX_SIZE_MAX_MB][24];
|
|
|
|
extern const uint8_t vp9_block2left_sb[TX_SIZE_MAX_SB][96];
|
|
|
|
extern const uint8_t vp9_block2above_sb[TX_SIZE_MAX_SB][96];
|
|
|
|
extern const uint8_t vp9_block2left_sb64[TX_SIZE_MAX_SB][384];
|
|
|
|
extern const uint8_t vp9_block2above_sb64[TX_SIZE_MAX_SB][384];
|
2012-12-07 00:40:57 +04:00
|
|
|
|
2013-03-07 21:17:35 +04:00
|
|
|
#define USE_ADST_FOR_I16X16_8X8 1
|
|
|
|
#define USE_ADST_FOR_I16X16_4X4 1
|
2012-11-16 03:14:38 +04:00
|
|
|
#define USE_ADST_FOR_I8X8_4X4 1
|
|
|
|
#define USE_ADST_PERIPHERY_ONLY 1
|
2013-03-07 21:17:35 +04:00
|
|
|
#define USE_ADST_FOR_SB 1
|
|
|
|
#define USE_ADST_FOR_REMOTE_EDGE 0
|
2012-11-16 03:14:38 +04:00
|
|
|
|
2013-03-06 03:18:06 +04:00
|
|
|
static TX_TYPE get_tx_type_4x4(const MACROBLOCKD *xd, int ib) {
|
2012-11-16 03:14:38 +04:00
|
|
|
// TODO(debargha): explore different patterns for ADST usage when blocksize
|
|
|
|
// is smaller than the prediction size
|
2012-10-16 03:41:41 +04:00
|
|
|
TX_TYPE tx_type = DCT_DCT;
|
2013-03-07 21:17:35 +04:00
|
|
|
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
|
2013-04-10 08:28:27 +04:00
|
|
|
const int wb = mb_width_log2(sb_type), hb = mb_height_log2(sb_type);
|
2013-03-07 21:17:35 +04:00
|
|
|
#if !USE_ADST_FOR_SB
|
2013-04-10 08:28:27 +04:00
|
|
|
if (sb_type > BLOCK_SIZE_MB16X16)
|
2013-03-06 03:18:06 +04:00
|
|
|
return tx_type;
|
2013-03-07 21:17:35 +04:00
|
|
|
#endif
|
2013-04-10 08:28:27 +04:00
|
|
|
if (ib >= (16 << (wb + hb))) // no chroma adst
|
2012-11-16 03:14:38 +04:00
|
|
|
return tx_type;
|
2013-02-12 03:58:22 +04:00
|
|
|
if (xd->lossless)
|
2013-02-12 09:14:46 +04:00
|
|
|
return DCT_DCT;
|
2013-04-11 23:16:35 +04:00
|
|
|
if (xd->mode_info_context->mbmi.mode == I4X4_PRED &&
|
2012-10-16 03:41:41 +04:00
|
|
|
xd->q_index < ACTIVE_HT) {
|
2013-03-06 03:18:06 +04:00
|
|
|
const BLOCKD *b = &xd->block[ib];
|
2012-10-10 00:19:15 +04:00
|
|
|
tx_type = txfm_map(
|
|
|
|
#if CONFIG_NEWBINTRAMODES
|
|
|
|
b->bmi.as_mode.first == B_CONTEXT_PRED ? b->bmi.as_mode.context :
|
|
|
|
#endif
|
|
|
|
b->bmi.as_mode.first);
|
2012-11-16 03:14:38 +04:00
|
|
|
} else if (xd->mode_info_context->mbmi.mode == I8X8_PRED &&
|
|
|
|
xd->q_index < ACTIVE_HT) {
|
2013-03-06 03:18:06 +04:00
|
|
|
const BLOCKD *b = &xd->block[ib];
|
2013-03-07 21:17:35 +04:00
|
|
|
const int ic = (ib & 10);
|
2012-11-16 03:14:38 +04:00
|
|
|
#if USE_ADST_FOR_I8X8_4X4
|
|
|
|
#if USE_ADST_PERIPHERY_ONLY
|
|
|
|
// Use ADST for periphery blocks only
|
2013-03-07 21:17:35 +04:00
|
|
|
const int inner = ib & 5;
|
2012-11-16 03:14:38 +04:00
|
|
|
b += ic - ib;
|
2013-03-07 21:17:35 +04:00
|
|
|
tx_type = txfm_map(pred_mode_conv(
|
|
|
|
(MB_PREDICTION_MODE)b->bmi.as_mode.first));
|
|
|
|
#if USE_ADST_FOR_REMOTE_EDGE
|
|
|
|
if (inner == 5)
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#else
|
|
|
|
if (inner == 1) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = ADST_DCT;
|
|
|
|
else if (tx_type == DCT_ADST) tx_type = DCT_DCT;
|
|
|
|
} else if (inner == 4) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = DCT_ADST;
|
|
|
|
else if (tx_type == ADST_DCT) tx_type = DCT_DCT;
|
|
|
|
} else if (inner == 5) {
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
}
|
|
|
|
#endif
|
2012-11-16 03:14:38 +04:00
|
|
|
#else
|
|
|
|
// Use ADST
|
2013-03-07 21:17:35 +04:00
|
|
|
b += ic - ib;
|
2012-11-16 03:14:38 +04:00
|
|
|
tx_type = txfm_map(pred_mode_conv(
|
|
|
|
(MB_PREDICTION_MODE)b->bmi.as_mode.first));
|
|
|
|
#endif
|
|
|
|
#else
|
|
|
|
// Use 2D DCT
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#endif
|
|
|
|
} else if (xd->mode_info_context->mbmi.mode < I8X8_PRED &&
|
|
|
|
xd->q_index < ACTIVE_HT) {
|
|
|
|
#if USE_ADST_FOR_I16X16_4X4
|
|
|
|
#if USE_ADST_PERIPHERY_ONLY
|
2013-04-10 08:28:27 +04:00
|
|
|
const int hmax = 4 << wb;
|
2013-03-07 21:17:35 +04:00
|
|
|
tx_type = txfm_map(pred_mode_conv(xd->mode_info_context->mbmi.mode));
|
|
|
|
#if USE_ADST_FOR_REMOTE_EDGE
|
|
|
|
if ((ib & (hmax - 1)) != 0 && ib >= hmax)
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#else
|
|
|
|
if (ib >= 1 && ib < hmax) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = ADST_DCT;
|
|
|
|
else if (tx_type == DCT_ADST) tx_type = DCT_DCT;
|
|
|
|
} else if (ib >= 1 && (ib & (hmax - 1)) == 0) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = DCT_ADST;
|
|
|
|
else if (tx_type == ADST_DCT) tx_type = DCT_DCT;
|
|
|
|
} else if (ib != 0) {
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
}
|
|
|
|
#endif
|
2012-11-16 03:14:38 +04:00
|
|
|
#else
|
|
|
|
// Use ADST
|
|
|
|
tx_type = txfm_map(pred_mode_conv(xd->mode_info_context->mbmi.mode));
|
|
|
|
#endif
|
|
|
|
#else
|
|
|
|
// Use 2D DCT
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#endif
|
2012-10-16 03:41:41 +04:00
|
|
|
}
|
|
|
|
return tx_type;
|
|
|
|
}
|
|
|
|
|
2013-03-06 03:18:06 +04:00
|
|
|
static TX_TYPE get_tx_type_8x8(const MACROBLOCKD *xd, int ib) {
|
2012-11-16 03:14:38 +04:00
|
|
|
// TODO(debargha): explore different patterns for ADST usage when blocksize
|
|
|
|
// is smaller than the prediction size
|
2012-10-16 03:41:41 +04:00
|
|
|
TX_TYPE tx_type = DCT_DCT;
|
2013-03-07 21:17:35 +04:00
|
|
|
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
|
2013-04-10 08:28:27 +04:00
|
|
|
const int wb = mb_width_log2(sb_type), hb = mb_height_log2(sb_type);
|
2013-03-07 21:17:35 +04:00
|
|
|
#if !USE_ADST_FOR_SB
|
2013-04-10 08:28:27 +04:00
|
|
|
if (sb_type > BLOCK_SIZE_MB16X16)
|
2012-11-16 03:14:38 +04:00
|
|
|
return tx_type;
|
2013-03-07 21:17:35 +04:00
|
|
|
#endif
|
2013-04-10 08:28:27 +04:00
|
|
|
if (ib >= (16 << (wb + hb))) // no chroma adst
|
2013-03-06 03:18:06 +04:00
|
|
|
return tx_type;
|
2012-10-16 03:41:41 +04:00
|
|
|
if (xd->mode_info_context->mbmi.mode == I8X8_PRED &&
|
|
|
|
xd->q_index < ACTIVE_HT8) {
|
2013-03-06 03:18:06 +04:00
|
|
|
const BLOCKD *b = &xd->block[ib];
|
2012-11-06 06:13:04 +04:00
|
|
|
// TODO(rbultje): MB_PREDICTION_MODE / B_PREDICTION_MODE should be merged
|
|
|
|
// or the relationship otherwise modified to address this type conversion.
|
|
|
|
tx_type = txfm_map(pred_mode_conv(
|
2012-11-16 03:14:38 +04:00
|
|
|
(MB_PREDICTION_MODE)b->bmi.as_mode.first));
|
|
|
|
} else if (xd->mode_info_context->mbmi.mode < I8X8_PRED &&
|
|
|
|
xd->q_index < ACTIVE_HT8) {
|
2013-03-07 21:17:35 +04:00
|
|
|
#if USE_ADST_FOR_I16X16_8X8
|
2012-11-16 03:14:38 +04:00
|
|
|
#if USE_ADST_PERIPHERY_ONLY
|
2013-04-10 08:28:27 +04:00
|
|
|
const int hmax = 4 << wb;
|
2013-03-07 21:17:35 +04:00
|
|
|
tx_type = txfm_map(pred_mode_conv(xd->mode_info_context->mbmi.mode));
|
|
|
|
#if USE_ADST_FOR_REMOTE_EDGE
|
|
|
|
if ((ib & (hmax - 1)) != 0 && ib >= hmax)
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#else
|
|
|
|
if (ib >= 1 && ib < hmax) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = ADST_DCT;
|
|
|
|
else if (tx_type == DCT_ADST) tx_type = DCT_DCT;
|
|
|
|
} else if (ib >= 1 && (ib & (hmax - 1)) == 0) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = DCT_ADST;
|
|
|
|
else if (tx_type == ADST_DCT) tx_type = DCT_DCT;
|
|
|
|
} else if (ib != 0) {
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
}
|
|
|
|
#endif
|
2012-11-16 03:14:38 +04:00
|
|
|
#else
|
|
|
|
// Use ADST
|
|
|
|
tx_type = txfm_map(pred_mode_conv(xd->mode_info_context->mbmi.mode));
|
|
|
|
#endif
|
|
|
|
#else
|
|
|
|
// Use 2D DCT
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#endif
|
2012-10-16 03:41:41 +04:00
|
|
|
}
|
|
|
|
return tx_type;
|
2012-08-02 20:07:33 +04:00
|
|
|
}
|
2012-09-22 01:20:15 +04:00
|
|
|
|
2013-03-06 03:18:06 +04:00
|
|
|
static TX_TYPE get_tx_type_16x16(const MACROBLOCKD *xd, int ib) {
|
2012-10-16 03:41:41 +04:00
|
|
|
TX_TYPE tx_type = DCT_DCT;
|
2013-03-07 21:17:35 +04:00
|
|
|
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
|
2013-04-10 08:28:27 +04:00
|
|
|
const int wb = mb_width_log2(sb_type), hb = mb_height_log2(sb_type);
|
2013-03-07 21:17:35 +04:00
|
|
|
#if !USE_ADST_FOR_SB
|
2013-04-10 08:28:27 +04:00
|
|
|
if (sb_type > BLOCK_SIZE_MB16X16)
|
2013-03-07 21:17:35 +04:00
|
|
|
return tx_type;
|
|
|
|
#endif
|
2013-04-10 08:28:27 +04:00
|
|
|
if (ib >= (16 << (wb + hb)))
|
2012-11-16 03:14:38 +04:00
|
|
|
return tx_type;
|
|
|
|
if (xd->mode_info_context->mbmi.mode < I8X8_PRED &&
|
2012-10-16 03:41:41 +04:00
|
|
|
xd->q_index < ACTIVE_HT16) {
|
|
|
|
tx_type = txfm_map(pred_mode_conv(xd->mode_info_context->mbmi.mode));
|
2013-03-07 21:17:35 +04:00
|
|
|
#if USE_ADST_PERIPHERY_ONLY
|
2013-04-10 08:28:27 +04:00
|
|
|
if (sb_type > BLOCK_SIZE_MB16X16) {
|
|
|
|
const int hmax = 4 << wb;
|
2013-03-07 21:17:35 +04:00
|
|
|
#if USE_ADST_FOR_REMOTE_EDGE
|
|
|
|
if ((ib & (hmax - 1)) != 0 && ib >= hmax)
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
#else
|
|
|
|
if (ib >= 1 && ib < hmax) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = ADST_DCT;
|
|
|
|
else if (tx_type == DCT_ADST) tx_type = DCT_DCT;
|
|
|
|
} else if (ib >= 1 && (ib & (hmax - 1)) == 0) {
|
|
|
|
if (tx_type == ADST_ADST) tx_type = DCT_ADST;
|
|
|
|
else if (tx_type == ADST_DCT) tx_type = DCT_DCT;
|
|
|
|
} else if (ib != 0) {
|
|
|
|
tx_type = DCT_DCT;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
#endif
|
2012-10-16 03:41:41 +04:00
|
|
|
}
|
|
|
|
return tx_type;
|
|
|
|
}
|
|
|
|
|
2013-02-20 23:36:31 +04:00
|
|
|
void vp9_build_block_doffsets(MACROBLOCKD *xd);
|
|
|
|
void vp9_setup_block_dptrs(MACROBLOCKD *xd);
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
static void update_blockd_bmi(MACROBLOCKD *xd) {
|
2013-02-27 23:17:38 +04:00
|
|
|
const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
|
|
|
|
|
2013-04-11 23:16:35 +04:00
|
|
|
if (mode == SPLITMV || mode == I8X8_PRED || mode == I4X4_PRED) {
|
2013-02-27 23:17:38 +04:00
|
|
|
int i;
|
|
|
|
for (i = 0; i < 16; i++)
|
2012-07-14 02:21:29 +04:00
|
|
|
xd->block[i].bmi = xd->mode_info_context->bmi[i];
|
|
|
|
}
|
2011-08-05 03:30:27 +04:00
|
|
|
}
|
2013-02-27 23:17:38 +04:00
|
|
|
|
2013-02-20 22:16:24 +04:00
|
|
|
static TX_SIZE get_uv_tx_size(const MACROBLOCKD *xd) {
|
2013-04-02 05:23:04 +04:00
|
|
|
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
|
|
|
|
const TX_SIZE size = mbmi->txfm_size;
|
|
|
|
const MB_PREDICTION_MODE mode = mbmi->mode;
|
|
|
|
|
|
|
|
switch (mbmi->sb_type) {
|
|
|
|
case BLOCK_SIZE_SB64X64:
|
|
|
|
return size;
|
2013-04-10 08:28:27 +04:00
|
|
|
#if CONFIG_SBSEGMENT
|
|
|
|
case BLOCK_SIZE_SB64X32:
|
|
|
|
case BLOCK_SIZE_SB32X64:
|
|
|
|
#endif
|
2013-04-02 05:23:04 +04:00
|
|
|
case BLOCK_SIZE_SB32X32:
|
|
|
|
if (size == TX_32X32)
|
|
|
|
return TX_16X16;
|
|
|
|
else
|
|
|
|
return size;
|
|
|
|
default:
|
|
|
|
if (size == TX_16X16)
|
|
|
|
return TX_8X8;
|
|
|
|
else if (size == TX_8X8 && (mode == I8X8_PRED || mode == SPLITMV))
|
|
|
|
return TX_4X4;
|
|
|
|
else
|
|
|
|
return size;
|
2013-02-20 22:16:24 +04:00
|
|
|
}
|
2013-04-02 05:23:04 +04:00
|
|
|
|
|
|
|
return size;
|
2013-02-20 22:16:24 +04:00
|
|
|
}
|
2013-03-27 02:23:30 +04:00
|
|
|
|
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
static int get_nzc_used(TX_SIZE tx_size) {
|
|
|
|
return (tx_size >= TX_16X16);
|
|
|
|
}
|
|
|
|
#endif
|
2013-04-03 01:50:40 +04:00
|
|
|
|
|
|
|
struct plane_block_idx {
|
|
|
|
int plane;
|
|
|
|
int block;
|
|
|
|
};
|
|
|
|
|
|
|
|
// TODO(jkoleszar): returning a struct so it can be used in a const context,
|
|
|
|
// expect to refactor this further later.
|
2013-04-04 23:03:27 +04:00
|
|
|
static INLINE struct plane_block_idx plane_block_idx(int y_blocks,
|
|
|
|
int b_idx) {
|
|
|
|
const int v_offset = y_blocks * 5 / 4;
|
2013-04-03 01:50:40 +04:00
|
|
|
struct plane_block_idx res;
|
|
|
|
|
2013-04-04 23:03:27 +04:00
|
|
|
if (b_idx < y_blocks) {
|
2013-04-03 01:50:40 +04:00
|
|
|
res.plane = 0;
|
|
|
|
res.block = b_idx;
|
|
|
|
} else if (b_idx < v_offset) {
|
|
|
|
res.plane = 1;
|
2013-04-04 23:03:27 +04:00
|
|
|
res.block = b_idx - y_blocks;
|
2013-04-03 01:50:40 +04:00
|
|
|
} else {
|
2013-04-04 23:03:27 +04:00
|
|
|
assert(b_idx < y_blocks * 3 / 2);
|
2013-04-03 01:50:40 +04:00
|
|
|
res.plane = 2;
|
|
|
|
res.block = b_idx - v_offset;
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2013-04-09 21:15:10 +04:00
|
|
|
/* TODO(jkoleszar): Probably best to remove instances that require this,
|
|
|
|
* as the data likely becomes per-plane and stored in the per-plane structures.
|
|
|
|
* This is a stub to work with the existing code.
|
|
|
|
*/
|
|
|
|
static INLINE int old_block_idx_4x4(MACROBLOCKD* const xd, int block_size_b,
|
|
|
|
int plane, int i) {
|
|
|
|
const int luma_blocks = 1 << block_size_b;
|
|
|
|
assert(xd->plane[0].subsampling_x == 0);
|
|
|
|
assert(xd->plane[0].subsampling_y == 0);
|
|
|
|
assert(xd->plane[1].subsampling_x == 1);
|
|
|
|
assert(xd->plane[1].subsampling_y == 1);
|
|
|
|
assert(xd->plane[2].subsampling_x == 1);
|
|
|
|
assert(xd->plane[2].subsampling_y == 1);
|
|
|
|
return plane == 0 ? i :
|
|
|
|
plane == 1 ? luma_blocks + i :
|
|
|
|
luma_blocks * 5 / 4 + i;
|
|
|
|
}
|
|
|
|
|
|
|
|
typedef void (*foreach_transformed_block_visitor)(int plane, int block,
|
|
|
|
int block_size_b,
|
|
|
|
int ss_txfrm_size,
|
|
|
|
void *arg);
|
|
|
|
static INLINE void foreach_transformed_block_in_plane(
|
|
|
|
const MACROBLOCKD* const xd, int block_size, int plane,
|
|
|
|
int is_split, foreach_transformed_block_visitor visit, void *arg) {
|
|
|
|
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
|
|
|
|
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
|
|
|
|
const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
|
|
|
|
const int block_size_b = block_size;
|
|
|
|
const int txfrm_size_b = tx_size * 2;
|
|
|
|
|
|
|
|
// subsampled size of the block
|
|
|
|
const int ss_sum = xd->plane[plane].subsampling_x +
|
|
|
|
xd->plane[plane].subsampling_y;
|
|
|
|
const int ss_block_size = block_size_b - ss_sum;
|
|
|
|
|
|
|
|
// size of the transform to use. scale the transform down if it's larger
|
|
|
|
// than the size of the subsampled data, or forced externally by the mb mode.
|
|
|
|
const int ss_max = MAX(xd->plane[plane].subsampling_x,
|
|
|
|
xd->plane[plane].subsampling_y);
|
|
|
|
const int ss_txfrm_size = txfrm_size_b > ss_block_size || is_split
|
|
|
|
? txfrm_size_b - ss_max * 2
|
|
|
|
: txfrm_size_b;
|
|
|
|
|
|
|
|
// TODO(jkoleszar): 1 may not be correct here with larger chroma planes.
|
|
|
|
const int inc = is_split ? 1 : (1 << ss_txfrm_size);
|
|
|
|
int i;
|
|
|
|
|
|
|
|
assert(txfrm_size_b <= block_size_b);
|
|
|
|
assert(ss_txfrm_size <= ss_block_size);
|
|
|
|
for (i = 0; i < (1 << ss_block_size); i += inc) {
|
|
|
|
visit(plane, i, block_size_b, ss_txfrm_size, arg);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static INLINE void foreach_transformed_block(
|
|
|
|
const MACROBLOCKD* const xd, int block_size,
|
|
|
|
foreach_transformed_block_visitor visit, void *arg) {
|
|
|
|
const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
|
|
|
|
const int is_split =
|
|
|
|
xd->mode_info_context->mbmi.txfm_size == TX_8X8 &&
|
|
|
|
(mode == I8X8_PRED || mode == SPLITMV);
|
|
|
|
int plane;
|
|
|
|
|
|
|
|
for (plane = 0; plane < MAX_MB_PLANE; plane++) {
|
|
|
|
const int is_split_chroma = is_split &&
|
|
|
|
xd->plane[plane].plane_type == PLANE_TYPE_UV;
|
|
|
|
|
|
|
|
foreach_transformed_block_in_plane(xd, block_size, plane, is_split_chroma,
|
|
|
|
visit, arg);
|
|
|
|
}
|
|
|
|
}
|
2013-04-03 01:50:40 +04:00
|
|
|
|
2013-04-11 22:14:31 +04:00
|
|
|
static INLINE void foreach_transformed_block_uv(
|
|
|
|
const MACROBLOCKD* const xd, int block_size,
|
|
|
|
foreach_transformed_block_visitor visit, void *arg) {
|
|
|
|
const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
|
|
|
|
const int is_split =
|
|
|
|
xd->mode_info_context->mbmi.txfm_size == TX_8X8 &&
|
|
|
|
(mode == I8X8_PRED || mode == SPLITMV);
|
|
|
|
int plane;
|
|
|
|
|
|
|
|
for (plane = 1; plane < MAX_MB_PLANE; plane++) {
|
|
|
|
foreach_transformed_block_in_plane(xd, block_size, plane, is_split,
|
|
|
|
visit, arg);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-12-19 03:31:19 +04:00
|
|
|
#endif // VP9_COMMON_VP9_BLOCKD_H_
|