xref: /aosp_15_r20/external/libaom/av1/encoder/encodemb.h (revision 77c1e3ccc04c968bd2bc212e87364f250e820521)
1 /*
2  * Copyright (c) 2016, Alliance for Open Media. All rights reserved.
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #ifndef AOM_AV1_ENCODER_ENCODEMB_H_
13 #define AOM_AV1_ENCODER_ENCODEMB_H_
14 
15 #include "config/aom_config.h"
16 
17 #include "av1/common/av1_common_int.h"
18 #include "av1/common/txb_common.h"
19 #include "av1/encoder/av1_quantize.h"
20 #include "av1/encoder/block.h"
21 #include "av1/encoder/tokenize.h"
22 #ifdef __cplusplus
23 extern "C" {
24 #endif
25 
26 enum {
27   AV1_XFORM_QUANT_FP = 0,
28   AV1_XFORM_QUANT_B = 1,
29   AV1_XFORM_QUANT_DC = 2,
30   AV1_XFORM_QUANT_SKIP_QUANT,
31   AV1_XFORM_QUANT_TYPES,
32 } UENUM1BYTE(AV1_XFORM_QUANT);
33 
34 // TODO(any): Merge OPT_TYPe and TRELLLIS_OPT_TYPE
35 // Available optimization types to optimize the quantized coefficients.
36 enum {
37   NONE_OPT = 0,            // No optimization.
38   TRELLIS_OPT = 1,         // Trellis optimization. See `av1_optimize_b()`.
39   DROPOUT_OPT = 2,         // Dropout optimization. See `av1_dropout_qcoeff()`.
40   TRELLIS_DROPOUT_OPT = 3  // Perform dropout after trellis optimization.
41 } UENUM1BYTE(OPT_TYPE);
42 
43 enum {
44   NO_TRELLIS_OPT,          // No trellis optimization
45   FULL_TRELLIS_OPT,        // Trellis optimization in all stages
46   FINAL_PASS_TRELLIS_OPT,  // Trellis optimization in only the final encode pass
47   NO_ESTIMATE_YRD_TRELLIS_OPT  // Disable trellis in estimate_yrd_for_sb
48 } UENUM1BYTE(TRELLIS_OPT_TYPE);
49 
50 struct optimize_ctx {
51   ENTROPY_CONTEXT ta[MAX_MB_PLANE][MAX_MIB_SIZE];
52   ENTROPY_CONTEXT tl[MAX_MB_PLANE][MAX_MIB_SIZE];
53 };
54 
55 struct encode_b_args {
56   const struct AV1_COMP *cpi;
57   MACROBLOCK *x;
58   struct optimize_ctx *ctx;
59   ENTROPY_CONTEXT *ta;
60   ENTROPY_CONTEXT *tl;
61   RUN_TYPE dry_run;
62   TRELLIS_OPT_TYPE enable_optimize_b;
63 };
64 
65 void av1_encode_sb(const struct AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
66                    RUN_TYPE dry_run);
67 
68 void av1_foreach_transformed_block_in_plane(
69     const MACROBLOCKD *const xd, BLOCK_SIZE plane_bsize, int plane,
70     foreach_transformed_block_visitor visit, void *arg);
71 
72 void av1_encode_sby_pass1(struct AV1_COMP *cpi, MACROBLOCK *x,
73                           BLOCK_SIZE bsize);
74 
75 void av1_setup_xform(const AV1_COMMON *cm, MACROBLOCK *x, TX_SIZE tx_size,
76                      TX_TYPE tx_type, TxfmParam *txfm_param);
77 void av1_setup_quant(TX_SIZE tx_size, int use_optimize_b, int xform_quant_idx,
78                      int use_quant_b_adapt, QUANT_PARAM *qparam);
79 void av1_setup_qmatrix(const CommonQuantParams *quant_params,
80                        const MACROBLOCKD *xd, int plane, TX_SIZE tx_size,
81                        TX_TYPE tx_type, QUANT_PARAM *qparam);
82 
83 void av1_xform_dc_only(MACROBLOCK *x, int plane, int block,
84                        TxfmParam *txfm_param, int64_t per_px_mean);
85 
86 void av1_xform_quant(MACROBLOCK *x, int plane, int block, int blk_row,
87                      int blk_col, BLOCK_SIZE plane_bsize, TxfmParam *txfm_param,
88                      const QUANT_PARAM *qparam);
89 
90 void av1_xform(MACROBLOCK *x, int plane, int block, int blk_row, int blk_col,
91                BLOCK_SIZE plane_bsize, TxfmParam *txfm_param);
92 
93 void av1_quant(MACROBLOCK *x, int plane, int block, TxfmParam *txfm_param,
94                const QUANT_PARAM *qparam);
95 
96 int av1_optimize_b(const struct AV1_COMP *cpi, MACROBLOCK *mb, int plane,
97                    int block, TX_SIZE tx_size, TX_TYPE tx_type,
98                    const TXB_CTX *const txb_ctx, int *rate_cost);
99 
100 // This function can be used as (i) a further optimization to reduce the
101 // redundancy of quantized coefficients (a.k.a., `qcoeff`) after trellis
102 // optimization, or (ii) an alternative to trellis optimization in high-speed
103 // compression mode (e.g., real-time mode under speed-6) due to its LOW time
104 // complexity. The rational behind is to drop out the may-be redundant quantized
105 // coefficient which is among a bunch of zeros. NOTE: This algorithm is not as
106 // accurate as trellis optimization since the hyper-parameters are hard-coded
107 // instead of dynamic search. More adaptive logic may improve the performance.
108 // This function should be applied to all or partical block cells.
109 // Inputs:
110 //   mb: Pointer to the MACROBLOCK to perform dropout on.
111 //   plane: Index of the plane to which the target block belongs.
112 //   block: Index of the target block.
113 //   tx_size: Transform size of the target block.
114 //   tx_type: Transform type of the target block. This field is particularly
115 //            used to find out the scan order of the block.
116 //   qindex: Quantization index used for target block. In general, all blocks
117 //           in a same plane share the same quantization index. This field is
118 //           particularly used to determine how many zeros should be used to
119 //           drop out a coefficient.
120 // Returns:
121 //   Nothing will be returned, but `qcoeff`, `dqcoeff`, `eob`, as well as
122 //   `txb_entropy_ctx`, which `mb` points to, may be modified by this function.
123 void av1_dropout_qcoeff(MACROBLOCK *mb, int plane, int block, TX_SIZE tx_size,
124                         TX_TYPE tx_type, int qindex);
125 // Same as above, with the number of zeroes needed before/after a coeff to drop
126 // it explicitly passed in, instead of being derived from qindex.
127 void av1_dropout_qcoeff_num(MACROBLOCK *mb, int plane, int block,
128                             TX_SIZE tx_size, TX_TYPE tx_type,
129                             int dropout_num_before, int dropout_num_after);
130 
131 void av1_subtract_block(BitDepthInfo bd_info, int rows, int cols, int16_t *diff,
132                         ptrdiff_t diff_stride, const uint8_t *src8,
133                         ptrdiff_t src_stride, const uint8_t *pred8,
134                         ptrdiff_t pred_stride);
135 
136 void av1_subtract_txb(MACROBLOCK *x, int plane, BLOCK_SIZE plane_bsize,
137                       int blk_col, int blk_row, TX_SIZE tx_size);
138 
139 void av1_subtract_plane(MACROBLOCK *x, BLOCK_SIZE plane_bsize, int plane);
140 
av1_set_txb_context(MACROBLOCK * x,int plane,int block,TX_SIZE tx_size,ENTROPY_CONTEXT * a,ENTROPY_CONTEXT * l)141 static inline void av1_set_txb_context(MACROBLOCK *x, int plane, int block,
142                                        TX_SIZE tx_size, ENTROPY_CONTEXT *a,
143                                        ENTROPY_CONTEXT *l) {
144   const uint8_t ctx = x->plane[plane].txb_entropy_ctx[block];
145   memset(a, ctx, tx_size_wide_unit[tx_size] * sizeof(*a));
146   memset(l, ctx, tx_size_high_unit[tx_size] * sizeof(*l));
147 }
148 
149 void av1_encode_intra_block_plane(const struct AV1_COMP *cpi, MACROBLOCK *x,
150                                   BLOCK_SIZE bsize, int plane, RUN_TYPE dry_run,
151                                   TRELLIS_OPT_TYPE enable_optimize_b);
152 
is_trellis_used(TRELLIS_OPT_TYPE optimize_b,RUN_TYPE dry_run)153 static inline int is_trellis_used(TRELLIS_OPT_TYPE optimize_b,
154                                   RUN_TYPE dry_run) {
155   if (optimize_b == NO_TRELLIS_OPT) return false;
156   if (optimize_b == FINAL_PASS_TRELLIS_OPT && dry_run != OUTPUT_ENABLED)
157     return false;
158   return true;
159 }
160 
161 // Scaling terms (precision of 12 bits) to perform tx-size specific
162 // normalization that is used in DCT_DCT forward transform.
163 // For transform blocks of 1:2 and 2:1       - sqrt(2) normalization is used
164 // For transform blocks of 1:4 and 4:1       - factor of 2 is used
165 // For transform blocks TX_8x8 and below     - an additional factor of 2 is used
166 // For transform blocks max(width,height)=64 - currently not supported
167 
168 static const uint16_t dc_coeff_scale[TX_SIZES_ALL] = {
169   1024, 2048, 4096, 4096, 0,    1448, 1448, 2896, 2896, 2896,
170   2896, 0,    0,    2048, 2048, 4096, 4096, 0,    0
171 };
172 
173 #ifdef __cplusplus
174 }  // extern "C"
175 #endif
176 
177 #endif  // AOM_AV1_ENCODER_ENCODEMB_H_
178