Lines Matching full:search
150 // Try the full image filter search with non-dual filter only.
165 CDEF_FULL_SEARCH, /**< Full search */
166 CDEF_FAST_SEARCH_LVL1, /**< Search among a subset of all possible filters. */
167 CDEF_FAST_SEARCH_LVL2, /**< Search reduced subset of filters than Level 1. */
168 CDEF_FAST_SEARCH_LVL3, /**< Search reduced subset of secondary filters than
170 CDEF_FAST_SEARCH_LVL4, /**< Search reduced subset of filters than Level 3. */
171 CDEF_FAST_SEARCH_LVL5, /**< Search reduced subset of filters than Level 4. */
178 // Terminate search early based on distortion so far compared to
220 // Turns off multi-winner mode. So we will do txfm search on either all modes
221 // if winner mode is off, or we will only on txfm search on a single winner
245 // Default transform search used in evaluation of best inter candidates
249 // Transform search in motion mode rd during MODE_EVAL stage.
251 // Transform search in compound type mode rd during MODE_EVAL stage.
253 // All transform search cases
261 // INT_MAX: Disable fast search.
263 // during mode search.
265 // mode search.
273 // and avoid doing full TX type search for such blocks.
279 // skip remaining transform type search when we found the rdcost of skip is
283 // Prune tx type search using previous frame stats.
285 // Prune tx type search using estimated RDcost
296 // Search partitions using RD criterion
355 * \brief This enumeration defines a variety of simple motion search based
378 * \brief This enumeration defines a variety of mesh search prune levels.
381 PRUNE_MESH_SEARCH_DISABLED = 0, /*!< Prune mesh search level 0. */
382 PRUNE_MESH_SEARCH_LVL_1 = 1, /*!< Prune mesh search level 1. */
383 PRUNE_MESH_SEARCH_LVL_2 = 2, /*!< Prune mesh search level 2. */
387 * \brief This enumeration defines inter search early termination index in
392 0, /*!< Early terminate inter mode search based on sse disabled. */
394 1, /*!< Early terminate inter mode search based on sse, index 1. */
396 2, /*!< Early terminate inter mode search based on sse, index 2. */
398 3, /*!< Early terminate inter mode search based on sse, index 3. */
400 4, /*!< Early terminate inter mode search based on sse, index 4. */
440 * Superres-auto mode search type:
492 * \brief Reduces the mv search window.
493 * By default, the initial search window is around
500 * \brief Skips the motion search when the zero mv has small sse.
510 * \brief Skips the motion search centered on 0,0 mv.
525 // Prune the intra modes search by tpl.
526 // If set to 0, we will search all intra modes from DC_PRED to PAETH_PRED.
527 // If set to 1, we only search DC_PRED, V_PRED, and H_PRED.
539 // When to stop subpel search.
542 // Which search method to use.
557 // Use SAD instead of SATD during intra/inter mode search.
559 // If set to 1, use SAD during intra/inter mode search for frames in the
561 // If set to 2, use SAD during intra/inter mode search for all frames.
563 // i.e., SATD is used during intra/inter mode search of the first GF group.
603 // Prune extended partition types search based on the current best partition
616 // Use a ML model to adaptively terminate partition search after trying
631 // simple motion search
645 // Partition search early breakout thresholds.
649 // Thresholds for ML based partition search breakout.
654 // simple motion search based pruning. QIDX_BASED_AGG_LVL1 corresponds to
655 // qindex based and simple motion search based pruning.
662 // Perform simple motion search before none_partition to decide if we
673 // Controls whether to reduce the number of motion search steps. If this is 0,
691 // Disable extended partition search if the current bsize is greater than the
701 // Prune extended partition search based on whether the split/rect partitions
702 // provided an improvement in the previous search.
704 // 1 : prune 1:4 partition search using winner info from split partitions
705 // 2 : prune 1:4 and AB partition search using split and HORZ/VERT info
744 // structure instead of recursive partition search. Therefore, this speed
748 // Terminate partition search for child partition,
764 // Prune rectangular split based on simple motion search split/no_split score.
768 // The current encoder adopts a DFS search for block partitions.
771 // AB partition could use previous rd information and skip mode search.
798 // the available information from previous rectangular search.
805 // The current partition search records the best rdcost so far and uses it
806 // in mode search and transform search to early skip when some criteria is
810 // search.
828 // Motion search method (Diamond, NSTEP, Hex, Big Diamond, Square, etc).
831 // Enable the use of faster, less accurate mv search method
837 // If this is set to 1, we limit the motion search range to 2 times the
842 // logarithmic search that keeps stepping at 1/2 pixel units until
847 // Maximum number of steps in logarithmic subpel search before giving up.
850 // When to stop subpel search.
853 // When to stop subpel search in simple motion search.
856 // If true, sub-pixel search uses the exact convolve function used for final
860 // Threshold for allowing exhaustive motion search.
869 // Reduce single motion search range based on MV result of prior ref_mv_idx.
872 // Prune mesh search.
875 // Use the rd cost around the best FULLPEL_MV to speed up subpel search
878 // Set the full pixel search level of obmc
883 // Accurate full pixel motion search based on TPL stats.
886 // Allow intrabc motion search
889 // Whether to downsample the rows in sad calculation during motion search.
892 // motion vector between skipping vs not skipping, motion search is redone
896 // odd rows for the starting MV is small. Redo motion search with sf off
902 // Enable/disable extensive joint motion search.
905 // Enable second best mv check in joint mv search.
911 // Skips full pixel search based on start mv of prior ref_mv_idx.
913 // 1: Skips the full pixel search upto 4 neighbor full-pel MV positions.
914 // 2: Skips the full pixel search upto 8 neighbor full-pel MV positions.
921 // Maximum number of iterations in WARPED_CAUSAL refinement search
927 // transform search and uses a model to estimate rd, while the final pass
928 // computes the full transform search. Two types of models are supported:
934 // Bypass transform search based on skip rd at following stages
935 // i. Compound type mode search
936 // ii. Motion mode search (mode evaluation and winner motion mode stage)
937 // iii. Transform search for best inter candidates
950 // Drop less likely to be picked reference frames in the RD search.
988 // 1. The other drl has the same mv during the SIMPLE_TRANSLATION search
1007 // Reuse the inter_intra_mode search result from NEARESTMV mode to other
1019 // Skip some ref frames in compound motion search by single motion search
1022 // Note: The search order might affect the result. It assumes that the single
1023 // reference modes are searched before compound modes. It is better to search
1027 // Instead of performing a full MV search, do a simple translation first
1028 // and only perform a full MV search on the motion vectors that performed
1032 // Only search compound modes with at least one "good" reference frame.
1062 // Based on previous ref_mv_idx search result, prune the following search.
1068 // Prune obmc search using previous frame stats.
1069 // INT_MAX : disable obmc search
1072 // Prune warped motion search using previous frame stats.
1075 // Variance threshold to enable/disable Interintra wedge search
1078 // Variance threshold to enable/disable Interinter wedge search
1081 // De-couple wedge and mode search during interintra RDO.
1087 // Enable/disable ME for interinter wedge search.
1111 // Model based breakout after interpolation filter search
1140 // Reuse masked compound type search results
1143 // Enable/disable fast search for wedge masks
1146 // Early breakout from transform search of inter modes
1149 // Limit number of inter modes for txfm search if a newmv mode gets
1159 // Prune warped motion search based on block size.
1162 // Do not search compound modes for ARF.
1175 // Do limited interpolation filter search for dual filters, since best choice
1183 // Check mv and ref_frames before search, if they are very close with previous
1184 // saved results, filter search can be skipped.
1191 // skip interpolation filter search for a block in chessboard pattern
1194 // adaptive interp_filter search to allow skip of certain filter types.
1198 // filter search.
1243 // prune palette search
1245 // 1: Perform coarse search to prune the palette colors. For winner colors,
1246 // neighbors are also evaluated using a finer search.
1247 // 2: Perform 2 way palette search from max colors to min colors (and min
1248 // colors to remaining colors) and terminate the search if current number of
1252 // Terminate early in luma palette_size search. Speed feature values indicate
1283 // best possible Chroma from Luma (CFL) parameter. Then we do a full RD search
1284 // near the best possible parameter. The search range is set here.
1290 // 33: Exhaustive rd search (33 == CFL_MAGS_SIZE). This mode should only
1313 // Terminate early in chroma palette_size search.
1331 // Init search depth for square and rectangular transform partitions.
1333 // 0 - search full tree, 1: search 1 level, 2: search the highest level only
1339 // If any dimension of a coding block size above 64, always search the
1349 // Shortcut the transform block partition and type search when the target
1355 // Prune level for tx_size_type search for inter based on rd model
1360 // Refine TX type after fast TX search.
1379 // Enable/disable early breakout during transform search of intra modes, by
1383 // 1) best_rd evolves during the search in choose_tx_size_type_from_rd()
1406 // tx_type search and compute distortion in pixel domain for final RD_STATS
1429 // Use hash table to store macroblock RD search results
1430 // to avoid repeated search on the same residue signal.
1443 // search method
1446 // Control transform size search level
1473 // If on, disables interpolation filter search in handle_inter_mode loop, and
1509 // prune sgr ep using binary search like mechanism
1518 // Range of loop restoration unit sizes to search
1531 // Prune self-guided loop restoration based on wiener search results
1547 // Disable the refinement search around the wiener filter coefficients.
1587 // For nonrd: Reduces ref frame search.
1588 // 0 - low level of search prune in non last frames
1589 // 1 - pruned search in non last frames
1590 // 2 - more pruned search in non last frames
1616 // Number of best inter modes to search transform. INT_MAX - search all.
1619 // Use interpolation filter search in non-RD mode decision.
1622 // Use simplified RD model for interpolation search and Intra
1625 // For nonrd mode: use hybrid intra mode search for intra only frames based on
1633 // fewer blocks will go through the palette search.
1664 // Parameter indicating initial search window to be used in full-pixel search
1732 // Prune inter mode search in rd path based on current block's temporal
1780 // Force to only use dct for palette search in nonrd pickmode.
1802 // Early terminate inter mode search based on sse in non-rd path.
1900 // Downgrades the block-level subpel motion search to
1902 // search performed well, zeromv has low sad or low source_var
1961 * Partition search speed features:
1966 * Motion search speed features:
1971 * Inter mode search speed features:
1976 * Interpolation filter search speed features:
1981 * Intra mode search speed features:
1986 * Transform size/type search speed features: