Commit graph

3936 commits

Author SHA1 Message Date
Joose Sainio 0b6f666a1b [mtt] remove lfnst dependency to depth 2023-08-15 13:00:08 +03:00
Joose Sainio 790b1fad48 wip 2023-08-15 13:00:07 +03:00
Joose Sainio 6a0864839c [mtt] Actually remove the last width dependency to depth 2023-08-15 13:00:06 +03:00
Joose Sainio dcf879e5ed [mtt] remove all rest usages of deriving width and height from depth 2023-08-15 12:59:39 +03:00
Joose Sainio 26dcadc149 [mtt] change most if not all of search hierarchy to use cu_loc_t 2023-08-15 12:47:11 +03:00
siivonek 0ec16967a1 [isp] Fix reference building. When ISP was in use, not enough samples were generated. Uninitialized memory was referenced. Fix some typos. 2023-08-14 12:21:30 +03:00
siivonek b16c404362 [isp] Remove some obsolete TODOs and old commented out code. 2023-08-14 12:21:30 +03:00
siivonek 95d73116f9 [isp] Fix some CI errors. Some const modifiers were discarded. 2023-08-14 12:21:29 +03:00
siivonek 90e2a17759 [lfnst] Fix LFNST error when MIP enabled. 2023-08-14 12:21:28 +03:00
siivonek 7005d222d5 [isp] Fix lfnst constraint check when ISP is used. Remove some obsolete comments. 2023-08-14 12:21:28 +03:00
siivonek 3c861e4c02 [isp] Fix search. Best LFNST and MTS modes were not selected correctly for ISP modes. 2023-08-14 12:21:27 +03:00
siivonek b4cc321349 [isp] Fix transform selection when MTS & ISP is used. Wrong transform was selected. Change mts parameter name to better reflect its purpose. 2023-08-14 12:21:26 +03:00
siivonek 85f6b00394 [isp] Add lfnst asserts. Fix error in MTS search. Fix chroma lfnst index when no coefficients present. 2023-08-14 12:21:26 +03:00
siivonek b9822398a0 [isp] Fix lfnst constraint checks when ISP is in use. Add some asserts. 2023-08-14 12:21:25 +03:00
siivonek 701257cdd2 [isp] Remove unnecessary code from forward dct 32. 2023-08-14 12:21:25 +03:00
siivonek 89db34d4e0 [isp] Use TR_MAX_WIDTH in ISP checks instead of parameter. 2023-08-14 12:21:24 +03:00
siivonek c4bc2d6b10 [isp] Limit ISP search to block size 32. Size 64 is not allowed. 2023-08-14 12:21:23 +03:00
siivonek 5713fbff1a [isp] Add ISP checks to search. LFNST can be used with ISP for larger blocks. Transform skip cannot be used with ISP. 2023-08-14 12:21:22 +03:00
siivonek 7282534879 [isp] Fix CI errors. 2023-08-14 12:21:22 +03:00
siivonek 01c4d1ddb0 [isp] Fix cabac issues. There are always four transform blocks even if there are only two ISP splits. Fix prediction issues. PDPC filter was applied when it should be disabled. Fix reference building issues. Left reference was built incorrectly for blocks with height 2. 2023-08-14 12:21:21 +03:00
siivonek b8e36bbc4a [isp] Fix storing cbfs for small ISP splits. Fix pdpc filtering. Cannot be used if width or height is less than 4. Fix dct related CI errors. 2023-08-14 12:21:20 +03:00
siivonek 99495c331b [isp] Fix some asserts to allow log2_dim 1 block sizes. Fix coefficient group scan order for small dimensions. 2023-08-14 12:21:19 +03:00
siivonek d39fddf0d8 [isp] Implement DCT for small blocks. 2023-08-14 12:21:19 +03:00
siivonek 910501012f [isp] Fix referene building for depth 2 blocks. Flip horizontal mode dimensions during prediction. Fix reference length during prediction when ISP enabled. 2023-08-14 12:21:18 +03:00
siivonek 7ba557af6b [isp] Fix luma cbf writing for ISP splits. Do not write luma cbf if first three splits had luma cbf 0. 2023-08-14 12:21:17 +03:00
siivonek a28e61eff7 [isp] Fix CI errors. 2023-08-14 12:21:17 +03:00
siivonek 4794104ecc [isp] Fix errors in reference building. Use cubic filter during prediction if ISP enabled. 2023-08-14 12:21:16 +03:00
Joose Sainio 662f31d61d [isp] Use correct coordinates for depth 4 chroma tu cost calculation 2023-08-14 12:17:36 +03:00
Joose Sainio 08942a5394 [tr-skip] fix transform skip flag writing 2023-08-14 12:17:35 +03:00
siivonek a261d4c5b3 [isp] WIP 2023-08-14 12:17:34 +03:00
siivonek 6340dfe4ce [isp] Fix mistake in pu_loc argument passing, was not used after passing. 2023-08-14 12:17:33 +03:00
Joose Sainio 88c33c0489 [lfnst] Fix lfnst constraint checking for the new coeff order 2023-08-14 12:17:33 +03:00
Joose Sainio e0e96068cc [lfnst] lfnst is not allowed for transform split 2023-08-14 12:17:32 +03:00
Joose Sainio cb7f9919e3 [jccr] Fix jccr coefficient copying 2023-08-14 12:17:32 +03:00
Joose Sainio 3e23fd0601 [cabac] fix cbf_y context for tr splits 2023-08-14 12:17:31 +03:00
siivonek 59292d8808 [isp] Add extra logic to reference building to accommodate ISP. Remove some asserts which were invalidated by ISP. 2023-08-14 12:17:31 +03:00
siivonek 33cd44f11b [isp] Fix chroma coeff writing for ISP. 2023-08-14 12:17:30 +03:00
siivonek d8d206365c [isp] Fix jccr coeffs. 2023-08-14 12:17:29 +03:00
siivonek 7398e58431 [isp] Fix coeff cost calculation. Coeff arrays were indexed wrongly. 2023-08-14 12:17:29 +03:00
siivonek d050efcb87 [isp] Fix error in last sig coeff function call. Height was not used. Fix cbf writing. Fix transform skip flag writing. 2023-08-14 12:17:28 +03:00
siivonek 33ae02aae0 [isp] Fix mistake in isp cbf writing. Loop index was increased twice. 2023-08-14 12:17:28 +03:00
siivonek 4a21039e23 [isp] Fix mistake in function declaration. 2023-08-14 12:17:27 +03:00
siivonek b8506c757c [isp] Convert functions to handle new coeff array order. Add function for getting coeff array subset. Fix assert. 2023-08-14 12:17:26 +03:00
siivonek 69dcb04c99 [isp] Use temporary coeff array when quantizing coeffs. After deriving coeffs, copy temp coeffs from linear order to correct arrays with cu order. 2023-08-14 12:17:26 +03:00
siivonek 0ae71feae4 [isp] Fix assert. 2023-08-14 12:17:25 +03:00
siivonek 2e8f008de4 [isp] Redo call hierarchy to include x, y coordinates. 2023-08-14 12:17:24 +03:00
siivonek 10f9b2be26 [isp] Keep lfnst constraint up to date during search. 2023-08-14 12:16:42 +03:00
siivonek 39f30563c5 [isp] Fix chroma width error when writing cu loc. Remove redundant IPS mode checks. 2023-08-14 12:16:42 +03:00
siivonek b53308f258 [isp] Fix mistake in setting cbfs. Skip stting if ISP is not used. 2023-08-14 12:16:41 +03:00
siivonek 56ebea7358 [isp] Set cbfs for isp splits after search. Add helper function for isp split number. 2023-08-14 12:16:40 +03:00
siivonek 510798cb3d [isp] Fix mistake in isp cabac write. Intra luma mpm flag bit was checking isp when it did not need to. 2023-08-14 12:16:40 +03:00
siivonek f86dc29ce7 [isp] Fix mistake in cost calculation. Remove some commented out code blocks. 2023-08-14 12:16:39 +03:00
siivonek bbb8faea98 [isp] Modify encode transform coeff func to handle non-square blocks, use cu_loc_t where possible. Fix mistake in mts idct generic. 2023-08-14 12:16:38 +03:00
siivonek 7062697beb [isp] Resolve TODOs. Make scan order tables const. 2023-08-14 12:16:37 +03:00
siivonek 93317cafa4 [isp] Write isp config bit to sps. 2023-08-14 12:16:37 +03:00
siivonek 182d0f4e66 [isp] Remove old_scan tables and related asserts. Fix coefficient group indexing. 2023-08-14 12:16:36 +03:00
siivonek f8641f7436 [isp] Fix assert. Implement coef cost calculation for isp splits. 2023-08-14 12:16:35 +03:00
siivonek ae0336fdfc [isp] Add non-square block handling to functions. 2023-08-14 12:16:34 +03:00
siivonek 031a758d6c [isp] Count isp cbfs. 2023-08-14 12:16:01 +03:00
siivonek 75175ee2e2 [isp] Fix isp search. 2023-08-14 12:16:00 +03:00
siivonek 8d914ce849 [isp] Implement coefficient encoding for isp splits. Make get_split_dim non static, it was needed elsewhere after all. 2023-08-14 12:15:59 +03:00
siivonek 573ecf80e3 [isp] Move can_use_lfnst_with_isp to intra.c. Remove duplicate functions. Move isp related functions from search to intra. Make isp_split_dim static. Move isp related defines from search to intra. 2023-08-14 12:15:58 +03:00
siivonek bcbd952dfd [isp] Add height handling to avx2 reconstruction. 2023-08-14 12:15:57 +03:00
siivonek 7c340fd92b [isp] Add height to inverse transform skip. 2023-08-14 12:15:56 +03:00
siivonek 318d925028 [isp] Add new convert_to_log2 table. Change all instances which used old convert_to_bit table to change dimensions into log2. 2023-08-14 12:15:55 +03:00
siivonek 6922157ed3 [isp] Fix quantization function calls. Some were not getting height as input. 2023-08-14 12:15:54 +03:00
siivonek 50ad91a94e [isp] Modify quantization functions to work with non-square blocks. 2023-08-14 12:15:53 +03:00
siivonek 31c8f1356f [isp] Add height to sig coeff group context calculation function. 2023-08-14 12:15:53 +03:00
siivonek 936256e750 [isp] Fix sig coeff flag context calculation function call. Width & height was swapped. 2023-08-14 12:15:52 +03:00
siivonek 9e7f4eac99 [isp] Change variable name 'type' to 'color'. 2023-08-14 12:15:51 +03:00
siivonek 09b905c6c4 [isp] Add height to get_tr_type function. 2023-08-14 12:15:51 +03:00
siivonek 8b7d573ae7 [isp] Add height to idct getter function. Check block dimensions in transform 2d functions. 2023-08-14 12:15:50 +03:00
siivonek 370bd07c55 [isp] Fix error in mts dct and idct. 2023-08-14 12:15:49 +03:00
siivonek 3a874ab5dd [isp] Comment out dct non square function. It is not needed since mts dct function will handle transform for non square blocks. 2023-08-14 12:15:49 +03:00
siivonek f9116441da [isp] Fix avx2 function call. Missing height parameter. 2023-08-14 12:15:48 +03:00
siivonek 6f756e831d [isp] Uncomment old scan order code to test against new one. Add assert to ensure old and new tables are the same. 2023-08-14 12:15:47 +03:00
siivonek c4d1f80f8f [isp] Fix error in scan order getter. Change define names to better reflect what they do. Add more accurate bookmark comments to scan order buffer table. 2023-08-14 12:15:46 +03:00
siivonek 8131e970e5 [isp] Modify existing scan table calls to use new getter. Add safety assert to getter. 2023-08-14 12:15:46 +03:00
siivonek 6ff9ae074e [isp] Add scan order getter. Add bookmark comments to scan order buffer. 2023-08-14 12:15:45 +03:00
siivonek 55d77c6b50 [isp] Add scan order tables for all possible block sizes. 2023-08-14 12:15:45 +03:00
siivonek 35271648db [isp] Fix some errors. Pass height to functions. Some WIP comments. 2023-08-14 12:15:44 +03:00
siivonek a9090c99b5 [isp] Fix error in inverse transform shifting. 2023-08-14 12:15:43 +03:00
siivonek cd7e091992 [isp] Fix mistake in transform if clause. 2023-08-14 12:15:43 +03:00
siivonek 6a3ddfd0bc [isp] Modify inverse transform to handle non-square blocks. 2023-08-14 12:15:42 +03:00
siivonek 626c9b02ea [isp] Modify transform and quantization functions to handle non-square blocks. Add strategy headers to CMakelist. 2023-08-14 12:15:41 +03:00
siivonek 06532dce02 [isp] Implement ISP search and partitioning. Add helper function for constructing cu_loc types. WIP stuff for transform. 2023-08-14 12:15:40 +03:00
siivonek 6236cc29be [isp] Fix avx2 function call. 2023-08-14 12:15:39 +03:00
siivonek ec4909095c [isp] Do not filter references if ISP is used. 2023-08-14 12:15:38 +03:00
siivonek 96df3ffd64 [isp] Change function calls to cu_loc_t. 2023-08-14 12:15:38 +03:00
siivonek 9406c5c31d [isp] Modify generic intra pred functions to handle non-square blocks. 2023-08-14 12:15:37 +03:00
siivonek 03b22e561e [isp] Add ISP command line option. 2023-08-14 12:15:36 +03:00
Marko Viitanen 1a1fea1a19 Merge branch 'implement_ibc' 2023-08-09 09:34:29 +03:00
Marko Viitanen 18b4a8be79 [ibc] Include the chroma in crc 2023-07-27 10:58:20 +03:00
Marko Viitanen 20875a9819 [ibc] Calculate hashes every 4 pixels and change the IBC costs a bit 2023-07-27 10:29:55 +03:00
Marko Viitanen 0fefd3f621 [ibc] Add hash based starting point finder for reqular inter search *experimental* 2023-07-24 22:55:52 +03:00
Marko Viitanen 6f4d538f4f [ibc] Clean up the ibc search, utilize hash based starting points if ibc=2 2023-07-24 22:07:22 +03:00
Marko Viitanen 6fe629e666 [ibc] A bit of cleanup and skip IBC search if cost is already less than 500 2023-07-21 20:40:00 +03:00
Marko Viitanen 3cef3c0119 Change the hardcoded general_level_idc from 5.2 to 6.3 2023-07-21 20:19:43 +03:00
Marko Viitanen 95dc4aa0cb [ibc] Fix the IBC buffer limitation, 256x64 pixels allowed 2023-07-21 20:15:24 +03:00
Marko Viitanen 8ff184a6b3 [ibc] Fill the IBC hashmap at the start of LCU search and use reverse map for "pos to hash" 2023-07-21 20:14:23 +03:00
Marko Viitanen 457d650f49 [ibc] Fix for CRC calculations
- Input for the 64bit crc intrinsic was 32bit
2023-07-19 09:57:24 +03:00
Marko Viitanen 15fb6f8183 [ibc] Add first version of the IBC hash search 2023-06-29 21:57:06 +03:00
Marko Viitanen 8cec02280f [ibc] Use IBC hashmap in LCU row basis 2023-06-28 23:06:04 +03:00
Marko Viitanen 76d66591c5 [ibc] Implement CRC for 8x8 block and generate a full hashmap at the frame load 2023-06-26 21:24:10 +03:00
Marko Viitanen 4b1f5ca7e2 [ibc] Add the hashmap to the frame and fix some small issues with hashmap and crc32c
- crc32c_4x4 strategy was not working, made some changes to the initialization
2023-06-22 21:44:49 +03:00
Marko Viitanen 30321e6dd4 [ibc] Fix uvg_hashmap_hash definition 2023-06-22 14:45:05 +03:00
Marko Viitanen a32a318d18 [ibc] Add CRC32C functions, with SSE 4.2 optimized CRC calculations 2023-06-22 14:39:56 +03:00
Marko Viitanen 7252befc17 [ibc] Add a hashmap implementation for IBC hash search 2023-06-22 14:39:56 +03:00
Marko Viitanen 68382f9e25 [ibc] Handle 4x4 block cases 2023-06-22 14:39:55 +03:00
Marko Viitanen 31fbf453c1 [ibc] Fix IBCFlag writing with I-frames and clean up some code 2023-06-22 14:39:55 +03:00
Marko Viitanen 8aded6406b [ibc] Fix issue in search 2023-05-08 12:26:06 +03:00
Marko Viitanen 34c7c432f9 [ibc] Fix deblocking for the IBC blocks 2023-05-08 11:58:40 +03:00
Joose Sainio e636db489f [cfg] Parameterize intra rough search granularity 2022-09-28 08:49:38 +03:00
Joose Sainio ed6a0528fe Further make things faster 2022-08-30 15:17:05 +03:00
Joose Sainio b0b2b0e536 Try making ultrafast all intra a bit faster 2022-08-29 14:11:08 +03:00
Marko Viitanen 6de2e2d581 [ibc] Fix some git merge issues and IBC merge candidate selection 2022-08-03 10:46:02 +03:00
Marko Viitanen 6a0e2a062d [ibc] Implement a proper search for IBC based on Inter search 2022-07-29 11:53:11 +03:00
Marko Viitanen 65c017c2f2 [ibc] Add check for above block in IBC search 2022-07-29 11:53:11 +03:00
Marko Viitanen 09e62a68fe [ibc] Fix merge candidate selection bug and IBC HMVP size reset at the start of the lcu row 2022-07-29 11:53:10 +03:00
Marko Viitanen d288cc46e9 [ibc] Fix coding of IBC in P and B slices, enable in search 2022-07-29 11:53:10 +03:00
Marko Viitanen 48584eead9 [ibc] Reset the jccr flags to fix a bug with IBC 2022-07-29 11:53:10 +03:00
Marko Viitanen 7ce01b4826 [ibc] Tune search costs a bit and revert debug vector scaling 2022-07-29 11:53:09 +03:00
Marko Viitanen 0fdf96fab2 [ibc] Change internal MV storage to INTERNAL_MV_PREC and code it as full-pel 2022-07-29 11:53:09 +03:00
Marko Viitanen cc4c757695 [ibc] Fix bugs on IBC reconstruction and add a simple search for I-frames 2022-07-29 11:53:08 +03:00
Marko Viitanen d9164f3cfe [ibc] Simplify the IBC merge candidate and mv cand selection 2022-07-29 11:53:08 +03:00
Marko Viitanen a46a4531a3 [ibc] Add HMVP for IBC and correct AMVP selection 2022-07-29 11:52:17 +03:00
Marko Viitanen dbc2006ba9 [ibc] Implement IBC reconstruction function when blocks are completely in the ibc buffer 2022-07-29 11:52:16 +03:00
Marko Viitanen b49d32af21 [ibc] Add IBC buffers 2022-07-29 11:52:15 +03:00
Marko Viitanen 6ec4c37b47 [ibc] Add IBC Flag context and code the bits, disable by default for now 2022-07-29 11:49:49 +03:00
Marko Viitanen 20d0a9b65e [ibc] Add --ibc parameter and config values for Intra Block Copy 2022-07-29 11:49:49 +03:00
Marko Viitanen cd2d4066d5 Fix scaled MV clipping and remove some unused variables 2022-07-28 13:59:11 +03:00
Marko Viitanen 3dd738ebb5 Fix mv_t rounding problems in some functions 2022-07-27 13:02:37 +03:00
Marko Viitanen 5ce1035291 [debug] Fix Motion Vector debug code not to overflow on videos not divisible by LCU_WIDTH 2022-07-27 12:48:39 +03:00
Marko Viitanen b7b7c22e44 Change mv_t to int32_t because of possible overflow in large videos 2022-07-27 12:48:39 +03:00
Joose Sainio ea32ef33ac [lfnst] handle transform skip correctly during search 2022-07-08 10:57:26 +03:00
Joose Sainio 03fb6ce92e [lfnst] Fix lfnst+tr_skip for dual tree 2022-07-08 10:57:26 +03:00
Joose Sainio 450cd00290 [mts] Fix cost calculation 2022-07-08 10:56:35 +03:00
Joose Sainio f9212b4e44 [mts] Don't do tr-skip when tr-skip is disabled 2022-07-06 15:15:28 +03:00
Joose Sainio dc7c8eeb41 [tr-skip] fix uvg_encode_ts_residual 2022-07-06 10:51:01 +03:00
Joose Sainio 427d611a00 [intra] Perform chroma search for rd2 2022-07-05 12:29:27 +03:00
Joose Sainio e2c34e7c25 [lfnst] Fix lfnst for --rd 2 2022-07-05 12:19:21 +03:00
Joose Sainio 02aa36f1a2 [tests] Fix final issue with avx2 satd and update test results 2022-07-05 10:28:59 +03:00
Joose Sainio 42adfb52a7 [satd] Satd scaling on avx2 implementations and re-enable satd tests 2022-07-05 09:34:59 +03:00
Joose Sainio 1f6a62e70e [fix-up] Force lfnst off when trying the mode from below depth block 2022-07-04 13:45:16 +03:00
Joose Sainio 3de4b99aec [jccr] Fix cost calculation 2022-07-04 13:41:14 +03:00
Joose Sainio 3a6414c31d [dual-tree] Fix deblock 2022-06-30 14:21:03 +03:00
Joose Sainio 5fefea025f [lfnst] get constarints for jccr mode 2022-06-29 16:35:55 +03:00
Joose Sainio b35a75b2eb [lfnst] Fix lfnst with rdoq 2022-06-29 16:25:25 +03:00
Joose Sainio 6ef532775b [intra] Fix various issues with cclm, mip, dual-tree, and lfnst 2022-06-29 15:09:34 +03:00
Joose Sainio 06d277bc78 [doc] update manpage and readme 2022-06-28 16:25:25 +03:00