Commit graph

4650 commits

Author SHA1 Message Date
siivonek b02fb1b1af Remove left shift from planar half functions. Implement the left shift with madd. Planar preds of width 4, 8 and 16 should work now without overflows. Add loop unroll macros to vertical half functions. Will be added to hor half functions later. 2024-05-09 13:06:15 +03:00
siivonek 0eb0f110c2 Add missing packus to the end of planar calculation. 2024-05-09 13:05:49 +03:00
siivonek 4ae234ef24 Implement 16xN planar prediction. 2024-05-09 13:05:49 +03:00
siivonek 4bbd5f4c9a Implement 4xN planar prediction. 2024-05-09 13:05:49 +03:00
siivonek e888be80e2 Add intra avx2 planar placeholder functions. Implement 8xN planar prediction. Note: does not work with height < 4 yet. Initial plan is to produce the planar prediction as two halves. This is subject to change at at this point, it seems only planar functions for different widths are needed. 2024-05-09 13:05:49 +03:00
siivonek c91b76a668 Add globals required by intra avx2 code. Remove unnecessary stuff from intra avx2. 2024-05-09 12:42:38 +03:00
Joose Sainio e5e32d67f4 [avx2] Remove a define that was never meant to be committed 2023-09-27 12:54:53 +03:00
Joose Sainio 9add13b705 update version an docs 2023-09-27 09:47:05 +03:00
Joose Sainio 84580aebb0 Merge branch 'release-prep' into master 2023-09-27 08:11:09 +03:00
Joose Sainio 4a1cd926fb [rdoq] Fix rdoq using uninitialized values that do not matter 2023-09-26 14:26:07 +03:00
Joose Sainio 079d7e9a1a [tests] Fix mts_tests.c to not consider irrelevant elements 2023-09-26 11:36:43 +03:00
Joose Sainio 69c1c948fa [cfg] Specify that MTT and ISP are currently experimental 2023-09-26 10:41:31 +03:00
Joose Sainio e32cf4fb52 [avx2] Re-enable disabled avx2 functions that do not work with non-square blocks 2023-09-26 10:38:29 +03:00
Joose Sainio ff77346527 [dct2] Remove unnecessary memsets 2023-09-26 09:57:47 +03:00
Joose Sainio 64d222d17c [dep_quant] Remove dead code and fix small issue 2023-09-26 09:42:30 +03:00
siivonek 284724398e Add some comments. 2023-09-26 09:21:49 +03:00
Joose Sainio 3d4e732952 [avx2] Fix issue with 16x32 inverse transform 2023-09-26 09:21:49 +03:00
Joose Sainio d62a3f888e [avx2] static all transform tables 2023-09-26 09:21:48 +03:00
Joose Sainio 1f9955bdda [avx2] Fix compilation errors 2023-09-26 09:21:35 +03:00
Joose Sainio 13d4313e02 [avx2] Mostly working 2023-09-26 09:21:29 +03:00
Joose Sainio b78f9aff17 [avx2] Inverses work when ISP is not enabled 2023-09-26 09:21:24 +03:00
siivonek 4dccbcc30d [avx2] Forward transforms seem to be working 2023-09-26 09:21:24 +03:00
Joose Sainio 19829da152 Disable all avx2 optimizations that cannot be used with mtt/isp 2023-09-26 09:21:23 +03:00
Joose Sainio 1c293b8253 pass context_store as pointer
This reverts commit 47c5ea3d5c.
2023-09-26 09:21:23 +03:00
Joose Sainio 2caf077cff Remove avx512 instrincis 2023-09-26 09:21:23 +03:00
Joose Sainio 254826d396 [avx2] Add comments 2023-09-26 09:21:19 +03:00
Joose Sainio f2fb641acb [avx2] Replace inefficient loop with AVX2 code 2023-09-26 09:21:19 +03:00
Joose Sainio bc24601369 [avx2] Improve avx2 version of update_common_context 2023-09-26 09:21:19 +03:00
Joose Sainio 915104cf10 [dep_quant] Change order of absLevels 2023-09-26 09:21:18 +03:00
Joose Sainio d850c346d6 [dep_quant] Change order of ctxInit 2023-09-26 09:21:18 +03:00
Joose Sainio a624988c91 [dep_quant] Separate abs levels and ctx init 2023-09-26 09:21:18 +03:00
Joose Sainio dda972c665 [avx2] Try to do lnz decision with avx2 2023-09-26 09:21:18 +03:00
Joose Sainio cf6f03b73b [avx2] This has worked but I'm pretty sure these should be unaligned 2023-09-26 09:20:56 +03:00
Joose Sainio b4c84e820c [avx2] Simplify 2023-09-26 09:20:56 +03:00
Joose Sainio 2811ce58f4 [avx2] AVX2 version of depquant now exactly matches scalar version 2023-09-26 09:20:56 +03:00
Joose Sainio 48ea4bff4d [dep_quant] Fix rate_estimator and quant_block init cases 2023-09-26 09:20:55 +03:00
Joose Sainio dfff9a8030 [avx2] Move dep quant stuff to strategies 2023-09-26 09:20:55 +03:00
Joose Sainio 0591342b3a [avx2] replace or 2023-09-26 09:20:38 +03:00
Joose Sainio 8b1d6fab59 [avx2] Replace loads and stores with non-avx512 stores 2023-09-26 09:20:37 +03:00
Joose Sainio 6d0a3fa5fc [avx2] Replace _mm_and_epi32 with _mm_and_si128 2023-09-26 09:20:37 +03:00
Joose Sainio 7fdc045690 [dep_quant] Clean up 2023-09-26 09:20:37 +03:00
Joose Sainio 8eb0f66734 [depquant] update_state_eos_avx2 working 2023-09-26 09:20:37 +03:00
Joose Sainio 00cc58bc55 [depquant] Only initialize rate_estimator when necessary 2023-09-26 09:20:37 +03:00
Joose Sainio 00f838306f [depquant] Initialize quant_block only when necessary 2023-09-26 09:20:37 +03:00
Joose Sainio 9e27b4056a [avx2] WIP update_state_eos_avx2 2023-09-26 09:20:36 +03:00
Joose Sainio c56350b8d6 [avx2] and last 2023-09-26 09:20:36 +03:00
Joose Sainio 9f69713c24 [depquant] remove an unnecessary memcpy 2023-09-26 09:20:36 +03:00
Joose Sainio aa48943c22 [avx2] Do decision cost comparison with avx2 2023-09-26 09:20:36 +03:00
Joose Sainio cd6110cfac [depquant] Pre calculate things
sig_ctx_offset gtx_ctx_offset cg_pos pos_y  pos_x next_sbb_right next_sbb_below
2023-09-26 09:20:30 +03:00
Joose Sainio 8f4c3cecbf [avx2] update_states_avx2 working 2023-09-26 09:20:29 +03:00