hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-28 11:44:08 +00:00
Code Issues Actions 1 Projects Releases Wiki Activity
1280 commits 3 branches 6 tags 10 MiB
Commit graph

1145 commits

Author SHA1 Message Date
Ari Koivula 38b224cf69 Change rest of cu split search costs to double. 2014-10-08 10:31:54 +03:00
Ari Koivula 17473624d3 Add transform tree bit costs for cbf_luma. 2014-10-08 10:31:54 +03:00
Ari Koivula 3b04d39db4 Take cabac bits into account on transform tree. 2014-10-08 10:31:54 +03:00
Ari Koivula 296f142d9e Retain coded block flag data during transform split search. 2014-10-08 10:31:54 +03:00
Ari Koivula 85dea10f3f Clean up transform split search. 
- Remove unnecessary checks and comment.
 2014-10-08 10:31:54 +03:00
Ari Koivula e1b801eb6f Add transform tree chroma cbf bits. 2014-10-08 10:31:23 +03:00
Ari Koivula 3868cc7ff1 Fix crash on inter search when --tr-depth-intra is used. 
- Transform splits meant for intra modes were used for inter when inter mode
  was chosen, which caused an assert to be triggered if the split transform
  block didn't have any coefficients.
 2014-10-03 19:29:06 +03:00
Ari Lemmetti bcf12567d0 Added some comments. 2014-10-03 17:51:58 +03:00
Ari Lemmetti fea517c2ae Misc code cleanup 2014-10-03 17:06:09 +03:00
Ari Lemmetti 85682c3b6a Removed unused transpose functions. 2014-10-03 11:39:31 +03:00
Ari Koivula 8a80845b91 Add chroma to transform split search. 2014-10-03 11:36:57 +03:00
Ari Koivula 51662e1081 Fix differences between cu_rd_cost_luma and rdo_cost_intra. 2014-10-03 11:36:57 +03:00
Ari Koivula bc7d7d5cb6 Add cu_info* as parameter to reconstruction functions. 
- This is required so these functions can be used for searching. When NULL
  is given they take the CU from LCU struct as they did previously.

Conflicts:
	src/search.c
 2014-10-03 11:36:56 +03:00
Ari Koivula ccc575e2c6 Disable transform tree bits. 2014-10-03 11:36:56 +03:00
Ari Koivula a0ab469c89 Disable rdo_cost_intra. 2014-10-03 11:36:56 +03:00
Ari Koivula c164978e21 Add FULL_CU_SPLIT_SEARCH macro for disabling cu split optimization. 2014-10-03 11:36:56 +03:00
Ari Koivula 549ac96438 Change costs to doubles to avoid rounding intermediate results. 
- Helps with debugging.
 2014-10-03 11:36:56 +03:00
Ari Koivula e591e89ade Add prediction mode to chroma reconstruction parameters. 
- Just like in luma.
 2014-10-03 11:36:56 +03:00
Ari Koivula f6272f06fc Unify signature for transform functions. 
- Some used block, coeff and some src, dst. Now all signatures are const input
  and non-const output.
 2014-10-03 11:21:43 +03:00
Ari Koivula b932cf4b21 Clean up avx2 dct macros. 2014-10-03 11:16:25 +03:00
Ari Koivula 47244a15c3 Merge branch 'dct-optimizations' 
Conflicts:
	src/strategies/avx2/dct-avx2.c
	src/strategies/generic/dct-generic.c
 2014-10-02 13:45:21 +03:00
Ari Lemmetti 61e1510480 Transform functions in dct-avx2.c are now generated with macros. 2014-10-02 13:24:30 +03:00
Ari Lemmetti 9407610555 Moved DCT / DST matrices to dct-generic.c 2014-10-02 13:24:30 +03:00
Ari Lemmetti 7255112bd8 Added transposed DCT/DST tables. Use them while calculating transforms instead of doing runtime transpose. Added separate functions for DST and IDST. 2014-10-02 13:24:30 +03:00
Ari Lemmetti e7bcb58846 Added 32x32 IDCT 2014-10-02 13:24:30 +03:00
Ari Lemmetti eacf173b7e Added 32x32 DCT for AVX2 2014-10-02 13:24:30 +03:00
Ari Lemmetti d2856a5d40 Added 32x32 transpose 2014-10-02 13:24:30 +03:00
Ari Lemmetti 7a33f08312 Added 16x16 DCT and IDCT for AVX2 2014-10-02 13:24:30 +03:00
Ari Lemmetti d2fe2a5391 Added 16x16 transpose 2014-10-02 13:24:30 +03:00
Ari Lemmetti d6af146a2e Added part of the functions 16x16 DCT needs 2014-10-02 13:24:30 +03:00
Ari Lemmetti aba3acdfff Added AVX2 optimized transforms for 4x4 and 8x8 blocks 2014-10-02 13:24:30 +03:00
Ari Lemmetti 5856f32d81 Fixed incorrect shift values for inverse transforms in generic strategy 2014-10-02 13:24:29 +03:00
Ari Lemmetti 41b032664d First version of 4x4 forward DCT 2014-10-02 13:24:29 +03:00
Ari Koivula 36232619ab Fix broken cabac contexts in wpp. 
- Fixes #84.
- The issue was caused by 241b9d6 naively copying the whole struct, which
  contains data other than just the contexts. Rather than reverting the
  change, the struct was refactored to have another struct that contained
  just the contexts.
 2014-09-24 01:02:52 +03:00
Ari Koivula 4e052d3f0f Wrap contexts of cabac_data inside cabac_data.ctx struct. 2014-09-24 01:02:37 +03:00
Ari Koivula b339004c4c Rename cabac_state.ctx to cur_ctx. 2014-09-24 01:02:28 +03:00
Ari Koivula 8b8b53fba5 Merge branch 'sao_cabac' 2014-09-22 10:28:30 +03:00
Ari Koivula bfa399c8fc Fix compiler warnings. 
- Non-parenthesized parameter in a macro.
- Unused variables.
- Wrong const qualifiers.
- Signed/unsigned comparison.
 2014-09-22 10:04:57 +03:00
Marko Viitanen 6f65a9cbbd Improved SAO merge decisions 2014-09-16 10:08:17 +03:00
Marko Viitanen 21df11ba4e Implemented SAO search for both chroma components 2014-09-15 16:07:31 +03:00
Marko Viitanen e8d1140a1a Check SAO band offset for both chroma components and better SAO chroma cabac costs 2014-09-15 16:07:31 +03:00
Marko Viitanen 0c92031e8a SAO merge checking cleanup 2014-09-15 16:07:31 +03:00
Marko Viitanen b274e7adcd Added cabac bit cost calculations to SAO search 2014-09-15 16:07:31 +03:00
Ari Koivula 5f732126c3 Add cabac bit costs float table. 2014-09-15 15:45:43 +03:00
Ari Koivula 0db7d8d20f test cu split cost 2014-09-15 15:42:03 +03:00
Ari Koivula 35b2e6f755 Add missing cabac context for chroma cbf. 
- The context was also missing from HM, but has been fixed in HM13.
 2014-09-15 15:41:44 +03:00
Ari Koivula 241b9d6adb Simplify cabac context copying. 
Conflicts:
	src/context.c
 2014-09-15 15:41:44 +03:00
darealshinji 61a414bced reposition colons in usage message to match with the rest 2014-09-15 03:40:18 +02:00
Ari Koivula 3c73892609 Fix transform split search. 
- Redo the search with the best mode to make sure the tr_depth parameters are
  correct.
 2014-09-11 10:56:53 +03:00
Ari Koivula 46b6b1243b Add --rd=3 mode and enable searching of intra depth 0. 
- intra_build_reference_border was overflowing at depth 0 because it uses
  arrays just large enough to accommodate 32x32 transforms, which is the
  biggest transform.
- For similar reasons search_intra_rough doesn't work at depth 0.
- The --rd=3 mode tries all modes with transform search. It also works without
  rough search so it was used to test depth 0 search. If --rd=3 is not on intra
  split at depth 0 is not searched for.

Conflicts:
	src/search.c
 2014-09-11 10:54:41 +03:00
Ari Koivula c5fa824347 Rebase transform split search. 2014-09-08 14:13:59 +03:00
Ari Koivula 79b86ce6e1 Add --tr-depth-intra command line option. 
Conflicts:
	src/encoder.c
 2014-09-04 13:42:24 +03:00
Marko Viitanen fe236de807 Fixed sps_max_dec_pic_buffering value to include current picture 2014-09-01 10:31:11 +03:00
Marko Viitanen dbcc8d65aa Removed duplicate function from RDOQ 2014-08-28 08:50:01 +03:00
Ari Koivula 931ec7301c Put slice delta QP to bitstream. 
- Before slice delta QP was always 0. Now if global->QP is changed before
  contexts are set, the delta qp is put to the bitstream, allowing for rough
  frame level rate control.
 2014-08-25 16:43:23 +03:00
Ari Koivula 4c3bbd4a35 Rewrite the SContruct. 
- Works with new /strategy/ structure.
- Change architecture selection to use arch= instead of construction target.
 2014-08-25 16:43:23 +03:00
Ari Lemmetti f88c3b6f37 Removed unnecessary if (both branches did the same thing) 2014-08-20 11:54:35 +03:00
Laurent Fasnacht f3c311fe1a Fix commit 8502f3d 2014-08-11 15:17:15 +02:00
Laurent Fasnacht f9bffe35a5 Log tile id in sad perf log 2014-08-11 11:57:08 +02:00
Laurent Fasnacht 6a937de9b2 Fix search_cu log 2014-08-11 11:57:08 +02:00
Laurent Fasnacht 8502f3d850 Improve logging 2014-08-11 11:57:07 +02:00
Laurent Fasnacht f1b303a2d2 Fix compilation errors 2014-08-11 09:53:06 +02:00
Ari Lemmetti 47e3bcfb50 Fixed incorrect shift values for inverse transforms in generic strategy 2014-08-07 16:01:30 +03:00
Ari Lemmetti 709520a233 Removed all AVX2 instructions from SATD functions. 
-Zero extend macro now returns results in 2 xmm registers instead of one ymm
 2014-07-31 13:25:28 +03:00
Ari Lemmetti 0beb278f5b Partial butterfly strategy is now called DCT strategy. Made changes to transform functions in preparation for optimizations. 
-Moved fast_forward_dst and fast_inverse_dst to DCT strategies
 2014-07-31 13:25:28 +03:00
Ari Lemmetti 6bf63bd171 Added AVX2 strategy for partial butterfly (no optimizations yet) 2014-07-31 13:25:28 +03:00
Ari Lemmetti faccc4f09b Partial butterfly functions now utilize the strategy selector 2014-07-31 13:25:28 +03:00
Ari Koivula c2fac805d7 Give HAVE_ALIGNED_STACK to yasm on windows. 
- Linux gets it through some other means but on windows it needs to be
  given explicitly.

- Fixes issue #78.
 2014-07-30 16:26:23 +03:00
Ari Koivula 669e99dd7f Improve intra SAD AVX2 intrinsics. 
- Moved implementations for different sizes to inline functions that are
  defined using each other, reducing the amount of redundant code.

- Performance of sad_8bit_32x32_avx2 improved by about 10% due to unrolling of
  the loop.
 2014-07-25 15:59:55 +03:00
Ari Koivula e00102f0ca Compile asm optimizations only if yasm is present. 2014-07-23 14:57:40 +03:00
Ari Lemmetti 85fb0784e4 Fixed intendentation and added some empty lines for readability 2014-07-23 12:32:27 +03:00
Ari Lemmetti bd6e89c1f0 Updated include directories and file names to Makefile 2014-07-22 15:36:54 +03:00
Ari Lemmetti 4f88ebce5a Added comments and made visual studio not to compile x86inc.asm 2014-07-22 15:07:57 +03:00
Ari Koivula cfd3636e08 Move some repetitive SATD asm into a macro. 
Conflicts:
	src/strategies/x86_avx/picture_x86.asm
 2014-07-22 12:46:39 +03:00
Ari Lemmetti c81639dd09 Removed old unused macro 2014-07-22 11:11:20 +03:00
Ari Lemmetti cf0797cafd Reordered and intended assembly code 2014-07-22 11:07:42 +03:00
Ari Lemmetti fea44c8234 Renaming AVX/asm files 
-Splitted SAD and SATD functions in separate files
 2014-07-21 18:02:01 +03:00
Ari Lemmetti a64df6f0d0 Merge branch 'asm' 
Conflicts:
	build/kvazaar_lib/kvazaar_lib.vcxproj.filters
	src/Makefile
	src/strategies/strategies-picture.c
 2014-07-21 16:41:09 +03:00
Ari Lemmetti 1be2c3aae5 Preparing push to master and misc 
-Removed unnecessary <math.h> headers
-Updated AVX/asm optimizations to match the new file hierarchy
-Makefile only compiles .asm files if KVAZAAR_DISABLE_YASM is not set to 1 and TARGET_CPU_ARCH is x86
 2014-07-21 12:39:56 +03:00
Ari Koivula a8f7103797 Add AVX2 implementations for sad_8bit_ 8x8, 16x16 and 32x32. 2014-07-18 18:27:30 +03:00
Ari Koivula 3daa5dd1f1 Add sse2 implementaton for sad_8bit_4x4. 2014-07-18 18:20:34 +03:00
Ari Koivula f49332c9b8 Add missing includes. 2014-07-18 17:56:15 +03:00
Ari Koivula 291817667f Tidy up the Makefile. 2014-07-18 17:31:18 +03:00
Ari Koivula e241866f43 Compile intrinsic functions with appropriate flags in gcc. 
- Remove -march=native as it's no longer necessary for intrinsics to work.
  Closes #77.

- I couldn't test altivec or sse4.1, but sse4.1 compiles so I expect it
  to work.
 2014-07-18 17:28:14 +03:00
Ari Koivula 5662621b3c Free threadqueue jobs when they are not needed. 
- Also add destroying the mutex when the job is freed.

- This makes Kvazaar no longer acquire thousands of OS handles on Windows.
 2014-07-16 16:51:20 +03:00
Ari Lemmetti 1e94262f85 Made AVX asm compatible with the changed system 
- x86inc.asm is now located in extras
- Removed unused cpu.asm/h
 2014-07-14 18:51:17 +03:00
Ari Lemmetti 683eda1183 Merge branch 'master' into asm 
Conflicts:
	build/kvazaar_lib/kvazaar_lib.vcxproj
	build/kvazaar_lib/kvazaar_lib.vcxproj.filters
	src/Makefile
	src/strategies/strategies-picture.c
 2014-07-14 16:42:33 +03:00
Ari Lemmetti 7f873e037c Updated Makefile to compile picture_x86.asm 2014-07-14 15:30:08 +03:00
Ari Lemmetti 2169f9ab8c Added AVX asm comments and fixes 
-Added vzeroupper to satd macro to prevent AVX-SSE transition penalties int picture_x86.asm
-Fixed the order of registers in zero extend macro in picture_x86.asm
-Fixed SATD checkers test pattern in satd_tests.c
 2014-07-14 14:43:36 +03:00
Ari Koivula 5d0df56c94 Move optimizations to their own compilation units according to target. 
- This is necessary in order to compile AVX intrinsics correctly in
  Visual Studio. Having everything in their own units should also make
  compiling normal C code with optimizations on easier.

- For now the makefile still relies on GCC __target__ attribute for compiling
  intrinsics.
 2014-07-11 17:26:19 +03:00
Ari Koivula f605d6c35b Align intra buffers to 32 bytes for 256 bit SIMD instructions. 2014-07-11 17:26:19 +03:00
Ari Koivula fbd03b706e Reconfigure VS project. 
- Moved compilation flag stuff from project file to the abstraction layer.

- Disabled randomized base address as unnecessary.

- Disable stack buffer security check from release.
 2014-07-11 17:26:19 +03:00
Laurent Fasnacht 72abc69b3d Measure time for SAD in _DEBUG mode 2014-07-08 11:42:58 +02:00
Laurent Fasnacht 1a318c714d log poc with new_frame 2014-07-08 11:42:19 +02:00
Laurent Fasnacht e64a692780 Add CU type in threadqueue.log 2014-07-08 09:06:31 +02:00
Laurent Fasnacht abfbb7cad3 Fix duplicate type key in threadqueue.log 2014-07-07 11:36:50 +02:00
Laurent Fasnacht 946e3b9651 Log search_cu to threadqueue.log 2014-07-07 10:50:05 +02:00
Laurent Fasnacht f62e571c15 Add missing info to threadqueue.log 2014-07-07 10:49:40 +02:00
Ari Lemmetti 048127c7e3 AVX assembly optimizations improved 2014-07-02 16:57:06 +03:00
Ari Koivula 7ecf78bb70 Use sqrt lambda cost for searches not using SSD. 
- Add encoder_state->global->cur_lambda_cost_sqrt.

- Use sqrt lambda for inter search and rough intra search.

- The effect on inter is around 10-20% bdrate. The effect on intra is smaller
  and non-existent when --rd=2 is enabled, as the intra search refinement was
  already done with SSD and correct lambda.
 2014-06-26 13:56:38 +03:00