The ffmpeg from ffvvc

When using bit depths higher than 10-bit, the sps_range_extension may be include in the SPS. This adds five new flags, primarily to increase the dynamic range. It is possible to use higher bit depths without the range extension, however their benefit will be reduced.

optimize put_uni_pixels_N_128x128 AVX2/SSE4 code

see #146 (comment)
we have a similar issue for put_pixels too, see #145 (comment)

Adding the transformed residual directly to the input.

See discussion at https://patchwork.ffmpeg.org/project/ffmpeg/patch/TYSPR06MB643369DCA30C6FACB0065E2CAA62A@TYSPR06MB6433.apcprd06.prod.outlook.com/#81890

Add github workflow

Check ARM peformance

Please help run performance tests for the following clips on ARM chip
comand line: ./ffmpeg -i xxx.266 -vsync 0 -f rawvideo /dev/null -y

and fill following table:

clips	FPS(3 times average)
RitualDance_1920x1080_60_10_420_32_LD.26
RitualDance_1920x1080_60_10_420_37_RA.266
Tango2_3840x2160_60_10_420_27_LD.266

top 10 functions for Tango2_3840x2160_60_10_420_27_LD.266

functions	percent
xxx	xx%

Thanks

Add AVX2 assembly code for inter predict

please add AVX for VVCInterDSPContext

function	8 bits	10 bits	12 bits
put	✅	✅	✅
put_uni	✅	✅	✅
put_uni_w	⬜	⬜	⬜
avg	✅	✅	✅
avg_w	✅	✅	✅
put_ciip	⬜	⬜	⬜
put_gpm	⬜	⬜	⬜
fetch_samples	⬜	⬜	⬜
bdof_fetch_samples	⬜	⬜	⬜
prof_grad_filter	⬜	⬜	⬜
apply_prof	⬜	⬜	⬜
apply_prof_uni	⬜	⬜	⬜
apply_prof_uni_w	⬜	⬜	⬜
apply_prof_bi	⬜	⬜	⬜
apply_prof_bi_w	⬜	⬜	⬜
apply_bdof	⬜	⬜	⬜
sad	⬜	⬜	⬜
dmvr	⬜	⬜	⬜

please add functions based on usage
11.96% ffmpeg_g [.] put_vvc_luma_hv_10
5.88% ffmpeg_g [.] alf_get_coeff_and_clip_10
5.25% ffmpeg_g [.] ff_vvc_inv_dct2_64
4.30% [kernel] [k] __lock_text_start
4.22% ffmpeg_g [.] ff_vvc_alf_filter_luma_w16_16bpc_avx2
3.46% ffmpeg_g [.] put_vvc_luma_bi_hv_10
3.45% ffmpeg_g [.] alf_filter_luma_vb_10
3.13% ffmpeg_g [.] vvc_loop_filter_luma_10
2.81% ffmpeg_g [.] lmcs_filter_luma_10
2.46% ffmpeg_g [.] put_vvc_luma_uni_hv_10
2.27% ffmpeg_g [.] put_vvc_chroma_hv_10
2.21% libc-2.31.so [.] 0x000000000018b733
2.05% libc-2.31.so [.] 0x000000000018bb41
1.95% ffmpeg_g [.] put_vvc_chroma_uni_hv_10
1.84% ffmpeg_g [.] put_vvc_chroma_bi_hv_10
1.81% ffmpeg_g [.] vvc_deblock_bs
1.41% ffmpeg_g [.] ff_vvc_predict_inter
1.25% libpthread-2.31.so [.] __pthread_mutex_lock
1.24% libpthread-2.31.so [.] __pthread_mutex_unlock
1.22% ffmpeg_g [.] ff_vvc_residual_coding
1.08% ffmpeg_g [.] alf_filter_cc_10
1.03% ffmpeg_g [.] apply_prof_uni_10
0.99% ffmpeg_g [.] ff_vvc_alf_filter
0.98% ffmpeg_g [.] ff_vvc_inv_dct2_32
0.94% ffmpeg_g [.] vvc_deblock_bs_luma_vertical
0.92% ffmpeg_g [.] add_residual_10

LTRP_A_ERICSSON failure with assembly optimisations enabled

The test LTRP_A_ERICSSON fails when assembly optimisations are enabled. This affects NASM and YASM on both Linux and Windows. This is not a duplicate of #26 - applying patch bab47ca fixes all valgrind memcheck errors, but the test still fails.

Intermittent Windows build failure

Windows CI build seems to fail randomly with unable to find mspdbcore.dll. Here is an example. Thought this had been fixed by 68c056f but seemingly not.

Support PCM

https://github.com/ffvvc/tests/tree/main/conformance/failed/BDPCM

Support crop

https://github.com/ffvvc/tests/tree/main/conformance/failed/CROP

Add AVX2 assembly code for LMCS filter

based on this LMCS consume about 2.81% time for Tango2_3840x2160_60_10_420_27_LD.266, maybe we can use VPGATHERDD to optimize it.

11.96% ffmpeg_g [.] put_vvc_luma_hv_10
5.88% ffmpeg_g [.] alf_get_coeff_and_clip_10
5.25% ffmpeg_g [.] ff_vvc_inv_dct2_64
4.30% [kernel] [k] __lock_text_start
4.22% ffmpeg_g [.] ff_vvc_alf_filter_luma_w16_16bpc_avx2
3.46% ffmpeg_g [.] put_vvc_luma_bi_hv_10
3.45% ffmpeg_g [.] alf_filter_luma_vb_10
3.13% ffmpeg_g [.] vvc_loop_filter_luma_10
2.81% ffmpeg_g [.] lmcs_filter_luma_10
2.46% ffmpeg_g [.] put_vvc_luma_uni_hv_10
2.27% ffmpeg_g [.] put_vvc_chroma_hv_10
2.21% libc-2.31.so [.] 0x000000000018b733
2.05% libc-2.31.so [.] 0x000000000018bb41
1.95% ffmpeg_g [.] put_vvc_chroma_uni_hv_10
1.84% ffmpeg_g [.] put_vvc_chroma_bi_hv_10
1.81% ffmpeg_g [.] vvc_deblock_bs
1.41% ffmpeg_g [.] ff_vvc_predict_inter
1.25% libpthread-2.31.so [.] __pthread_mutex_lock
1.24% libpthread-2.31.so [.] __pthread_mutex_unlock
1.22% ffmpeg_g [.] ff_vvc_residual_coding
1.08% ffmpeg_g [.] alf_filter_cc_10
1.03% ffmpeg_g [.] apply_prof_uni_10
0.99% ffmpeg_g [.] ff_vvc_alf_filter
0.98% ffmpeg_g [.] ff_vvc_inv_dct2_32
0.94% ffmpeg_g [.] vvc_deblock_bs_luma_vertical
0.92% ffmpeg_g [.] add_residual_10

Support subpic

https://github.com/ffvvc/tests/tree/main/conformance/failed/SUBPIC

valgrind issue for SAO filter.

run "valgrind ./ffmpeg_g -i SLICES_A_HUAWEI_3.bit -f rawvideo /dev/null -y"

you will get

==5154== Use of uninitialised value of size 8
==5154== at 0x46E433: sao_edge_filter_10 (vvcdsp_template.c:200)
==5154== by 0x50ACE9: ff_vvc_sao_filter (vvc_filter.c:309)
==5154== by 0x52C1DB: run_sao (vvc_thread.c:447)
==5154== by 0x52B462: ff_vvc_task_run (vvc_thread.c:567)
==5154== by 0x500AE9: executor_worker_task (executor.c:79)
==5154== by 0x49C8608: start_thread (pthread_create.c:477)
==5154== by 0x4B02132: clone (clone.S:95)

It's realted to sao temporary buffer lc->sao_buffer, once you memset the buffer here , valgrind will not report any issue.

Windows multithreading test failure

Had some time so did a little bit more research on the Windows CI (#52) test failure.

memsetting the entire lc->sao_buffer like bab47ca does not fix the issue, so I don't think the issue is related to #26. With this change, valgrind and clang's address sanitiser don't report any memory issues.

I have compiled FFmpeg directly with MSVC/MSYS2 (i.e. not via FFVS-Project-Generator) and the problem is similar so I don't think it's anything to do with the build files. I haven't yet got the gcc/MSYS2 toolchain or MinGW gcc cross-compilation working unfortunately.

I can't get the LTRP_A_ERICSSON_3 failure to reproduce on my machine, so I don't think there's anything special about this test. The tests which fail most frequently on my machine are:

LMCS_A_Dolby_3.bit
WPP_B_Sharp_2.bit

The failures only occur when running tests concurrently, they do not occur when running the tests individually or when running tests using a single thread. I don't know whether this points towards libavcodec/vvc_thread.c at all? ~~This line is part of what is preventing cross-compilation at the moment. Should it not be testing for compiler using _MSC_VER or something instead of checking the OS?~~ See #57 for fix. Don't believe this is related.

Enable checkasm on windows.

Based on this #60 (comment).
It's highly possible the github windows runner has AVX2 support.
Please help enable chehckasm test on windows

Support fields

https://github.com/ffvvc/tests/tree/main/conformance/failed/FIELD

List of assembly code that can be reused from hevc

http://ffmpeg.org/pipermail/ffmpeg-devel/2023-January/305656.html

remove min_scan_x, min_scan_y and find a better way to get max_scan_x and max_scan_y.

see https://patchwork.ffmpeg.org/project/ffmpeg/patch/TYSPR06MB64337C13ECE97978A56C1152AA62A@TYSPR06MB6433.apcprd06.prod.outlook.com/#81903

Refact Executor to reduce shedule cost

The current executor will use a single queue for not-ready and ready tasks.
So, every task posted to the queue may wake all threads.
Better split it into two queues: 1 for not ready tasks. 1 for ready tasks. it is protected by different locks
The thread adds a task to the executor who will also check readiness for all tasks in the not-ready queue.
once it is ready it will add to the ready queue.

the target is reduce the lock an unlock usage in perf top result
11.96% ffmpeg_g [.] put_vvc_luma_hv_10
5.88% ffmpeg_g [.] alf_get_coeff_and_clip_10
5.25% ffmpeg_g [.] ff_vvc_inv_dct2_64
4.30% [kernel] [k] __lock_text_start
4.22% ffmpeg_g [.] ff_vvc_alf_filter_luma_w16_16bpc_avx2
3.46% ffmpeg_g [.] put_vvc_luma_bi_hv_10
3.45% ffmpeg_g [.] alf_filter_luma_vb_10
3.13% ffmpeg_g [.] vvc_loop_filter_luma_10
2.81% ffmpeg_g [.] lmcs_filter_luma_10
2.46% ffmpeg_g [.] put_vvc_luma_uni_hv_10
2.27% ffmpeg_g [.] put_vvc_chroma_hv_10
2.21% libc-2.31.so [.] 0x000000000018b733
2.05% libc-2.31.so [.] 0x000000000018bb41
1.95% ffmpeg_g [.] put_vvc_chroma_uni_hv_10
1.84% ffmpeg_g [.] put_vvc_chroma_bi_hv_10
1.81% ffmpeg_g [.] vvc_deblock_bs
1.41% ffmpeg_g [.] ff_vvc_predict_inter
1.25% libpthread-2.31.so [.] __pthread_mutex_lock
1.24% libpthread-2.31.so [.] __pthread_mutex_unlock
1.22% ffmpeg_g [.] ff_vvc_residual_coding
1.08% ffmpeg_g [.] alf_filter_cc_10
1.03% ffmpeg_g [.] apply_prof_uni_10
0.99% ffmpeg_g [.] ff_vvc_alf_filter
0.98% ffmpeg_g [.] ff_vvc_inv_dct2_32
0.94% ffmpeg_g [.] vvc_deblock_bs_luma_vertical
0.92% ffmpeg_g [.] add_residual_10

see https://github.com/ffvvc/FFmpeg/wiki/Introduction-to-the-code-structures#executorc for ref.

4:4:4 inter- failure

Conformance tests utilising 4:4:4 colour and inter-prediction are currently failing.

[vvc @ 0000010ed137d040] Skipping NAL unit 12 0KB sq=    0B f=0/0
[vvc @ 0000010ed137d040] Skipping NAL unit 13
[vvc @ 0000010ed137d040] Skipping NAL unit 24 0KB sq=    0B f=0/0
[vvc @ 0000010ed1374250] Intra Block Copy is not implemented. Update your FFmpeg version to the newest one from Git. 
If the problem still occurs, it means that your file has a feature which has not been implemented.
[vvc @ 0000010ed137d040] Skipping NAL unit 24
    Last message repeated 3 times
[vvc @ 0000010ed1374610] Intra Block Copy is not implemented. Update your FFmpeg version to the newest one from Git. 
If the problem still occurs, it means that your file has a feature which has not been implemented.
[vvc @ 0000010ed137d040] Skipping NAL unit 24
[vvc @ 0000010ed137d040] Skipping NAL unit 24 0KB sq=    0B f=0/0
    Last message repeated 3 times
Input #0, vvc, from 'carvideo_yuv422p10le.vvc':
  Duration: N/A, bitrate: N/A
  Stream #0:0: Video: vvc (Main 10 4:4:4), yuv422p10le(tv), 640x360, 25 fps, 1200k tbr, 1200k tbn

https://i.ibb.co/H7rPrVt/uiecfbv.png

  Stream #0:0: Video: vvc (Main 10 4:4:4), none(tv), 640x360, 25 fps, 1200k tbr, 1200k tbn
[vvc @ 00000179d5c3fbf0] Skipping NAL unit 12
[vvc @ 00000179d5c3fbf0] Skipping NAL unit 13
[vvc @ 00000179d5c3fbf0] Skipping NAL unit 24
[vvc @ 00000179ce69d330] Palette is not implemented. Update your FFmpeg version to the newest one from Git. If the problem still occurs, it means that your file has a feature which has not been implemented.
[vvc @ 00000179d5c3fbf0] Skipping NAL unit 24
    Last message repeated 3 times
[vvc @ 00000179d4c2c100] Palette is not implemented. Update your FFmpeg version to the newest one from Git. If the problem still occurs, it means that your file has a feature which has not been implemented.
[vvc @ 00000179d5c3fbf0] Skipping NAL unit 24

green screenshot

I understand that VVC decoder is only 10bit.

LInk codec ffmpeg | JVETVVC 21.0
https://www.sendspace.com/file/xn02cy

Go through ffmpeg's upstreeam check list

https://ffmpeg.org/developer.html#New-codecs-or-formats-checklist
https://ffmpeg.org/developer.html#Patch-submission-checklist

add prefix for Executor and Task

see http://ffmpeg.org/pipermail/ffmpeg-devel/2023-January/305656.html

Add AVX2 assembly code for SAO filter

VVC SAO near same as HEVC. We can fully reuse HEVC SAO code for VVC.
Please add AVX2 for VVCSAODSPContext

Set up windows ci

Please use https://github.com/ShiftMediaProject/FFVS-Project-Generator#generating-custom-project-files to set up windows ci

Write a ducument to introduce code structures

sao asm mismatch with hevc

There are several code mismatches with hevc.
like https://github.com/ffvvc/FFmpeg/blob/main/libavcodec/x86/vvc_sao.asm#L193, the EDGE_SRCSTRIDE
and
https://github.com/ffvvc/FFmpeg/blob/main/libavcodec/x86/vvc_sao.asm#L173
@zackerthescar do you know why?

thank you

Optimze ALF

For https://github.com/ffvvc/tests/blob/main/performance/RitualDance_1920x1080_60_10_420_37_RA.266, ALF use about 30~40% cpu resources. May need to vectorize the alf filter

the following table is an old one, please do
"./ffmpeg -i RitualDance_1920x1080_60_10_420_37_RA.266 -f rawvideo /dev/null -y" by yourself

28.82% ffmpeg_g ffmpeg_g [.] alf_filter_luma_10
7.43% ffmpeg_g ffmpeg_g [.] put_vvc_luma_hv_10
5.48% ffmpeg_g ffmpeg_g [.] ff_vvc_inv_dct2_64
5.30% ffmpeg_g ffmpeg_g [.] alf_get_coeff_and_clip_10
3.31% ffmpeg_g ffmpeg_g [.] inv_dct2.constprop.4
3.11% ffmpeg_g ffmpeg_g [.] alf_filter_chroma_10
3.04% ffmpeg_g ffmpeg_g [.] lmcs_filter_luma_10
2.50% ffmpeg_g ffmpeg_g [.] ff_vvc_deblocking_ctb_boundary_strengths
2.28% ffmpeg_g [kernel.kallsyms] [k] clear_page_erms
1.81% ffmpeg_g ffmpeg_g [.] put_vvc_luma_bi_hv_10
1.81% ffmpeg_g libc-2.31.so [.] 0x000000000018ba51
1.52% ffmpeg_g ffmpeg_g [.] ff_vvc_inv_dst7_32
1.50% ffmpeg_g ffmpeg_g [.] vvc_loop_filter_luma_10
1.46% ffmpeg_g ffmpeg_g [.] put_vvc_luma_uni_hv_10
1.37% ffmpeg_g [kernel.kallsyms] [k] __lock_text_start
1.30% ffmpeg_g ffmpeg_g [.] itransform.isra.0
1.27% ffmpeg_g ffmpeg_g [.] hls_coding_unit
1.27% ffmpeg_g ffmpeg_g [.] put_vvc_chroma_hv_10
1.27% ffmpeg_g ffmpeg_g [.] put_vvc_chroma_uni_hv_10
1.25% ffmpeg_g libc-2.31.so [.] 0x000000000018b643
1.17% ffmpeg_g ffmpeg_g [.] ff_vvc_residual_coding
1.08% ffmpeg_g ffmpeg_g [.] ff_vvc_deblock_filter
1.01% ffmpeg_g ffmpeg_g [.] ff_vvc_inv_dct2_32
0.85% ffmpeg_g ffmpeg_g [.] alf_filter_cc_10
0.79% ffmpeg_g ffmpeg_g [.] put_vvc_chroma_bi_hv_10
0.69% ffmpeg_g ffmpeg_g [.] alf_prepare_buffer
0.65% ffmpeg_g ffmpeg_g [.] intra_pred_10
0.55% ffmpeg_g ffmpeg_g [.] apply_prof_uni_10
0.53% ffmpeg_g libc-2.31.so [.] 0x000000000018b7b3
0.47% ffmpeg_g ffmpeg_g [.] ff_vvc_apply_dmvr_info_ctb
0.47% ffmpeg_g ffmpeg_g [.] put_vvc_luma_v_10

use tb->max_scan_x and tb->max_scan_y to optimze dct8 and dst7

by the nature of transformation, We only have a few no zero coeffes. the position is recorded by tb->max_scan_x and tb->max_scan_y.

We can use this to optimize

FFmpeg/libavcodec/vvc_itx_1d.c

Line 146 in 3cb136d

    
           static void matrix_mul(int *out, const ptrdiff_t out_stride, const int *in, const ptrdiff_t in_stride, const int8_t* matrix, const int size)

see https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/-/blob/master/source/Lib/CommonLib/TrQuant_EMT.cpp#L886 for reference.

./libavcodec/x86/mathops.h:125: Error: operand type mismatch for `shr'

A recent update to binutils has broken FFmpeg compilation: https://trac.ffmpeg.org/ticket/10405. There is a fix in FFmpeg master, so we will need to merge/rebase to allow building with binutils >= 2.41.

Use avpriv_report_missing_feature instead of av_assert0

see http://ffmpeg.org/pipermail/ffmpeg-devel/2023-January/305656.html

random md5 mismatch issue

TEMPSCAL_A_Panasonic_4.bit:
https://github.com/ffvvc/FFmpeg/actions/runs/5445305861/jobs/9904363017 c code only
https://github.com/ffvvc/FFmpeg/actions/runs/5865552084/job/15902678600 c code only

TREE_C_HHI_3.bit:
https://github.com/ffvvc/FFmpeg/actions/runs/5538781956/jobs/10109072165

Use dav1d style HIGHBD_DECL_SUFFIX to reduce code size and remove duplicate code in asm.

see comments from http://ffmpeg.org/pipermail/ffmpeg-devel/2023-January/305660.html

Make inter and parser process blockless.

Current inter-predict is blockable, it will wait for another frame's pixels to ready. It may have the following issues:

It will capture the executor's thread, and reduce the worker number.
If all threads are blocked, It maybe introduces a deadlock, cause of #17

Unrolling idct2 code

FFmpeg/libavcodec/vvc_itx_1d.c

Line 125 in 3cb136d

static void inv_dct2(int *out, const int size, const ptrdiff_t out_stride,

the idct2 code is not optimzed. Please refer to https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/-/blob/master/source/Lib/CommonLib/TrQuant_EMT.cpp#L164 to optimze it.

non monotonically increasing dts

run ./ffmpeg -i LTRP_A_ERICSSON_3.bit test.yuv -vsync 0
the rawvideo muxer will report:

[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 63
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 64
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 65
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 66
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 67
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 68
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 69
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 70
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 71
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 72
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 73
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 74
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 75
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 76
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 77
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 78
[rawvideo @ 0x55b65f1d6c80] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 79 >= 79
It may be related to some timestamp issues

Support RPR

https://github.com/ffvvc/tests/tree/main/conformance/failed/RPR

Random timeout fail for conformance test.

see
https://github.com/ffvvc/FFmpeg/actions/runs/3922992619
https://github.com/ffvvc/FFmpeg/actions/runs/3923078300
https://github.com/ffvvc/FFmpeg/actions/runs/3923169524
this never happened in my local, may need find a server to reproduce it

Fix #ifdef WIN32 for atomic_compare_exchange_strong

https://github.com/ffvvc/FFmpeg/blob/main/libavcodec/vvc_thread.c#L570

This is a FFMPEG Windows compatibility issue, mainly because https://github.com/ffvvc/FFmpeg/blob/main/compat/atomics/win32/stdatomic.h#L104 does not follow the spec, it accepts automic_int(intptr_t) instead of int.

Support weight predict

https://github.com/ffvvc/tests/tree/main/conformance/failed/WP

Reduce temporary buffers VVCLocalContext

We have many temporary buffers in VVCLocalContext,

FFmpeg/libavcodec/vvc_ctu.h

Lines 279 to 291 in b4e85c1

    
           /* +7 is for subpixel interpolation, *2 for high bit depths */ 
        
           DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2]; 
        
           /* The extended size between the new edge emu buffer is abused by SAO */ 
        
           DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer2)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2]; 
        
           DECLARE_ALIGNED(32, int16_t, tmp)[MAX_PB_SIZE * MAX_PB_SIZE]; 
        
           DECLARE_ALIGNED(32, int16_t, tmp1)[MAX_PB_SIZE * MAX_PB_SIZE]; 
        
           DECLARE_ALIGNED(32, uint8_t, ciip_tmp1)[MAX_PB_SIZE * MAX_PB_SIZE * 2]; 
        
           DECLARE_ALIGNED(32, uint8_t, ciip_tmp2)[MAX_PB_SIZE * MAX_PB_SIZE * 2]; 
        
           DECLARE_ALIGNED(32, uint8_t, sao_buffer)[(MAX_CTU_SIZE + 2 * SAO_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2]; 
        
           DECLARE_ALIGNED(32, uint8_t, alf_buffer_luma)[(MAX_CTU_SIZE + 2 * ALF_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2]; 
        
           DECLARE_ALIGNED(32, uint8_t, alf_buffer_chroma)[(MAX_CTU_SIZE + 2 * ALF_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2];

We do not need most of them since we have different local contexts for inter, recon, deblock, sao and alf.

2~4 of the buffers may be enough for the current code. Please help remove sao_buffer, alf_buffer_luma and other buffers

Add AVX2 assembly code for ALF filter

add AVX2 support for functions for VVCALFDSPContext

function	8 bits	10 bits	12 bits
filter[LUMA]	✅#44	✅#42	⬜
filter[CHROMA]	✅#44	✅#42	⬜
filter_cc	⬜	⬜	⬜
classify	✅#58	✅#58	⬜
recon_coeff_and_clip	⬜	⬜	⬜

In gcc 11.3.1 and higher. I have such message:
I added this fix plus #88

c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:133:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
133 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:142:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
142 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:301:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
301 | DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
| ^~~~~~~~~~~~~~~
In file included from vvcdsp.c:24:
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:133:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
133 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:142:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
142 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:301:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
301 | DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
| ^~~~~~~~~~~~~~~
In gcc 11.3.1 and higher. I have such message:
I added this fix plus #88

c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:133:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
133 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:142:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
142 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:301:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
301 | DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
| ^~~~~~~~~~~~~~~
In file included from vvcdsp.c:24:
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:133:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
133 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:142:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
142 | DECLARE_ALIGNED(4, Mv, mv)[2]; ///< mvL0, vvL1
| ^~~~~~~~~~~~~~~
c:\gcc1131\x86_64-w64-mingw32\include\libavcodec\vvc_ctu.h:301:5: error: expected specifier-qualifier-list before 'DECLARE_ALIGNED'
301 | DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
| ^~~~~~~~~~~~~~~

How to setup:
#91 (comment)

Add Coverity scan

https://scan.coverity.com/


	/* +7 is for subpixel interpolation, 2 for high bit depths /
	DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
	/* The extended size between the new edge emu buffer is abused by SAO */
	DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer2)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
	DECLARE_ALIGNED(32, int16_t, tmp)[MAX_PB_SIZE * MAX_PB_SIZE];
	DECLARE_ALIGNED(32, int16_t, tmp1)[MAX_PB_SIZE * MAX_PB_SIZE];
	DECLARE_ALIGNED(32, uint8_t, ciip_tmp1)[MAX_PB_SIZE * MAX_PB_SIZE * 2];
	DECLARE_ALIGNED(32, uint8_t, ciip_tmp2)[MAX_PB_SIZE * MAX_PB_SIZE * 2];
	DECLARE_ALIGNED(32, uint8_t, sao_buffer)[(MAX_CTU_SIZE + 2 * SAO_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2];
	DECLARE_ALIGNED(32, uint8_t, alf_buffer_luma)[(MAX_CTU_SIZE + 2 * ALF_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2];
	DECLARE_ALIGNED(32, uint8_t, alf_buffer_chroma)[(MAX_CTU_SIZE + 2 * ALF_PADDING_SIZE) * EDGE_EMU_BUFFER_STRIDE * 2];

ffvvc / ffmpeg Goto Github PK

ffmpeg's People

Contributors

Stargazers

Watchers

Forkers

ffmpeg's Issues

Recommend Projects

Recommend Topics

Recommend Org