Re: [FFmpeg-trac] #10856(undetermined:new): Regression: VA-API + QSV transcoding fail, with large number of errors
#10856: Regression: VA-API + QSV transcoding fail, with large number of errors
-+-
Reporter: eero-t |Owner: (none)
Type: defect | Status: new
Priority: important|Component:
| undetermined
Version: git-master | Resolution:
Keywords: regression | Blocked By:
Blocking: | Reproduced by developer: 0
Analyzed by developer: 0|
-+-
Comment (by eero-t):
FYI: During this same time period, I've had also Intel {{{media-driver}}}
update with the following new commits:
{{{
$ git log --format=oneline
eab411768e61a46e096793291f5f992eecf76cd8...a3c3b6ec43ba71aa8d0701df4d052be08adae107
a3c3b6ec43ba71aa8d0701df4d052be08adae107 [Script] Add workflow permissions
b7d3727aedc1279a87077bade5c1ba58137b8751 [Media Common] Bump up version to
24.1.3
0f10db04d6582a6d909f6c641cc6f4d5200eead0 media_libva_decoder: don't return
early
288df99b2da41e48ea928b6f2d8eca2e5b6e408b [Media Common] [VP] HW swizzling
rework on new platforms
878d16175637dc369e68e1ebad57c77a2f273572 [VP] Enable Preemption
55d1e8bceb0f8d3a4b1f128bb938a8506e5c5718 [Encode] Improve effiiciency of
Fast Dump usage
f3e38626762ddb1db62b8ab66d392f885342868a [Decode] Add the surface format
check for jpeg decode in Linux
e33ab705259cb12901c6a930efa17e4eb5ea660e [Decode] refine the MMIO reg
report logical for AV1
f5bc040244e88bd7651dde254ca6f0fb90c42acf [Media Interface] fix mem leak on
g12_lp
}}}
But as MSDK transcoding tool works still fine for the same transcoding
operations as what are done with FFmpeg, that's unlikely to affect this.
--
Ticket URL:
FFmpeg
FFmpeg issue tracker___
FFmpeg-trac mailing list
FFmpeg-trac@avcodec.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-trac
To unsubscribe, visit link above, or email
ffmpeg-trac-requ...@ffmpeg.org with subject "unsubscribe".
Re: [FFmpeg-trac] #10856(undetermined:new): Regression: VA-API + QSV transcoding fail, with large number of errors
#10856: Regression: VA-API + QSV transcoding fail, with large number of errors
-+-
Reporter: eero-t |Owner: (none)
Type: defect | Status: new
Priority: important|Component:
| undetermined
Version: git-master | Resolution:
Keywords: regression | Blocked By:
Blocking: | Reproduced by developer: 0
Analyzed by developer: 0|
-+-
Comment (by eero-t):
Before the problematic commit(s), output looked like this:
{{{
ffmpeg version N-113425-gac06190a5a Copyright (c) 2000-2024 the FFmpeg
developers
built with gcc 11 (Ubuntu 11.4.0-1ubuntu1~22.04)
configuration: --prefix=/opt/install/Nightly_3469 --enable-libmfx
--enable-vaapi --enable-sdl2 --disable-libx265 --disable-libx264
--disable-libvpx --enable-libvorbis --enable-libopus --disable-libmp3lame
--disable-libass --disable-sndio --enable-libfreetype --enable-gpl
--disable-doc
libavutil 58. 36.101 / 58. 36.101
libavcodec 60. 38.100 / 60. 38.100
libavformat60. 20.100 / 60. 20.100
libavdevice60. 4.100 / 60. 4.100
libavfilter 9. 17.100 / 9. 17.100
libswscale 7. 6.100 / 7. 6.100
libswresample 4. 13.100 / 4. 13.100
libpostproc57. 4.100 / 57. 4.100
-vsync is deprecated. Use -fps_mode
[AVHWDeviceContext @ 0x55c971633dc0] libva: VA-API version 1.21.0
[AVHWDeviceContext @ 0x55c971633dc0] libva: User requested driver 'iHD'
[AVHWDeviceContext @ 0x55c971633dc0] libva: Trying to open
/opt/install/Nightly_3469/lib/dri/iHD_drv_video.so
[AVHWDeviceContext @ 0x55c971633dc0] libva: Found init function
__vaDriverInit_1_21
[AVHWDeviceContext @ 0x55c971633dc0] libva: va_openDriver() returns 0
[AVHWDeviceContext @ 0x55c971633dc0] Initialised VAAPI connection: version
1.21
[AVHWDeviceContext @ 0x55c971633dc0] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.2 (eab411768).
[AVHWDeviceContext @ 0x55c971633dc0] Driver not found in known nonstandard
list, using standard behaviour.
[AVHWDeviceContext @ 0x55c9716338c0] Use Intel(R) Media SDK to create MFX
session, API version is 1.35, the required implementation version is 1.3
[AVHWDeviceContext @ 0x55c9716338c0] Initialize MFX session:
implementation version is 1.35
[h264 @ 0x55c97171b9c0] Reinit context to 1280x720, pix_fmt: yuv420p
[h264 @ 0x55c97171a340] max_analyze_duration 50 reached at 52
microseconds st:0
Input #0, h264, from 'input/1280x720p_29.97_10mb_h264_cabac.264':
Duration: N/A, bitrate: N/A
Stream #0:0: Video: h264 (High), 1 reference frame, yuv420p(tv, bt709,
progressive, left), 1280x720 [SAR 1:1 DAR 16:9], 25 fps, 59.94 tbr, 1200k
tbn
[out#0/h264 @ 0x55c97171fc40] No explicit maps, mapping streams
automatically...
[vost#0:0/h264_qsv @ 0x55c9717202c0] Created video stream from input
stream 0:0
[h264_mp4toannexb @ 0x55c9718180c0] The input looks like it is Annex B
already
Stream mapping:
Stream #0:0 -> #0:0 (h264 (h264_qsv) -> h264 (h264_qsv))
[vost#0:0/h264_qsv @ 0x55c9717202c0] Starting thread...
[vf#0:0 @ 0x55c97173ac00] Starting thread...
[vist#0:0/h264 @ 0x55c97171fe80] Starting thread...
[in#0/h264 @ 0x55c97171a180] Starting thread...
Press [q] to stop, [?] for help
[AVHWDeviceContext @ 0x7fda54002840] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.2 (eab411768).
[AVHWDeviceContext @ 0x7fda54002840] Driver not found in known nonstandard
list, using standard behaviour.
[h264_qsv @ 0x55c97173c080] Decoder: output is video memory surface
[h264_qsv @ 0x55c97173c080] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[AVHWDeviceContext @ 0x7fda54007440] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.2 (eab411768).
[AVHWDeviceContext @ 0x7fda54007440] Driver not found in known nonstandard
list, using standard behaviour.
[h264_qsv @ 0x55c97173c080] Decoder: output is video memory surface
[h264_qsv @ 0x55c97173c080] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[graph 0 input from stream 0:0 @ 0x7fda4c003340] w:1280 h:720 pixfmt:qsv
tb:1/120 fr:3/1001 sar:1/1 csp:bt709 range:tv
[h264_qsv @ 0x55c971721900] Using input frames context (format qsv) with
h264_qsv encoder.
[h264_qsv @ 0x55c971721900] Encoder: input is video memory surface
[h264_qsv @ 0x55c971721900] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[h264_qsv @ 0x55c971721900] Using the variable bitrate (VBR) ratecontrol
method
[h264_qsv @ 0x55c971721900] MFMode:2
[h264_qsv @ 0x55c971721900] profile: avc high; level: 32
[h264_qsv @ 0x55c971721900
Re: [FFmpeg-trac] #10856(undetermined:new): Regression: VA-API + QSV transcoding fail, with large number of errors
#10856: Regression: VA-API + QSV transcoding fail, with large number of errors
-+-
Reporter: eero-t |Owner: (none)
Type: defect | Status: new
Priority: important|Component:
| undetermined
Version: git-master | Resolution:
Keywords: regression | Blocked By:
Blocking: | Reproduced by developer: 0
Analyzed by developer: 0|
-+-
Comment (by eero-t):
With QSV, warnings are about allocation failures:
{{{
ffmpeg version N-113614-g4f80441455 Copyright (c) 2000-2024 the FFmpeg
developers
built with gcc 11 (Ubuntu 11.4.0-1ubuntu1~22.04)
configuration: --prefix=/opt/install/Nightly_3486 --enable-libmfx
--enable-vaapi --enable-sdl2 --disable-libx265 --disable-libx264
--disable-libvpx --enable-libvorbis --enable-libopus --disable-libmp3lame
--disable-libass --disable-sndio --enable-libfreetype --enable-gpl
--disable-doc
libavutil 58. 36.101 / 58. 36.101
libavcodec 60. 39.100 / 60. 39.100
libavformat60. 20.100 / 60. 20.100
libavdevice60. 4.100 / 60. 4.100
libavfilter 9. 17.100 / 9. 17.100
libswscale 7. 6.100 / 7. 6.100
libswresample 4. 13.100 / 4. 13.100
libpostproc57. 4.100 / 57. 4.100
-vsync is deprecated. Use -fps_mode
[AVHWDeviceContext @ 0x56221bed3dc0] libva: VA-API version 1.21.0
[AVHWDeviceContext @ 0x56221bed3dc0] libva: User requested driver 'iHD'
[AVHWDeviceContext @ 0x56221bed3dc0] libva: Trying to open
/opt/install/Nightly_3486/lib/dri/iHD_drv_video.so
[AVHWDeviceContext @ 0x56221bed3dc0] libva: Found init function
__vaDriverInit_1_21
[AVHWDeviceContext @ 0x56221bed3dc0] libva: va_openDriver() returns 0
[AVHWDeviceContext @ 0x56221bed3dc0] Initialised VAAPI connection: version
1.21
[AVHWDeviceContext @ 0x56221bed3dc0] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.4 (35b955d7a).
[AVHWDeviceContext @ 0x56221bed3dc0] Driver not found in known nonstandard
list, using standard behaviour.
[AVHWDeviceContext @ 0x56221bed38c0] Use Intel(R) Media SDK to create MFX
session, API version is 1.35, the required implementation version is 1.3
[AVHWDeviceContext @ 0x56221bed38c0] Initialize MFX session:
implementation version is 1.35
[h264 @ 0x56221bfbb980] Reinit context to 1280x720, pix_fmt: yuv420p
[h264 @ 0x56221bfba340] max_analyze_duration 50 reached at 52
microseconds st:0
Input #0, h264, from 'input/1280x720p_29.97_10mb_h264_cabac.264':
Duration: N/A, bitrate: N/A
Stream #0:0: Video: h264 (High), 1 reference frame, yuv420p(tv, bt709,
progressive, left), 1280x720 (0x0) [SAR 1:1 DAR 16:9], 25 fps, 59.94 tbr,
1200k tbn
[out#0/h264 @ 0x56221bfbfc00] No explicit maps, mapping streams
automatically...
[vost#0:0/h264_qsv @ 0x56221bfc0280] Created video stream from input
stream 0:0
[h264_mp4toannexb @ 0x56221bfdac40] The input looks like it is Annex B
already
Stream mapping:
Stream #0:0 -> #0:0 (h264 (h264_qsv) -> h264 (h264_qsv))
[vost#0:0/h264_qsv @ 0x56221bfc0280] Starting thread...
[vf#0:0 @ 0x56221bfd4440] Starting thread...
[vist#0:0/h264 @ 0x56221bfc0cc0] [dec:h264_qsv @ 0x56221bfc2bc0] Starting
thread...
[in#0/h264 @ 0x56221bfba180] Starting thread...
Press [q] to stop, [?] for help
[AVHWDeviceContext @ 0x7f693c002840] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.4 (35b955d7a).
[AVHWDeviceContext @ 0x7f693c002840] Driver not found in known nonstandard
list, using standard behaviour.
[h264_qsv @ 0x56221bfda040] Decoder: output is video memory surface
[h264_qsv @ 0x56221bfda040] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[AVHWDeviceContext @ 0x7f693c007440] VAAPI driver: Intel iHD driver for
Intel(R) Gen Graphics - 24.1.4 (35b955d7a).
[AVHWDeviceContext @ 0x7f693c007440] Driver not found in known nonstandard
list, using standard behaviour.
[h264_qsv @ 0x56221bfda040] Decoder: output is video memory surface
[h264_qsv @ 0x56221bfda040] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[graph 0 input from stream 0:0 @ 0x7f6934003340] w:1280 h:720 pixfmt:qsv
tb:1/120 fr:3/1001 sar:1/1 csp:bt709 range:tv
[graph 0 input from stream 0:0 @ 0x7f6934003340] video frame properties
congruent with link at pts_time: 0
[h264_qsv @ 0x56221bfc1d80] Using input frames context (format qsv) with
h264_qsv encoder.
[h264_qsv @ 0x56221bfc1d80] Encoder: input is video memory surface
[h264_qsv @ 0x56221bfc1d80] Use Intel(R) Media SDK to create MFX session,
the required implementation version is 1.35
[h264_qsv @ 0x56221bfc1d80] Using the variable bitrate (VBR) rate
[FFmpeg-trac] #10856(undetermined:new): Regression: VA-API + QSV transcoding fail, with large number of errors
#10856: Regression: VA-API + QSV transcoding fail, with large number of errors
-+-
Reporter: eero-t | Type: defect
Status: new | Priority: important
Component: | Version: git-
undetermined | master
Keywords: regression | Blocked By:
Blocking: | Reproduced by developer: 0
Analyzed by developer: 0|
-+-
Between following FFmpeg git master commits:
{{{
$ git log --format=oneline
ac06190a5a11f2b170e7719d769d7c0d65bff3e0...887a7817b6671cd5dce9ba9bf97b88afdbbe7137
887a7817b6671cd5dce9ba9bf97b88afdbbe7137 lavc: move bitstream filters into
bsf/ subdir
e0da916b8f5b079a4865eef7f64863f50785463d fftools/ffmpeg: optimize inter-
thread queue sizes
341d0419e1291d5af524405d21d564aaf23dd409 avcodec/texturedsp: Factor common
code out
4f58372c10dfc5d8c4ae46738b5a1c79633ee5ff avcodec/hap: Avoid unnecessary
opt.h inclusion
a3fc9fb9fbc0c7154d874cc6740345bfc0102282 avcodec/texturedsp: Add separate
TextureDSPEncContext
8f791304f2e7bc70d6d6c404301a5c3a1de7da57 avcodec/dxvenc, hap(dec|enc):
Move TextureDSPContext to stack
e1d1304b4b30e442cb0b348ba486003b40740af3 avcodec/texturedspenc: Remove
unused rgtc1_u_alpha encoding func
916f01674178b0949b8d432b0d192169cc98b733 avcodec/dxvenc: Fix data races
with slice threading
555879ca7cf50f265c4beae365cc5142f3eb660e avcodec/dxvenc: Don't cast const
away
d52e8ed114301b44f2a1ff526c6164c43da35749 libavfilter/vf_dnn_detect: Use
class confidence to filt boxes
3de38b9da5c2ffddcf1c532bca78f989b0474494 libavfilter/dnn_interface: use
dims to represent shapes
c695de56b5ba8b2436e455c2284159759ca444d3 libavfilter/dnn_bakcend_openvino:
Add automatic input/output detection
0c517fcbe8ee864c8390dd08696ec63621a4f3e0 lavc/vvc: Fix emulation
prevention byte handling
bab7f91c36ab0386f269d8fd4787dbf5dbbcb01d tests/fate/ffmpeg: add a test for
the issue fixed in previous commit
1ea2e757f7cba714eefd27efb2f9285078bba4a3 fftools/ffmpeg_filter: do not end
filtering when a graph input EOFs
00013341dfd674b72acb8d93fb6b3f3dc7ebc42e fftools/ffmpeg_sched: add filter
API to signal EOF on input
eb4584f994fb9bcfe2a305dada99f8dd5805aae0 avcodec/vvc_ps: remove duplicated
enum
763e31a8d3e3def53fc22828e75df54127e9487e lavc/vvc: Clamp shift RHS
cb7b4ee02491662e0aeee26479e8add955cd9b43 lavc/vvc: Use av_log2 when
destination is integer
8c99a1429a19268e91463e21fdf2f61b8a678135 avcodec/d3d12va_mpeg2|vc1: remove
the unused macros
8b41e9cfbe64c95c04d131632f099910296db48b avcodec/d3d12va_decode: check
existance before assigning a new index
}}}
FFmpeg transcoding broke, at least for following transcoding commands
(which do not use HW scaling):
* H.265=>H.265 (4K): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
Netflix_FoodMarket_4096x2160_10bit_420_100mbs_600.h265 -c:v hevc_vaapi
-b:v 20M -an -vframes 300 -y 0099_4K20.h265}}}
* H.264=>H.264 (HD): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
1920x1080i_29.97_20mb_mpeg2_high.mpv -c:v h264_vaapi -b:v 6000K
-compression_level 7 -an -vframes 2400 -y 0024_HD17i7_1.0.h264}}}
* H.264=>H.264 ("HD"): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
1280x720p_29.97_10mb_h264_cabac.264 -c:v h264_vaapi -b:v 2000K
-compression_level 4 -an -vframes 2400 -y 0027_HD20_1.0.h264}}}
* H.264=>H.264 (SD): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
720x480p_30.00_4mb_h264_cabac_180s.264 -c:v h264_vaapi -b:v 2000K
-compression_level 4 -an -vframes 4800 -y 0038_SD03_1.0.h264}}}
* H.264=>MPEG2 (SD): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
720x480p_30.00_4mb_h264_cabac_180s.264 -c:v mpeg2_vaapi -b:v 2000K
-compression_level 4 -an -vframes 2400 -y 0039_SD03MP2_1.0.mpg}}}
Identical transcoding with HW scale-down added still works OK though, at
least for:
* H.264=>H.264 ("HD"): {{{ffmpeg -loglevel verbose -vsync passthrough
-fpsprobesize 300 -analyzeduration 500K -hwaccel vaapi -vaapi_device
/dev/dri/renderD128 -hwaccel_output_format vaapi -i
1280x720p_29.97_10mb_h264_cabac.264 -c:v h264_vaapi -b:v 800K -vf
scale_vaapi=w=352:h=240,fps=15 -compression_level 4 -an -vframes
