[Ffmpeg-devel-irc] ffmpeg-devel.log.20180413

Sat Apr 14 03:05:03 EEST 2018

[01:58:52 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:05a92668942f: avcodec/dnxhddec: Check dc vlc
[01:58:53 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:d81bfc042cea: avcodec/h264_slice: Do not attempt to render into frames already output
[01:58:54 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:e4afafaa7eff: avcodec/jpeg2000dsp: Fix integer overflows in ict_int()
[01:58:55 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:ba2f8469ba95: avcodec/opus_parser: Check payload_len in parse_opus_ts_header()
[01:58:56 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:347744121bc8: avcodec/diracdec: Fix integer overflow with quant
[01:58:57 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:bf7715fd29d5: avcodec/dirac_dwt: Fix overflows in COMPOSE_HAARiH0/COMPOSE_HAARiL0
[01:58:58 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:014ba5c73758: avcodec/h264addpx_template: Fixes integer overflows
[01:58:59 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:d8e501de42d9: avcodec/arm/sbrdsp_neon: Use a free register instead of putting 2 things in one
[01:59:00 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:7965ffbeb111: avcodec/utils: Avoid hardcoding duplicated types in sizeof()
[01:59:01 CEST] <cone-163> ffmpeg 03Carl Eugen Hoyos 07release/3.3:68ca7aefb26c: configure: bump year
[01:59:02 CEST] <cone-163> ffmpeg 03Nikolas Bowe 07release/3.3:980fe1b7a6fc: avformat/matroskadec: Fix float-cast-overflow undefined behavior in matroska_parse_tracks()
[01:59:03 CEST] <cone-163> ffmpeg 03Nikolas Bowe 07release/3.3:68f9a9fe4bcb: avformat/lrcdec: Fix memory leak in lrc_read_header()
[01:59:04 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:eecb0ecf51ed: avcodec/ac3dec_fixed: Fix integer overflow in scale_coefs()
[01:59:05 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:488a6ae7e5e8: avcodec/jpeg2000: Check sum of sizes of band->prec before allocating
[01:59:06 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:a9fb2f8c80d4: avcodec/ulti: Check number of blocks at init
[01:59:07 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:60bc7a6ec08a: avcodec/snowdec: Fix integer overflow before htaps check
[01:59:08 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:162bbc22e203: avcodec/truemotion2: Fix integer overflow in TM2_RECALC_BLOCK()
[01:59:09 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:fb27cebc93cb: avcodec/hevc_cabac: Move prefix check in coeff_abs_level_remaining_decode() down
[01:59:10 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:ce82d4722bde: avcodec/dxtory: Fix bits left checks
[01:59:11 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:dbe356a00974: avcodec/mjpegdec: Fix integer overflow in DC dequantization
[01:59:12 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:aec30d0da9dc: avcodec/hevc_cabac: Check prefix so as to avoid invalid shifts in coeff_abs_level_remaining_decode()
[01:59:13 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:d74839d793eb: avfilter/vf_transpose: Fix used plane count.
[01:59:14 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:61a911d0073b: avcodec/mpeg4videodec: Check mb_num also against 0
[01:59:15 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:093c80747ba4: avcodec/get_bits: Document the return code of get_vlc2()
[01:59:16 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:90cfaff0bbc7: avcodec/mpeg4videodec: Avoid possibly aliasing violating casts
[01:59:17 CEST] <cone-163> ffmpeg 03Aman Gupta 07release/3.3:bb5748ec9d24: avcodec/hevc_ps: extract SPS fields required for hvcC construction
[01:59:18 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:4019c2a67c6a: avcodec/hevc_ps: Check log2_sao_offset_scale_*
[01:59:19 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:38384cdd991c: avcodec/indeo5: Do not leave frame_type set to an invalid value
[01:59:20 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:55f9c21363af: avcodec/dirac_dwt: Fix several integer overflows
[01:59:21 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:d07a0ae1af03: avcodec/exr: Fix memleaks in decode_header()
[01:59:22 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:a97335b1b3ae: avcodec/aacsbr_fixed: Fix overflows in rounding in sbr_hf_assemble()
[01:59:23 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:4d4656e8cd09: avcodec/wavpack: Fix integer overflow in FFABS
[01:59:24 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:0322f781777d: avcodec/utvideodec: Fix bytes left check in decode_frame()
[01:59:25 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:838d02fcff8d: avcodec/huffyuvdec: Check input buffer size
[01:59:26 CEST] <cone-163> ffmpeg 03Nekopanda 07release/3.3:66881cf2b58b: avcodec/mpeg2dec: Fix field selection for skipped macroblocks
[01:59:27 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:55e6c6b5feb7: avcodec/scpr: Fix reading a pixel before the first
[01:59:28 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:65fc03589fb4: avcodec/pafvideo: Check allocated frame size
[01:59:29 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:0df3ad79c7d8: avcodec/vp3: Check eob_run
[01:59:30 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:f03616d2a438: avcodec/mpeg4videodec: Ignore multiple VOL headers
[01:59:31 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:76a886dae364: avcodec/vp3: Error out on invalid num_coeffs in unpack_vlcs()
[01:59:32 CEST] <cone-163> ffmpeg 03Xiaohan Wang 07release/3.3:e8fb74c0c9cd: avcodec/h264_cavlc: Set valid qscale value in ff_h264_decode_mb_cavlc()
[01:59:33 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:878fc42a906d: avcodec/h264_cabac: Tighten allowed coeff_abs range
[01:59:34 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:519d48c861a9: avutil/common: Fix integer overflow in av_clip_uint8_c() and av_clip_uint16_c()
[01:59:35 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:659a23e89fba: avcodec/exr: Check remaining bits in last get code loop
[01:59:36 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:2a85ead5a374: avcodec/vp8: Check for bitstream end before vp7_fade_frame()
[01:59:37 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:1a4f8de03d57: avcodec/jpeg2000dec: Use av_image_check_size2()
[01:59:38 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:37cd7f337530: avcodec/hevcdec: Check luma/chroma_log2_weight_denom
[01:59:39 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:93a16aebf2ac: avcodec/dirac_dwt_template: Fix Integer overflow in horizontal_compose_dd137i()
[01:59:40 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:af5c12c0290b: avcodec/dxtory: Remove code that corrupts dimensions
[01:59:41 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:205689ae8a5e: avcodec/diracdec: Use int64 in global mv to prevent overflow
[01:59:42 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:1a387f1ce6ec: avcodec/jpeg2000dwt: Fix integer overflows in sr_1d53()
[01:59:43 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:b172815c3c9b: avcodec/aacdec_templat: Fix integer overflow in apply_ltp()
[01:59:44 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:4018d8586ffe: avcodec/h264_parse: Clear invalid chroma weights in ff_h264_pred_weight_table()
[01:59:45 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:b9d5b1f05dad: avcodec/diracdec: Fix integer overflow in mv computation
[01:59:46 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:f3562ee6fc84: avcodec/cavsdec: Check alpha/beta offset
[01:59:47 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:bcc6d409283a: avcodec/smc: Check input packet size
[01:59:48 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:bafb13dc0fd6: avcodec/utvideodec: Check subsample factors
[01:59:49 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:5909508e8df3: avcodec/bintext: sanity check dimensions
[01:59:50 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:4d45d5b606a1: avcodec/exr: fix invalid shift in unpack_14()
[01:59:51 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:c3e774784bac: avcodec/g2meet: Check tile dimensions with av_image_check_size2()
[01:59:52 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:e908a595db8e: avcodec/truemotion2rt: Check input buffer size
[01:59:53 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:6daa205cd3fe: libavformat/oggparsevorbis: Fix memleak on multiple headers
[01:59:54 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:fef832c188c0: avformat/oggdec: Fix metadata memleak on multiple headers
[01:59:55 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:e89d8ed7cb3f: avformat/utils: Fix integer overflow of fps_first/last_dts
[01:59:56 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:42bd425205f4: avformat/oggparsetheora: Do not adjust AV_NOPTS_VALUE
[01:59:57 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:8cf7205a72f5: avformat/avidec: Fix integer overflow in cum_len check
[01:59:58 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:e0a08c833dd8: avformat/oggparseogm: Fix undefined shift in ogm_packet()
[01:59:59 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:e61cdbe27167: avformat/oggparseogm: Check lb against psize
[02:00:00 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:3a29fda42ab7: avformat/oggparsedaala: Do not adjust AV_NOPTS_VALUE
[02:00:01 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:edfe9ae63b19: avformat/mov: Fix integer overflows related to sample_duration
[02:00:02 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:02bf1c617cdd: avcodec/wmalosslessdec: Reset num_saved_bits on error path
[02:00:03 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:719b9b673cfb: avcodec/nuv: Check for minimum input size for uncomprssed and rtjpeg
[02:00:04 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:5b586f0bc87d: avcodec/nuv: rtjpeg with dimensions less than 16 would result in no decoded pixels thus reject it
[02:00:05 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:a89b45b492dd: avformat/mov: Check STSC and remove invalid entries
[02:00:06 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:899d40c17fef: avcodec/get_bits: Make sure the input bitstream with padding can be addressed
[02:00:07 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:8a89cce37288: avformat/mov: Move +1 in check to avoid hypothetical overflow in add_ctts_entry()
[02:00:08 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:769cb8973817: avcodec/tableprint_vlc: Fix build failure with --enable-hardcoded-tables
[02:00:09 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:d79b274acc3a: avcodec/wmalosslessdec: Fix null pointer dereference in decode_frame()
[02:00:10 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:15d4dc0da1e9: avcodec/utvideodec: Set pro flag based on fourcc
[02:00:11 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:49336482fd04: avfilter/vf_signature: use av_strlcpy()
[02:00:12 CEST] <cone-163> ffmpeg 03heimdallr 07release/3.3:3f949b7a6461: avcodec/imgconvert: Fix loss mask bug in avcodec_find_best_pix_fmt_of_list()
[02:00:13 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:1d3c141af4f0: avcodec/cscd: Error out when LZ* decompression fails
[02:00:14 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:ae2c159b87f2: avcodec/aacdec_fixed: Fix integer overflow in apply_independent_coupling_fixed()
[02:00:15 CEST] <cone-163> ffmpeg 03Hendrik Schreiber 07release/3.3:30c26bb78f6a: swresample/swresample: Fix for seg fault in swr_convert_internal() -> sum2_float during dithering.
[02:00:16 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:23f13130f12a: libavcodec/rv34: error out earlier on missing references
[02:00:17 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:0e5946d5b247: avcodec/aacsbr_fixed: Fix integer overflow in sbr_hf_assemble()
[02:00:18 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:c9452899db10: avcodec/movtextdec: Check style_start/end
[02:00:19 CEST] <cone-163> ffmpeg 03Matt Wolenetz 07release/3.3:9b1a4da66899: lavc/libopusdec: Allow avcodec_open2 to call .close
[02:00:20 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:9b56e65e8c12: avformat/mov: Fix extradata memleak
[02:00:21 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:41458534aa88: avcodec/h264_slice: Fix integer overflow with last_poc
[02:00:22 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:6de499caed33: avcodec/mjpegdec: Check input buffer size.
[02:00:23 CEST] <cone-163> ffmpeg 03James Almer 07release/3.3:3e3704da0c71: avformat/utils: fix mixed declarations and code
[02:00:24 CEST] <cone-163> ffmpeg 03Michael Niedermayer 07release/3.3:7af6fba1455a: Update for 3.3.7
[03:03:04 CEST] <Chloe> wm4: rewriting libswscale huh
[03:03:22 CEST] <Chloe> That sounds interesting/fun
[03:04:05 CEST] <wm4> and extremely hard
[03:05:26 CEST] <Chloe> There was that other library i wanted to move into the only filter which uses it as well
[08:42:22 CEST] <cone-100> ffmpeg 03Paul B Mahol 07master:9e406326687b: avfilter: add vfrdet filter
[09:21:43 CEST] <cone-100> ffmpeg 03Timo Rothenpieler 07master:9e857aa00e10: avcodec/nvdec_hevc: add support for new extended sps/pps flags from SDK 8.1
[09:40:48 CEST] <cone-100> ffmpeg 03Steven Liu 07master:987026adda79: avformat/hlsenc: add option hls_delete_threshold
[10:53:15 CEST] <BtbN> Can someone take a look at the logic in my nvenc/dec version check macros? https://github.com/BtbN/FFmpeg/commit/cc763532613e84143872c9f5dff159cbd2b40383.patch
[11:15:51 CEST] <nevcairiel> BtbN: i prefer brackets when mixing OR and AND statements, but otherwise it looks fine to me
[11:16:29 CEST] <BtbN> You mean extra brackets around the && part?
[11:16:54 CEST] <nevcairiel> yeah
[11:17:05 CEST] <BtbN> Yeah, thought about that as well. I'll add them
[11:23:22 CEST] <cone-100> ffmpeg 03Timo Rothenpieler 07master:9f3902f10734: avcodec/nv{enc,dec}: use sane version checking macro
[11:23:52 CEST] <BtbN> I wonder if this QP map mode they added with the new SDK is of any value to ffmpeg
[11:24:02 CEST] <BtbN> seems rather hard to expose it
[11:28:21 CEST] <cone-100> ffmpeg 03Hendrik Leppkes 07master:b4159d959394: avcodec/dxva2: add VP9 10-bit Profile2 mode mappings
[11:28:22 CEST] <cone-100> ffmpeg 03Hendrik Leppkes 07master:7e20c8f5ebe9: avcodec/vp9: add profile 2 10-bit DXVA2/D3D11 decoding support
[11:29:02 CEST] <BtbN> I also wonder if there'd be any downside to enabling bframe references by default if they're supported
[11:29:42 CEST] <nevcairiel> should probably see how it impacts quality
[11:29:57 CEST] <BtbN> It should improve it, shouldn't it?
[11:31:22 CEST] <nevcairiel> trust, but verify :)
[11:31:48 CEST] <BtbN> vmav was the visual quality detection thing, right?
[11:31:56 CEST] <BtbN> or vmaf, rather
[11:32:03 CEST] <nevcairiel> yea the new metric from netflix
[11:33:00 CEST] <BtbN> hm, just remembered that the PC im in front of right now is Intel...
[12:09:29 CEST] <kierank> wm4: do you run mpv twitter?
[12:13:45 CEST] <durandal_1707> no, some other folk
[12:40:01 CEST] <durandal_1707> what happened to api for libsws which would use AVFrames directly/for configuration
[12:52:56 CEST] <Chloe> Seriously though, what is ffmpeg 4.0 waiting on?
[12:53:21 CEST] <Chloe> Or whats blocking it?*
[12:53:39 CEST] <durandal_1707> jamrial and michaelni know more, there is topic on ml
[12:53:41 CEST] <nevcairiel> there is a thread on the ML discussing that
[12:54:27 CEST] <robUx4> avcodec/vp9: add profile 2 10-bit DXVA2/D3D11 decoding support < cool !
[12:55:31 CEST] <Chloe> nevcairiel: ah right. Will check it
[12:55:35 CEST] <robUx4> nevcairiel: no versioning for this addition ?
[12:58:20 CEST] <nevcairiel> hwaccel stuff works mostly transparently, i see no huge reason to do that
[12:58:51 CEST] <nevcairiel> even if you use get_format yourself, either you get called with a hwaccel format or you don't, not sure why you would need to know beforehand
[13:05:40 CEST] <Chloe> Mmh theres an email from jamrial at 4 am this morning which refuses to display
[13:05:51 CEST] <Chloe> (In the ffmpeg 4.0 thread)
[13:25:52 CEST] <durandal_1707> Chloe: what? you are no longer subscribed to ml?
[13:27:06 CEST] <Chloe> durandal_1707: i am. My mail client is broken i think (nothing new)
[13:36:53 CEST] <cone-100> ffmpeg 03Paul B Mahol 07master:4078291d32c2: avcodec/magicyuvenc: mark as experimental
[16:09:20 CEST] <cone-100> ffmpeg 03James Almer 07master:e7a2f1f8d835: configure: disable direct stripping in OpenBSD
[16:09:21 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:876f9ac9a682: configure: make the C11 atomics check more robust
[19:07:39 CEST] <BtbN> hm, is the frame->key_frame logic for h264 flawed, or am I misunderstanding the flag? It's not set for all frames for which ff_h264_get_slice_type() returns an I Frame
[19:08:22 CEST] <nevcairiel> key frame is more flexible then that
[19:09:14 CEST] <nevcairiel> basically, it marks clean cut points
[19:09:28 CEST] <BtbN> Nvidit just sent me a patch that basically does h->cur_pic_ptr->f->key_frame = (2 == ff_h264_get_slice_type(h->slice_ctx));
[19:09:30 CEST] <nevcairiel> IDR frames, I frames with recovery, e tc
[19:09:31 CEST] <BtbN> *Nvidia
[19:09:44 CEST] <BtbN> Because it confuses their decoder
[19:10:02 CEST] <nevcairiel> no changing h264 logic for confused hardware decoders
[19:10:03 CEST] <nevcairiel> :)
[19:10:43 CEST] <nevcairiel> I stands for "Intra" afterall, not "Key"
[19:11:07 CEST] <BtbN> Yeah, so "ctx->pic_params.intra_pic_flag = frame->key_frame;" is just wrong
[19:12:44 CEST] <nevcairiel> yeah if you want intra you should check the slice type, not the key_frame type
[19:12:54 CEST] <BtbN> That code is in generic nvdec.c
[19:13:50 CEST] <BtbN> Can't it just be frame->pict_type == AV_PICTURE_TYPE_I?
[19:14:00 CEST] <nevcairiel> but why not set that field in the codec specific things
[19:14:01 CEST] <BtbN> Or is that not always accurate?
[19:14:08 CEST] <nevcairiel> h264 already has that field and sets it to zero
[19:14:14 CEST] <nevcairiel> just fill in a condition and done
[19:14:29 CEST] <BtbN> It would mean moving generic logic to every single individual codec
[19:15:02 CEST] <nevcairiel> assuming that you can do this generically
[19:15:40 CEST] <nevcairiel> IntraPicFlag in dxva2 for example acts on a slice level
[19:15:48 CEST] <nevcairiel> frame->pict_type may not
[19:16:07 CEST] <BtbN> cuvid wants it per frame
[19:16:23 CEST] <nevcairiel> its sned per-frame as well, but you need to make sure there is only intra slices in there
[19:16:38 CEST] <nevcairiel> not sure if frame->pict_type is accurate like that
[19:16:51 CEST] <nevcairiel> safer to just check the slices, since you handle those anyway
[19:17:22 CEST] <BtbN> https://bpaste.net/show/7d3f4159254c this is weird
[19:17:31 CEST] <BtbN> why do h264 and hevc just set it to 0?
[19:17:48 CEST] <Rathann> hi guys, is there any benefit in building FFmpeg with -march=haswell instead of a generic x86_64 build?
[19:18:25 CEST] <Rathann> I'm considering if it's worth it to take advantage of https://clearlinux.org/blogs/transparent-use-library-packages-optimized-intel-architecture
[19:18:32 CEST] <BtbN> nevcairiel, how would I check the slices? The nvdec only has per-frame parameters.
[19:19:04 CEST] <nevcairiel> do what dxva2 does, set it to 1 initially and then on each slice check the slice type and unset it if required
[19:19:16 CEST] <BtbN> so iterate over slice_ctx?
[19:19:23 CEST] <nevcairiel> no, in the slice callback
[19:19:25 CEST] <nevcairiel> see dxva2_h264_decode_slice
[19:19:31 CEST] <nevcairiel> adapt for nvdec_h264_decode_slice
[19:19:51 CEST] <wm4> Rathann: generally not since most "hot" code paths have asm written for them anyway
[19:20:02 CEST] <BtbN> Is slice_ctx[0] always a new slice in there?
[19:20:10 CEST] <nevcairiel> its always the current slice
[19:20:29 CEST] <nevcairiel> iirc there is only more of those for slice threading
[19:20:33 CEST] <nevcairiel> which hardware doesnt use
[19:20:43 CEST] <Rathann> wm4: ok, thanks for the advice
[19:20:59 CEST] <jamrial> Rathann: only in a bunch of modules like sha/md5 afaik, since for those the compiler can use bmi2 instructions
[19:21:27 CEST] <nevcairiel> it would be easy enough to benchmark your designed use-case anyway, but the impact is likely to be minimal
[19:21:34 CEST] <jamrial> but aside from that, modern gcc already optimizes code for modern intel cpus when using generic
[19:22:09 CEST] <BtbN> nevcairiel, how does that even work? Is it re-alloced and moved up each time?
[19:22:19 CEST] <nevcairiel> its decoded and cleared
[19:22:23 CEST] <nevcairiel> slices are processed serially
[19:22:37 CEST] <BtbN> So there is never more than one slice in there, despite nb_slices being a thing?
[19:22:44 CEST] <jamrial> we disable tree vectorize, so the fact -march=haswell enables compiler generated avx instructions doesn't really have much of an effect
[19:22:45 CEST] <nevcairiel> unless slice threading
[19:23:52 CEST] <Rathann> jamrial: thank you
[19:25:20 CEST] <nevcairiel> BtbN: specifically, there is nb_slice_ctx to control this, which is always 1 for hwaccel use (and only not 1 for slice threading cases)
[19:26:34 CEST] <BtbN> I wonder if nvidia even tested their patch
[19:26:49 CEST] <nevcairiel> why do they send it to you and not the ML anyway
[19:26:54 CEST] <nevcairiel> i want some frowning too
[19:27:01 CEST] <BtbN> I asked them the same, they said thanks for noticing
[19:27:07 CEST] <BtbN> Not sure what to make of it
[19:27:09 CEST] <nevcairiel> lol
[19:27:11 CEST] <JEEB> lol
[19:27:48 CEST] <BtbN> I guess future patches will be sent to ffmpeg-devel though, at least I hope so
[19:27:59 CEST] <nevcairiel> they did that in the past at least to s ome degree
[19:28:02 CEST] <BtbN> the changes they made are essentially a no-op, since it's overwritten like 5 lines later
[19:28:12 CEST] <BtbN> yeah, it's a new guy sending the patches
[19:28:22 CEST] <BtbN> Pretty sure it's just an oversight on his side
[19:28:41 CEST] <BtbN> and english is not his native language, so "thanks for noticing" might be meant differently than it sounds
[19:29:29 CEST] <BtbN> https://bpaste.net/show/273968f55be3 that's the patch they sent me
[19:29:52 CEST] <BtbN> it seems confused in more than one way. -force_key_frames for a decoder?
[19:30:02 CEST] <wm4> wow yoda conditions
[19:30:03 CEST] <nevcairiel> thats one huge mess
[19:30:40 CEST] <BtbN> 5 lines after ff_nvdec_start_frame the intra_pic_flag it sets is overwritten as well
[19:30:49 CEST] <BtbN> I guess the whole patch is about altering the key_frame logic?
[19:31:11 CEST] <nevcairiel> i dont think intra_pic_flag is particularly required for the nvdec decoder anyway, its usually only used as an optimization for the decoder, not a required property
[19:31:39 CEST] <BtbN> """-force_key_frames "source" option forces I frames in encoded sequence to be at same positions as in decoded sequence.""" was the mail header
[19:31:41 CEST] <nevcairiel> and actual key frames should already be marked as such by h264dec, so it seems all a bit weird
[19:31:57 CEST] <BtbN> They are unhappy about not all I frames being marked I guess
[19:32:04 CEST] <BtbN> but only IDR
[19:32:06 CEST] <nevcairiel> yeah well cant fix that
[19:32:31 CEST] <nevcairiel> key frames are more then intra frames in the more complex codecs
[19:32:46 CEST] <nevcairiel> and a hwaccel has no business changing such flags
[19:33:45 CEST] <BtbN> does HEVC not have similar logic?
[19:34:11 CEST] <BtbN> dxva2 just seems to be using IS_IRAP(h)
[19:35:01 CEST] <nevcairiel> like I said, IntraPicFlag is usually not a mandatory field in the spec, so I might have cheaped out on that
[19:36:14 CEST] <BtbN> well, for HEVC it's a global HEVCContext flag
[19:38:57 CEST] <BtbN> vp8/vp9 nvdec doesn't bother setting it at all
[19:41:45 CEST] <cone-100> ffmpeg 03Gyan Doshi 07master:6b28e551d009: MAINTAINERS: add myself to documentation section
[19:43:20 CEST] <BtbN> lets see what they have to say about that version
[21:44:54 CEST] <klaxa> ok, so i want to replace my malloc()s and realloc()s with the av_* equivalents, now the docs for av_realloc() say: >Warning    Unlike av_malloc(), the returned pointer is not guaranteed to be correctly aligned. 
[21:46:02 CEST] <cone-100> ffmpeg 03Bela Bodecs 07master:3e1204b94d1a: avformat/utils: Stream specifier enhancement 2.
[21:46:03 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:5c62a3dbda15: tests/fate/filter-video: Use a bitexact/lossless input sample for fate-filter-metadata-silencedetect
[21:46:25 CEST] <klaxa> my wifi died just as i sent this line... so possibly lost
[21:46:28 CEST] <klaxa> ok, so i want to replace my malloc()s and realloc()s with the av_* equivalents, now the docs for av_realloc() say: >Warning    Unlike av_malloc(), the returned pointer is not guaranteed to be correctly aligned.
[21:46:52 CEST] <klaxa> am i correct to assume this has only performance implications and i can use it just like realloc()?
[21:48:43 CEST] <klaxa> also thinking about replacing my heap-allocated array by avutil's dynamic array, seems to fill the need exactly
[21:49:18 CEST] <nevcairiel> some buffers need alignment for SIMD
[21:49:21 CEST] <wm4> using av_malloc for anything other than buffers that SIMD code runs on is pretty fringe anyway
[21:49:25 CEST] <nevcairiel> if they dont have it, shit may crash
[21:49:43 CEST] <wm4> so if you ask me we should prefer malloc() whenever possible
[21:50:18 CEST] <klaxa> ok, michaelni said on the ML mixing av_free() and free() may be not good
[21:50:30 CEST] <klaxa> so should i try to change everything to malloc then?
[21:50:41 CEST] <nevcairiel> just dont accidentlaly give anything to ffmpeg that it might require to be a llocated by av_malloc
[21:50:42 CEST] <klaxa> but i think some libav* things want av_free()
[21:50:56 CEST] <nevcairiel> ie. things it frees or reallocs
[21:51:12 CEST] <wm4> you (probably) can't call av_free on a malloc() pointer and vice versa and the only reason we cargo cult av_malloc everywhere is due to some vague consistency arguments
[21:51:23 CEST] <michaelni> its much safer to use av_*alloc() & av_free* than mixing malloc and av_malloc
[21:51:27 CEST] <wm4> (even if av_realloc destroys alignment...)
[21:52:30 CEST] <michaelni> and you cannot just use malloc() as av_malloc()/av_free() is needed by the API at a few points
[21:53:35 CEST] <nevcairiel> i prefer using av_malloc in the entire module that access av* libraries so such problems dont happen
[21:53:45 CEST] <nevcairiel> but up to everyone to use  their own style
[21:54:11 CEST] <klaxa> well it certainly feels cleaner to only use one function for the same task
[21:54:27 CEST] <klaxa> and it avoids confusion probably
[21:58:02 CEST] <jamrial> wm4: no, as that could easily generate crashes. using av_free for a buffer allocated with malloc may end up badly, like on windows where av_free -> _aligned_free
[21:59:17 CEST] <jamrial> also, the allocation limit for all av_*alloc functions can be customized with av_max_alloc(), so it's not just a #define av_malloc malloc kind of situation
[22:01:17 CEST] <wm4> av_max_alloc is global state and shouldn't exist
[22:01:22 CEST] <wm4> it's idiotic
[22:35:13 CEST] <klaxa> huh, kinda shot myself in the foot with my segment_save() function :P i think i only used it for debugging, but it does seem useful after all
[22:35:43 CEST] <klaxa> well it should support saving in any format or protocol supported by ffmpeg then, guess that means it needs its own muxing as well
[22:56:46 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:c48ceff786bd: avformat/utils: Fix integer overflow in end time calculation in update_stream_timings()
[22:56:47 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:f6304af2341d: avcodec/truemotion2: Propagate out of bounds error from GET_TOK()
[22:56:48 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:544324827e01: avcodec/utils: Enforce minimum width also for VP5/6
[22:56:49 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07master:37d46dc21d70: avformat/utils: Check cur_dts in update_initial_timestamps() more
[23:32:56 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07release/3.3:01645d591567: avformat/utils: Fix integer overflow in end time calculation in update_stream_timings()
[23:32:57 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07release/3.3:d7bda29820f0: avcodec/truemotion2: Propagate out of bounds error from GET_TOK()
[23:32:58 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07release/3.3:b8ef20becf92: avcodec/utils: Enforce minimum width also for VP5/6
[23:32:59 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07release/3.3:d4b5ffeb57f2: avformat/utils: Check cur_dts in update_initial_timestamps() more
[23:33:00 CEST] <cone-100> ffmpeg 03Michael Niedermayer 07release/3.3:93e2cb451141: Changelog: update
[23:59:18 CEST] <jamrial> wbs: can you look at "[PATCH] configure: fix clang-cl detection" ?
[23:59:48 CEST] <jamrial> i'm surprised people are saying it's needed, considering you've been working on clang cl stuff lately
[00:00:00 CEST] --- Sat Apr 14 2018