[Ffmpeg-devel-irc] ffmpeg-devel.log.20181226

[burek]

[10:47:17 CET] <cone-590> ffmpeg 03Paul B Mahol 07master:530fc345ec73: avfilter/af_sofalizer: add fltp sample format support
[10:47:17 CET] <cone-590> ffmpeg 03Paul B Mahol 07master:8e4e2c9e40b1: avfilter/af_sofalizer: use float constants
[10:47:17 CET] <cone-590> ffmpeg 03Paul B Mahol 07master:60e9007be2e9: avfilter/af_sofalizer: fix typo in comments
[10:47:17 CET] <cone-590> ffmpeg 03Paul B Mahol 07master:92ed9316bba1: avfilter/af_sofalizer: fix regression after 7ea4b928a264
[10:47:17 CET] <cone-590> ffmpeg 03Paul B Mahol 07master:9ce96a744d6f: avfilter/af_headphone: fix regression after 7c201e420
[12:15:48 CET] <cone-590> ffmpeg 03hwrenx 07master:7499d0b0159f: lavc/libdavs2: enable multithread
[12:22:30 CET] <cone-590> ffmpeg 03vkot 07master:3ee735901e08: avformat/hls.c: Properly free prev_segments dynarray after playlist parsing
[12:32:29 CET] <cone-590> ffmpeg 03Adrian Guzowski 07master:c822d70c3d2d: Fix usage of temp_file flag in hls_flags option.
[12:32:30 CET] <cone-590> ffmpeg 03Steven Liu 07master:cef857da481e: avformat/hlsenc: remove unused variable to fix compiler warning
[14:40:30 CET] <jamrial> j-b: https://aomedia.googlesource.com/aom/+/9b1252eab0616d2c1f6d7990c6256441c0b6483f they reverted it
[14:40:56 CET] <j-b> jamrial: wow
[14:41:02 CET] <j-b> I'm a bit surprised.
[14:42:46 CET] <jamrial> it was a pretty bad break, affecting the codec control IDs which are used by everyone. it wasn't like the previous breaks where they removed random unused defines inherited from vpx
[14:43:42 CET] <j-b> jamrial: reminds me that we should make it official that dav1d API will be frozen soon.
[14:44:07 CET] <j-b> There is still a new call needed IIRC and the timestamp discussion, but then it should be frozen
[16:09:34 CET] <jamrial> durandal_1707: wtf was that reply?
[17:21:15 CET] <kierank> michaelni: your 6s timeout is too large
[17:21:21 CET] <kierank> need to make it smaller
[17:35:59 CET] <ramiro> I got a crash on fate-4xm-1 due to a buffer not being aligned enough for avx's clear_blocks().
[17:36:07 CET] <ramiro> I'm too lazy to send a patch, but here's the fix: https://gist.github.com/ramiropolla/86497b62a66e8dc4cbe179db6a7d62b8
[17:36:53 CET] <ramiro> (also in the diff is a similar change for dnxhddec.c, but since it's the first element in the struct it will be correctly aligned anyways)
[17:50:28 CET] <JEEB> ramiro: I guess you don't have git send-email configured? :)
[17:50:55 CET] <JEEB> also I think we had some macro for the alignment
[17:52:28 CET] <JEEB> I remember lately some patch changed some numeric magical value to the alignment requirement value which then differs between AVX2/AVX etc
[17:53:21 CET] <jamrial> the blockdsp api requires 32byte alignment, so that patch is correct
[17:53:42 CET] <JEEB> yes, I am just trying to quickly git grep for the macro
[17:53:55 CET] <JEEB> because I remember there was a single macro that would get bumped by new SIMDs
[17:55:02 CET] <JEEB> I know av_cpu_max_align is a thing in cpu.h, but I did remember something else being around, too
[17:55:30 CET] <jamrial> there's the padding define, but that one is currently 64
[17:55:55 CET] <JEEB> yes, because AVX2
[17:55:58 CET] <JEEB> (or 512?)
[17:56:45 CET] <jamrial> 512 :p
[17:57:13 CET] <jamrial> STRIDE_ALIGN is 16, 32 or 64 depending on configure time options and host
[17:57:21 CET] <JEEB> yup
[17:59:28 CET] <cone-665> ffmpeg 03Ramiro Polla 07master:4d34205bec70: avcodec/4xm: fix block alignment
[17:59:28 CET] <cone-665> ffmpeg 03Ramiro Polla 07master:c268712895f6: avcodec/dnxhddec: fix block alignment
[18:01:12 CET] <JEEB> anyways, my comments mostly stemmed from the thing that I notice every now and then that we have missed alignment requirements and have to bump things in various places. so in a way when such a thing happens my first reaction is not to go for the actual currently required alignment, but instead the current max SIMD alignment requirement (if such exists like STRIDE_ALIGN), since that way every time that 
[18:01:18 CET] <JEEB> gets bumped everything gets bumped as well
[18:01:18 CET] <JEEB> I hope this sort of logic makes sense :P
[18:06:59 CET] <kierank> that was the idea a few years ago but mru didn't like it
[18:07:06 CET] <kierank> forcing arm users to align to 32-bytes
[18:08:14 CET] <ramiro> JEEB: dumping a patch on IRC is faster than git send-email :)
[18:08:21 CET] <ramiro> jamrial: thanks!
[18:08:27 CET] <atomnuker> you know at configure time the arch so you can figure the max padding too
[20:29:21 CET] <cone-665> ffmpeg 03Lauri Kasanen 07master:8dd9df9ecd25: swscale/output: Altivec-optimize float yuv2plane1
[21:59:21 CET] <cone-665> ffmpeg 03Paul B Mahol 07master:8443462e4a56: avfilter/af_afir: remove unused variable
[23:08:04 CET] <cone-665> ffmpeg 03Paul B Mahol 07master:d702600350de: avfilter/af_afir: remove dead store variable
[00:00:00 CET] --- Thu Dec 27 2018