[Ffmpeg-devel-irc] ffmpeg-devel.log.20190216

[burek]

[17:52:41 CET] <philipl> BtbN: I think your cuda_check rework breaks any cuda_check users that directly link cuda.
[17:53:14 CET] <philipl> The vf_scale_cuda won't compile for me here, because the inline function references tcuGetErrorName etc and those are only defined in the dynamic loader header and that cannot be included at the same time as cuda.h
[17:54:39 CET] <philipl> It probably requires a completely separate impl, which i guess is fine.
[17:58:13 CET] <jamrial> can't you make the function avpriv and share it between libs?
[18:01:55 CET] <philipl> jamrial: that is where we started...
[18:02:22 CET] <jamrial> you were using ff_ and #including the c file on every library
[18:02:43 CET] <nevcairiel> please avoid avpriv
[18:02:54 CET] <nevcairiel> what exactly was the problem with the include way anyway?
[18:02:58 CET] <nevcairiel> that works fine for many other functions
[18:24:26 CET] <philipl> I don't know what the case was that failed. The report was symbol duplication. Could that happen if someone statically links all libav* into a program?
[18:53:51 CET] <philipl> BtbN: https://github.com/philipl/FFmpeg/commit/f028d813709cb14bdf08de898bc560e41e630763
[19:26:09 CET] <cone-203> ffmpeg 03Philip Langdale 07master:f4ea930a1192: avcodec/hevc_ps: Expose all SPS and PPS range extension flags
[19:26:10 CET] <cone-203> ffmpeg 03Philip Langdale 07master:e06ccfbe1d33: avcodec/nvdec: Add support for decoding HEVC 4:4:4 content
[19:26:11 CET] <cone-203> ffmpeg 03Philip Langdale 07master:83c7ac2e47ef: avcodec/nvdec: Explicitly mark codecs that support 444 output formats
[19:26:12 CET] <cone-203> ffmpeg 03Philip Langdale 07master:317b7b06fd97: avcodec/cuviddec: Add support for decoding HEVC 4:4:4 content
[19:44:35 CET] <cone-203> ffmpeg 03Philip Langdale 07master:d6fc5dc24aa0: avcodec/version: Bump micro-version for nvdec/cuviddec changes
[20:03:30 CET] <nevcairiel> the way this stuff usually works is that in static builds, only one of the .c files is actually build, and all static libraries reference the same function (ie. from libavutil or whichever is the "primary" one), which works since static has no problems referenceing ff* symbols from another archive
[20:03:38 CET] <nevcairiel> and the duplicate .c f iles are only build with dynamic builds
[20:03:55 CET] <nevcairiel> if this is not how it was done, then it was done wrong in the first place :)
[20:04:29 CET] <nevcairiel> if you in fact have the same ff symbol in every .a file, there is a good chance it'll collide in some toolchain
[20:05:01 CET] <philipl> That is very likely what happened.
[20:05:33 CET] <philipl> There was no conditional logic for static vs dynamic.
[20:05:39 CET] <nevcairiel> sounds to me like someone suggested thats how we solve it, but then not enough research was done  how exactly its done
[20:07:32 CET] <philipl> Yep.
[20:07:51 CET] <nevcairiel> http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavformat/Makefile;h=c010fc83f977ac46c189466fd5ce2b1d2add0caf;hb=HEAD#l64 for example
[20:07:59 CET] <nevcairiel> log2_tab.c is only included in shared builds
[20:08:30 CET] <philipl> Yes, I was just looking.
[20:08:59 CET] <nevcairiel> unfortunately makefile logic is a bit limited so having two conditions to include a fi le (ie. cuda and shared) gets a bit annoying
[20:09:02 CET] <philipl> To go back and do this correctly, it would need to be conditioned on both SHARED and CUDA
[20:09:05 CET] <philipl> yeah
[20:09:22 CET] <philipl> I guess we could add a specific condition in configure for it
[20:30:43 CET] <cone-203> ffmpeg 03Michael Niedermayer 07master:1f686d023b95: avcodec/mpeg4videodec: Clear interlaced_dct for studio profile
[20:30:44 CET] <cone-203> ffmpeg 03gxw 07master:1466dc140d49: avcodec/mips: [loongson] optimize theora decoding with mmi.
[00:00:00 CET] --- Sun Feb 17 2019