[Ffmpeg-devel-irc] ffmpeg-devel.log.20190519

burek burek021 at gmail.com
Mon May 20 03:05:03 EEST 2019


[00:41:28 CEST] <cone-864> ffmpeg 03Marton Balint 07master:c61d16cf3b81: Revert "lavf/utils: Allow url credentials to contain a slash."
[11:08:52 CEST] <cone-080> ffmpeg 03Michael Niedermayer 07master:fbdb3aa17993: avcodec/v210dec: Fix alignment check for AVX2
[18:55:25 CEST] <Lynne> do all CPUs which support avx support fma3?
[18:56:44 CEST] <JEEB> good question. I heard some CPUs having FMA3 removed (or was it FMA4 if there's such?)
[18:58:17 CEST] <Lynne> fma4 is amd's thing which isn't removed but isn't flagged as supported in their new CPUs last I heard
[18:58:53 CEST] <JEEB> ah
[18:58:57 CEST] <JEEB> that was it then
[18:59:23 CEST] <nevcairiel> AVX1 is definitely older then FMA3
[18:59:31 CEST] <nevcairiel> with AVX2 there may b e some overlap
[19:00:01 CEST] <nevcairiel> ie. Sandy Bridge introduced AVX, Haswell only FMA3
[19:09:00 CEST] <jamrial> all cpus that support avx2 also support fma3, but not the other way around
[19:09:25 CEST] <jamrial> all cpus that support fma3 also support avx, but not the other way around
[19:11:11 CEST] <jamrial> so to answer the question, no. Sandy and Ivy bridge both support avx but not fma3
[20:46:47 CEST] <Gramner> avx2 implies fma3 though
[20:47:14 CEST] <nevcairiel> in practice, or also strictly as a rule?
[20:47:22 CEST] <Gramner> I don'
[20:47:57 CEST] <Gramner> I don't think there's technically any requirements for anything. you could probably make an x86 cpu with sse2 but not sse
[20:48:13 CEST] <nevcairiel> not like  anyone else is going to make a x86 cpu at this point
[20:48:40 CEST] <Gramner> but of course for sanity it makes sense to assume that earlier instruction sets are supported for some cases
[20:48:48 CEST] <Gramner> and every cpu with avx2 also has fma3
[20:49:26 CEST] <durandal_1707> do we have heap implemented somewhere?
[00:00:00 CEST] --- Mon May 20 2019


More information about the Ffmpeg-devel-irc mailing list