On Sat, Jun 3, 2017 at 12:57 AM, Amaury LeLeyzour <amaury.leleyz...@arm.com> wrote: > Hi there, > > > Here's a patch requested by Google in the context of the Chromium project. > See: > > https://bugs.chromium.org/p/chromium/issues/detail?id=686903 > > > The ARMv8 deprecates some IT block usage that this patch tries to address. > I've ran some regression tests (the part that works on a native ARM build), > and it looks fine, but some tests are failing for a vanilla top of the tree > ffmpeg, so it's hard for me to confirm unless you trust me because I have > tested the only relevant part of this patch thoroughly with my own unit > tests. This patch does not add any new regressions. >
Your implementation of FASTDIV looks almost like the original macro we have in the generic code, if there is no need for an optimized assembly, then it should be removed entirely, and fallback to the generic macro. On that note, wouldn't older systems still benefit from having the assembly over the C code, and it should just be disabled conditionally if needed? A similar thing goes for av_clipl_int32_arm, we generally don't like "platform optimized" C code, either we can tune the original C code to be fast, or we write CPU specific assembly. Same remark about older systems applies. - Hendrik _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
