Hi, 2015-03-12 16:17 GMT+01:00 Michael Niedermayer <michae...@gmx.at>: > i belive any motion vector that points left outside the picture will > trigger this one
OK I thought it was a magnitude issue, it's actually mvx being negative. That's indeed mighty weird and didn't know about this. > the compiler should optimize the extra operation out, ive > confirmed this before posting the patch in some cases but i didnt > check all Funny that it actually restores the original operation. I really wonder what architecture would have this undefined (excluding overflow situations) >> If we are going to overengineer such issues, we could have something like: >> #if HAVE_FSANITIZED_SHIFT_PLEASURING >> # define LEGAL_SHIFT(a, b, type) ( (a) * (((type)1) << (b) ) >> #else >> # define LEGAL_SHIFT(a, b, type) ( (a) << (b) ) >> #endif > > that could be done it would make the code less readable though Let's not do this, it seems that's what the compiler already does, so... But I'd be curious to know what theorical compilers and/or theorical archictures are affected. -- Christophe _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel