2019-02-06 20:17 GMT+01:00, Michael Niedermayer <mich...@niedermayer.cc>: > On Wed, Feb 06, 2019 at 01:57:37PM +0100, Carl Eugen Hoyos wrote: >> 2019-02-03 18:21 GMT+01:00, Michael Niedermayer <mich...@niedermayer.cc>: >> > Fixes: 1377/clusterfuzz-testcase-minimized-5487049807233024 >> > Fixes: assertion failure in sbr_sum_square_c() >> >> This changes the output, no? > > in normal cases, no it should not > It may be that one can construct a case thats not yet in the range > where the old code was undefined, the new code is defined by > 1 in a Quintillion differently and then the decoder output which is > not 64bit integers still has a off by 1 difference. > That could be, or it might be not, i dont know > > You can certainly have a differenceof the old code crashing with > assertion failure though if thst is considered "different" > >> Why are the asm implementations not affected? > > Please correct me if iam wrong but > i think only the float but not the integer version > of this have been optimized.
Thank you for the explanation! Carl Eugen _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel