On Wed, Apr 20, 2016 at 12:35 AM, Mark Thompson <[email protected]> wrote: > On 19/04/16 21:36, Luca Barbato wrote: >> - *f->adaptcoeffs = ((res & ((~0UL) << 31)) ^ ((~0UL) << 30)) >> >> >> + *f->adaptcoeffs = ((res & INT32_MIN) ^ (INT32_MIN >> 1)) >> >> (25 + (absres <= f->avg*3) + (absres <= >> f->avg*4/3)); > > This is still relying on implementation-defined behaviour by right-shifting a > negative signed integer (and on int being 32 bits). The compiler can still > do the wrong thing if it wants, but at least it now has to have documented > that it will in advance. >
Thats because for some reason he replaced the second part with INT32_MIN as well, the original commit I linked in the other thread didn't rely on implementation defined behavior anymore. - Hendrik _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
