On Wed, Jan 22, 2020 at 4:23 PM Carl Eugen Hoyos <[email protected]> wrote:
> Am Mi., 22. Jan. 2020 um 16:21 Uhr schrieb Andreas Rheinhardt > <[email protected]>: > > > > On Wed, Jan 22, 2020 at 4:06 PM Carl Eugen Hoyos <[email protected]> > wrote: > > > > > Am Mi., 22. Jan. 2020 um 15:57 Uhr schrieb Andreas Rheinhardt > > > <[email protected]>: > > > > > > > > when directly accessing a buffer via a pointer to uint8_t cast to a > > > > pointer to uint64_t/uint32_t. This happened only if > HAVE_FAST_UNALIGNED > > > > was set, so it was ok, but UBSan nevertheless complained about > unaligned > > > > accesses. So simply use AV_RNxx to read the buffer; this also > improves > > > > readability. > > > > > > Is there a speed impact? > > > > > The assembly produced by both Clang as well as GCC was completely > > unchanged (for non-UBSan builds). > > (on x86?) > > Yes: x86-64. - Andreas _______________________________________________ ffmpeg-devel mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
