On Sat, Aug 31, 2024 at 10:40 PM Michael Niedermayer <mich...@niedermayer.cc> wrote: > On Fri, Aug 30, 2024 at 08:56:55PM +0200, Ramiro Polla wrote: > > A55 A76 > > deinterleave_bytes_c: 70342.0 34497.5 > > deinterleave_bytes_neon: 21594.5 ( 3.26x) 5535.2 ( 6.23x) > > deinterleave_bytes_aligned_c: 71340.8 34651.2 > > deinterleave_bytes_aligned_neon: 8616.8 ( 8.28x) 3996.2 ( 8.67x) > > --- > > libswscale/aarch64/rgb2rgb.c | 4 ++ > > libswscale/aarch64/rgb2rgb_neon.S | 59 +++++++++++++++++++++++ > > tests/checkasm/sw_rgb.c | 77 +++++++++++++++++++++++++++++++ > > 3 files changed, 140 insertions(+) > > this breaks fate on x86-64 > > Test checkasm-sw_rgb failed. Look at tests/data/fate/checkasm-sw_rgb.err for > details.
The sse2/avx implementations of deinterleaveBytes use LOOP_NVXX_TO_UV, which checks for alignment on src (and can read unaligned data) but expects dst to be aligned. Should the unaligned versions of these functions be modified to support writing to unaligned data? _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
