[FFmpeg-devel] [aarch64] improve performance of ff_yuv2planeX_8_neon

[Clément Bœsch]

On Tue, Dec 10, 2019 at 04:38:25PM -0600, Sebastian Pop wrote:
> Hi,
> 
> This patch rewrites the innermost loop of ff_yuv2planeX_8_neon to avoid zips and
> horizontal adds by using fused multiply adds. The patch also uses ld1r to load
> one element and replicate it across all lanes of the vector. The patch also
> improves the clipping code by removing the shift right instructions and
> performing the shift with the shift-right narrow instructions.
> 
> I see 8% better performance on an m6g instance with neoverse-n1 CPUs:
> $ ffmpeg -nostats -f lavfi -i testsrc2=4k:d=2 -vf
> bench=start,scale=1024x1024,bench=stop -f null -
> before: t:0.014015 avg:0.014096 max:0.015018 min:0.013971
> after:  t:0.012985 avg:0.013013 max:0.013996 min:0.012818
> 
> Tested with `make check` on aarch64-linux.
> 
> Please let me know how I can improve the patch.
> 

Looks nice. I can't test currently but LGTM.

If you're still looking for room of improvement, maybe it makes sense to
increase the filterAlign for NEON and process more at once (seems to be set at
2 currently for the vertical filter):

        const int filterAlign = X86_MMX(cpu_flags)     ? 2 :
                                PPC_ALTIVEC(cpu_flags) ? 8 :
                                have_neon(cpu_flags)   ? 2 : 1;

I can't tell if it would open room for improvement, but might be something to
experiment with eventually.

Thanks for your work, sorry for the delay, and happy new year.

-- 
Clément B.