x265 encoding on arm64

[nakTT]

Originally Posted by Blue_MiSfit

Maybe too slow for you, but ARM is rapidly becoming more and more prevalent as ARM chips can occupy an interesting quadrant on the power / speed curve. I imagine with thorough assembly optimization a modern ARM server CPU could outperform a modern x86_64 CPU in terms of efficiency.

If this wasn't the case we probably wouldn't see AWS, Apple, and Microsoft all investing in their own ARM silicon.

Granted, HEVC compression is a very specific use case

I'm sure everyone would agree on the potential that ARM design has. But currently the performance is indeed poor. I believe it is mainly due to the lack of optimization for this architecture.

And of course, i'm speaking as someone who love to see the rise of ARM to give some serious competition in the area that has always been dominated by x86.

[Ritsuka]

HandBrake has got a patch with additional neon optimisations that has not been merged in x265 yet: https://github.com/HandBrake/HandBra...or-arm64.patch

Any help to test it on a non Apple cpu, or help to get it merged in x265 is welcomed.

[benwaggoner]

There have been a ton of ARM optimizations added to x265 over the years. It might not be as optimized as for x64, but there is plenty there.

It's likely more that ARM systems just haven't been intended for the sort of sustained load highly parallel workloads like has been available for x64. Apple's M1 certainly is promising for what ARM could potentially do, but still has a much, much smaller number of cores than available from Intel and AMD.

I'm not sure how AVX? compares with NEON for SIMD, but I know Intel has viewed SW encoding as a material workflow in their designs.