The SSE3 and NEON SIMD implementations of IDCT mirror each other very closely.
It would be great to experiment with using the nightly-only Portable SIMD APIs, and see if there is anything missing for a portable implementation that would support both of those targets. If there are some incompatibilities, it's best to find out now, before the API is stabilized and while we can still affect it.
The SSE3 and NEON SIMD implementations of IDCT mirror each other very closely.
It would be great to experiment with using the nightly-only Portable SIMD APIs, and see if there is anything missing for a portable implementation that would support both of those targets. If there are some incompatibilities, it's best to find out now, before the API is stabilized and while we can still affect it.