It would be quite helpful if there was frame interpolation in MetalFX, similar to what NV and AMD are adding to their upscaling technologies. I have found MetalFX temporal upscaling critical to a real-time ray traced application, and frame interpolation could halve the compute cost, letting me try more expensive RT algorithms. I am hoping this technology will debut at WWDC 2023, if not then maybe during the winter round of OS updates.

I'm trying to optimize code for the M1 GPU family (Apple 7 variants where F32 compute power equals F16 compute power). xcrun metal-opt mentions a compile option called -fma-shff-hoist-depth-agx2=<uint>. There's something called fma.shff. I'd like to know what the "SHFF" acronym stands for. Shuffle? Shift? Perhaps it's related to SIMD-group reductions.

I am a Metal developer who frequently uses the Metal feature set tables. In 2020 just after the iPhone 12 event, the feature set tables were updated immediately to reflect the new A14 GPU. Although Apple released a video detailing the A15's new features and updated the MSL specification, they haven't added A15 to the feature set tables. That would we really helpful incase there were any small additions from A14 -> A15 that weren't documented in the video. Also, the M1 is in the Apple7 family, but it isn't on the feature set tables because they were last updated before the M1 event. Is Apple waiting until the M2 event in April to add Apple8 to the feature set tables?