For all the people eager to eat more CPU cycles, I’ve replaced the old and obsolete bSNES filter API with a new one: https://github.com/libretro/RetroArch/pull/649
A sample filter is in the pull request. Note that the API adds easy support for both multithreading and SIMD-awareness (no need to dick around with CPUID checking and stuff, you get a bitmask of all available SIMD instruction sets).