the bit_matrix implementation seems to use sse2 instructions (code using AVX is unconditionally disabled and labeled as buggy). The AES implementation seems to use sse2 and AES-NI instructions.
During the last meeting, we discussed using simde as an abstraction library. The problem however is, that it doesn't support AES. There is however a library which supports this use case: sse2neon. While it's not as broad as simde, it supports exactly what we need to enable Android support.
Related: #9 Motion currently uses x86 simd instructions in three places:
the bit_matrix implementation seems to use sse2 instructions (code using AVX is unconditionally disabled and labeled as buggy). The AES implementation seems to use sse2 and AES-NI instructions.
During the last meeting, we discussed using simde as an abstraction library. The problem however is, that it doesn't support AES. There is however a library which supports this use case: sse2neon. While it's not as broad as simde, it supports exactly what we need to enable Android support.