Regarding upload CGImage pixel data to MTLTexture, i also tried vImage acclerated API, the performance bottleneck still lies on vConvert_PermuteChannels_ARGB8888_CV_vec, considering the function name start with "v", i guess it's already SIMD like instruction set accelerated. and channel permute operation looks like can't be avoided.
Topic:
Graphics & Games
SubTopic:
General
Tags: