4 ms·
Indeed. Worth noting that Accelerate.framework provides fast and correct bulk f16 <-> f32 conversions as `vImageConvert_Planar16FtoPlanarF` and `vImageConvert_P
by stephencanon 3y ago
Indeed. Worth noting that Accelerate.framework provides fast and correct bulk f16 <-> f32 conversions as `vImageConvert_Planar16FtoPlanarF` and `vImageConvert_PlanarFtoPlanar16F`, and that the arm conversion instructions are unconditionally available for apps that compile for arm64 (they're part of the base ARMv8 ISA), so any _new_ code shouldn't need to worry about this.