ARM neon intrinsic speed difference (in: DSP and Plugin Devel…)

By: Marvinh
20 Aug 06:53

I don’t know if Mac developers noticed this but my neon code was super slow . I have seen some posts online people say don’t bother with intrinsics

Instead of doing the actual functions calls to multiple add or multiply accumulate add

Take your types and just use the overloaded operators

[code]
Float32X4_t a = load( imaging_float_ptr)

Float32X4_t b = load( imaging_float_ptr+4)

float scalar = 0.5f;

a + (b-a)*scalar

[/code]

It makes a world of difference and is faster than compiler…