While working on #233 I realised that the native build of the library can only default to one VPU implementation (i.e. can either emulate xs3a VPU or vx4b VPU). Right now it falls back to using xs3a one with no way to switch to vx4b. I guess if we really want to use the library so model the VPU precision natively, we'd have to support both architectures building natively.
While working on #233 I realised that the native build of the library can only default to one VPU implementation (i.e. can either emulate xs3a VPU or vx4b VPU). Right now it falls back to using xs3a one with no way to switch to vx4b. I guess if we really want to use the library so model the VPU precision natively, we'd have to support both architectures building natively.