Open ambrad opened 6 years ago
Start with divergence_sphere_update, which will give the clearest signal. If that works, use it in the HV kernels, too.
VR (especially compute_remap) is also memory bw bound, so it should also be investigated if divergence_sphere_update goes well
compute_remap
divergence_sphere_update
Start with divergence_sphere_update, which will give the clearest signal. If that works, use it in the HV kernels, too.