Open cgarling opened 6 months ago
Wrote an alternate loop for gauss2d_integral_halfpix
in commit 0c77e2dc599b20a5574c65552b7a76822c946aca timing went from 26 -> 36 ns, significant difference. Going to stay on 1.10 with LoopVectorization for a bit, revisit at later time. This was just an example though, as I believe only GaussianPSFAsymmetric
is currently used in bin_cmd_smooth
.
We get 2x speedup from LV in the kernel evaluations loops for bin_cmd_smooth
as well.
Seeing deprectation warnings on Julia nightly in CI but all tests passing for now.
LV is being deprecated for Julia 1.11+ and so we should consider moving away from it now. Should fall back to basic
@simd
or even remove the implementations that use these custom loops rather than themul!
calls.